2 * linux/kernel/power/snapshot.c
4 * This file provide system snapshot/restore functionality.
6 * Copyright (C) 1998-2005 Pavel Machek <pavel@suse.cz>
8 * This file is released under the GPLv2, and is based on swsusp.c.
13 #include <linux/version.h>
14 #include <linux/module.h>
16 #include <linux/suspend.h>
17 #include <linux/smp_lock.h>
18 #include <linux/delay.h>
19 #include <linux/bitops.h>
20 #include <linux/spinlock.h>
21 #include <linux/kernel.h>
23 #include <linux/device.h>
24 #include <linux/bootmem.h>
25 #include <linux/syscalls.h>
26 #include <linux/console.h>
27 #include <linux/highmem.h>
29 #include <asm/uaccess.h>
30 #include <asm/mmu_context.h>
31 #include <asm/pgtable.h>
32 #include <asm/tlbflush.h>
37 struct pbe *pagedir_nosave;
38 static unsigned int nr_copy_pages;
39 static unsigned int nr_meta_pages;
40 static unsigned long *buffer;
43 unsigned int count_highmem_pages(void)
46 unsigned long zone_pfn;
50 if (is_highmem(zone)) {
51 mark_free_pages(zone);
52 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; zone_pfn++) {
54 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
57 page = pfn_to_page(pfn);
58 if (PageReserved(page))
60 if (PageNosaveFree(page))
71 struct highmem_page *next;
74 static struct highmem_page *highmem_copy;
76 static int save_highmem_zone(struct zone *zone)
78 unsigned long zone_pfn;
79 mark_free_pages(zone);
80 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
82 struct highmem_page *save;
84 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
90 page = pfn_to_page(pfn);
92 * This condition results from rvmalloc() sans vmalloc_32()
93 * and architectural memory reservations. This should be
94 * corrected eventually when the cases giving rise to this
95 * are better understood.
97 if (PageReserved(page))
99 BUG_ON(PageNosave(page));
100 if (PageNosaveFree(page))
102 save = kmalloc(sizeof(struct highmem_page), GFP_ATOMIC);
105 save->next = highmem_copy;
107 save->data = (void *) get_zeroed_page(GFP_ATOMIC);
112 kaddr = kmap_atomic(page, KM_USER0);
113 memcpy(save->data, kaddr, PAGE_SIZE);
114 kunmap_atomic(kaddr, KM_USER0);
120 int save_highmem(void)
125 pr_debug("swsusp: Saving Highmem");
127 for_each_zone (zone) {
128 if (is_highmem(zone))
129 res = save_highmem_zone(zone);
137 int restore_highmem(void)
139 printk("swsusp: Restoring Highmem\n");
140 while (highmem_copy) {
141 struct highmem_page *save = highmem_copy;
143 highmem_copy = save->next;
145 kaddr = kmap_atomic(save->page, KM_USER0);
146 memcpy(kaddr, save->data, PAGE_SIZE);
147 kunmap_atomic(kaddr, KM_USER0);
148 free_page((long) save->data);
154 static inline unsigned int count_highmem_pages(void) {return 0;}
155 static inline int save_highmem(void) {return 0;}
156 static inline int restore_highmem(void) {return 0;}
159 static int pfn_is_nosave(unsigned long pfn)
161 unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
162 unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
163 return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
167 * saveable - Determine whether a page should be cloned or not.
170 * We save a page if it's Reserved, and not in the range of pages
171 * statically defined as 'unsaveable', or if it isn't reserved, and
172 * isn't part of a free chunk of pages.
175 static int saveable(struct zone *zone, unsigned long *zone_pfn)
177 unsigned long pfn = *zone_pfn + zone->zone_start_pfn;
183 page = pfn_to_page(pfn);
184 BUG_ON(PageReserved(page) && PageNosave(page));
185 if (PageNosave(page))
187 if (PageReserved(page) && pfn_is_nosave(pfn))
189 if (PageNosaveFree(page))
195 unsigned int count_data_pages(void)
198 unsigned long zone_pfn;
201 for_each_zone (zone) {
202 if (is_highmem(zone))
204 mark_free_pages(zone);
205 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
206 n += saveable(zone, &zone_pfn);
211 static void copy_data_pages(struct pbe *pblist)
214 unsigned long zone_pfn;
218 for_each_zone (zone) {
219 if (is_highmem(zone))
221 mark_free_pages(zone);
222 /* This is necessary for swsusp_free() */
223 for_each_pb_page (p, pblist)
224 SetPageNosaveFree(virt_to_page(p));
225 for_each_pbe (p, pblist)
226 SetPageNosaveFree(virt_to_page(p->address));
227 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
228 if (saveable(zone, &zone_pfn)) {
230 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
232 pbe->orig_address = (unsigned long)page_address(page);
233 /* copy_page is not usable for copying task structs. */
234 memcpy((void *)pbe->address, (void *)pbe->orig_address, PAGE_SIZE);
244 * free_pagedir - free pages allocated with alloc_pagedir()
247 static void free_pagedir(struct pbe *pblist, int clear_nosave_free)
252 pbe = (pblist + PB_PAGE_SKIP)->next;
253 ClearPageNosave(virt_to_page(pblist));
254 if (clear_nosave_free)
255 ClearPageNosaveFree(virt_to_page(pblist));
256 free_page((unsigned long)pblist);
262 * fill_pb_page - Create a list of PBEs on a given memory page
265 static inline void fill_pb_page(struct pbe *pbpage)
270 pbpage += PB_PAGE_SKIP;
273 while (++p < pbpage);
277 * create_pbe_list - Create a list of PBEs on top of a given chain
278 * of memory pages allocated with alloc_pagedir()
281 static inline void create_pbe_list(struct pbe *pblist, unsigned int nr_pages)
283 struct pbe *pbpage, *p;
284 unsigned int num = PBES_PER_PAGE;
286 for_each_pb_page (pbpage, pblist) {
290 fill_pb_page(pbpage);
291 num += PBES_PER_PAGE;
294 for (num -= PBES_PER_PAGE - 1, p = pbpage; num < nr_pages; p++, num++)
300 static unsigned int unsafe_pages;
303 * @safe_needed - on resume, for storing the PBE list and the image,
304 * we can only use memory pages that do not conflict with the pages
305 * used before suspend.
307 * The unsafe pages are marked with the PG_nosave_free flag
308 * and we count them using unsafe_pages
311 static inline void *alloc_image_page(gfp_t gfp_mask, int safe_needed)
315 res = (void *)get_zeroed_page(gfp_mask);
317 while (res && PageNosaveFree(virt_to_page(res))) {
318 /* The page is unsafe, mark it for swsusp_free() */
319 SetPageNosave(virt_to_page(res));
321 res = (void *)get_zeroed_page(gfp_mask);
324 SetPageNosave(virt_to_page(res));
325 SetPageNosaveFree(virt_to_page(res));
330 unsigned long get_safe_page(gfp_t gfp_mask)
332 return (unsigned long)alloc_image_page(gfp_mask, 1);
336 * alloc_pagedir - Allocate the page directory.
338 * First, determine exactly how many pages we need and
341 * We arrange the pages in a chain: each page is an array of PBES_PER_PAGE
342 * struct pbe elements (pbes) and the last element in the page points
345 * On each page we set up a list of struct_pbe elements.
348 static struct pbe *alloc_pagedir(unsigned int nr_pages, gfp_t gfp_mask,
352 struct pbe *pblist, *pbe;
357 pblist = alloc_image_page(gfp_mask, safe_needed);
358 /* FIXME: rewrite this ugly loop */
359 for (pbe = pblist, num = PBES_PER_PAGE; pbe && num < nr_pages;
360 pbe = pbe->next, num += PBES_PER_PAGE) {
362 pbe->next = alloc_image_page(gfp_mask, safe_needed);
364 if (!pbe) { /* get_zeroed_page() failed */
365 free_pagedir(pblist, 1);
368 create_pbe_list(pblist, nr_pages);
373 * Free pages we allocated for suspend. Suspend pages are alocated
374 * before atomic copy, so we need to free them after resume.
377 void swsusp_free(void)
380 unsigned long zone_pfn;
382 for_each_zone(zone) {
383 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
384 if (pfn_valid(zone_pfn + zone->zone_start_pfn)) {
386 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
387 if (PageNosave(page) && PageNosaveFree(page)) {
388 ClearPageNosave(page);
389 ClearPageNosaveFree(page);
390 free_page((long) page_address(page));
396 pagedir_nosave = NULL;
402 * enough_free_mem - Make sure we enough free memory to snapshot.
404 * Returns TRUE or FALSE after checking the number of available
408 static int enough_free_mem(unsigned int nr_pages)
414 if (!is_highmem(zone))
415 n += zone->free_pages;
416 pr_debug("swsusp: available memory: %u pages\n", n);
417 return n > (nr_pages + PAGES_FOR_IO +
418 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
421 static int alloc_data_pages(struct pbe *pblist, gfp_t gfp_mask, int safe_needed)
425 for_each_pbe (p, pblist) {
426 p->address = (unsigned long)alloc_image_page(gfp_mask, safe_needed);
433 static struct pbe *swsusp_alloc(unsigned int nr_pages)
437 if (!(pblist = alloc_pagedir(nr_pages, GFP_ATOMIC | __GFP_COLD, 0))) {
438 printk(KERN_ERR "suspend: Allocating pagedir failed.\n");
442 if (alloc_data_pages(pblist, GFP_ATOMIC | __GFP_COLD, 0)) {
443 printk(KERN_ERR "suspend: Allocating image pages failed.\n");
451 asmlinkage int swsusp_save(void)
453 unsigned int nr_pages;
455 pr_debug("swsusp: critical section: \n");
458 nr_pages = count_data_pages();
459 printk("swsusp: Need to copy %u pages\n", nr_pages);
461 pr_debug("swsusp: pages needed: %u + %lu + %u, free: %u\n",
463 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE,
464 PAGES_FOR_IO, nr_free_pages());
466 if (!enough_free_mem(nr_pages)) {
467 printk(KERN_ERR "swsusp: Not enough free memory\n");
471 pagedir_nosave = swsusp_alloc(nr_pages);
475 /* During allocating of suspend pagedir, new cold pages may appear.
479 copy_data_pages(pagedir_nosave);
482 * End of critical section. From now on, we can write to memory,
483 * but we should not touch disk. This specially means we must _not_
484 * touch swap space! Except we must write out our image of course.
487 nr_copy_pages = nr_pages;
488 nr_meta_pages = (nr_pages * sizeof(long) + PAGE_SIZE - 1) >> PAGE_SHIFT;
490 printk("swsusp: critical section/: done (%d pages copied)\n", nr_pages);
494 static void init_header(struct swsusp_info *info)
496 memset(info, 0, sizeof(struct swsusp_info));
497 info->version_code = LINUX_VERSION_CODE;
498 info->num_physpages = num_physpages;
499 memcpy(&info->uts, &system_utsname, sizeof(system_utsname));
500 info->cpus = num_online_cpus();
501 info->image_pages = nr_copy_pages;
502 info->pages = nr_copy_pages + nr_meta_pages + 1;
503 info->size = info->pages;
504 info->size <<= PAGE_SHIFT;
508 * pack_orig_addresses - the .orig_address fields of the PBEs from the
509 * list starting at @pbe are stored in the array @buf[] (1 page)
512 static inline struct pbe *pack_orig_addresses(unsigned long *buf, struct pbe *pbe)
516 for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
517 buf[j] = pbe->orig_address;
521 for (; j < PAGE_SIZE / sizeof(long); j++)
527 * snapshot_read_next - used for reading the system memory snapshot.
529 * On the first call to it @handle should point to a zeroed
530 * snapshot_handle structure. The structure gets updated and a pointer
531 * to it should be passed to this function every next time.
533 * The @count parameter should contain the number of bytes the caller
534 * wants to read from the snapshot. It must not be zero.
536 * On success the function returns a positive number. Then, the caller
537 * is allowed to read up to the returned number of bytes from the memory
538 * location computed by the data_of() macro. The number returned
539 * may be smaller than @count, but this only happens if the read would
540 * cross a page boundary otherwise.
542 * The function returns 0 to indicate the end of data stream condition,
543 * and a negative number is returned on error. In such cases the
544 * structure pointed to by @handle is not updated and should not be used
548 int snapshot_read_next(struct snapshot_handle *handle, size_t count)
550 if (handle->page > nr_meta_pages + nr_copy_pages)
553 /* This makes the buffer be freed by swsusp_free() */
554 buffer = alloc_image_page(GFP_ATOMIC, 0);
558 if (!handle->offset) {
559 init_header((struct swsusp_info *)buffer);
560 handle->buffer = buffer;
561 handle->pbe = pagedir_nosave;
563 if (handle->prev < handle->page) {
564 if (handle->page <= nr_meta_pages) {
565 handle->pbe = pack_orig_addresses(buffer, handle->pbe);
567 handle->pbe = pagedir_nosave;
569 handle->buffer = (void *)handle->pbe->address;
570 handle->pbe = handle->pbe->next;
572 handle->prev = handle->page;
574 handle->buf_offset = handle->page_offset;
575 if (handle->page_offset + count >= PAGE_SIZE) {
576 count = PAGE_SIZE - handle->page_offset;
577 handle->page_offset = 0;
580 handle->page_offset += count;
582 handle->offset += count;
587 * mark_unsafe_pages - mark the pages that cannot be used for storing
588 * the image during resume, because they conflict with the pages that
589 * had been used before suspend
592 static int mark_unsafe_pages(struct pbe *pblist)
595 unsigned long zone_pfn;
598 if (!pblist) /* a sanity check */
601 /* Clear page flags */
602 for_each_zone (zone) {
603 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
604 if (pfn_valid(zone_pfn + zone->zone_start_pfn))
605 ClearPageNosaveFree(pfn_to_page(zone_pfn +
606 zone->zone_start_pfn));
609 /* Mark orig addresses */
610 for_each_pbe (p, pblist) {
611 if (virt_addr_valid(p->orig_address))
612 SetPageNosaveFree(virt_to_page(p->orig_address));
622 static void copy_page_backup_list(struct pbe *dst, struct pbe *src)
624 /* We assume both lists contain the same number of elements */
626 dst->orig_address = src->orig_address;
632 static int check_header(struct swsusp_info *info)
636 if (info->version_code != LINUX_VERSION_CODE)
637 reason = "kernel version";
638 if (info->num_physpages != num_physpages)
639 reason = "memory size";
640 if (strcmp(info->uts.sysname,system_utsname.sysname))
641 reason = "system type";
642 if (strcmp(info->uts.release,system_utsname.release))
643 reason = "kernel release";
644 if (strcmp(info->uts.version,system_utsname.version))
646 if (strcmp(info->uts.machine,system_utsname.machine))
649 printk(KERN_ERR "swsusp: Resume mismatch: %s\n", reason);
656 * load header - check the image header and copy data from it
659 static int load_header(struct snapshot_handle *handle,
660 struct swsusp_info *info)
665 error = check_header(info);
667 pblist = alloc_pagedir(info->image_pages, GFP_ATOMIC, 0);
670 pagedir_nosave = pblist;
671 handle->pbe = pblist;
672 nr_copy_pages = info->image_pages;
673 nr_meta_pages = info->pages - info->image_pages - 1;
679 * unpack_orig_addresses - copy the elements of @buf[] (1 page) to
680 * the PBEs in the list starting at @pbe
683 static inline struct pbe *unpack_orig_addresses(unsigned long *buf,
688 for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
689 pbe->orig_address = buf[j];
696 * prepare_image - use metadata contained in the PBE list
697 * pointed to by pagedir_nosave to mark the pages that will
698 * be overwritten in the process of restoring the system
699 * memory state from the image ("unsafe" pages) and allocate
700 * memory for the image
702 * The idea is to allocate the PBE list first and then
703 * allocate as many pages as it's needed for the image data,
704 * but not to assign these pages to the PBEs initially.
705 * Instead, we just mark them as allocated and create a list
706 * of "safe" which will be used later
710 struct safe_page *next;
711 char padding[PAGE_SIZE - sizeof(void *)];
714 static struct safe_page *safe_pages;
716 static int prepare_image(struct snapshot_handle *handle)
719 unsigned int nr_pages = nr_copy_pages;
720 struct pbe *p, *pblist = NULL;
723 error = mark_unsafe_pages(p);
725 pblist = alloc_pagedir(nr_pages, GFP_ATOMIC, 1);
727 copy_page_backup_list(pblist, p);
733 if (!error && nr_pages > unsafe_pages) {
734 nr_pages -= unsafe_pages;
736 struct safe_page *ptr;
738 ptr = (struct safe_page *)get_zeroed_page(GFP_ATOMIC);
743 if (!PageNosaveFree(virt_to_page(ptr))) {
744 /* The page is "safe", add it to the list */
745 ptr->next = safe_pages;
748 /* Mark the page as allocated */
749 SetPageNosave(virt_to_page(ptr));
750 SetPageNosaveFree(virt_to_page(ptr));
754 pagedir_nosave = pblist;
762 static void *get_buffer(struct snapshot_handle *handle)
764 struct pbe *pbe = handle->pbe, *last = handle->last_pbe;
765 struct page *page = virt_to_page(pbe->orig_address);
767 if (PageNosave(page) && PageNosaveFree(page)) {
769 * We have allocated the "original" page frame and we can
770 * use it directly to store the read page
773 if (last && last->next)
775 return (void *)pbe->orig_address;
778 * The "original" page frame has not been allocated and we have to
779 * use a "safe" page frame to store the read page
781 pbe->address = (unsigned long)safe_pages;
782 safe_pages = safe_pages->next;
785 handle->last_pbe = pbe;
786 return (void *)pbe->address;
790 * snapshot_write_next - used for writing the system memory snapshot.
792 * On the first call to it @handle should point to a zeroed
793 * snapshot_handle structure. The structure gets updated and a pointer
794 * to it should be passed to this function every next time.
796 * The @count parameter should contain the number of bytes the caller
797 * wants to write to the image. It must not be zero.
799 * On success the function returns a positive number. Then, the caller
800 * is allowed to write up to the returned number of bytes to the memory
801 * location computed by the data_of() macro. The number returned
802 * may be smaller than @count, but this only happens if the write would
803 * cross a page boundary otherwise.
805 * The function returns 0 to indicate the "end of file" condition,
806 * and a negative number is returned on error. In such cases the
807 * structure pointed to by @handle is not updated and should not be used
811 int snapshot_write_next(struct snapshot_handle *handle, size_t count)
815 if (handle->prev && handle->page > nr_meta_pages + nr_copy_pages)
818 /* This makes the buffer be freed by swsusp_free() */
819 buffer = alloc_image_page(GFP_ATOMIC, 0);
824 handle->buffer = buffer;
825 if (handle->prev < handle->page) {
827 error = load_header(handle, (struct swsusp_info *)buffer);
830 } else if (handle->prev <= nr_meta_pages) {
831 handle->pbe = unpack_orig_addresses(buffer, handle->pbe);
833 error = prepare_image(handle);
836 handle->pbe = pagedir_nosave;
837 handle->last_pbe = NULL;
838 handle->buffer = get_buffer(handle);
841 handle->pbe = handle->pbe->next;
842 handle->buffer = get_buffer(handle);
844 handle->prev = handle->page;
846 handle->buf_offset = handle->page_offset;
847 if (handle->page_offset + count >= PAGE_SIZE) {
848 count = PAGE_SIZE - handle->page_offset;
849 handle->page_offset = 0;
852 handle->page_offset += count;
854 handle->offset += count;
858 int snapshot_image_loaded(struct snapshot_handle *handle)
860 return !(!handle->pbe || handle->pbe->next || !nr_copy_pages ||
861 handle->page <= nr_meta_pages + nr_copy_pages);