2 * linux/kernel/power/swsusp.c
4 * This file is to realize architecture-independent
5 * machine suspend feature using pretty near only high-level routines
7 * Copyright (C) 1998-2001 Gabor Kuti <seasons@fornax.hu>
8 * Copyright (C) 1998,2001-2004 Pavel Machek <pavel@suse.cz>
10 * This file is released under the GPLv2.
12 * I'd like to thank the following people for their work:
14 * Pavel Machek <pavel@ucw.cz>:
15 * Modifications, defectiveness pointing, being with me at the very beginning,
16 * suspend to swap space, stop all tasks. Port to 2.4.18-ac and 2.5.17.
18 * Steve Doddi <dirk@loth.demon.co.uk>:
19 * Support the possibility of hardware state restoring.
21 * Raph <grey.havens@earthling.net>:
22 * Support for preserving states of network devices and virtual console
23 * (including X and svgatextmode)
25 * Kurt Garloff <garloff@suse.de>:
26 * Straightened the critical function in order to prevent compilers from
27 * playing tricks with local variables.
29 * Andreas Mohr <a.mohr@mailto.de>
31 * Alex Badea <vampire@go.ro>:
34 * Andreas Steinmetz <ast@domdv.de>:
35 * Added encrypted suspend option
37 * More state savers are welcome. Especially for the scsi layer...
39 * For TODOs,FIXMEs also look in Documentation/power/swsusp.txt
42 #include <linux/module.h>
44 #include <linux/suspend.h>
45 #include <linux/smp_lock.h>
46 #include <linux/file.h>
47 #include <linux/utsname.h>
48 #include <linux/version.h>
49 #include <linux/delay.h>
50 #include <linux/reboot.h>
51 #include <linux/bitops.h>
52 #include <linux/vt_kern.h>
53 #include <linux/kbd_kern.h>
54 #include <linux/keyboard.h>
55 #include <linux/spinlock.h>
56 #include <linux/genhd.h>
57 #include <linux/kernel.h>
58 #include <linux/major.h>
59 #include <linux/swap.h>
61 #include <linux/device.h>
62 #include <linux/buffer_head.h>
63 #include <linux/swapops.h>
64 #include <linux/bootmem.h>
65 #include <linux/syscalls.h>
66 #include <linux/console.h>
67 #include <linux/highmem.h>
68 #include <linux/bio.h>
69 #include <linux/mount.h>
71 #include <asm/uaccess.h>
72 #include <asm/mmu_context.h>
73 #include <asm/pgtable.h>
74 #include <asm/tlbflush.h>
77 #include <linux/random.h>
78 #include <linux/crypto.h>
79 #include <asm/scatterlist.h>
87 /* References to section boundaries */
88 extern const void __nosave_begin, __nosave_end;
90 /* Variables to be preserved over suspend */
91 static int nr_copy_pages_check;
93 extern char resume_file[];
95 /* Local variables that should not be affected by save */
96 static unsigned int nr_copy_pages __nosavedata = 0;
98 /* Suspend pagedir is allocated before final copy, therefore it
99 must be freed after resume
101 Warning: this is evil. There are actually two pagedirs at time of
102 resume. One is "pagedir_save", which is empty frame allocated at
103 time of suspend, that must be freed. Second is "pagedir_nosave",
104 allocated at time of resume, that travels through memory not to
105 collide with anything.
107 Warning: this is even more evil than it seems. Pagedirs this file
108 talks about are completely different from page directories used by
111 suspend_pagedir_t *pagedir_nosave __nosavedata = NULL;
112 static suspend_pagedir_t *pagedir_save;
114 #define SWSUSP_SIG "S1SUSPEND"
116 static struct swsusp_header {
117 char reserved[PAGE_SIZE - 20 - MAXKEY - MAXIV - sizeof(swp_entry_t)];
118 u8 key_iv[MAXKEY+MAXIV];
119 swp_entry_t swsusp_info;
122 } __attribute__((packed, aligned(PAGE_SIZE))) swsusp_header;
124 static struct swsusp_info swsusp_info;
127 * XXX: We try to keep some more pages free so that I/O operations succeed
128 * without paging. Might this be more?
130 #define PAGES_FOR_IO 512
136 /* We memorize in swapfile_used what swap devices are used for suspension */
137 #define SWAPFILE_UNUSED 0
138 #define SWAPFILE_SUSPEND 1 /* This is the suspending device */
139 #define SWAPFILE_IGNORED 2 /* Those are other swap devices ignored for suspension */
141 static unsigned short swapfile_used[MAX_SWAPFILES];
142 static unsigned short root_swap;
144 static int write_page(unsigned long addr, swp_entry_t * loc);
145 static int bio_read_page(pgoff_t page_off, void * page);
147 static u8 key_iv[MAXKEY+MAXIV];
149 #ifdef CONFIG_SWSUSP_ENCRYPT
151 static int crypto_init(int mode, void **mem)
156 struct crypto_tfm *tfm;
158 modemsg = mode ? "suspend not possible" : "resume not possible";
160 tfm = crypto_alloc_tfm(CIPHER, CRYPTO_TFM_MODE_CBC);
162 printk(KERN_ERR "swsusp: no tfm, %s\n", modemsg);
167 if(MAXKEY < crypto_tfm_alg_min_keysize(tfm)) {
168 printk(KERN_ERR "swsusp: key buffer too small, %s\n", modemsg);
174 get_random_bytes(key_iv, MAXKEY+MAXIV);
176 len = crypto_tfm_alg_max_keysize(tfm);
180 if (crypto_cipher_setkey(tfm, key_iv, len)) {
181 printk(KERN_ERR "swsusp: key setup failure, %s\n", modemsg);
182 error = -EKEYREJECTED;
186 len = crypto_tfm_alg_ivsize(tfm);
189 printk(KERN_ERR "swsusp: iv buffer too small, %s\n", modemsg);
194 crypto_cipher_set_iv(tfm, key_iv+MAXKEY, len);
200 fail: crypto_free_tfm(tfm);
204 static __inline__ void crypto_exit(void *mem)
206 crypto_free_tfm((struct crypto_tfm *)mem);
209 static __inline__ int crypto_write(struct pbe *p, void *mem)
212 struct scatterlist src, dst;
214 src.page = virt_to_page(p->address);
216 src.length = PAGE_SIZE;
217 dst.page = virt_to_page((void *)&swsusp_header);
219 dst.length = PAGE_SIZE;
221 error = crypto_cipher_encrypt((struct crypto_tfm *)mem, &dst, &src,
225 error = write_page((unsigned long)&swsusp_header,
230 static __inline__ int crypto_read(struct pbe *p, void *mem)
233 struct scatterlist src, dst;
235 error = bio_read_page(swp_offset(p->swap_address), (void *)p->address);
238 src.length = PAGE_SIZE;
240 dst.length = PAGE_SIZE;
241 src.page = dst.page = virt_to_page((void *)p->address);
243 error = crypto_cipher_decrypt((struct crypto_tfm *)mem, &dst,
249 static __inline__ int crypto_init(int mode, void *mem)
254 static __inline__ void crypto_exit(void *mem)
258 static __inline__ int crypto_write(struct pbe *p, void *mem)
260 return write_page(p->address, &(p->swap_address));
263 static __inline__ int crypto_read(struct pbe *p, void *mem)
265 return bio_read_page(swp_offset(p->swap_address), (void *)p->address);
269 static int mark_swapfiles(swp_entry_t prev)
273 rw_swap_page_sync(READ,
274 swp_entry(root_swap, 0),
275 virt_to_page((unsigned long)&swsusp_header));
276 if (!memcmp("SWAP-SPACE",swsusp_header.sig, 10) ||
277 !memcmp("SWAPSPACE2",swsusp_header.sig, 10)) {
278 memcpy(swsusp_header.orig_sig,swsusp_header.sig, 10);
279 memcpy(swsusp_header.sig,SWSUSP_SIG, 10);
280 memcpy(swsusp_header.key_iv, key_iv, MAXKEY+MAXIV);
281 swsusp_header.swsusp_info = prev;
282 error = rw_swap_page_sync(WRITE,
283 swp_entry(root_swap, 0),
284 virt_to_page((unsigned long)
287 pr_debug("swsusp: Partition is not swap space.\n");
294 * Check whether the swap device is the specified resume
295 * device, irrespective of whether they are specified by
298 * (Thus, device inode aliasing is allowed. You can say /dev/hda4
299 * instead of /dev/ide/host0/bus0/target0/lun0/part4 [if using devfs]
300 * and they'll be considered the same device. This is *necessary* for
301 * devfs, since the resume code can only recognize the form /dev/hda4,
302 * but the suspend code would see the long name.)
304 static int is_resume_device(const struct swap_info_struct *swap_info)
306 struct file *file = swap_info->swap_file;
307 struct inode *inode = file->f_dentry->d_inode;
309 return S_ISBLK(inode->i_mode) &&
310 swsusp_resume_device == MKDEV(imajor(inode), iminor(inode));
313 static int swsusp_swap_check(void) /* This is called before saving image */
317 len=strlen(resume_file);
320 spin_lock(&swap_lock);
321 for (i=0; i<MAX_SWAPFILES; i++) {
322 if (!(swap_info[i].flags & SWP_WRITEOK)) {
323 swapfile_used[i]=SWAPFILE_UNUSED;
326 printk(KERN_WARNING "resume= option should be used to set suspend device" );
327 if (root_swap == 0xFFFF) {
328 swapfile_used[i] = SWAPFILE_SUSPEND;
331 swapfile_used[i] = SWAPFILE_IGNORED;
333 /* we ignore all swap devices that are not the resume_file */
334 if (is_resume_device(&swap_info[i])) {
335 swapfile_used[i] = SWAPFILE_SUSPEND;
338 swapfile_used[i] = SWAPFILE_IGNORED;
343 spin_unlock(&swap_lock);
344 return (root_swap != 0xffff) ? 0 : -ENODEV;
348 * This is called after saving image so modification
349 * will be lost after resume... and that's what we want.
350 * we make the device unusable. A new call to
351 * lock_swapdevices can unlock the devices.
353 static void lock_swapdevices(void)
357 spin_lock(&swap_lock);
358 for (i = 0; i< MAX_SWAPFILES; i++)
359 if (swapfile_used[i] == SWAPFILE_IGNORED) {
360 swap_info[i].flags ^= SWP_WRITEOK;
362 spin_unlock(&swap_lock);
366 * write_page - Write one page to a fresh swap location.
367 * @addr: Address we're writing.
368 * @loc: Place to store the entry we used.
370 * Allocate a new swap entry and 'sync' it. Note we discard -EIO
371 * errors. That is an artifact left over from swsusp. It did not
372 * check the return of rw_swap_page_sync() at all, since most pages
373 * written back to swap would return -EIO.
374 * This is a partial improvement, since we will at least return other
375 * errors, though we need to eventually fix the damn code.
377 static int write_page(unsigned long addr, swp_entry_t * loc)
382 entry = get_swap_page();
383 if (swp_offset(entry) &&
384 swapfile_used[swp_type(entry)] == SWAPFILE_SUSPEND) {
385 error = rw_swap_page_sync(WRITE, entry,
397 * data_free - Free the swap entries used by the saved image.
399 * Walk the list of used swap entries and free each one.
400 * This is only used for cleanup when suspend fails.
402 static void data_free(void)
407 for_each_pbe(p, pagedir_nosave) {
408 entry = p->swap_address;
417 * data_write - Write saved image to swap.
419 * Walk the list of pages in the image and sync each one to swap.
421 static int data_write(void)
423 int error = 0, i = 0;
424 unsigned int mod = nr_copy_pages / 100;
428 if ((error = crypto_init(1, &tfm)))
434 printk( "Writing data to swap (%d pages)... ", nr_copy_pages );
435 for_each_pbe (p, pagedir_nosave) {
437 printk( "\b\b\b\b%3d%%", i / mod );
438 if ((error = crypto_write(p, tfm))) {
444 printk("\b\b\b\bdone\n");
449 static void dump_info(void)
451 pr_debug(" swsusp: Version: %u\n",swsusp_info.version_code);
452 pr_debug(" swsusp: Num Pages: %ld\n",swsusp_info.num_physpages);
453 pr_debug(" swsusp: UTS Sys: %s\n",swsusp_info.uts.sysname);
454 pr_debug(" swsusp: UTS Node: %s\n",swsusp_info.uts.nodename);
455 pr_debug(" swsusp: UTS Release: %s\n",swsusp_info.uts.release);
456 pr_debug(" swsusp: UTS Version: %s\n",swsusp_info.uts.version);
457 pr_debug(" swsusp: UTS Machine: %s\n",swsusp_info.uts.machine);
458 pr_debug(" swsusp: UTS Domain: %s\n",swsusp_info.uts.domainname);
459 pr_debug(" swsusp: CPUs: %d\n",swsusp_info.cpus);
460 pr_debug(" swsusp: Image: %ld Pages\n",swsusp_info.image_pages);
461 pr_debug(" swsusp: Pagedir: %ld Pages\n",swsusp_info.pagedir_pages);
464 static void init_header(void)
466 memset(&swsusp_info, 0, sizeof(swsusp_info));
467 swsusp_info.version_code = LINUX_VERSION_CODE;
468 swsusp_info.num_physpages = num_physpages;
469 memcpy(&swsusp_info.uts, &system_utsname, sizeof(system_utsname));
471 swsusp_info.suspend_pagedir = pagedir_nosave;
472 swsusp_info.cpus = num_online_cpus();
473 swsusp_info.image_pages = nr_copy_pages;
476 static int close_swap(void)
482 error = write_page((unsigned long)&swsusp_info, &entry);
485 error = mark_swapfiles(entry);
492 * free_pagedir_entries - Free pages used by the page directory.
494 * This is used during suspend for error recovery.
497 static void free_pagedir_entries(void)
501 for (i = 0; i < swsusp_info.pagedir_pages; i++)
502 swap_free(swsusp_info.pagedir[i]);
507 * write_pagedir - Write the array of pages holding the page directory.
508 * @last: Last swap entry we write (needed for header).
511 static int write_pagedir(void)
517 printk( "Writing pagedir...");
518 for_each_pb_page (pbe, pagedir_nosave) {
519 if ((error = write_page((unsigned long)pbe, &swsusp_info.pagedir[n++])))
523 swsusp_info.pagedir_pages = n;
524 printk("done (%u pages)\n", n);
529 * write_suspend_image - Write entire image and metadata.
532 static int write_suspend_image(void)
537 if ((error = data_write()))
540 if ((error = write_pagedir()))
543 if ((error = close_swap()))
546 memset(key_iv, 0, MAXKEY+MAXIV);
549 free_pagedir_entries();
556 #ifdef CONFIG_HIGHMEM
557 struct highmem_page {
560 struct highmem_page *next;
563 static struct highmem_page *highmem_copy;
565 static int save_highmem_zone(struct zone *zone)
567 unsigned long zone_pfn;
568 mark_free_pages(zone);
569 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
571 struct highmem_page *save;
573 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
579 page = pfn_to_page(pfn);
581 * PageReserved results from rvmalloc() sans vmalloc_32()
582 * and architectural memory reservations.
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
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
594 * XXX: must remove usage of PageReserved!
596 if (PageReserved(page))
598 BUG_ON(PageNosave(page));
599 if (PageNosaveFree(page))
601 save = kmalloc(sizeof(struct highmem_page), GFP_ATOMIC);
604 save->next = highmem_copy;
606 save->data = (void *) get_zeroed_page(GFP_ATOMIC);
611 kaddr = kmap_atomic(page, KM_USER0);
612 memcpy(save->data, kaddr, PAGE_SIZE);
613 kunmap_atomic(kaddr, KM_USER0);
618 #endif /* CONFIG_HIGHMEM */
621 static int save_highmem(void)
623 #ifdef CONFIG_HIGHMEM
627 pr_debug("swsusp: Saving Highmem\n");
628 for_each_zone (zone) {
629 if (is_highmem(zone))
630 res = save_highmem_zone(zone);
638 static int restore_highmem(void)
640 #ifdef CONFIG_HIGHMEM
641 printk("swsusp: Restoring Highmem\n");
642 while (highmem_copy) {
643 struct highmem_page *save = highmem_copy;
645 highmem_copy = save->next;
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);
658 static int pfn_is_nosave(unsigned long pfn)
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);
666 * saveable - Determine whether a page should be cloned or not.
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.
674 static int saveable(struct zone * zone, unsigned long * zone_pfn)
676 unsigned long pfn = *zone_pfn + zone->zone_start_pfn;
682 page = pfn_to_page(pfn);
683 if (PageNosave(page))
685 if (pfn_is_nosave(pfn)) {
686 pr_debug("[nosave pfn 0x%lx]", pfn);
689 if (PageNosaveFree(page))
695 static void count_data_pages(void)
698 unsigned long zone_pfn;
702 for_each_zone (zone) {
703 if (is_highmem(zone))
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);
712 static void copy_data_pages(void)
715 unsigned long zone_pfn;
716 struct pbe * pbe = pagedir_nosave;
718 pr_debug("copy_data_pages(): pages to copy: %d\n", nr_copy_pages);
719 for_each_zone (zone) {
720 if (is_highmem(zone))
722 mark_free_pages(zone);
723 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
724 if (saveable(zone, &zone_pfn)) {
726 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
728 pbe->orig_address = (long) page_address(page);
729 /* copy_page is not usable for copying task structs. */
730 memcpy((void *)pbe->address, (void *)pbe->orig_address, PAGE_SIZE);
740 * calc_nr - Determine the number of pages needed for a pbe list.
743 static int calc_nr(int nr_copy)
745 return nr_copy + (nr_copy+PBES_PER_PAGE-2)/(PBES_PER_PAGE-1);
749 * free_pagedir - free pages allocated with alloc_pagedir()
752 static inline void free_pagedir(struct pbe *pblist)
757 pbe = (pblist + PB_PAGE_SKIP)->next;
758 free_page((unsigned long)pblist);
764 * fill_pb_page - Create a list of PBEs on a given memory page
767 static inline void fill_pb_page(struct pbe *pbpage)
772 pbpage += PB_PAGE_SKIP;
775 while (++p < pbpage);
779 * create_pbe_list - Create a list of PBEs on top of a given chain
780 * of memory pages allocated with alloc_pagedir()
783 static void create_pbe_list(struct pbe *pblist, unsigned nr_pages)
785 struct pbe *pbpage, *p;
786 unsigned num = PBES_PER_PAGE;
788 for_each_pb_page (pbpage, pblist) {
792 fill_pb_page(pbpage);
793 num += PBES_PER_PAGE;
796 for (num -= PBES_PER_PAGE - 1, p = pbpage; num < nr_pages; p++, num++)
800 pr_debug("create_pbe_list(): initialized %d PBEs\n", num);
804 * alloc_pagedir - Allocate the page directory.
806 * First, determine exactly how many pages we need and
809 * We arrange the pages in a chain: each page is an array of PBES_PER_PAGE
810 * struct pbe elements (pbes) and the last element in the page points
813 * On each page we set up a list of struct_pbe elements.
816 static struct pbe * alloc_pagedir(unsigned nr_pages)
819 struct pbe *pblist, *pbe;
824 pr_debug("alloc_pagedir(): nr_pages = %d\n", nr_pages);
825 pblist = (struct pbe *)get_zeroed_page(GFP_ATOMIC | __GFP_COLD);
826 for (pbe = pblist, num = PBES_PER_PAGE; pbe && num < nr_pages;
827 pbe = pbe->next, num += PBES_PER_PAGE) {
829 pbe->next = (struct pbe *)get_zeroed_page(GFP_ATOMIC | __GFP_COLD);
831 if (!pbe) { /* get_zeroed_page() failed */
832 free_pagedir(pblist);
839 * free_image_pages - Free pages allocated for snapshot
842 static void free_image_pages(void)
846 for_each_pbe (p, pagedir_save) {
848 ClearPageNosave(virt_to_page(p->address));
849 free_page(p->address);
856 * alloc_image_pages - Allocate pages for the snapshot.
859 static int alloc_image_pages(void)
863 for_each_pbe (p, pagedir_save) {
864 p->address = get_zeroed_page(GFP_ATOMIC | __GFP_COLD);
867 SetPageNosave(virt_to_page(p->address));
872 /* Free pages we allocated for suspend. Suspend pages are alocated
873 * before atomic copy, so we need to free them after resume.
875 void swsusp_free(void)
877 BUG_ON(PageNosave(virt_to_page(pagedir_save)));
878 BUG_ON(PageNosaveFree(virt_to_page(pagedir_save)));
880 free_pagedir(pagedir_save);
885 * enough_free_mem - Make sure we enough free memory to snapshot.
887 * Returns TRUE or FALSE after checking the number of available
891 static int enough_free_mem(void)
893 if (nr_free_pages() < (nr_copy_pages + PAGES_FOR_IO)) {
894 pr_debug("swsusp: Not enough free pages: Have %d\n",
903 * enough_swap - Make sure we have enough swap to save the image.
905 * Returns TRUE or FALSE after checking the total amount of swap
908 * FIXME: si_swapinfo(&i) returns all swap devices information.
909 * We should only consider resume_device.
912 static int enough_swap(void)
917 if (i.freeswap < (nr_copy_pages + PAGES_FOR_IO)) {
918 pr_debug("swsusp: Not enough swap. Need %ld\n",i.freeswap);
924 static int swsusp_alloc(void)
928 pagedir_nosave = NULL;
929 nr_copy_pages = calc_nr(nr_copy_pages);
930 nr_copy_pages_check = nr_copy_pages;
932 pr_debug("suspend: (pages needed: %d + %d free: %d)\n",
933 nr_copy_pages, PAGES_FOR_IO, nr_free_pages());
935 if (!enough_free_mem())
941 if (MAX_PBES < nr_copy_pages / PBES_PER_PAGE +
942 !!(nr_copy_pages % PBES_PER_PAGE))
945 if (!(pagedir_save = alloc_pagedir(nr_copy_pages))) {
946 printk(KERN_ERR "suspend: Allocating pagedir failed.\n");
949 create_pbe_list(pagedir_save, nr_copy_pages);
950 pagedir_nosave = pagedir_save;
951 if ((error = alloc_image_pages())) {
952 printk(KERN_ERR "suspend: Allocating image pages failed.\n");
960 static int suspend_prepare_image(void)
964 pr_debug("swsusp: critical section: \n");
965 if (save_highmem()) {
966 printk(KERN_CRIT "Suspend machine: Not enough free pages for highmem\n");
973 printk("swsusp: Need to copy %u pages\n", nr_copy_pages);
975 error = swsusp_alloc();
979 /* During allocating of suspend pagedir, new cold pages may appear.
986 * End of critical section. From now on, we can write to memory,
987 * but we should not touch disk. This specially means we must _not_
988 * touch swap space! Except we must write out our image of course.
991 printk("swsusp: critical section/: done (%d pages copied)\n", nr_copy_pages );
996 /* It is important _NOT_ to umount filesystems at this point. We want
997 * them synced (in case something goes wrong) but we DO not want to mark
998 * filesystem clean: it is not. (And it does not matter, if we resume
999 * correctly, we'll mark system clean, anyway.)
1001 int swsusp_write(void)
1006 error = write_suspend_image();
1007 /* This will unlock ignored swap devices since writing is finished */
1014 extern asmlinkage int swsusp_arch_suspend(void);
1015 extern asmlinkage int swsusp_arch_resume(void);
1018 asmlinkage int swsusp_save(void)
1020 return suspend_prepare_image();
1023 int swsusp_suspend(void)
1026 if ((error = arch_prepare_suspend()))
1028 local_irq_disable();
1029 /* At this point, device_suspend() has been called, but *not*
1030 * device_power_down(). We *must* device_power_down() now.
1031 * Otherwise, drivers for some devices (e.g. interrupt controllers)
1032 * become desynchronized with the actual state of the hardware
1033 * at resume time, and evil weirdness ensues.
1035 if ((error = device_power_down(PMSG_FREEZE))) {
1036 printk(KERN_ERR "Some devices failed to power down, aborting suspend\n");
1041 if ((error = swsusp_swap_check())) {
1042 printk(KERN_ERR "swsusp: cannot find swap device, try swapon -a.\n");
1048 save_processor_state();
1049 if ((error = swsusp_arch_suspend()))
1050 printk(KERN_ERR "Error %d suspending\n", error);
1051 /* Restore control flow magically appears here */
1052 restore_processor_state();
1053 BUG_ON (nr_copy_pages_check != nr_copy_pages);
1060 int swsusp_resume(void)
1063 local_irq_disable();
1064 if (device_power_down(PMSG_FREEZE))
1065 printk(KERN_ERR "Some devices failed to power down, very bad\n");
1066 /* We'll ignore saved state, but this gets preempt count (etc) right */
1067 save_processor_state();
1068 error = swsusp_arch_resume();
1069 /* Code below is only ever reached in case of failure. Otherwise
1070 * execution continues at place where swsusp_arch_suspend was called
1073 restore_processor_state();
1075 touch_softlockup_watchdog();
1082 * On resume, for storing the PBE list and the image,
1083 * we can only use memory pages that do not conflict with the pages
1084 * which had been used before suspend.
1086 * We don't know which pages are usable until we allocate them.
1088 * Allocated but unusable (ie eaten) memory pages are linked together
1089 * to create a list, so that we can free them easily
1091 * We could have used a type other than (void *)
1092 * for this purpose, but ...
1094 static void **eaten_memory = NULL;
1096 static inline void eat_page(void *page)
1101 eaten_memory = page;
1105 unsigned long get_usable_page(gfp_t gfp_mask)
1109 m = get_zeroed_page(gfp_mask);
1110 while (!PageNosaveFree(virt_to_page(m))) {
1111 eat_page((void *)m);
1112 m = get_zeroed_page(gfp_mask);
1119 void free_eaten_memory(void)
1127 m = (unsigned long)c;
1132 eaten_memory = NULL;
1133 pr_debug("swsusp: %d unused pages freed\n", i);
1137 * check_pagedir - We ensure here that pages that the PBEs point to
1138 * won't collide with pages where we're going to restore from the loaded
1142 static int check_pagedir(struct pbe *pblist)
1146 /* This is necessary, so that we can free allocated pages
1147 * in case of failure
1149 for_each_pbe (p, pblist)
1152 for_each_pbe (p, pblist) {
1153 p->address = get_usable_page(GFP_ATOMIC);
1161 * swsusp_pagedir_relocate - It is possible, that some memory pages
1162 * occupied by the list of PBEs collide with pages where we're going to
1163 * restore from the loaded pages later. We relocate them here.
1166 static struct pbe * swsusp_pagedir_relocate(struct pbe *pblist)
1169 unsigned long zone_pfn;
1170 struct pbe *pbpage, *tail, *p;
1172 int rel = 0, error = 0;
1174 if (!pblist) /* a sanity check */
1177 pr_debug("swsusp: Relocating pagedir (%lu pages to check)\n",
1178 swsusp_info.pagedir_pages);
1180 /* Set page flags */
1182 for_each_zone (zone) {
1183 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
1184 SetPageNosaveFree(pfn_to_page(zone_pfn +
1185 zone->zone_start_pfn));
1188 /* Clear orig addresses */
1190 for_each_pbe (p, pblist)
1191 ClearPageNosaveFree(virt_to_page(p->orig_address));
1193 tail = pblist + PB_PAGE_SKIP;
1195 /* Relocate colliding pages */
1197 for_each_pb_page (pbpage, pblist) {
1198 if (!PageNosaveFree(virt_to_page((unsigned long)pbpage))) {
1199 m = (void *)get_usable_page(GFP_ATOMIC | __GFP_COLD);
1204 memcpy(m, (void *)pbpage, PAGE_SIZE);
1205 if (pbpage == pblist)
1206 pblist = (struct pbe *)m;
1208 tail->next = (struct pbe *)m;
1210 eat_page((void *)pbpage);
1211 pbpage = (struct pbe *)m;
1213 /* We have to link the PBEs again */
1215 for (p = pbpage; p < pbpage + PB_PAGE_SKIP; p++)
1216 if (p->next) /* needed to save the end */
1221 tail = pbpage + PB_PAGE_SKIP;
1225 printk("\nswsusp: Out of memory\n\n");
1226 free_pagedir(pblist);
1227 free_eaten_memory();
1229 /* Is this even worth handling? It should never ever happen, and we
1230 have just lost user's state, anyway... */
1232 printk("swsusp: Relocated %d pages\n", rel);
1238 * Using bio to read from swap.
1239 * This code requires a bit more work than just using buffer heads
1240 * but, it is the recommended way for 2.5/2.6.
1241 * The following are to signal the beginning and end of I/O. Bios
1242 * finish asynchronously, while we want them to happen synchronously.
1243 * A simple atomic_t, and a wait loop take care of this problem.
1246 static atomic_t io_done = ATOMIC_INIT(0);
1248 static int end_io(struct bio * bio, unsigned int num, int err)
1250 if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1251 panic("I/O error reading memory image");
1252 atomic_set(&io_done, 0);
1256 static struct block_device * resume_bdev;
1259 * submit - submit BIO request.
1260 * @rw: READ or WRITE.
1261 * @off physical offset of page.
1262 * @page: page we're reading or writing.
1264 * Straight from the textbook - allocate and initialize the bio.
1265 * If we're writing, make sure the page is marked as dirty.
1266 * Then submit it and wait.
1269 static int submit(int rw, pgoff_t page_off, void * page)
1274 bio = bio_alloc(GFP_ATOMIC, 1);
1277 bio->bi_sector = page_off * (PAGE_SIZE >> 9);
1279 bio->bi_bdev = resume_bdev;
1280 bio->bi_end_io = end_io;
1282 if (bio_add_page(bio, virt_to_page(page), PAGE_SIZE, 0) < PAGE_SIZE) {
1283 printk("swsusp: ERROR: adding page to bio at %ld\n",page_off);
1289 bio_set_pages_dirty(bio);
1291 atomic_set(&io_done, 1);
1292 submit_bio(rw | (1 << BIO_RW_SYNC), bio);
1293 while (atomic_read(&io_done))
1301 static int bio_read_page(pgoff_t page_off, void * page)
1303 return submit(READ, page_off, page);
1306 static int bio_write_page(pgoff_t page_off, void * page)
1308 return submit(WRITE, page_off, page);
1312 * Sanity check if this image makes sense with this kernel/swap context
1313 * I really don't think that it's foolproof but more than nothing..
1316 static const char * sanity_check(void)
1319 if (swsusp_info.version_code != LINUX_VERSION_CODE)
1320 return "kernel version";
1321 if (swsusp_info.num_physpages != num_physpages)
1322 return "memory size";
1323 if (strcmp(swsusp_info.uts.sysname,system_utsname.sysname))
1324 return "system type";
1325 if (strcmp(swsusp_info.uts.release,system_utsname.release))
1326 return "kernel release";
1327 if (strcmp(swsusp_info.uts.version,system_utsname.version))
1329 if (strcmp(swsusp_info.uts.machine,system_utsname.machine))
1332 /* We can't use number of online CPUs when we use hotplug to remove them ;-))) */
1333 if (swsusp_info.cpus != num_possible_cpus())
1334 return "number of cpus";
1340 static int check_header(void)
1342 const char * reason = NULL;
1345 if ((error = bio_read_page(swp_offset(swsusp_header.swsusp_info), &swsusp_info)))
1348 /* Is this same machine? */
1349 if ((reason = sanity_check())) {
1350 printk(KERN_ERR "swsusp: Resume mismatch: %s\n",reason);
1353 nr_copy_pages = swsusp_info.image_pages;
1357 static int check_sig(void)
1361 memset(&swsusp_header, 0, sizeof(swsusp_header));
1362 if ((error = bio_read_page(0, &swsusp_header)))
1364 if (!memcmp(SWSUSP_SIG, swsusp_header.sig, 10)) {
1365 memcpy(swsusp_header.sig, swsusp_header.orig_sig, 10);
1366 memcpy(key_iv, swsusp_header.key_iv, MAXKEY+MAXIV);
1367 memset(swsusp_header.key_iv, 0, MAXKEY+MAXIV);
1370 * Reset swap signature now.
1372 error = bio_write_page(0, &swsusp_header);
1377 pr_debug("swsusp: Signature found, resuming\n");
1382 * data_read - Read image pages from swap.
1384 * You do not need to check for overlaps, check_pagedir()
1388 static int data_read(struct pbe *pblist)
1393 int mod = swsusp_info.image_pages / 100;
1396 if ((error = crypto_init(0, &tfm)))
1402 printk("swsusp: Reading image data (%lu pages): ",
1403 swsusp_info.image_pages);
1405 for_each_pbe (p, pblist) {
1407 printk("\b\b\b\b%3d%%", i / mod);
1409 if ((error = crypto_read(p, tfm))) {
1416 printk("\b\b\b\bdone\n");
1422 * read_pagedir - Read page backup list pages from swap
1425 static int read_pagedir(struct pbe *pblist)
1427 struct pbe *pbpage, *p;
1434 printk("swsusp: Reading pagedir (%lu pages)\n",
1435 swsusp_info.pagedir_pages);
1437 for_each_pb_page (pbpage, pblist) {
1438 unsigned long offset = swp_offset(swsusp_info.pagedir[i++]);
1442 p = (pbpage + PB_PAGE_SKIP)->next;
1443 error = bio_read_page(offset, (void *)pbpage);
1444 (pbpage + PB_PAGE_SKIP)->next = p;
1451 free_pagedir(pblist);
1453 BUG_ON(i != swsusp_info.pagedir_pages);
1459 static int check_suspend_image(void)
1463 if ((error = check_sig()))
1466 if ((error = check_header()))
1472 static int read_suspend_image(void)
1477 if (!(p = alloc_pagedir(nr_copy_pages)))
1480 if ((error = read_pagedir(p)))
1483 create_pbe_list(p, nr_copy_pages);
1485 if (!(pagedir_nosave = swsusp_pagedir_relocate(p)))
1488 /* Allocate memory for the image and read the data from swap */
1490 error = check_pagedir(pagedir_nosave);
1493 error = data_read(pagedir_nosave);
1495 if (error) { /* We fail cleanly */
1496 free_eaten_memory();
1497 for_each_pbe (p, pagedir_nosave)
1499 free_page(p->address);
1502 free_pagedir(pagedir_nosave);
1508 * swsusp_check - Check for saved image in swap
1511 int swsusp_check(void)
1515 resume_bdev = open_by_devnum(swsusp_resume_device, FMODE_READ);
1516 if (!IS_ERR(resume_bdev)) {
1517 set_blocksize(resume_bdev, PAGE_SIZE);
1518 error = check_suspend_image();
1520 blkdev_put(resume_bdev);
1522 error = PTR_ERR(resume_bdev);
1525 pr_debug("swsusp: resume file found\n");
1527 pr_debug("swsusp: Error %d check for resume file\n", error);
1532 * swsusp_read - Read saved image from swap.
1535 int swsusp_read(void)
1539 if (IS_ERR(resume_bdev)) {
1540 pr_debug("swsusp: block device not initialised\n");
1541 return PTR_ERR(resume_bdev);
1544 error = read_suspend_image();
1545 blkdev_put(resume_bdev);
1546 memset(key_iv, 0, MAXKEY+MAXIV);
1549 pr_debug("swsusp: Reading resume file was successful\n");
1551 pr_debug("swsusp: Error %d resuming\n", error);
1556 * swsusp_close - close swap device.
1559 void swsusp_close(void)
1561 if (IS_ERR(resume_bdev)) {
1562 pr_debug("swsusp: block device not initialised\n");
1566 blkdev_put(resume_bdev);