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
  67 #include <asm/uaccess.h>
  68 #include <asm/mmu_context.h>
  69 #include <asm/pgtable.h>
  70 #include <asm/io.h>
  71
  72 #include "power.h"
  73
  74 /* References to section boundaries */
  75 extern char __nosave_begin, __nosave_end;
  76
  77 extern int is_head_of_free_region(struct page *);
  78
  79 /* Variables to be preserved over suspend */
  80 int pagedir_order_check;
  81 int nr_copy_pages_check;
  82
  83 extern char resume_file[];
  84 static dev_t resume_device;
  85 /* Local variables that should not be affected by save */
  86 unsigned int nr_copy_pages __nosavedata = 0;
  87
  88 /* Suspend pagedir is allocated before final copy, therefore it
  89    must be freed after resume
  90
  91    Warning: this is evil. There are actually two pagedirs at time of
  92    resume. One is "pagedir_save", which is empty frame allocated at
  93    time of suspend, that must be freed. Second is "pagedir_nosave",
  94    allocated at time of resume, that travels through memory not to
  95    collide with anything.
  96
  97    Warning: this is even more evil than it seems. Pagedirs this file
  98    talks about are completely different from page directories used by
  99    MMU hardware.
 100  */
 101 suspend_pagedir_t *pagedir_nosave __nosavedata = NULL;
 102 static suspend_pagedir_t *pagedir_save;
 103 static int pagedir_order __nosavedata = 0;
 104
 105 #define SWSUSP_SIG      "S1SUSPEND"
 106
 107 struct swsusp_header {
 108         char reserved[PAGE_SIZE - 20 - sizeof(swp_entry_t)];
 109         swp_entry_t swsusp_info;
 110         char    orig_sig[10];
 111         char    sig[10];
 112 } __attribute__((packed, aligned(PAGE_SIZE))) swsusp_header;
 113
 114 struct swsusp_info swsusp_info;
 115
 116 /*
 117  * XXX: We try to keep some more pages free so that I/O operations succeed
 118  * without paging. Might this be more?
 119  */
 120 #define PAGES_FOR_IO    512
 121
 122 /*
 123  * Saving part...
 124  */
 125
 126 /* We memorize in swapfile_used what swap devices are used for suspension */
 127 #define SWAPFILE_UNUSED    0
 128 #define SWAPFILE_SUSPEND   1    /* This is the suspending device */
 129 #define SWAPFILE_IGNORED   2    /* Those are other swap devices ignored for suspension */
 130
 131 static unsigned short swapfile_used[MAX_SWAPFILES];
 132 static unsigned short root_swap;
 133
 134 static int mark_swapfiles(swp_entry_t prev)
 135 {
 136         int error;
 137
 138         rw_swap_page_sync(READ,
 139                           swp_entry(root_swap, 0),
 140                           virt_to_page((unsigned long)&swsusp_header));
 141         if (!memcmp("SWAP-SPACE",swsusp_header.sig, 10) ||
 142             !memcmp("SWAPSPACE2",swsusp_header.sig, 10)) {
 143                 memcpy(swsusp_header.orig_sig,swsusp_header.sig, 10);
 144                 memcpy(swsusp_header.sig,SWSUSP_SIG, 10);
 145                 swsusp_header.swsusp_info = prev;
 146                 error = rw_swap_page_sync(WRITE,
 147                                           swp_entry(root_swap, 0),
 148                                           virt_to_page((unsigned long)
 149                                                        &swsusp_header));
 150         } else {
 151                 pr_debug("swsusp: Partition is not swap space.\n");
 152                 error = -ENODEV;
 153         }
 154         return error;
 155 }
 156
 157 /*
 158  * Check whether the swap device is the specified resume
 159  * device, irrespective of whether they are specified by
 160  * identical names.
 161  *
 162  * (Thus, device inode aliasing is allowed.  You can say /dev/hda4
 163  * instead of /dev/ide/host0/bus0/target0/lun0/part4 [if using devfs]
 164  * and they'll be considered the same device.  This is *necessary* for
 165  * devfs, since the resume code can only recognize the form /dev/hda4,
 166  * but the suspend code would see the long name.)
 167  */
 168 static int is_resume_device(const struct swap_info_struct *swap_info)
 169 {
 170         struct file *file = swap_info->swap_file;
 171         struct inode *inode = file->f_dentry->d_inode;
 172
 173         return S_ISBLK(inode->i_mode) &&
 174                 resume_device == MKDEV(imajor(inode), iminor(inode));
 175 }
 176
 177 int swsusp_swap_check(void) /* This is called before saving image */
 178 {
 179         int i, len;
 180
 181         len=strlen(resume_file);
 182         root_swap = 0xFFFF;
 183
 184         swap_list_lock();
 185         for(i=0; i<MAX_SWAPFILES; i++) {
 186                 if (swap_info[i].flags == 0) {
 187                         swapfile_used[i]=SWAPFILE_UNUSED;
 188                 } else {
 189                         if(!len) {
 190                                 printk(KERN_WARNING "resume= option should be used to set suspend device" );
 191                                 if(root_swap == 0xFFFF) {
 192                                         swapfile_used[i] = SWAPFILE_SUSPEND;
 193                                         root_swap = i;
 194                                 } else
 195                                         swapfile_used[i] = SWAPFILE_IGNORED;
 196                         } else {
 197                                 /* we ignore all swap devices that are not the resume_file */
 198                                 if (is_resume_device(&swap_info[i])) {
 199                                         swapfile_used[i] = SWAPFILE_SUSPEND;
 200                                         root_swap = i;
 201                                 } else {
 202                                         swapfile_used[i] = SWAPFILE_IGNORED;
 203                                 }
 204                         }
 205                 }
 206         }
 207         swap_list_unlock();
 208         return (root_swap != 0xffff) ? 0 : -ENODEV;
 209 }
 210
 211 /**
 212  * This is called after saving image so modification
 213  * will be lost after resume... and that's what we want.
 214  * we make the device unusable. A new call to
 215  * lock_swapdevices can unlock the devices.
 216  */
 217 static void lock_swapdevices(void)
 218 {
 219         int i;
 220
 221         swap_list_lock();
 222         for(i = 0; i< MAX_SWAPFILES; i++)
 223                 if(swapfile_used[i] == SWAPFILE_IGNORED) {
 224                         swap_info[i].flags ^= 0xFF;
 225                 }
 226         swap_list_unlock();
 227 }
 228
 229
 230
 231 /**
 232  *      write_swap_page - Write one page to a fresh swap location.
 233  *      @addr:  Address we're writing.
 234  *      @loc:   Place to store the entry we used.
 235  *
 236  *      Allocate a new swap entry and 'sync' it. Note we discard -EIO
 237  *      errors. That is an artifact left over from swsusp. It did not
 238  *      check the return of rw_swap_page_sync() at all, since most pages
 239  *      written back to swap would return -EIO.
 240  *      This is a partial improvement, since we will at least return other
 241  *      errors, though we need to eventually fix the damn code.
 242  */
 243
 244 static int write_page(unsigned long addr, swp_entry_t * loc)
 245 {
 246         swp_entry_t entry;
 247         int error = 0;
 248
 249         entry = get_swap_page();
 250         if (swp_offset(entry) &&
 251             swapfile_used[swp_type(entry)] == SWAPFILE_SUSPEND) {
 252                 error = rw_swap_page_sync(WRITE, entry,
 253                                           virt_to_page(addr));
 254                 if (error == -EIO)
 255                         error = 0;
 256                 if (!error)
 257                         *loc = entry;
 258         } else
 259                 error = -ENOSPC;
 260         return error;
 261 }
 262
 263
 264 /**
 265  *      data_free - Free the swap entries used by the saved image.
 266  *
 267  *      Walk the list of used swap entries and free each one.
 268  *      This is only used for cleanup when suspend fails.
 269  */
 270
 271 static void data_free(void)
 272 {
 273         swp_entry_t entry;
 274         int i;
 275
 276         for (i = 0; i < nr_copy_pages; i++) {
 277                 entry = (pagedir_nosave + i)->swap_address;
 278                 if (entry.val)
 279                         swap_free(entry);
 280                 else
 281                         break;
 282                 (pagedir_nosave + i)->swap_address = (swp_entry_t){0};
 283         }
 284 }
 285
 286
 287 /**
 288  *      data_write - Write saved image to swap.
 289  *
 290  *      Walk the list of pages in the image and sync each one to swap.
 291  */
 292
 293 static int data_write(void)
 294 {
 295         int error = 0;
 296         int i;
 297
 298         printk( "Writing data to swap (%d pages): ", nr_copy_pages );
 299         for (i = 0; i < nr_copy_pages && !error; i++) {
 300                 if (!(i%100))
 301                         printk( "." );
 302                 error = write_page((pagedir_nosave+i)->address,
 303                                           &((pagedir_nosave+i)->swap_address));
 304         }
 305         printk(" %d Pages done.\n",i);
 306         return error;
 307 }
 308
 309 static void dump_info(void)
 310 {
 311         pr_debug(" swsusp: Version: %u\n",swsusp_info.version_code);
 312         pr_debug(" swsusp: Num Pages: %ld\n",swsusp_info.num_physpages);
 313         pr_debug(" swsusp: UTS Sys: %s\n",swsusp_info.uts.sysname);
 314         pr_debug(" swsusp: UTS Node: %s\n",swsusp_info.uts.nodename);
 315         pr_debug(" swsusp: UTS Release: %s\n",swsusp_info.uts.release);
 316         pr_debug(" swsusp: UTS Version: %s\n",swsusp_info.uts.version);
 317         pr_debug(" swsusp: UTS Machine: %s\n",swsusp_info.uts.machine);
 318         pr_debug(" swsusp: UTS Domain: %s\n",swsusp_info.uts.domainname);
 319         pr_debug(" swsusp: CPUs: %d\n",swsusp_info.cpus);
 320         pr_debug(" swsusp: Image: %ld Pages\n",swsusp_info.image_pages);
 321         pr_debug(" swsusp: Pagedir: %ld Pages\n",swsusp_info.pagedir_pages);
 322 }
 323
 324 static void init_header(void)
 325 {
 326         memset(&swsusp_info,0,sizeof(swsusp_info));
 327         swsusp_info.version_code = LINUX_VERSION_CODE;
 328         swsusp_info.num_physpages = num_physpages;
 329         memcpy(&swsusp_info.uts,&system_utsname,sizeof(system_utsname));
 330
 331         swsusp_info.suspend_pagedir = pagedir_nosave;
 332         swsusp_info.cpus = num_online_cpus();
 333         swsusp_info.image_pages = nr_copy_pages;
 334         dump_info();
 335 }
 336
 337 static int close_swap(void)
 338 {
 339         swp_entry_t entry;
 340         int error;
 341
 342         error = write_page((unsigned long)&swsusp_info,&entry);
 343         if (!error) {
 344                 printk( "S" );
 345                 error = mark_swapfiles(entry);
 346                 printk( "|\n" );
 347         }
 348         return error;
 349 }
 350
 351 /**
 352  *      free_pagedir_entries - Free pages used by the page directory.
 353  *
 354  *      This is used during suspend for error recovery.
 355  */
 356
 357 static void free_pagedir_entries(void)
 358 {
 359         int i;
 360
 361         for (i = 0; i < swsusp_info.pagedir_pages; i++)
 362                 swap_free(swsusp_info.pagedir[i]);
 363 }
 364
 365
 366 /**
 367  *      write_pagedir - Write the array of pages holding the page directory.
 368  *      @last:  Last swap entry we write (needed for header).
 369  */
 370
 371 static int write_pagedir(void)
 372 {
 373         unsigned long addr = (unsigned long)pagedir_nosave;
 374         int error = 0;
 375         int n = SUSPEND_PD_PAGES(nr_copy_pages);
 376         int i;
 377
 378         swsusp_info.pagedir_pages = n;
 379         printk( "Writing pagedir (%d pages)\n", n);
 380         for (i = 0; i < n && !error; i++, addr += PAGE_SIZE)
 381                 error = write_page(addr, &swsusp_info.pagedir[i]);
 382         return error;
 383 }
 384
 385 /**
 386  *      write_suspend_image - Write entire image and metadata.
 387  *
 388  */
 389
 390 static int write_suspend_image(void)
 391 {
 392         int error;
 393
 394         init_header();
 395         if ((error = data_write()))
 396                 goto FreeData;
 397
 398         if ((error = write_pagedir()))
 399                 goto FreePagedir;
 400
 401         if ((error = close_swap()))
 402                 goto FreePagedir;
 403  Done:
 404         return error;
 405  FreePagedir:
 406         free_pagedir_entries();
 407  FreeData:
 408         data_free();
 409         goto Done;
 410 }
 411
 412
 413 #ifdef CONFIG_HIGHMEM
 414 struct highmem_page {
 415         char *data;
 416         struct page *page;
 417         struct highmem_page *next;
 418 };
 419
 420 struct highmem_page *highmem_copy = NULL;
 421
 422 static int save_highmem_zone(struct zone *zone)
 423 {
 424         unsigned long zone_pfn;
 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                 int chunk_size;
 431
 432                 if (!(pfn%1000))
 433                         printk(".");
 434                 if (!pfn_valid(pfn))
 435                         continue;
 436                 page = pfn_to_page(pfn);
 437                 /*
 438                  * This condition results from rvmalloc() sans vmalloc_32()
 439                  * and architectural memory reservations. This should be
 440                  * corrected eventually when the cases giving rise to this
 441                  * are better understood.
 442                  */
 443                 if (PageReserved(page)) {
 444                         printk("highmem reserved page?!\n");
 445                         continue;
 446                 }
 447                 if ((chunk_size = is_head_of_free_region(page))) {
 448                         pfn += chunk_size - 1;
 449                         zone_pfn += chunk_size - 1;
 450                         continue;
 451                 }
 452                 save = kmalloc(sizeof(struct highmem_page), GFP_ATOMIC);
 453                 if (!save)
 454                         return -ENOMEM;
 455                 save->next = highmem_copy;
 456                 save->page = page;
 457                 save->data = (void *) get_zeroed_page(GFP_ATOMIC);
 458                 if (!save->data) {
 459                         kfree(save);
 460                         return -ENOMEM;
 461                 }
 462                 kaddr = kmap_atomic(page, KM_USER0);
 463                 memcpy(save->data, kaddr, PAGE_SIZE);
 464                 kunmap_atomic(kaddr, KM_USER0);
 465                 highmem_copy = save;
 466         }
 467         return 0;
 468 }
 469 #endif /* CONFIG_HIGHMEM */
 470
 471
 472 static int save_highmem(void)
 473 {
 474 #ifdef CONFIG_HIGHMEM
 475         struct zone *zone;
 476         int res = 0;
 477
 478         pr_debug("swsusp: Saving Highmem\n");
 479         for_each_zone(zone) {
 480                 if (is_highmem(zone))
 481                         res = save_highmem_zone(zone);
 482                 if (res)
 483                         return res;
 484         }
 485 #endif
 486         return 0;
 487 }
 488
 489 static int restore_highmem(void)
 490 {
 491 #ifdef CONFIG_HIGHMEM
 492         printk("swsusp: Restoring Highmem\n");
 493         while (highmem_copy) {
 494                 struct highmem_page *save = highmem_copy;
 495                 void *kaddr;
 496                 highmem_copy = save->next;
 497
 498                 kaddr = kmap_atomic(save->page, KM_USER0);
 499                 memcpy(kaddr, save->data, PAGE_SIZE);
 500                 kunmap_atomic(kaddr, KM_USER0);
 501                 free_page((long) save->data);
 502                 kfree(save);
 503         }
 504 #endif
 505         return 0;
 506 }
 507
 508
 509 static int pfn_is_nosave(unsigned long pfn)
 510 {
 511         unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
 512         unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
 513         return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
 514 }
 515
 516 /**
 517  *      saveable - Determine whether a page should be cloned or not.
 518  *      @pfn:   The page
 519  *
 520  *      We save a page if it's Reserved, and not in the range of pages
 521  *      statically defined as 'unsaveable', or if it isn't reserved, and
 522  *      isn't part of a free chunk of pages.
 523  *      If it is part of a free chunk, we update @pfn to point to the last
 524  *      page of the chunk.
 525  */
 526
 527 static int saveable(struct zone * zone, unsigned long * zone_pfn)
 528 {
 529         unsigned long pfn = *zone_pfn + zone->zone_start_pfn;
 530         unsigned long chunk_size;
 531         struct page * page;
 532
 533         if (!pfn_valid(pfn))
 534                 return 0;
 535
 536         if (!(pfn%1000))
 537                 printk(".");
 538         page = pfn_to_page(pfn);
 539         BUG_ON(PageReserved(page) && PageNosave(page));
 540         if (PageNosave(page))
 541                 return 0;
 542         if (PageReserved(page) && pfn_is_nosave(pfn)) {
 543                 pr_debug("[nosave pfn 0x%lx]", pfn);
 544                 return 0;
 545         }
 546         if ((chunk_size = is_head_of_free_region(page))) {
 547                 *zone_pfn += chunk_size - 1;
 548                 return 0;
 549         }
 550
 551         return 1;
 552 }
 553
 554 static void count_data_pages(void)
 555 {
 556         struct zone *zone;
 557         unsigned long zone_pfn;
 558
 559         nr_copy_pages = 0;
 560
 561         for_each_zone(zone) {
 562                 if (!is_highmem(zone)) {
 563                         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 564                                 nr_copy_pages += saveable(zone, &zone_pfn);
 565                 }
 566         }
 567 }
 568
 569
 570 static void copy_data_pages(void)
 571 {
 572         struct zone *zone;
 573         unsigned long zone_pfn;
 574         struct pbe * pbe = pagedir_nosave;
 575
 576         for_each_zone(zone) {
 577                 if (!is_highmem(zone))
 578                         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
 579                                 if (saveable(zone, &zone_pfn)) {
 580                                         struct page * page;
 581                                         page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
 582                                         pbe->orig_address = (long) page_address(page);
 583                                         /* copy_page is no usable for copying task structs. */
 584                                         memcpy((void *)pbe->address, (void *)pbe->orig_address, PAGE_SIZE);
 585                                         pbe++;
 586                                 }
 587                         }
 588         }
 589 }
 590
 591
 592 static void free_suspend_pagedir_zone(struct zone *zone, unsigned long pagedir)
 593 {
 594         unsigned long zone_pfn, pagedir_end, pagedir_pfn, pagedir_end_pfn;
 595         pagedir_end = pagedir + (PAGE_SIZE << pagedir_order);
 596         pagedir_pfn = __pa(pagedir) >> PAGE_SHIFT;
 597         pagedir_end_pfn = __pa(pagedir_end) >> PAGE_SHIFT;
 598         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
 599                 struct page *page;
 600                 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
 601                 if (!pfn_valid(pfn))
 602                         continue;
 603                 page = pfn_to_page(pfn);
 604                 if (!TestClearPageNosave(page))
 605                         continue;
 606                 else if (pfn >= pagedir_pfn && pfn < pagedir_end_pfn)
 607                         continue;
 608                 __free_page(page);
 609         }
 610 }
 611
 612 void swsusp_free(void)
 613 {
 614         unsigned long p = (unsigned long)pagedir_save;
 615         struct zone *zone;
 616         for_each_zone(zone) {
 617                 if (!is_highmem(zone))
 618                         free_suspend_pagedir_zone(zone, p);
 619         }
 620         free_pages(p, pagedir_order);
 621 }
 622
 623
 624 /**
 625  *      calc_order - Determine the order of allocation needed for pagedir_save.
 626  *
 627  *      This looks tricky, but is just subtle. Please fix it some time.
 628  *      Since there are %nr_copy_pages worth of pages in the snapshot, we need
 629  *      to allocate enough contiguous space to hold
 630  *              (%nr_copy_pages * sizeof(struct pbe)),
 631  *      which has the saved/orig locations of the page..
 632  *
 633  *      SUSPEND_PD_PAGES() tells us how many pages we need to hold those
 634  *      structures, then we call get_bitmask_order(), which will tell us the
 635  *      last bit set in the number, starting with 1. (If we need 30 pages, that
 636  *      is 0x0000001e in hex. The last bit is the 5th, which is the order we
 637  *      would use to allocate 32 contiguous pages).
 638  *
 639  *      Since we also need to save those pages, we add the number of pages that
 640  *      we need to nr_copy_pages, and in case of an overflow, do the
 641  *      calculation again to update the number of pages needed.
 642  *
 643  *      With this model, we will tend to waste a lot of memory if we just cross
 644  *      an order boundary. Plus, the higher the order of allocation that we try
 645  *      to do, the more likely we are to fail in a low-memory situtation
 646  *      (though we're unlikely to get this far in such a case, since swsusp
 647  *      requires half of memory to be free anyway).
 648  */
 649
 650
 651 static void calc_order(void)
 652 {
 653         int diff = 0;
 654         int order = 0;
 655
 656         do {
 657                 diff = get_bitmask_order(SUSPEND_PD_PAGES(nr_copy_pages)) - order;
 658                 if (diff) {
 659                         order += diff;
 660                         nr_copy_pages += 1 << diff;
 661                 }
 662         } while(diff);
 663         pagedir_order = order;
 664 }
 665
 666
 667 /**
 668  *      alloc_pagedir - Allocate the page directory.
 669  *
 670  *      First, determine exactly how many contiguous pages we need and
 671  *      allocate them.
 672  */
 673
 674 static int alloc_pagedir(void)
 675 {
 676         calc_order();
 677         pagedir_save = (suspend_pagedir_t *)__get_free_pages(GFP_ATOMIC | __GFP_COLD,
 678                                                              pagedir_order);
 679         if (!pagedir_save)
 680                 return -ENOMEM;
 681         memset(pagedir_save, 0, (1 << pagedir_order) * PAGE_SIZE);
 682         pagedir_nosave = pagedir_save;
 683         return 0;
 684 }
 685
 686
 687 /**
 688  *      alloc_image_pages - Allocate pages for the snapshot.
 689  *
 690  */
 691
 692 static int alloc_image_pages(void)
 693 {
 694         struct pbe * p;
 695         int i;
 696
 697         for (i = 0, p = pagedir_save; i < nr_copy_pages; i++, p++) {
 698                 p->address = get_zeroed_page(GFP_ATOMIC | __GFP_COLD);
 699                 if(!p->address)
 700                         goto Error;
 701                 SetPageNosave(virt_to_page(p->address));
 702         }
 703         return 0;
 704  Error:
 705         do {
 706                 if (p->address)
 707                         free_page(p->address);
 708                 p->address = 0;
 709         } while (p-- > pagedir_save);
 710         return -ENOMEM;
 711 }
 712
 713
 714 /**
 715  *      enough_free_mem - Make sure we enough free memory to snapshot.
 716  *
 717  *      Returns TRUE or FALSE after checking the number of available
 718  *      free pages.
 719  */
 720
 721 static int enough_free_mem(void)
 722 {
 723         if (nr_free_pages() < (nr_copy_pages + PAGES_FOR_IO)) {
 724                 pr_debug("swsusp: Not enough free pages: Have %d\n",
 725                          nr_free_pages());
 726                 return 0;
 727         }
 728         return 1;
 729 }
 730
 731
 732 /**
 733  *      enough_swap - Make sure we have enough swap to save the image.
 734  *
 735  *      Returns TRUE or FALSE after checking the total amount of swap
 736  *      space avaiable.
 737  *
 738  *      FIXME: si_swapinfo(&i) returns all swap devices information.
 739  *      We should only consider resume_device.
 740  */
 741
 742 static int enough_swap(void)
 743 {
 744         struct sysinfo i;
 745
 746         si_swapinfo(&i);
 747         if (i.freeswap < (nr_copy_pages + PAGES_FOR_IO))  {
 748                 pr_debug("swsusp: Not enough swap. Need %ld\n",i.freeswap);
 749                 return 0;
 750         }
 751         return 1;
 752 }
 753
 754 static int swsusp_alloc(void)
 755 {
 756         int error;
 757
 758         pr_debug("suspend: (pages needed: %d + %d free: %d)\n",
 759                  nr_copy_pages, PAGES_FOR_IO, nr_free_pages());
 760
 761         pagedir_nosave = NULL;
 762         if (!enough_free_mem())
 763                 return -ENOMEM;
 764
 765         if (!enough_swap())
 766                 return -ENOSPC;
 767
 768         if ((error = alloc_pagedir())) {
 769                 pr_debug("suspend: Allocating pagedir failed.\n");
 770                 return error;
 771         }
 772         if ((error = alloc_image_pages())) {
 773                 pr_debug("suspend: Allocating image pages failed.\n");
 774                 swsusp_free();
 775                 return error;
 776         }
 777
 778         nr_copy_pages_check = nr_copy_pages;
 779         pagedir_order_check = pagedir_order;
 780         return 0;
 781 }
 782
 783 int suspend_prepare_image(void)
 784 {
 785         unsigned int nr_needed_pages = 0;
 786         int error;
 787
 788         pr_debug("swsusp: critical section: \n");
 789         if (save_highmem()) {
 790                 printk(KERN_CRIT "Suspend machine: Not enough free pages for highmem\n");
 791                 return -ENOMEM;
 792         }
 793
 794         drain_local_pages();
 795         count_data_pages();
 796         printk("swsusp: Need to copy %u pages\n",nr_copy_pages);
 797         nr_needed_pages = nr_copy_pages + PAGES_FOR_IO;
 798
 799         error = swsusp_alloc();
 800         if (error)
 801                 return error;
 802
 803         /* During allocating of suspend pagedir, new cold pages may appear.
 804          * Kill them.
 805          */
 806         drain_local_pages();
 807         copy_data_pages();
 808
 809         /*
 810          * End of critical section. From now on, we can write to memory,
 811          * but we should not touch disk. This specially means we must _not_
 812          * touch swap space! Except we must write out our image of course.
 813          */
 814
 815         printk("swsusp: critical section/: done (%d pages copied)\n", nr_copy_pages );
 816         return 0;
 817 }
 818
 819
 820 /* It is important _NOT_ to umount filesystems at this point. We want
 821  * them synced (in case something goes wrong) but we DO not want to mark
 822  * filesystem clean: it is not. (And it does not matter, if we resume
 823  * correctly, we'll mark system clean, anyway.)
 824  */
 825 int swsusp_write(void)
 826 {
 827         int error;
 828         device_resume();
 829         lock_swapdevices();
 830         error = write_suspend_image();
 831         /* This will unlock ignored swap devices since writing is finished */
 832         lock_swapdevices();
 833         return error;
 834
 835 }
 836
 837
 838 extern asmlinkage int swsusp_arch_suspend(void);
 839 extern asmlinkage int swsusp_arch_resume(void);
 840
 841
 842 asmlinkage int swsusp_save(void)
 843 {
 844         int error = 0;
 845
 846         if ((error = swsusp_swap_check()))
 847                 return error;
 848         return suspend_prepare_image();
 849 }
 850
 851 int swsusp_suspend(void)
 852 {
 853         int error;
 854         if ((error = arch_prepare_suspend()))
 855                 return error;
 856         local_irq_disable();
 857         save_processor_state();
 858         error = swsusp_arch_suspend();
 859         /* Restore control flow magically appears here */
 860         restore_processor_state();
 861         restore_highmem();
 862         local_irq_enable();
 863         return error;
 864 }
 865
 866
 867 asmlinkage int swsusp_restore(void)
 868 {
 869         BUG_ON (nr_copy_pages_check != nr_copy_pages);
 870         BUG_ON (pagedir_order_check != pagedir_order);
 871
 872         /* Even mappings of "global" things (vmalloc) need to be fixed */
 873         __flush_tlb_global();
 874         return 0;
 875 }
 876
 877 int swsusp_resume(void)
 878 {
 879         int error;
 880         local_irq_disable();
 881         /* We'll ignore saved state, but this gets preempt count (etc) right */
 882         save_processor_state();
 883         error = swsusp_arch_resume();
 884         /* Code below is only ever reached in case of failure. Otherwise
 885          * execution continues at place where swsusp_arch_suspend was called
 886          */
 887         BUG_ON(!error);
 888         restore_processor_state();
 889         restore_highmem();
 890         local_irq_enable();
 891         return error;
 892 }
 893
 894
 895
 896 /* More restore stuff */
 897
 898 #define does_collide(addr) does_collide_order(pagedir_nosave, addr, 0)
 899
 900 /*
 901  * Returns true if given address/order collides with any orig_address
 902  */
 903 static int __init does_collide_order(suspend_pagedir_t *pagedir, unsigned long addr,
 904                 int order)
 905 {
 906         int i;
 907         unsigned long addre = addr + (PAGE_SIZE<<order);
 908
 909         for (i=0; i < nr_copy_pages; i++)
 910                 if ((pagedir+i)->orig_address >= addr &&
 911                         (pagedir+i)->orig_address < addre)
 912                         return 1;
 913
 914         return 0;
 915 }
 916
 917 /*
 918  * We check here that pagedir & pages it points to won't collide with pages
 919  * where we're going to restore from the loaded pages later
 920  */
 921 static int __init check_pagedir(void)
 922 {
 923         int i;
 924
 925         for(i=0; i < nr_copy_pages; i++) {
 926                 unsigned long addr;
 927
 928                 do {
 929                         addr = get_zeroed_page(GFP_ATOMIC);
 930                         if(!addr)
 931                                 return -ENOMEM;
 932                 } while (does_collide(addr));
 933
 934                 (pagedir_nosave+i)->address = addr;
 935         }
 936         return 0;
 937 }
 938
 939 static int __init swsusp_pagedir_relocate(void)
 940 {
 941         /*
 942          * We have to avoid recursion (not to overflow kernel stack),
 943          * and that's why code looks pretty cryptic
 944          */
 945         suspend_pagedir_t *old_pagedir = pagedir_nosave;
 946         void **eaten_memory = NULL;
 947         void **c = eaten_memory, *m, *f;
 948         int ret = 0;
 949
 950         printk("Relocating pagedir ");
 951
 952         if (!does_collide_order(old_pagedir, (unsigned long)old_pagedir, pagedir_order)) {
 953                 printk("not necessary\n");
 954                 return check_pagedir();
 955         }
 956
 957         while ((m = (void *) __get_free_pages(GFP_ATOMIC, pagedir_order)) != NULL) {
 958                 if (!does_collide_order(old_pagedir, (unsigned long)m, pagedir_order))
 959                         break;
 960                 eaten_memory = m;
 961                 printk( "." );
 962                 *eaten_memory = c;
 963                 c = eaten_memory;
 964         }
 965
 966         if (!m) {
 967                 printk("out of memory\n");
 968                 ret = -ENOMEM;
 969         } else {
 970                 pagedir_nosave =
 971                         memcpy(m, old_pagedir, PAGE_SIZE << pagedir_order);
 972         }
 973
 974         c = eaten_memory;
 975         while (c) {
 976                 printk(":");
 977                 f = c;
 978                 c = *c;
 979                 free_pages((unsigned long)f, pagedir_order);
 980         }
 981         printk("|\n");
 982         return check_pagedir();
 983 }
 984
 985 /**
 986  *      Using bio to read from swap.
 987  *      This code requires a bit more work than just using buffer heads
 988  *      but, it is the recommended way for 2.5/2.6.
 989  *      The following are to signal the beginning and end of I/O. Bios
 990  *      finish asynchronously, while we want them to happen synchronously.
 991  *      A simple atomic_t, and a wait loop take care of this problem.
 992  */
 993
 994 static atomic_t io_done = ATOMIC_INIT(0);
 995
 996 static void start_io(void)
 997 {
 998         atomic_set(&io_done,1);
 999 }
1000
1001 static int end_io(struct bio * bio, unsigned int num, int err)
1002 {
1003         atomic_set(&io_done,0);
1004         return 0;
1005 }
1006
1007 static void wait_io(void)
1008 {
1009         while(atomic_read(&io_done))
1010                 io_schedule();
1011 }
1012
1013
1014 static struct block_device * resume_bdev;
1015
1016 /**
1017  *      submit - submit BIO request.
1018  *      @rw:    READ or WRITE.
1019  *      @off    physical offset of page.
1020  *      @page:  page we're reading or writing.
1021  *
1022  *      Straight from the textbook - allocate and initialize the bio.
1023  *      If we're writing, make sure the page is marked as dirty.
1024  *      Then submit it and wait.
1025  */
1026
1027 static int submit(int rw, pgoff_t page_off, void * page)
1028 {
1029         int error = 0;
1030         struct bio * bio;
1031
1032         bio = bio_alloc(GFP_ATOMIC, 1);
1033         if (!bio)
1034                 return -ENOMEM;
1035         bio->bi_sector = page_off * (PAGE_SIZE >> 9);
1036         bio_get(bio);
1037         bio->bi_bdev = resume_bdev;
1038         bio->bi_end_io = end_io;
1039
1040         if (bio_add_page(bio, virt_to_page(page), PAGE_SIZE, 0) < PAGE_SIZE) {
1041                 printk("swsusp: ERROR: adding page to bio at %ld\n",page_off);
1042                 error = -EFAULT;
1043                 goto Done;
1044         }
1045
1046         if (rw == WRITE)
1047                 bio_set_pages_dirty(bio);
1048         start_io();
1049         submit_bio(rw | (1 << BIO_RW_SYNC), bio);
1050         wait_io();
1051  Done:
1052         bio_put(bio);
1053         return error;
1054 }
1055
1056 int bio_read_page(pgoff_t page_off, void * page)
1057 {
1058         return submit(READ, page_off, page);
1059 }
1060
1061 int bio_write_page(pgoff_t page_off, void * page)
1062 {
1063         return submit(WRITE, page_off, page);
1064 }
1065
1066 /*
1067  * Sanity check if this image makes sense with this kernel/swap context
1068  * I really don't think that it's foolproof but more than nothing..
1069  */
1070
1071 static const char * __init sanity_check(void)
1072 {
1073         dump_info();
1074         if(swsusp_info.version_code != LINUX_VERSION_CODE)
1075                 return "kernel version";
1076         if(swsusp_info.num_physpages != num_physpages)
1077                 return "memory size";
1078         if (strcmp(swsusp_info.uts.sysname,system_utsname.sysname))
1079                 return "system type";
1080         if (strcmp(swsusp_info.uts.release,system_utsname.release))
1081                 return "kernel release";
1082         if (strcmp(swsusp_info.uts.version,system_utsname.version))
1083                 return "version";
1084         if (strcmp(swsusp_info.uts.machine,system_utsname.machine))
1085                 return "machine";
1086         if(swsusp_info.cpus != num_online_cpus())
1087                 return "number of cpus";
1088         return NULL;
1089 }
1090
1091
1092 static int __init check_header(void)
1093 {
1094         const char * reason = NULL;
1095         int error;
1096
1097         if ((error = bio_read_page(swp_offset(swsusp_header.swsusp_info), &swsusp_info)))
1098                 return error;
1099
1100         /* Is this same machine? */
1101         if ((reason = sanity_check())) {
1102                 printk(KERN_ERR "swsusp: Resume mismatch: %s\n",reason);
1103                 return -EPERM;
1104         }
1105         nr_copy_pages = swsusp_info.image_pages;
1106         return error;
1107 }
1108
1109 static int __init check_sig(void)
1110 {
1111         int error;
1112
1113         memset(&swsusp_header, 0, sizeof(swsusp_header));
1114         if ((error = bio_read_page(0, &swsusp_header)))
1115                 return error;
1116         if (!memcmp(SWSUSP_SIG, swsusp_header.sig, 10)) {
1117                 memcpy(swsusp_header.sig, swsusp_header.orig_sig, 10);
1118
1119                 /*
1120                  * Reset swap signature now.
1121                  */
1122                 error = bio_write_page(0, &swsusp_header);
1123         } else {
1124                 pr_debug(KERN_ERR "swsusp: Invalid partition type.\n");
1125                 return -EINVAL;
1126         }
1127         if (!error)
1128                 pr_debug("swsusp: Signature found, resuming\n");
1129         return error;
1130 }
1131
1132 /**
1133  *      swsusp_read_data - Read image pages from swap.
1134  *
1135  *      You do not need to check for overlaps, check_pagedir()
1136  *      already did that.
1137  */
1138
1139 static int __init data_read(void)
1140 {
1141         struct pbe * p;
1142         int error;
1143         int i;
1144
1145         if ((error = swsusp_pagedir_relocate()))
1146                 return error;
1147
1148         printk( "Reading image data (%d pages): ", nr_copy_pages );
1149         for(i = 0, p = pagedir_nosave; i < nr_copy_pages && !error; i++, p++) {
1150                 if (!(i%100))
1151                         printk( "." );
1152                 error = bio_read_page(swp_offset(p->swap_address),
1153                                   (void *)p->address);
1154         }
1155         printk(" %d done.\n",i);
1156         return error;
1157
1158 }
1159
1160 extern dev_t __init name_to_dev_t(const char *line);
1161
1162 static int __init read_pagedir(void)
1163 {
1164         unsigned long addr;
1165         int i, n = swsusp_info.pagedir_pages;
1166         int error = 0;
1167
1168         pagedir_order = get_bitmask_order(n);
1169
1170         addr =__get_free_pages(GFP_ATOMIC, pagedir_order);
1171         if (!addr)
1172                 return -ENOMEM;
1173         pagedir_nosave = (struct pbe *)addr;
1174
1175         pr_debug("pmdisk: Reading pagedir (%d Pages)\n",n);
1176
1177         for (i = 0; i < n && !error; i++, addr += PAGE_SIZE) {
1178                 unsigned long offset = swp_offset(swsusp_info.pagedir[i]);
1179                 if (offset)
1180                         error = bio_read_page(offset, (void *)addr);
1181                 else
1182                         error = -EFAULT;
1183         }
1184         if (error)
1185                 free_pages((unsigned long)pagedir_nosave, pagedir_order);
1186         return error;
1187 }
1188
1189 static int __init read_suspend_image(void)
1190 {
1191         int error = 0;
1192
1193         if ((error = check_sig()))
1194                 return error;
1195         if ((error = check_header()))
1196                 return error;
1197         if ((error = read_pagedir()))
1198                 return error;
1199         if ((error = data_read()))
1200                 free_pages((unsigned long)pagedir_nosave, pagedir_order);
1201         return error;
1202 }
1203
1204 /**
1205  *      pmdisk_read - Read saved image from swap.
1206  */
1207
1208 int __init swsusp_read(void)
1209 {
1210         int error;
1211
1212         if (!strlen(resume_file))
1213                 return -ENOENT;
1214
1215         resume_device = name_to_dev_t(resume_file);
1216         pr_debug("swsusp: Resume From Partition: %s\n", resume_file);
1217
1218         resume_bdev = open_by_devnum(resume_device, FMODE_READ);
1219         if (!IS_ERR(resume_bdev)) {
1220                 set_blocksize(resume_bdev, PAGE_SIZE);
1221                 error = read_suspend_image();
1222                 blkdev_put(resume_bdev);
1223         } else
1224                 error = PTR_ERR(resume_bdev);
1225
1226         if (!error)
1227                 pr_debug("Reading resume file was successful\n");
1228         else
1229                 pr_debug("pmdisk: Error %d resuming\n", error);
1230         return error;
1231 }