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/tlbflush.h>
  71 #include <asm/io.h>
  72
  73 #include "power.h"
  74
  75 /* References to section boundaries */
  76 extern const void __nosave_begin, __nosave_end;
  77
  78 /* Variables to be preserved over suspend */
  79 static int pagedir_order_check;
  80 static int nr_copy_pages_check;
  81
  82 extern char resume_file[];
  83 static dev_t resume_device;
  84 /* Local variables that should not be affected by save */
  85 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 static int pagedir_order __nosavedata = 0;
 103
 104 #define SWSUSP_SIG      "S1SUSPEND"
 105
 106 static struct swsusp_header {
 107         char reserved[PAGE_SIZE - 20 - sizeof(swp_entry_t)];
 108         swp_entry_t swsusp_info;
 109         char    orig_sig[10];
 110         char    sig[10];
 111 } __attribute__((packed, aligned(PAGE_SIZE))) swsusp_header;
 112
 113 static struct swsusp_info swsusp_info;
 114
 115 /*
 116  * XXX: We try to keep some more pages free so that I/O operations succeed
 117  * without paging. Might this be more?
 118  */
 119 #define PAGES_FOR_IO    512
 120
 121 /*
 122  * Saving part...
 123  */
 124
 125 /* We memorize in swapfile_used what swap devices are used for suspension */
 126 #define SWAPFILE_UNUSED    0
 127 #define SWAPFILE_SUSPEND   1    /* This is the suspending device */
 128 #define SWAPFILE_IGNORED   2    /* Those are other swap devices ignored for suspension */
 129
 130 static unsigned short swapfile_used[MAX_SWAPFILES];
 131 static unsigned short root_swap;
 132
 133 static int mark_swapfiles(swp_entry_t prev)
 134 {
 135         int error;
 136
 137         rw_swap_page_sync(READ,
 138                           swp_entry(root_swap, 0),
 139                           virt_to_page((unsigned long)&swsusp_header));
 140         if (!memcmp("SWAP-SPACE",swsusp_header.sig, 10) ||
 141             !memcmp("SWAPSPACE2",swsusp_header.sig, 10)) {
 142                 memcpy(swsusp_header.orig_sig,swsusp_header.sig, 10);
 143                 memcpy(swsusp_header.sig,SWSUSP_SIG, 10);
 144                 swsusp_header.swsusp_info = prev;
 145                 error = rw_swap_page_sync(WRITE,
 146                                           swp_entry(root_swap, 0),
 147                                           virt_to_page((unsigned long)
 148                                                        &swsusp_header));
 149         } else {
 150                 pr_debug("swsusp: Partition is not swap space.\n");
 151                 error = -ENODEV;
 152         }
 153         return error;
 154 }
 155
 156 /*
 157  * Check whether the swap device is the specified resume
 158  * device, irrespective of whether they are specified by
 159  * identical names.
 160  *
 161  * (Thus, device inode aliasing is allowed.  You can say /dev/hda4
 162  * instead of /dev/ide/host0/bus0/target0/lun0/part4 [if using devfs]
 163  * and they'll be considered the same device.  This is *necessary* for
 164  * devfs, since the resume code can only recognize the form /dev/hda4,
 165  * but the suspend code would see the long name.)
 166  */
 167 static int is_resume_device(const struct swap_info_struct *swap_info)
 168 {
 169         struct file *file = swap_info->swap_file;
 170         struct inode *inode = file->f_dentry->d_inode;
 171
 172         return S_ISBLK(inode->i_mode) &&
 173                 resume_device == MKDEV(imajor(inode), iminor(inode));
 174 }
 175
 176 static int swsusp_swap_check(void) /* This is called before saving image */
 177 {
 178         int i, len;
 179
 180         len=strlen(resume_file);
 181         root_swap = 0xFFFF;
 182
 183         swap_list_lock();
 184         for(i=0; i<MAX_SWAPFILES; i++) {
 185                 if (swap_info[i].flags == 0) {
 186                         swapfile_used[i]=SWAPFILE_UNUSED;
 187                 } else {
 188                         if(!len) {
 189                                 printk(KERN_WARNING "resume= option should be used to set suspend device" );
 190                                 if(root_swap == 0xFFFF) {
 191                                         swapfile_used[i] = SWAPFILE_SUSPEND;
 192                                         root_swap = i;
 193                                 } else
 194                                         swapfile_used[i] = SWAPFILE_IGNORED;
 195                         } else {
 196                                 /* we ignore all swap devices that are not the resume_file */
 197                                 if (is_resume_device(&swap_info[i])) {
 198                                         swapfile_used[i] = SWAPFILE_SUSPEND;
 199                                         root_swap = i;
 200                                 } else {
 201                                         swapfile_used[i] = SWAPFILE_IGNORED;
 202                                 }
 203                         }
 204                 }
 205         }
 206         swap_list_unlock();
 207         return (root_swap != 0xffff) ? 0 : -ENODEV;
 208 }
 209
 210 /**
 211  * This is called after saving image so modification
 212  * will be lost after resume... and that's what we want.
 213  * we make the device unusable. A new call to
 214  * lock_swapdevices can unlock the devices.
 215  */
 216 static void lock_swapdevices(void)
 217 {
 218         int i;
 219
 220         swap_list_lock();
 221         for(i = 0; i< MAX_SWAPFILES; i++)
 222                 if(swapfile_used[i] == SWAPFILE_IGNORED) {
 223                         swap_info[i].flags ^= 0xFF;
 224                 }
 225         swap_list_unlock();
 226 }
 227
 228
 229
 230 /**
 231  *      write_swap_page - Write one page to a fresh swap location.
 232  *      @addr:  Address we're writing.
 233  *      @loc:   Place to store the entry we used.
 234  *
 235  *      Allocate a new swap entry and 'sync' it. Note we discard -EIO
 236  *      errors. That is an artifact left over from swsusp. It did not
 237  *      check the return of rw_swap_page_sync() at all, since most pages
 238  *      written back to swap would return -EIO.
 239  *      This is a partial improvement, since we will at least return other
 240  *      errors, though we need to eventually fix the damn code.
 241  */
 242
 243 static int write_page(unsigned long addr, swp_entry_t * loc)
 244 {
 245         swp_entry_t entry;
 246         int error = 0;
 247
 248         entry = get_swap_page();
 249         if (swp_offset(entry) &&
 250             swapfile_used[swp_type(entry)] == SWAPFILE_SUSPEND) {
 251                 error = rw_swap_page_sync(WRITE, entry,
 252                                           virt_to_page(addr));
 253                 if (error == -EIO)
 254                         error = 0;
 255                 if (!error)
 256                         *loc = entry;
 257         } else
 258                 error = -ENOSPC;
 259         return error;
 260 }
 261
 262
 263 /**
 264  *      data_free - Free the swap entries used by the saved image.
 265  *
 266  *      Walk the list of used swap entries and free each one.
 267  *      This is only used for cleanup when suspend fails.
 268  */
 269
 270 static void data_free(void)
 271 {
 272         swp_entry_t entry;
 273         int i;
 274
 275         for (i = 0; i < nr_copy_pages; i++) {
 276                 entry = (pagedir_nosave + i)->swap_address;
 277                 if (entry.val)
 278                         swap_free(entry);
 279                 else
 280                         break;
 281                 (pagedir_nosave + i)->swap_address = (swp_entry_t){0};
 282         }
 283 }
 284
 285
 286 /**
 287  *      data_write - Write saved image to swap.
 288  *
 289  *      Walk the list of pages in the image and sync each one to swap.
 290  */
 291
 292 static int data_write(void)
 293 {
 294         int error = 0;
 295         int i;
 296         unsigned int mod = nr_copy_pages / 100;
 297
 298         if (!mod)
 299                 mod = 1;
 300
 301         printk( "Writing data to swap (%d pages)...     ", nr_copy_pages );
 302         for (i = 0; i < nr_copy_pages && !error; i++) {
 303                 if (!(i%mod))
 304                         printk( "\b\b\b\b%3d%%", i / mod );
 305                 error = write_page((pagedir_nosave+i)->address,
 306                                           &((pagedir_nosave+i)->swap_address));
 307         }
 308         printk("\b\b\b\bdone\n");
 309         return error;
 310 }
 311
 312 static void dump_info(void)
 313 {
 314         pr_debug(" swsusp: Version: %u\n",swsusp_info.version_code);
 315         pr_debug(" swsusp: Num Pages: %ld\n",swsusp_info.num_physpages);
 316         pr_debug(" swsusp: UTS Sys: %s\n",swsusp_info.uts.sysname);
 317         pr_debug(" swsusp: UTS Node: %s\n",swsusp_info.uts.nodename);
 318         pr_debug(" swsusp: UTS Release: %s\n",swsusp_info.uts.release);
 319         pr_debug(" swsusp: UTS Version: %s\n",swsusp_info.uts.version);
 320         pr_debug(" swsusp: UTS Machine: %s\n",swsusp_info.uts.machine);
 321         pr_debug(" swsusp: UTS Domain: %s\n",swsusp_info.uts.domainname);
 322         pr_debug(" swsusp: CPUs: %d\n",swsusp_info.cpus);
 323         pr_debug(" swsusp: Image: %ld Pages\n",swsusp_info.image_pages);
 324         pr_debug(" swsusp: Pagedir: %ld Pages\n",swsusp_info.pagedir_pages);
 325 }
 326
 327 static void init_header(void)
 328 {
 329         memset(&swsusp_info,0,sizeof(swsusp_info));
 330         swsusp_info.version_code = LINUX_VERSION_CODE;
 331         swsusp_info.num_physpages = num_physpages;
 332         memcpy(&swsusp_info.uts,&system_utsname,sizeof(system_utsname));
 333
 334         swsusp_info.suspend_pagedir = pagedir_nosave;
 335         swsusp_info.cpus = num_online_cpus();
 336         swsusp_info.image_pages = nr_copy_pages;
 337         dump_info();
 338 }
 339
 340 static int close_swap(void)
 341 {
 342         swp_entry_t entry;
 343         int error;
 344
 345         error = write_page((unsigned long)&swsusp_info,&entry);
 346         if (!error) {
 347                 printk( "S" );
 348                 error = mark_swapfiles(entry);
 349                 printk( "|\n" );
 350         }
 351         return error;
 352 }
 353
 354 /**
 355  *      free_pagedir_entries - Free pages used by the page directory.
 356  *
 357  *      This is used during suspend for error recovery.
 358  */
 359
 360 static void free_pagedir_entries(void)
 361 {
 362         int i;
 363
 364         for (i = 0; i < swsusp_info.pagedir_pages; i++)
 365                 swap_free(swsusp_info.pagedir[i]);
 366 }
 367
 368
 369 /**
 370  *      write_pagedir - Write the array of pages holding the page directory.
 371  *      @last:  Last swap entry we write (needed for header).
 372  */
 373
 374 static int write_pagedir(void)
 375 {
 376         unsigned long addr = (unsigned long)pagedir_nosave;
 377         int error = 0;
 378         int n = SUSPEND_PD_PAGES(nr_copy_pages);
 379         int i;
 380
 381         swsusp_info.pagedir_pages = n;
 382         printk( "Writing pagedir (%d pages)\n", n);
 383         for (i = 0; i < n && !error; i++, addr += PAGE_SIZE)
 384                 error = write_page(addr, &swsusp_info.pagedir[i]);
 385         return error;
 386 }
 387
 388 /**
 389  *      write_suspend_image - Write entire image and metadata.
 390  *
 391  */
 392
 393 static int write_suspend_image(void)
 394 {
 395         int error;
 396
 397         init_header();
 398         if ((error = data_write()))
 399                 goto FreeData;
 400
 401         if ((error = write_pagedir()))
 402                 goto FreePagedir;
 403
 404         if ((error = close_swap()))
 405                 goto FreePagedir;
 406  Done:
 407         return error;
 408  FreePagedir:
 409         free_pagedir_entries();
 410  FreeData:
 411         data_free();
 412         goto Done;
 413 }
 414
 415
 416 #ifdef CONFIG_HIGHMEM
 417 struct highmem_page {
 418         char *data;
 419         struct page *page;
 420         struct highmem_page *next;
 421 };
 422
 423 static struct highmem_page *highmem_copy;
 424
 425 static int save_highmem_zone(struct zone *zone)
 426 {
 427         unsigned long zone_pfn;
 428         mark_free_pages(zone);
 429         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
 430                 struct page *page;
 431                 struct highmem_page *save;
 432                 void *kaddr;
 433                 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
 434
 435                 if (!(pfn%1000))
 436                         printk(".");
 437                 if (!pfn_valid(pfn))
 438                         continue;
 439                 page = pfn_to_page(pfn);
 440                 /*
 441                  * This condition results from rvmalloc() sans vmalloc_32()
 442                  * and architectural memory reservations. This should be
 443                  * corrected eventually when the cases giving rise to this
 444                  * are better understood.
 445                  */
 446                 if (PageReserved(page)) {
 447                         printk("highmem reserved page?!\n");
 448                         continue;
 449                 }
 450                 BUG_ON(PageNosave(page));
 451                 if (PageNosaveFree(page))
 452                         continue;
 453                 save = kmalloc(sizeof(struct highmem_page), GFP_ATOMIC);
 454                 if (!save)
 455                         return -ENOMEM;
 456                 save->next = highmem_copy;
 457                 save->page = page;
 458                 save->data = (void *) get_zeroed_page(GFP_ATOMIC);
 459                 if (!save->data) {
 460                         kfree(save);
 461                         return -ENOMEM;
 462                 }
 463                 kaddr = kmap_atomic(page, KM_USER0);
 464                 memcpy(save->data, kaddr, PAGE_SIZE);
 465                 kunmap_atomic(kaddr, KM_USER0);
 466                 highmem_copy = save;
 467         }
 468         return 0;
 469 }
 470 #endif /* CONFIG_HIGHMEM */
 471
 472
 473 static int save_highmem(void)
 474 {
 475 #ifdef CONFIG_HIGHMEM
 476         struct zone *zone;
 477         int res = 0;
 478
 479         pr_debug("swsusp: Saving Highmem\n");
 480         for_each_zone(zone) {
 481                 if (is_highmem(zone))
 482                         res = save_highmem_zone(zone);
 483                 if (res)
 484                         return res;
 485         }
 486 #endif
 487         return 0;
 488 }
 489
 490 static int restore_highmem(void)
 491 {
 492 #ifdef CONFIG_HIGHMEM
 493         printk("swsusp: Restoring Highmem\n");
 494         while (highmem_copy) {
 495                 struct highmem_page *save = highmem_copy;
 496                 void *kaddr;
 497                 highmem_copy = save->next;
 498
 499                 kaddr = kmap_atomic(save->page, KM_USER0);
 500                 memcpy(kaddr, save->data, PAGE_SIZE);
 501                 kunmap_atomic(kaddr, KM_USER0);
 502                 free_page((long) save->data);
 503                 kfree(save);
 504         }
 505 #endif
 506         return 0;
 507 }
 508
 509
 510 static int pfn_is_nosave(unsigned long pfn)
 511 {
 512         unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
 513         unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
 514         return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
 515 }
 516
 517 /**
 518  *      saveable - Determine whether a page should be cloned or not.
 519  *      @pfn:   The page
 520  *
 521  *      We save a page if it's Reserved, and not in the range of pages
 522  *      statically defined as 'unsaveable', or if it isn't reserved, and
 523  *      isn't part of a free chunk of pages.
 524  */
 525
 526 static int saveable(struct zone * zone, unsigned long * zone_pfn)
 527 {
 528         unsigned long pfn = *zone_pfn + zone->zone_start_pfn;
 529         struct page * page;
 530
 531         if (!pfn_valid(pfn))
 532                 return 0;
 533
 534         page = pfn_to_page(pfn);
 535         BUG_ON(PageReserved(page) && PageNosave(page));
 536         if (PageNosave(page))
 537                 return 0;
 538         if (PageReserved(page) && pfn_is_nosave(pfn)) {
 539                 pr_debug("[nosave pfn 0x%lx]", pfn);
 540                 return 0;
 541         }
 542         if (PageNosaveFree(page))
 543                 return 0;
 544
 545         return 1;
 546 }
 547
 548 static void count_data_pages(void)
 549 {
 550         struct zone *zone;
 551         unsigned long zone_pfn;
 552
 553         nr_copy_pages = 0;
 554
 555         for_each_zone(zone) {
 556                 if (is_highmem(zone))
 557                         continue;
 558                 mark_free_pages(zone);
 559                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 560                         nr_copy_pages += saveable(zone, &zone_pfn);
 561         }
 562 }
 563
 564
 565 static void copy_data_pages(void)
 566 {
 567         struct zone *zone;
 568         unsigned long zone_pfn;
 569         struct pbe * pbe = pagedir_nosave;
 570         int to_copy = nr_copy_pages;
 571
 572         for_each_zone(zone) {
 573                 if (is_highmem(zone))
 574                         continue;
 575                 mark_free_pages(zone);
 576                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
 577                         if (saveable(zone, &zone_pfn)) {
 578                                 struct page * page;
 579                                 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
 580                                 pbe->orig_address = (long) page_address(page);
 581                                 /* copy_page is not usable for copying task structs. */
 582                                 memcpy((void *)pbe->address, (void *)pbe->orig_address, PAGE_SIZE);
 583                                 pbe++;
 584                                 to_copy--;
 585                         }
 586                 }
 587         }
 588         BUG_ON(to_copy);
 589 }
 590
 591
 592 /**
 593  *      calc_order - Determine the order of allocation needed for pagedir_save.
 594  *
 595  *      This looks tricky, but is just subtle. Please fix it some time.
 596  *      Since there are %nr_copy_pages worth of pages in the snapshot, we need
 597  *      to allocate enough contiguous space to hold
 598  *              (%nr_copy_pages * sizeof(struct pbe)),
 599  *      which has the saved/orig locations of the page..
 600  *
 601  *      SUSPEND_PD_PAGES() tells us how many pages we need to hold those
 602  *      structures, then we call get_bitmask_order(), which will tell us the
 603  *      last bit set in the number, starting with 1. (If we need 30 pages, that
 604  *      is 0x0000001e in hex. The last bit is the 5th, which is the order we
 605  *      would use to allocate 32 contiguous pages).
 606  *
 607  *      Since we also need to save those pages, we add the number of pages that
 608  *      we need to nr_copy_pages, and in case of an overflow, do the
 609  *      calculation again to update the number of pages needed.
 610  *
 611  *      With this model, we will tend to waste a lot of memory if we just cross
 612  *      an order boundary. Plus, the higher the order of allocation that we try
 613  *      to do, the more likely we are to fail in a low-memory situtation
 614  *      (though we're unlikely to get this far in such a case, since swsusp
 615  *      requires half of memory to be free anyway).
 616  */
 617
 618
 619 static void calc_order(void)
 620 {
 621         int diff = 0;
 622         int order = 0;
 623
 624         do {
 625                 diff = get_bitmask_order(SUSPEND_PD_PAGES(nr_copy_pages)) - order;
 626                 if (diff) {
 627                         order += diff;
 628                         nr_copy_pages += 1 << diff;
 629                 }
 630         } while(diff);
 631         pagedir_order = order;
 632 }
 633
 634
 635 /**
 636  *      alloc_pagedir - Allocate the page directory.
 637  *
 638  *      First, determine exactly how many contiguous pages we need and
 639  *      allocate them.
 640  */
 641
 642 static int alloc_pagedir(void)
 643 {
 644         calc_order();
 645         pagedir_save = (suspend_pagedir_t *)__get_free_pages(GFP_ATOMIC | __GFP_COLD,
 646                                                              pagedir_order);
 647         if (!pagedir_save)
 648                 return -ENOMEM;
 649         memset(pagedir_save, 0, (1 << pagedir_order) * PAGE_SIZE);
 650         pagedir_nosave = pagedir_save;
 651         return 0;
 652 }
 653
 654 /**
 655  *      free_image_pages - Free pages allocated for snapshot
 656  */
 657
 658 static void free_image_pages(void)
 659 {
 660         struct pbe * p;
 661         int i;
 662
 663         p = pagedir_save;
 664         for (i = 0, p = pagedir_save; i < nr_copy_pages; i++, p++) {
 665                 if (p->address) {
 666                         ClearPageNosave(virt_to_page(p->address));
 667                         free_page(p->address);
 668                         p->address = 0;
 669                 }
 670         }
 671 }
 672
 673 /**
 674  *      alloc_image_pages - Allocate pages for the snapshot.
 675  *
 676  */
 677
 678 static int alloc_image_pages(void)
 679 {
 680         struct pbe * p;
 681         int i;
 682
 683         for (i = 0, p = pagedir_save; i < nr_copy_pages; i++, p++) {
 684                 p->address = get_zeroed_page(GFP_ATOMIC | __GFP_COLD);
 685                 if (!p->address)
 686                         return -ENOMEM;
 687                 SetPageNosave(virt_to_page(p->address));
 688         }
 689         return 0;
 690 }
 691
 692 void swsusp_free(void)
 693 {
 694         BUG_ON(PageNosave(virt_to_page(pagedir_save)));
 695         BUG_ON(PageNosaveFree(virt_to_page(pagedir_save)));
 696         free_image_pages();
 697         free_pages((unsigned long) pagedir_save, pagedir_order);
 698 }
 699
 700
 701 /**
 702  *      enough_free_mem - Make sure we enough free memory to snapshot.
 703  *
 704  *      Returns TRUE or FALSE after checking the number of available
 705  *      free pages.
 706  */
 707
 708 static int enough_free_mem(void)
 709 {
 710         if (nr_free_pages() < (nr_copy_pages + PAGES_FOR_IO)) {
 711                 pr_debug("swsusp: Not enough free pages: Have %d\n",
 712                          nr_free_pages());
 713                 return 0;
 714         }
 715         return 1;
 716 }
 717
 718
 719 /**
 720  *      enough_swap - Make sure we have enough swap to save the image.
 721  *
 722  *      Returns TRUE or FALSE after checking the total amount of swap
 723  *      space avaiable.
 724  *
 725  *      FIXME: si_swapinfo(&i) returns all swap devices information.
 726  *      We should only consider resume_device.
 727  */
 728
 729 static int enough_swap(void)
 730 {
 731         struct sysinfo i;
 732
 733         si_swapinfo(&i);
 734         if (i.freeswap < (nr_copy_pages + PAGES_FOR_IO))  {
 735                 pr_debug("swsusp: Not enough swap. Need %ld\n",i.freeswap);
 736                 return 0;
 737         }
 738         return 1;
 739 }
 740
 741 static int swsusp_alloc(void)
 742 {
 743         int error;
 744
 745         pr_debug("suspend: (pages needed: %d + %d free: %d)\n",
 746                  nr_copy_pages, PAGES_FOR_IO, nr_free_pages());
 747
 748         pagedir_nosave = NULL;
 749         if (!enough_free_mem())
 750                 return -ENOMEM;
 751
 752         if (!enough_swap())
 753                 return -ENOSPC;
 754
 755         if ((error = alloc_pagedir())) {
 756                 printk(KERN_ERR "suspend: Allocating pagedir failed.\n");
 757                 return error;
 758         }
 759         if ((error = alloc_image_pages())) {
 760                 printk(KERN_ERR "suspend: Allocating image pages failed.\n");
 761                 swsusp_free();
 762                 return error;
 763         }
 764
 765         nr_copy_pages_check = nr_copy_pages;
 766         pagedir_order_check = pagedir_order;
 767         return 0;
 768 }
 769
 770 static int suspend_prepare_image(void)
 771 {
 772         int error;
 773
 774         pr_debug("swsusp: critical section: \n");
 775         if (save_highmem()) {
 776                 printk(KERN_CRIT "Suspend machine: Not enough free pages for highmem\n");
 777                 restore_highmem();
 778                 return -ENOMEM;
 779         }
 780
 781         drain_local_pages();
 782         count_data_pages();
 783         printk("swsusp: Need to copy %u pages\n",nr_copy_pages);
 784
 785         error = swsusp_alloc();
 786         if (error)
 787                 return error;
 788
 789         /* During allocating of suspend pagedir, new cold pages may appear.
 790          * Kill them.
 791          */
 792         drain_local_pages();
 793         copy_data_pages();
 794
 795         /*
 796          * End of critical section. From now on, we can write to memory,
 797          * but we should not touch disk. This specially means we must _not_
 798          * touch swap space! Except we must write out our image of course.
 799          */
 800
 801         printk("swsusp: critical section/: done (%d pages copied)\n", nr_copy_pages );
 802         return 0;
 803 }
 804
 805
 806 /* It is important _NOT_ to umount filesystems at this point. We want
 807  * them synced (in case something goes wrong) but we DO not want to mark
 808  * filesystem clean: it is not. (And it does not matter, if we resume
 809  * correctly, we'll mark system clean, anyway.)
 810  */
 811 int swsusp_write(void)
 812 {
 813         int error;
 814         device_resume();
 815         lock_swapdevices();
 816         error = write_suspend_image();
 817         /* This will unlock ignored swap devices since writing is finished */
 818         lock_swapdevices();
 819         return error;
 820
 821 }
 822
 823
 824 extern asmlinkage int swsusp_arch_suspend(void);
 825 extern asmlinkage int swsusp_arch_resume(void);
 826
 827
 828 asmlinkage int swsusp_save(void)
 829 {
 830         int error = 0;
 831
 832         if ((error = swsusp_swap_check())) {
 833                 printk(KERN_ERR "swsusp: FATAL: cannot find swap device, try "
 834                                 "swapon -a!\n");
 835                 return error;
 836         }
 837         return suspend_prepare_image();
 838 }
 839
 840 int swsusp_suspend(void)
 841 {
 842         int error;
 843         if ((error = arch_prepare_suspend()))
 844                 return error;
 845         local_irq_disable();
 846         /* At this point, device_suspend() has been called, but *not*
 847          * device_power_down(). We *must* device_power_down() now.
 848          * Otherwise, drivers for some devices (e.g. interrupt controllers)
 849          * become desynchronized with the actual state of the hardware
 850          * at resume time, and evil weirdness ensues.
 851          */
 852         if ((error = device_power_down(PMSG_FREEZE))) {
 853                 local_irq_enable();
 854                 return error;
 855         }
 856         save_processor_state();
 857         error = swsusp_arch_suspend();
 858         /* Restore control flow magically appears here */
 859         restore_processor_state();
 860         restore_highmem();
 861         device_power_up();
 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         device_power_down(PMSG_FREEZE);
 882         /* We'll ignore saved state, but this gets preempt count (etc) right */
 883         save_processor_state();
 884         error = swsusp_arch_resume();
 885         /* Code below is only ever reached in case of failure. Otherwise
 886          * execution continues at place where swsusp_arch_suspend was called
 887          */
 888         BUG_ON(!error);
 889         restore_processor_state();
 890         restore_highmem();
 891         device_power_up();
 892         local_irq_enable();
 893         return error;
 894 }
 895
 896 /* More restore stuff */
 897
 898 /*
 899  * Returns true if given address/order collides with any orig_address
 900  */
 901 static int __init does_collide_order(unsigned long addr, int order)
 902 {
 903         int i;
 904
 905         for (i=0; i < (1<<order); i++)
 906                 if (!PageNosaveFree(virt_to_page(addr + i * PAGE_SIZE)))
 907                         return 1;
 908         return 0;
 909 }
 910
 911 /*
 912  * We check here that pagedir & pages it points to won't collide with pages
 913  * where we're going to restore from the loaded pages later
 914  */
 915 static int __init check_pagedir(void)
 916 {
 917         int i;
 918
 919         for(i=0; i < nr_copy_pages; i++) {
 920                 unsigned long addr;
 921
 922                 do {
 923                         addr = get_zeroed_page(GFP_ATOMIC);
 924                         if(!addr)
 925                                 return -ENOMEM;
 926                 } while (does_collide_order(addr, 0));
 927
 928                 (pagedir_nosave+i)->address = addr;
 929         }
 930         return 0;
 931 }
 932
 933 static int __init swsusp_pagedir_relocate(void)
 934 {
 935         /*
 936          * We have to avoid recursion (not to overflow kernel stack),
 937          * and that's why code looks pretty cryptic
 938          */
 939         suspend_pagedir_t *old_pagedir = pagedir_nosave;
 940         void **eaten_memory = NULL;
 941         void **c = eaten_memory, *m, *f;
 942         int ret = 0;
 943         struct zone *zone;
 944         int i;
 945         struct pbe *p;
 946         unsigned long zone_pfn;
 947
 948         printk("Relocating pagedir ");
 949
 950         /* Set page flags */
 951
 952         for_each_zone(zone) {
 953                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 954                         SetPageNosaveFree(pfn_to_page(zone_pfn +
 955                                         zone->zone_start_pfn));
 956         }
 957
 958         /* Clear orig address */
 959
 960         for(i = 0, p = pagedir_nosave; i < nr_copy_pages; i++, p++) {
 961                 ClearPageNosaveFree(virt_to_page(p->orig_address));
 962         }
 963
 964         if (!does_collide_order((unsigned long)old_pagedir, pagedir_order)) {
 965                 printk("not necessary\n");
 966                 return check_pagedir();
 967         }
 968
 969         while ((m = (void *) __get_free_pages(GFP_ATOMIC, pagedir_order)) != NULL) {
 970                 if (!does_collide_order((unsigned long)m, pagedir_order))
 971                         break;
 972                 eaten_memory = m;
 973                 printk( "." );
 974                 *eaten_memory = c;
 975                 c = eaten_memory;
 976         }
 977
 978         if (!m) {
 979                 printk("out of memory\n");
 980                 ret = -ENOMEM;
 981         } else {
 982                 pagedir_nosave =
 983                         memcpy(m, old_pagedir, PAGE_SIZE << pagedir_order);
 984         }
 985
 986         c = eaten_memory;
 987         while (c) {
 988                 printk(":");
 989                 f = c;
 990                 c = *c;
 991                 free_pages((unsigned long)f, pagedir_order);
 992         }
 993         if (ret)
 994                 return ret;
 995         printk("|\n");
 996         return check_pagedir();
 997 }
 998
 999 /**
1000  *      Using bio to read from swap.
1001  *      This code requires a bit more work than just using buffer heads
1002  *      but, it is the recommended way for 2.5/2.6.
1003  *      The following are to signal the beginning and end of I/O. Bios
1004  *      finish asynchronously, while we want them to happen synchronously.
1005  *      A simple atomic_t, and a wait loop take care of this problem.
1006  */
1007
1008 static atomic_t io_done = ATOMIC_INIT(0);
1009
1010 static int end_io(struct bio * bio, unsigned int num, int err)
1011 {
1012         if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1013                 panic("I/O error reading memory image");
1014         atomic_set(&io_done, 0);
1015         return 0;
1016 }
1017
1018 static struct block_device * resume_bdev;
1019
1020 /**
1021  *      submit - submit BIO request.
1022  *      @rw:    READ or WRITE.
1023  *      @off    physical offset of page.
1024  *      @page:  page we're reading or writing.
1025  *
1026  *      Straight from the textbook - allocate and initialize the bio.
1027  *      If we're writing, make sure the page is marked as dirty.
1028  *      Then submit it and wait.
1029  */
1030
1031 static int submit(int rw, pgoff_t page_off, void * page)
1032 {
1033         int error = 0;
1034         struct bio * bio;
1035
1036         bio = bio_alloc(GFP_ATOMIC, 1);
1037         if (!bio)
1038                 return -ENOMEM;
1039         bio->bi_sector = page_off * (PAGE_SIZE >> 9);
1040         bio_get(bio);
1041         bio->bi_bdev = resume_bdev;
1042         bio->bi_end_io = end_io;
1043
1044         if (bio_add_page(bio, virt_to_page(page), PAGE_SIZE, 0) < PAGE_SIZE) {
1045                 printk("swsusp: ERROR: adding page to bio at %ld\n",page_off);
1046                 error = -EFAULT;
1047                 goto Done;
1048         }
1049
1050         if (rw == WRITE)
1051                 bio_set_pages_dirty(bio);
1052
1053         atomic_set(&io_done, 1);
1054         submit_bio(rw | (1 << BIO_RW_SYNC), bio);
1055         while (atomic_read(&io_done))
1056                 yield();
1057
1058  Done:
1059         bio_put(bio);
1060         return error;
1061 }
1062
1063 static int bio_read_page(pgoff_t page_off, void * page)
1064 {
1065         return submit(READ, page_off, page);
1066 }
1067
1068 static int bio_write_page(pgoff_t page_off, void * page)
1069 {
1070         return submit(WRITE, page_off, page);
1071 }
1072
1073 /*
1074  * Sanity check if this image makes sense with this kernel/swap context
1075  * I really don't think that it's foolproof but more than nothing..
1076  */
1077
1078 static const char * __init sanity_check(void)
1079 {
1080         dump_info();
1081         if(swsusp_info.version_code != LINUX_VERSION_CODE)
1082                 return "kernel version";
1083         if(swsusp_info.num_physpages != num_physpages)
1084                 return "memory size";
1085         if (strcmp(swsusp_info.uts.sysname,system_utsname.sysname))
1086                 return "system type";
1087         if (strcmp(swsusp_info.uts.release,system_utsname.release))
1088                 return "kernel release";
1089         if (strcmp(swsusp_info.uts.version,system_utsname.version))
1090                 return "version";
1091         if (strcmp(swsusp_info.uts.machine,system_utsname.machine))
1092                 return "machine";
1093         if(swsusp_info.cpus != num_online_cpus())
1094                 return "number of cpus";
1095         return NULL;
1096 }
1097
1098
1099 static int __init check_header(void)
1100 {
1101         const char * reason = NULL;
1102         int error;
1103
1104         if ((error = bio_read_page(swp_offset(swsusp_header.swsusp_info), &swsusp_info)))
1105                 return error;
1106
1107         /* Is this same machine? */
1108         if ((reason = sanity_check())) {
1109                 printk(KERN_ERR "swsusp: Resume mismatch: %s\n",reason);
1110                 return -EPERM;
1111         }
1112         nr_copy_pages = swsusp_info.image_pages;
1113         pagedir_order = get_bitmask_order(SUSPEND_PD_PAGES(nr_copy_pages));
1114         return error;
1115 }
1116
1117 static int __init check_sig(void)
1118 {
1119         int error;
1120
1121         memset(&swsusp_header, 0, sizeof(swsusp_header));
1122         if ((error = bio_read_page(0, &swsusp_header)))
1123                 return error;
1124         if (!memcmp(SWSUSP_SIG, swsusp_header.sig, 10)) {
1125                 memcpy(swsusp_header.sig, swsusp_header.orig_sig, 10);
1126
1127                 /*
1128                  * Reset swap signature now.
1129                  */
1130                 error = bio_write_page(0, &swsusp_header);
1131         } else {
1132                 pr_debug(KERN_ERR "swsusp: Suspend partition has wrong signature?\n");
1133                 return -EINVAL;
1134         }
1135         if (!error)
1136                 pr_debug("swsusp: Signature found, resuming\n");
1137         return error;
1138 }
1139
1140 /**
1141  *      swsusp_read_data - Read image pages from swap.
1142  *
1143  *      You do not need to check for overlaps, check_pagedir()
1144  *      already did that.
1145  */
1146
1147 static int __init data_read(void)
1148 {
1149         struct pbe * p;
1150         int error;
1151         int i;
1152         int mod = nr_copy_pages / 100;
1153
1154         if (!mod)
1155                 mod = 1;
1156
1157         if ((error = swsusp_pagedir_relocate()))
1158                 return error;
1159
1160         printk( "Reading image data (%d pages):     ", nr_copy_pages );
1161         for(i = 0, p = pagedir_nosave; i < nr_copy_pages && !error; i++, p++) {
1162                 if (!(i%mod))
1163                         printk( "\b\b\b\b%3d%%", i / mod );
1164                 error = bio_read_page(swp_offset(p->swap_address),
1165                                   (void *)p->address);
1166         }
1167         printk(" %d done.\n",i);
1168         return error;
1169
1170 }
1171
1172 extern dev_t __init name_to_dev_t(const char *line);
1173
1174 static int __init read_pagedir(void)
1175 {
1176         unsigned long addr;
1177         int i, n = swsusp_info.pagedir_pages;
1178         int error = 0;
1179
1180         addr = __get_free_pages(GFP_ATOMIC, pagedir_order);
1181         if (!addr)
1182                 return -ENOMEM;
1183         pagedir_nosave = (struct pbe *)addr;
1184
1185         pr_debug("swsusp: Reading pagedir (%d Pages)\n",n);
1186
1187         for (i = 0; i < n && !error; i++, addr += PAGE_SIZE) {
1188                 unsigned long offset = swp_offset(swsusp_info.pagedir[i]);
1189                 if (offset)
1190                         error = bio_read_page(offset, (void *)addr);
1191                 else
1192                         error = -EFAULT;
1193         }
1194         if (error)
1195                 free_pages((unsigned long)pagedir_nosave, pagedir_order);
1196         return error;
1197 }
1198
1199 static int __init read_suspend_image(void)
1200 {
1201         int error = 0;
1202
1203         if ((error = check_sig()))
1204                 return error;
1205         if ((error = check_header()))
1206                 return error;
1207         if ((error = read_pagedir()))
1208                 return error;
1209         if ((error = data_read()))
1210                 free_pages((unsigned long)pagedir_nosave, pagedir_order);
1211         return error;
1212 }
1213
1214 /**
1215  *      swsusp_read - Read saved image from swap.
1216  */
1217
1218 int __init swsusp_read(void)
1219 {
1220         int error;
1221
1222         if (!strlen(resume_file))
1223                 return -ENOENT;
1224
1225         resume_device = name_to_dev_t(resume_file);
1226         pr_debug("swsusp: Resume From Partition: %s\n", resume_file);
1227
1228         resume_bdev = open_by_devnum(resume_device, FMODE_READ);
1229         if (!IS_ERR(resume_bdev)) {
1230                 set_blocksize(resume_bdev, PAGE_SIZE);
1231                 error = read_suspend_image();
1232                 blkdev_put(resume_bdev);
1233         } else
1234                 error = PTR_ERR(resume_bdev);
1235
1236         if (!error)
1237                 pr_debug("Reading resume file was successful\n");
1238         else
1239                 pr_debug("swsusp: Error %d resuming\n", error);
1240         return error;
1241 }