Commit | Line | Data |
---|---|---|
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> |
1da177e4 LT |
52 | |
53 | #include "power.h" | |
54 | ||
ca0aec0f | 55 | /* |
853609b6 | 56 | * Preferred image size in bytes (tunable via /sys/power/image_size). |
ca0aec0f | 57 | * When it is set to N, swsusp will do its best to ensure the image |
853609b6 | 58 | * size will not exceed N bytes, but if that is impossible, it will |
ca0aec0f RW |
59 | * try to create the smallest image possible. |
60 | */ | |
853609b6 | 61 | unsigned long image_size = 500 * 1024 * 1024; |
ca0aec0f | 62 | |
f577eb30 RW |
63 | int in_suspend __nosavedata = 0; |
64 | ||
1da177e4 | 65 | /** |
f577eb30 RW |
66 | * The following functions are used for tracing the allocated |
67 | * swap pages, so that they can be freed in case of an error. | |
7088a5c0 | 68 | * |
f577eb30 | 69 | * The functions operate on a linked bitmap structure defined |
61159a31 | 70 | * in power.h |
1da177e4 | 71 | */ |
7088a5c0 | 72 | |
61159a31 | 73 | void free_bitmap(struct bitmap_page *bitmap) |
1da177e4 | 74 | { |
f577eb30 | 75 | struct bitmap_page *bp; |
1da177e4 | 76 | |
f577eb30 RW |
77 | while (bitmap) { |
78 | bp = bitmap->next; | |
79 | free_page((unsigned long)bitmap); | |
80 | bitmap = bp; | |
7088a5c0 RW |
81 | } |
82 | } | |
83 | ||
61159a31 | 84 | struct bitmap_page *alloc_bitmap(unsigned int nr_bits) |
7088a5c0 | 85 | { |
f577eb30 RW |
86 | struct bitmap_page *bitmap, *bp; |
87 | unsigned int n; | |
7088a5c0 | 88 | |
f577eb30 | 89 | if (!nr_bits) |
7088a5c0 RW |
90 | return NULL; |
91 | ||
f577eb30 RW |
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); | |
7088a5c0 RW |
99 | return NULL; |
100 | } | |
1da177e4 | 101 | } |
f577eb30 | 102 | return bitmap; |
1da177e4 LT |
103 | } |
104 | ||
f577eb30 | 105 | static int bitmap_set(struct bitmap_page *bitmap, unsigned long bit) |
1da177e4 | 106 | { |
f577eb30 RW |
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++; | |
7088a5c0 | 122 | } |
f577eb30 RW |
123 | bitmap->chunks[n] |= (1UL << bit); |
124 | return 0; | |
7088a5c0 | 125 | } |
1da177e4 | 126 | |
61159a31 | 127 | unsigned long alloc_swap_page(int swap, struct bitmap_page *bitmap) |
7088a5c0 | 128 | { |
f577eb30 RW |
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 | } | |
7088a5c0 | 137 | } |
f577eb30 | 138 | return offset; |
7088a5c0 | 139 | } |
1da177e4 | 140 | |
61159a31 | 141 | void free_all_swap_pages(int swap, struct bitmap_page *bitmap) |
7088a5c0 | 142 | { |
f577eb30 RW |
143 | unsigned int bit, n; |
144 | unsigned long test; | |
7088a5c0 | 145 | |
f577eb30 RW |
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; | |
1da177e4 | 155 | } |
7088a5c0 RW |
156 | } |
157 | ||
72a97e08 RW |
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 | |
d6277db4 RW |
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 | } | |
72a97e08 RW |
174 | |
175 | int swsusp_shrink_memory(void) | |
176 | { | |
b3a93a25 | 177 | long size, tmp; |
72a97e08 RW |
178 | struct zone *zone; |
179 | unsigned long pages = 0; | |
180 | unsigned int i = 0; | |
181 | char *p = "-\\|/"; | |
182 | ||
183 | printk("Shrinking memory... "); | |
184 | do { | |
ce4ab001 | 185 | size = 2 * count_special_pages(); |
b3a93a25 RW |
186 | size += size / 50 + count_data_pages(); |
187 | size += (size + PBES_PER_PAGE - 1) / PBES_PER_PAGE + | |
72a97e08 | 188 | PAGES_FOR_IO; |
b3a93a25 | 189 | tmp = size; |
72a97e08 | 190 | for_each_zone (zone) |
a938c356 | 191 | if (!is_highmem(zone) && populated_zone(zone)) { |
72a97e08 | 192 | tmp -= zone->free_pages; |
a938c356 RW |
193 | tmp += zone->lowmem_reserve[ZONE_NORMAL]; |
194 | } | |
72a97e08 | 195 | if (tmp > 0) { |
d6277db4 | 196 | tmp = __shrink_memory(tmp); |
72a97e08 RW |
197 | if (!tmp) |
198 | return -ENOMEM; | |
199 | pages += tmp; | |
853609b6 | 200 | } else if (size > image_size / PAGE_SIZE) { |
d6277db4 | 201 | tmp = __shrink_memory(size - (image_size / PAGE_SIZE)); |
b3a93a25 | 202 | pages += tmp; |
72a97e08 | 203 | } |
72a97e08 RW |
204 | printk("\b%c", p[i++%4]); |
205 | } while (tmp > 0); | |
206 | printk("\bdone (%lu pages freed)\n", pages); | |
207 | ||
208 | return 0; | |
209 | } | |
210 | ||
1da177e4 LT |
211 | int swsusp_suspend(void) |
212 | { | |
213 | int error; | |
0fbeb5a4 | 214 | |
1da177e4 LT |
215 | if ((error = arch_prepare_suspend())) |
216 | return error; | |
217 | local_irq_disable(); | |
218 | /* At this point, device_suspend() has been called, but *not* | |
219 | * device_power_down(). We *must* device_power_down() now. | |
220 | * Otherwise, drivers for some devices (e.g. interrupt controllers) | |
221 | * become desynchronized with the actual state of the hardware | |
222 | * at resume time, and evil weirdness ensues. | |
223 | */ | |
224 | if ((error = device_power_down(PMSG_FREEZE))) { | |
99dc7d63 | 225 | printk(KERN_ERR "Some devices failed to power down, aborting suspend\n"); |
0fbeb5a4 | 226 | goto Enable_irqs; |
1da177e4 | 227 | } |
47b724f3 | 228 | |
ce4ab001 | 229 | if ((error = save_special_mem())) { |
0fbeb5a4 RW |
230 | printk(KERN_ERR "swsusp: Not enough free pages for highmem\n"); |
231 | goto Restore_highmem; | |
47b724f3 PM |
232 | } |
233 | ||
1da177e4 LT |
234 | save_processor_state(); |
235 | if ((error = swsusp_arch_suspend())) | |
99dc7d63 | 236 | printk(KERN_ERR "Error %d suspending\n", error); |
1da177e4 LT |
237 | /* Restore control flow magically appears here */ |
238 | restore_processor_state(); | |
0fbeb5a4 | 239 | Restore_highmem: |
ce4ab001 | 240 | restore_special_mem(); |
1da177e4 | 241 | device_power_up(); |
0fbeb5a4 | 242 | Enable_irqs: |
1da177e4 LT |
243 | local_irq_enable(); |
244 | return error; | |
245 | } | |
246 | ||
247 | int swsusp_resume(void) | |
248 | { | |
249 | int error; | |
250 | local_irq_disable(); | |
251 | if (device_power_down(PMSG_FREEZE)) | |
252 | printk(KERN_ERR "Some devices failed to power down, very bad\n"); | |
253 | /* We'll ignore saved state, but this gets preempt count (etc) right */ | |
254 | save_processor_state(); | |
255 | error = swsusp_arch_resume(); | |
256 | /* Code below is only ever reached in case of failure. Otherwise | |
257 | * execution continues at place where swsusp_arch_suspend was called | |
258 | */ | |
259 | BUG_ON(!error); | |
2c1b4a5c RW |
260 | /* The only reason why swsusp_arch_resume() can fail is memory being |
261 | * very tight, so we have to free it as soon as we can to avoid | |
262 | * subsequent failures | |
263 | */ | |
264 | swsusp_free(); | |
1da177e4 | 265 | restore_processor_state(); |
ce4ab001 | 266 | restore_special_mem(); |
8446f1d3 | 267 | touch_softlockup_watchdog(); |
1da177e4 LT |
268 | device_power_up(); |
269 | local_irq_enable(); | |
270 | return error; | |
271 | } |