diff options
-rw-r--r-- | include/linux/suspend.h | 6 | ||||
-rw-r--r-- | kernel/power/Makefile | 2 | ||||
-rw-r--r-- | kernel/power/power.h | 17 | ||||
-rw-r--r-- | kernel/power/snapshot.c | 464 | ||||
-rw-r--r-- | kernel/power/swsusp.c | 440 |
5 files changed, 492 insertions, 437 deletions
diff --git a/include/linux/suspend.h b/include/linux/suspend.h index ba448c760168..380915e9563d 100644 --- a/include/linux/suspend.h +++ b/include/linux/suspend.h | |||
@@ -74,4 +74,10 @@ void __restore_processor_state(struct saved_context *ctxt); | |||
74 | extern unsigned long get_usable_page(gfp_t gfp_mask); | 74 | extern unsigned long get_usable_page(gfp_t gfp_mask); |
75 | extern void free_eaten_memory(void); | 75 | extern void free_eaten_memory(void); |
76 | 76 | ||
77 | /* | ||
78 | * XXX: We try to keep some more pages free so that I/O operations succeed | ||
79 | * without paging. Might this be more? | ||
80 | */ | ||
81 | #define PAGES_FOR_IO 512 | ||
82 | |||
77 | #endif /* _LINUX_SWSUSP_H */ | 83 | #endif /* _LINUX_SWSUSP_H */ |
diff --git a/kernel/power/Makefile b/kernel/power/Makefile index 2f438d0eaa13..c71eb4579c07 100644 --- a/kernel/power/Makefile +++ b/kernel/power/Makefile | |||
@@ -4,7 +4,7 @@ EXTRA_CFLAGS += -DDEBUG | |||
4 | endif | 4 | endif |
5 | 5 | ||
6 | obj-y := main.o process.o console.o pm.o | 6 | obj-y := main.o process.o console.o pm.o |
7 | obj-$(CONFIG_SOFTWARE_SUSPEND) += swsusp.o disk.o | 7 | obj-$(CONFIG_SOFTWARE_SUSPEND) += swsusp.o disk.o snapshot.o |
8 | 8 | ||
9 | obj-$(CONFIG_SUSPEND_SMP) += smp.o | 9 | obj-$(CONFIG_SUSPEND_SMP) += smp.o |
10 | 10 | ||
diff --git a/kernel/power/power.h b/kernel/power/power.h index 6748de23e83c..e54dd8435de7 100644 --- a/kernel/power/power.h +++ b/kernel/power/power.h | |||
@@ -53,3 +53,20 @@ extern void thaw_processes(void); | |||
53 | 53 | ||
54 | extern int pm_prepare_console(void); | 54 | extern int pm_prepare_console(void); |
55 | extern void pm_restore_console(void); | 55 | extern void pm_restore_console(void); |
56 | |||
57 | |||
58 | /* References to section boundaries */ | ||
59 | extern const void __nosave_begin, __nosave_end; | ||
60 | |||
61 | extern unsigned int nr_copy_pages; | ||
62 | extern suspend_pagedir_t *pagedir_nosave; | ||
63 | extern suspend_pagedir_t *pagedir_save; | ||
64 | |||
65 | extern asmlinkage int swsusp_arch_suspend(void); | ||
66 | extern asmlinkage int swsusp_arch_resume(void); | ||
67 | |||
68 | extern int restore_highmem(void); | ||
69 | extern void free_pagedir(struct pbe *pblist); | ||
70 | extern struct pbe * alloc_pagedir(unsigned nr_pages); | ||
71 | extern void create_pbe_list(struct pbe *pblist, unsigned nr_pages); | ||
72 | extern int enough_swap(void); | ||
diff --git a/kernel/power/snapshot.c b/kernel/power/snapshot.c new file mode 100644 index 000000000000..0f0a7f306b0d --- /dev/null +++ b/kernel/power/snapshot.c | |||
@@ -0,0 +1,464 @@ | |||
1 | /* | ||
2 | * linux/kernel/power/swsusp.c | ||
3 | * | ||
4 | * This file is to realize architecture-independent | ||
5 | * machine suspend feature using pretty near only high-level routines | ||
6 | * | ||
7 | * Copyright (C) 1998-2005 Pavel Machek <pavel@suse.cz> | ||
8 | * | ||
9 | * This file is released under the GPLv2, and is based on swsusp.c. | ||
10 | * | ||
11 | */ | ||
12 | |||
13 | |||
14 | #include <linux/module.h> | ||
15 | #include <linux/mm.h> | ||
16 | #include <linux/suspend.h> | ||
17 | #include <linux/smp_lock.h> | ||
18 | #include <linux/file.h> | ||
19 | #include <linux/utsname.h> | ||
20 | #include <linux/version.h> | ||
21 | #include <linux/delay.h> | ||
22 | #include <linux/reboot.h> | ||
23 | #include <linux/bitops.h> | ||
24 | #include <linux/vt_kern.h> | ||
25 | #include <linux/kbd_kern.h> | ||
26 | #include <linux/keyboard.h> | ||
27 | #include <linux/spinlock.h> | ||
28 | #include <linux/genhd.h> | ||
29 | #include <linux/kernel.h> | ||
30 | #include <linux/major.h> | ||
31 | #include <linux/swap.h> | ||
32 | #include <linux/pm.h> | ||
33 | #include <linux/device.h> | ||
34 | #include <linux/buffer_head.h> | ||
35 | #include <linux/swapops.h> | ||
36 | #include <linux/bootmem.h> | ||
37 | #include <linux/syscalls.h> | ||
38 | #include <linux/console.h> | ||
39 | #include <linux/highmem.h> | ||
40 | #include <linux/bio.h> | ||
41 | #include <linux/mount.h> | ||
42 | |||
43 | #include <asm/uaccess.h> | ||
44 | #include <asm/mmu_context.h> | ||
45 | #include <asm/pgtable.h> | ||
46 | #include <asm/tlbflush.h> | ||
47 | #include <asm/io.h> | ||
48 | |||
49 | #include <linux/random.h> | ||
50 | #include <linux/crypto.h> | ||
51 | #include <asm/scatterlist.h> | ||
52 | |||
53 | #include "power.h" | ||
54 | |||
55 | |||
56 | |||
57 | |||
58 | #ifdef CONFIG_HIGHMEM | ||
59 | struct highmem_page { | ||
60 | char *data; | ||
61 | struct page *page; | ||
62 | struct highmem_page *next; | ||
63 | }; | ||
64 | |||
65 | static struct highmem_page *highmem_copy; | ||
66 | |||
67 | static int save_highmem_zone(struct zone *zone) | ||
68 | { | ||
69 | unsigned long zone_pfn; | ||
70 | mark_free_pages(zone); | ||
71 | for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) { | ||
72 | struct page *page; | ||
73 | struct highmem_page *save; | ||
74 | void *kaddr; | ||
75 | unsigned long pfn = zone_pfn + zone->zone_start_pfn; | ||
76 | |||
77 | if (!(pfn%1000)) | ||
78 | printk("."); | ||
79 | if (!pfn_valid(pfn)) | ||
80 | continue; | ||
81 | page = pfn_to_page(pfn); | ||
82 | /* | ||
83 | * This condition results from rvmalloc() sans vmalloc_32() | ||
84 | * and architectural memory reservations. This should be | ||
85 | * corrected eventually when the cases giving rise to this | ||
86 | * are better understood. | ||
87 | */ | ||
88 | if (PageReserved(page)) { | ||
89 | printk("highmem reserved page?!\n"); | ||
90 | continue; | ||
91 | } | ||
92 | BUG_ON(PageNosave(page)); | ||
93 | if (PageNosaveFree(page)) | ||
94 | continue; | ||
95 | save = kmalloc(sizeof(struct highmem_page), GFP_ATOMIC); | ||
96 | if (!save) | ||
97 | return -ENOMEM; | ||
98 | save->next = highmem_copy; | ||
99 | save->page = page; | ||
100 | save->data = (void *) get_zeroed_page(GFP_ATOMIC); | ||
101 | if (!save->data) { | ||
102 | kfree(save); | ||
103 | return -ENOMEM; | ||
104 | } | ||
105 | kaddr = kmap_atomic(page, KM_USER0); | ||
106 | memcpy(save->data, kaddr, PAGE_SIZE); | ||
107 | kunmap_atomic(kaddr, KM_USER0); | ||
108 | highmem_copy = save; | ||
109 | } | ||
110 | return 0; | ||
111 | } | ||
112 | #endif /* CONFIG_HIGHMEM */ | ||
113 | |||
114 | |||
115 | static int save_highmem(void) | ||
116 | { | ||
117 | #ifdef CONFIG_HIGHMEM | ||
118 | struct zone *zone; | ||
119 | int res = 0; | ||
120 | |||
121 | pr_debug("swsusp: Saving Highmem\n"); | ||
122 | for_each_zone (zone) { | ||
123 | if (is_highmem(zone)) | ||
124 | res = save_highmem_zone(zone); | ||
125 | if (res) | ||
126 | return res; | ||
127 | } | ||
128 | #endif | ||
129 | return 0; | ||
130 | } | ||
131 | |||
132 | int restore_highmem(void) | ||
133 | { | ||
134 | #ifdef CONFIG_HIGHMEM | ||
135 | printk("swsusp: Restoring Highmem\n"); | ||
136 | while (highmem_copy) { | ||
137 | struct highmem_page *save = highmem_copy; | ||
138 | void *kaddr; | ||
139 | highmem_copy = save->next; | ||
140 | |||
141 | kaddr = kmap_atomic(save->page, KM_USER0); | ||
142 | memcpy(kaddr, save->data, PAGE_SIZE); | ||
143 | kunmap_atomic(kaddr, KM_USER0); | ||
144 | free_page((long) save->data); | ||
145 | kfree(save); | ||
146 | } | ||
147 | #endif | ||
148 | return 0; | ||
149 | } | ||
150 | |||
151 | |||
152 | static int pfn_is_nosave(unsigned long pfn) | ||
153 | { | ||
154 | unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT; | ||
155 | unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT; | ||
156 | return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn); | ||
157 | } | ||
158 | |||
159 | /** | ||
160 | * saveable - Determine whether a page should be cloned or not. | ||
161 | * @pfn: The page | ||
162 | * | ||
163 | * We save a page if it's Reserved, and not in the range of pages | ||
164 | * statically defined as 'unsaveable', or if it isn't reserved, and | ||
165 | * isn't part of a free chunk of pages. | ||
166 | */ | ||
167 | |||
168 | static int saveable(struct zone * zone, unsigned long * zone_pfn) | ||
169 | { | ||
170 | unsigned long pfn = *zone_pfn + zone->zone_start_pfn; | ||
171 | struct page * page; | ||
172 | |||
173 | if (!pfn_valid(pfn)) | ||
174 | return 0; | ||
175 | |||
176 | page = pfn_to_page(pfn); | ||
177 | BUG_ON(PageReserved(page) && PageNosave(page)); | ||
178 | if (PageNosave(page)) | ||
179 | return 0; | ||
180 | if (PageReserved(page) && pfn_is_nosave(pfn)) { | ||
181 | pr_debug("[nosave pfn 0x%lx]", pfn); | ||
182 | return 0; | ||
183 | } | ||
184 | if (PageNosaveFree(page)) | ||
185 | return 0; | ||
186 | |||
187 | return 1; | ||
188 | } | ||
189 | |||
190 | static void count_data_pages(void) | ||
191 | { | ||
192 | struct zone *zone; | ||
193 | unsigned long zone_pfn; | ||
194 | |||
195 | nr_copy_pages = 0; | ||
196 | |||
197 | for_each_zone (zone) { | ||
198 | if (is_highmem(zone)) | ||
199 | continue; | ||
200 | mark_free_pages(zone); | ||
201 | for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) | ||
202 | nr_copy_pages += saveable(zone, &zone_pfn); | ||
203 | } | ||
204 | } | ||
205 | |||
206 | static void copy_data_pages(void) | ||
207 | { | ||
208 | struct zone *zone; | ||
209 | unsigned long zone_pfn; | ||
210 | struct pbe *pbe = pagedir_nosave, *p; | ||
211 | |||
212 | pr_debug("copy_data_pages(): pages to copy: %d\n", nr_copy_pages); | ||
213 | for_each_zone (zone) { | ||
214 | if (is_highmem(zone)) | ||
215 | continue; | ||
216 | mark_free_pages(zone); | ||
217 | /* This is necessary for swsusp_free() */ | ||
218 | for_each_pb_page (p, pagedir_nosave) | ||
219 | SetPageNosaveFree(virt_to_page(p)); | ||
220 | for_each_pbe(p, pagedir_nosave) | ||
221 | SetPageNosaveFree(virt_to_page(p->address)); | ||
222 | for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) { | ||
223 | if (saveable(zone, &zone_pfn)) { | ||
224 | struct page * page; | ||
225 | page = pfn_to_page(zone_pfn + zone->zone_start_pfn); | ||
226 | BUG_ON(!pbe); | ||
227 | pbe->orig_address = (unsigned long)page_address(page); | ||
228 | /* copy_page is not usable for copying task structs. */ | ||
229 | memcpy((void *)pbe->address, (void *)pbe->orig_address, PAGE_SIZE); | ||
230 | pbe = pbe->next; | ||
231 | } | ||
232 | } | ||
233 | } | ||
234 | BUG_ON(pbe); | ||
235 | } | ||
236 | |||
237 | |||
238 | /** | ||
239 | * free_pagedir - free pages allocated with alloc_pagedir() | ||
240 | */ | ||
241 | |||
242 | void free_pagedir(struct pbe *pblist) | ||
243 | { | ||
244 | struct pbe *pbe; | ||
245 | |||
246 | while (pblist) { | ||
247 | pbe = (pblist + PB_PAGE_SKIP)->next; | ||
248 | ClearPageNosave(virt_to_page(pblist)); | ||
249 | ClearPageNosaveFree(virt_to_page(pblist)); | ||
250 | free_page((unsigned long)pblist); | ||
251 | pblist = pbe; | ||
252 | } | ||
253 | } | ||
254 | |||
255 | /** | ||
256 | * fill_pb_page - Create a list of PBEs on a given memory page | ||
257 | */ | ||
258 | |||
259 | static inline void fill_pb_page(struct pbe *pbpage) | ||
260 | { | ||
261 | struct pbe *p; | ||
262 | |||
263 | p = pbpage; | ||
264 | pbpage += PB_PAGE_SKIP; | ||
265 | do | ||
266 | p->next = p + 1; | ||
267 | while (++p < pbpage); | ||
268 | } | ||
269 | |||
270 | /** | ||
271 | * create_pbe_list - Create a list of PBEs on top of a given chain | ||
272 | * of memory pages allocated with alloc_pagedir() | ||
273 | */ | ||
274 | |||
275 | void create_pbe_list(struct pbe *pblist, unsigned nr_pages) | ||
276 | { | ||
277 | struct pbe *pbpage, *p; | ||
278 | unsigned num = PBES_PER_PAGE; | ||
279 | |||
280 | for_each_pb_page (pbpage, pblist) { | ||
281 | if (num >= nr_pages) | ||
282 | break; | ||
283 | |||
284 | fill_pb_page(pbpage); | ||
285 | num += PBES_PER_PAGE; | ||
286 | } | ||
287 | if (pbpage) { | ||
288 | for (num -= PBES_PER_PAGE - 1, p = pbpage; num < nr_pages; p++, num++) | ||
289 | p->next = p + 1; | ||
290 | p->next = NULL; | ||
291 | } | ||
292 | pr_debug("create_pbe_list(): initialized %d PBEs\n", num); | ||
293 | } | ||
294 | |||
295 | static void *alloc_image_page(void) | ||
296 | { | ||
297 | void *res = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_COLD); | ||
298 | if (res) { | ||
299 | SetPageNosave(virt_to_page(res)); | ||
300 | SetPageNosaveFree(virt_to_page(res)); | ||
301 | } | ||
302 | return res; | ||
303 | } | ||
304 | |||
305 | /** | ||
306 | * alloc_pagedir - Allocate the page directory. | ||
307 | * | ||
308 | * First, determine exactly how many pages we need and | ||
309 | * allocate them. | ||
310 | * | ||
311 | * We arrange the pages in a chain: each page is an array of PBES_PER_PAGE | ||
312 | * struct pbe elements (pbes) and the last element in the page points | ||
313 | * to the next page. | ||
314 | * | ||
315 | * On each page we set up a list of struct_pbe elements. | ||
316 | */ | ||
317 | |||
318 | struct pbe * alloc_pagedir(unsigned nr_pages) | ||
319 | { | ||
320 | unsigned num; | ||
321 | struct pbe *pblist, *pbe; | ||
322 | |||
323 | if (!nr_pages) | ||
324 | return NULL; | ||
325 | |||
326 | pr_debug("alloc_pagedir(): nr_pages = %d\n", nr_pages); | ||
327 | pblist = (struct pbe *)alloc_image_page(); | ||
328 | /* FIXME: rewrite this ugly loop */ | ||
329 | for (pbe = pblist, num = PBES_PER_PAGE; pbe && num < nr_pages; | ||
330 | pbe = pbe->next, num += PBES_PER_PAGE) { | ||
331 | pbe += PB_PAGE_SKIP; | ||
332 | pbe->next = (struct pbe *)alloc_image_page(); | ||
333 | } | ||
334 | if (!pbe) { /* get_zeroed_page() failed */ | ||
335 | free_pagedir(pblist); | ||
336 | pblist = NULL; | ||
337 | } | ||
338 | return pblist; | ||
339 | } | ||
340 | |||
341 | /** | ||
342 | * Free pages we allocated for suspend. Suspend pages are alocated | ||
343 | * before atomic copy, so we need to free them after resume. | ||
344 | */ | ||
345 | |||
346 | void swsusp_free(void) | ||
347 | { | ||
348 | struct zone *zone; | ||
349 | unsigned long zone_pfn; | ||
350 | |||
351 | for_each_zone(zone) { | ||
352 | for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) | ||
353 | if (pfn_valid(zone_pfn + zone->zone_start_pfn)) { | ||
354 | struct page * page; | ||
355 | page = pfn_to_page(zone_pfn + zone->zone_start_pfn); | ||
356 | if (PageNosave(page) && PageNosaveFree(page)) { | ||
357 | ClearPageNosave(page); | ||
358 | ClearPageNosaveFree(page); | ||
359 | free_page((long) page_address(page)); | ||
360 | } | ||
361 | } | ||
362 | } | ||
363 | } | ||
364 | |||
365 | |||
366 | /** | ||
367 | * enough_free_mem - Make sure we enough free memory to snapshot. | ||
368 | * | ||
369 | * Returns TRUE or FALSE after checking the number of available | ||
370 | * free pages. | ||
371 | */ | ||
372 | |||
373 | static int enough_free_mem(void) | ||
374 | { | ||
375 | pr_debug("swsusp: available memory: %u pages\n", nr_free_pages()); | ||
376 | return nr_free_pages() > (nr_copy_pages + PAGES_FOR_IO + | ||
377 | nr_copy_pages/PBES_PER_PAGE + !!(nr_copy_pages%PBES_PER_PAGE)); | ||
378 | } | ||
379 | |||
380 | |||
381 | static int swsusp_alloc(void) | ||
382 | { | ||
383 | struct pbe * p; | ||
384 | |||
385 | pagedir_nosave = NULL; | ||
386 | |||
387 | if (MAX_PBES < nr_copy_pages / PBES_PER_PAGE + | ||
388 | !!(nr_copy_pages % PBES_PER_PAGE)) | ||
389 | return -ENOSPC; | ||
390 | |||
391 | if (!(pagedir_save = alloc_pagedir(nr_copy_pages))) { | ||
392 | printk(KERN_ERR "suspend: Allocating pagedir failed.\n"); | ||
393 | return -ENOMEM; | ||
394 | } | ||
395 | create_pbe_list(pagedir_save, nr_copy_pages); | ||
396 | pagedir_nosave = pagedir_save; | ||
397 | |||
398 | for_each_pbe (p, pagedir_save) { | ||
399 | p->address = (unsigned long)alloc_image_page(); | ||
400 | if (!p->address) { | ||
401 | printk(KERN_ERR "suspend: Allocating image pages failed.\n"); | ||
402 | swsusp_free(); | ||
403 | return -ENOMEM; | ||
404 | } | ||
405 | } | ||
406 | |||
407 | return 0; | ||
408 | } | ||
409 | |||
410 | static int suspend_prepare_image(void) | ||
411 | { | ||
412 | int error; | ||
413 | |||
414 | pr_debug("swsusp: critical section: \n"); | ||
415 | if (save_highmem()) { | ||
416 | printk(KERN_CRIT "swsusp: Not enough free pages for highmem\n"); | ||
417 | restore_highmem(); | ||
418 | return -ENOMEM; | ||
419 | } | ||
420 | |||
421 | drain_local_pages(); | ||
422 | count_data_pages(); | ||
423 | printk("swsusp: Need to copy %u pages\n", nr_copy_pages); | ||
424 | |||
425 | pr_debug("swsusp: pages needed: %u + %lu + %u, free: %u\n", | ||
426 | nr_copy_pages, | ||
427 | nr_copy_pages/PBES_PER_PAGE + !!(nr_copy_pages%PBES_PER_PAGE), | ||
428 | PAGES_FOR_IO, nr_free_pages()); | ||
429 | |||
430 | if (!enough_free_mem()) { | ||
431 | printk(KERN_ERR "swsusp: Not enough free memory\n"); | ||
432 | return -ENOMEM; | ||
433 | } | ||
434 | |||
435 | if (!enough_swap()) { | ||
436 | printk(KERN_ERR "swsusp: Not enough free swap\n"); | ||
437 | return -ENOSPC; | ||
438 | } | ||
439 | |||
440 | error = swsusp_alloc(); | ||
441 | if (error) | ||
442 | return error; | ||
443 | |||
444 | /* During allocating of suspend pagedir, new cold pages may appear. | ||
445 | * Kill them. | ||
446 | */ | ||
447 | drain_local_pages(); | ||
448 | copy_data_pages(); | ||
449 | |||
450 | /* | ||
451 | * End of critical section. From now on, we can write to memory, | ||
452 | * but we should not touch disk. This specially means we must _not_ | ||
453 | * touch swap space! Except we must write out our image of course. | ||
454 | */ | ||
455 | |||
456 | printk("swsusp: critical section/: done (%d pages copied)\n", nr_copy_pages ); | ||
457 | return 0; | ||
458 | } | ||
459 | |||
460 | |||
461 | asmlinkage int swsusp_save(void) | ||
462 | { | ||
463 | return suspend_prepare_image(); | ||
464 | } | ||
diff --git a/kernel/power/swsusp.c b/kernel/power/swsusp.c index ae46506e2137..fc50b5d2dd26 100644 --- a/kernel/power/swsusp.c +++ b/kernel/power/swsusp.c | |||
@@ -5,7 +5,7 @@ | |||
5 | * machine suspend feature using pretty near only high-level routines | 5 | * machine suspend feature using pretty near only high-level routines |
6 | * | 6 | * |
7 | * Copyright (C) 1998-2001 Gabor Kuti <seasons@fornax.hu> | 7 | * Copyright (C) 1998-2001 Gabor Kuti <seasons@fornax.hu> |
8 | * Copyright (C) 1998,2001-2004 Pavel Machek <pavel@suse.cz> | 8 | * Copyright (C) 1998,2001-2005 Pavel Machek <pavel@suse.cz> |
9 | * | 9 | * |
10 | * This file is released under the GPLv2. | 10 | * This file is released under the GPLv2. |
11 | * | 11 | * |
@@ -84,16 +84,10 @@ | |||
84 | #define MAXKEY 32 | 84 | #define MAXKEY 32 |
85 | #define MAXIV 32 | 85 | #define MAXIV 32 |
86 | 86 | ||
87 | /* References to section boundaries */ | ||
88 | extern const void __nosave_begin, __nosave_end; | ||
89 | |||
90 | /* Variables to be preserved over suspend */ | ||
91 | static int nr_copy_pages_check; | ||
92 | |||
93 | extern char resume_file[]; | 87 | extern char resume_file[]; |
94 | 88 | ||
95 | /* Local variables that should not be affected by save */ | 89 | /* Local variables that should not be affected by save */ |
96 | static unsigned int nr_copy_pages __nosavedata = 0; | 90 | unsigned int nr_copy_pages __nosavedata = 0; |
97 | 91 | ||
98 | /* Suspend pagedir is allocated before final copy, therefore it | 92 | /* Suspend pagedir is allocated before final copy, therefore it |
99 | must be freed after resume | 93 | must be freed after resume |
@@ -109,7 +103,7 @@ static unsigned int nr_copy_pages __nosavedata = 0; | |||
109 | MMU hardware. | 103 | MMU hardware. |
110 | */ | 104 | */ |
111 | suspend_pagedir_t *pagedir_nosave __nosavedata = NULL; | 105 | suspend_pagedir_t *pagedir_nosave __nosavedata = NULL; |
112 | static suspend_pagedir_t *pagedir_save; | 106 | suspend_pagedir_t *pagedir_save; |
113 | 107 | ||
114 | #define SWSUSP_SIG "S1SUSPEND" | 108 | #define SWSUSP_SIG "S1SUSPEND" |
115 | 109 | ||
@@ -124,12 +118,6 @@ static struct swsusp_header { | |||
124 | static struct swsusp_info swsusp_info; | 118 | static struct swsusp_info swsusp_info; |
125 | 119 | ||
126 | /* | 120 | /* |
127 | * XXX: We try to keep some more pages free so that I/O operations succeed | ||
128 | * without paging. Might this be more? | ||
129 | */ | ||
130 | #define PAGES_FOR_IO 512 | ||
131 | |||
132 | /* | ||
133 | * Saving part... | 121 | * Saving part... |
134 | */ | 122 | */ |
135 | 123 | ||
@@ -552,335 +540,6 @@ static int write_suspend_image(void) | |||
552 | goto Done; | 540 | goto Done; |
553 | } | 541 | } |
554 | 542 | ||
555 | |||
556 | #ifdef CONFIG_HIGHMEM | ||
557 | struct highmem_page { | ||
558 | char *data; | ||
559 | struct page *page; | ||
560 | struct highmem_page *next; | ||
561 | }; | ||
562 | |||
563 | static struct highmem_page *highmem_copy; | ||
564 | |||
565 | static int save_highmem_zone(struct zone *zone) | ||
566 | { | ||
567 | unsigned long zone_pfn; | ||
568 | mark_free_pages(zone); | ||
569 | for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) { | ||
570 | struct page *page; | ||
571 | struct highmem_page *save; | ||
572 | void *kaddr; | ||
573 | unsigned long pfn = zone_pfn + zone->zone_start_pfn; | ||
574 | |||
575 | if (!(pfn%1000)) | ||
576 | printk("."); | ||
577 | if (!pfn_valid(pfn)) | ||
578 | continue; | ||
579 | page = pfn_to_page(pfn); | ||
580 | /* | ||
581 | * PageReserved results from rvmalloc() sans vmalloc_32() | ||
582 | * and architectural memory reservations. | ||
583 | * | ||
584 | * rvmalloc should not cause this, because all implementations | ||
585 | * appear to always be using vmalloc_32 on architectures with | ||
586 | * highmem. This is a good thing, because we would like to save | ||
587 | * rvmalloc pages. | ||
588 | * | ||
589 | * It appears to be triggered by pages which do not point to | ||
590 | * valid memory (see arch/i386/mm/init.c:one_highpage_init(), | ||
591 | * which sets PageReserved if the page does not point to valid | ||
592 | * RAM. | ||
593 | * | ||
594 | * XXX: must remove usage of PageReserved! | ||
595 | */ | ||
596 | if (PageReserved(page)) | ||
597 | continue; | ||
598 | BUG_ON(PageNosave(page)); | ||
599 | if (PageNosaveFree(page)) | ||
600 | continue; | ||
601 | save = kmalloc(sizeof(struct highmem_page), GFP_ATOMIC); | ||
602 | if (!save) | ||
603 | return -ENOMEM; | ||
604 | save->next = highmem_copy; | ||
605 | save->page = page; | ||
606 | save->data = (void *) get_zeroed_page(GFP_ATOMIC); | ||
607 | if (!save->data) { | ||
608 | kfree(save); | ||
609 | return -ENOMEM; | ||
610 | } | ||
611 | kaddr = kmap_atomic(page, KM_USER0); | ||
612 | memcpy(save->data, kaddr, PAGE_SIZE); | ||
613 | kunmap_atomic(kaddr, KM_USER0); | ||
614 | highmem_copy = save; | ||
615 | } | ||
616 | return 0; | ||
617 | } | ||
618 | #endif /* CONFIG_HIGHMEM */ | ||
619 | |||
620 | |||
621 | static int save_highmem(void) | ||
622 | { | ||
623 | #ifdef CONFIG_HIGHMEM | ||
624 | struct zone *zone; | ||
625 | int res = 0; | ||
626 | |||
627 | pr_debug("swsusp: Saving Highmem\n"); | ||
628 | for_each_zone (zone) { | ||
629 | if (is_highmem(zone)) | ||
630 | res = save_highmem_zone(zone); | ||
631 | if (res) | ||
632 | return res; | ||
633 | } | ||
634 | #endif | ||
635 | return 0; | ||
636 | } | ||
637 | |||
638 | static int restore_highmem(void) | ||
639 | { | ||
640 | #ifdef CONFIG_HIGHMEM | ||
641 | printk("swsusp: Restoring Highmem\n"); | ||
642 | while (highmem_copy) { | ||
643 | struct highmem_page *save = highmem_copy; | ||
644 | void *kaddr; | ||
645 | highmem_copy = save->next; | ||
646 | |||
647 | kaddr = kmap_atomic(save->page, KM_USER0); | ||
648 | memcpy(kaddr, save->data, PAGE_SIZE); | ||
649 | kunmap_atomic(kaddr, KM_USER0); | ||
650 | free_page((long) save->data); | ||
651 | kfree(save); | ||
652 | } | ||
653 | #endif | ||
654 | return 0; | ||
655 | } | ||
656 | |||
657 | |||
658 | static int pfn_is_nosave(unsigned long pfn) | ||
659 | { | ||
660 | unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT; | ||
661 | unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT; | ||
662 | return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn); | ||
663 | } | ||
664 | |||
665 | /** | ||
666 | * saveable - Determine whether a page should be cloned or not. | ||
667 | * @pfn: The page | ||
668 | * | ||
669 | * We save a page if it's Reserved, and not in the range of pages | ||
670 | * statically defined as 'unsaveable', or if it isn't reserved, and | ||
671 | * isn't part of a free chunk of pages. | ||
672 | */ | ||
673 | |||
674 | static int saveable(struct zone * zone, unsigned long * zone_pfn) | ||
675 | { | ||
676 | unsigned long pfn = *zone_pfn + zone->zone_start_pfn; | ||
677 | struct page * page; | ||
678 | |||
679 | if (!pfn_valid(pfn)) | ||
680 | return 0; | ||
681 | |||
682 | page = pfn_to_page(pfn); | ||
683 | if (PageNosave(page)) | ||
684 | return 0; | ||
685 | if (pfn_is_nosave(pfn)) { | ||
686 | pr_debug("[nosave pfn 0x%lx]", pfn); | ||
687 | return 0; | ||
688 | } | ||
689 | if (PageNosaveFree(page)) | ||
690 | return 0; | ||
691 | |||
692 | return 1; | ||
693 | } | ||
694 | |||
695 | static void count_data_pages(void) | ||
696 | { | ||
697 | struct zone *zone; | ||
698 | unsigned long zone_pfn; | ||
699 | |||
700 | nr_copy_pages = 0; | ||
701 | |||
702 | for_each_zone (zone) { | ||
703 | if (is_highmem(zone)) | ||
704 | continue; | ||
705 | mark_free_pages(zone); | ||
706 | for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) | ||
707 | nr_copy_pages += saveable(zone, &zone_pfn); | ||
708 | } | ||
709 | } | ||
710 | |||
711 | static void copy_data_pages(void) | ||
712 | { | ||
713 | struct zone *zone; | ||
714 | unsigned long zone_pfn; | ||
715 | struct pbe *pbe = pagedir_nosave, *p; | ||
716 | |||
717 | pr_debug("copy_data_pages(): pages to copy: %d\n", nr_copy_pages); | ||
718 | for_each_zone (zone) { | ||
719 | if (is_highmem(zone)) | ||
720 | continue; | ||
721 | mark_free_pages(zone); | ||
722 | /* This is necessary for swsusp_free() */ | ||
723 | for_each_pb_page (p, pagedir_nosave) | ||
724 | SetPageNosaveFree(virt_to_page(p)); | ||
725 | for_each_pbe(p, pagedir_nosave) | ||
726 | SetPageNosaveFree(virt_to_page(p->address)); | ||
727 | for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) { | ||
728 | if (saveable(zone, &zone_pfn)) { | ||
729 | struct page * page; | ||
730 | page = pfn_to_page(zone_pfn + zone->zone_start_pfn); | ||
731 | BUG_ON(!pbe); | ||
732 | pbe->orig_address = (unsigned long)page_address(page); | ||
733 | /* copy_page is not usable for copying task structs. */ | ||
734 | memcpy((void *)pbe->address, (void *)pbe->orig_address, PAGE_SIZE); | ||
735 | pbe = pbe->next; | ||
736 | } | ||
737 | } | ||
738 | } | ||
739 | BUG_ON(pbe); | ||
740 | } | ||
741 | |||
742 | |||
743 | /** | ||
744 | * free_pagedir - free pages allocated with alloc_pagedir() | ||
745 | */ | ||
746 | |||
747 | static inline void free_pagedir(struct pbe *pblist) | ||
748 | { | ||
749 | struct pbe *pbe; | ||
750 | |||
751 | while (pblist) { | ||
752 | pbe = (pblist + PB_PAGE_SKIP)->next; | ||
753 | ClearPageNosave(virt_to_page(pblist)); | ||
754 | ClearPageNosaveFree(virt_to_page(pblist)); | ||
755 | free_page((unsigned long)pblist); | ||
756 | pblist = pbe; | ||
757 | } | ||
758 | } | ||
759 | |||
760 | /** | ||
761 | * fill_pb_page - Create a list of PBEs on a given memory page | ||
762 | */ | ||
763 | |||
764 | static inline void fill_pb_page(struct pbe *pbpage) | ||
765 | { | ||
766 | struct pbe *p; | ||
767 | |||
768 | p = pbpage; | ||
769 | pbpage += PB_PAGE_SKIP; | ||
770 | do | ||
771 | p->next = p + 1; | ||
772 | while (++p < pbpage); | ||
773 | } | ||
774 | |||
775 | /** | ||
776 | * create_pbe_list - Create a list of PBEs on top of a given chain | ||
777 | * of memory pages allocated with alloc_pagedir() | ||
778 | */ | ||
779 | |||
780 | static void create_pbe_list(struct pbe *pblist, unsigned nr_pages) | ||
781 | { | ||
782 | struct pbe *pbpage, *p; | ||
783 | unsigned num = PBES_PER_PAGE; | ||
784 | |||
785 | for_each_pb_page (pbpage, pblist) { | ||
786 | if (num >= nr_pages) | ||
787 | break; | ||
788 | |||
789 | fill_pb_page(pbpage); | ||
790 | num += PBES_PER_PAGE; | ||
791 | } | ||
792 | if (pbpage) { | ||
793 | for (num -= PBES_PER_PAGE - 1, p = pbpage; num < nr_pages; p++, num++) | ||
794 | p->next = p + 1; | ||
795 | p->next = NULL; | ||
796 | } | ||
797 | pr_debug("create_pbe_list(): initialized %d PBEs\n", num); | ||
798 | } | ||
799 | |||
800 | static void *alloc_image_page(void) | ||
801 | { | ||
802 | void *res = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_COLD); | ||
803 | if (res) { | ||
804 | SetPageNosave(virt_to_page(res)); | ||
805 | SetPageNosaveFree(virt_to_page(res)); | ||
806 | } | ||
807 | return res; | ||
808 | } | ||
809 | |||
810 | /** | ||
811 | * alloc_pagedir - Allocate the page directory. | ||
812 | * | ||
813 | * First, determine exactly how many pages we need and | ||
814 | * allocate them. | ||
815 | * | ||
816 | * We arrange the pages in a chain: each page is an array of PBES_PER_PAGE | ||
817 | * struct pbe elements (pbes) and the last element in the page points | ||
818 | * to the next page. | ||
819 | * | ||
820 | * On each page we set up a list of struct_pbe elements. | ||
821 | */ | ||
822 | |||
823 | static struct pbe * alloc_pagedir(unsigned nr_pages) | ||
824 | { | ||
825 | unsigned num; | ||
826 | struct pbe *pblist, *pbe; | ||
827 | |||
828 | if (!nr_pages) | ||
829 | return NULL; | ||
830 | |||
831 | pr_debug("alloc_pagedir(): nr_pages = %d\n", nr_pages); | ||
832 | pblist = (struct pbe *)alloc_image_page(); | ||
833 | for (pbe = pblist, num = PBES_PER_PAGE; pbe && num < nr_pages; | ||
834 | pbe = pbe->next, num += PBES_PER_PAGE) { | ||
835 | pbe += PB_PAGE_SKIP; | ||
836 | pbe->next = (struct pbe *)alloc_image_page(); | ||
837 | } | ||
838 | if (!pbe) { /* get_zeroed_page() failed */ | ||
839 | free_pagedir(pblist); | ||
840 | pblist = NULL; | ||
841 | } | ||
842 | return pblist; | ||
843 | } | ||
844 | |||
845 | /** | ||
846 | * Free pages we allocated for suspend. Suspend pages are alocated | ||
847 | * before atomic copy, so we need to free them after resume. | ||
848 | */ | ||
849 | |||
850 | void swsusp_free(void) | ||
851 | { | ||
852 | struct zone *zone; | ||
853 | unsigned long zone_pfn; | ||
854 | |||
855 | for_each_zone(zone) { | ||
856 | for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) | ||
857 | if (pfn_valid(zone_pfn + zone->zone_start_pfn)) { | ||
858 | struct page * page; | ||
859 | page = pfn_to_page(zone_pfn + zone->zone_start_pfn); | ||
860 | if (PageNosave(page) && PageNosaveFree(page)) { | ||
861 | ClearPageNosave(page); | ||
862 | ClearPageNosaveFree(page); | ||
863 | free_page((long) page_address(page)); | ||
864 | } | ||
865 | } | ||
866 | } | ||
867 | } | ||
868 | |||
869 | /** | ||
870 | * enough_free_mem - Make sure we enough free memory to snapshot. | ||
871 | * | ||
872 | * Returns TRUE or FALSE after checking the number of available | ||
873 | * free pages. | ||
874 | */ | ||
875 | |||
876 | static int enough_free_mem(void) | ||
877 | { | ||
878 | pr_debug("swsusp: available memory: %u pages\n", nr_free_pages()); | ||
879 | return nr_free_pages() > (nr_copy_pages + PAGES_FOR_IO + | ||
880 | nr_copy_pages/PBES_PER_PAGE + !!(nr_copy_pages%PBES_PER_PAGE)); | ||
881 | } | ||
882 | |||
883 | |||
884 | /** | 543 | /** |
885 | * enough_swap - Make sure we have enough swap to save the image. | 544 | * enough_swap - Make sure we have enough swap to save the image. |
886 | * | 545 | * |
@@ -891,7 +550,7 @@ static int enough_free_mem(void) | |||
891 | * We should only consider resume_device. | 550 | * We should only consider resume_device. |
892 | */ | 551 | */ |
893 | 552 | ||
894 | static int enough_swap(void) | 553 | int enough_swap(void) |
895 | { | 554 | { |
896 | struct sysinfo i; | 555 | struct sysinfo i; |
897 | 556 | ||
@@ -901,88 +560,6 @@ static int enough_swap(void) | |||
901 | nr_copy_pages/PBES_PER_PAGE + !!(nr_copy_pages%PBES_PER_PAGE)); | 560 | nr_copy_pages/PBES_PER_PAGE + !!(nr_copy_pages%PBES_PER_PAGE)); |
902 | } | 561 | } |
903 | 562 | ||
904 | static int swsusp_alloc(void) | ||
905 | { | ||
906 | struct pbe * p; | ||
907 | |||
908 | pagedir_nosave = NULL; | ||
909 | |||
910 | if (!(pagedir_save = alloc_pagedir(nr_copy_pages))) { | ||
911 | printk(KERN_ERR "suspend: Allocating pagedir failed.\n"); | ||
912 | return -ENOMEM; | ||
913 | } | ||
914 | create_pbe_list(pagedir_save, nr_copy_pages); | ||
915 | pagedir_nosave = pagedir_save; | ||
916 | |||
917 | for_each_pbe (p, pagedir_save) { | ||
918 | p->address = (unsigned long)alloc_image_page(); | ||
919 | if (!p->address) { | ||
920 | printk(KERN_ERR "suspend: Allocating image pages failed.\n"); | ||
921 | swsusp_free(); | ||
922 | return -ENOMEM; | ||
923 | } | ||
924 | } | ||
925 | |||
926 | return 0; | ||
927 | } | ||
928 | |||
929 | static int suspend_prepare_image(void) | ||
930 | { | ||
931 | int error; | ||
932 | |||
933 | pr_debug("swsusp: critical section: \n"); | ||
934 | if (save_highmem()) { | ||
935 | printk(KERN_CRIT "swsusp: Not enough free pages for highmem\n"); | ||
936 | restore_highmem(); | ||
937 | return -ENOMEM; | ||
938 | } | ||
939 | |||
940 | drain_local_pages(); | ||
941 | count_data_pages(); | ||
942 | printk("swsusp: Need to copy %u pages\n", nr_copy_pages); | ||
943 | nr_copy_pages_check = nr_copy_pages; | ||
944 | |||
945 | pr_debug("swsusp: pages needed: %u + %lu + %u, free: %u\n", | ||
946 | nr_copy_pages, | ||
947 | nr_copy_pages/PBES_PER_PAGE + !!(nr_copy_pages%PBES_PER_PAGE), | ||
948 | PAGES_FOR_IO, nr_free_pages()); | ||
949 | |||
950 | if (!enough_free_mem()) { | ||
951 | printk(KERN_ERR "swsusp: Not enough free memory\n"); | ||
952 | return -ENOMEM; | ||
953 | } | ||
954 | |||
955 | if (MAX_PBES < nr_copy_pages / PBES_PER_PAGE + | ||
956 | !!(nr_copy_pages % PBES_PER_PAGE)) { | ||
957 | printk(KERN_ERR "swsusp: Too many image pages\n"); | ||
958 | return -ENOSPC; | ||
959 | } | ||
960 | |||
961 | if (!enough_swap()) { | ||
962 | printk(KERN_ERR "swsusp: Not enough free swap\n"); | ||
963 | return -ENOSPC; | ||
964 | } | ||
965 | |||
966 | error = swsusp_alloc(); | ||
967 | if (error) | ||
968 | return error; | ||
969 | |||
970 | /* During allocating of suspend pagedir, new cold pages may appear. | ||
971 | * Kill them. | ||
972 | */ | ||
973 | drain_local_pages(); | ||
974 | copy_data_pages(); | ||
975 | |||
976 | /* | ||
977 | * End of critical section. From now on, we can write to memory, | ||
978 | * but we should not touch disk. This specially means we must _not_ | ||
979 | * touch swap space! Except we must write out our image of course. | ||
980 | */ | ||
981 | |||
982 | printk("swsusp: critical section/: done (%d pages copied)\n", nr_copy_pages ); | ||
983 | return 0; | ||
984 | } | ||
985 | |||
986 | 563 | ||
987 | /* It is important _NOT_ to umount filesystems at this point. We want | 564 | /* It is important _NOT_ to umount filesystems at this point. We want |
988 | * them synced (in case something goes wrong) but we DO not want to mark | 565 | * them synced (in case something goes wrong) but we DO not want to mark |
@@ -1002,14 +579,6 @@ int swsusp_write(void) | |||
1002 | } | 579 | } |
1003 | 580 | ||
1004 | 581 | ||
1005 | extern asmlinkage int swsusp_arch_suspend(void); | ||
1006 | extern asmlinkage int swsusp_arch_resume(void); | ||
1007 | |||
1008 | |||
1009 | asmlinkage int swsusp_save(void) | ||
1010 | { | ||
1011 | return suspend_prepare_image(); | ||
1012 | } | ||
1013 | 582 | ||
1014 | int swsusp_suspend(void) | 583 | int swsusp_suspend(void) |
1015 | { | 584 | { |
@@ -1041,7 +610,6 @@ int swsusp_suspend(void) | |||
1041 | printk(KERN_ERR "Error %d suspending\n", error); | 610 | printk(KERN_ERR "Error %d suspending\n", error); |
1042 | /* Restore control flow magically appears here */ | 611 | /* Restore control flow magically appears here */ |
1043 | restore_processor_state(); | 612 | restore_processor_state(); |
1044 | BUG_ON (nr_copy_pages_check != nr_copy_pages); | ||
1045 | restore_highmem(); | 613 | restore_highmem(); |
1046 | device_power_up(); | 614 | device_power_up(); |
1047 | local_irq_enable(); | 615 | local_irq_enable(); |