err.no Git - linux-2.6/blob - kernel/power/swsusp.c

   1 /*
   2  * linux/kernel/power/swsusp.c
   3  *
   4  * This file is to realize architecture-independent
   5  * machine suspend feature using pretty near only high-level routines
   6  *
   7  * Copyright (C) 1998-2001 Gabor Kuti <seasons@fornax.hu>
   8  * Copyright (C) 1998,2001-2004 Pavel Machek <pavel@suse.cz>
   9  *
  10  * This file is released under the GPLv2.
  11  *
  12  * I'd like to thank the following people for their work:
  13  *
  14  * Pavel Machek <pavel@ucw.cz>:
  15  * Modifications, defectiveness pointing, being with me at the very beginning,
  16  * suspend to swap space, stop all tasks. Port to 2.4.18-ac and 2.5.17.
  17  *
  18  * Steve Doddi <dirk@loth.demon.co.uk>:
  19  * Support the possibility of hardware state restoring.
  20  *
  21  * Raph <grey.havens@earthling.net>:
  22  * Support for preserving states of network devices and virtual console
  23  * (including X and svgatextmode)
  24  *
  25  * Kurt Garloff <garloff@suse.de>:
  26  * Straightened the critical function in order to prevent compilers from
  27  * playing tricks with local variables.
  28  *
  29  * Andreas Mohr <a.mohr@mailto.de>
  30  *
  31  * Alex Badea <vampire@go.ro>:
  32  * Fixed runaway init
  33  *
  34  * More state savers are welcome. Especially for the scsi layer...
  35  *
  36  * For TODOs,FIXMEs also look in Documentation/power/swsusp.txt
  37  */
  38
  39 #include <linux/module.h>
  40 #include <linux/mm.h>
  41 #include <linux/suspend.h>
  42 #include <linux/smp_lock.h>
  43 #include <linux/file.h>
  44 #include <linux/utsname.h>
  45 #include <linux/version.h>
  46 #include <linux/delay.h>
  47 #include <linux/reboot.h>
  48 #include <linux/bitops.h>
  49 #include <linux/vt_kern.h>
  50 #include <linux/kbd_kern.h>
  51 #include <linux/keyboard.h>
  52 #include <linux/spinlock.h>
  53 #include <linux/genhd.h>
  54 #include <linux/kernel.h>
  55 #include <linux/major.h>
  56 #include <linux/swap.h>
  57 #include <linux/pm.h>
  58 #include <linux/device.h>
  59 #include <linux/buffer_head.h>
  60 #include <linux/swapops.h>
  61 #include <linux/bootmem.h>
  62 #include <linux/syscalls.h>
  63 #include <linux/console.h>
  64 #include <linux/highmem.h>
  65 #include <linux/bio.h>
  66 #include <linux/mount.h>
  67
  68 #include <asm/uaccess.h>
  69 #include <asm/mmu_context.h>
  70 #include <asm/pgtable.h>
  71 #include <asm/tlbflush.h>
  72 #include <asm/io.h>
  73
  74 #include "power.h"
  75
  76 /* References to section boundaries */
  77 extern const void __nosave_begin, __nosave_end;
  78
  79 /* Variables to be preserved over suspend */
  80 static int nr_copy_pages_check;
  81
  82 extern char resume_file[];
  83
  84 /* Local variables that should not be affected by save */
  85 static unsigned int nr_copy_pages __nosavedata = 0;
  86
  87 /* Suspend pagedir is allocated before final copy, therefore it
  88    must be freed after resume
  89
  90    Warning: this is evil. There are actually two pagedirs at time of
  91    resume. One is "pagedir_save", which is empty frame allocated at
  92    time of suspend, that must be freed. Second is "pagedir_nosave",
  93    allocated at time of resume, that travels through memory not to
  94    collide with anything.
  95
  96    Warning: this is even more evil than it seems. Pagedirs this file
  97    talks about are completely different from page directories used by
  98    MMU hardware.
  99  */
 100 suspend_pagedir_t *pagedir_nosave __nosavedata = NULL;
 101 static suspend_pagedir_t *pagedir_save;
 102
 103 #define SWSUSP_SIG      "S1SUSPEND"
 104
 105 static struct swsusp_header {
 106         char reserved[PAGE_SIZE - 20 - sizeof(swp_entry_t)];
 107         swp_entry_t swsusp_info;
 108         char    orig_sig[10];
 109         char    sig[10];
 110 } __attribute__((packed, aligned(PAGE_SIZE))) swsusp_header;
 111
 112 static struct swsusp_info swsusp_info;
 113
 114 /*
 115  * XXX: We try to keep some more pages free so that I/O operations succeed
 116  * without paging. Might this be more?
 117  */
 118 #define PAGES_FOR_IO    512
 119
 120 /*
 121  * Saving part...
 122  */
 123
 124 /* We memorize in swapfile_used what swap devices are used for suspension */
 125 #define SWAPFILE_UNUSED    0
 126 #define SWAPFILE_SUSPEND   1    /* This is the suspending device */
 127 #define SWAPFILE_IGNORED   2    /* Those are other swap devices ignored for suspension */
 128
 129 static unsigned short swapfile_used[MAX_SWAPFILES];
 130 static unsigned short root_swap;
 131
 132 static int mark_swapfiles(swp_entry_t prev)
 133 {
 134         int error;
 135
 136         rw_swap_page_sync(READ,
 137                           swp_entry(root_swap, 0),
 138                           virt_to_page((unsigned long)&swsusp_header));
 139         if (!memcmp("SWAP-SPACE",swsusp_header.sig, 10) ||
 140             !memcmp("SWAPSPACE2",swsusp_header.sig, 10)) {
 141                 memcpy(swsusp_header.orig_sig,swsusp_header.sig, 10);
 142                 memcpy(swsusp_header.sig,SWSUSP_SIG, 10);
 143                 swsusp_header.swsusp_info = prev;
 144                 error = rw_swap_page_sync(WRITE,
 145                                           swp_entry(root_swap, 0),
 146                                           virt_to_page((unsigned long)
 147                                                        &swsusp_header));
 148         } else {
 149                 pr_debug("swsusp: Partition is not swap space.\n");
 150                 error = -ENODEV;
 151         }
 152         return error;
 153 }
 154
 155 /*
 156  * Check whether the swap device is the specified resume
 157  * device, irrespective of whether they are specified by
 158  * identical names.
 159  *
 160  * (Thus, device inode aliasing is allowed.  You can say /dev/hda4
 161  * instead of /dev/ide/host0/bus0/target0/lun0/part4 [if using devfs]
 162  * and they'll be considered the same device.  This is *necessary* for
 163  * devfs, since the resume code can only recognize the form /dev/hda4,
 164  * but the suspend code would see the long name.)
 165  */
 166 static int is_resume_device(const struct swap_info_struct *swap_info)
 167 {
 168         struct file *file = swap_info->swap_file;
 169         struct inode *inode = file->f_dentry->d_inode;
 170
 171         return S_ISBLK(inode->i_mode) &&
 172                 swsusp_resume_device == MKDEV(imajor(inode), iminor(inode));
 173 }
 174
 175 static int swsusp_swap_check(void) /* This is called before saving image */
 176 {
 177         int i, len;
 178
 179         len=strlen(resume_file);
 180         root_swap = 0xFFFF;
 181
 182         spin_lock(&swap_lock);
 183         for (i=0; i<MAX_SWAPFILES; i++) {
 184                 if (!(swap_info[i].flags & SWP_WRITEOK)) {
 185                         swapfile_used[i]=SWAPFILE_UNUSED;
 186                 } else {
 187                         if (!len) {
 188                                 printk(KERN_WARNING "resume= option should be used to set suspend device" );
 189                                 if (root_swap == 0xFFFF) {
 190                                         swapfile_used[i] = SWAPFILE_SUSPEND;
 191                                         root_swap = i;
 192                                 } else
 193                                         swapfile_used[i] = SWAPFILE_IGNORED;
 194                         } else {
 195                                 /* we ignore all swap devices that are not the resume_file */
 196                                 if (is_resume_device(&swap_info[i])) {
 197                                         swapfile_used[i] = SWAPFILE_SUSPEND;
 198                                         root_swap = i;
 199                                 } else {
 200                                         swapfile_used[i] = SWAPFILE_IGNORED;
 201                                 }
 202                         }
 203                 }
 204         }
 205         spin_unlock(&swap_lock);
 206         return (root_swap != 0xffff) ? 0 : -ENODEV;
 207 }
 208
 209 /**
 210  * This is called after saving image so modification
 211  * will be lost after resume... and that's what we want.
 212  * we make the device unusable. A new call to
 213  * lock_swapdevices can unlock the devices.
 214  */
 215 static void lock_swapdevices(void)
 216 {
 217         int i;
 218
 219         spin_lock(&swap_lock);
 220         for (i = 0; i< MAX_SWAPFILES; i++)
 221                 if (swapfile_used[i] == SWAPFILE_IGNORED) {
 222                         swap_info[i].flags ^= SWP_WRITEOK;
 223                 }
 224         spin_unlock(&swap_lock);
 225 }
 226
 227 /**
 228  *      write_swap_page - Write one page to a fresh swap location.
 229  *      @addr:  Address we're writing.
 230  *      @loc:   Place to store the entry we used.
 231  *
 232  *      Allocate a new swap entry and 'sync' it. Note we discard -EIO
 233  *      errors. That is an artifact left over from swsusp. It did not
 234  *      check the return of rw_swap_page_sync() at all, since most pages
 235  *      written back to swap would return -EIO.
 236  *      This is a partial improvement, since we will at least return other
 237  *      errors, though we need to eventually fix the damn code.
 238  */
 239 static int write_page(unsigned long addr, swp_entry_t * loc)
 240 {
 241         swp_entry_t entry;
 242         int error = 0;
 243
 244         entry = get_swap_page();
 245         if (swp_offset(entry) &&
 246             swapfile_used[swp_type(entry)] == SWAPFILE_SUSPEND) {
 247                 error = rw_swap_page_sync(WRITE, entry,
 248                                           virt_to_page(addr));
 249                 if (error == -EIO)
 250                         error = 0;
 251                 if (!error)
 252                         *loc = entry;
 253         } else
 254                 error = -ENOSPC;
 255         return error;
 256 }
 257
 258 /**
 259  *      data_free - Free the swap entries used by the saved image.
 260  *
 261  *      Walk the list of used swap entries and free each one.
 262  *      This is only used for cleanup when suspend fails.
 263  */
 264 static void data_free(void)
 265 {
 266         swp_entry_t entry;
 267         int i;
 268
 269         for (i = 0; i < nr_copy_pages; i++) {
 270                 entry = (pagedir_nosave + i)->swap_address;
 271                 if (entry.val)
 272                         swap_free(entry);
 273                 else
 274                         break;
 275                 (pagedir_nosave + i)->swap_address = (swp_entry_t){0};
 276         }
 277 }
 278
 279 /**
 280  *      data_write - Write saved image to swap.
 281  *
 282  *      Walk the list of pages in the image and sync each one to swap.
 283  */
 284 static int data_write(void)
 285 {
 286         int error = 0, i = 0;
 287         unsigned int mod = nr_copy_pages / 100;
 288         struct pbe *p;
 289
 290         if (!mod)
 291                 mod = 1;
 292
 293         printk( "Writing data to swap (%d pages)...     ", nr_copy_pages );
 294         for_each_pbe (p, pagedir_nosave) {
 295                 if (!(i%mod))
 296                         printk( "\b\b\b\b%3d%%", i / mod );
 297                 if ((error = write_page(p->address, &(p->swap_address))))
 298                         return error;
 299                 i++;
 300         }
 301         printk("\b\b\b\bdone\n");
 302         return error;
 303 }
 304
 305 static void dump_info(void)
 306 {
 307         pr_debug(" swsusp: Version: %u\n",swsusp_info.version_code);
 308         pr_debug(" swsusp: Num Pages: %ld\n",swsusp_info.num_physpages);
 309         pr_debug(" swsusp: UTS Sys: %s\n",swsusp_info.uts.sysname);
 310         pr_debug(" swsusp: UTS Node: %s\n",swsusp_info.uts.nodename);
 311         pr_debug(" swsusp: UTS Release: %s\n",swsusp_info.uts.release);
 312         pr_debug(" swsusp: UTS Version: %s\n",swsusp_info.uts.version);
 313         pr_debug(" swsusp: UTS Machine: %s\n",swsusp_info.uts.machine);
 314         pr_debug(" swsusp: UTS Domain: %s\n",swsusp_info.uts.domainname);
 315         pr_debug(" swsusp: CPUs: %d\n",swsusp_info.cpus);
 316         pr_debug(" swsusp: Image: %ld Pages\n",swsusp_info.image_pages);
 317         pr_debug(" swsusp: Pagedir: %ld Pages\n",swsusp_info.pagedir_pages);
 318 }
 319
 320 static void init_header(void)
 321 {
 322         memset(&swsusp_info, 0, sizeof(swsusp_info));
 323         swsusp_info.version_code = LINUX_VERSION_CODE;
 324         swsusp_info.num_physpages = num_physpages;
 325         memcpy(&swsusp_info.uts, &system_utsname, sizeof(system_utsname));
 326
 327         swsusp_info.suspend_pagedir = pagedir_nosave;
 328         swsusp_info.cpus = num_online_cpus();
 329         swsusp_info.image_pages = nr_copy_pages;
 330 }
 331
 332 static int close_swap(void)
 333 {
 334         swp_entry_t entry;
 335         int error;
 336
 337         dump_info();
 338         error = write_page((unsigned long)&swsusp_info, &entry);
 339         if (!error) {
 340                 printk( "S" );
 341                 error = mark_swapfiles(entry);
 342                 printk( "|\n" );
 343         }
 344         return error;
 345 }
 346
 347 /**
 348  *      free_pagedir_entries - Free pages used by the page directory.
 349  *
 350  *      This is used during suspend for error recovery.
 351  */
 352
 353 static void free_pagedir_entries(void)
 354 {
 355         int i;
 356
 357         for (i = 0; i < swsusp_info.pagedir_pages; i++)
 358                 swap_free(swsusp_info.pagedir[i]);
 359 }
 360
 361
 362 /**
 363  *      write_pagedir - Write the array of pages holding the page directory.
 364  *      @last:  Last swap entry we write (needed for header).
 365  */
 366
 367 static int write_pagedir(void)
 368 {
 369         int error = 0;
 370         unsigned n = 0;
 371         struct pbe * pbe;
 372
 373         printk( "Writing pagedir...");
 374         for_each_pb_page (pbe, pagedir_nosave) {
 375                 if ((error = write_page((unsigned long)pbe, &swsusp_info.pagedir[n++])))
 376                         return error;
 377         }
 378
 379         swsusp_info.pagedir_pages = n;
 380         printk("done (%u pages)\n", n);
 381         return error;
 382 }
 383
 384 /**
 385  *      write_suspend_image - Write entire image and metadata.
 386  *
 387  */
 388
 389 static int write_suspend_image(void)
 390 {
 391         int error;
 392
 393         init_header();
 394         if ((error = data_write()))
 395                 goto FreeData;
 396
 397         if ((error = write_pagedir()))
 398                 goto FreePagedir;
 399
 400         if ((error = close_swap()))
 401                 goto FreePagedir;
 402  Done:
 403         return error;
 404  FreePagedir:
 405         free_pagedir_entries();
 406  FreeData:
 407         data_free();
 408         goto Done;
 409 }
 410
 411
 412 #ifdef CONFIG_HIGHMEM
 413 struct highmem_page {
 414         char *data;
 415         struct page *page;
 416         struct highmem_page *next;
 417 };
 418
 419 static struct highmem_page *highmem_copy;
 420
 421 static int save_highmem_zone(struct zone *zone)
 422 {
 423         unsigned long zone_pfn;
 424         mark_free_pages(zone);
 425         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
 426                 struct page *page;
 427                 struct highmem_page *save;
 428                 void *kaddr;
 429                 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
 430
 431                 if (!(pfn%1000))
 432                         printk(".");
 433                 if (!pfn_valid(pfn))
 434                         continue;
 435                 page = pfn_to_page(pfn);
 436                 /*
 437                  * This condition results from rvmalloc() sans vmalloc_32()
 438                  * and architectural memory reservations. This should be
 439                  * corrected eventually when the cases giving rise to this
 440                  * are better understood.
 441                  */
 442                 if (PageReserved(page)) {
 443                         printk("highmem reserved page?!\n");
 444                         continue;
 445                 }
 446                 BUG_ON(PageNosave(page));
 447                 if (PageNosaveFree(page))
 448                         continue;
 449                 save = kmalloc(sizeof(struct highmem_page), GFP_ATOMIC);
 450                 if (!save)
 451                         return -ENOMEM;
 452                 save->next = highmem_copy;
 453                 save->page = page;
 454                 save->data = (void *) get_zeroed_page(GFP_ATOMIC);
 455                 if (!save->data) {
 456                         kfree(save);
 457                         return -ENOMEM;
 458                 }
 459                 kaddr = kmap_atomic(page, KM_USER0);
 460                 memcpy(save->data, kaddr, PAGE_SIZE);
 461                 kunmap_atomic(kaddr, KM_USER0);
 462                 highmem_copy = save;
 463         }
 464         return 0;
 465 }
 466 #endif /* CONFIG_HIGHMEM */
 467
 468
 469 static int save_highmem(void)
 470 {
 471 #ifdef CONFIG_HIGHMEM
 472         struct zone *zone;
 473         int res = 0;
 474
 475         pr_debug("swsusp: Saving Highmem\n");
 476         for_each_zone (zone) {
 477                 if (is_highmem(zone))
 478                         res = save_highmem_zone(zone);
 479                 if (res)
 480                         return res;
 481         }
 482 #endif
 483         return 0;
 484 }
 485
 486 static int restore_highmem(void)
 487 {
 488 #ifdef CONFIG_HIGHMEM
 489         printk("swsusp: Restoring Highmem\n");
 490         while (highmem_copy) {
 491                 struct highmem_page *save = highmem_copy;
 492                 void *kaddr;
 493                 highmem_copy = save->next;
 494
 495                 kaddr = kmap_atomic(save->page, KM_USER0);
 496                 memcpy(kaddr, save->data, PAGE_SIZE);
 497                 kunmap_atomic(kaddr, KM_USER0);
 498                 free_page((long) save->data);
 499                 kfree(save);
 500         }
 501 #endif
 502         return 0;
 503 }
 504
 505
 506 static int pfn_is_nosave(unsigned long pfn)
 507 {
 508         unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
 509         unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
 510         return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
 511 }
 512
 513 /**
 514  *      saveable - Determine whether a page should be cloned or not.
 515  *      @pfn:   The page
 516  *
 517  *      We save a page if it's Reserved, and not in the range of pages
 518  *      statically defined as 'unsaveable', or if it isn't reserved, and
 519  *      isn't part of a free chunk of pages.
 520  */
 521
 522 static int saveable(struct zone * zone, unsigned long * zone_pfn)
 523 {
 524         unsigned long pfn = *zone_pfn + zone->zone_start_pfn;
 525         struct page * page;
 526
 527         if (!pfn_valid(pfn))
 528                 return 0;
 529
 530         page = pfn_to_page(pfn);
 531         BUG_ON(PageReserved(page) && PageNosave(page));
 532         if (PageNosave(page))
 533                 return 0;
 534         if (PageReserved(page) && pfn_is_nosave(pfn)) {
 535                 pr_debug("[nosave pfn 0x%lx]", pfn);
 536                 return 0;
 537         }
 538         if (PageNosaveFree(page))
 539                 return 0;
 540
 541         return 1;
 542 }
 543
 544 static void count_data_pages(void)
 545 {
 546         struct zone *zone;
 547         unsigned long zone_pfn;
 548
 549         nr_copy_pages = 0;
 550
 551         for_each_zone (zone) {
 552                 if (is_highmem(zone))
 553                         continue;
 554                 mark_free_pages(zone);
 555                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 556                         nr_copy_pages += saveable(zone, &zone_pfn);
 557         }
 558 }
 559
 560
 561 static void copy_data_pages(void)
 562 {
 563         struct zone *zone;
 564         unsigned long zone_pfn;
 565         struct pbe * pbe = pagedir_nosave;
 566
 567         pr_debug("copy_data_pages(): pages to copy: %d\n", nr_copy_pages);
 568         for_each_zone (zone) {
 569                 if (is_highmem(zone))
 570                         continue;
 571                 mark_free_pages(zone);
 572                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
 573                         if (saveable(zone, &zone_pfn)) {
 574                                 struct page * page;
 575                                 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
 576                                 BUG_ON(!pbe);
 577                                 pbe->orig_address = (long) page_address(page);
 578                                 /* copy_page is not usable for copying task structs. */
 579                                 memcpy((void *)pbe->address, (void *)pbe->orig_address, PAGE_SIZE);
 580                                 pbe = pbe->next;
 581                         }
 582                 }
 583         }
 584         BUG_ON(pbe);
 585 }
 586
 587
 588 /**
 589  *      calc_nr - Determine the number of pages needed for a pbe list.
 590  */
 591
 592 static int calc_nr(int nr_copy)
 593 {
 594         int extra = 0;
 595         int mod = !!(nr_copy % PBES_PER_PAGE);
 596         int diff = (nr_copy / PBES_PER_PAGE) + mod;
 597
 598         do {
 599                 extra += diff;
 600                 nr_copy += diff;
 601                 mod = !!(nr_copy % PBES_PER_PAGE);
 602                 diff = (nr_copy / PBES_PER_PAGE) + mod - extra;
 603         } while (diff > 0);
 604
 605         return nr_copy;
 606 }
 607
 608 /**
 609  *      free_pagedir - free pages allocated with alloc_pagedir()
 610  */
 611
 612 static inline void free_pagedir(struct pbe *pblist)
 613 {
 614         struct pbe *pbe;
 615
 616         while (pblist) {
 617                 pbe = (pblist + PB_PAGE_SKIP)->next;
 618                 free_page((unsigned long)pblist);
 619                 pblist = pbe;
 620         }
 621 }
 622
 623 /**
 624  *      fill_pb_page - Create a list of PBEs on a given memory page
 625  */
 626
 627 static inline void fill_pb_page(struct pbe *pbpage)
 628 {
 629         struct pbe *p;
 630
 631         p = pbpage;
 632         pbpage += PB_PAGE_SKIP;
 633         do
 634                 p->next = p + 1;
 635         while (++p < pbpage);
 636 }
 637
 638 /**
 639  *      create_pbe_list - Create a list of PBEs on top of a given chain
 640  *      of memory pages allocated with alloc_pagedir()
 641  */
 642
 643 static void create_pbe_list(struct pbe *pblist, unsigned nr_pages)
 644 {
 645         struct pbe *pbpage, *p;
 646         unsigned num = PBES_PER_PAGE;
 647
 648         for_each_pb_page (pbpage, pblist) {
 649                 if (num >= nr_pages)
 650                         break;
 651
 652                 fill_pb_page(pbpage);
 653                 num += PBES_PER_PAGE;
 654         }
 655         if (pbpage) {
 656                 for (num -= PBES_PER_PAGE - 1, p = pbpage; num < nr_pages; p++, num++)
 657                         p->next = p + 1;
 658                 p->next = NULL;
 659         }
 660         pr_debug("create_pbe_list(): initialized %d PBEs\n", num);
 661 }
 662
 663 /**
 664  *      alloc_pagedir - Allocate the page directory.
 665  *
 666  *      First, determine exactly how many pages we need and
 667  *      allocate them.
 668  *
 669  *      We arrange the pages in a chain: each page is an array of PBES_PER_PAGE
 670  *      struct pbe elements (pbes) and the last element in the page points
 671  *      to the next page.
 672  *
 673  *      On each page we set up a list of struct_pbe elements.
 674  */
 675
 676 static struct pbe * alloc_pagedir(unsigned nr_pages)
 677 {
 678         unsigned num;
 679         struct pbe *pblist, *pbe;
 680
 681         if (!nr_pages)
 682                 return NULL;
 683
 684         pr_debug("alloc_pagedir(): nr_pages = %d\n", nr_pages);
 685         pblist = (struct pbe *)get_zeroed_page(GFP_ATOMIC | __GFP_COLD);
 686         for (pbe = pblist, num = PBES_PER_PAGE; pbe && num < nr_pages;
 687                         pbe = pbe->next, num += PBES_PER_PAGE) {
 688                 pbe += PB_PAGE_SKIP;
 689                 pbe->next = (struct pbe *)get_zeroed_page(GFP_ATOMIC | __GFP_COLD);
 690         }
 691         if (!pbe) { /* get_zeroed_page() failed */
 692                 free_pagedir(pblist);
 693                 pblist = NULL;
 694         }
 695         return pblist;
 696 }
 697
 698 /**
 699  *      free_image_pages - Free pages allocated for snapshot
 700  */
 701
 702 static void free_image_pages(void)
 703 {
 704         struct pbe * p;
 705
 706         for_each_pbe (p, pagedir_save) {
 707                 if (p->address) {
 708                         ClearPageNosave(virt_to_page(p->address));
 709                         free_page(p->address);
 710                         p->address = 0;
 711                 }
 712         }
 713 }
 714
 715 /**
 716  *      alloc_image_pages - Allocate pages for the snapshot.
 717  */
 718
 719 static int alloc_image_pages(void)
 720 {
 721         struct pbe * p;
 722
 723         for_each_pbe (p, pagedir_save) {
 724                 p->address = get_zeroed_page(GFP_ATOMIC | __GFP_COLD);
 725                 if (!p->address)
 726                         return -ENOMEM;
 727                 SetPageNosave(virt_to_page(p->address));
 728         }
 729         return 0;
 730 }
 731
 732 void swsusp_free(void)
 733 {
 734         BUG_ON(PageNosave(virt_to_page(pagedir_save)));
 735         BUG_ON(PageNosaveFree(virt_to_page(pagedir_save)));
 736         free_image_pages();
 737         free_pagedir(pagedir_save);
 738 }
 739
 740
 741 /**
 742  *      enough_free_mem - Make sure we enough free memory to snapshot.
 743  *
 744  *      Returns TRUE or FALSE after checking the number of available
 745  *      free pages.
 746  */
 747
 748 static int enough_free_mem(void)
 749 {
 750         if (nr_free_pages() < (nr_copy_pages + PAGES_FOR_IO)) {
 751                 pr_debug("swsusp: Not enough free pages: Have %d\n",
 752                          nr_free_pages());
 753                 return 0;
 754         }
 755         return 1;
 756 }
 757
 758
 759 /**
 760  *      enough_swap - Make sure we have enough swap to save the image.
 761  *
 762  *      Returns TRUE or FALSE after checking the total amount of swap
 763  *      space avaiable.
 764  *
 765  *      FIXME: si_swapinfo(&i) returns all swap devices information.
 766  *      We should only consider resume_device.
 767  */
 768
 769 static int enough_swap(void)
 770 {
 771         struct sysinfo i;
 772
 773         si_swapinfo(&i);
 774         if (i.freeswap < (nr_copy_pages + PAGES_FOR_IO))  {
 775                 pr_debug("swsusp: Not enough swap. Need %ld\n",i.freeswap);
 776                 return 0;
 777         }
 778         return 1;
 779 }
 780
 781 static int swsusp_alloc(void)
 782 {
 783         int error;
 784
 785         pagedir_nosave = NULL;
 786         nr_copy_pages = calc_nr(nr_copy_pages);
 787
 788         pr_debug("suspend: (pages needed: %d + %d free: %d)\n",
 789                  nr_copy_pages, PAGES_FOR_IO, nr_free_pages());
 790
 791         if (!enough_free_mem())
 792                 return -ENOMEM;
 793
 794         if (!enough_swap())
 795                 return -ENOSPC;
 796
 797         if (!(pagedir_save = alloc_pagedir(nr_copy_pages))) {
 798                 printk(KERN_ERR "suspend: Allocating pagedir failed.\n");
 799                 return -ENOMEM;
 800         }
 801         create_pbe_list(pagedir_save, nr_copy_pages);
 802         pagedir_nosave = pagedir_save;
 803         if ((error = alloc_image_pages())) {
 804                 printk(KERN_ERR "suspend: Allocating image pages failed.\n");
 805                 swsusp_free();
 806                 return error;
 807         }
 808
 809         nr_copy_pages_check = nr_copy_pages;
 810         return 0;
 811 }
 812
 813 static int suspend_prepare_image(void)
 814 {
 815         int error;
 816
 817         pr_debug("swsusp: critical section: \n");
 818         if (save_highmem()) {
 819                 printk(KERN_CRIT "Suspend machine: Not enough free pages for highmem\n");
 820                 restore_highmem();
 821                 return -ENOMEM;
 822         }
 823
 824         drain_local_pages();
 825         count_data_pages();
 826         printk("swsusp: Need to copy %u pages\n", nr_copy_pages);
 827
 828         error = swsusp_alloc();
 829         if (error)
 830                 return error;
 831
 832         /* During allocating of suspend pagedir, new cold pages may appear.
 833          * Kill them.
 834          */
 835         drain_local_pages();
 836         copy_data_pages();
 837
 838         /*
 839          * End of critical section. From now on, we can write to memory,
 840          * but we should not touch disk. This specially means we must _not_
 841          * touch swap space! Except we must write out our image of course.
 842          */
 843
 844         printk("swsusp: critical section/: done (%d pages copied)\n", nr_copy_pages );
 845         return 0;
 846 }
 847
 848
 849 /* It is important _NOT_ to umount filesystems at this point. We want
 850  * them synced (in case something goes wrong) but we DO not want to mark
 851  * filesystem clean: it is not. (And it does not matter, if we resume
 852  * correctly, we'll mark system clean, anyway.)
 853  */
 854 int swsusp_write(void)
 855 {
 856         int error;
 857         device_resume();
 858         lock_swapdevices();
 859         error = write_suspend_image();
 860         /* This will unlock ignored swap devices since writing is finished */
 861         lock_swapdevices();
 862         return error;
 863
 864 }
 865
 866
 867 extern asmlinkage int swsusp_arch_suspend(void);
 868 extern asmlinkage int swsusp_arch_resume(void);
 869
 870
 871 asmlinkage int swsusp_save(void)
 872 {
 873         return suspend_prepare_image();
 874 }
 875
 876 int swsusp_suspend(void)
 877 {
 878         int error;
 879         if ((error = arch_prepare_suspend()))
 880                 return error;
 881         local_irq_disable();
 882         /* At this point, device_suspend() has been called, but *not*
 883          * device_power_down(). We *must* device_power_down() now.
 884          * Otherwise, drivers for some devices (e.g. interrupt controllers)
 885          * become desynchronized with the actual state of the hardware
 886          * at resume time, and evil weirdness ensues.
 887          */
 888         if ((error = device_power_down(PMSG_FREEZE))) {
 889                 local_irq_enable();
 890                 return error;
 891         }
 892
 893         if ((error = swsusp_swap_check())) {
 894                 printk(KERN_ERR "swsusp: FATAL: cannot find swap device, try "
 895                                 "swapon -a!\n");
 896                 local_irq_enable();
 897                 return error;
 898         }
 899
 900         save_processor_state();
 901         if ((error = swsusp_arch_suspend()))
 902                 printk("Error %d suspending\n", error);
 903         /* Restore control flow magically appears here */
 904         restore_processor_state();
 905         BUG_ON (nr_copy_pages_check != nr_copy_pages);
 906         restore_highmem();
 907         device_power_up();
 908         local_irq_enable();
 909         return error;
 910 }
 911
 912 int swsusp_resume(void)
 913 {
 914         int error;
 915         local_irq_disable();
 916         if (device_power_down(PMSG_FREEZE))
 917                 printk(KERN_ERR "Some devices failed to power down, very bad\n");
 918         /* We'll ignore saved state, but this gets preempt count (etc) right */
 919         save_processor_state();
 920         error = swsusp_arch_resume();
 921         /* Code below is only ever reached in case of failure. Otherwise
 922          * execution continues at place where swsusp_arch_suspend was called
 923          */
 924         BUG_ON(!error);
 925         restore_processor_state();
 926         restore_highmem();
 927         device_power_up();
 928         local_irq_enable();
 929         return error;
 930 }
 931
 932 /**
 933  *      On resume, for storing the PBE list and the image,
 934  *      we can only use memory pages that do not conflict with the pages
 935  *      which had been used before suspend.
 936  *
 937  *      We don't know which pages are usable until we allocate them.
 938  *
 939  *      Allocated but unusable (ie eaten) memory pages are linked together
 940  *      to create a list, so that we can free them easily
 941  *
 942  *      We could have used a type other than (void *)
 943  *      for this purpose, but ...
 944  */
 945 static void **eaten_memory = NULL;
 946
 947 static inline void eat_page(void *page)
 948 {
 949         void **c;
 950
 951         c = eaten_memory;
 952         eaten_memory = page;
 953         *eaten_memory = c;
 954 }
 955
 956 static unsigned long get_usable_page(unsigned gfp_mask)
 957 {
 958         unsigned long m;
 959
 960         m = get_zeroed_page(gfp_mask);
 961         while (!PageNosaveFree(virt_to_page(m))) {
 962                 eat_page((void *)m);
 963                 m = get_zeroed_page(gfp_mask);
 964                 if (!m)
 965                         break;
 966         }
 967         return m;
 968 }
 969
 970 static void free_eaten_memory(void)
 971 {
 972         unsigned long m;
 973         void **c;
 974         int i = 0;
 975
 976         c = eaten_memory;
 977         while (c) {
 978                 m = (unsigned long)c;
 979                 c = *c;
 980                 free_page(m);
 981                 i++;
 982         }
 983         eaten_memory = NULL;
 984         pr_debug("swsusp: %d unused pages freed\n", i);
 985 }
 986
 987 /**
 988  *      check_pagedir - We ensure here that pages that the PBEs point to
 989  *      won't collide with pages where we're going to restore from the loaded
 990  *      pages later
 991  */
 992
 993 static int check_pagedir(struct pbe *pblist)
 994 {
 995         struct pbe *p;
 996
 997         /* This is necessary, so that we can free allocated pages
 998          * in case of failure
 999          */
1000         for_each_pbe (p, pblist)
1001                 p->address = 0UL;
1002
1003         for_each_pbe (p, pblist) {
1004                 p->address = get_usable_page(GFP_ATOMIC);
1005                 if (!p->address)
1006                         return -ENOMEM;
1007         }
1008         return 0;
1009 }
1010
1011 /**
1012  *      swsusp_pagedir_relocate - It is possible, that some memory pages
1013  *      occupied by the list of PBEs collide with pages where we're going to
1014  *      restore from the loaded pages later.  We relocate them here.
1015  */
1016
1017 static struct pbe * swsusp_pagedir_relocate(struct pbe *pblist)
1018 {
1019         struct zone *zone;
1020         unsigned long zone_pfn;
1021         struct pbe *pbpage, *tail, *p;
1022         void *m;
1023         int rel = 0, error = 0;
1024
1025         if (!pblist) /* a sanity check */
1026                 return NULL;
1027
1028         pr_debug("swsusp: Relocating pagedir (%lu pages to check)\n",
1029                         swsusp_info.pagedir_pages);
1030
1031         /* Set page flags */
1032
1033         for_each_zone (zone) {
1034                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
1035                         SetPageNosaveFree(pfn_to_page(zone_pfn +
1036                                         zone->zone_start_pfn));
1037         }
1038
1039         /* Clear orig addresses */
1040
1041         for_each_pbe (p, pblist)
1042                 ClearPageNosaveFree(virt_to_page(p->orig_address));
1043
1044         tail = pblist + PB_PAGE_SKIP;
1045
1046         /* Relocate colliding pages */
1047
1048         for_each_pb_page (pbpage, pblist) {
1049                 if (!PageNosaveFree(virt_to_page((unsigned long)pbpage))) {
1050                         m = (void *)get_usable_page(GFP_ATOMIC | __GFP_COLD);
1051                         if (!m) {
1052                                 error = -ENOMEM;
1053                                 break;
1054                         }
1055                         memcpy(m, (void *)pbpage, PAGE_SIZE);
1056                         if (pbpage == pblist)
1057                                 pblist = (struct pbe *)m;
1058                         else
1059                                 tail->next = (struct pbe *)m;
1060
1061                         eat_page((void *)pbpage);
1062                         pbpage = (struct pbe *)m;
1063
1064                         /* We have to link the PBEs again */
1065
1066                         for (p = pbpage; p < pbpage + PB_PAGE_SKIP; p++)
1067                                 if (p->next) /* needed to save the end */
1068                                         p->next = p + 1;
1069
1070                         rel++;
1071                 }
1072                 tail = pbpage + PB_PAGE_SKIP;
1073         }
1074
1075         if (error) {
1076                 printk("\nswsusp: Out of memory\n\n");
1077                 free_pagedir(pblist);
1078                 free_eaten_memory();
1079                 pblist = NULL;
1080         }
1081         else
1082                 printk("swsusp: Relocated %d pages\n", rel);
1083
1084         return pblist;
1085 }
1086
1087 /*
1088  *      Using bio to read from swap.
1089  *      This code requires a bit more work than just using buffer heads
1090  *      but, it is the recommended way for 2.5/2.6.
1091  *      The following are to signal the beginning and end of I/O. Bios
1092  *      finish asynchronously, while we want them to happen synchronously.
1093  *      A simple atomic_t, and a wait loop take care of this problem.
1094  */
1095
1096 static atomic_t io_done = ATOMIC_INIT(0);
1097
1098 static int end_io(struct bio * bio, unsigned int num, int err)
1099 {
1100         if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1101                 panic("I/O error reading memory image");
1102         atomic_set(&io_done, 0);
1103         return 0;
1104 }
1105
1106 static struct block_device * resume_bdev;
1107
1108 /**
1109  *      submit - submit BIO request.
1110  *      @rw:    READ or WRITE.
1111  *      @off    physical offset of page.
1112  *      @page:  page we're reading or writing.
1113  *
1114  *      Straight from the textbook - allocate and initialize the bio.
1115  *      If we're writing, make sure the page is marked as dirty.
1116  *      Then submit it and wait.
1117  */
1118
1119 static int submit(int rw, pgoff_t page_off, void * page)
1120 {
1121         int error = 0;
1122         struct bio * bio;
1123
1124         bio = bio_alloc(GFP_ATOMIC, 1);
1125         if (!bio)
1126                 return -ENOMEM;
1127         bio->bi_sector = page_off * (PAGE_SIZE >> 9);
1128         bio_get(bio);
1129         bio->bi_bdev = resume_bdev;
1130         bio->bi_end_io = end_io;
1131
1132         if (bio_add_page(bio, virt_to_page(page), PAGE_SIZE, 0) < PAGE_SIZE) {
1133                 printk("swsusp: ERROR: adding page to bio at %ld\n",page_off);
1134                 error = -EFAULT;
1135                 goto Done;
1136         }
1137
1138         if (rw == WRITE)
1139                 bio_set_pages_dirty(bio);
1140
1141         atomic_set(&io_done, 1);
1142         submit_bio(rw | (1 << BIO_RW_SYNC), bio);
1143         while (atomic_read(&io_done))
1144                 yield();
1145
1146  Done:
1147         bio_put(bio);
1148         return error;
1149 }
1150
1151 static int bio_read_page(pgoff_t page_off, void * page)
1152 {
1153         return submit(READ, page_off, page);
1154 }
1155
1156 static int bio_write_page(pgoff_t page_off, void * page)
1157 {
1158         return submit(WRITE, page_off, page);
1159 }
1160
1161 /*
1162  * Sanity check if this image makes sense with this kernel/swap context
1163  * I really don't think that it's foolproof but more than nothing..
1164  */
1165
1166 static const char * sanity_check(void)
1167 {
1168         dump_info();
1169         if (swsusp_info.version_code != LINUX_VERSION_CODE)
1170                 return "kernel version";
1171         if (swsusp_info.num_physpages != num_physpages)
1172                 return "memory size";
1173         if (strcmp(swsusp_info.uts.sysname,system_utsname.sysname))
1174                 return "system type";
1175         if (strcmp(swsusp_info.uts.release,system_utsname.release))
1176                 return "kernel release";
1177         if (strcmp(swsusp_info.uts.version,system_utsname.version))
1178                 return "version";
1179         if (strcmp(swsusp_info.uts.machine,system_utsname.machine))
1180                 return "machine";
1181 #if 0
1182         if(swsusp_info.cpus != num_online_cpus())
1183                 return "number of cpus";
1184 #endif
1185         return NULL;
1186 }
1187
1188
1189 static int check_header(void)
1190 {
1191         const char * reason = NULL;
1192         int error;
1193
1194         if ((error = bio_read_page(swp_offset(swsusp_header.swsusp_info), &swsusp_info)))
1195                 return error;
1196
1197         /* Is this same machine? */
1198         if ((reason = sanity_check())) {
1199                 printk(KERN_ERR "swsusp: Resume mismatch: %s\n",reason);
1200                 return -EPERM;
1201         }
1202         nr_copy_pages = swsusp_info.image_pages;
1203         return error;
1204 }
1205
1206 static int check_sig(void)
1207 {
1208         int error;
1209
1210         memset(&swsusp_header, 0, sizeof(swsusp_header));
1211         if ((error = bio_read_page(0, &swsusp_header)))
1212                 return error;
1213         if (!memcmp(SWSUSP_SIG, swsusp_header.sig, 10)) {
1214                 memcpy(swsusp_header.sig, swsusp_header.orig_sig, 10);
1215
1216                 /*
1217                  * Reset swap signature now.
1218                  */
1219                 error = bio_write_page(0, &swsusp_header);
1220         } else {
1221                 printk(KERN_ERR "swsusp: Suspend partition has wrong signature?\n");
1222                 return -EINVAL;
1223         }
1224         if (!error)
1225                 pr_debug("swsusp: Signature found, resuming\n");
1226         return error;
1227 }
1228
1229 /**
1230  *      data_read - Read image pages from swap.
1231  *
1232  *      You do not need to check for overlaps, check_pagedir()
1233  *      already did that.
1234  */
1235
1236 static int data_read(struct pbe *pblist)
1237 {
1238         struct pbe * p;
1239         int error = 0;
1240         int i = 0;
1241         int mod = swsusp_info.image_pages / 100;
1242
1243         if (!mod)
1244                 mod = 1;
1245
1246         printk("swsusp: Reading image data (%lu pages):     ",
1247                         swsusp_info.image_pages);
1248
1249         for_each_pbe (p, pblist) {
1250                 if (!(i % mod))
1251                         printk("\b\b\b\b%3d%%", i / mod);
1252
1253                 error = bio_read_page(swp_offset(p->swap_address),
1254                                   (void *)p->address);
1255                 if (error)
1256                         return error;
1257
1258                 i++;
1259         }
1260         printk("\b\b\b\bdone\n");
1261         return error;
1262 }
1263
1264 /**
1265  *      read_pagedir - Read page backup list pages from swap
1266  */
1267
1268 static int read_pagedir(struct pbe *pblist)
1269 {
1270         struct pbe *pbpage, *p;
1271         unsigned i = 0;
1272         int error;
1273
1274         if (!pblist)
1275                 return -EFAULT;
1276
1277         printk("swsusp: Reading pagedir (%lu pages)\n",
1278                         swsusp_info.pagedir_pages);
1279
1280         for_each_pb_page (pbpage, pblist) {
1281                 unsigned long offset = swp_offset(swsusp_info.pagedir[i++]);
1282
1283                 error = -EFAULT;
1284                 if (offset) {
1285                         p = (pbpage + PB_PAGE_SKIP)->next;
1286                         error = bio_read_page(offset, (void *)pbpage);
1287                         (pbpage + PB_PAGE_SKIP)->next = p;
1288                 }
1289                 if (error)
1290                         break;
1291         }
1292
1293         if (error)
1294                 free_page((unsigned long)pblist);
1295
1296         BUG_ON(i != swsusp_info.pagedir_pages);
1297
1298         return error;
1299 }
1300
1301
1302 static int check_suspend_image(void)
1303 {
1304         int error = 0;
1305
1306         if ((error = check_sig()))
1307                 return error;
1308
1309         if ((error = check_header()))
1310                 return error;
1311
1312         return 0;
1313 }
1314
1315 static int read_suspend_image(void)
1316 {
1317         int error = 0;
1318         struct pbe *p;
1319
1320         if (!(p = alloc_pagedir(nr_copy_pages)))
1321                 return -ENOMEM;
1322
1323         if ((error = read_pagedir(p)))
1324                 return error;
1325
1326         create_pbe_list(p, nr_copy_pages);
1327
1328         if (!(pagedir_nosave = swsusp_pagedir_relocate(p)))
1329                 return -ENOMEM;
1330
1331         /* Allocate memory for the image and read the data from swap */
1332
1333         error = check_pagedir(pagedir_nosave);
1334         free_eaten_memory();
1335         if (!error)
1336                 error = data_read(pagedir_nosave);
1337
1338         if (error) { /* We fail cleanly */
1339                 for_each_pbe (p, pagedir_nosave)
1340                         if (p->address) {
1341                                 free_page(p->address);
1342                                 p->address = 0UL;
1343                         }
1344                 free_pagedir(pagedir_nosave);
1345         }
1346         return error;
1347 }
1348
1349 /**
1350  *      swsusp_check - Check for saved image in swap
1351  */
1352
1353 int swsusp_check(void)
1354 {
1355         int error;
1356
1357         resume_bdev = open_by_devnum(swsusp_resume_device, FMODE_READ);
1358         if (!IS_ERR(resume_bdev)) {
1359                 set_blocksize(resume_bdev, PAGE_SIZE);
1360                 error = check_suspend_image();
1361                 if (error)
1362                     blkdev_put(resume_bdev);
1363         } else
1364                 error = PTR_ERR(resume_bdev);
1365
1366         if (!error)
1367                 pr_debug("swsusp: resume file found\n");
1368         else
1369                 pr_debug("swsusp: Error %d check for resume file\n", error);
1370         return error;
1371 }
1372
1373 /**
1374  *      swsusp_read - Read saved image from swap.
1375  */
1376
1377 int swsusp_read(void)
1378 {
1379         int error;
1380
1381         if (IS_ERR(resume_bdev)) {
1382                 pr_debug("swsusp: block device not initialised\n");
1383                 return PTR_ERR(resume_bdev);
1384         }
1385
1386         error = read_suspend_image();
1387         blkdev_put(resume_bdev);
1388
1389         if (!error)
1390                 pr_debug("swsusp: Reading resume file was successful\n");
1391         else
1392                 pr_debug("swsusp: Error %d resuming\n", error);
1393         return error;
1394 }
1395
1396 /**
1397  *      swsusp_close - close swap device.
1398  */
1399
1400 void swsusp_close(void)
1401 {
1402         if (IS_ERR(resume_bdev)) {
1403                 pr_debug("swsusp: block device not initialised\n");
1404                 return;
1405         }
1406
1407         blkdev_put(resume_bdev);
1408 }