[PATCH] swsusp: rework swsusp_suspend
[linux-2.6] / kernel / power / snapshot.c
1 /*
2  * linux/kernel/power/snapshot.c
3  *
4  * This file provide system snapshot/restore functionality.
5  *
6  * Copyright (C) 1998-2005 Pavel Machek <pavel@suse.cz>
7  *
8  * This file is released under the GPLv2, and is based on swsusp.c.
9  *
10  */
11
12
13 #include <linux/module.h>
14 #include <linux/mm.h>
15 #include <linux/suspend.h>
16 #include <linux/smp_lock.h>
17 #include <linux/delay.h>
18 #include <linux/bitops.h>
19 #include <linux/spinlock.h>
20 #include <linux/kernel.h>
21 #include <linux/pm.h>
22 #include <linux/device.h>
23 #include <linux/bootmem.h>
24 #include <linux/syscalls.h>
25 #include <linux/console.h>
26 #include <linux/highmem.h>
27
28 #include <asm/uaccess.h>
29 #include <asm/mmu_context.h>
30 #include <asm/pgtable.h>
31 #include <asm/tlbflush.h>
32 #include <asm/io.h>
33
34 #include "power.h"
35
36 #ifdef CONFIG_HIGHMEM
37 struct highmem_page {
38         char *data;
39         struct page *page;
40         struct highmem_page *next;
41 };
42
43 static struct highmem_page *highmem_copy;
44
45 static int save_highmem_zone(struct zone *zone)
46 {
47         unsigned long zone_pfn;
48         mark_free_pages(zone);
49         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
50                 struct page *page;
51                 struct highmem_page *save;
52                 void *kaddr;
53                 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
54
55                 if (!(pfn%1000))
56                         printk(".");
57                 if (!pfn_valid(pfn))
58                         continue;
59                 page = pfn_to_page(pfn);
60                 /*
61                  * This condition results from rvmalloc() sans vmalloc_32()
62                  * and architectural memory reservations. This should be
63                  * corrected eventually when the cases giving rise to this
64                  * are better understood.
65                  */
66                 if (PageReserved(page)) {
67                         printk("highmem reserved page?!\n");
68                         continue;
69                 }
70                 BUG_ON(PageNosave(page));
71                 if (PageNosaveFree(page))
72                         continue;
73                 save = kmalloc(sizeof(struct highmem_page), GFP_ATOMIC);
74                 if (!save)
75                         return -ENOMEM;
76                 save->next = highmem_copy;
77                 save->page = page;
78                 save->data = (void *) get_zeroed_page(GFP_ATOMIC);
79                 if (!save->data) {
80                         kfree(save);
81                         return -ENOMEM;
82                 }
83                 kaddr = kmap_atomic(page, KM_USER0);
84                 memcpy(save->data, kaddr, PAGE_SIZE);
85                 kunmap_atomic(kaddr, KM_USER0);
86                 highmem_copy = save;
87         }
88         return 0;
89 }
90
91 int save_highmem(void)
92 {
93         struct zone *zone;
94         int res = 0;
95
96         pr_debug("swsusp: Saving Highmem\n");
97         for_each_zone (zone) {
98                 if (is_highmem(zone))
99                         res = save_highmem_zone(zone);
100                 if (res)
101                         return res;
102         }
103         return 0;
104 }
105
106 int restore_highmem(void)
107 {
108         printk("swsusp: Restoring Highmem\n");
109         while (highmem_copy) {
110                 struct highmem_page *save = highmem_copy;
111                 void *kaddr;
112                 highmem_copy = save->next;
113
114                 kaddr = kmap_atomic(save->page, KM_USER0);
115                 memcpy(kaddr, save->data, PAGE_SIZE);
116                 kunmap_atomic(kaddr, KM_USER0);
117                 free_page((long) save->data);
118                 kfree(save);
119         }
120         return 0;
121 }
122 #endif
123
124 static int pfn_is_nosave(unsigned long pfn)
125 {
126         unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
127         unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
128         return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
129 }
130
131 /**
132  *      saveable - Determine whether a page should be cloned or not.
133  *      @pfn:   The page
134  *
135  *      We save a page if it's Reserved, and not in the range of pages
136  *      statically defined as 'unsaveable', or if it isn't reserved, and
137  *      isn't part of a free chunk of pages.
138  */
139
140 static int saveable(struct zone *zone, unsigned long *zone_pfn)
141 {
142         unsigned long pfn = *zone_pfn + zone->zone_start_pfn;
143         struct page *page;
144
145         if (!pfn_valid(pfn))
146                 return 0;
147
148         page = pfn_to_page(pfn);
149         BUG_ON(PageReserved(page) && PageNosave(page));
150         if (PageNosave(page))
151                 return 0;
152         if (PageReserved(page) && pfn_is_nosave(pfn)) {
153                 pr_debug("[nosave pfn 0x%lx]", pfn);
154                 return 0;
155         }
156         if (PageNosaveFree(page))
157                 return 0;
158
159         return 1;
160 }
161
162 static unsigned count_data_pages(void)
163 {
164         struct zone *zone;
165         unsigned long zone_pfn;
166         unsigned int n = 0;
167
168         for_each_zone (zone) {
169                 if (is_highmem(zone))
170                         continue;
171                 mark_free_pages(zone);
172                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
173                         n += saveable(zone, &zone_pfn);
174         }
175         return n;
176 }
177
178 static void copy_data_pages(struct pbe *pblist)
179 {
180         struct zone *zone;
181         unsigned long zone_pfn;
182         struct pbe *pbe, *p;
183
184         pbe = pblist;
185         for_each_zone (zone) {
186                 if (is_highmem(zone))
187                         continue;
188                 mark_free_pages(zone);
189                 /* This is necessary for swsusp_free() */
190                 for_each_pb_page (p, pblist)
191                         SetPageNosaveFree(virt_to_page(p));
192                 for_each_pbe (p, pblist)
193                         SetPageNosaveFree(virt_to_page(p->address));
194                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
195                         if (saveable(zone, &zone_pfn)) {
196                                 struct page *page;
197                                 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
198                                 BUG_ON(!pbe);
199                                 pbe->orig_address = (unsigned long)page_address(page);
200                                 /* copy_page is not usable for copying task structs. */
201                                 memcpy((void *)pbe->address, (void *)pbe->orig_address, PAGE_SIZE);
202                                 pbe = pbe->next;
203                         }
204                 }
205         }
206         BUG_ON(pbe);
207 }
208
209
210 /**
211  *      free_pagedir - free pages allocated with alloc_pagedir()
212  */
213
214 void free_pagedir(struct pbe *pblist)
215 {
216         struct pbe *pbe;
217
218         while (pblist) {
219                 pbe = (pblist + PB_PAGE_SKIP)->next;
220                 ClearPageNosave(virt_to_page(pblist));
221                 ClearPageNosaveFree(virt_to_page(pblist));
222                 free_page((unsigned long)pblist);
223                 pblist = pbe;
224         }
225 }
226
227 /**
228  *      fill_pb_page - Create a list of PBEs on a given memory page
229  */
230
231 static inline void fill_pb_page(struct pbe *pbpage)
232 {
233         struct pbe *p;
234
235         p = pbpage;
236         pbpage += PB_PAGE_SKIP;
237         do
238                 p->next = p + 1;
239         while (++p < pbpage);
240 }
241
242 /**
243  *      create_pbe_list - Create a list of PBEs on top of a given chain
244  *      of memory pages allocated with alloc_pagedir()
245  */
246
247 void create_pbe_list(struct pbe *pblist, unsigned int nr_pages)
248 {
249         struct pbe *pbpage, *p;
250         unsigned int num = PBES_PER_PAGE;
251
252         for_each_pb_page (pbpage, pblist) {
253                 if (num >= nr_pages)
254                         break;
255
256                 fill_pb_page(pbpage);
257                 num += PBES_PER_PAGE;
258         }
259         if (pbpage) {
260                 for (num -= PBES_PER_PAGE - 1, p = pbpage; num < nr_pages; p++, num++)
261                         p->next = p + 1;
262                 p->next = NULL;
263         }
264         pr_debug("create_pbe_list(): initialized %d PBEs\n", num);
265 }
266
267 /**
268  *      @safe_needed - on resume, for storing the PBE list and the image,
269  *      we can only use memory pages that do not conflict with the pages
270  *      which had been used before suspend.
271  *
272  *      The unsafe pages are marked with the PG_nosave_free flag
273  *
274  *      Allocated but unusable (ie eaten) memory pages should be marked
275  *      so that swsusp_free() can release them
276  */
277
278 static inline void *alloc_image_page(gfp_t gfp_mask, int safe_needed)
279 {
280         void *res;
281
282         if (safe_needed)
283                 do {
284                         res = (void *)get_zeroed_page(gfp_mask);
285                         if (res && PageNosaveFree(virt_to_page(res)))
286                                 /* This is for swsusp_free() */
287                                 SetPageNosave(virt_to_page(res));
288                 } while (res && PageNosaveFree(virt_to_page(res)));
289         else
290                 res = (void *)get_zeroed_page(gfp_mask);
291         if (res) {
292                 SetPageNosave(virt_to_page(res));
293                 SetPageNosaveFree(virt_to_page(res));
294         }
295         return res;
296 }
297
298 unsigned long get_safe_page(gfp_t gfp_mask)
299 {
300         return (unsigned long)alloc_image_page(gfp_mask, 1);
301 }
302
303 /**
304  *      alloc_pagedir - Allocate the page directory.
305  *
306  *      First, determine exactly how many pages we need and
307  *      allocate them.
308  *
309  *      We arrange the pages in a chain: each page is an array of PBES_PER_PAGE
310  *      struct pbe elements (pbes) and the last element in the page points
311  *      to the next page.
312  *
313  *      On each page we set up a list of struct_pbe elements.
314  */
315
316 struct pbe *alloc_pagedir(unsigned int nr_pages, gfp_t gfp_mask, int safe_needed)
317 {
318         unsigned int num;
319         struct pbe *pblist, *pbe;
320
321         if (!nr_pages)
322                 return NULL;
323
324         pr_debug("alloc_pagedir(): nr_pages = %d\n", nr_pages);
325         pblist = alloc_image_page(gfp_mask, safe_needed);
326         /* FIXME: rewrite this ugly loop */
327         for (pbe = pblist, num = PBES_PER_PAGE; pbe && num < nr_pages;
328                         pbe = pbe->next, num += PBES_PER_PAGE) {
329                 pbe += PB_PAGE_SKIP;
330                 pbe->next = alloc_image_page(gfp_mask, safe_needed);
331         }
332         if (!pbe) { /* get_zeroed_page() failed */
333                 free_pagedir(pblist);
334                 pblist = NULL;
335         }
336         return pblist;
337 }
338
339 /**
340  * Free pages we allocated for suspend. Suspend pages are alocated
341  * before atomic copy, so we need to free them after resume.
342  */
343
344 void swsusp_free(void)
345 {
346         struct zone *zone;
347         unsigned long zone_pfn;
348
349         for_each_zone(zone) {
350                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
351                         if (pfn_valid(zone_pfn + zone->zone_start_pfn)) {
352                                 struct page *page;
353                                 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
354                                 if (PageNosave(page) && PageNosaveFree(page)) {
355                                         ClearPageNosave(page);
356                                         ClearPageNosaveFree(page);
357                                         free_page((long) page_address(page));
358                                 }
359                         }
360         }
361 }
362
363
364 /**
365  *      enough_free_mem - Make sure we enough free memory to snapshot.
366  *
367  *      Returns TRUE or FALSE after checking the number of available
368  *      free pages.
369  */
370
371 static int enough_free_mem(unsigned int nr_pages)
372 {
373         pr_debug("swsusp: available memory: %u pages\n", nr_free_pages());
374         return nr_free_pages() > (nr_pages + PAGES_FOR_IO +
375                 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
376 }
377
378 int alloc_data_pages(struct pbe *pblist, gfp_t gfp_mask, int safe_needed)
379 {
380         struct pbe *p;
381
382         for_each_pbe (p, pblist) {
383                 p->address = (unsigned long)alloc_image_page(gfp_mask, safe_needed);
384                 if (!p->address)
385                         return -ENOMEM;
386         }
387         return 0;
388 }
389
390 static struct pbe *swsusp_alloc(unsigned int nr_pages)
391 {
392         struct pbe *pblist;
393
394         if (!(pblist = alloc_pagedir(nr_pages, GFP_ATOMIC | __GFP_COLD, 0))) {
395                 printk(KERN_ERR "suspend: Allocating pagedir failed.\n");
396                 return NULL;
397         }
398         create_pbe_list(pblist, nr_pages);
399
400         if (alloc_data_pages(pblist, GFP_ATOMIC | __GFP_COLD, 0)) {
401                 printk(KERN_ERR "suspend: Allocating image pages failed.\n");
402                 swsusp_free();
403                 return NULL;
404         }
405
406         return pblist;
407 }
408
409 asmlinkage int swsusp_save(void)
410 {
411         unsigned int nr_pages;
412
413         pr_debug("swsusp: critical section: \n");
414
415         drain_local_pages();
416         nr_pages = count_data_pages();
417         printk("swsusp: Need to copy %u pages\n", nr_pages);
418
419         pr_debug("swsusp: pages needed: %u + %lu + %u, free: %u\n",
420                  nr_pages,
421                  (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE,
422                  PAGES_FOR_IO, nr_free_pages());
423
424         /* This is needed because of the fixed size of swsusp_info */
425         if (MAX_PBES < (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE)
426                 return -ENOSPC;
427
428         if (!enough_free_mem(nr_pages)) {
429                 printk(KERN_ERR "swsusp: Not enough free memory\n");
430                 return -ENOMEM;
431         }
432
433         pagedir_nosave = swsusp_alloc(nr_pages);
434         if (!pagedir_nosave)
435                 return -ENOMEM;
436
437         /* During allocating of suspend pagedir, new cold pages may appear.
438          * Kill them.
439          */
440         drain_local_pages();
441         copy_data_pages(pagedir_nosave);
442
443         /*
444          * End of critical section. From now on, we can write to memory,
445          * but we should not touch disk. This specially means we must _not_
446          * touch swap space! Except we must write out our image of course.
447          */
448
449         nr_copy_pages = nr_pages;
450
451         printk("swsusp: critical section/: done (%d pages copied)\n", nr_pages);
452         return 0;
453 }