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         unsigned int mod = nr_copy_pages / 100;
 298
 299         if (!mod)
 300                 mod = 1;
 301
 302         printk( "Writing data to swap (%d pages)...     ", nr_copy_pages );
 303         for (i = 0; i < nr_copy_pages && !error; i++) {
 304                 if (!(i%mod))
 305                         printk( "\b\b\b\b%3d%%", i / mod );
 306                 error = write_page((pagedir_nosave+i)->address,
 307                                           &((pagedir_nosave+i)->swap_address));
 308         }
 309         printk("\b\b\b\bdone\n");
 310         return error;
 311 }
 312
 313 static void dump_info(void)
 314 {
 315         pr_debug(" swsusp: Version: %u\n",swsusp_info.version_code);
 316         pr_debug(" swsusp: Num Pages: %ld\n",swsusp_info.num_physpages);
 317         pr_debug(" swsusp: UTS Sys: %s\n",swsusp_info.uts.sysname);
 318         pr_debug(" swsusp: UTS Node: %s\n",swsusp_info.uts.nodename);
 319         pr_debug(" swsusp: UTS Release: %s\n",swsusp_info.uts.release);
 320         pr_debug(" swsusp: UTS Version: %s\n",swsusp_info.uts.version);
 321         pr_debug(" swsusp: UTS Machine: %s\n",swsusp_info.uts.machine);
 322         pr_debug(" swsusp: UTS Domain: %s\n",swsusp_info.uts.domainname);
 323         pr_debug(" swsusp: CPUs: %d\n",swsusp_info.cpus);
 324         pr_debug(" swsusp: Image: %ld Pages\n",swsusp_info.image_pages);
 325         pr_debug(" swsusp: Pagedir: %ld Pages\n",swsusp_info.pagedir_pages);
 326 }
 327
 328 static void init_header(void)
 329 {
 330         memset(&swsusp_info,0,sizeof(swsusp_info));
 331         swsusp_info.version_code = LINUX_VERSION_CODE;
 332         swsusp_info.num_physpages = num_physpages;
 333         memcpy(&swsusp_info.uts,&system_utsname,sizeof(system_utsname));
 334
 335         swsusp_info.suspend_pagedir = pagedir_nosave;
 336         swsusp_info.cpus = num_online_cpus();
 337         swsusp_info.image_pages = nr_copy_pages;
 338         dump_info();
 339 }
 340
 341 static int close_swap(void)
 342 {
 343         swp_entry_t entry;
 344         int error;
 345
 346         error = write_page((unsigned long)&swsusp_info,&entry);
 347         if (!error) {
 348                 printk( "S" );
 349                 error = mark_swapfiles(entry);
 350                 printk( "|\n" );
 351         }
 352         return error;
 353 }
 354
 355 /**
 356  *      free_pagedir_entries - Free pages used by the page directory.
 357  *
 358  *      This is used during suspend for error recovery.
 359  */
 360
 361 static void free_pagedir_entries(void)
 362 {
 363         int i;
 364
 365         for (i = 0; i < swsusp_info.pagedir_pages; i++)
 366                 swap_free(swsusp_info.pagedir[i]);
 367 }
 368
 369
 370 /**
 371  *      write_pagedir - Write the array of pages holding the page directory.
 372  *      @last:  Last swap entry we write (needed for header).
 373  */
 374
 375 static int write_pagedir(void)
 376 {
 377         unsigned long addr = (unsigned long)pagedir_nosave;
 378         int error = 0;
 379         int n = SUSPEND_PD_PAGES(nr_copy_pages);
 380         int i;
 381
 382         swsusp_info.pagedir_pages = n;
 383         printk( "Writing pagedir (%d pages)\n", n);
 384         for (i = 0; i < n && !error; i++, addr += PAGE_SIZE)
 385                 error = write_page(addr, &swsusp_info.pagedir[i]);
 386         return error;
 387 }
 388
 389 /**
 390  *      write_suspend_image - Write entire image and metadata.
 391  *
 392  */
 393
 394 static int write_suspend_image(void)
 395 {
 396         int error;
 397
 398         init_header();
 399         if ((error = data_write()))
 400                 goto FreeData;
 401
 402         if ((error = write_pagedir()))
 403                 goto FreePagedir;
 404
 405         if ((error = close_swap()))
 406                 goto FreePagedir;
 407  Done:
 408         return error;
 409  FreePagedir:
 410         free_pagedir_entries();
 411  FreeData:
 412         data_free();
 413         goto Done;
 414 }
 415
 416
 417 #ifdef CONFIG_HIGHMEM
 418 struct highmem_page {
 419         char *data;
 420         struct page *page;
 421         struct highmem_page *next;
 422 };
 423
 424 struct highmem_page *highmem_copy = NULL;
 425
 426 static int save_highmem_zone(struct zone *zone)
 427 {
 428         unsigned long zone_pfn;
 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                 int chunk_size;
 435
 436                 if (!(pfn%1000))
 437                         printk(".");
 438                 if (!pfn_valid(pfn))
 439                         continue;
 440                 page = pfn_to_page(pfn);
 441                 /*
 442                  * This condition results from rvmalloc() sans vmalloc_32()
 443                  * and architectural memory reservations. This should be
 444                  * corrected eventually when the cases giving rise to this
 445                  * are better understood.
 446                  */
 447                 if (PageReserved(page)) {
 448                         printk("highmem reserved page?!\n");
 449                         continue;
 450                 }
 451                 if ((chunk_size = is_head_of_free_region(page))) {
 452                         pfn += chunk_size - 1;
 453                         zone_pfn += chunk_size - 1;
 454                         continue;
 455                 }
 456                 save = kmalloc(sizeof(struct highmem_page), GFP_ATOMIC);
 457                 if (!save)
 458                         return -ENOMEM;
 459                 save->next = highmem_copy;
 460                 save->page = page;
 461                 save->data = (void *) get_zeroed_page(GFP_ATOMIC);
 462                 if (!save->data) {
 463                         kfree(save);
 464                         return -ENOMEM;
 465                 }
 466                 kaddr = kmap_atomic(page, KM_USER0);
 467                 memcpy(save->data, kaddr, PAGE_SIZE);
 468                 kunmap_atomic(kaddr, KM_USER0);
 469                 highmem_copy = save;
 470         }
 471         return 0;
 472 }
 473 #endif /* CONFIG_HIGHMEM */
 474
 475
 476 static int save_highmem(void)
 477 {
 478 #ifdef CONFIG_HIGHMEM
 479         struct zone *zone;
 480         int res = 0;
 481
 482         pr_debug("swsusp: Saving Highmem\n");
 483         for_each_zone(zone) {
 484                 if (is_highmem(zone))
 485                         res = save_highmem_zone(zone);
 486                 if (res)
 487                         return res;
 488         }
 489 #endif
 490         return 0;
 491 }
 492
 493 static int restore_highmem(void)
 494 {
 495 #ifdef CONFIG_HIGHMEM
 496         printk("swsusp: Restoring Highmem\n");
 497         while (highmem_copy) {
 498                 struct highmem_page *save = highmem_copy;
 499                 void *kaddr;
 500                 highmem_copy = save->next;
 501
 502                 kaddr = kmap_atomic(save->page, KM_USER0);
 503                 memcpy(kaddr, save->data, PAGE_SIZE);
 504                 kunmap_atomic(kaddr, KM_USER0);
 505                 free_page((long) save->data);
 506                 kfree(save);
 507         }
 508 #endif
 509         return 0;
 510 }
 511
 512
 513 static int pfn_is_nosave(unsigned long pfn)
 514 {
 515         unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
 516         unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
 517         return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
 518 }
 519
 520 /**
 521  *      saveable - Determine whether a page should be cloned or not.
 522  *      @pfn:   The page
 523  *
 524  *      We save a page if it's Reserved, and not in the range of pages
 525  *      statically defined as 'unsaveable', or if it isn't reserved, and
 526  *      isn't part of a free chunk of pages.
 527  *      If it is part of a free chunk, we update @pfn to point to the last
 528  *      page of the chunk.
 529  */
 530
 531 static int saveable(struct zone * zone, unsigned long * zone_pfn)
 532 {
 533         unsigned long pfn = *zone_pfn + zone->zone_start_pfn;
 534         unsigned long chunk_size;
 535         struct page * page;
 536
 537         if (!pfn_valid(pfn))
 538                 return 0;
 539
 540         if (!(pfn%1000))
 541                 printk(".");
 542         page = pfn_to_page(pfn);
 543         BUG_ON(PageReserved(page) && PageNosave(page));
 544         if (PageNosave(page))
 545                 return 0;
 546         if (PageReserved(page) && pfn_is_nosave(pfn)) {
 547                 pr_debug("[nosave pfn 0x%lx]", pfn);
 548                 return 0;
 549         }
 550         if ((chunk_size = is_head_of_free_region(page))) {
 551                 *zone_pfn += chunk_size - 1;
 552                 return 0;
 553         }
 554
 555         return 1;
 556 }
 557
 558 static void count_data_pages(void)
 559 {
 560         struct zone *zone;
 561         unsigned long zone_pfn;
 562
 563         nr_copy_pages = 0;
 564
 565         for_each_zone(zone) {
 566                 if (!is_highmem(zone)) {
 567                         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 568                                 nr_copy_pages += saveable(zone, &zone_pfn);
 569                 }
 570         }
 571 }
 572
 573
 574 static void copy_data_pages(void)
 575 {
 576         struct zone *zone;
 577         unsigned long zone_pfn;
 578         struct pbe * pbe = pagedir_nosave;
 579
 580         for_each_zone(zone) {
 581                 if (!is_highmem(zone))
 582                         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
 583                                 if (saveable(zone, &zone_pfn)) {
 584                                         struct page * page;
 585                                         page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
 586                                         pbe->orig_address = (long) page_address(page);
 587                                         /* copy_page is no usable for copying task structs. */
 588                                         memcpy((void *)pbe->address, (void *)pbe->orig_address, PAGE_SIZE);
 589                                         pbe++;
 590                                 }
 591                         }
 592         }
 593 }
 594
 595
 596 static void free_suspend_pagedir_zone(struct zone *zone, unsigned long pagedir)
 597 {
 598         unsigned long zone_pfn, pagedir_end, pagedir_pfn, pagedir_end_pfn;
 599         pagedir_end = pagedir + (PAGE_SIZE << pagedir_order);
 600         pagedir_pfn = __pa(pagedir) >> PAGE_SHIFT;
 601         pagedir_end_pfn = __pa(pagedir_end) >> PAGE_SHIFT;
 602         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
 603                 struct page *page;
 604                 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
 605                 if (!pfn_valid(pfn))
 606                         continue;
 607                 page = pfn_to_page(pfn);
 608                 if (!TestClearPageNosave(page))
 609                         continue;
 610                 else if (pfn >= pagedir_pfn && pfn < pagedir_end_pfn)
 611                         continue;
 612                 __free_page(page);
 613         }
 614 }
 615
 616 void swsusp_free(void)
 617 {
 618         unsigned long p = (unsigned long)pagedir_save;
 619         struct zone *zone;
 620         for_each_zone(zone) {
 621                 if (!is_highmem(zone))
 622                         free_suspend_pagedir_zone(zone, p);
 623         }
 624         free_pages(p, pagedir_order);
 625 }
 626
 627
 628 /**
 629  *      calc_order - Determine the order of allocation needed for pagedir_save.
 630  *
 631  *      This looks tricky, but is just subtle. Please fix it some time.
 632  *      Since there are %nr_copy_pages worth of pages in the snapshot, we need
 633  *      to allocate enough contiguous space to hold
 634  *              (%nr_copy_pages * sizeof(struct pbe)),
 635  *      which has the saved/orig locations of the page..
 636  *
 637  *      SUSPEND_PD_PAGES() tells us how many pages we need to hold those
 638  *      structures, then we call get_bitmask_order(), which will tell us the
 639  *      last bit set in the number, starting with 1. (If we need 30 pages, that
 640  *      is 0x0000001e in hex. The last bit is the 5th, which is the order we
 641  *      would use to allocate 32 contiguous pages).
 642  *
 643  *      Since we also need to save those pages, we add the number of pages that
 644  *      we need to nr_copy_pages, and in case of an overflow, do the
 645  *      calculation again to update the number of pages needed.
 646  *
 647  *      With this model, we will tend to waste a lot of memory if we just cross
 648  *      an order boundary. Plus, the higher the order of allocation that we try
 649  *      to do, the more likely we are to fail in a low-memory situtation
 650  *      (though we're unlikely to get this far in such a case, since swsusp
 651  *      requires half of memory to be free anyway).
 652  */
 653
 654
 655 static void calc_order(void)
 656 {
 657         int diff = 0;
 658         int order = 0;
 659
 660         do {
 661                 diff = get_bitmask_order(SUSPEND_PD_PAGES(nr_copy_pages)) - order;
 662                 if (diff) {
 663                         order += diff;
 664                         nr_copy_pages += 1 << diff;
 665                 }
 666         } while(diff);
 667         pagedir_order = order;
 668 }
 669
 670
 671 /**
 672  *      alloc_pagedir - Allocate the page directory.
 673  *
 674  *      First, determine exactly how many contiguous pages we need and
 675  *      allocate them.
 676  */
 677
 678 static int alloc_pagedir(void)
 679 {
 680         calc_order();
 681         pagedir_save = (suspend_pagedir_t *)__get_free_pages(GFP_ATOMIC | __GFP_COLD,
 682                                                              pagedir_order);
 683         if (!pagedir_save)
 684                 return -ENOMEM;
 685         memset(pagedir_save, 0, (1 << pagedir_order) * PAGE_SIZE);
 686         pagedir_nosave = pagedir_save;
 687         return 0;
 688 }
 689
 690
 691 /**
 692  *      alloc_image_pages - Allocate pages for the snapshot.
 693  *
 694  */
 695
 696 static int alloc_image_pages(void)
 697 {
 698         struct pbe * p;
 699         int i;
 700
 701         for (i = 0, p = pagedir_save; i < nr_copy_pages; i++, p++) {
 702                 p->address = get_zeroed_page(GFP_ATOMIC | __GFP_COLD);
 703                 if(!p->address)
 704                         goto Error;
 705                 SetPageNosave(virt_to_page(p->address));
 706         }
 707         return 0;
 708  Error:
 709         do {
 710                 if (p->address)
 711                         free_page(p->address);
 712                 p->address = 0;
 713         } while (p-- > pagedir_save);
 714         return -ENOMEM;
 715 }
 716
 717
 718 /**
 719  *      enough_free_mem - Make sure we enough free memory to snapshot.
 720  *
 721  *      Returns TRUE or FALSE after checking the number of available
 722  *      free pages.
 723  */
 724
 725 static int enough_free_mem(void)
 726 {
 727         if (nr_free_pages() < (nr_copy_pages + PAGES_FOR_IO)) {
 728                 pr_debug("swsusp: Not enough free pages: Have %d\n",
 729                          nr_free_pages());
 730                 return 0;
 731         }
 732         return 1;
 733 }
 734
 735
 736 /**
 737  *      enough_swap - Make sure we have enough swap to save the image.
 738  *
 739  *      Returns TRUE or FALSE after checking the total amount of swap
 740  *      space avaiable.
 741  *
 742  *      FIXME: si_swapinfo(&i) returns all swap devices information.
 743  *      We should only consider resume_device.
 744  */
 745
 746 static int enough_swap(void)
 747 {
 748         struct sysinfo i;
 749
 750         si_swapinfo(&i);
 751         if (i.freeswap < (nr_copy_pages + PAGES_FOR_IO))  {
 752                 pr_debug("swsusp: Not enough swap. Need %ld\n",i.freeswap);
 753                 return 0;
 754         }
 755         return 1;
 756 }
 757
 758 static int swsusp_alloc(void)
 759 {
 760         int error;
 761
 762         pr_debug("suspend: (pages needed: %d + %d free: %d)\n",
 763                  nr_copy_pages, PAGES_FOR_IO, nr_free_pages());
 764
 765         pagedir_nosave = NULL;
 766         if (!enough_free_mem())
 767                 return -ENOMEM;
 768
 769         if (!enough_swap())
 770                 return -ENOSPC;
 771
 772         if ((error = alloc_pagedir())) {
 773                 pr_debug("suspend: Allocating pagedir failed.\n");
 774                 return error;
 775         }
 776         if ((error = alloc_image_pages())) {
 777                 pr_debug("suspend: Allocating image pages failed.\n");
 778                 swsusp_free();
 779                 return error;
 780         }
 781
 782         nr_copy_pages_check = nr_copy_pages;
 783         pagedir_order_check = pagedir_order;
 784         return 0;
 785 }
 786
 787 int suspend_prepare_image(void)
 788 {
 789         unsigned int nr_needed_pages;
 790         int error;
 791
 792         pr_debug("swsusp: critical section: \n");
 793         if (save_highmem()) {
 794                 printk(KERN_CRIT "Suspend machine: Not enough free pages for highmem\n");
 795                 restore_highmem();
 796                 return -ENOMEM;
 797         }
 798
 799         drain_local_pages();
 800         count_data_pages();
 801         printk("swsusp: Need to copy %u pages\n",nr_copy_pages);
 802         nr_needed_pages = nr_copy_pages + PAGES_FOR_IO;
 803
 804         error = swsusp_alloc();
 805         if (error)
 806                 return error;
 807
 808         /* During allocating of suspend pagedir, new cold pages may appear.
 809          * Kill them.
 810          */
 811         drain_local_pages();
 812         copy_data_pages();
 813
 814         /*
 815          * End of critical section. From now on, we can write to memory,
 816          * but we should not touch disk. This specially means we must _not_
 817          * touch swap space! Except we must write out our image of course.
 818          */
 819
 820         printk("swsusp: critical section/: done (%d pages copied)\n", nr_copy_pages );
 821         return 0;
 822 }
 823
 824
 825 /* It is important _NOT_ to umount filesystems at this point. We want
 826  * them synced (in case something goes wrong) but we DO not want to mark
 827  * filesystem clean: it is not. (And it does not matter, if we resume
 828  * correctly, we'll mark system clean, anyway.)
 829  */
 830 int swsusp_write(void)
 831 {
 832         int error;
 833         device_resume();
 834         lock_swapdevices();
 835         error = write_suspend_image();
 836         /* This will unlock ignored swap devices since writing is finished */
 837         lock_swapdevices();
 838         return error;
 839
 840 }
 841
 842
 843 extern asmlinkage int swsusp_arch_suspend(void);
 844 extern asmlinkage int swsusp_arch_resume(void);
 845
 846
 847 asmlinkage int swsusp_save(void)
 848 {
 849         int error = 0;
 850
 851         if ((error = swsusp_swap_check())) {
 852                 printk(KERN_ERR "swsusp: FATAL: cannot find swap device, try "
 853                                 "swapon -a!\n");
 854                 return error;
 855         }
 856         return suspend_prepare_image();
 857 }
 858
 859 int swsusp_suspend(void)
 860 {
 861         int error;
 862         if ((error = arch_prepare_suspend()))
 863                 return error;
 864         local_irq_disable();
 865         save_processor_state();
 866         error = swsusp_arch_suspend();
 867         /* Restore control flow magically appears here */
 868         restore_processor_state();
 869         restore_highmem();
 870         local_irq_enable();
 871         return error;
 872 }
 873
 874
 875 asmlinkage int swsusp_restore(void)
 876 {
 877         BUG_ON (nr_copy_pages_check != nr_copy_pages);
 878         BUG_ON (pagedir_order_check != pagedir_order);
 879
 880         /* Even mappings of "global" things (vmalloc) need to be fixed */
 881         __flush_tlb_global();
 882         return 0;
 883 }
 884
 885 int swsusp_resume(void)
 886 {
 887         int error;
 888         local_irq_disable();
 889         /* We'll ignore saved state, but this gets preempt count (etc) right */
 890         save_processor_state();
 891         error = swsusp_arch_resume();
 892         /* Code below is only ever reached in case of failure. Otherwise
 893          * execution continues at place where swsusp_arch_suspend was called
 894          */
 895         BUG_ON(!error);
 896         restore_processor_state();
 897         restore_highmem();
 898         local_irq_enable();
 899         return error;
 900 }
 901
 902
 903
 904 /* More restore stuff */
 905
 906 #define does_collide(addr) does_collide_order(pagedir_nosave, addr, 0)
 907
 908 /*
 909  * Returns true if given address/order collides with any orig_address
 910  */
 911 static int __init does_collide_order(suspend_pagedir_t *pagedir, unsigned long addr,
 912                 int order)
 913 {
 914         int i;
 915         unsigned long addre = addr + (PAGE_SIZE<<order);
 916
 917         for (i=0; i < nr_copy_pages; i++)
 918                 if ((pagedir+i)->orig_address >= addr &&
 919                         (pagedir+i)->orig_address < addre)
 920                         return 1;
 921
 922         return 0;
 923 }
 924
 925 /*
 926  * We check here that pagedir & pages it points to won't collide with pages
 927  * where we're going to restore from the loaded pages later
 928  */
 929 static int __init check_pagedir(void)
 930 {
 931         int i;
 932
 933         for(i=0; i < nr_copy_pages; i++) {
 934                 unsigned long addr;
 935
 936                 do {
 937                         addr = get_zeroed_page(GFP_ATOMIC);
 938                         if(!addr)
 939                                 return -ENOMEM;
 940                 } while (does_collide(addr));
 941
 942                 (pagedir_nosave+i)->address = addr;
 943         }
 944         return 0;
 945 }
 946
 947 static int __init swsusp_pagedir_relocate(void)
 948 {
 949         /*
 950          * We have to avoid recursion (not to overflow kernel stack),
 951          * and that's why code looks pretty cryptic
 952          */
 953         suspend_pagedir_t *old_pagedir = pagedir_nosave;
 954         void **eaten_memory = NULL;
 955         void **c = eaten_memory, *m, *f;
 956         int ret = 0;
 957
 958         printk("Relocating pagedir ");
 959
 960         if (!does_collide_order(old_pagedir, (unsigned long)old_pagedir, pagedir_order)) {
 961                 printk("not necessary\n");
 962                 return check_pagedir();
 963         }
 964
 965         while ((m = (void *) __get_free_pages(GFP_ATOMIC, pagedir_order)) != NULL) {
 966                 if (!does_collide_order(old_pagedir, (unsigned long)m, pagedir_order))
 967                         break;
 968                 eaten_memory = m;
 969                 printk( "." );
 970                 *eaten_memory = c;
 971                 c = eaten_memory;
 972         }
 973
 974         if (!m) {
 975                 printk("out of memory\n");
 976                 ret = -ENOMEM;
 977         } else {
 978                 pagedir_nosave =
 979                         memcpy(m, old_pagedir, PAGE_SIZE << pagedir_order);
 980         }
 981
 982         c = eaten_memory;
 983         while (c) {
 984                 printk(":");
 985                 f = c;
 986                 c = *c;
 987                 free_pages((unsigned long)f, pagedir_order);
 988         }
 989         if (ret)
 990                 return ret;
 991         printk("|\n");
 992         return check_pagedir();
 993 }
 994
 995 /**
 996  *      Using bio to read from swap.
 997  *      This code requires a bit more work than just using buffer heads
 998  *      but, it is the recommended way for 2.5/2.6.
 999  *      The following are to signal the beginning and end of I/O. Bios
1000  *      finish asynchronously, while we want them to happen synchronously.
1001  *      A simple atomic_t, and a wait loop take care of this problem.
1002  */
1003
1004 static atomic_t io_done = ATOMIC_INIT(0);
1005
1006 static void start_io(void)
1007 {
1008         atomic_set(&io_done,1);
1009 }
1010
1011 static int end_io(struct bio * bio, unsigned int num, int err)
1012 {
1013         atomic_set(&io_done,0);
1014         return 0;
1015 }
1016
1017 static void wait_io(void)
1018 {
1019         while(atomic_read(&io_done))
1020                 io_schedule();
1021 }
1022
1023
1024 static struct block_device * resume_bdev;
1025
1026 /**
1027  *      submit - submit BIO request.
1028  *      @rw:    READ or WRITE.
1029  *      @off    physical offset of page.
1030  *      @page:  page we're reading or writing.
1031  *
1032  *      Straight from the textbook - allocate and initialize the bio.
1033  *      If we're writing, make sure the page is marked as dirty.
1034  *      Then submit it and wait.
1035  */
1036
1037 static int submit(int rw, pgoff_t page_off, void * page)
1038 {
1039         int error = 0;
1040         struct bio * bio;
1041
1042         bio = bio_alloc(GFP_ATOMIC, 1);
1043         if (!bio)
1044                 return -ENOMEM;
1045         bio->bi_sector = page_off * (PAGE_SIZE >> 9);
1046         bio_get(bio);
1047         bio->bi_bdev = resume_bdev;
1048         bio->bi_end_io = end_io;
1049
1050         if (bio_add_page(bio, virt_to_page(page), PAGE_SIZE, 0) < PAGE_SIZE) {
1051                 printk("swsusp: ERROR: adding page to bio at %ld\n",page_off);
1052                 error = -EFAULT;
1053                 goto Done;
1054         }
1055
1056         if (rw == WRITE)
1057                 bio_set_pages_dirty(bio);
1058         start_io();
1059         submit_bio(rw | (1 << BIO_RW_SYNC), bio);
1060         wait_io();
1061  Done:
1062         bio_put(bio);
1063         return error;
1064 }
1065
1066 int bio_read_page(pgoff_t page_off, void * page)
1067 {
1068         return submit(READ, page_off, page);
1069 }
1070
1071 int bio_write_page(pgoff_t page_off, void * page)
1072 {
1073         return submit(WRITE, page_off, page);
1074 }
1075
1076 /*
1077  * Sanity check if this image makes sense with this kernel/swap context
1078  * I really don't think that it's foolproof but more than nothing..
1079  */
1080
1081 static const char * __init sanity_check(void)
1082 {
1083         dump_info();
1084         if(swsusp_info.version_code != LINUX_VERSION_CODE)
1085                 return "kernel version";
1086         if(swsusp_info.num_physpages != num_physpages)
1087                 return "memory size";
1088         if (strcmp(swsusp_info.uts.sysname,system_utsname.sysname))
1089                 return "system type";
1090         if (strcmp(swsusp_info.uts.release,system_utsname.release))
1091                 return "kernel release";
1092         if (strcmp(swsusp_info.uts.version,system_utsname.version))
1093                 return "version";
1094         if (strcmp(swsusp_info.uts.machine,system_utsname.machine))
1095                 return "machine";
1096         if(swsusp_info.cpus != num_online_cpus())
1097                 return "number of cpus";
1098         return NULL;
1099 }
1100
1101
1102 static int __init check_header(void)
1103 {
1104         const char * reason = NULL;
1105         int error;
1106
1107         if ((error = bio_read_page(swp_offset(swsusp_header.swsusp_info), &swsusp_info)))
1108                 return error;
1109
1110         /* Is this same machine? */
1111         if ((reason = sanity_check())) {
1112                 printk(KERN_ERR "swsusp: Resume mismatch: %s\n",reason);
1113                 return -EPERM;
1114         }
1115         nr_copy_pages = swsusp_info.image_pages;
1116         return error;
1117 }
1118
1119 static int __init check_sig(void)
1120 {
1121         int error;
1122
1123         memset(&swsusp_header, 0, sizeof(swsusp_header));
1124         if ((error = bio_read_page(0, &swsusp_header)))
1125                 return error;
1126         if (!memcmp(SWSUSP_SIG, swsusp_header.sig, 10)) {
1127                 memcpy(swsusp_header.sig, swsusp_header.orig_sig, 10);
1128
1129                 /*
1130                  * Reset swap signature now.
1131                  */
1132                 error = bio_write_page(0, &swsusp_header);
1133         } else {
1134                 pr_debug(KERN_ERR "swsusp: Suspend partition has wrong signature?\n");
1135                 return -EINVAL;
1136         }
1137         if (!error)
1138                 pr_debug("swsusp: Signature found, resuming\n");
1139         return error;
1140 }
1141
1142 /**
1143  *      swsusp_read_data - Read image pages from swap.
1144  *
1145  *      You do not need to check for overlaps, check_pagedir()
1146  *      already did that.
1147  */
1148
1149 static int __init data_read(void)
1150 {
1151         struct pbe * p;
1152         int error;
1153         int i;
1154         int mod = nr_copy_pages / 100;
1155
1156         if (!mod)
1157                 mod = 1;
1158
1159         if ((error = swsusp_pagedir_relocate()))
1160                 return error;
1161
1162         printk( "Reading image data (%d pages):     ", nr_copy_pages );
1163         for(i = 0, p = pagedir_nosave; i < nr_copy_pages && !error; i++, p++) {
1164                 if (!(i%mod))
1165                         printk( "\b\b\b\b%3d%%", i / mod );
1166                 error = bio_read_page(swp_offset(p->swap_address),
1167                                   (void *)p->address);
1168         }
1169         printk(" %d done.\n",i);
1170         return error;
1171
1172 }
1173
1174 extern dev_t __init name_to_dev_t(const char *line);
1175
1176 static int __init read_pagedir(void)
1177 {
1178         unsigned long addr;
1179         int i, n = swsusp_info.pagedir_pages;
1180         int error = 0;
1181
1182         pagedir_order = get_bitmask_order(n);
1183
1184         addr =__get_free_pages(GFP_ATOMIC, pagedir_order);
1185         if (!addr)
1186                 return -ENOMEM;
1187         pagedir_nosave = (struct pbe *)addr;
1188
1189         pr_debug("pmdisk: Reading pagedir (%d Pages)\n",n);
1190
1191         for (i = 0; i < n && !error; i++, addr += PAGE_SIZE) {
1192                 unsigned long offset = swp_offset(swsusp_info.pagedir[i]);
1193                 if (offset)
1194                         error = bio_read_page(offset, (void *)addr);
1195                 else
1196                         error = -EFAULT;
1197         }
1198         if (error)
1199                 free_pages((unsigned long)pagedir_nosave, pagedir_order);
1200         return error;
1201 }
1202
1203 static int __init read_suspend_image(void)
1204 {
1205         int error = 0;
1206
1207         if ((error = check_sig()))
1208                 return error;
1209         if ((error = check_header()))
1210                 return error;
1211         if ((error = read_pagedir()))
1212                 return error;
1213         if ((error = data_read()))
1214                 free_pages((unsigned long)pagedir_nosave, pagedir_order);
1215         return error;
1216 }
1217
1218 /**
1219  *      pmdisk_read - Read saved image from swap.
1220  */
1221
1222 int __init swsusp_read(void)
1223 {
1224         int error;
1225
1226         if (!strlen(resume_file))
1227                 return -ENOENT;
1228
1229         resume_device = name_to_dev_t(resume_file);
1230         pr_debug("swsusp: Resume From Partition: %s\n", resume_file);
1231
1232         resume_bdev = open_by_devnum(resume_device, FMODE_READ);
1233         if (!IS_ERR(resume_bdev)) {
1234                 set_blocksize(resume_bdev, PAGE_SIZE);
1235                 error = read_suspend_image();
1236                 blkdev_put(resume_bdev);
1237         } else
1238                 error = PTR_ERR(resume_bdev);
1239
1240         if (!error)
1241                 pr_debug("Reading resume file was successful\n");
1242         else
1243                 pr_debug("pmdisk: Error %d resuming\n", error);
1244         return error;
1245 }