kernel/power/swsusp.c

   1 /*
   2  * linux/kernel/power/swsusp.c
   3  *
   4  * This file provides code to write suspend image to swap and read it back.
   5  *
   6  * Copyright (C) 1998-2001 Gabor Kuti <seasons@fornax.hu>
   7  * Copyright (C) 1998,2001-2005 Pavel Machek <pavel@suse.cz>
   8  *
   9  * This file is released under the GPLv2.
  10  *
  11  * I'd like to thank the following people for their work:
  12  *
  13  * Pavel Machek <pavel@ucw.cz>:
  14  * Modifications, defectiveness pointing, being with me at the very beginning,
  15  * suspend to swap space, stop all tasks. Port to 2.4.18-ac and 2.5.17.
  16  *
  17  * Steve Doddi <dirk@loth.demon.co.uk>:
  18  * Support the possibility of hardware state restoring.
  19  *
  20  * Raph <grey.havens@earthling.net>:
  21  * Support for preserving states of network devices and virtual console
  22  * (including X and svgatextmode)
  23  *
  24  * Kurt Garloff <garloff@suse.de>:
  25  * Straightened the critical function in order to prevent compilers from
  26  * playing tricks with local variables.
  27  *
  28  * Andreas Mohr <a.mohr@mailto.de>
  29  *
  30  * Alex Badea <vampire@go.ro>:
  31  * Fixed runaway init
  32  *
  33  * Rafael J. Wysocki <rjw@sisk.pl>
  34  * Added the swap map data structure and reworked the handling of swap
  35  *
  36  * More state savers are welcome. Especially for the scsi layer...
  37  *
  38  * For TODOs,FIXMEs also look in Documentation/power/swsusp.txt
  39  */
  40
  41 #include <linux/module.h>
  42 #include <linux/mm.h>
  43 #include <linux/suspend.h>
  44 #include <linux/smp_lock.h>
  45 #include <linux/file.h>
  46 #include <linux/utsname.h>
  47 #include <linux/version.h>
  48 #include <linux/delay.h>
  49 #include <linux/bitops.h>
  50 #include <linux/spinlock.h>
  51 #include <linux/genhd.h>
  52 #include <linux/kernel.h>
  53 #include <linux/major.h>
  54 #include <linux/swap.h>
  55 #include <linux/pm.h>
  56 #include <linux/device.h>
  57 #include <linux/buffer_head.h>
  58 #include <linux/swapops.h>
  59 #include <linux/bootmem.h>
  60 #include <linux/syscalls.h>
  61 #include <linux/highmem.h>
  62 #include <linux/bio.h>
  63
  64 #include <asm/uaccess.h>
  65 #include <asm/mmu_context.h>
  66 #include <asm/pgtable.h>
  67 #include <asm/tlbflush.h>
  68 #include <asm/io.h>
  69
  70 #include "power.h"
  71
  72 /*
  73  * Preferred image size in MB (tunable via /sys/power/image_size).
  74  * When it is set to N, swsusp will do its best to ensure the image
  75  * size will not exceed N MB, but if that is impossible, it will
  76  * try to create the smallest image possible.
  77  */
  78 unsigned int image_size = 500;
  79
  80 #ifdef CONFIG_HIGHMEM
  81 unsigned int count_highmem_pages(void);
  82 int save_highmem(void);
  83 int restore_highmem(void);
  84 #else
  85 static int save_highmem(void) { return 0; }
  86 static int restore_highmem(void) { return 0; }
  87 static unsigned int count_highmem_pages(void) { return 0; }
  88 #endif
  89
  90 extern char resume_file[];
  91
  92 #define SWSUSP_SIG      "S1SUSPEND"
  93
  94 static struct swsusp_header {
  95         char reserved[PAGE_SIZE - 20 - sizeof(swp_entry_t)];
  96         swp_entry_t swsusp_info;
  97         char    orig_sig[10];
  98         char    sig[10];
  99 } __attribute__((packed, aligned(PAGE_SIZE))) swsusp_header;
 100
 101 static struct swsusp_info swsusp_info;
 102
 103 /*
 104  * Saving part...
 105  */
 106
 107 /* We memorize in swapfile_used what swap devices are used for suspension */
 108 #define SWAPFILE_UNUSED    0
 109 #define SWAPFILE_SUSPEND   1    /* This is the suspending device */
 110 #define SWAPFILE_IGNORED   2    /* Those are other swap devices ignored for suspension */
 111
 112 static unsigned short swapfile_used[MAX_SWAPFILES];
 113 static unsigned short root_swap;
 114
 115 static int mark_swapfiles(swp_entry_t prev)
 116 {
 117         int error;
 118
 119         rw_swap_page_sync(READ,
 120                           swp_entry(root_swap, 0),
 121                           virt_to_page((unsigned long)&swsusp_header));
 122         if (!memcmp("SWAP-SPACE",swsusp_header.sig, 10) ||
 123             !memcmp("SWAPSPACE2",swsusp_header.sig, 10)) {
 124                 memcpy(swsusp_header.orig_sig,swsusp_header.sig, 10);
 125                 memcpy(swsusp_header.sig,SWSUSP_SIG, 10);
 126                 swsusp_header.swsusp_info = prev;
 127                 error = rw_swap_page_sync(WRITE,
 128                                           swp_entry(root_swap, 0),
 129                                           virt_to_page((unsigned long)
 130                                                        &swsusp_header));
 131         } else {
 132                 pr_debug("swsusp: Partition is not swap space.\n");
 133                 error = -ENODEV;
 134         }
 135         return error;
 136 }
 137
 138 /*
 139  * Check whether the swap device is the specified resume
 140  * device, irrespective of whether they are specified by
 141  * identical names.
 142  *
 143  * (Thus, device inode aliasing is allowed.  You can say /dev/hda4
 144  * instead of /dev/ide/host0/bus0/target0/lun0/part4 [if using devfs]
 145  * and they'll be considered the same device.  This is *necessary* for
 146  * devfs, since the resume code can only recognize the form /dev/hda4,
 147  * but the suspend code would see the long name.)
 148  */
 149 static int is_resume_device(const struct swap_info_struct *swap_info)
 150 {
 151         struct file *file = swap_info->swap_file;
 152         struct inode *inode = file->f_dentry->d_inode;
 153
 154         return S_ISBLK(inode->i_mode) &&
 155                 swsusp_resume_device == MKDEV(imajor(inode), iminor(inode));
 156 }
 157
 158 static int swsusp_swap_check(void) /* This is called before saving image */
 159 {
 160         int i, len;
 161
 162         len=strlen(resume_file);
 163         root_swap = 0xFFFF;
 164
 165         spin_lock(&swap_lock);
 166         for (i=0; i<MAX_SWAPFILES; i++) {
 167                 if (!(swap_info[i].flags & SWP_WRITEOK)) {
 168                         swapfile_used[i]=SWAPFILE_UNUSED;
 169                 } else {
 170                         if (!len) {
 171                                 printk(KERN_WARNING "resume= option should be used to set suspend device" );
 172                                 if (root_swap == 0xFFFF) {
 173                                         swapfile_used[i] = SWAPFILE_SUSPEND;
 174                                         root_swap = i;
 175                                 } else
 176                                         swapfile_used[i] = SWAPFILE_IGNORED;
 177                         } else {
 178                                 /* we ignore all swap devices that are not the resume_file */
 179                                 if (is_resume_device(&swap_info[i])) {
 180                                         swapfile_used[i] = SWAPFILE_SUSPEND;
 181                                         root_swap = i;
 182                                 } else {
 183                                         swapfile_used[i] = SWAPFILE_IGNORED;
 184                                 }
 185                         }
 186                 }
 187         }
 188         spin_unlock(&swap_lock);
 189         return (root_swap != 0xffff) ? 0 : -ENODEV;
 190 }
 191
 192 /**
 193  * This is called after saving image so modification
 194  * will be lost after resume... and that's what we want.
 195  * we make the device unusable. A new call to
 196  * lock_swapdevices can unlock the devices.
 197  */
 198 static void lock_swapdevices(void)
 199 {
 200         int i;
 201
 202         spin_lock(&swap_lock);
 203         for (i = 0; i< MAX_SWAPFILES; i++)
 204                 if (swapfile_used[i] == SWAPFILE_IGNORED) {
 205                         swap_info[i].flags ^= SWP_WRITEOK;
 206                 }
 207         spin_unlock(&swap_lock);
 208 }
 209
 210 /**
 211  *      write_page - Write one page to a fresh swap location.
 212  *      @addr:  Address we're writing.
 213  *      @loc:   Place to store the entry we used.
 214  *
 215  *      Allocate a new swap entry and 'sync' it. Note we discard -EIO
 216  *      errors. That is an artifact left over from swsusp. It did not
 217  *      check the return of rw_swap_page_sync() at all, since most pages
 218  *      written back to swap would return -EIO.
 219  *      This is a partial improvement, since we will at least return other
 220  *      errors, though we need to eventually fix the damn code.
 221  */
 222 static int write_page(unsigned long addr, swp_entry_t *loc)
 223 {
 224         swp_entry_t entry;
 225         int error = 0;
 226
 227         entry = get_swap_page();
 228         if (swp_offset(entry) &&
 229             swapfile_used[swp_type(entry)] == SWAPFILE_SUSPEND) {
 230                 error = rw_swap_page_sync(WRITE, entry,
 231                                           virt_to_page(addr));
 232                 if (error == -EIO)
 233                         error = 0;
 234                 if (!error)
 235                         *loc = entry;
 236         } else
 237                 error = -ENOSPC;
 238         return error;
 239 }
 240
 241 /**
 242  *      Swap map-handling functions
 243  *
 244  *      The swap map is a data structure used for keeping track of each page
 245  *      written to the swap.  It consists of many swap_map_page structures
 246  *      that contain each an array of MAP_PAGE_SIZE swap entries.
 247  *      These structures are linked together with the help of either the
 248  *      .next (in memory) or the .next_swap (in swap) member.
 249  *
 250  *      The swap map is created during suspend.  At that time we need to keep
 251  *      it in memory, because we have to free all of the allocated swap
 252  *      entries if an error occurs.  The memory needed is preallocated
 253  *      so that we know in advance if there's enough of it.
 254  *
 255  *      The first swap_map_page structure is filled with the swap entries that
 256  *      correspond to the first MAP_PAGE_SIZE data pages written to swap and
 257  *      so on.  After the all of the data pages have been written, the order
 258  *      of the swap_map_page structures in the map is reversed so that they
 259  *      can be read from swap in the original order.  This causes the data
 260  *      pages to be loaded in exactly the same order in which they have been
 261  *      saved.
 262  *
 263  *      During resume we only need to use one swap_map_page structure
 264  *      at a time, which means that we only need to use two memory pages for
 265  *      reading the image - one for reading the swap_map_page structures
 266  *      and the second for reading the data pages from swap.
 267  */
 268
 269 #define MAP_PAGE_SIZE   ((PAGE_SIZE - sizeof(swp_entry_t) - sizeof(void *)) \
 270                         / sizeof(swp_entry_t))
 271
 272 struct swap_map_page {
 273         swp_entry_t             entries[MAP_PAGE_SIZE];
 274         swp_entry_t             next_swap;
 275         struct swap_map_page    *next;
 276 };
 277
 278 static inline void free_swap_map(struct swap_map_page *swap_map)
 279 {
 280         struct swap_map_page *swp;
 281
 282         while (swap_map) {
 283                 swp = swap_map->next;
 284                 free_page((unsigned long)swap_map);
 285                 swap_map = swp;
 286         }
 287 }
 288
 289 static struct swap_map_page *alloc_swap_map(unsigned int nr_pages)
 290 {
 291         struct swap_map_page *swap_map, *swp;
 292         unsigned n = 0;
 293
 294         if (!nr_pages)
 295                 return NULL;
 296
 297         pr_debug("alloc_swap_map(): nr_pages = %d\n", nr_pages);
 298         swap_map = (struct swap_map_page *)get_zeroed_page(GFP_ATOMIC);
 299         swp = swap_map;
 300         for (n = MAP_PAGE_SIZE; n < nr_pages; n += MAP_PAGE_SIZE) {
 301                 swp->next = (struct swap_map_page *)get_zeroed_page(GFP_ATOMIC);
 302                 swp = swp->next;
 303                 if (!swp) {
 304                         free_swap_map(swap_map);
 305                         return NULL;
 306                 }
 307         }
 308         return swap_map;
 309 }
 310
 311 /**
 312  *      reverse_swap_map - reverse the order of pages in the swap map
 313  *      @swap_map
 314  */
 315
 316 static inline struct swap_map_page *reverse_swap_map(struct swap_map_page *swap_map)
 317 {
 318         struct swap_map_page *prev, *next;
 319
 320         prev = NULL;
 321         while (swap_map) {
 322                 next = swap_map->next;
 323                 swap_map->next = prev;
 324                 prev = swap_map;
 325                 swap_map = next;
 326         }
 327         return prev;
 328 }
 329
 330 /**
 331  *      free_swap_map_entries - free the swap entries allocated to store
 332  *      the swap map @swap_map (this is only called in case of an error)
 333  */
 334 static inline void free_swap_map_entries(struct swap_map_page *swap_map)
 335 {
 336         while (swap_map) {
 337                 if (swap_map->next_swap.val)
 338                         swap_free(swap_map->next_swap);
 339                 swap_map = swap_map->next;
 340         }
 341 }
 342
 343 /**
 344  *      save_swap_map - save the swap map used for tracing the data pages
 345  *      stored in the swap
 346  */
 347
 348 static int save_swap_map(struct swap_map_page *swap_map, swp_entry_t *start)
 349 {
 350         swp_entry_t entry = (swp_entry_t){0};
 351         int error;
 352
 353         while (swap_map) {
 354                 swap_map->next_swap = entry;
 355                 if ((error = write_page((unsigned long)swap_map, &entry)))
 356                         return error;
 357                 swap_map = swap_map->next;
 358         }
 359         *start = entry;
 360         return 0;
 361 }
 362
 363 /**
 364  *      free_image_entries - free the swap entries allocated to store
 365  *      the image data pages (this is only called in case of an error)
 366  */
 367
 368 static inline void free_image_entries(struct swap_map_page *swp)
 369 {
 370         unsigned k;
 371
 372         while (swp) {
 373                 for (k = 0; k < MAP_PAGE_SIZE; k++)
 374                         if (swp->entries[k].val)
 375                                 swap_free(swp->entries[k]);
 376                 swp = swp->next;
 377         }
 378 }
 379
 380 /**
 381  *      The swap_map_handle structure is used for handling the swap map in
 382  *      a file-alike way
 383  */
 384
 385 struct swap_map_handle {
 386         struct swap_map_page *cur;
 387         unsigned int k;
 388 };
 389
 390 static inline void init_swap_map_handle(struct swap_map_handle *handle,
 391                                         struct swap_map_page *map)
 392 {
 393         handle->cur = map;
 394         handle->k = 0;
 395 }
 396
 397 static inline int swap_map_write_page(struct swap_map_handle *handle,
 398                                       unsigned long addr)
 399 {
 400         int error;
 401
 402         error = write_page(addr, handle->cur->entries + handle->k);
 403         if (error)
 404                 return error;
 405         if (++handle->k >= MAP_PAGE_SIZE) {
 406                 handle->cur = handle->cur->next;
 407                 handle->k = 0;
 408         }
 409         return 0;
 410 }
 411
 412 /**
 413  *      save_image_data - save the data pages pointed to by the PBEs
 414  *      from the list @pblist using the swap map handle @handle
 415  *      (assume there are @nr_pages data pages to save)
 416  */
 417
 418 static int save_image_data(struct pbe *pblist,
 419                            struct swap_map_handle *handle,
 420                            unsigned int nr_pages)
 421 {
 422         unsigned int m;
 423         struct pbe *p;
 424         int error = 0;
 425
 426         printk("Saving image data pages (%u pages) ...     ", nr_pages);
 427         m = nr_pages / 100;
 428         if (!m)
 429                 m = 1;
 430         nr_pages = 0;
 431         for_each_pbe (p, pblist) {
 432                 error = swap_map_write_page(handle, p->address);
 433                 if (error)
 434                         break;
 435                 if (!(nr_pages % m))
 436                         printk("\b\b\b\b%3d%%", nr_pages / m);
 437                 nr_pages++;
 438         }
 439         if (!error)
 440                 printk("\b\b\b\bdone\n");
 441         return error;
 442 }
 443
 444 static void dump_info(void)
 445 {
 446         pr_debug(" swsusp: Version: %u\n",swsusp_info.version_code);
 447         pr_debug(" swsusp: Num Pages: %ld\n",swsusp_info.num_physpages);
 448         pr_debug(" swsusp: UTS Sys: %s\n",swsusp_info.uts.sysname);
 449         pr_debug(" swsusp: UTS Node: %s\n",swsusp_info.uts.nodename);
 450         pr_debug(" swsusp: UTS Release: %s\n",swsusp_info.uts.release);
 451         pr_debug(" swsusp: UTS Version: %s\n",swsusp_info.uts.version);
 452         pr_debug(" swsusp: UTS Machine: %s\n",swsusp_info.uts.machine);
 453         pr_debug(" swsusp: UTS Domain: %s\n",swsusp_info.uts.domainname);
 454         pr_debug(" swsusp: CPUs: %d\n",swsusp_info.cpus);
 455         pr_debug(" swsusp: Image: %ld Pages\n",swsusp_info.image_pages);
 456         pr_debug(" swsusp: Total: %ld Pages\n", swsusp_info.pages);
 457 }
 458
 459 static void init_header(unsigned int nr_pages)
 460 {
 461         memset(&swsusp_info, 0, sizeof(swsusp_info));
 462         swsusp_info.version_code = LINUX_VERSION_CODE;
 463         swsusp_info.num_physpages = num_physpages;
 464         memcpy(&swsusp_info.uts, &system_utsname, sizeof(system_utsname));
 465
 466         swsusp_info.cpus = num_online_cpus();
 467         swsusp_info.image_pages = nr_pages;
 468         swsusp_info.pages = nr_pages +
 469                 ((nr_pages * sizeof(long) + PAGE_SIZE - 1) >> PAGE_SHIFT);
 470 }
 471
 472 static int close_swap(void)
 473 {
 474         swp_entry_t entry;
 475         int error;
 476
 477         dump_info();
 478         error = write_page((unsigned long)&swsusp_info, &entry);
 479         if (!error) {
 480                 printk( "S" );
 481                 error = mark_swapfiles(entry);
 482                 printk( "|\n" );
 483         }
 484         return error;
 485 }
 486
 487 /**
 488  *      pack_orig_addresses - the .orig_address fields of the PBEs from the
 489  *      list starting at @pbe are stored in the array @buf[] (1 page)
 490  */
 491
 492 static inline struct pbe *pack_orig_addresses(unsigned long *buf,
 493                                               struct pbe *pbe)
 494 {
 495         int j;
 496
 497         for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
 498                 buf[j] = pbe->orig_address;
 499                 pbe = pbe->next;
 500         }
 501         if (!pbe)
 502                 for (; j < PAGE_SIZE / sizeof(long); j++)
 503                         buf[j] = 0;
 504         return pbe;
 505 }
 506
 507 /**
 508  *      save_image_metadata - save the .orig_address fields of the PBEs
 509  *      from the list @pblist using the swap map handle @handle
 510  */
 511
 512 static int save_image_metadata(struct pbe *pblist,
 513                                struct swap_map_handle *handle)
 514 {
 515         unsigned long *buf;
 516         unsigned int n = 0;
 517         struct pbe *p;
 518         int error = 0;
 519
 520         printk("Saving image metadata ... ");
 521         buf = (unsigned long *)get_zeroed_page(GFP_ATOMIC);
 522         if (!buf)
 523                 return -ENOMEM;
 524         p = pblist;
 525         while (p) {
 526                 p = pack_orig_addresses(buf, p);
 527                 error = swap_map_write_page(handle, (unsigned long)buf);
 528                 if (error)
 529                         break;
 530                 n++;
 531         }
 532         free_page((unsigned long)buf);
 533         if (!error)
 534                 printk("done (%u pages saved)\n", n);
 535         return error;
 536 }
 537
 538 /**
 539  *      enough_swap - Make sure we have enough swap to save the image.
 540  *
 541  *      Returns TRUE or FALSE after checking the total amount of swap
 542  *      space avaiable.
 543  *
 544  *      FIXME: si_swapinfo(&i) returns all swap devices information.
 545  *      We should only consider resume_device.
 546  */
 547
 548 static int enough_swap(unsigned int nr_pages)
 549 {
 550         struct sysinfo i;
 551
 552         si_swapinfo(&i);
 553         pr_debug("swsusp: available swap: %lu pages\n", i.freeswap);
 554         return i.freeswap > (nr_pages + PAGES_FOR_IO +
 555                 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
 556 }
 557
 558 /**
 559  *      write_suspend_image - Write entire image and metadata.
 560  */
 561 static int write_suspend_image(struct pbe *pblist, unsigned int nr_pages)
 562 {
 563         struct swap_map_page *swap_map;
 564         struct swap_map_handle handle;
 565         int error;
 566
 567         if (!enough_swap(nr_pages)) {
 568                 printk(KERN_ERR "swsusp: Not enough free swap\n");
 569                 return -ENOSPC;
 570         }
 571
 572         init_header(nr_pages);
 573         swap_map = alloc_swap_map(swsusp_info.pages);
 574         if (!swap_map)
 575                 return -ENOMEM;
 576         init_swap_map_handle(&handle, swap_map);
 577
 578         error = save_image_metadata(pblist, &handle);
 579         if (!error)
 580                 error = save_image_data(pblist, &handle, nr_pages);
 581         if (error)
 582                 goto Free_image_entries;
 583
 584         swap_map = reverse_swap_map(swap_map);
 585         error = save_swap_map(swap_map, &swsusp_info.start);
 586         if (error)
 587                 goto Free_map_entries;
 588
 589         error = close_swap();
 590         if (error)
 591                 goto Free_map_entries;
 592
 593 Free_swap_map:
 594         free_swap_map(swap_map);
 595         return error;
 596
 597 Free_map_entries:
 598         free_swap_map_entries(swap_map);
 599 Free_image_entries:
 600         free_image_entries(swap_map);
 601         goto Free_swap_map;
 602 }
 603
 604 /* It is important _NOT_ to umount filesystems at this point. We want
 605  * them synced (in case something goes wrong) but we DO not want to mark
 606  * filesystem clean: it is not. (And it does not matter, if we resume
 607  * correctly, we'll mark system clean, anyway.)
 608  */
 609 int swsusp_write(struct pbe *pblist, unsigned int nr_pages)
 610 {
 611         int error;
 612
 613         if ((error = swsusp_swap_check())) {
 614                 printk(KERN_ERR "swsusp: cannot find swap device, try swapon -a.\n");
 615                 return error;
 616         }
 617         lock_swapdevices();
 618         error = write_suspend_image(pblist, nr_pages);
 619         /* This will unlock ignored swap devices since writing is finished */
 620         lock_swapdevices();
 621         return error;
 622 }
 623
 624 /**
 625  *      swsusp_shrink_memory -  Try to free as much memory as needed
 626  *
 627  *      ... but do not OOM-kill anyone
 628  *
 629  *      Notice: all userland should be stopped before it is called, or
 630  *      livelock is possible.
 631  */
 632
 633 #define SHRINK_BITE     10000
 634
 635 int swsusp_shrink_memory(void)
 636 {
 637         long size, tmp;
 638         struct zone *zone;
 639         unsigned long pages = 0;
 640         unsigned int i = 0;
 641         char *p = "-\\|/";
 642
 643         printk("Shrinking memory...  ");
 644         do {
 645                 size = 2 * count_highmem_pages();
 646                 size += size / 50 + count_data_pages();
 647                 size += (size + PBES_PER_PAGE - 1) / PBES_PER_PAGE +
 648                         PAGES_FOR_IO;
 649                 tmp = size;
 650                 for_each_zone (zone)
 651                         if (!is_highmem(zone))
 652                                 tmp -= zone->free_pages;
 653                 if (tmp > 0) {
 654                         tmp = shrink_all_memory(SHRINK_BITE);
 655                         if (!tmp)
 656                                 return -ENOMEM;
 657                         pages += tmp;
 658                 } else if (size > (image_size * 1024 * 1024) / PAGE_SIZE) {
 659                         tmp = shrink_all_memory(SHRINK_BITE);
 660                         pages += tmp;
 661                 }
 662                 printk("\b%c", p[i++%4]);
 663         } while (tmp > 0);
 664         printk("\bdone (%lu pages freed)\n", pages);
 665
 666         return 0;
 667 }
 668
 669 int swsusp_suspend(void)
 670 {
 671         int error;
 672
 673         if ((error = arch_prepare_suspend()))
 674                 return error;
 675         local_irq_disable();
 676         /* At this point, device_suspend() has been called, but *not*
 677          * device_power_down(). We *must* device_power_down() now.
 678          * Otherwise, drivers for some devices (e.g. interrupt controllers)
 679          * become desynchronized with the actual state of the hardware
 680          * at resume time, and evil weirdness ensues.
 681          */
 682         if ((error = device_power_down(PMSG_FREEZE))) {
 683                 printk(KERN_ERR "Some devices failed to power down, aborting suspend\n");
 684                 goto Enable_irqs;
 685         }
 686
 687         if ((error = save_highmem())) {
 688                 printk(KERN_ERR "swsusp: Not enough free pages for highmem\n");
 689                 goto Restore_highmem;
 690         }
 691
 692         save_processor_state();
 693         if ((error = swsusp_arch_suspend()))
 694                 printk(KERN_ERR "Error %d suspending\n", error);
 695         /* Restore control flow magically appears here */
 696         restore_processor_state();
 697 Restore_highmem:
 698         restore_highmem();
 699         device_power_up();
 700 Enable_irqs:
 701         local_irq_enable();
 702         return error;
 703 }
 704
 705 int swsusp_resume(void)
 706 {
 707         int error;
 708         local_irq_disable();
 709         if (device_power_down(PMSG_FREEZE))
 710                 printk(KERN_ERR "Some devices failed to power down, very bad\n");
 711         /* We'll ignore saved state, but this gets preempt count (etc) right */
 712         save_processor_state();
 713         error = swsusp_arch_resume();
 714         /* Code below is only ever reached in case of failure. Otherwise
 715          * execution continues at place where swsusp_arch_suspend was called
 716          */
 717         BUG_ON(!error);
 718         /* The only reason why swsusp_arch_resume() can fail is memory being
 719          * very tight, so we have to free it as soon as we can to avoid
 720          * subsequent failures
 721          */
 722         swsusp_free();
 723         restore_processor_state();
 724         restore_highmem();
 725         touch_softlockup_watchdog();
 726         device_power_up();
 727         local_irq_enable();
 728         return error;
 729 }
 730
 731 /**
 732  *      mark_unsafe_pages - mark the pages that cannot be used for storing
 733  *      the image during resume, because they conflict with the pages that
 734  *      had been used before suspend
 735  */
 736
 737 static void mark_unsafe_pages(struct pbe *pblist)
 738 {
 739         struct zone *zone;
 740         unsigned long zone_pfn;
 741         struct pbe *p;
 742
 743         if (!pblist) /* a sanity check */
 744                 return;
 745
 746         /* Clear page flags */
 747         for_each_zone (zone) {
 748                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 749                         if (pfn_valid(zone_pfn + zone->zone_start_pfn))
 750                                 ClearPageNosaveFree(pfn_to_page(zone_pfn +
 751                                         zone->zone_start_pfn));
 752         }
 753
 754         /* Mark orig addresses */
 755         for_each_pbe (p, pblist)
 756                 SetPageNosaveFree(virt_to_page(p->orig_address));
 757
 758 }
 759
 760 static void copy_page_backup_list(struct pbe *dst, struct pbe *src)
 761 {
 762         /* We assume both lists contain the same number of elements */
 763         while (src) {
 764                 dst->orig_address = src->orig_address;
 765                 dst = dst->next;
 766                 src = src->next;
 767         }
 768 }
 769
 770 /*
 771  *      Using bio to read from swap.
 772  *      This code requires a bit more work than just using buffer heads
 773  *      but, it is the recommended way for 2.5/2.6.
 774  *      The following are to signal the beginning and end of I/O. Bios
 775  *      finish asynchronously, while we want them to happen synchronously.
 776  *      A simple atomic_t, and a wait loop take care of this problem.
 777  */
 778
 779 static atomic_t io_done = ATOMIC_INIT(0);
 780
 781 static int end_io(struct bio *bio, unsigned int num, int err)
 782 {
 783         if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
 784                 panic("I/O error reading memory image");
 785         atomic_set(&io_done, 0);
 786         return 0;
 787 }
 788
 789 static struct block_device *resume_bdev;
 790
 791 /**
 792  *      submit - submit BIO request.
 793  *      @rw:    READ or WRITE.
 794  *      @off    physical offset of page.
 795  *      @page:  page we're reading or writing.
 796  *
 797  *      Straight from the textbook - allocate and initialize the bio.
 798  *      If we're writing, make sure the page is marked as dirty.
 799  *      Then submit it and wait.
 800  */
 801
 802 static int submit(int rw, pgoff_t page_off, void *page)
 803 {
 804         int error = 0;
 805         struct bio *bio;
 806
 807         bio = bio_alloc(GFP_ATOMIC, 1);
 808         if (!bio)
 809                 return -ENOMEM;
 810         bio->bi_sector = page_off * (PAGE_SIZE >> 9);
 811         bio_get(bio);
 812         bio->bi_bdev = resume_bdev;
 813         bio->bi_end_io = end_io;
 814
 815         if (bio_add_page(bio, virt_to_page(page), PAGE_SIZE, 0) < PAGE_SIZE) {
 816                 printk("swsusp: ERROR: adding page to bio at %ld\n",page_off);
 817                 error = -EFAULT;
 818                 goto Done;
 819         }
 820
 821         if (rw == WRITE)
 822                 bio_set_pages_dirty(bio);
 823
 824         atomic_set(&io_done, 1);
 825         submit_bio(rw | (1 << BIO_RW_SYNC), bio);
 826         while (atomic_read(&io_done))
 827                 yield();
 828
 829  Done:
 830         bio_put(bio);
 831         return error;
 832 }
 833
 834 static int bio_read_page(pgoff_t page_off, void *page)
 835 {
 836         return submit(READ, page_off, page);
 837 }
 838
 839 static int bio_write_page(pgoff_t page_off, void *page)
 840 {
 841         return submit(WRITE, page_off, page);
 842 }
 843
 844 /**
 845  *      The following functions allow us to read data using a swap map
 846  *      in a file-alike way
 847  */
 848
 849 static inline void release_swap_map_reader(struct swap_map_handle *handle)
 850 {
 851         if (handle->cur)
 852                 free_page((unsigned long)handle->cur);
 853         handle->cur = NULL;
 854 }
 855
 856 static inline int get_swap_map_reader(struct swap_map_handle *handle,
 857                                       swp_entry_t start)
 858 {
 859         int error;
 860
 861         if (!swp_offset(start))
 862                 return -EINVAL;
 863         handle->cur = (struct swap_map_page *)get_zeroed_page(GFP_ATOMIC);
 864         if (!handle->cur)
 865                 return -ENOMEM;
 866         error = bio_read_page(swp_offset(start), handle->cur);
 867         if (error) {
 868                 release_swap_map_reader(handle);
 869                 return error;
 870         }
 871         handle->k = 0;
 872         return 0;
 873 }
 874
 875 static inline int swap_map_read_page(struct swap_map_handle *handle, void *buf)
 876 {
 877         unsigned long offset;
 878         int error;
 879
 880         if (!handle->cur)
 881                 return -EINVAL;
 882         offset = swp_offset(handle->cur->entries[handle->k]);
 883         if (!offset)
 884                 return -EINVAL;
 885         error = bio_read_page(offset, buf);
 886         if (error)
 887                 return error;
 888         if (++handle->k >= MAP_PAGE_SIZE) {
 889                 handle->k = 0;
 890                 offset = swp_offset(handle->cur->next_swap);
 891                 if (!offset)
 892                         release_swap_map_reader(handle);
 893                 else
 894                         error = bio_read_page(offset, handle->cur);
 895         }
 896         return error;
 897 }
 898
 899 /*
 900  * Sanity check if this image makes sense with this kernel/swap context
 901  * I really don't think that it's foolproof but more than nothing..
 902  */
 903
 904 static const char *sanity_check(void)
 905 {
 906         dump_info();
 907         if (swsusp_info.version_code != LINUX_VERSION_CODE)
 908                 return "kernel version";
 909         if (swsusp_info.num_physpages != num_physpages)
 910                 return "memory size";
 911         if (strcmp(swsusp_info.uts.sysname,system_utsname.sysname))
 912                 return "system type";
 913         if (strcmp(swsusp_info.uts.release,system_utsname.release))
 914                 return "kernel release";
 915         if (strcmp(swsusp_info.uts.version,system_utsname.version))
 916                 return "version";
 917         if (strcmp(swsusp_info.uts.machine,system_utsname.machine))
 918                 return "machine";
 919 #if 0
 920         /* We can't use number of online CPUs when we use hotplug to remove them ;-))) */
 921         if (swsusp_info.cpus != num_possible_cpus())
 922                 return "number of cpus";
 923 #endif
 924         return NULL;
 925 }
 926
 927 static int check_header(void)
 928 {
 929         const char *reason = NULL;
 930         int error;
 931
 932         if ((error = bio_read_page(swp_offset(swsusp_header.swsusp_info), &swsusp_info)))
 933                 return error;
 934
 935         /* Is this same machine? */
 936         if ((reason = sanity_check())) {
 937                 printk(KERN_ERR "swsusp: Resume mismatch: %s\n",reason);
 938                 return -EPERM;
 939         }
 940         return error;
 941 }
 942
 943 static int check_sig(void)
 944 {
 945         int error;
 946
 947         memset(&swsusp_header, 0, sizeof(swsusp_header));
 948         if ((error = bio_read_page(0, &swsusp_header)))
 949                 return error;
 950         if (!memcmp(SWSUSP_SIG, swsusp_header.sig, 10)) {
 951                 memcpy(swsusp_header.sig, swsusp_header.orig_sig, 10);
 952
 953                 /*
 954                  * Reset swap signature now.
 955                  */
 956                 error = bio_write_page(0, &swsusp_header);
 957         } else {
 958                 return -EINVAL;
 959         }
 960         if (!error)
 961                 pr_debug("swsusp: Signature found, resuming\n");
 962         return error;
 963 }
 964
 965 /**
 966  *      load_image_data - load the image data using the swap map handle
 967  *      @handle and store them using the page backup list @pblist
 968  *      (assume there are @nr_pages pages to load)
 969  */
 970
 971 static int load_image_data(struct pbe *pblist,
 972                            struct swap_map_handle *handle,
 973                            unsigned int nr_pages)
 974 {
 975         int error;
 976         unsigned int m;
 977         struct pbe *p;
 978
 979         if (!pblist)
 980                 return -EINVAL;
 981         printk("Loading image data pages (%u pages) ...     ", nr_pages);
 982         m = nr_pages / 100;
 983         if (!m)
 984                 m = 1;
 985         nr_pages = 0;
 986         p = pblist;
 987         while (p) {
 988                 error = swap_map_read_page(handle, (void *)p->address);
 989                 if (error)
 990                         break;
 991                 p = p->next;
 992                 if (!(nr_pages % m))
 993                         printk("\b\b\b\b%3d%%", nr_pages / m);
 994                 nr_pages++;
 995         }
 996         if (!error)
 997                 printk("\b\b\b\bdone\n");
 998         return error;
 999 }
1000
1001 /**
1002  *      unpack_orig_addresses - copy the elements of @buf[] (1 page) to
1003  *      the PBEs in the list starting at @pbe
1004  */
1005
1006 static inline struct pbe *unpack_orig_addresses(unsigned long *buf,
1007                                                 struct pbe *pbe)
1008 {
1009         int j;
1010
1011         for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
1012                 pbe->orig_address = buf[j];
1013                 pbe = pbe->next;
1014         }
1015         return pbe;
1016 }
1017
1018 /**
1019  *      load_image_metadata - load the image metadata using the swap map
1020  *      handle @handle and put them into the PBEs in the list @pblist
1021  */
1022
1023 static int load_image_metadata(struct pbe *pblist, struct swap_map_handle *handle)
1024 {
1025         struct pbe *p;
1026         unsigned long *buf;
1027         unsigned int n = 0;
1028         int error = 0;
1029
1030         printk("Loading image metadata ... ");
1031         buf = (unsigned long *)get_zeroed_page(GFP_ATOMIC);
1032         if (!buf)
1033                 return -ENOMEM;
1034         p = pblist;
1035         while (p) {
1036                 error = swap_map_read_page(handle, buf);
1037                 if (error)
1038                         break;
1039                 p = unpack_orig_addresses(buf, p);
1040                 n++;
1041         }
1042         free_page((unsigned long)buf);
1043         if (!error)
1044                 printk("done (%u pages loaded)\n", n);
1045         return error;
1046 }
1047
1048 static int check_suspend_image(void)
1049 {
1050         int error = 0;
1051
1052         if ((error = check_sig()))
1053                 return error;
1054
1055         if ((error = check_header()))
1056                 return error;
1057
1058         return 0;
1059 }
1060
1061 static int read_suspend_image(struct pbe **pblist_ptr)
1062 {
1063         int error = 0;
1064         struct pbe *p, *pblist;
1065         struct swap_map_handle handle;
1066         unsigned int nr_pages = swsusp_info.image_pages;
1067
1068         p = alloc_pagedir(nr_pages, GFP_ATOMIC, 0);
1069         if (!p)
1070                 return -ENOMEM;
1071         error = get_swap_map_reader(&handle, swsusp_info.start);
1072         if (error)
1073                 /* The PBE list at p will be released by swsusp_free() */
1074                 return error;
1075         error = load_image_metadata(p, &handle);
1076         if (!error) {
1077                 mark_unsafe_pages(p);
1078                 pblist = alloc_pagedir(nr_pages, GFP_ATOMIC, 1);
1079                 if (pblist)
1080                         copy_page_backup_list(pblist, p);
1081                 free_pagedir(p);
1082                 if (!pblist)
1083                         error = -ENOMEM;
1084
1085                 /* Allocate memory for the image and read the data from swap */
1086                 if (!error)
1087                         error = alloc_data_pages(pblist, GFP_ATOMIC, 1);
1088                 if (!error) {
1089                         release_eaten_pages();
1090                         error = load_image_data(pblist, &handle, nr_pages);
1091                 }
1092                 if (!error)
1093                         *pblist_ptr = pblist;
1094         }
1095         release_swap_map_reader(&handle);
1096         return error;
1097 }
1098
1099 /**
1100  *      swsusp_check - Check for saved image in swap
1101  */
1102
1103 int swsusp_check(void)
1104 {
1105         int error;
1106
1107         resume_bdev = open_by_devnum(swsusp_resume_device, FMODE_READ);
1108         if (!IS_ERR(resume_bdev)) {
1109                 set_blocksize(resume_bdev, PAGE_SIZE);
1110                 error = check_suspend_image();
1111                 if (error)
1112                     blkdev_put(resume_bdev);
1113         } else
1114                 error = PTR_ERR(resume_bdev);
1115
1116         if (!error)
1117                 pr_debug("swsusp: resume file found\n");
1118         else
1119                 pr_debug("swsusp: Error %d check for resume file\n", error);
1120         return error;
1121 }
1122
1123 /**
1124  *      swsusp_read - Read saved image from swap.
1125  */
1126
1127 int swsusp_read(struct pbe **pblist_ptr)
1128 {
1129         int error;
1130
1131         if (IS_ERR(resume_bdev)) {
1132                 pr_debug("swsusp: block device not initialised\n");
1133                 return PTR_ERR(resume_bdev);
1134         }
1135
1136         error = read_suspend_image(pblist_ptr);
1137         blkdev_put(resume_bdev);
1138
1139         if (!error)
1140                 pr_debug("swsusp: Reading resume file was successful\n");
1141         else
1142                 pr_debug("swsusp: Error %d resuming\n", error);
1143         return error;
1144 }
1145
1146 /**
1147  *      swsusp_close - close swap device.
1148  */
1149
1150 void swsusp_close(void)
1151 {
1152         if (IS_ERR(resume_bdev)) {
1153                 pr_debug("swsusp: block device not initialised\n");
1154                 return;
1155         }
1156
1157         blkdev_put(resume_bdev);
1158 }