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 #ifdef CONFIG_HIGHMEM
  66 unsigned int count_highmem_pages(void);
  67 int save_highmem(void);
  68 int restore_highmem(void);
  69 #else
  70 static inline int save_highmem(void) { return 0; }
  71 static inline int restore_highmem(void) { return 0; }
  72 static inline unsigned int count_highmem_pages(void) { return 0; }
  73 #endif
  74
  75 /**
  76  *      The following functions are used for tracing the allocated
  77  *      swap pages, so that they can be freed in case of an error.
  78  *
  79  *      The functions operate on a linked bitmap structure defined
  80  *      in power.h
  81  */
  82
  83 void free_bitmap(struct bitmap_page *bitmap)
  84 {
  85         struct bitmap_page *bp;
  86
  87         while (bitmap) {
  88                 bp = bitmap->next;
  89                 free_page((unsigned long)bitmap);
  90                 bitmap = bp;
  91         }
  92 }
  93
  94 struct bitmap_page *alloc_bitmap(unsigned int nr_bits)
  95 {
  96         struct bitmap_page *bitmap, *bp;
  97         unsigned int n;
  98
  99         if (!nr_bits)
 100                 return NULL;
 101
 102         bitmap = (struct bitmap_page *)get_zeroed_page(GFP_KERNEL);
 103         bp = bitmap;
 104         for (n = BITMAP_PAGE_BITS; n < nr_bits; n += BITMAP_PAGE_BITS) {
 105                 bp->next = (struct bitmap_page *)get_zeroed_page(GFP_KERNEL);
 106                 bp = bp->next;
 107                 if (!bp) {
 108                         free_bitmap(bitmap);
 109                         return NULL;
 110                 }
 111         }
 112         return bitmap;
 113 }
 114
 115 static int bitmap_set(struct bitmap_page *bitmap, unsigned long bit)
 116 {
 117         unsigned int n;
 118
 119         n = BITMAP_PAGE_BITS;
 120         while (bitmap && n <= bit) {
 121                 n += BITMAP_PAGE_BITS;
 122                 bitmap = bitmap->next;
 123         }
 124         if (!bitmap)
 125                 return -EINVAL;
 126         n -= BITMAP_PAGE_BITS;
 127         bit -= n;
 128         n = 0;
 129         while (bit >= BITS_PER_CHUNK) {
 130                 bit -= BITS_PER_CHUNK;
 131                 n++;
 132         }
 133         bitmap->chunks[n] |= (1UL << bit);
 134         return 0;
 135 }
 136
 137 unsigned long alloc_swap_page(int swap, struct bitmap_page *bitmap)
 138 {
 139         unsigned long offset;
 140
 141         offset = swp_offset(get_swap_page_of_type(swap));
 142         if (offset) {
 143                 if (bitmap_set(bitmap, offset)) {
 144                         swap_free(swp_entry(swap, offset));
 145                         offset = 0;
 146                 }
 147         }
 148         return offset;
 149 }
 150
 151 void free_all_swap_pages(int swap, struct bitmap_page *bitmap)
 152 {
 153         unsigned int bit, n;
 154         unsigned long test;
 155
 156         bit = 0;
 157         while (bitmap) {
 158                 for (n = 0; n < BITMAP_PAGE_CHUNKS; n++)
 159                         for (test = 1UL; test; test <<= 1) {
 160                                 if (bitmap->chunks[n] & test)
 161                                         swap_free(swp_entry(swap, bit));
 162                                 bit++;
 163                         }
 164                 bitmap = bitmap->next;
 165         }
 166 }
 167
 168 /**
 169  *      swsusp_shrink_memory -  Try to free as much memory as needed
 170  *
 171  *      ... but do not OOM-kill anyone
 172  *
 173  *      Notice: all userland should be stopped before it is called, or
 174  *      livelock is possible.
 175  */
 176
 177 #define SHRINK_BITE     10000
 178 static inline unsigned long __shrink_memory(long tmp)
 179 {
 180         if (tmp > SHRINK_BITE)
 181                 tmp = SHRINK_BITE;
 182         return shrink_all_memory(tmp);
 183 }
 184
 185 int swsusp_shrink_memory(void)
 186 {
 187         long size, tmp;
 188         struct zone *zone;
 189         unsigned long pages = 0;
 190         unsigned int i = 0;
 191         char *p = "-\\|/";
 192
 193         printk("Shrinking memory...  ");
 194         do {
 195                 size = 2 * count_highmem_pages();
 196                 size += size / 50 + count_data_pages() + PAGES_FOR_IO;
 197                 tmp = size;
 198                 for_each_zone (zone)
 199                         if (!is_highmem(zone) && populated_zone(zone)) {
 200                                 tmp -= zone->free_pages;
 201                                 tmp += zone->lowmem_reserve[ZONE_NORMAL];
 202                                 tmp += snapshot_additional_pages(zone);
 203                         }
 204                 if (tmp > 0) {
 205                         tmp = __shrink_memory(tmp);
 206                         if (!tmp)
 207                                 return -ENOMEM;
 208                         pages += tmp;
 209                 } else if (size > image_size / PAGE_SIZE) {
 210                         tmp = __shrink_memory(size - (image_size / PAGE_SIZE));
 211                         pages += tmp;
 212                 }
 213                 printk("\b%c", p[i++%4]);
 214         } while (tmp > 0);
 215         printk("\bdone (%lu pages freed)\n", pages);
 216
 217         return 0;
 218 }
 219
 220 int swsusp_suspend(void)
 221 {
 222         int error;
 223
 224         if ((error = arch_prepare_suspend()))
 225                 return error;
 226         local_irq_disable();
 227         /* At this point, device_suspend() has been called, but *not*
 228          * device_power_down(). We *must* device_power_down() now.
 229          * Otherwise, drivers for some devices (e.g. interrupt controllers)
 230          * become desynchronized with the actual state of the hardware
 231          * at resume time, and evil weirdness ensues.
 232          */
 233         if ((error = device_power_down(PMSG_FREEZE))) {
 234                 printk(KERN_ERR "Some devices failed to power down, aborting suspend\n");
 235                 goto Enable_irqs;
 236         }
 237
 238         if ((error = save_highmem())) {
 239                 printk(KERN_ERR "swsusp: Not enough free pages for highmem\n");
 240                 goto Restore_highmem;
 241         }
 242
 243         save_processor_state();
 244         if ((error = swsusp_arch_suspend()))
 245                 printk(KERN_ERR "Error %d suspending\n", error);
 246         /* Restore control flow magically appears here */
 247         restore_processor_state();
 248 Restore_highmem:
 249         restore_highmem();
 250         /* NOTE:  device_power_up() is just a resume() for devices
 251          * that suspended with irqs off ... no overall powerup.
 252          */
 253         device_power_up();
 254 Enable_irqs:
 255         local_irq_enable();
 256         return error;
 257 }
 258
 259 int swsusp_resume(void)
 260 {
 261         int error;
 262
 263         local_irq_disable();
 264         /* NOTE:  device_power_down() is just a suspend() with irqs off;
 265          * it has no special "power things down" semantics
 266          */
 267         if (device_power_down(PMSG_PRETHAW))
 268                 printk(KERN_ERR "Some devices failed to power down, very bad\n");
 269         /* We'll ignore saved state, but this gets preempt count (etc) right */
 270         save_processor_state();
 271         error = swsusp_arch_resume();
 272         /* Code below is only ever reached in case of failure. Otherwise
 273          * execution continues at place where swsusp_arch_suspend was called
 274          */
 275         BUG_ON(!error);
 276         /* The only reason why swsusp_arch_resume() can fail is memory being
 277          * very tight, so we have to free it as soon as we can to avoid
 278          * subsequent failures
 279          */
 280         swsusp_free();
 281         restore_processor_state();
 282         restore_highmem();
 283         touch_softlockup_watchdog();
 284         device_power_up();
 285         local_irq_enable();
 286         return error;
 287 }