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
  53 #include "power.h"
  54
  55 /*
  56  * Preferred image size in bytes (tunable via /sys/power/image_size).
  57  * When it is set to N, swsusp will do its best to ensure the image
  58  * size will not exceed N bytes, but if that is impossible, it will
  59  * try to create the smallest image possible.
  60  */
  61 unsigned long image_size = 500 * 1024 * 1024;
  62
  63 int in_suspend __nosavedata = 0;
  64
  65 /**
  66  *      The following functions are used for tracing the allocated
  67  *      swap pages, so that they can be freed in case of an error.
  68  *
  69  *      The functions operate on a linked bitmap structure defined
  70  *      in power.h
  71  */
  72
  73 void free_bitmap(struct bitmap_page *bitmap)
  74 {
  75         struct bitmap_page *bp;
  76
  77         while (bitmap) {
  78                 bp = bitmap->next;
  79                 free_page((unsigned long)bitmap);
  80                 bitmap = bp;
  81         }
  82 }
  83
  84 struct bitmap_page *alloc_bitmap(unsigned int nr_bits)
  85 {
  86         struct bitmap_page *bitmap, *bp;
  87         unsigned int n;
  88
  89         if (!nr_bits)
  90                 return NULL;
  91
  92         bitmap = (struct bitmap_page *)get_zeroed_page(GFP_KERNEL);
  93         bp = bitmap;
  94         for (n = BITMAP_PAGE_BITS; n < nr_bits; n += BITMAP_PAGE_BITS) {
  95                 bp->next = (struct bitmap_page *)get_zeroed_page(GFP_KERNEL);
  96                 bp = bp->next;
  97                 if (!bp) {
  98                         free_bitmap(bitmap);
  99                         return NULL;
 100                 }
 101         }
 102         return bitmap;
 103 }
 104
 105 static int bitmap_set(struct bitmap_page *bitmap, unsigned long bit)
 106 {
 107         unsigned int n;
 108
 109         n = BITMAP_PAGE_BITS;
 110         while (bitmap && n <= bit) {
 111                 n += BITMAP_PAGE_BITS;
 112                 bitmap = bitmap->next;
 113         }
 114         if (!bitmap)
 115                 return -EINVAL;
 116         n -= BITMAP_PAGE_BITS;
 117         bit -= n;
 118         n = 0;
 119         while (bit >= BITS_PER_CHUNK) {
 120                 bit -= BITS_PER_CHUNK;
 121                 n++;
 122         }
 123         bitmap->chunks[n] |= (1UL << bit);
 124         return 0;
 125 }
 126
 127 unsigned long alloc_swap_page(int swap, struct bitmap_page *bitmap)
 128 {
 129         unsigned long offset;
 130
 131         offset = swp_offset(get_swap_page_of_type(swap));
 132         if (offset) {
 133                 if (bitmap_set(bitmap, offset)) {
 134                         swap_free(swp_entry(swap, offset));
 135                         offset = 0;
 136                 }
 137         }
 138         return offset;
 139 }
 140
 141 void free_all_swap_pages(int swap, struct bitmap_page *bitmap)
 142 {
 143         unsigned int bit, n;
 144         unsigned long test;
 145
 146         bit = 0;
 147         while (bitmap) {
 148                 for (n = 0; n < BITMAP_PAGE_CHUNKS; n++)
 149                         for (test = 1UL; test; test <<= 1) {
 150                                 if (bitmap->chunks[n] & test)
 151                                         swap_free(swp_entry(swap, bit));
 152                                 bit++;
 153                         }
 154                 bitmap = bitmap->next;
 155         }
 156 }
 157
 158 /**
 159  *      swsusp_shrink_memory -  Try to free as much memory as needed
 160  *
 161  *      ... but do not OOM-kill anyone
 162  *
 163  *      Notice: all userland should be stopped before it is called, or
 164  *      livelock is possible.
 165  */
 166
 167 #define SHRINK_BITE     10000
 168 static inline unsigned long __shrink_memory(long tmp)
 169 {
 170         if (tmp > SHRINK_BITE)
 171                 tmp = SHRINK_BITE;
 172         return shrink_all_memory(tmp);
 173 }
 174
 175 int swsusp_shrink_memory(void)
 176 {
 177         long size, tmp;
 178         struct zone *zone;
 179         unsigned long pages = 0;
 180         unsigned int i = 0;
 181         char *p = "-\\|/";
 182
 183         printk("Shrinking memory...  ");
 184         do {
 185                 size = 2 * count_special_pages();
 186                 size += size / 50 + count_data_pages();
 187                 size += (size + PBES_PER_PAGE - 1) / PBES_PER_PAGE +
 188                         PAGES_FOR_IO;
 189                 tmp = size;
 190                 for_each_zone (zone)
 191                         if (!is_highmem(zone))
 192                                 tmp -= zone->free_pages;
 193                 if (tmp > 0) {
 194                         tmp = __shrink_memory(tmp);
 195                         if (!tmp)
 196                                 return -ENOMEM;
 197                         pages += tmp;
 198                 } else if (size > image_size / PAGE_SIZE) {
 199                         tmp = __shrink_memory(size - (image_size / PAGE_SIZE));
 200                         pages += tmp;
 201                 }
 202                 printk("\b%c", p[i++%4]);
 203         } while (tmp > 0);
 204         printk("\bdone (%lu pages freed)\n", pages);
 205
 206         return 0;
 207 }
 208
 209 int swsusp_suspend(void)
 210 {
 211         int error;
 212
 213         if ((error = arch_prepare_suspend()))
 214                 return error;
 215         local_irq_disable();
 216         /* At this point, device_suspend() has been called, but *not*
 217          * device_power_down(). We *must* device_power_down() now.
 218          * Otherwise, drivers for some devices (e.g. interrupt controllers)
 219          * become desynchronized with the actual state of the hardware
 220          * at resume time, and evil weirdness ensues.
 221          */
 222         if ((error = device_power_down(PMSG_FREEZE))) {
 223                 printk(KERN_ERR "Some devices failed to power down, aborting suspend\n");
 224                 goto Enable_irqs;
 225         }
 226
 227         if ((error = save_special_mem())) {
 228                 printk(KERN_ERR "swsusp: Not enough free pages for highmem\n");
 229                 goto Restore_highmem;
 230         }
 231
 232         save_processor_state();
 233         if ((error = swsusp_arch_suspend()))
 234                 printk(KERN_ERR "Error %d suspending\n", error);
 235         /* Restore control flow magically appears here */
 236         restore_processor_state();
 237 Restore_highmem:
 238         restore_special_mem();
 239         device_power_up();
 240 Enable_irqs:
 241         local_irq_enable();
 242         return error;
 243 }
 244
 245 int swsusp_resume(void)
 246 {
 247         int error;
 248         local_irq_disable();
 249         if (device_power_down(PMSG_FREEZE))
 250                 printk(KERN_ERR "Some devices failed to power down, very bad\n");
 251         /* We'll ignore saved state, but this gets preempt count (etc) right */
 252         save_processor_state();
 253         error = swsusp_arch_resume();
 254         /* Code below is only ever reached in case of failure. Otherwise
 255          * execution continues at place where swsusp_arch_suspend was called
 256          */
 257         BUG_ON(!error);
 258         /* The only reason why swsusp_arch_resume() can fail is memory being
 259          * very tight, so we have to free it as soon as we can to avoid
 260          * subsequent failures
 261          */
 262         swsusp_free();
 263         restore_processor_state();
 264         restore_special_mem();
 265         touch_softlockup_watchdog();
 266         device_power_up();
 267         local_irq_enable();
 268         return error;
 269 }