Merge branch 'proc-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/adobriyan...
[linux-2.6-block.git] / kernel / power / swsusp.c
CommitLineData
1da177e4
LT
1/*
2 * linux/kernel/power/swsusp.c
3 *
96bc7aec 4 * This file provides code to write suspend image to swap and read it back.
1da177e4
LT
5 *
6 * Copyright (C) 1998-2001 Gabor Kuti <seasons@fornax.hu>
25761b6e 7 * Copyright (C) 1998,2001-2005 Pavel Machek <pavel@suse.cz>
1da177e4
LT
8 *
9 * This file is released under the GPLv2.
10 *
11 * I'd like to thank the following people for their work:
2e4d5822 12 *
1da177e4
LT
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 *
2e4d5822 17 * Steve Doddi <dirk@loth.demon.co.uk>:
1da177e4
LT
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 *
7088a5c0 33 * Rafael J. Wysocki <rjw@sisk.pl>
61159a31 34 * Reworked the freeing of memory and the handling of swap
7088a5c0 35 *
1da177e4
LT
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
1da177e4
LT
41#include <linux/mm.h>
42#include <linux/suspend.h>
1da177e4 43#include <linux/spinlock.h>
1da177e4
LT
44#include <linux/kernel.h>
45#include <linux/major.h>
46#include <linux/swap.h>
47#include <linux/pm.h>
1da177e4
LT
48#include <linux/swapops.h>
49#include <linux/bootmem.h>
50#include <linux/syscalls.h>
1da177e4 51#include <linux/highmem.h>
0d3a9abe 52#include <linux/time.h>
d1d241cc 53#include <linux/rbtree.h>
1da177e4
LT
54
55#include "power.h"
56
ca0aec0f 57/*
853609b6 58 * Preferred image size in bytes (tunable via /sys/power/image_size).
ca0aec0f 59 * When it is set to N, swsusp will do its best to ensure the image
853609b6 60 * size will not exceed N bytes, but if that is impossible, it will
ca0aec0f
RW
61 * try to create the smallest image possible.
62 */
853609b6 63unsigned long image_size = 500 * 1024 * 1024;
ca0aec0f 64
f577eb30
RW
65int in_suspend __nosavedata = 0;
66
1da177e4 67/**
f577eb30
RW
68 * The following functions are used for tracing the allocated
69 * swap pages, so that they can be freed in case of an error.
1da177e4 70 */
7088a5c0 71
d1d241cc
RW
72struct swsusp_extent {
73 struct rb_node node;
74 unsigned long start;
75 unsigned long end;
76};
1da177e4 77
d1d241cc 78static struct rb_root swsusp_extents = RB_ROOT;
7088a5c0 79
d1d241cc 80static int swsusp_extents_insert(unsigned long swap_offset)
7088a5c0 81{
d1d241cc
RW
82 struct rb_node **new = &(swsusp_extents.rb_node);
83 struct rb_node *parent = NULL;
84 struct swsusp_extent *ext;
85
86 /* Figure out where to put the new node */
87 while (*new) {
88 ext = container_of(*new, struct swsusp_extent, node);
89 parent = *new;
90 if (swap_offset < ext->start) {
91 /* Try to merge */
92 if (swap_offset == ext->start - 1) {
93 ext->start--;
94 return 0;
95 }
96 new = &((*new)->rb_left);
97 } else if (swap_offset > ext->end) {
98 /* Try to merge */
99 if (swap_offset == ext->end + 1) {
100 ext->end++;
101 return 0;
102 }
103 new = &((*new)->rb_right);
104 } else {
105 /* It already is in the tree */
106 return -EINVAL;
7088a5c0 107 }
1da177e4 108 }
d1d241cc
RW
109 /* Add the new node and rebalance the tree. */
110 ext = kzalloc(sizeof(struct swsusp_extent), GFP_KERNEL);
111 if (!ext)
112 return -ENOMEM;
113
114 ext->start = swap_offset;
115 ext->end = swap_offset;
116 rb_link_node(&ext->node, parent, new);
117 rb_insert_color(&ext->node, &swsusp_extents);
f577eb30 118 return 0;
7088a5c0 119}
1da177e4 120
d1d241cc
RW
121/**
122 * alloc_swapdev_block - allocate a swap page and register that it has
123 * been allocated, so that it can be freed in case of an error.
124 */
125
126sector_t alloc_swapdev_block(int swap)
7088a5c0 127{
f577eb30
RW
128 unsigned long offset;
129
130 offset = swp_offset(get_swap_page_of_type(swap));
131 if (offset) {
d1d241cc 132 if (swsusp_extents_insert(offset))
f577eb30 133 swap_free(swp_entry(swap, offset));
3aef83e0
RW
134 else
135 return swapdev_block(swap, offset);
7088a5c0 136 }
3aef83e0 137 return 0;
7088a5c0 138}
1da177e4 139
d1d241cc
RW
140/**
141 * free_all_swap_pages - free swap pages allocated for saving image data.
142 * It also frees the extents used to register which swap entres had been
143 * allocated.
144 */
145
146void free_all_swap_pages(int swap)
7088a5c0 147{
d1d241cc
RW
148 struct rb_node *node;
149
150 while ((node = swsusp_extents.rb_node)) {
151 struct swsusp_extent *ext;
152 unsigned long offset;
153
154 ext = container_of(node, struct swsusp_extent, node);
155 rb_erase(node, &swsusp_extents);
156 for (offset = ext->start; offset <= ext->end; offset++)
157 swap_free(swp_entry(swap, offset));
158
159 kfree(ext);
1da177e4 160 }
7088a5c0
RW
161}
162
d1d241cc
RW
163int swsusp_swap_in_use(void)
164{
165 return (swsusp_extents.rb_node != NULL);
166}
167
0d3a9abe
RW
168/**
169 * swsusp_show_speed - print the time elapsed between two events represented by
170 * @start and @stop
171 *
172 * @nr_pages - number of pages processed between @start and @stop
173 * @msg - introductory message to print
174 */
175
176void swsusp_show_speed(struct timeval *start, struct timeval *stop,
177 unsigned nr_pages, char *msg)
178{
179 s64 elapsed_centisecs64;
180 int centisecs;
181 int k;
182 int kps;
183
184 elapsed_centisecs64 = timeval_to_ns(stop) - timeval_to_ns(start);
185 do_div(elapsed_centisecs64, NSEC_PER_SEC / 100);
186 centisecs = elapsed_centisecs64;
187 if (centisecs == 0)
188 centisecs = 1; /* avoid div-by-zero */
189 k = nr_pages * (PAGE_SIZE / 1024);
190 kps = (k * 100) / centisecs;
23976728
RW
191 printk(KERN_INFO "PM: %s %d kbytes in %d.%02d seconds (%d.%02d MB/s)\n",
192 msg, k,
0d3a9abe
RW
193 centisecs / 100, centisecs % 100,
194 kps / 1000, (kps % 1000) / 10);
195}
196
72a97e08
RW
197/**
198 * swsusp_shrink_memory - Try to free as much memory as needed
199 *
200 * ... but do not OOM-kill anyone
201 *
202 * Notice: all userland should be stopped before it is called, or
203 * livelock is possible.
204 */
205
206#define SHRINK_BITE 10000
d6277db4
RW
207static inline unsigned long __shrink_memory(long tmp)
208{
209 if (tmp > SHRINK_BITE)
210 tmp = SHRINK_BITE;
211 return shrink_all_memory(tmp);
212}
72a97e08
RW
213
214int swsusp_shrink_memory(void)
215{
8357376d 216 long tmp;
72a97e08
RW
217 struct zone *zone;
218 unsigned long pages = 0;
219 unsigned int i = 0;
220 char *p = "-\\|/";
0d3a9abe 221 struct timeval start, stop;
72a97e08 222
23976728 223 printk(KERN_INFO "PM: Shrinking memory... ");
0d3a9abe 224 do_gettimeofday(&start);
72a97e08 225 do {
8357376d
RW
226 long size, highmem_size;
227
228 highmem_size = count_highmem_pages();
56f99bcb 229 size = count_data_pages() + PAGES_FOR_IO + SPARE_PAGES;
b3a93a25 230 tmp = size;
8357376d 231 size += highmem_size;
72a97e08 232 for_each_zone (zone)
8357376d 233 if (populated_zone(zone)) {
c75fd0ee 234 tmp += snapshot_additional_pages(zone);
8357376d 235 if (is_highmem(zone)) {
d23ad423
CL
236 highmem_size -=
237 zone_page_state(zone, NR_FREE_PAGES);
8357376d 238 } else {
d23ad423 239 tmp -= zone_page_state(zone, NR_FREE_PAGES);
8357376d 240 tmp += zone->lowmem_reserve[ZONE_NORMAL];
8357376d 241 }
a938c356 242 }
8357376d
RW
243
244 if (highmem_size < 0)
245 highmem_size = 0;
246
247 tmp += highmem_size;
72a97e08 248 if (tmp > 0) {
d6277db4 249 tmp = __shrink_memory(tmp);
72a97e08
RW
250 if (!tmp)
251 return -ENOMEM;
252 pages += tmp;
853609b6 253 } else if (size > image_size / PAGE_SIZE) {
d6277db4 254 tmp = __shrink_memory(size - (image_size / PAGE_SIZE));
b3a93a25 255 pages += tmp;
72a97e08 256 }
72a97e08
RW
257 printk("\b%c", p[i++%4]);
258 } while (tmp > 0);
0d3a9abe 259 do_gettimeofday(&stop);
72a97e08 260 printk("\bdone (%lu pages freed)\n", pages);
0d3a9abe 261 swsusp_show_speed(&start, &stop, pages, "Freed");
72a97e08
RW
262
263 return 0;
264}
3f4b0ef7
RW
265
266/*
267 * Platforms, like ACPI, may want us to save some memory used by them during
268 * hibernation and to restore the contents of this memory during the subsequent
269 * resume. The code below implements a mechanism allowing us to do that.
270 */
271
272struct nvs_page {
273 unsigned long phys_start;
274 unsigned int size;
275 void *kaddr;
276 void *data;
277 struct list_head node;
278};
279
280static LIST_HEAD(nvs_list);
281
282/**
283 * hibernate_nvs_register - register platform NVS memory region to save
284 * @start - physical address of the region
285 * @size - size of the region
286 *
287 * The NVS region need not be page-aligned (both ends) and we arrange
288 * things so that the data from page-aligned addresses in this region will
289 * be copied into separate RAM pages.
290 */
291int hibernate_nvs_register(unsigned long start, unsigned long size)
292{
293 struct nvs_page *entry, *next;
294
295 while (size > 0) {
296 unsigned int nr_bytes;
297
298 entry = kzalloc(sizeof(struct nvs_page), GFP_KERNEL);
299 if (!entry)
300 goto Error;
301
302 list_add_tail(&entry->node, &nvs_list);
303 entry->phys_start = start;
304 nr_bytes = PAGE_SIZE - (start & ~PAGE_MASK);
305 entry->size = (size < nr_bytes) ? size : nr_bytes;
306
307 start += entry->size;
308 size -= entry->size;
309 }
310 return 0;
311
312 Error:
313 list_for_each_entry_safe(entry, next, &nvs_list, node) {
314 list_del(&entry->node);
315 kfree(entry);
316 }
317 return -ENOMEM;
318}
319
320/**
321 * hibernate_nvs_free - free data pages allocated for saving NVS regions
322 */
323void hibernate_nvs_free(void)
324{
325 struct nvs_page *entry;
326
327 list_for_each_entry(entry, &nvs_list, node)
328 if (entry->data) {
329 free_page((unsigned long)entry->data);
330 entry->data = NULL;
331 if (entry->kaddr) {
332 iounmap(entry->kaddr);
333 entry->kaddr = NULL;
334 }
335 }
336}
337
338/**
339 * hibernate_nvs_alloc - allocate memory necessary for saving NVS regions
340 */
341int hibernate_nvs_alloc(void)
342{
343 struct nvs_page *entry;
344
345 list_for_each_entry(entry, &nvs_list, node) {
346 entry->data = (void *)__get_free_page(GFP_KERNEL);
347 if (!entry->data) {
348 hibernate_nvs_free();
349 return -ENOMEM;
350 }
351 }
352 return 0;
353}
354
355/**
356 * hibernate_nvs_save - save NVS memory regions
357 */
358void hibernate_nvs_save(void)
359{
360 struct nvs_page *entry;
361
362 printk(KERN_INFO "PM: Saving platform NVS memory\n");
363
364 list_for_each_entry(entry, &nvs_list, node)
365 if (entry->data) {
366 entry->kaddr = ioremap(entry->phys_start, entry->size);
367 memcpy(entry->data, entry->kaddr, entry->size);
368 }
369}
370
371/**
372 * hibernate_nvs_restore - restore NVS memory regions
373 *
374 * This function is going to be called with interrupts disabled, so it
375 * cannot iounmap the virtual addresses used to access the NVS region.
376 */
377void hibernate_nvs_restore(void)
378{
379 struct nvs_page *entry;
380
381 printk(KERN_INFO "PM: Restoring platform NVS memory\n");
382
383 list_for_each_entry(entry, &nvs_list, node)
384 if (entry->data)
385 memcpy(entry->kaddr, entry->data, entry->size);
386}