Pileus Git - ~andy/linux/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 struct arch_saveable_page {
  43         unsigned long start;
  44         unsigned long end;
  45         char *data;
  46         struct arch_saveable_page *next;
  47 };
  48 static struct arch_saveable_page *arch_pages;
  49
  50 int swsusp_add_arch_pages(unsigned long start, unsigned long end)
  51 {
  52         struct arch_saveable_page *tmp;
  53
  54         while (start < end) {
  55                 tmp = kzalloc(sizeof(struct arch_saveable_page), GFP_KERNEL);
  56                 if (!tmp)
  57                         return -ENOMEM;
  58                 tmp->start = start;
  59                 tmp->end = ((start >> PAGE_SHIFT) + 1) << PAGE_SHIFT;
  60                 if (tmp->end > end)
  61                         tmp->end = end;
  62                 tmp->next = arch_pages;
  63                 start = tmp->end;
  64                 arch_pages = tmp;
  65         }
  66         return 0;
  67 }
  68
  69 static unsigned int count_arch_pages(void)
  70 {
  71         unsigned int count = 0;
  72         struct arch_saveable_page *tmp = arch_pages;
  73         while (tmp) {
  74                 count++;
  75                 tmp = tmp->next;
  76         }
  77         return count;
  78 }
  79
  80 static int save_arch_mem(void)
  81 {
  82         char *kaddr;
  83         struct arch_saveable_page *tmp = arch_pages;
  84         int offset;
  85
  86         pr_debug("swsusp: Saving arch specific memory");
  87         while (tmp) {
  88                 tmp->data = (char *)__get_free_page(GFP_ATOMIC);
  89                 if (!tmp->data)
  90                         return -ENOMEM;
  91                 offset = tmp->start - (tmp->start & PAGE_MASK);
  92                 /* arch pages might haven't a 'struct page' */
  93                 kaddr = kmap_atomic_pfn(tmp->start >> PAGE_SHIFT, KM_USER0);
  94                 memcpy(tmp->data + offset, kaddr + offset,
  95                         tmp->end - tmp->start);
  96                 kunmap_atomic(kaddr, KM_USER0);
  97
  98                 tmp = tmp->next;
  99         }
 100         return 0;
 101 }
 102
 103 static int restore_arch_mem(void)
 104 {
 105         char *kaddr;
 106         struct arch_saveable_page *tmp = arch_pages;
 107         int offset;
 108
 109         while (tmp) {
 110                 if (!tmp->data)
 111                         continue;
 112                 offset = tmp->start - (tmp->start & PAGE_MASK);
 113                 kaddr = kmap_atomic_pfn(tmp->start >> PAGE_SHIFT, KM_USER0);
 114                 memcpy(kaddr + offset, tmp->data + offset,
 115                         tmp->end - tmp->start);
 116                 kunmap_atomic(kaddr, KM_USER0);
 117                 free_page((long)tmp->data);
 118                 tmp->data = NULL;
 119                 tmp = tmp->next;
 120         }
 121         return 0;
 122 }
 123
 124 #ifdef CONFIG_HIGHMEM
 125 static unsigned int count_highmem_pages(void)
 126 {
 127         struct zone *zone;
 128         unsigned long zone_pfn;
 129         unsigned int n = 0;
 130
 131         for_each_zone (zone)
 132                 if (is_highmem(zone)) {
 133                         mark_free_pages(zone);
 134                         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; zone_pfn++) {
 135                                 struct page *page;
 136                                 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
 137                                 if (!pfn_valid(pfn))
 138                                         continue;
 139                                 page = pfn_to_page(pfn);
 140                                 if (PageReserved(page))
 141                                         continue;
 142                                 if (PageNosaveFree(page))
 143                                         continue;
 144                                 n++;
 145                         }
 146                 }
 147         return n;
 148 }
 149
 150 struct highmem_page {
 151         char *data;
 152         struct page *page;
 153         struct highmem_page *next;
 154 };
 155
 156 static struct highmem_page *highmem_copy;
 157
 158 static int save_highmem_zone(struct zone *zone)
 159 {
 160         unsigned long zone_pfn;
 161         mark_free_pages(zone);
 162         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
 163                 struct page *page;
 164                 struct highmem_page *save;
 165                 void *kaddr;
 166                 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
 167
 168                 if (!(pfn%10000))
 169                         printk(".");
 170                 if (!pfn_valid(pfn))
 171                         continue;
 172                 page = pfn_to_page(pfn);
 173                 /*
 174                  * This condition results from rvmalloc() sans vmalloc_32()
 175                  * and architectural memory reservations. This should be
 176                  * corrected eventually when the cases giving rise to this
 177                  * are better understood.
 178                  */
 179                 if (PageReserved(page))
 180                         continue;
 181                 BUG_ON(PageNosave(page));
 182                 if (PageNosaveFree(page))
 183                         continue;
 184                 save = kmalloc(sizeof(struct highmem_page), GFP_ATOMIC);
 185                 if (!save)
 186                         return -ENOMEM;
 187                 save->next = highmem_copy;
 188                 save->page = page;
 189                 save->data = (void *) get_zeroed_page(GFP_ATOMIC);
 190                 if (!save->data) {
 191                         kfree(save);
 192                         return -ENOMEM;
 193                 }
 194                 kaddr = kmap_atomic(page, KM_USER0);
 195                 memcpy(save->data, kaddr, PAGE_SIZE);
 196                 kunmap_atomic(kaddr, KM_USER0);
 197                 highmem_copy = save;
 198         }
 199         return 0;
 200 }
 201
 202 static int save_highmem(void)
 203 {
 204         struct zone *zone;
 205         int res = 0;
 206
 207         pr_debug("swsusp: Saving Highmem");
 208         drain_local_pages();
 209         for_each_zone (zone) {
 210                 if (is_highmem(zone))
 211                         res = save_highmem_zone(zone);
 212                 if (res)
 213                         return res;
 214         }
 215         printk("\n");
 216         return 0;
 217 }
 218
 219 static int restore_highmem(void)
 220 {
 221         printk("swsusp: Restoring Highmem\n");
 222         while (highmem_copy) {
 223                 struct highmem_page *save = highmem_copy;
 224                 void *kaddr;
 225                 highmem_copy = save->next;
 226
 227                 kaddr = kmap_atomic(save->page, KM_USER0);
 228                 memcpy(kaddr, save->data, PAGE_SIZE);
 229                 kunmap_atomic(kaddr, KM_USER0);
 230                 free_page((long) save->data);
 231                 kfree(save);
 232         }
 233         return 0;
 234 }
 235 #else
 236 static inline unsigned int count_highmem_pages(void) {return 0;}
 237 static inline int save_highmem(void) {return 0;}
 238 static inline int restore_highmem(void) {return 0;}
 239 #endif
 240
 241 unsigned int count_special_pages(void)
 242 {
 243         return count_arch_pages() + count_highmem_pages();
 244 }
 245
 246 int save_special_mem(void)
 247 {
 248         int ret;
 249         ret = save_arch_mem();
 250         if (!ret)
 251                 ret = save_highmem();
 252         return ret;
 253 }
 254
 255 int restore_special_mem(void)
 256 {
 257         int ret;
 258         ret = restore_arch_mem();
 259         if (!ret)
 260                 ret = restore_highmem();
 261         return ret;
 262 }
 263
 264 static int pfn_is_nosave(unsigned long pfn)
 265 {
 266         unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
 267         unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
 268         return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
 269 }
 270
 271 /**
 272  *      saveable - Determine whether a page should be cloned or not.
 273  *      @pfn:   The page
 274  *
 275  *      We save a page if it's Reserved, and not in the range of pages
 276  *      statically defined as 'unsaveable', or if it isn't reserved, and
 277  *      isn't part of a free chunk of pages.
 278  */
 279
 280 static int saveable(struct zone *zone, unsigned long *zone_pfn)
 281 {
 282         unsigned long pfn = *zone_pfn + zone->zone_start_pfn;
 283         struct page *page;
 284
 285         if (!pfn_valid(pfn))
 286                 return 0;
 287
 288         page = pfn_to_page(pfn);
 289         if (PageNosave(page))
 290                 return 0;
 291         if (PageReserved(page) && pfn_is_nosave(pfn))
 292                 return 0;
 293         if (PageNosaveFree(page))
 294                 return 0;
 295
 296         return 1;
 297 }
 298
 299 unsigned int count_data_pages(void)
 300 {
 301         struct zone *zone;
 302         unsigned long zone_pfn;
 303         unsigned int n = 0;
 304
 305         for_each_zone (zone) {
 306                 if (is_highmem(zone))
 307                         continue;
 308                 mark_free_pages(zone);
 309                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 310                         n += saveable(zone, &zone_pfn);
 311         }
 312         return n;
 313 }
 314
 315 static void copy_data_pages(struct pbe *pblist)
 316 {
 317         struct zone *zone;
 318         unsigned long zone_pfn;
 319         struct pbe *pbe, *p;
 320
 321         pbe = pblist;
 322         for_each_zone (zone) {
 323                 if (is_highmem(zone))
 324                         continue;
 325                 mark_free_pages(zone);
 326                 /* This is necessary for swsusp_free() */
 327                 for_each_pb_page (p, pblist)
 328                         SetPageNosaveFree(virt_to_page(p));
 329                 for_each_pbe (p, pblist)
 330                         SetPageNosaveFree(virt_to_page(p->address));
 331                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
 332                         if (saveable(zone, &zone_pfn)) {
 333                                 struct page *page;
 334                                 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
 335                                 BUG_ON(!pbe);
 336                                 pbe->orig_address = (unsigned long)page_address(page);
 337                                 /* copy_page is not usable for copying task structs. */
 338                                 memcpy((void *)pbe->address, (void *)pbe->orig_address, PAGE_SIZE);
 339                                 pbe = pbe->next;
 340                         }
 341                 }
 342         }
 343         BUG_ON(pbe);
 344 }
 345
 346
 347 /**
 348  *      free_pagedir - free pages allocated with alloc_pagedir()
 349  */
 350
 351 static void free_pagedir(struct pbe *pblist, int clear_nosave_free)
 352 {
 353         struct pbe *pbe;
 354
 355         while (pblist) {
 356                 pbe = (pblist + PB_PAGE_SKIP)->next;
 357                 ClearPageNosave(virt_to_page(pblist));
 358                 if (clear_nosave_free)
 359                         ClearPageNosaveFree(virt_to_page(pblist));
 360                 free_page((unsigned long)pblist);
 361                 pblist = pbe;
 362         }
 363 }
 364
 365 /**
 366  *      fill_pb_page - Create a list of PBEs on a given memory page
 367  */
 368
 369 static inline void fill_pb_page(struct pbe *pbpage)
 370 {
 371         struct pbe *p;
 372
 373         p = pbpage;
 374         pbpage += PB_PAGE_SKIP;
 375         do
 376                 p->next = p + 1;
 377         while (++p < pbpage);
 378 }
 379
 380 /**
 381  *      create_pbe_list - Create a list of PBEs on top of a given chain
 382  *      of memory pages allocated with alloc_pagedir()
 383  */
 384
 385 static inline void create_pbe_list(struct pbe *pblist, unsigned int nr_pages)
 386 {
 387         struct pbe *pbpage, *p;
 388         unsigned int num = PBES_PER_PAGE;
 389
 390         for_each_pb_page (pbpage, pblist) {
 391                 if (num >= nr_pages)
 392                         break;
 393
 394                 fill_pb_page(pbpage);
 395                 num += PBES_PER_PAGE;
 396         }
 397         if (pbpage) {
 398                 for (num -= PBES_PER_PAGE - 1, p = pbpage; num < nr_pages; p++, num++)
 399                         p->next = p + 1;
 400                 p->next = NULL;
 401         }
 402 }
 403
 404 static unsigned int unsafe_pages;
 405
 406 /**
 407  *      @safe_needed - on resume, for storing the PBE list and the image,
 408  *      we can only use memory pages that do not conflict with the pages
 409  *      used before suspend.
 410  *
 411  *      The unsafe pages are marked with the PG_nosave_free flag
 412  *      and we count them using unsafe_pages
 413  */
 414
 415 static inline void *alloc_image_page(gfp_t gfp_mask, int safe_needed)
 416 {
 417         void *res;
 418
 419         res = (void *)get_zeroed_page(gfp_mask);
 420         if (safe_needed)
 421                 while (res && PageNosaveFree(virt_to_page(res))) {
 422                         /* The page is unsafe, mark it for swsusp_free() */
 423                         SetPageNosave(virt_to_page(res));
 424                         unsafe_pages++;
 425                         res = (void *)get_zeroed_page(gfp_mask);
 426                 }
 427         if (res) {
 428                 SetPageNosave(virt_to_page(res));
 429                 SetPageNosaveFree(virt_to_page(res));
 430         }
 431         return res;
 432 }
 433
 434 unsigned long get_safe_page(gfp_t gfp_mask)
 435 {
 436         return (unsigned long)alloc_image_page(gfp_mask, 1);
 437 }
 438
 439 /**
 440  *      alloc_pagedir - Allocate the page directory.
 441  *
 442  *      First, determine exactly how many pages we need and
 443  *      allocate them.
 444  *
 445  *      We arrange the pages in a chain: each page is an array of PBES_PER_PAGE
 446  *      struct pbe elements (pbes) and the last element in the page points
 447  *      to the next page.
 448  *
 449  *      On each page we set up a list of struct_pbe elements.
 450  */
 451
 452 static struct pbe *alloc_pagedir(unsigned int nr_pages, gfp_t gfp_mask,
 453                                  int safe_needed)
 454 {
 455         unsigned int num;
 456         struct pbe *pblist, *pbe;
 457
 458         if (!nr_pages)
 459                 return NULL;
 460
 461         pblist = alloc_image_page(gfp_mask, safe_needed);
 462         /* FIXME: rewrite this ugly loop */
 463         for (pbe = pblist, num = PBES_PER_PAGE; pbe && num < nr_pages;
 464                         pbe = pbe->next, num += PBES_PER_PAGE) {
 465                 pbe += PB_PAGE_SKIP;
 466                 pbe->next = alloc_image_page(gfp_mask, safe_needed);
 467         }
 468         if (!pbe) { /* get_zeroed_page() failed */
 469                 free_pagedir(pblist, 1);
 470                 pblist = NULL;
 471         } else
 472                 create_pbe_list(pblist, nr_pages);
 473         return pblist;
 474 }
 475
 476 /**
 477  * Free pages we allocated for suspend. Suspend pages are alocated
 478  * before atomic copy, so we need to free them after resume.
 479  */
 480
 481 void swsusp_free(void)
 482 {
 483         struct zone *zone;
 484         unsigned long zone_pfn;
 485
 486         for_each_zone(zone) {
 487                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 488                         if (pfn_valid(zone_pfn + zone->zone_start_pfn)) {
 489                                 struct page *page;
 490                                 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
 491                                 if (PageNosave(page) && PageNosaveFree(page)) {
 492                                         ClearPageNosave(page);
 493                                         ClearPageNosaveFree(page);
 494                                         free_page((long) page_address(page));
 495                                 }
 496                         }
 497         }
 498         nr_copy_pages = 0;
 499         nr_meta_pages = 0;
 500         pagedir_nosave = NULL;
 501         buffer = NULL;
 502 }
 503
 504
 505 /**
 506  *      enough_free_mem - Make sure we enough free memory to snapshot.
 507  *
 508  *      Returns TRUE or FALSE after checking the number of available
 509  *      free pages.
 510  */
 511
 512 static int enough_free_mem(unsigned int nr_pages)
 513 {
 514         struct zone *zone;
 515         unsigned int n = 0;
 516
 517         for_each_zone (zone)
 518                 if (!is_highmem(zone))
 519                         n += zone->free_pages;
 520         pr_debug("swsusp: available memory: %u pages\n", n);
 521         return n > (nr_pages + PAGES_FOR_IO +
 522                 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
 523 }
 524
 525 static int alloc_data_pages(struct pbe *pblist, gfp_t gfp_mask, int safe_needed)
 526 {
 527         struct pbe *p;
 528
 529         for_each_pbe (p, pblist) {
 530                 p->address = (unsigned long)alloc_image_page(gfp_mask, safe_needed);
 531                 if (!p->address)
 532                         return -ENOMEM;
 533         }
 534         return 0;
 535 }
 536
 537 static struct pbe *swsusp_alloc(unsigned int nr_pages)
 538 {
 539         struct pbe *pblist;
 540
 541         if (!(pblist = alloc_pagedir(nr_pages, GFP_ATOMIC | __GFP_COLD, 0))) {
 542                 printk(KERN_ERR "suspend: Allocating pagedir failed.\n");
 543                 return NULL;
 544         }
 545
 546         if (alloc_data_pages(pblist, GFP_ATOMIC | __GFP_COLD, 0)) {
 547                 printk(KERN_ERR "suspend: Allocating image pages failed.\n");
 548                 swsusp_free();
 549                 return NULL;
 550         }
 551
 552         return pblist;
 553 }
 554
 555 asmlinkage int swsusp_save(void)
 556 {
 557         unsigned int nr_pages;
 558
 559         pr_debug("swsusp: critical section: \n");
 560
 561         drain_local_pages();
 562         nr_pages = count_data_pages();
 563         printk("swsusp: Need to copy %u pages\n", nr_pages);
 564
 565         pr_debug("swsusp: pages needed: %u + %lu + %u, free: %u\n",
 566                  nr_pages,
 567                  (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE,
 568                  PAGES_FOR_IO, nr_free_pages());
 569
 570         if (!enough_free_mem(nr_pages)) {
 571                 printk(KERN_ERR "swsusp: Not enough free memory\n");
 572                 return -ENOMEM;
 573         }
 574
 575         pagedir_nosave = swsusp_alloc(nr_pages);
 576         if (!pagedir_nosave)
 577                 return -ENOMEM;
 578
 579         /* During allocating of suspend pagedir, new cold pages may appear.
 580          * Kill them.
 581          */
 582         drain_local_pages();
 583         copy_data_pages(pagedir_nosave);
 584
 585         /*
 586          * End of critical section. From now on, we can write to memory,
 587          * but we should not touch disk. This specially means we must _not_
 588          * touch swap space! Except we must write out our image of course.
 589          */
 590
 591         nr_copy_pages = nr_pages;
 592         nr_meta_pages = (nr_pages * sizeof(long) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 593
 594         printk("swsusp: critical section/: done (%d pages copied)\n", nr_pages);
 595         return 0;
 596 }
 597
 598 static void init_header(struct swsusp_info *info)
 599 {
 600         memset(info, 0, sizeof(struct swsusp_info));
 601         info->version_code = LINUX_VERSION_CODE;
 602         info->num_physpages = num_physpages;
 603         memcpy(&info->uts, &system_utsname, sizeof(system_utsname));
 604         info->cpus = num_online_cpus();
 605         info->image_pages = nr_copy_pages;
 606         info->pages = nr_copy_pages + nr_meta_pages + 1;
 607         info->size = info->pages;
 608         info->size <<= PAGE_SHIFT;
 609 }
 610
 611 /**
 612  *      pack_orig_addresses - the .orig_address fields of the PBEs from the
 613  *      list starting at @pbe are stored in the array @buf[] (1 page)
 614  */
 615
 616 static inline struct pbe *pack_orig_addresses(unsigned long *buf, struct pbe *pbe)
 617 {
 618         int j;
 619
 620         for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
 621                 buf[j] = pbe->orig_address;
 622                 pbe = pbe->next;
 623         }
 624         if (!pbe)
 625                 for (; j < PAGE_SIZE / sizeof(long); j++)
 626                         buf[j] = 0;
 627         return pbe;
 628 }
 629
 630 /**
 631  *      snapshot_read_next - used for reading the system memory snapshot.
 632  *
 633  *      On the first call to it @handle should point to a zeroed
 634  *      snapshot_handle structure.  The structure gets updated and a pointer
 635  *      to it should be passed to this function every next time.
 636  *
 637  *      The @count parameter should contain the number of bytes the caller
 638  *      wants to read from the snapshot.  It must not be zero.
 639  *
 640  *      On success the function returns a positive number.  Then, the caller
 641  *      is allowed to read up to the returned number of bytes from the memory
 642  *      location computed by the data_of() macro.  The number returned
 643  *      may be smaller than @count, but this only happens if the read would
 644  *      cross a page boundary otherwise.
 645  *
 646  *      The function returns 0 to indicate the end of data stream condition,
 647  *      and a negative number is returned on error.  In such cases the
 648  *      structure pointed to by @handle is not updated and should not be used
 649  *      any more.
 650  */
 651
 652 int snapshot_read_next(struct snapshot_handle *handle, size_t count)
 653 {
 654         if (handle->page > nr_meta_pages + nr_copy_pages)
 655                 return 0;
 656         if (!buffer) {
 657                 /* This makes the buffer be freed by swsusp_free() */
 658                 buffer = alloc_image_page(GFP_ATOMIC, 0);
 659                 if (!buffer)
 660                         return -ENOMEM;
 661         }
 662         if (!handle->offset) {
 663                 init_header((struct swsusp_info *)buffer);
 664                 handle->buffer = buffer;
 665                 handle->pbe = pagedir_nosave;
 666         }
 667         if (handle->prev < handle->page) {
 668                 if (handle->page <= nr_meta_pages) {
 669                         handle->pbe = pack_orig_addresses(buffer, handle->pbe);
 670                         if (!handle->pbe)
 671                                 handle->pbe = pagedir_nosave;
 672                 } else {
 673                         handle->buffer = (void *)handle->pbe->address;
 674                         handle->pbe = handle->pbe->next;
 675                 }
 676                 handle->prev = handle->page;
 677         }
 678         handle->buf_offset = handle->page_offset;
 679         if (handle->page_offset + count >= PAGE_SIZE) {
 680                 count = PAGE_SIZE - handle->page_offset;
 681                 handle->page_offset = 0;
 682                 handle->page++;
 683         } else {
 684                 handle->page_offset += count;
 685         }
 686         handle->offset += count;
 687         return count;
 688 }
 689
 690 /**
 691  *      mark_unsafe_pages - mark the pages that cannot be used for storing
 692  *      the image during resume, because they conflict with the pages that
 693  *      had been used before suspend
 694  */
 695
 696 static int mark_unsafe_pages(struct pbe *pblist)
 697 {
 698         struct zone *zone;
 699         unsigned long zone_pfn;
 700         struct pbe *p;
 701
 702         if (!pblist) /* a sanity check */
 703                 return -EINVAL;
 704
 705         /* Clear page flags */
 706         for_each_zone (zone) {
 707                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 708                         if (pfn_valid(zone_pfn + zone->zone_start_pfn))
 709                                 ClearPageNosaveFree(pfn_to_page(zone_pfn +
 710                                         zone->zone_start_pfn));
 711         }
 712
 713         /* Mark orig addresses */
 714         for_each_pbe (p, pblist) {
 715                 if (virt_addr_valid(p->orig_address))
 716                         SetPageNosaveFree(virt_to_page(p->orig_address));
 717                 else
 718                         return -EFAULT;
 719         }
 720
 721         unsafe_pages = 0;
 722
 723         return 0;
 724 }
 725
 726 static void copy_page_backup_list(struct pbe *dst, struct pbe *src)
 727 {
 728         /* We assume both lists contain the same number of elements */
 729         while (src) {
 730                 dst->orig_address = src->orig_address;
 731                 dst = dst->next;
 732                 src = src->next;
 733         }
 734 }
 735
 736 static int check_header(struct swsusp_info *info)
 737 {
 738         char *reason = NULL;
 739
 740         if (info->version_code != LINUX_VERSION_CODE)
 741                 reason = "kernel version";
 742         if (info->num_physpages != num_physpages)
 743                 reason = "memory size";
 744         if (strcmp(info->uts.sysname,system_utsname.sysname))
 745                 reason = "system type";
 746         if (strcmp(info->uts.release,system_utsname.release))
 747                 reason = "kernel release";
 748         if (strcmp(info->uts.version,system_utsname.version))
 749                 reason = "version";
 750         if (strcmp(info->uts.machine,system_utsname.machine))
 751                 reason = "machine";
 752         if (reason) {
 753                 printk(KERN_ERR "swsusp: Resume mismatch: %s\n", reason);
 754                 return -EPERM;
 755         }
 756         return 0;
 757 }
 758
 759 /**
 760  *      load header - check the image header and copy data from it
 761  */
 762
 763 static int load_header(struct snapshot_handle *handle,
 764                               struct swsusp_info *info)
 765 {
 766         int error;
 767         struct pbe *pblist;
 768
 769         error = check_header(info);
 770         if (!error) {
 771                 pblist = alloc_pagedir(info->image_pages, GFP_ATOMIC, 0);
 772                 if (!pblist)
 773                         return -ENOMEM;
 774                 pagedir_nosave = pblist;
 775                 handle->pbe = pblist;
 776                 nr_copy_pages = info->image_pages;
 777                 nr_meta_pages = info->pages - info->image_pages - 1;
 778         }
 779         return error;
 780 }
 781
 782 /**
 783  *      unpack_orig_addresses - copy the elements of @buf[] (1 page) to
 784  *      the PBEs in the list starting at @pbe
 785  */
 786
 787 static inline struct pbe *unpack_orig_addresses(unsigned long *buf,
 788                                                 struct pbe *pbe)
 789 {
 790         int j;
 791
 792         for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
 793                 pbe->orig_address = buf[j];
 794                 pbe = pbe->next;
 795         }
 796         return pbe;
 797 }
 798
 799 /**
 800  *      prepare_image - use metadata contained in the PBE list
 801  *      pointed to by pagedir_nosave to mark the pages that will
 802  *      be overwritten in the process of restoring the system
 803  *      memory state from the image ("unsafe" pages) and allocate
 804  *      memory for the image
 805  *
 806  *      The idea is to allocate the PBE list first and then
 807  *      allocate as many pages as it's needed for the image data,
 808  *      but not to assign these pages to the PBEs initially.
 809  *      Instead, we just mark them as allocated and create a list
 810  *      of "safe" which will be used later
 811  */
 812
 813 struct safe_page {
 814         struct safe_page *next;
 815         char padding[PAGE_SIZE - sizeof(void *)];
 816 };
 817
 818 static struct safe_page *safe_pages;
 819
 820 static int prepare_image(struct snapshot_handle *handle)
 821 {
 822         int error = 0;
 823         unsigned int nr_pages = nr_copy_pages;
 824         struct pbe *p, *pblist = NULL;
 825
 826         p = pagedir_nosave;
 827         error = mark_unsafe_pages(p);
 828         if (!error) {
 829                 pblist = alloc_pagedir(nr_pages, GFP_ATOMIC, 1);
 830                 if (pblist)
 831                         copy_page_backup_list(pblist, p);
 832                 free_pagedir(p, 0);
 833                 if (!pblist)
 834                         error = -ENOMEM;
 835         }
 836         safe_pages = NULL;
 837         if (!error && nr_pages > unsafe_pages) {
 838                 nr_pages -= unsafe_pages;
 839                 while (nr_pages--) {
 840                         struct safe_page *ptr;
 841
 842                         ptr = (struct safe_page *)get_zeroed_page(GFP_ATOMIC);
 843                         if (!ptr) {
 844                                 error = -ENOMEM;
 845                                 break;
 846                         }
 847                         if (!PageNosaveFree(virt_to_page(ptr))) {
 848                                 /* The page is "safe", add it to the list */
 849                                 ptr->next = safe_pages;
 850                                 safe_pages = ptr;
 851                         }
 852                         /* Mark the page as allocated */
 853                         SetPageNosave(virt_to_page(ptr));
 854                         SetPageNosaveFree(virt_to_page(ptr));
 855                 }
 856         }
 857         if (!error) {
 858                 pagedir_nosave = pblist;
 859         } else {
 860                 handle->pbe = NULL;
 861                 swsusp_free();
 862         }
 863         return error;
 864 }
 865
 866 static void *get_buffer(struct snapshot_handle *handle)
 867 {
 868         struct pbe *pbe = handle->pbe, *last = handle->last_pbe;
 869         struct page *page = virt_to_page(pbe->orig_address);
 870
 871         if (PageNosave(page) && PageNosaveFree(page)) {
 872                 /*
 873                  * We have allocated the "original" page frame and we can
 874                  * use it directly to store the read page
 875                  */
 876                 pbe->address = 0;
 877                 if (last && last->next)
 878                         last->next = NULL;
 879                 return (void *)pbe->orig_address;
 880         }
 881         /*
 882          * The "original" page frame has not been allocated and we have to
 883          * use a "safe" page frame to store the read page
 884          */
 885         pbe->address = (unsigned long)safe_pages;
 886         safe_pages = safe_pages->next;
 887         if (last)
 888                 last->next = pbe;
 889         handle->last_pbe = pbe;
 890         return (void *)pbe->address;
 891 }
 892
 893 /**
 894  *      snapshot_write_next - used for writing the system memory snapshot.
 895  *
 896  *      On the first call to it @handle should point to a zeroed
 897  *      snapshot_handle structure.  The structure gets updated and a pointer
 898  *      to it should be passed to this function every next time.
 899  *
 900  *      The @count parameter should contain the number of bytes the caller
 901  *      wants to write to the image.  It must not be zero.
 902  *
 903  *      On success the function returns a positive number.  Then, the caller
 904  *      is allowed to write up to the returned number of bytes to the memory
 905  *      location computed by the data_of() macro.  The number returned
 906  *      may be smaller than @count, but this only happens if the write would
 907  *      cross a page boundary otherwise.
 908  *
 909  *      The function returns 0 to indicate the "end of file" condition,
 910  *      and a negative number is returned on error.  In such cases the
 911  *      structure pointed to by @handle is not updated and should not be used
 912  *      any more.
 913  */
 914
 915 int snapshot_write_next(struct snapshot_handle *handle, size_t count)
 916 {
 917         int error = 0;
 918
 919         if (handle->prev && handle->page > nr_meta_pages + nr_copy_pages)
 920                 return 0;
 921         if (!buffer) {
 922                 /* This makes the buffer be freed by swsusp_free() */
 923                 buffer = alloc_image_page(GFP_ATOMIC, 0);
 924                 if (!buffer)
 925                         return -ENOMEM;
 926         }
 927         if (!handle->offset)
 928                 handle->buffer = buffer;
 929         if (handle->prev < handle->page) {
 930                 if (!handle->prev) {
 931                         error = load_header(handle, (struct swsusp_info *)buffer);
 932                         if (error)
 933                                 return error;
 934                 } else if (handle->prev <= nr_meta_pages) {
 935                         handle->pbe = unpack_orig_addresses(buffer, handle->pbe);
 936                         if (!handle->pbe) {
 937                                 error = prepare_image(handle);
 938                                 if (error)
 939                                         return error;
 940                                 handle->pbe = pagedir_nosave;
 941                                 handle->last_pbe = NULL;
 942                                 handle->buffer = get_buffer(handle);
 943                         }
 944                 } else {
 945                         handle->pbe = handle->pbe->next;
 946                         handle->buffer = get_buffer(handle);
 947                 }
 948                 handle->prev = handle->page;
 949         }
 950         handle->buf_offset = handle->page_offset;
 951         if (handle->page_offset + count >= PAGE_SIZE) {
 952                 count = PAGE_SIZE - handle->page_offset;
 953                 handle->page_offset = 0;
 954                 handle->page++;
 955         } else {
 956                 handle->page_offset += count;
 957         }
 958         handle->offset += count;
 959         return count;
 960 }
 961
 962 int snapshot_image_loaded(struct snapshot_handle *handle)
 963 {
 964         return !(!handle->pbe || handle->pbe->next || !nr_copy_pages ||
 965                 handle->page <= nr_meta_pages + nr_copy_pages);
 966 }