git.oblomov.eu Git - linux-2.6/blob - 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 /**
 160  *      @safe_needed - on resume, for storing the PBE list and the image,
 161  *      we can only use memory pages that do not conflict with the pages
 162  *      used before suspend.
 163  *
 164  *      The unsafe pages are marked with the PG_nosave_free flag
 165  *      and we count them using unsafe_pages
 166  */
 167
 168 static unsigned int unsafe_pages;
 169
 170 static void *alloc_image_page(gfp_t gfp_mask, int safe_needed)
 171 {
 172         void *res;
 173
 174         res = (void *)get_zeroed_page(gfp_mask);
 175         if (safe_needed)
 176                 while (res && PageNosaveFree(virt_to_page(res))) {
 177                         /* The page is unsafe, mark it for swsusp_free() */
 178                         SetPageNosave(virt_to_page(res));
 179                         unsafe_pages++;
 180                         res = (void *)get_zeroed_page(gfp_mask);
 181                 }
 182         if (res) {
 183                 SetPageNosave(virt_to_page(res));
 184                 SetPageNosaveFree(virt_to_page(res));
 185         }
 186         return res;
 187 }
 188
 189 unsigned long get_safe_page(gfp_t gfp_mask)
 190 {
 191         return (unsigned long)alloc_image_page(gfp_mask, 1);
 192 }
 193
 194 /**
 195  *      free_image_page - free page represented by @addr, allocated with
 196  *      alloc_image_page (page flags set by it must be cleared)
 197  */
 198
 199 static inline void free_image_page(void *addr, int clear_nosave_free)
 200 {
 201         ClearPageNosave(virt_to_page(addr));
 202         if (clear_nosave_free)
 203                 ClearPageNosaveFree(virt_to_page(addr));
 204         free_page((unsigned long)addr);
 205 }
 206
 207 /**
 208  *      pfn_is_nosave - check if given pfn is in the 'nosave' section
 209  */
 210
 211 static inline int pfn_is_nosave(unsigned long pfn)
 212 {
 213         unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
 214         unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
 215         return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
 216 }
 217
 218 /**
 219  *      saveable - Determine whether a page should be cloned or not.
 220  *      @pfn:   The page
 221  *
 222  *      We save a page if it isn't Nosave, and is not in the range of pages
 223  *      statically defined as 'unsaveable', and it
 224  *      isn't a part of a free chunk of pages.
 225  */
 226
 227 static struct page *saveable_page(unsigned long pfn)
 228 {
 229         struct page *page;
 230
 231         if (!pfn_valid(pfn))
 232                 return NULL;
 233
 234         page = pfn_to_page(pfn);
 235
 236         if (PageNosave(page))
 237                 return NULL;
 238         if (PageReserved(page) && pfn_is_nosave(pfn))
 239                 return NULL;
 240         if (PageNosaveFree(page))
 241                 return NULL;
 242
 243         return page;
 244 }
 245
 246 unsigned int count_data_pages(void)
 247 {
 248         struct zone *zone;
 249         unsigned long pfn, max_zone_pfn;
 250         unsigned int n = 0;
 251
 252         for_each_zone (zone) {
 253                 if (is_highmem(zone))
 254                         continue;
 255                 mark_free_pages(zone);
 256                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
 257                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
 258                         n += !!saveable_page(pfn);
 259         }
 260         return n;
 261 }
 262
 263 static inline void copy_data_page(long *dst, long *src)
 264 {
 265         int n;
 266
 267         /* copy_page and memcpy are not usable for copying task structs. */
 268         for (n = PAGE_SIZE / sizeof(long); n; n--)
 269                 *dst++ = *src++;
 270 }
 271
 272 static void copy_data_pages(struct pbe *pblist)
 273 {
 274         struct zone *zone;
 275         unsigned long pfn, max_zone_pfn;
 276         struct pbe *pbe;
 277
 278         pbe = pblist;
 279         for_each_zone (zone) {
 280                 if (is_highmem(zone))
 281                         continue;
 282                 mark_free_pages(zone);
 283                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
 284                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) {
 285                         struct page *page = saveable_page(pfn);
 286
 287                         if (page) {
 288                                 void *ptr = page_address(page);
 289
 290                                 BUG_ON(!pbe);
 291                                 copy_data_page((void *)pbe->address, ptr);
 292                                 pbe->orig_address = (unsigned long)ptr;
 293                                 pbe = pbe->next;
 294                         }
 295                 }
 296         }
 297         BUG_ON(pbe);
 298 }
 299
 300 /**
 301  *      free_pagedir - free pages allocated with alloc_pagedir()
 302  */
 303
 304 static void free_pagedir(struct pbe *pblist, int clear_nosave_free)
 305 {
 306         struct pbe *pbe;
 307
 308         while (pblist) {
 309                 pbe = (pblist + PB_PAGE_SKIP)->next;
 310                 free_image_page(pblist, clear_nosave_free);
 311                 pblist = pbe;
 312         }
 313 }
 314
 315 /**
 316  *      fill_pb_page - Create a list of PBEs on a given memory page
 317  */
 318
 319 static inline void fill_pb_page(struct pbe *pbpage)
 320 {
 321         struct pbe *p;
 322
 323         p = pbpage;
 324         pbpage += PB_PAGE_SKIP;
 325         do
 326                 p->next = p + 1;
 327         while (++p < pbpage);
 328 }
 329
 330 /**
 331  *      create_pbe_list - Create a list of PBEs on top of a given chain
 332  *      of memory pages allocated with alloc_pagedir()
 333  */
 334
 335 static inline void create_pbe_list(struct pbe *pblist, unsigned int nr_pages)
 336 {
 337         struct pbe *pbpage, *p;
 338         unsigned int num = PBES_PER_PAGE;
 339
 340         for_each_pb_page (pbpage, pblist) {
 341                 if (num >= nr_pages)
 342                         break;
 343
 344                 fill_pb_page(pbpage);
 345                 num += PBES_PER_PAGE;
 346         }
 347         if (pbpage) {
 348                 for (num -= PBES_PER_PAGE - 1, p = pbpage; num < nr_pages; p++, num++)
 349                         p->next = p + 1;
 350                 p->next = NULL;
 351         }
 352 }
 353
 354 /**
 355  *      alloc_pagedir - Allocate the page directory.
 356  *
 357  *      First, determine exactly how many pages we need and
 358  *      allocate them.
 359  *
 360  *      We arrange the pages in a chain: each page is an array of PBES_PER_PAGE
 361  *      struct pbe elements (pbes) and the last element in the page points
 362  *      to the next page.
 363  *
 364  *      On each page we set up a list of struct_pbe elements.
 365  */
 366
 367 static struct pbe *alloc_pagedir(unsigned int nr_pages, gfp_t gfp_mask,
 368                                  int safe_needed)
 369 {
 370         unsigned int num;
 371         struct pbe *pblist, *pbe;
 372
 373         if (!nr_pages)
 374                 return NULL;
 375
 376         pblist = alloc_image_page(gfp_mask, safe_needed);
 377         /* FIXME: rewrite this ugly loop */
 378         for (pbe = pblist, num = PBES_PER_PAGE; pbe && num < nr_pages;
 379                         pbe = pbe->next, num += PBES_PER_PAGE) {
 380                 pbe += PB_PAGE_SKIP;
 381                 pbe->next = alloc_image_page(gfp_mask, safe_needed);
 382         }
 383         if (!pbe) { /* get_zeroed_page() failed */
 384                 free_pagedir(pblist, 1);
 385                 pblist = NULL;
 386         } else
 387                 create_pbe_list(pblist, nr_pages);
 388         return pblist;
 389 }
 390
 391 /**
 392  * Free pages we allocated for suspend. Suspend pages are alocated
 393  * before atomic copy, so we need to free them after resume.
 394  */
 395
 396 void swsusp_free(void)
 397 {
 398         struct zone *zone;
 399         unsigned long pfn, max_zone_pfn;
 400
 401         for_each_zone(zone) {
 402                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
 403                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
 404                         if (pfn_valid(pfn)) {
 405                                 struct page *page = pfn_to_page(pfn);
 406
 407                                 if (PageNosave(page) && PageNosaveFree(page)) {
 408                                         ClearPageNosave(page);
 409                                         ClearPageNosaveFree(page);
 410                                         free_page((long) page_address(page));
 411                                 }
 412                         }
 413         }
 414         nr_copy_pages = 0;
 415         nr_meta_pages = 0;
 416         pagedir_nosave = NULL;
 417         buffer = NULL;
 418 }
 419
 420
 421 /**
 422  *      enough_free_mem - Make sure we enough free memory to snapshot.
 423  *
 424  *      Returns TRUE or FALSE after checking the number of available
 425  *      free pages.
 426  */
 427
 428 static int enough_free_mem(unsigned int nr_pages)
 429 {
 430         struct zone *zone;
 431         unsigned int n = 0;
 432
 433         for_each_zone (zone)
 434                 if (!is_highmem(zone))
 435                         n += zone->free_pages;
 436         pr_debug("swsusp: available memory: %u pages\n", n);
 437         return n > (nr_pages + PAGES_FOR_IO +
 438                 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
 439 }
 440
 441 static int alloc_data_pages(struct pbe *pblist, gfp_t gfp_mask, int safe_needed)
 442 {
 443         struct pbe *p;
 444
 445         for_each_pbe (p, pblist) {
 446                 p->address = (unsigned long)alloc_image_page(gfp_mask, safe_needed);
 447                 if (!p->address)
 448                         return -ENOMEM;
 449         }
 450         return 0;
 451 }
 452
 453 static struct pbe *swsusp_alloc(unsigned int nr_pages)
 454 {
 455         struct pbe *pblist;
 456
 457         if (!(pblist = alloc_pagedir(nr_pages, GFP_ATOMIC | __GFP_COLD, 0))) {
 458                 printk(KERN_ERR "suspend: Allocating pagedir failed.\n");
 459                 return NULL;
 460         }
 461
 462         if (alloc_data_pages(pblist, GFP_ATOMIC | __GFP_COLD, 0)) {
 463                 printk(KERN_ERR "suspend: Allocating image pages failed.\n");
 464                 swsusp_free();
 465                 return NULL;
 466         }
 467
 468         return pblist;
 469 }
 470
 471 asmlinkage int swsusp_save(void)
 472 {
 473         unsigned int nr_pages;
 474
 475         pr_debug("swsusp: critical section: \n");
 476
 477         drain_local_pages();
 478         nr_pages = count_data_pages();
 479         printk("swsusp: Need to copy %u pages\n", nr_pages);
 480
 481         pr_debug("swsusp: pages needed: %u + %lu + %u, free: %u\n",
 482                  nr_pages,
 483                  (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE,
 484                  PAGES_FOR_IO, nr_free_pages());
 485
 486         if (!enough_free_mem(nr_pages)) {
 487                 printk(KERN_ERR "swsusp: Not enough free memory\n");
 488                 return -ENOMEM;
 489         }
 490
 491         pagedir_nosave = swsusp_alloc(nr_pages);
 492         if (!pagedir_nosave)
 493                 return -ENOMEM;
 494
 495         /* During allocating of suspend pagedir, new cold pages may appear.
 496          * Kill them.
 497          */
 498         drain_local_pages();
 499         copy_data_pages(pagedir_nosave);
 500
 501         /*
 502          * End of critical section. From now on, we can write to memory,
 503          * but we should not touch disk. This specially means we must _not_
 504          * touch swap space! Except we must write out our image of course.
 505          */
 506
 507         nr_copy_pages = nr_pages;
 508         nr_meta_pages = (nr_pages * sizeof(long) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 509
 510         printk("swsusp: critical section/: done (%d pages copied)\n", nr_pages);
 511         return 0;
 512 }
 513
 514 static void init_header(struct swsusp_info *info)
 515 {
 516         memset(info, 0, sizeof(struct swsusp_info));
 517         info->version_code = LINUX_VERSION_CODE;
 518         info->num_physpages = num_physpages;
 519         memcpy(&info->uts, &system_utsname, sizeof(system_utsname));
 520         info->cpus = num_online_cpus();
 521         info->image_pages = nr_copy_pages;
 522         info->pages = nr_copy_pages + nr_meta_pages + 1;
 523         info->size = info->pages;
 524         info->size <<= PAGE_SHIFT;
 525 }
 526
 527 /**
 528  *      pack_orig_addresses - the .orig_address fields of the PBEs from the
 529  *      list starting at @pbe are stored in the array @buf[] (1 page)
 530  */
 531
 532 static inline struct pbe *pack_orig_addresses(unsigned long *buf, struct pbe *pbe)
 533 {
 534         int j;
 535
 536         for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
 537                 buf[j] = pbe->orig_address;
 538                 pbe = pbe->next;
 539         }
 540         if (!pbe)
 541                 for (; j < PAGE_SIZE / sizeof(long); j++)
 542                         buf[j] = 0;
 543         return pbe;
 544 }
 545
 546 /**
 547  *      snapshot_read_next - used for reading the system memory snapshot.
 548  *
 549  *      On the first call to it @handle should point to a zeroed
 550  *      snapshot_handle structure.  The structure gets updated and a pointer
 551  *      to it should be passed to this function every next time.
 552  *
 553  *      The @count parameter should contain the number of bytes the caller
 554  *      wants to read from the snapshot.  It must not be zero.
 555  *
 556  *      On success the function returns a positive number.  Then, the caller
 557  *      is allowed to read up to the returned number of bytes from the memory
 558  *      location computed by the data_of() macro.  The number returned
 559  *      may be smaller than @count, but this only happens if the read would
 560  *      cross a page boundary otherwise.
 561  *
 562  *      The function returns 0 to indicate the end of data stream condition,
 563  *      and a negative number is returned on error.  In such cases the
 564  *      structure pointed to by @handle is not updated and should not be used
 565  *      any more.
 566  */
 567
 568 int snapshot_read_next(struct snapshot_handle *handle, size_t count)
 569 {
 570         if (handle->cur > nr_meta_pages + nr_copy_pages)
 571                 return 0;
 572         if (!buffer) {
 573                 /* This makes the buffer be freed by swsusp_free() */
 574                 buffer = alloc_image_page(GFP_ATOMIC, 0);
 575                 if (!buffer)
 576                         return -ENOMEM;
 577         }
 578         if (!handle->offset) {
 579                 init_header((struct swsusp_info *)buffer);
 580                 handle->buffer = buffer;
 581                 handle->pbe = pagedir_nosave;
 582         }
 583         if (handle->prev < handle->cur) {
 584                 if (handle->cur <= nr_meta_pages) {
 585                         handle->pbe = pack_orig_addresses(buffer, handle->pbe);
 586                         if (!handle->pbe)
 587                                 handle->pbe = pagedir_nosave;
 588                 } else {
 589                         handle->buffer = (void *)handle->pbe->address;
 590                         handle->pbe = handle->pbe->next;
 591                 }
 592                 handle->prev = handle->cur;
 593         }
 594         handle->buf_offset = handle->cur_offset;
 595         if (handle->cur_offset + count >= PAGE_SIZE) {
 596                 count = PAGE_SIZE - handle->cur_offset;
 597                 handle->cur_offset = 0;
 598                 handle->cur++;
 599         } else {
 600                 handle->cur_offset += count;
 601         }
 602         handle->offset += count;
 603         return count;
 604 }
 605
 606 /**
 607  *      mark_unsafe_pages - mark the pages that cannot be used for storing
 608  *      the image during resume, because they conflict with the pages that
 609  *      had been used before suspend
 610  */
 611
 612 static int mark_unsafe_pages(struct pbe *pblist)
 613 {
 614         struct zone *zone;
 615         unsigned long pfn, max_zone_pfn;
 616         struct pbe *p;
 617
 618         if (!pblist) /* a sanity check */
 619                 return -EINVAL;
 620
 621         /* Clear page flags */
 622         for_each_zone (zone) {
 623                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
 624                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
 625                         if (pfn_valid(pfn))
 626                                 ClearPageNosaveFree(pfn_to_page(pfn));
 627         }
 628
 629         /* Mark orig addresses */
 630         for_each_pbe (p, pblist) {
 631                 if (virt_addr_valid(p->orig_address))
 632                         SetPageNosaveFree(virt_to_page(p->orig_address));
 633                 else
 634                         return -EFAULT;
 635         }
 636
 637         unsafe_pages = 0;
 638
 639         return 0;
 640 }
 641
 642 static void copy_page_backup_list(struct pbe *dst, struct pbe *src)
 643 {
 644         /* We assume both lists contain the same number of elements */
 645         while (src) {
 646                 dst->orig_address = src->orig_address;
 647                 dst = dst->next;
 648                 src = src->next;
 649         }
 650 }
 651
 652 static int check_header(struct swsusp_info *info)
 653 {
 654         char *reason = NULL;
 655
 656         if (info->version_code != LINUX_VERSION_CODE)
 657                 reason = "kernel version";
 658         if (info->num_physpages != num_physpages)
 659                 reason = "memory size";
 660         if (strcmp(info->uts.sysname,system_utsname.sysname))
 661                 reason = "system type";
 662         if (strcmp(info->uts.release,system_utsname.release))
 663                 reason = "kernel release";
 664         if (strcmp(info->uts.version,system_utsname.version))
 665                 reason = "version";
 666         if (strcmp(info->uts.machine,system_utsname.machine))
 667                 reason = "machine";
 668         if (reason) {
 669                 printk(KERN_ERR "swsusp: Resume mismatch: %s\n", reason);
 670                 return -EPERM;
 671         }
 672         return 0;
 673 }
 674
 675 /**
 676  *      load header - check the image header and copy data from it
 677  */
 678
 679 static int load_header(struct snapshot_handle *handle,
 680                               struct swsusp_info *info)
 681 {
 682         int error;
 683         struct pbe *pblist;
 684
 685         error = check_header(info);
 686         if (!error) {
 687                 pblist = alloc_pagedir(info->image_pages, GFP_ATOMIC, 0);
 688                 if (!pblist)
 689                         return -ENOMEM;
 690                 pagedir_nosave = pblist;
 691                 handle->pbe = pblist;
 692                 nr_copy_pages = info->image_pages;
 693                 nr_meta_pages = info->pages - info->image_pages - 1;
 694         }
 695         return error;
 696 }
 697
 698 /**
 699  *      unpack_orig_addresses - copy the elements of @buf[] (1 page) to
 700  *      the PBEs in the list starting at @pbe
 701  */
 702
 703 static inline struct pbe *unpack_orig_addresses(unsigned long *buf,
 704                                                 struct pbe *pbe)
 705 {
 706         int j;
 707
 708         for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
 709                 pbe->orig_address = buf[j];
 710                 pbe = pbe->next;
 711         }
 712         return pbe;
 713 }
 714
 715 /**
 716  *      prepare_image - use metadata contained in the PBE list
 717  *      pointed to by pagedir_nosave to mark the pages that will
 718  *      be overwritten in the process of restoring the system
 719  *      memory state from the image ("unsafe" pages) and allocate
 720  *      memory for the image
 721  *
 722  *      The idea is to allocate the PBE list first and then
 723  *      allocate as many pages as it's needed for the image data,
 724  *      but not to assign these pages to the PBEs initially.
 725  *      Instead, we just mark them as allocated and create a list
 726  *      of "safe" which will be used later
 727  */
 728
 729 struct safe_page {
 730         struct safe_page *next;
 731         char padding[PAGE_SIZE - sizeof(void *)];
 732 };
 733
 734 static struct safe_page *safe_pages;
 735
 736 static int prepare_image(struct snapshot_handle *handle)
 737 {
 738         int error = 0;
 739         unsigned int nr_pages = nr_copy_pages;
 740         struct pbe *p, *pblist = NULL;
 741
 742         p = pagedir_nosave;
 743         error = mark_unsafe_pages(p);
 744         if (!error) {
 745                 pblist = alloc_pagedir(nr_pages, GFP_ATOMIC, 1);
 746                 if (pblist)
 747                         copy_page_backup_list(pblist, p);
 748                 free_pagedir(p, 0);
 749                 if (!pblist)
 750                         error = -ENOMEM;
 751         }
 752         safe_pages = NULL;
 753         if (!error && nr_pages > unsafe_pages) {
 754                 nr_pages -= unsafe_pages;
 755                 while (nr_pages--) {
 756                         struct safe_page *ptr;
 757
 758                         ptr = (struct safe_page *)get_zeroed_page(GFP_ATOMIC);
 759                         if (!ptr) {
 760                                 error = -ENOMEM;
 761                                 break;
 762                         }
 763                         if (!PageNosaveFree(virt_to_page(ptr))) {
 764                                 /* The page is "safe", add it to the list */
 765                                 ptr->next = safe_pages;
 766                                 safe_pages = ptr;
 767                         }
 768                         /* Mark the page as allocated */
 769                         SetPageNosave(virt_to_page(ptr));
 770                         SetPageNosaveFree(virt_to_page(ptr));
 771                 }
 772         }
 773         if (!error) {
 774                 pagedir_nosave = pblist;
 775         } else {
 776                 handle->pbe = NULL;
 777                 swsusp_free();
 778         }
 779         return error;
 780 }
 781
 782 static void *get_buffer(struct snapshot_handle *handle)
 783 {
 784         struct pbe *pbe = handle->pbe, *last = handle->last_pbe;
 785         struct page *page = virt_to_page(pbe->orig_address);
 786
 787         if (PageNosave(page) && PageNosaveFree(page)) {
 788                 /*
 789                  * We have allocated the "original" page frame and we can
 790                  * use it directly to store the read page
 791                  */
 792                 pbe->address = 0;
 793                 if (last && last->next)
 794                         last->next = NULL;
 795                 return (void *)pbe->orig_address;
 796         }
 797         /*
 798          * The "original" page frame has not been allocated and we have to
 799          * use a "safe" page frame to store the read page
 800          */
 801         pbe->address = (unsigned long)safe_pages;
 802         safe_pages = safe_pages->next;
 803         if (last)
 804                 last->next = pbe;
 805         handle->last_pbe = pbe;
 806         return (void *)pbe->address;
 807 }
 808
 809 /**
 810  *      snapshot_write_next - used for writing the system memory snapshot.
 811  *
 812  *      On the first call to it @handle should point to a zeroed
 813  *      snapshot_handle structure.  The structure gets updated and a pointer
 814  *      to it should be passed to this function every next time.
 815  *
 816  *      The @count parameter should contain the number of bytes the caller
 817  *      wants to write to the image.  It must not be zero.
 818  *
 819  *      On success the function returns a positive number.  Then, the caller
 820  *      is allowed to write up to the returned number of bytes to the memory
 821  *      location computed by the data_of() macro.  The number returned
 822  *      may be smaller than @count, but this only happens if the write would
 823  *      cross a page boundary otherwise.
 824  *
 825  *      The function returns 0 to indicate the "end of file" condition,
 826  *      and a negative number is returned on error.  In such cases the
 827  *      structure pointed to by @handle is not updated and should not be used
 828  *      any more.
 829  */
 830
 831 int snapshot_write_next(struct snapshot_handle *handle, size_t count)
 832 {
 833         int error = 0;
 834
 835         if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages)
 836                 return 0;
 837         if (!buffer) {
 838                 /* This makes the buffer be freed by swsusp_free() */
 839                 buffer = alloc_image_page(GFP_ATOMIC, 0);
 840                 if (!buffer)
 841                         return -ENOMEM;
 842         }
 843         if (!handle->offset)
 844                 handle->buffer = buffer;
 845         handle->sync_read = 1;
 846         if (handle->prev < handle->cur) {
 847                 if (!handle->prev) {
 848                         error = load_header(handle,
 849                                         (struct swsusp_info *)buffer);
 850                         if (error)
 851                                 return error;
 852                 } else if (handle->prev <= nr_meta_pages) {
 853                         handle->pbe = unpack_orig_addresses(buffer,
 854                                                         handle->pbe);
 855                         if (!handle->pbe) {
 856                                 error = prepare_image(handle);
 857                                 if (error)
 858                                         return error;
 859                                 handle->pbe = pagedir_nosave;
 860                                 handle->last_pbe = NULL;
 861                                 handle->buffer = get_buffer(handle);
 862                                 handle->sync_read = 0;
 863                         }
 864                 } else {
 865                         handle->pbe = handle->pbe->next;
 866                         handle->buffer = get_buffer(handle);
 867                         handle->sync_read = 0;
 868                 }
 869                 handle->prev = handle->cur;
 870         }
 871         handle->buf_offset = handle->cur_offset;
 872         if (handle->cur_offset + count >= PAGE_SIZE) {
 873                 count = PAGE_SIZE - handle->cur_offset;
 874                 handle->cur_offset = 0;
 875                 handle->cur++;
 876         } else {
 877                 handle->cur_offset += count;
 878         }
 879         handle->offset += count;
 880         return count;
 881 }
 882
 883 int snapshot_image_loaded(struct snapshot_handle *handle)
 884 {
 885         return !(!handle->pbe || handle->pbe->next || !nr_copy_pages ||
 886                 handle->cur <= nr_meta_pages + nr_copy_pages);
 887 }