SAFE public projects git trees. - safe/jmp/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 static inline 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 isn't Nosave, and is not in the range of pages
 171  *      statically defined as 'unsaveable', and it
 172  *      isn't a part of a free chunk of pages.
 173  */
 174
 175 static struct page *saveable_page(unsigned long pfn)
 176 {
 177         struct page *page;
 178
 179         if (!pfn_valid(pfn))
 180                 return NULL;
 181
 182         page = pfn_to_page(pfn);
 183
 184         if (PageNosave(page))
 185                 return NULL;
 186         if (PageReserved(page) && pfn_is_nosave(pfn))
 187                 return NULL;
 188         if (PageNosaveFree(page))
 189                 return NULL;
 190
 191         return page;
 192 }
 193
 194 unsigned int count_data_pages(void)
 195 {
 196         struct zone *zone;
 197         unsigned long pfn, max_zone_pfn;
 198         unsigned int n = 0;
 199
 200         for_each_zone (zone) {
 201                 if (is_highmem(zone))
 202                         continue;
 203                 mark_free_pages(zone);
 204                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
 205                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
 206                         n += !!saveable_page(pfn);
 207         }
 208         return n;
 209 }
 210
 211 static inline void copy_data_page(long *dst, long *src)
 212 {
 213         int n;
 214
 215         /* copy_page and memcpy are not usable for copying task structs. */
 216         for (n = PAGE_SIZE / sizeof(long); n; n--)
 217                 *dst++ = *src++;
 218 }
 219
 220 static void copy_data_pages(struct pbe *pblist)
 221 {
 222         struct zone *zone;
 223         unsigned long pfn, max_zone_pfn;
 224         struct pbe *pbe;
 225
 226         pbe = pblist;
 227         for_each_zone (zone) {
 228                 if (is_highmem(zone))
 229                         continue;
 230                 mark_free_pages(zone);
 231                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
 232                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) {
 233                         struct page *page = saveable_page(pfn);
 234
 235                         if (page) {
 236                                 void *ptr = page_address(page);
 237
 238                                 BUG_ON(!pbe);
 239                                 copy_data_page((void *)pbe->address, ptr);
 240                                 pbe->orig_address = (unsigned long)ptr;
 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 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 pfn, max_zone_pfn;
 387
 388         for_each_zone(zone) {
 389                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
 390                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
 391                         if (pfn_valid(pfn)) {
 392                                 struct page *page = pfn_to_page(pfn);
 393
 394                                 if (PageNosave(page) && PageNosaveFree(page)) {
 395                                         ClearPageNosave(page);
 396                                         ClearPageNosaveFree(page);
 397                                         free_page((long) page_address(page));
 398                                 }
 399                         }
 400         }
 401         nr_copy_pages = 0;
 402         nr_meta_pages = 0;
 403         pagedir_nosave = NULL;
 404         buffer = NULL;
 405 }
 406
 407
 408 /**
 409  *      enough_free_mem - Make sure we enough free memory to snapshot.
 410  *
 411  *      Returns TRUE or FALSE after checking the number of available
 412  *      free pages.
 413  */
 414
 415 static int enough_free_mem(unsigned int nr_pages)
 416 {
 417         struct zone *zone;
 418         unsigned int n = 0;
 419
 420         for_each_zone (zone)
 421                 if (!is_highmem(zone))
 422                         n += zone->free_pages;
 423         pr_debug("swsusp: available memory: %u pages\n", n);
 424         return n > (nr_pages + PAGES_FOR_IO +
 425                 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
 426 }
 427
 428 static int alloc_data_pages(struct pbe *pblist, gfp_t gfp_mask, int safe_needed)
 429 {
 430         struct pbe *p;
 431
 432         for_each_pbe (p, pblist) {
 433                 p->address = (unsigned long)alloc_image_page(gfp_mask, safe_needed);
 434                 if (!p->address)
 435                         return -ENOMEM;
 436         }
 437         return 0;
 438 }
 439
 440 static struct pbe *swsusp_alloc(unsigned int nr_pages)
 441 {
 442         struct pbe *pblist;
 443
 444         if (!(pblist = alloc_pagedir(nr_pages, GFP_ATOMIC | __GFP_COLD, 0))) {
 445                 printk(KERN_ERR "suspend: Allocating pagedir failed.\n");
 446                 return NULL;
 447         }
 448
 449         if (alloc_data_pages(pblist, GFP_ATOMIC | __GFP_COLD, 0)) {
 450                 printk(KERN_ERR "suspend: Allocating image pages failed.\n");
 451                 swsusp_free();
 452                 return NULL;
 453         }
 454
 455         return pblist;
 456 }
 457
 458 asmlinkage int swsusp_save(void)
 459 {
 460         unsigned int nr_pages;
 461
 462         pr_debug("swsusp: critical section: \n");
 463
 464         drain_local_pages();
 465         nr_pages = count_data_pages();
 466         printk("swsusp: Need to copy %u pages\n", nr_pages);
 467
 468         pr_debug("swsusp: pages needed: %u + %lu + %u, free: %u\n",
 469                  nr_pages,
 470                  (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE,
 471                  PAGES_FOR_IO, nr_free_pages());
 472
 473         if (!enough_free_mem(nr_pages)) {
 474                 printk(KERN_ERR "swsusp: Not enough free memory\n");
 475                 return -ENOMEM;
 476         }
 477
 478         pagedir_nosave = swsusp_alloc(nr_pages);
 479         if (!pagedir_nosave)
 480                 return -ENOMEM;
 481
 482         /* During allocating of suspend pagedir, new cold pages may appear.
 483          * Kill them.
 484          */
 485         drain_local_pages();
 486         copy_data_pages(pagedir_nosave);
 487
 488         /*
 489          * End of critical section. From now on, we can write to memory,
 490          * but we should not touch disk. This specially means we must _not_
 491          * touch swap space! Except we must write out our image of course.
 492          */
 493
 494         nr_copy_pages = nr_pages;
 495         nr_meta_pages = (nr_pages * sizeof(long) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 496
 497         printk("swsusp: critical section/: done (%d pages copied)\n", nr_pages);
 498         return 0;
 499 }
 500
 501 static void init_header(struct swsusp_info *info)
 502 {
 503         memset(info, 0, sizeof(struct swsusp_info));
 504         info->version_code = LINUX_VERSION_CODE;
 505         info->num_physpages = num_physpages;
 506         memcpy(&info->uts, &system_utsname, sizeof(system_utsname));
 507         info->cpus = num_online_cpus();
 508         info->image_pages = nr_copy_pages;
 509         info->pages = nr_copy_pages + nr_meta_pages + 1;
 510         info->size = info->pages;
 511         info->size <<= PAGE_SHIFT;
 512 }
 513
 514 /**
 515  *      pack_orig_addresses - the .orig_address fields of the PBEs from the
 516  *      list starting at @pbe are stored in the array @buf[] (1 page)
 517  */
 518
 519 static inline struct pbe *pack_orig_addresses(unsigned long *buf, struct pbe *pbe)
 520 {
 521         int j;
 522
 523         for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
 524                 buf[j] = pbe->orig_address;
 525                 pbe = pbe->next;
 526         }
 527         if (!pbe)
 528                 for (; j < PAGE_SIZE / sizeof(long); j++)
 529                         buf[j] = 0;
 530         return pbe;
 531 }
 532
 533 /**
 534  *      snapshot_read_next - used for reading the system memory snapshot.
 535  *
 536  *      On the first call to it @handle should point to a zeroed
 537  *      snapshot_handle structure.  The structure gets updated and a pointer
 538  *      to it should be passed to this function every next time.
 539  *
 540  *      The @count parameter should contain the number of bytes the caller
 541  *      wants to read from the snapshot.  It must not be zero.
 542  *
 543  *      On success the function returns a positive number.  Then, the caller
 544  *      is allowed to read up to the returned number of bytes from the memory
 545  *      location computed by the data_of() macro.  The number returned
 546  *      may be smaller than @count, but this only happens if the read would
 547  *      cross a page boundary otherwise.
 548  *
 549  *      The function returns 0 to indicate the end of data stream condition,
 550  *      and a negative number is returned on error.  In such cases the
 551  *      structure pointed to by @handle is not updated and should not be used
 552  *      any more.
 553  */
 554
 555 int snapshot_read_next(struct snapshot_handle *handle, size_t count)
 556 {
 557         if (handle->cur > nr_meta_pages + nr_copy_pages)
 558                 return 0;
 559         if (!buffer) {
 560                 /* This makes the buffer be freed by swsusp_free() */
 561                 buffer = alloc_image_page(GFP_ATOMIC, 0);
 562                 if (!buffer)
 563                         return -ENOMEM;
 564         }
 565         if (!handle->offset) {
 566                 init_header((struct swsusp_info *)buffer);
 567                 handle->buffer = buffer;
 568                 handle->pbe = pagedir_nosave;
 569         }
 570         if (handle->prev < handle->cur) {
 571                 if (handle->cur <= nr_meta_pages) {
 572                         handle->pbe = pack_orig_addresses(buffer, handle->pbe);
 573                         if (!handle->pbe)
 574                                 handle->pbe = pagedir_nosave;
 575                 } else {
 576                         handle->buffer = (void *)handle->pbe->address;
 577                         handle->pbe = handle->pbe->next;
 578                 }
 579                 handle->prev = handle->cur;
 580         }
 581         handle->buf_offset = handle->cur_offset;
 582         if (handle->cur_offset + count >= PAGE_SIZE) {
 583                 count = PAGE_SIZE - handle->cur_offset;
 584                 handle->cur_offset = 0;
 585                 handle->cur++;
 586         } else {
 587                 handle->cur_offset += count;
 588         }
 589         handle->offset += count;
 590         return count;
 591 }
 592
 593 /**
 594  *      mark_unsafe_pages - mark the pages that cannot be used for storing
 595  *      the image during resume, because they conflict with the pages that
 596  *      had been used before suspend
 597  */
 598
 599 static int mark_unsafe_pages(struct pbe *pblist)
 600 {
 601         struct zone *zone;
 602         unsigned long pfn, max_zone_pfn;
 603         struct pbe *p;
 604
 605         if (!pblist) /* a sanity check */
 606                 return -EINVAL;
 607
 608         /* Clear page flags */
 609         for_each_zone (zone) {
 610                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
 611                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
 612                         if (pfn_valid(pfn))
 613                                 ClearPageNosaveFree(pfn_to_page(pfn));
 614         }
 615
 616         /* Mark orig addresses */
 617         for_each_pbe (p, pblist) {
 618                 if (virt_addr_valid(p->orig_address))
 619                         SetPageNosaveFree(virt_to_page(p->orig_address));
 620                 else
 621                         return -EFAULT;
 622         }
 623
 624         unsafe_pages = 0;
 625
 626         return 0;
 627 }
 628
 629 static void copy_page_backup_list(struct pbe *dst, struct pbe *src)
 630 {
 631         /* We assume both lists contain the same number of elements */
 632         while (src) {
 633                 dst->orig_address = src->orig_address;
 634                 dst = dst->next;
 635                 src = src->next;
 636         }
 637 }
 638
 639 static int check_header(struct swsusp_info *info)
 640 {
 641         char *reason = NULL;
 642
 643         if (info->version_code != LINUX_VERSION_CODE)
 644                 reason = "kernel version";
 645         if (info->num_physpages != num_physpages)
 646                 reason = "memory size";
 647         if (strcmp(info->uts.sysname,system_utsname.sysname))
 648                 reason = "system type";
 649         if (strcmp(info->uts.release,system_utsname.release))
 650                 reason = "kernel release";
 651         if (strcmp(info->uts.version,system_utsname.version))
 652                 reason = "version";
 653         if (strcmp(info->uts.machine,system_utsname.machine))
 654                 reason = "machine";
 655         if (reason) {
 656                 printk(KERN_ERR "swsusp: Resume mismatch: %s\n", reason);
 657                 return -EPERM;
 658         }
 659         return 0;
 660 }
 661
 662 /**
 663  *      load header - check the image header and copy data from it
 664  */
 665
 666 static int load_header(struct snapshot_handle *handle,
 667                               struct swsusp_info *info)
 668 {
 669         int error;
 670         struct pbe *pblist;
 671
 672         error = check_header(info);
 673         if (!error) {
 674                 pblist = alloc_pagedir(info->image_pages, GFP_ATOMIC, 0);
 675                 if (!pblist)
 676                         return -ENOMEM;
 677                 pagedir_nosave = pblist;
 678                 handle->pbe = pblist;
 679                 nr_copy_pages = info->image_pages;
 680                 nr_meta_pages = info->pages - info->image_pages - 1;
 681         }
 682         return error;
 683 }
 684
 685 /**
 686  *      unpack_orig_addresses - copy the elements of @buf[] (1 page) to
 687  *      the PBEs in the list starting at @pbe
 688  */
 689
 690 static inline struct pbe *unpack_orig_addresses(unsigned long *buf,
 691                                                 struct pbe *pbe)
 692 {
 693         int j;
 694
 695         for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
 696                 pbe->orig_address = buf[j];
 697                 pbe = pbe->next;
 698         }
 699         return pbe;
 700 }
 701
 702 /**
 703  *      prepare_image - use metadata contained in the PBE list
 704  *      pointed to by pagedir_nosave to mark the pages that will
 705  *      be overwritten in the process of restoring the system
 706  *      memory state from the image ("unsafe" pages) and allocate
 707  *      memory for the image
 708  *
 709  *      The idea is to allocate the PBE list first and then
 710  *      allocate as many pages as it's needed for the image data,
 711  *      but not to assign these pages to the PBEs initially.
 712  *      Instead, we just mark them as allocated and create a list
 713  *      of "safe" which will be used later
 714  */
 715
 716 struct safe_page {
 717         struct safe_page *next;
 718         char padding[PAGE_SIZE - sizeof(void *)];
 719 };
 720
 721 static struct safe_page *safe_pages;
 722
 723 static int prepare_image(struct snapshot_handle *handle)
 724 {
 725         int error = 0;
 726         unsigned int nr_pages = nr_copy_pages;
 727         struct pbe *p, *pblist = NULL;
 728
 729         p = pagedir_nosave;
 730         error = mark_unsafe_pages(p);
 731         if (!error) {
 732                 pblist = alloc_pagedir(nr_pages, GFP_ATOMIC, 1);
 733                 if (pblist)
 734                         copy_page_backup_list(pblist, p);
 735                 free_pagedir(p, 0);
 736                 if (!pblist)
 737                         error = -ENOMEM;
 738         }
 739         safe_pages = NULL;
 740         if (!error && nr_pages > unsafe_pages) {
 741                 nr_pages -= unsafe_pages;
 742                 while (nr_pages--) {
 743                         struct safe_page *ptr;
 744
 745                         ptr = (struct safe_page *)get_zeroed_page(GFP_ATOMIC);
 746                         if (!ptr) {
 747                                 error = -ENOMEM;
 748                                 break;
 749                         }
 750                         if (!PageNosaveFree(virt_to_page(ptr))) {
 751                                 /* The page is "safe", add it to the list */
 752                                 ptr->next = safe_pages;
 753                                 safe_pages = ptr;
 754                         }
 755                         /* Mark the page as allocated */
 756                         SetPageNosave(virt_to_page(ptr));
 757                         SetPageNosaveFree(virt_to_page(ptr));
 758                 }
 759         }
 760         if (!error) {
 761                 pagedir_nosave = pblist;
 762         } else {
 763                 handle->pbe = NULL;
 764                 swsusp_free();
 765         }
 766         return error;
 767 }
 768
 769 static void *get_buffer(struct snapshot_handle *handle)
 770 {
 771         struct pbe *pbe = handle->pbe, *last = handle->last_pbe;
 772         struct page *page = virt_to_page(pbe->orig_address);
 773
 774         if (PageNosave(page) && PageNosaveFree(page)) {
 775                 /*
 776                  * We have allocated the "original" page frame and we can
 777                  * use it directly to store the read page
 778                  */
 779                 pbe->address = 0;
 780                 if (last && last->next)
 781                         last->next = NULL;
 782                 return (void *)pbe->orig_address;
 783         }
 784         /*
 785          * The "original" page frame has not been allocated and we have to
 786          * use a "safe" page frame to store the read page
 787          */
 788         pbe->address = (unsigned long)safe_pages;
 789         safe_pages = safe_pages->next;
 790         if (last)
 791                 last->next = pbe;
 792         handle->last_pbe = pbe;
 793         return (void *)pbe->address;
 794 }
 795
 796 /**
 797  *      snapshot_write_next - used for writing the system memory snapshot.
 798  *
 799  *      On the first call to it @handle should point to a zeroed
 800  *      snapshot_handle structure.  The structure gets updated and a pointer
 801  *      to it should be passed to this function every next time.
 802  *
 803  *      The @count parameter should contain the number of bytes the caller
 804  *      wants to write to the image.  It must not be zero.
 805  *
 806  *      On success the function returns a positive number.  Then, the caller
 807  *      is allowed to write up to the returned number of bytes to the memory
 808  *      location computed by the data_of() macro.  The number returned
 809  *      may be smaller than @count, but this only happens if the write would
 810  *      cross a page boundary otherwise.
 811  *
 812  *      The function returns 0 to indicate the "end of file" condition,
 813  *      and a negative number is returned on error.  In such cases the
 814  *      structure pointed to by @handle is not updated and should not be used
 815  *      any more.
 816  */
 817
 818 int snapshot_write_next(struct snapshot_handle *handle, size_t count)
 819 {
 820         int error = 0;
 821
 822         if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages)
 823                 return 0;
 824         if (!buffer) {
 825                 /* This makes the buffer be freed by swsusp_free() */
 826                 buffer = alloc_image_page(GFP_ATOMIC, 0);
 827                 if (!buffer)
 828                         return -ENOMEM;
 829         }
 830         if (!handle->offset)
 831                 handle->buffer = buffer;
 832         handle->sync_read = 1;
 833         if (handle->prev < handle->cur) {
 834                 if (!handle->prev) {
 835                         error = load_header(handle,
 836                                         (struct swsusp_info *)buffer);
 837                         if (error)
 838                                 return error;
 839                 } else if (handle->prev <= nr_meta_pages) {
 840                         handle->pbe = unpack_orig_addresses(buffer,
 841                                                         handle->pbe);
 842                         if (!handle->pbe) {
 843                                 error = prepare_image(handle);
 844                                 if (error)
 845                                         return error;
 846                                 handle->pbe = pagedir_nosave;
 847                                 handle->last_pbe = NULL;
 848                                 handle->buffer = get_buffer(handle);
 849                                 handle->sync_read = 0;
 850                         }
 851                 } else {
 852                         handle->pbe = handle->pbe->next;
 853                         handle->buffer = get_buffer(handle);
 854                         handle->sync_read = 0;
 855                 }
 856                 handle->prev = handle->cur;
 857         }
 858         handle->buf_offset = handle->cur_offset;
 859         if (handle->cur_offset + count >= PAGE_SIZE) {
 860                 count = PAGE_SIZE - handle->cur_offset;
 861                 handle->cur_offset = 0;
 862                 handle->cur++;
 863         } else {
 864                 handle->cur_offset += count;
 865         }
 866         handle->offset += count;
 867         return count;
 868 }
 869
 870 int snapshot_image_loaded(struct snapshot_handle *handle)
 871 {
 872         return !(!handle->pbe || handle->pbe->next || !nr_copy_pages ||
 873                 handle->cur <= nr_meta_pages + nr_copy_pages);
 874 }