SAFE public projects git trees. - safe/jmp/linux-2.6/blob - kernel/power/snapshot.c

   1 /*
   2  * linux/kernel/power/snapshot.c
   3  *
   4  * This file provide system snapshot/restore functionality.
   5  *
   6  * Copyright (C) 1998-2005 Pavel Machek <pavel@suse.cz>
   7  *
   8  * This file is released under the GPLv2, and is based on swsusp.c.
   9  *
  10  */
  11
  12
  13 #include <linux/module.h>
  14 #include <linux/mm.h>
  15 #include <linux/suspend.h>
  16 #include <linux/smp_lock.h>
  17 #include <linux/delay.h>
  18 #include <linux/bitops.h>
  19 #include <linux/spinlock.h>
  20 #include <linux/kernel.h>
  21 #include <linux/pm.h>
  22 #include <linux/device.h>
  23 #include <linux/bootmem.h>
  24 #include <linux/syscalls.h>
  25 #include <linux/console.h>
  26 #include <linux/highmem.h>
  27
  28 #include <asm/uaccess.h>
  29 #include <asm/mmu_context.h>
  30 #include <asm/pgtable.h>
  31 #include <asm/tlbflush.h>
  32 #include <asm/io.h>
  33
  34 #include "power.h"
  35
  36 #ifdef CONFIG_HIGHMEM
  37 struct highmem_page {
  38         char *data;
  39         struct page *page;
  40         struct highmem_page *next;
  41 };
  42
  43 static struct highmem_page *highmem_copy;
  44
  45 static int save_highmem_zone(struct zone *zone)
  46 {
  47         unsigned long zone_pfn;
  48         mark_free_pages(zone);
  49         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
  50                 struct page *page;
  51                 struct highmem_page *save;
  52                 void *kaddr;
  53                 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
  54
  55                 if (!(pfn%1000))
  56                         printk(".");
  57                 if (!pfn_valid(pfn))
  58                         continue;
  59                 page = pfn_to_page(pfn);
  60                 /*
  61                  * This condition results from rvmalloc() sans vmalloc_32()
  62                  * and architectural memory reservations. This should be
  63                  * corrected eventually when the cases giving rise to this
  64                  * are better understood.
  65                  */
  66                 if (PageReserved(page)) {
  67                         printk("highmem reserved page?!\n");
  68                         continue;
  69                 }
  70                 BUG_ON(PageNosave(page));
  71                 if (PageNosaveFree(page))
  72                         continue;
  73                 save = kmalloc(sizeof(struct highmem_page), GFP_ATOMIC);
  74                 if (!save)
  75                         return -ENOMEM;
  76                 save->next = highmem_copy;
  77                 save->page = page;
  78                 save->data = (void *) get_zeroed_page(GFP_ATOMIC);
  79                 if (!save->data) {
  80                         kfree(save);
  81                         return -ENOMEM;
  82                 }
  83                 kaddr = kmap_atomic(page, KM_USER0);
  84                 memcpy(save->data, kaddr, PAGE_SIZE);
  85                 kunmap_atomic(kaddr, KM_USER0);
  86                 highmem_copy = save;
  87         }
  88         return 0;
  89 }
  90
  91
  92 static int save_highmem(void)
  93 {
  94         struct zone *zone;
  95         int res = 0;
  96
  97         pr_debug("swsusp: Saving Highmem\n");
  98         for_each_zone (zone) {
  99                 if (is_highmem(zone))
 100                         res = save_highmem_zone(zone);
 101                 if (res)
 102                         return res;
 103         }
 104         return 0;
 105 }
 106
 107 int restore_highmem(void)
 108 {
 109         printk("swsusp: Restoring Highmem\n");
 110         while (highmem_copy) {
 111                 struct highmem_page *save = highmem_copy;
 112                 void *kaddr;
 113                 highmem_copy = save->next;
 114
 115                 kaddr = kmap_atomic(save->page, KM_USER0);
 116                 memcpy(kaddr, save->data, PAGE_SIZE);
 117                 kunmap_atomic(kaddr, KM_USER0);
 118                 free_page((long) save->data);
 119                 kfree(save);
 120         }
 121         return 0;
 122 }
 123 #else
 124 static int save_highmem(void) { return 0; }
 125 int restore_highmem(void) { return 0; }
 126 #endif /* CONFIG_HIGHMEM */
 127
 128
 129 static int pfn_is_nosave(unsigned long pfn)
 130 {
 131         unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
 132         unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
 133         return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
 134 }
 135
 136 /**
 137  *      saveable - Determine whether a page should be cloned or not.
 138  *      @pfn:   The page
 139  *
 140  *      We save a page if it's Reserved, and not in the range of pages
 141  *      statically defined as 'unsaveable', or if it isn't reserved, and
 142  *      isn't part of a free chunk of pages.
 143  */
 144
 145 static int saveable(struct zone *zone, unsigned long *zone_pfn)
 146 {
 147         unsigned long pfn = *zone_pfn + zone->zone_start_pfn;
 148         struct page *page;
 149
 150         if (!pfn_valid(pfn))
 151                 return 0;
 152
 153         page = pfn_to_page(pfn);
 154         BUG_ON(PageReserved(page) && PageNosave(page));
 155         if (PageNosave(page))
 156                 return 0;
 157         if (PageReserved(page) && pfn_is_nosave(pfn)) {
 158                 pr_debug("[nosave pfn 0x%lx]", pfn);
 159                 return 0;
 160         }
 161         if (PageNosaveFree(page))
 162                 return 0;
 163
 164         return 1;
 165 }
 166
 167 static unsigned count_data_pages(void)
 168 {
 169         struct zone *zone;
 170         unsigned long zone_pfn;
 171         unsigned int n = 0;
 172
 173         for_each_zone (zone) {
 174                 if (is_highmem(zone))
 175                         continue;
 176                 mark_free_pages(zone);
 177                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 178                         n += saveable(zone, &zone_pfn);
 179         }
 180         return n;
 181 }
 182
 183 static void copy_data_pages(struct pbe *pblist)
 184 {
 185         struct zone *zone;
 186         unsigned long zone_pfn;
 187         struct pbe *pbe, *p;
 188
 189         pbe = pblist;
 190         for_each_zone (zone) {
 191                 if (is_highmem(zone))
 192                         continue;
 193                 mark_free_pages(zone);
 194                 /* This is necessary for swsusp_free() */
 195                 for_each_pb_page (p, pblist)
 196                         SetPageNosaveFree(virt_to_page(p));
 197                 for_each_pbe (p, pblist)
 198                         SetPageNosaveFree(virt_to_page(p->address));
 199                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
 200                         if (saveable(zone, &zone_pfn)) {
 201                                 struct page *page;
 202                                 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
 203                                 BUG_ON(!pbe);
 204                                 pbe->orig_address = (unsigned long)page_address(page);
 205                                 /* copy_page is not usable for copying task structs. */
 206                                 memcpy((void *)pbe->address, (void *)pbe->orig_address, PAGE_SIZE);
 207                                 pbe = pbe->next;
 208                         }
 209                 }
 210         }
 211         BUG_ON(pbe);
 212 }
 213
 214
 215 /**
 216  *      free_pagedir - free pages allocated with alloc_pagedir()
 217  */
 218
 219 static void free_pagedir(struct pbe *pblist)
 220 {
 221         struct pbe *pbe;
 222
 223         while (pblist) {
 224                 pbe = (pblist + PB_PAGE_SKIP)->next;
 225                 ClearPageNosave(virt_to_page(pblist));
 226                 ClearPageNosaveFree(virt_to_page(pblist));
 227                 free_page((unsigned long)pblist);
 228                 pblist = pbe;
 229         }
 230 }
 231
 232 /**
 233  *      fill_pb_page - Create a list of PBEs on a given memory page
 234  */
 235
 236 static inline void fill_pb_page(struct pbe *pbpage)
 237 {
 238         struct pbe *p;
 239
 240         p = pbpage;
 241         pbpage += PB_PAGE_SKIP;
 242         do
 243                 p->next = p + 1;
 244         while (++p < pbpage);
 245 }
 246
 247 /**
 248  *      create_pbe_list - Create a list of PBEs on top of a given chain
 249  *      of memory pages allocated with alloc_pagedir()
 250  */
 251
 252 void create_pbe_list(struct pbe *pblist, unsigned int nr_pages)
 253 {
 254         struct pbe *pbpage, *p;
 255         unsigned int num = PBES_PER_PAGE;
 256
 257         for_each_pb_page (pbpage, pblist) {
 258                 if (num >= nr_pages)
 259                         break;
 260
 261                 fill_pb_page(pbpage);
 262                 num += PBES_PER_PAGE;
 263         }
 264         if (pbpage) {
 265                 for (num -= PBES_PER_PAGE - 1, p = pbpage; num < nr_pages; p++, num++)
 266                         p->next = p + 1;
 267                 p->next = NULL;
 268         }
 269         pr_debug("create_pbe_list(): initialized %d PBEs\n", num);
 270 }
 271
 272 /**
 273  *      @safe_needed - on resume, for storing the PBE list and the image,
 274  *      we can only use memory pages that do not conflict with the pages
 275  *      which had been used before suspend.
 276  *
 277  *      The unsafe pages are marked with the PG_nosave_free flag
 278  *
 279  *      Allocated but unusable (ie eaten) memory pages should be marked
 280  *      so that swsusp_free() can release them
 281  */
 282
 283 static inline void *alloc_image_page(gfp_t gfp_mask, int safe_needed)
 284 {
 285         void *res;
 286
 287         if (safe_needed)
 288                 do {
 289                         res = (void *)get_zeroed_page(gfp_mask);
 290                         if (res && PageNosaveFree(virt_to_page(res)))
 291                                 /* This is for swsusp_free() */
 292                                 SetPageNosave(virt_to_page(res));
 293                 } while (res && PageNosaveFree(virt_to_page(res)));
 294         else
 295                 res = (void *)get_zeroed_page(gfp_mask);
 296         if (res) {
 297                 SetPageNosave(virt_to_page(res));
 298                 SetPageNosaveFree(virt_to_page(res));
 299         }
 300         return res;
 301 }
 302
 303 unsigned long get_safe_page(gfp_t gfp_mask)
 304 {
 305         return (unsigned long)alloc_image_page(gfp_mask, 1);
 306 }
 307
 308 /**
 309  *      alloc_pagedir - Allocate the page directory.
 310  *
 311  *      First, determine exactly how many pages we need and
 312  *      allocate them.
 313  *
 314  *      We arrange the pages in a chain: each page is an array of PBES_PER_PAGE
 315  *      struct pbe elements (pbes) and the last element in the page points
 316  *      to the next page.
 317  *
 318  *      On each page we set up a list of struct_pbe elements.
 319  */
 320
 321 struct pbe *alloc_pagedir(unsigned int nr_pages, gfp_t gfp_mask, int safe_needed)
 322 {
 323         unsigned int num;
 324         struct pbe *pblist, *pbe;
 325
 326         if (!nr_pages)
 327                 return NULL;
 328
 329         pr_debug("alloc_pagedir(): nr_pages = %d\n", nr_pages);
 330         pblist = alloc_image_page(gfp_mask, safe_needed);
 331         /* FIXME: rewrite this ugly loop */
 332         for (pbe = pblist, num = PBES_PER_PAGE; pbe && num < nr_pages;
 333                         pbe = pbe->next, num += PBES_PER_PAGE) {
 334                 pbe += PB_PAGE_SKIP;
 335                 pbe->next = alloc_image_page(gfp_mask, safe_needed);
 336         }
 337         if (!pbe) { /* get_zeroed_page() failed */
 338                 free_pagedir(pblist);
 339                 pblist = NULL;
 340         }
 341         return pblist;
 342 }
 343
 344 /**
 345  * Free pages we allocated for suspend. Suspend pages are alocated
 346  * before atomic copy, so we need to free them after resume.
 347  */
 348
 349 void swsusp_free(void)
 350 {
 351         struct zone *zone;
 352         unsigned long zone_pfn;
 353
 354         for_each_zone(zone) {
 355                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 356                         if (pfn_valid(zone_pfn + zone->zone_start_pfn)) {
 357                                 struct page *page;
 358                                 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
 359                                 if (PageNosave(page) && PageNosaveFree(page)) {
 360                                         ClearPageNosave(page);
 361                                         ClearPageNosaveFree(page);
 362                                         free_page((long) page_address(page));
 363                                 }
 364                         }
 365         }
 366 }
 367
 368
 369 /**
 370  *      enough_free_mem - Make sure we enough free memory to snapshot.
 371  *
 372  *      Returns TRUE or FALSE after checking the number of available
 373  *      free pages.
 374  */
 375
 376 static int enough_free_mem(unsigned int nr_pages)
 377 {
 378         pr_debug("swsusp: available memory: %u pages\n", nr_free_pages());
 379         return nr_free_pages() > (nr_pages + PAGES_FOR_IO +
 380                 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
 381 }
 382
 383 int alloc_data_pages(struct pbe *pblist, gfp_t gfp_mask, int safe_needed)
 384 {
 385         struct pbe *p;
 386
 387         for_each_pbe (p, pblist) {
 388                 p->address = (unsigned long)alloc_image_page(gfp_mask, safe_needed);
 389                 if (!p->address)
 390                         return -ENOMEM;
 391         }
 392         return 0;
 393 }
 394
 395 static struct pbe *swsusp_alloc(unsigned int nr_pages)
 396 {
 397         struct pbe *pblist;
 398
 399         if (!(pblist = alloc_pagedir(nr_pages, GFP_ATOMIC | __GFP_COLD, 0))) {
 400                 printk(KERN_ERR "suspend: Allocating pagedir failed.\n");
 401                 return NULL;
 402         }
 403         create_pbe_list(pblist, nr_pages);
 404
 405         if (alloc_data_pages(pblist, GFP_ATOMIC | __GFP_COLD, 0)) {
 406                 printk(KERN_ERR "suspend: Allocating image pages failed.\n");
 407                 swsusp_free();
 408                 return NULL;
 409         }
 410
 411         return pblist;
 412 }
 413
 414 asmlinkage int swsusp_save(void)
 415 {
 416         unsigned int nr_pages;
 417
 418         pr_debug("swsusp: critical section: \n");
 419         if (save_highmem()) {
 420                 printk(KERN_CRIT "swsusp: Not enough free pages for highmem\n");
 421                 restore_highmem();
 422                 return -ENOMEM;
 423         }
 424
 425         drain_local_pages();
 426         nr_pages = count_data_pages();
 427         printk("swsusp: Need to copy %u pages\n", nr_pages);
 428
 429         pr_debug("swsusp: pages needed: %u + %lu + %u, free: %u\n",
 430                  nr_pages,
 431                  (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE,
 432                  PAGES_FOR_IO, nr_free_pages());
 433
 434         /* This is needed because of the fixed size of swsusp_info */
 435         if (MAX_PBES < (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE)
 436                 return -ENOSPC;
 437
 438         if (!enough_free_mem(nr_pages)) {
 439                 printk(KERN_ERR "swsusp: Not enough free memory\n");
 440                 return -ENOMEM;
 441         }
 442
 443         if (!enough_swap(nr_pages)) {
 444                 printk(KERN_ERR "swsusp: Not enough free swap\n");
 445                 return -ENOSPC;
 446         }
 447
 448         pagedir_nosave = swsusp_alloc(nr_pages);
 449         if (!pagedir_nosave)
 450                 return -ENOMEM;
 451
 452         /* During allocating of suspend pagedir, new cold pages may appear.
 453          * Kill them.
 454          */
 455         drain_local_pages();
 456         copy_data_pages(pagedir_nosave);
 457
 458         /*
 459          * End of critical section. From now on, we can write to memory,
 460          * but we should not touch disk. This specially means we must _not_
 461          * touch swap space! Except we must write out our image of course.
 462          */
 463
 464         nr_copy_pages = nr_pages;
 465
 466         printk("swsusp: critical section/: done (%d pages copied)\n", nr_pages);
 467         return 0;
 468 }