kernel/power/swsusp.c

   1 /*
   2  * linux/kernel/power/swsusp.c
   3  *
   4  * This file provides code to write suspend image to swap and read it back.
   5  *
   6  * Copyright (C) 1998-2001 Gabor Kuti <seasons@fornax.hu>
   7  * Copyright (C) 1998,2001-2005 Pavel Machek <pavel@suse.cz>
   8  *
   9  * This file is released under the GPLv2.
  10  *
  11  * I'd like to thank the following people for their work:
  12  *
  13  * Pavel Machek <pavel@ucw.cz>:
  14  * Modifications, defectiveness pointing, being with me at the very beginning,
  15  * suspend to swap space, stop all tasks. Port to 2.4.18-ac and 2.5.17.
  16  *
  17  * Steve Doddi <dirk@loth.demon.co.uk>:
  18  * Support the possibility of hardware state restoring.
  19  *
  20  * Raph <grey.havens@earthling.net>:
  21  * Support for preserving states of network devices and virtual console
  22  * (including X and svgatextmode)
  23  *
  24  * Kurt Garloff <garloff@suse.de>:
  25  * Straightened the critical function in order to prevent compilers from
  26  * playing tricks with local variables.
  27  *
  28  * Andreas Mohr <a.mohr@mailto.de>
  29  *
  30  * Alex Badea <vampire@go.ro>:
  31  * Fixed runaway init
  32  *
  33  * Rafael J. Wysocki <rjw@sisk.pl>
  34  * Reworked the freeing of memory and the handling of swap
  35  *
  36  * More state savers are welcome. Especially for the scsi layer...
  37  *
  38  * For TODOs,FIXMEs also look in Documentation/power/swsusp.txt
  39  */
  40
  41 #include <linux/mm.h>
  42 #include <linux/suspend.h>
  43 #include <linux/spinlock.h>
  44 #include <linux/kernel.h>
  45 #include <linux/major.h>
  46 #include <linux/swap.h>
  47 #include <linux/pm.h>
  48 #include <linux/swapops.h>
  49 #include <linux/bootmem.h>
  50 #include <linux/syscalls.h>
  51 #include <linux/highmem.h>
  52 #include <linux/time.h>
  53
  54 #include "power.h"
  55
  56 /*
  57  * Preferred image size in bytes (tunable via /sys/power/image_size).
  58  * When it is set to N, swsusp will do its best to ensure the image
  59  * size will not exceed N bytes, but if that is impossible, it will
  60  * try to create the smallest image possible.
  61  */
  62 unsigned long image_size = 500 * 1024 * 1024;
  63
  64 int in_suspend __nosavedata = 0;
  65
  66 #ifdef CONFIG_HIGHMEM
  67 unsigned int count_highmem_pages(void);
  68 int restore_highmem(void);
  69 #else
  70 static inline int restore_highmem(void) { return 0; }
  71 static inline unsigned int count_highmem_pages(void) { return 0; }
  72 #endif
  73
  74 /**
  75  *      The following functions are used for tracing the allocated
  76  *      swap pages, so that they can be freed in case of an error.
  77  *
  78  *      The functions operate on a linked bitmap structure defined
  79  *      in power.h
  80  */
  81
  82 void free_bitmap(struct bitmap_page *bitmap)
  83 {
  84         struct bitmap_page *bp;
  85
  86         while (bitmap) {
  87                 bp = bitmap->next;
  88                 free_page((unsigned long)bitmap);
  89                 bitmap = bp;
  90         }
  91 }
  92
  93 struct bitmap_page *alloc_bitmap(unsigned int nr_bits)
  94 {
  95         struct bitmap_page *bitmap, *bp;
  96         unsigned int n;
  97
  98         if (!nr_bits)
  99                 return NULL;
 100
 101         bitmap = (struct bitmap_page *)get_zeroed_page(GFP_KERNEL);
 102         bp = bitmap;
 103         for (n = BITMAP_PAGE_BITS; n < nr_bits; n += BITMAP_PAGE_BITS) {
 104                 bp->next = (struct bitmap_page *)get_zeroed_page(GFP_KERNEL);
 105                 bp = bp->next;
 106                 if (!bp) {
 107                         free_bitmap(bitmap);
 108                         return NULL;
 109                 }
 110         }
 111         return bitmap;
 112 }
 113
 114 static int bitmap_set(struct bitmap_page *bitmap, unsigned long bit)
 115 {
 116         unsigned int n;
 117
 118         n = BITMAP_PAGE_BITS;
 119         while (bitmap && n <= bit) {
 120                 n += BITMAP_PAGE_BITS;
 121                 bitmap = bitmap->next;
 122         }
 123         if (!bitmap)
 124                 return -EINVAL;
 125         n -= BITMAP_PAGE_BITS;
 126         bit -= n;
 127         n = 0;
 128         while (bit >= BITS_PER_CHUNK) {
 129                 bit -= BITS_PER_CHUNK;
 130                 n++;
 131         }
 132         bitmap->chunks[n] |= (1UL << bit);
 133         return 0;
 134 }
 135
 136 sector_t alloc_swapdev_block(int swap, struct bitmap_page *bitmap)
 137 {
 138         unsigned long offset;
 139
 140         offset = swp_offset(get_swap_page_of_type(swap));
 141         if (offset) {
 142                 if (bitmap_set(bitmap, offset))
 143                         swap_free(swp_entry(swap, offset));
 144                 else
 145                         return swapdev_block(swap, offset);
 146         }
 147         return 0;
 148 }
 149
 150 void free_all_swap_pages(int swap, struct bitmap_page *bitmap)
 151 {
 152         unsigned int bit, n;
 153         unsigned long test;
 154
 155         bit = 0;
 156         while (bitmap) {
 157                 for (n = 0; n < BITMAP_PAGE_CHUNKS; n++)
 158                         for (test = 1UL; test; test <<= 1) {
 159                                 if (bitmap->chunks[n] & test)
 160                                         swap_free(swp_entry(swap, bit));
 161                                 bit++;
 162                         }
 163                 bitmap = bitmap->next;
 164         }
 165 }
 166
 167 /**
 168  *      swsusp_show_speed - print the time elapsed between two events represented by
 169  *      @start and @stop
 170  *
 171  *      @nr_pages -     number of pages processed between @start and @stop
 172  *      @msg -          introductory message to print
 173  */
 174
 175 void swsusp_show_speed(struct timeval *start, struct timeval *stop,
 176                         unsigned nr_pages, char *msg)
 177 {
 178         s64 elapsed_centisecs64;
 179         int centisecs;
 180         int k;
 181         int kps;
 182
 183         elapsed_centisecs64 = timeval_to_ns(stop) - timeval_to_ns(start);
 184         do_div(elapsed_centisecs64, NSEC_PER_SEC / 100);
 185         centisecs = elapsed_centisecs64;
 186         if (centisecs == 0)
 187                 centisecs = 1;  /* avoid div-by-zero */
 188         k = nr_pages * (PAGE_SIZE / 1024);
 189         kps = (k * 100) / centisecs;
 190         printk("%s %d kbytes in %d.%02d seconds (%d.%02d MB/s)\n", msg, k,
 191                         centisecs / 100, centisecs % 100,
 192                         kps / 1000, (kps % 1000) / 10);
 193 }
 194
 195 /**
 196  *      swsusp_shrink_memory -  Try to free as much memory as needed
 197  *
 198  *      ... but do not OOM-kill anyone
 199  *
 200  *      Notice: all userland should be stopped before it is called, or
 201  *      livelock is possible.
 202  */
 203
 204 #define SHRINK_BITE     10000
 205 static inline unsigned long __shrink_memory(long tmp)
 206 {
 207         if (tmp > SHRINK_BITE)
 208                 tmp = SHRINK_BITE;
 209         return shrink_all_memory(tmp);
 210 }
 211
 212 int swsusp_shrink_memory(void)
 213 {
 214         long tmp;
 215         struct zone *zone;
 216         unsigned long pages = 0;
 217         unsigned int i = 0;
 218         char *p = "-\\|/";
 219         struct timeval start, stop;
 220
 221         printk("Shrinking memory...  ");
 222         do_gettimeofday(&start);
 223         do {
 224                 long size, highmem_size;
 225
 226                 highmem_size = count_highmem_pages();
 227                 size = count_data_pages() + PAGES_FOR_IO;
 228                 tmp = size;
 229                 size += highmem_size;
 230                 for_each_zone (zone)
 231                         if (populated_zone(zone)) {
 232                                 tmp += snapshot_additional_pages(zone);
 233                                 if (is_highmem(zone)) {
 234                                         highmem_size -=
 235                                         zone_page_state(zone, NR_FREE_PAGES);
 236                                 } else {
 237                                         tmp -= zone_page_state(zone, NR_FREE_PAGES);
 238                                         tmp += zone->lowmem_reserve[ZONE_NORMAL];
 239                                 }
 240                         }
 241
 242                 if (highmem_size < 0)
 243                         highmem_size = 0;
 244
 245                 tmp += highmem_size;
 246                 if (tmp > 0) {
 247                         tmp = __shrink_memory(tmp);
 248                         if (!tmp)
 249                                 return -ENOMEM;
 250                         pages += tmp;
 251                 } else if (size > image_size / PAGE_SIZE) {
 252                         tmp = __shrink_memory(size - (image_size / PAGE_SIZE));
 253                         pages += tmp;
 254                 }
 255                 printk("\b%c", p[i++%4]);
 256         } while (tmp > 0);
 257         do_gettimeofday(&stop);
 258         printk("\bdone (%lu pages freed)\n", pages);
 259         swsusp_show_speed(&start, &stop, pages, "Freed");
 260
 261         return 0;
 262 }
 263
 264 int swsusp_suspend(void)
 265 {
 266         int error;
 267
 268         if ((error = arch_prepare_suspend()))
 269                 return error;
 270
 271         local_irq_disable();
 272         /* At this point, device_suspend() has been called, but *not*
 273          * device_power_down(). We *must* device_power_down() now.
 274          * Otherwise, drivers for some devices (e.g. interrupt controllers)
 275          * become desynchronized with the actual state of the hardware
 276          * at resume time, and evil weirdness ensues.
 277          */
 278         if ((error = device_power_down(PMSG_FREEZE))) {
 279                 printk(KERN_ERR "Some devices failed to power down, aborting suspend\n");
 280                 goto Enable_irqs;
 281         }
 282
 283         save_processor_state();
 284         if ((error = swsusp_arch_suspend()))
 285                 printk(KERN_ERR "Error %d suspending\n", error);
 286         /* Restore control flow magically appears here */
 287         restore_processor_state();
 288         /* NOTE:  device_power_up() is just a resume() for devices
 289          * that suspended with irqs off ... no overall powerup.
 290          */
 291         device_power_up();
 292  Enable_irqs:
 293         local_irq_enable();
 294         return error;
 295 }
 296
 297 int swsusp_resume(void)
 298 {
 299         int error;
 300
 301         local_irq_disable();
 302         /* NOTE:  device_power_down() is just a suspend() with irqs off;
 303          * it has no special "power things down" semantics
 304          */
 305         if (device_power_down(PMSG_PRETHAW))
 306                 printk(KERN_ERR "Some devices failed to power down, very bad\n");
 307         /* We'll ignore saved state, but this gets preempt count (etc) right */
 308         save_processor_state();
 309         error = restore_highmem();
 310         if (!error) {
 311                 error = swsusp_arch_resume();
 312                 /* The code below is only ever reached in case of a failure.
 313                  * Otherwise execution continues at place where
 314                  * swsusp_arch_suspend() was called
 315                  */
 316                 BUG_ON(!error);
 317                 /* This call to restore_highmem() undos the previous one */
 318                 restore_highmem();
 319         }
 320         /* The only reason why swsusp_arch_resume() can fail is memory being
 321          * very tight, so we have to free it as soon as we can to avoid
 322          * subsequent failures
 323          */
 324         swsusp_free();
 325         restore_processor_state();
 326         touch_softlockup_watchdog();
 327         device_power_up();
 328         local_irq_enable();
 329         return error;
 330 }