2 * linux/kernel/power/swsusp.c
4 * This file provides code to write suspend image to swap and read it back.
6 * Copyright (C) 1998-2001 Gabor Kuti <seasons@fornax.hu>
7 * Copyright (C) 1998,2001-2005 Pavel Machek <pavel@suse.cz>
9 * This file is released under the GPLv2.
11 * I'd like to thank the following people for their work:
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.
17 * Steve Doddi <dirk@loth.demon.co.uk>:
18 * Support the possibility of hardware state restoring.
20 * Raph <grey.havens@earthling.net>:
21 * Support for preserving states of network devices and virtual console
22 * (including X and svgatextmode)
24 * Kurt Garloff <garloff@suse.de>:
25 * Straightened the critical function in order to prevent compilers from
26 * playing tricks with local variables.
28 * Andreas Mohr <a.mohr@mailto.de>
30 * Alex Badea <vampire@go.ro>:
33 * Rafael J. Wysocki <rjw@sisk.pl>
34 * Reworked the freeing of memory and the handling of swap
36 * More state savers are welcome. Especially for the scsi layer...
38 * For TODOs,FIXMEs also look in Documentation/power/swsusp.txt
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>
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 #include <linux/rbtree.h>
58 * Preferred image size in bytes (tunable via /sys/power/image_size).
59 * When it is set to N, swsusp will do its best to ensure the image
60 * size will not exceed N bytes, but if that is impossible, it will
61 * try to create the smallest image possible.
63 unsigned long image_size
= 500 * 1024 * 1024;
65 int in_suspend __nosavedata
= 0;
68 * The following functions are used for tracing the allocated
69 * swap pages, so that they can be freed in case of an error.
72 struct swsusp_extent
{
78 static struct rb_root swsusp_extents
= RB_ROOT
;
80 static int swsusp_extents_insert(unsigned long swap_offset
)
82 struct rb_node
**new = &(swsusp_extents
.rb_node
);
83 struct rb_node
*parent
= NULL
;
84 struct swsusp_extent
*ext
;
86 /* Figure out where to put the new node */
88 ext
= container_of(*new, struct swsusp_extent
, node
);
90 if (swap_offset
< ext
->start
) {
92 if (swap_offset
== ext
->start
- 1) {
96 new = &((*new)->rb_left
);
97 } else if (swap_offset
> ext
->end
) {
99 if (swap_offset
== ext
->end
+ 1) {
103 new = &((*new)->rb_right
);
105 /* It already is in the tree */
109 /* Add the new node and rebalance the tree. */
110 ext
= kzalloc(sizeof(struct swsusp_extent
), GFP_KERNEL
);
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
);
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.
126 sector_t
alloc_swapdev_block(int swap
)
128 unsigned long offset
;
130 offset
= swp_offset(get_swap_page_of_type(swap
));
132 if (swsusp_extents_insert(offset
))
133 swap_free(swp_entry(swap
, offset
));
135 return swapdev_block(swap
, offset
);
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
146 void free_all_swap_pages(int swap
)
148 struct rb_node
*node
;
150 while ((node
= swsusp_extents
.rb_node
)) {
151 struct swsusp_extent
*ext
;
152 unsigned long offset
;
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
));
163 int swsusp_swap_in_use(void)
165 return (swsusp_extents
.rb_node
!= NULL
);
169 * swsusp_show_speed - print the time elapsed between two events represented by
172 * @nr_pages - number of pages processed between @start and @stop
173 * @msg - introductory message to print
176 void swsusp_show_speed(struct timeval
*start
, struct timeval
*stop
,
177 unsigned nr_pages
, char *msg
)
179 s64 elapsed_centisecs64
;
184 elapsed_centisecs64
= timeval_to_ns(stop
) - timeval_to_ns(start
);
185 do_div(elapsed_centisecs64
, NSEC_PER_SEC
/ 100);
186 centisecs
= elapsed_centisecs64
;
188 centisecs
= 1; /* avoid div-by-zero */
189 k
= nr_pages
* (PAGE_SIZE
/ 1024);
190 kps
= (k
* 100) / centisecs
;
191 printk(KERN_INFO
"PM: %s %d kbytes in %d.%02d seconds (%d.%02d MB/s)\n",
193 centisecs
/ 100, centisecs
% 100,
194 kps
/ 1000, (kps
% 1000) / 10);
198 * swsusp_shrink_memory - Try to free as much memory as needed
200 * ... but do not OOM-kill anyone
202 * Notice: all userland should be stopped before it is called, or
203 * livelock is possible.
206 #define SHRINK_BITE 10000
207 static inline unsigned long __shrink_memory(long tmp
)
209 if (tmp
> SHRINK_BITE
)
211 return shrink_all_memory(tmp
);
214 int swsusp_shrink_memory(void)
218 unsigned long pages
= 0;
221 struct timeval start
, stop
;
223 printk(KERN_INFO
"PM: Shrinking memory... ");
224 do_gettimeofday(&start
);
226 long size
, highmem_size
;
228 highmem_size
= count_highmem_pages();
229 size
= count_data_pages() + PAGES_FOR_IO
+ SPARE_PAGES
;
231 size
+= highmem_size
;
233 if (populated_zone(zone
)) {
234 tmp
+= snapshot_additional_pages(zone
);
235 if (is_highmem(zone
)) {
237 zone_page_state(zone
, NR_FREE_PAGES
);
239 tmp
-= zone_page_state(zone
, NR_FREE_PAGES
);
240 tmp
+= zone
->lowmem_reserve
[ZONE_NORMAL
];
244 if (highmem_size
< 0)
249 tmp
= __shrink_memory(tmp
);
253 } else if (size
> image_size
/ PAGE_SIZE
) {
254 tmp
= __shrink_memory(size
- (image_size
/ PAGE_SIZE
));
257 printk("\b%c", p
[i
++%4]);
259 do_gettimeofday(&stop
);
260 printk("\bdone (%lu pages freed)\n", pages
);
261 swsusp_show_speed(&start
, &stop
, pages
, "Freed");
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.
273 unsigned long phys_start
;
277 struct list_head node
;
280 static LIST_HEAD(nvs_list
);
283 * hibernate_nvs_register - register platform NVS memory region to save
284 * @start - physical address of the region
285 * @size - size of the region
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.
291 int hibernate_nvs_register(unsigned long start
, unsigned long size
)
293 struct nvs_page
*entry
, *next
;
296 unsigned int nr_bytes
;
298 entry
= kzalloc(sizeof(struct nvs_page
), GFP_KERNEL
);
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
;
307 start
+= entry
->size
;
313 list_for_each_entry_safe(entry
, next
, &nvs_list
, node
) {
314 list_del(&entry
->node
);
321 * hibernate_nvs_free - free data pages allocated for saving NVS regions
323 void hibernate_nvs_free(void)
325 struct nvs_page
*entry
;
327 list_for_each_entry(entry
, &nvs_list
, node
)
329 free_page((unsigned long)entry
->data
);
332 iounmap(entry
->kaddr
);
339 * hibernate_nvs_alloc - allocate memory necessary for saving NVS regions
341 int hibernate_nvs_alloc(void)
343 struct nvs_page
*entry
;
345 list_for_each_entry(entry
, &nvs_list
, node
) {
346 entry
->data
= (void *)__get_free_page(GFP_KERNEL
);
348 hibernate_nvs_free();
356 * hibernate_nvs_save - save NVS memory regions
358 void hibernate_nvs_save(void)
360 struct nvs_page
*entry
;
362 printk(KERN_INFO
"PM: Saving platform NVS memory\n");
364 list_for_each_entry(entry
, &nvs_list
, node
)
366 entry
->kaddr
= ioremap(entry
->phys_start
, entry
->size
);
367 memcpy(entry
->data
, entry
->kaddr
, entry
->size
);
372 * hibernate_nvs_restore - restore NVS memory regions
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.
377 void hibernate_nvs_restore(void)
379 struct nvs_page
*entry
;
381 printk(KERN_INFO
"PM: Restoring platform NVS memory\n");
383 list_for_each_entry(entry
, &nvs_list
, node
)
385 memcpy(entry
->kaddr
, entry
->data
, entry
->size
);