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 /* List of PBEs used for creating and restoring the suspend image */
  38 struct pbe *restore_pblist;
  39
  40 static unsigned int nr_copy_pages;
  41 static unsigned int nr_meta_pages;
  42 static unsigned long *buffer;
  43
  44 #ifdef CONFIG_HIGHMEM
  45 unsigned int count_highmem_pages(void)
  46 {
  47         struct zone *zone;
  48         unsigned long zone_pfn;
  49         unsigned int n = 0;
  50
  51         for_each_zone (zone)
  52                 if (is_highmem(zone)) {
  53                         mark_free_pages(zone);
  54                         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; zone_pfn++) {
  55                                 struct page *page;
  56                                 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
  57                                 if (!pfn_valid(pfn))
  58                                         continue;
  59                                 page = pfn_to_page(pfn);
  60                                 if (PageReserved(page))
  61                                         continue;
  62                                 if (PageNosaveFree(page))
  63                                         continue;
  64                                 n++;
  65                         }
  66                 }
  67         return n;
  68 }
  69
  70 struct highmem_page {
  71         char *data;
  72         struct page *page;
  73         struct highmem_page *next;
  74 };
  75
  76 static struct highmem_page *highmem_copy;
  77
  78 static int save_highmem_zone(struct zone *zone)
  79 {
  80         unsigned long zone_pfn;
  81         mark_free_pages(zone);
  82         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
  83                 struct page *page;
  84                 struct highmem_page *save;
  85                 void *kaddr;
  86                 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
  87
  88                 if (!(pfn%10000))
  89                         printk(".");
  90                 if (!pfn_valid(pfn))
  91                         continue;
  92                 page = pfn_to_page(pfn);
  93                 /*
  94                  * This condition results from rvmalloc() sans vmalloc_32()
  95                  * and architectural memory reservations. This should be
  96                  * corrected eventually when the cases giving rise to this
  97                  * are better understood.
  98                  */
  99                 if (PageReserved(page))
 100                         continue;
 101                 BUG_ON(PageNosave(page));
 102                 if (PageNosaveFree(page))
 103                         continue;
 104                 save = kmalloc(sizeof(struct highmem_page), GFP_ATOMIC);
 105                 if (!save)
 106                         return -ENOMEM;
 107                 save->next = highmem_copy;
 108                 save->page = page;
 109                 save->data = (void *) get_zeroed_page(GFP_ATOMIC);
 110                 if (!save->data) {
 111                         kfree(save);
 112                         return -ENOMEM;
 113                 }
 114                 kaddr = kmap_atomic(page, KM_USER0);
 115                 memcpy(save->data, kaddr, PAGE_SIZE);
 116                 kunmap_atomic(kaddr, KM_USER0);
 117                 highmem_copy = save;
 118         }
 119         return 0;
 120 }
 121
 122 int save_highmem(void)
 123 {
 124         struct zone *zone;
 125         int res = 0;
 126
 127         pr_debug("swsusp: Saving Highmem");
 128         drain_local_pages();
 129         for_each_zone (zone) {
 130                 if (is_highmem(zone))
 131                         res = save_highmem_zone(zone);
 132                 if (res)
 133                         return res;
 134         }
 135         printk("\n");
 136         return 0;
 137 }
 138
 139 int restore_highmem(void)
 140 {
 141         printk("swsusp: Restoring Highmem\n");
 142         while (highmem_copy) {
 143                 struct highmem_page *save = highmem_copy;
 144                 void *kaddr;
 145                 highmem_copy = save->next;
 146
 147                 kaddr = kmap_atomic(save->page, KM_USER0);
 148                 memcpy(kaddr, save->data, PAGE_SIZE);
 149                 kunmap_atomic(kaddr, KM_USER0);
 150                 free_page((long) save->data);
 151                 kfree(save);
 152         }
 153         return 0;
 154 }
 155 #else
 156 static inline unsigned int count_highmem_pages(void) {return 0;}
 157 static inline int save_highmem(void) {return 0;}
 158 static inline int restore_highmem(void) {return 0;}
 159 #endif
 160
 161 /**
 162  *      @safe_needed - on resume, for storing the PBE list and the image,
 163  *      we can only use memory pages that do not conflict with the pages
 164  *      used before suspend.
 165  *
 166  *      The unsafe pages are marked with the PG_nosave_free flag
 167  *      and we count them using unsafe_pages
 168  */
 169
 170 #define PG_ANY          0
 171 #define PG_SAFE         1
 172 #define PG_UNSAFE_CLEAR 1
 173 #define PG_UNSAFE_KEEP  0
 174
 175 static unsigned int unsafe_pages;
 176
 177 static void *alloc_image_page(gfp_t gfp_mask, int safe_needed)
 178 {
 179         void *res;
 180
 181         res = (void *)get_zeroed_page(gfp_mask);
 182         if (safe_needed)
 183                 while (res && PageNosaveFree(virt_to_page(res))) {
 184                         /* The page is unsafe, mark it for swsusp_free() */
 185                         SetPageNosave(virt_to_page(res));
 186                         unsafe_pages++;
 187                         res = (void *)get_zeroed_page(gfp_mask);
 188                 }
 189         if (res) {
 190                 SetPageNosave(virt_to_page(res));
 191                 SetPageNosaveFree(virt_to_page(res));
 192         }
 193         return res;
 194 }
 195
 196 unsigned long get_safe_page(gfp_t gfp_mask)
 197 {
 198         return (unsigned long)alloc_image_page(gfp_mask, PG_SAFE);
 199 }
 200
 201 /**
 202  *      free_image_page - free page represented by @addr, allocated with
 203  *      alloc_image_page (page flags set by it must be cleared)
 204  */
 205
 206 static inline void free_image_page(void *addr, int clear_nosave_free)
 207 {
 208         ClearPageNosave(virt_to_page(addr));
 209         if (clear_nosave_free)
 210                 ClearPageNosaveFree(virt_to_page(addr));
 211         free_page((unsigned long)addr);
 212 }
 213
 214 /**
 215  *      pfn_is_nosave - check if given pfn is in the 'nosave' section
 216  */
 217
 218 static inline int pfn_is_nosave(unsigned long pfn)
 219 {
 220         unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
 221         unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
 222         return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
 223 }
 224
 225 /**
 226  *      saveable - Determine whether a page should be cloned or not.
 227  *      @pfn:   The page
 228  *
 229  *      We save a page if it isn't Nosave, and is not in the range of pages
 230  *      statically defined as 'unsaveable', and it
 231  *      isn't a part of a free chunk of pages.
 232  */
 233
 234 static struct page *saveable_page(unsigned long pfn)
 235 {
 236         struct page *page;
 237
 238         if (!pfn_valid(pfn))
 239                 return NULL;
 240
 241         page = pfn_to_page(pfn);
 242
 243         if (PageNosave(page))
 244                 return NULL;
 245         if (PageReserved(page) && pfn_is_nosave(pfn))
 246                 return NULL;
 247         if (PageNosaveFree(page))
 248                 return NULL;
 249
 250         return page;
 251 }
 252
 253 unsigned int count_data_pages(void)
 254 {
 255         struct zone *zone;
 256         unsigned long pfn, max_zone_pfn;
 257         unsigned int n = 0;
 258
 259         for_each_zone (zone) {
 260                 if (is_highmem(zone))
 261                         continue;
 262                 mark_free_pages(zone);
 263                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
 264                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
 265                         n += !!saveable_page(pfn);
 266         }
 267         return n;
 268 }
 269
 270 static inline void copy_data_page(long *dst, long *src)
 271 {
 272         int n;
 273
 274         /* copy_page and memcpy are not usable for copying task structs. */
 275         for (n = PAGE_SIZE / sizeof(long); n; n--)
 276                 *dst++ = *src++;
 277 }
 278
 279 static void copy_data_pages(struct pbe *pblist)
 280 {
 281         struct zone *zone;
 282         unsigned long pfn, max_zone_pfn;
 283         struct pbe *pbe;
 284
 285         pbe = pblist;
 286         for_each_zone (zone) {
 287                 if (is_highmem(zone))
 288                         continue;
 289                 mark_free_pages(zone);
 290                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
 291                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) {
 292                         struct page *page = saveable_page(pfn);
 293
 294                         if (page) {
 295                                 void *ptr = page_address(page);
 296
 297                                 BUG_ON(!pbe);
 298                                 copy_data_page((void *)pbe->address, ptr);
 299                                 pbe->orig_address = (unsigned long)ptr;
 300                                 pbe = pbe->next;
 301                         }
 302                 }
 303         }
 304         BUG_ON(pbe);
 305 }
 306
 307 /**
 308  *      free_pagedir - free pages allocated with alloc_pagedir()
 309  */
 310
 311 static void free_pagedir(struct pbe *pblist, int clear_nosave_free)
 312 {
 313         struct pbe *pbe;
 314
 315         while (pblist) {
 316                 pbe = (pblist + PB_PAGE_SKIP)->next;
 317                 free_image_page(pblist, clear_nosave_free);
 318                 pblist = pbe;
 319         }
 320 }
 321
 322 /**
 323  *      fill_pb_page - Create a list of PBEs on a given memory page
 324  */
 325
 326 static inline void fill_pb_page(struct pbe *pbpage, unsigned int n)
 327 {
 328         struct pbe *p;
 329
 330         p = pbpage;
 331         pbpage += n - 1;
 332         do
 333                 p->next = p + 1;
 334         while (++p < pbpage);
 335 }
 336
 337 /**
 338  *      create_pbe_list - Create a list of PBEs on top of a given chain
 339  *      of memory pages allocated with alloc_pagedir()
 340  *
 341  *      This function assumes that pages allocated by alloc_image_page() will
 342  *      always be zeroed.
 343  */
 344
 345 static inline void create_pbe_list(struct pbe *pblist, unsigned int nr_pages)
 346 {
 347         struct pbe *pbpage;
 348         unsigned int num = PBES_PER_PAGE;
 349
 350         for_each_pb_page (pbpage, pblist) {
 351                 if (num >= nr_pages)
 352                         break;
 353
 354                 fill_pb_page(pbpage, PBES_PER_PAGE);
 355                 num += PBES_PER_PAGE;
 356         }
 357         if (pbpage) {
 358                 num -= PBES_PER_PAGE;
 359                 fill_pb_page(pbpage, nr_pages - num);
 360         }
 361 }
 362
 363 /**
 364  *      alloc_pagedir - Allocate the page directory.
 365  *
 366  *      First, determine exactly how many pages we need and
 367  *      allocate them.
 368  *
 369  *      We arrange the pages in a chain: each page is an array of PBES_PER_PAGE
 370  *      struct pbe elements (pbes) and the last element in the page points
 371  *      to the next page.
 372  *
 373  *      On each page we set up a list of struct_pbe elements.
 374  */
 375
 376 static struct pbe *alloc_pagedir(unsigned int nr_pages, gfp_t gfp_mask,
 377                                  int safe_needed)
 378 {
 379         unsigned int num;
 380         struct pbe *pblist, *pbe;
 381
 382         if (!nr_pages)
 383                 return NULL;
 384
 385         pblist = alloc_image_page(gfp_mask, safe_needed);
 386         pbe = pblist;
 387         for (num = PBES_PER_PAGE; num < nr_pages; num += PBES_PER_PAGE) {
 388                 if (!pbe) {
 389                         free_pagedir(pblist, PG_UNSAFE_CLEAR);
 390                         return NULL;
 391                 }
 392                 pbe += PB_PAGE_SKIP;
 393                 pbe->next = alloc_image_page(gfp_mask, safe_needed);
 394                 pbe = pbe->next;
 395         }
 396         create_pbe_list(pblist, nr_pages);
 397         return pblist;
 398 }
 399
 400 /**
 401  * Free pages we allocated for suspend. Suspend pages are alocated
 402  * before atomic copy, so we need to free them after resume.
 403  */
 404
 405 void swsusp_free(void)
 406 {
 407         struct zone *zone;
 408         unsigned long pfn, max_zone_pfn;
 409
 410         for_each_zone(zone) {
 411                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
 412                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
 413                         if (pfn_valid(pfn)) {
 414                                 struct page *page = pfn_to_page(pfn);
 415
 416                                 if (PageNosave(page) && PageNosaveFree(page)) {
 417                                         ClearPageNosave(page);
 418                                         ClearPageNosaveFree(page);
 419                                         free_page((long) page_address(page));
 420                                 }
 421                         }
 422         }
 423         nr_copy_pages = 0;
 424         nr_meta_pages = 0;
 425         restore_pblist = NULL;
 426         buffer = NULL;
 427 }
 428
 429
 430 /**
 431  *      enough_free_mem - Make sure we enough free memory to snapshot.
 432  *
 433  *      Returns TRUE or FALSE after checking the number of available
 434  *      free pages.
 435  */
 436
 437 static int enough_free_mem(unsigned int nr_pages)
 438 {
 439         struct zone *zone;
 440         unsigned int n = 0;
 441
 442         for_each_zone (zone)
 443                 if (!is_highmem(zone))
 444                         n += zone->free_pages;
 445         pr_debug("swsusp: available memory: %u pages\n", n);
 446         return n > (nr_pages + PAGES_FOR_IO +
 447                 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
 448 }
 449
 450 static int alloc_data_pages(struct pbe *pblist, gfp_t gfp_mask, int safe_needed)
 451 {
 452         struct pbe *p;
 453
 454         for_each_pbe (p, pblist) {
 455                 p->address = (unsigned long)alloc_image_page(gfp_mask, safe_needed);
 456                 if (!p->address)
 457                         return -ENOMEM;
 458         }
 459         return 0;
 460 }
 461
 462 static struct pbe *swsusp_alloc(unsigned int nr_pages)
 463 {
 464         struct pbe *pblist;
 465
 466         pblist = alloc_pagedir(nr_pages, GFP_ATOMIC | __GFP_COLD, PG_ANY);
 467         if (!pblist) {
 468                 printk(KERN_ERR "suspend: Allocating pagedir failed.\n");
 469                 return NULL;
 470         }
 471
 472         if (alloc_data_pages(pblist, GFP_ATOMIC | __GFP_COLD, PG_ANY)) {
 473                 printk(KERN_ERR "suspend: Allocating image pages failed.\n");
 474                 swsusp_free();
 475                 return NULL;
 476         }
 477
 478         return pblist;
 479 }
 480
 481 asmlinkage int swsusp_save(void)
 482 {
 483         unsigned int nr_pages;
 484
 485         pr_debug("swsusp: critical section: \n");
 486
 487         drain_local_pages();
 488         nr_pages = count_data_pages();
 489         printk("swsusp: Need to copy %u pages\n", nr_pages);
 490
 491         pr_debug("swsusp: pages needed: %u + %lu + %u, free: %u\n",
 492                  nr_pages,
 493                  (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE,
 494                  PAGES_FOR_IO, nr_free_pages());
 495
 496         if (!enough_free_mem(nr_pages)) {
 497                 printk(KERN_ERR "swsusp: Not enough free memory\n");
 498                 return -ENOMEM;
 499         }
 500
 501         restore_pblist = swsusp_alloc(nr_pages);
 502         if (!restore_pblist)
 503                 return -ENOMEM;
 504
 505         /* During allocating of suspend pagedir, new cold pages may appear.
 506          * Kill them.
 507          */
 508         drain_local_pages();
 509         copy_data_pages(restore_pblist);
 510
 511         /*
 512          * End of critical section. From now on, we can write to memory,
 513          * but we should not touch disk. This specially means we must _not_
 514          * touch swap space! Except we must write out our image of course.
 515          */
 516
 517         nr_copy_pages = nr_pages;
 518         nr_meta_pages = (nr_pages * sizeof(long) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 519
 520         printk("swsusp: critical section/: done (%d pages copied)\n", nr_pages);
 521         return 0;
 522 }
 523
 524 static void init_header(struct swsusp_info *info)
 525 {
 526         memset(info, 0, sizeof(struct swsusp_info));
 527         info->version_code = LINUX_VERSION_CODE;
 528         info->num_physpages = num_physpages;
 529         memcpy(&info->uts, &system_utsname, sizeof(system_utsname));
 530         info->cpus = num_online_cpus();
 531         info->image_pages = nr_copy_pages;
 532         info->pages = nr_copy_pages + nr_meta_pages + 1;
 533         info->size = info->pages;
 534         info->size <<= PAGE_SHIFT;
 535 }
 536
 537 /**
 538  *      pack_orig_addresses - the .orig_address fields of the PBEs from the
 539  *      list starting at @pbe are stored in the array @buf[] (1 page)
 540  */
 541
 542 static inline struct pbe *pack_orig_addresses(unsigned long *buf, struct pbe *pbe)
 543 {
 544         int j;
 545
 546         for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
 547                 buf[j] = pbe->orig_address;
 548                 pbe = pbe->next;
 549         }
 550         if (!pbe)
 551                 for (; j < PAGE_SIZE / sizeof(long); j++)
 552                         buf[j] = 0;
 553         return pbe;
 554 }
 555
 556 /**
 557  *      snapshot_read_next - used for reading the system memory snapshot.
 558  *
 559  *      On the first call to it @handle should point to a zeroed
 560  *      snapshot_handle structure.  The structure gets updated and a pointer
 561  *      to it should be passed to this function every next time.
 562  *
 563  *      The @count parameter should contain the number of bytes the caller
 564  *      wants to read from the snapshot.  It must not be zero.
 565  *
 566  *      On success the function returns a positive number.  Then, the caller
 567  *      is allowed to read up to the returned number of bytes from the memory
 568  *      location computed by the data_of() macro.  The number returned
 569  *      may be smaller than @count, but this only happens if the read would
 570  *      cross a page boundary otherwise.
 571  *
 572  *      The function returns 0 to indicate the end of data stream condition,
 573  *      and a negative number is returned on error.  In such cases the
 574  *      structure pointed to by @handle is not updated and should not be used
 575  *      any more.
 576  */
 577
 578 int snapshot_read_next(struct snapshot_handle *handle, size_t count)
 579 {
 580         if (handle->cur > nr_meta_pages + nr_copy_pages)
 581                 return 0;
 582         if (!buffer) {
 583                 /* This makes the buffer be freed by swsusp_free() */
 584                 buffer = alloc_image_page(GFP_ATOMIC, PG_ANY);
 585                 if (!buffer)
 586                         return -ENOMEM;
 587         }
 588         if (!handle->offset) {
 589                 init_header((struct swsusp_info *)buffer);
 590                 handle->buffer = buffer;
 591                 handle->pbe = restore_pblist;
 592         }
 593         if (handle->prev < handle->cur) {
 594                 if (handle->cur <= nr_meta_pages) {
 595                         handle->pbe = pack_orig_addresses(buffer, handle->pbe);
 596                         if (!handle->pbe)
 597                                 handle->pbe = restore_pblist;
 598                 } else {
 599                         handle->buffer = (void *)handle->pbe->address;
 600                         handle->pbe = handle->pbe->next;
 601                 }
 602                 handle->prev = handle->cur;
 603         }
 604         handle->buf_offset = handle->cur_offset;
 605         if (handle->cur_offset + count >= PAGE_SIZE) {
 606                 count = PAGE_SIZE - handle->cur_offset;
 607                 handle->cur_offset = 0;
 608                 handle->cur++;
 609         } else {
 610                 handle->cur_offset += count;
 611         }
 612         handle->offset += count;
 613         return count;
 614 }
 615
 616 /**
 617  *      mark_unsafe_pages - mark the pages that cannot be used for storing
 618  *      the image during resume, because they conflict with the pages that
 619  *      had been used before suspend
 620  */
 621
 622 static int mark_unsafe_pages(struct pbe *pblist)
 623 {
 624         struct zone *zone;
 625         unsigned long pfn, max_zone_pfn;
 626         struct pbe *p;
 627
 628         if (!pblist) /* a sanity check */
 629                 return -EINVAL;
 630
 631         /* Clear page flags */
 632         for_each_zone (zone) {
 633                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
 634                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
 635                         if (pfn_valid(pfn))
 636                                 ClearPageNosaveFree(pfn_to_page(pfn));
 637         }
 638
 639         /* Mark orig addresses */
 640         for_each_pbe (p, pblist) {
 641                 if (virt_addr_valid(p->orig_address))
 642                         SetPageNosaveFree(virt_to_page(p->orig_address));
 643                 else
 644                         return -EFAULT;
 645         }
 646
 647         unsafe_pages = 0;
 648
 649         return 0;
 650 }
 651
 652 static void copy_page_backup_list(struct pbe *dst, struct pbe *src)
 653 {
 654         /* We assume both lists contain the same number of elements */
 655         while (src) {
 656                 dst->orig_address = src->orig_address;
 657                 dst = dst->next;
 658                 src = src->next;
 659         }
 660 }
 661
 662 static int check_header(struct swsusp_info *info)
 663 {
 664         char *reason = NULL;
 665
 666         if (info->version_code != LINUX_VERSION_CODE)
 667                 reason = "kernel version";
 668         if (info->num_physpages != num_physpages)
 669                 reason = "memory size";
 670         if (strcmp(info->uts.sysname,system_utsname.sysname))
 671                 reason = "system type";
 672         if (strcmp(info->uts.release,system_utsname.release))
 673                 reason = "kernel release";
 674         if (strcmp(info->uts.version,system_utsname.version))
 675                 reason = "version";
 676         if (strcmp(info->uts.machine,system_utsname.machine))
 677                 reason = "machine";
 678         if (reason) {
 679                 printk(KERN_ERR "swsusp: Resume mismatch: %s\n", reason);
 680                 return -EPERM;
 681         }
 682         return 0;
 683 }
 684
 685 /**
 686  *      load header - check the image header and copy data from it
 687  */
 688
 689 static int load_header(struct snapshot_handle *handle,
 690                               struct swsusp_info *info)
 691 {
 692         int error;
 693         struct pbe *pblist;
 694
 695         error = check_header(info);
 696         if (!error) {
 697                 pblist = alloc_pagedir(info->image_pages, GFP_ATOMIC, PG_ANY);
 698                 if (!pblist)
 699                         return -ENOMEM;
 700                 restore_pblist = pblist;
 701                 handle->pbe = pblist;
 702                 nr_copy_pages = info->image_pages;
 703                 nr_meta_pages = info->pages - info->image_pages - 1;
 704         }
 705         return error;
 706 }
 707
 708 /**
 709  *      unpack_orig_addresses - copy the elements of @buf[] (1 page) to
 710  *      the PBEs in the list starting at @pbe
 711  */
 712
 713 static inline struct pbe *unpack_orig_addresses(unsigned long *buf,
 714                                                 struct pbe *pbe)
 715 {
 716         int j;
 717
 718         for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
 719                 pbe->orig_address = buf[j];
 720                 pbe = pbe->next;
 721         }
 722         return pbe;
 723 }
 724
 725 /**
 726  *      prepare_image - use metadata contained in the PBE list
 727  *      pointed to by restore_pblist to mark the pages that will
 728  *      be overwritten in the process of restoring the system
 729  *      memory state from the image ("unsafe" pages) and allocate
 730  *      memory for the image
 731  *
 732  *      The idea is to allocate the PBE list first and then
 733  *      allocate as many pages as it's needed for the image data,
 734  *      but not to assign these pages to the PBEs initially.
 735  *      Instead, we just mark them as allocated and create a list
 736  *      of "safe" which will be used later
 737  */
 738
 739 struct safe_page {
 740         struct safe_page *next;
 741         char padding[PAGE_SIZE - sizeof(void *)];
 742 };
 743
 744 static struct safe_page *safe_pages;
 745
 746 static int prepare_image(struct snapshot_handle *handle)
 747 {
 748         int error = 0;
 749         unsigned int nr_pages = nr_copy_pages;
 750         struct pbe *p, *pblist = NULL;
 751
 752         p = restore_pblist;
 753         error = mark_unsafe_pages(p);
 754         if (!error) {
 755                 pblist = alloc_pagedir(nr_pages, GFP_ATOMIC, PG_SAFE);
 756                 if (pblist)
 757                         copy_page_backup_list(pblist, p);
 758                 free_pagedir(p, PG_UNSAFE_KEEP);
 759                 if (!pblist)
 760                         error = -ENOMEM;
 761         }
 762         safe_pages = NULL;
 763         if (!error && nr_pages > unsafe_pages) {
 764                 nr_pages -= unsafe_pages;
 765                 while (nr_pages--) {
 766                         struct safe_page *ptr;
 767
 768                         ptr = (struct safe_page *)get_zeroed_page(GFP_ATOMIC);
 769                         if (!ptr) {
 770                                 error = -ENOMEM;
 771                                 break;
 772                         }
 773                         if (!PageNosaveFree(virt_to_page(ptr))) {
 774                                 /* The page is "safe", add it to the list */
 775                                 ptr->next = safe_pages;
 776                                 safe_pages = ptr;
 777                         }
 778                         /* Mark the page as allocated */
 779                         SetPageNosave(virt_to_page(ptr));
 780                         SetPageNosaveFree(virt_to_page(ptr));
 781                 }
 782         }
 783         if (!error) {
 784                 restore_pblist = pblist;
 785         } else {
 786                 handle->pbe = NULL;
 787                 swsusp_free();
 788         }
 789         return error;
 790 }
 791
 792 static void *get_buffer(struct snapshot_handle *handle)
 793 {
 794         struct pbe *pbe = handle->pbe, *last = handle->last_pbe;
 795         struct page *page = virt_to_page(pbe->orig_address);
 796
 797         if (PageNosave(page) && PageNosaveFree(page)) {
 798                 /*
 799                  * We have allocated the "original" page frame and we can
 800                  * use it directly to store the read page
 801                  */
 802                 pbe->address = 0;
 803                 if (last && last->next)
 804                         last->next = NULL;
 805                 return (void *)pbe->orig_address;
 806         }
 807         /*
 808          * The "original" page frame has not been allocated and we have to
 809          * use a "safe" page frame to store the read page
 810          */
 811         pbe->address = (unsigned long)safe_pages;
 812         safe_pages = safe_pages->next;
 813         if (last)
 814                 last->next = pbe;
 815         handle->last_pbe = pbe;
 816         return (void *)pbe->address;
 817 }
 818
 819 /**
 820  *      snapshot_write_next - used for writing the system memory snapshot.
 821  *
 822  *      On the first call to it @handle should point to a zeroed
 823  *      snapshot_handle structure.  The structure gets updated and a pointer
 824  *      to it should be passed to this function every next time.
 825  *
 826  *      The @count parameter should contain the number of bytes the caller
 827  *      wants to write to the image.  It must not be zero.
 828  *
 829  *      On success the function returns a positive number.  Then, the caller
 830  *      is allowed to write up to the returned number of bytes to the memory
 831  *      location computed by the data_of() macro.  The number returned
 832  *      may be smaller than @count, but this only happens if the write would
 833  *      cross a page boundary otherwise.
 834  *
 835  *      The function returns 0 to indicate the "end of file" condition,
 836  *      and a negative number is returned on error.  In such cases the
 837  *      structure pointed to by @handle is not updated and should not be used
 838  *      any more.
 839  */
 840
 841 int snapshot_write_next(struct snapshot_handle *handle, size_t count)
 842 {
 843         int error = 0;
 844
 845         if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages)
 846                 return 0;
 847         if (!buffer) {
 848                 /* This makes the buffer be freed by swsusp_free() */
 849                 buffer = alloc_image_page(GFP_ATOMIC, PG_ANY);
 850                 if (!buffer)
 851                         return -ENOMEM;
 852         }
 853         if (!handle->offset)
 854                 handle->buffer = buffer;
 855         handle->sync_read = 1;
 856         if (handle->prev < handle->cur) {
 857                 if (!handle->prev) {
 858                         error = load_header(handle,
 859                                         (struct swsusp_info *)buffer);
 860                         if (error)
 861                                 return error;
 862                 } else if (handle->prev <= nr_meta_pages) {
 863                         handle->pbe = unpack_orig_addresses(buffer,
 864                                                         handle->pbe);
 865                         if (!handle->pbe) {
 866                                 error = prepare_image(handle);
 867                                 if (error)
 868                                         return error;
 869                                 handle->pbe = restore_pblist;
 870                                 handle->last_pbe = NULL;
 871                                 handle->buffer = get_buffer(handle);
 872                                 handle->sync_read = 0;
 873                         }
 874                 } else {
 875                         handle->pbe = handle->pbe->next;
 876                         handle->buffer = get_buffer(handle);
 877                         handle->sync_read = 0;
 878                 }
 879                 handle->prev = handle->cur;
 880         }
 881         handle->buf_offset = handle->cur_offset;
 882         if (handle->cur_offset + count >= PAGE_SIZE) {
 883                 count = PAGE_SIZE - handle->cur_offset;
 884                 handle->cur_offset = 0;
 885                 handle->cur++;
 886         } else {
 887                 handle->cur_offset += count;
 888         }
 889         handle->offset += count;
 890         return count;
 891 }
 892
 893 int snapshot_image_loaded(struct snapshot_handle *handle)
 894 {
 895         return !(!handle->pbe || handle->pbe->next || !nr_copy_pages ||
 896                 handle->cur <= nr_meta_pages + nr_copy_pages);
 897 }