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;
42 unsigned int count_highmem_pages(void)
45 unsigned long zone_pfn;
49 if (is_highmem(zone)) {
50 mark_free_pages(zone);
51 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; zone_pfn++) {
53 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
56 page = pfn_to_page(pfn);
57 if (PageReserved(page))
59 if (PageNosaveFree(page))
70 struct highmem_page *next;
73 static struct highmem_page *highmem_copy;
75 static int save_highmem_zone(struct zone *zone)
77 unsigned long zone_pfn;
78 mark_free_pages(zone);
79 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
81 struct highmem_page *save;
83 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
89 page = pfn_to_page(pfn);
91 * This condition results from rvmalloc() sans vmalloc_32()
92 * and architectural memory reservations. This should be
93 * corrected eventually when the cases giving rise to this
94 * are better understood.
96 if (PageReserved(page))
98 BUG_ON(PageNosave(page));
99 if (PageNosaveFree(page))
101 save = kmalloc(sizeof(struct highmem_page), GFP_ATOMIC);
104 save->next = highmem_copy;
106 save->data = (void *) get_zeroed_page(GFP_ATOMIC);
111 kaddr = kmap_atomic(page, KM_USER0);
112 memcpy(save->data, kaddr, PAGE_SIZE);
113 kunmap_atomic(kaddr, KM_USER0);
119 int save_highmem(void)
124 pr_debug("swsusp: Saving Highmem\n");
125 for_each_zone (zone) {
126 if (is_highmem(zone))
127 res = save_highmem_zone(zone);
134 int restore_highmem(void)
136 printk("swsusp: Restoring Highmem\n");
137 while (highmem_copy) {
138 struct highmem_page *save = highmem_copy;
140 highmem_copy = save->next;
142 kaddr = kmap_atomic(save->page, KM_USER0);
143 memcpy(kaddr, save->data, PAGE_SIZE);
144 kunmap_atomic(kaddr, KM_USER0);
145 free_page((long) save->data);
152 static int pfn_is_nosave(unsigned long pfn)
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);
160 * saveable - Determine whether a page should be cloned or not.
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.
168 static int saveable(struct zone *zone, unsigned long *zone_pfn)
170 unsigned long pfn = *zone_pfn + zone->zone_start_pfn;
176 page = pfn_to_page(pfn);
177 BUG_ON(PageReserved(page) && PageNosave(page));
178 if (PageNosave(page))
180 if (PageReserved(page) && pfn_is_nosave(pfn))
182 if (PageNosaveFree(page))
188 unsigned int count_data_pages(void)
191 unsigned long zone_pfn;
194 for_each_zone (zone) {
195 if (is_highmem(zone))
197 mark_free_pages(zone);
198 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
199 n += saveable(zone, &zone_pfn);
204 static void copy_data_pages(struct pbe *pblist)
207 unsigned long zone_pfn;
211 for_each_zone (zone) {
212 if (is_highmem(zone))
214 mark_free_pages(zone);
215 /* This is necessary for swsusp_free() */
216 for_each_pb_page (p, pblist)
217 SetPageNosaveFree(virt_to_page(p));
218 for_each_pbe (p, pblist)
219 SetPageNosaveFree(virt_to_page(p->address));
220 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
221 if (saveable(zone, &zone_pfn)) {
223 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
225 pbe->orig_address = (unsigned long)page_address(page);
226 /* copy_page is not usable for copying task structs. */
227 memcpy((void *)pbe->address, (void *)pbe->orig_address, PAGE_SIZE);
237 * free_pagedir - free pages allocated with alloc_pagedir()
240 static void free_pagedir(struct pbe *pblist)
245 pbe = (pblist + PB_PAGE_SKIP)->next;
246 ClearPageNosave(virt_to_page(pblist));
247 ClearPageNosaveFree(virt_to_page(pblist));
248 free_page((unsigned long)pblist);
254 * fill_pb_page - Create a list of PBEs on a given memory page
257 static inline void fill_pb_page(struct pbe *pbpage)
262 pbpage += PB_PAGE_SKIP;
265 while (++p < pbpage);
269 * create_pbe_list - Create a list of PBEs on top of a given chain
270 * of memory pages allocated with alloc_pagedir()
273 static inline void create_pbe_list(struct pbe *pblist, unsigned int nr_pages)
275 struct pbe *pbpage, *p;
276 unsigned int num = PBES_PER_PAGE;
278 for_each_pb_page (pbpage, pblist) {
282 fill_pb_page(pbpage);
283 num += PBES_PER_PAGE;
286 for (num -= PBES_PER_PAGE - 1, p = pbpage; num < nr_pages; p++, num++)
293 * On resume it is necessary to trace and eventually free the unsafe
294 * pages that have been allocated, because they are needed for I/O
295 * (on x86-64 we likely will "eat" these pages once again while
296 * creating the temporary page translation tables)
300 struct eaten_page *next;
301 char padding[PAGE_SIZE - sizeof(void *)];
304 static struct eaten_page *eaten_pages = NULL;
306 static void release_eaten_pages(void)
308 struct eaten_page *p, *q;
313 /* We don't want swsusp_free() to free this page again */
314 ClearPageNosave(virt_to_page(p));
315 free_page((unsigned long)p);
322 * @safe_needed - on resume, for storing the PBE list and the image,
323 * we can only use memory pages that do not conflict with the pages
324 * which had been used before suspend.
326 * The unsafe pages are marked with the PG_nosave_free flag
328 * Allocated but unusable (ie eaten) memory pages should be marked
329 * so that swsusp_free() can release them
332 static inline void *alloc_image_page(gfp_t gfp_mask, int safe_needed)
338 res = (void *)get_zeroed_page(gfp_mask);
339 if (res && PageNosaveFree(virt_to_page(res))) {
340 /* This is for swsusp_free() */
341 SetPageNosave(virt_to_page(res));
342 ((struct eaten_page *)res)->next = eaten_pages;
345 } while (res && PageNosaveFree(virt_to_page(res)));
347 res = (void *)get_zeroed_page(gfp_mask);
349 SetPageNosave(virt_to_page(res));
350 SetPageNosaveFree(virt_to_page(res));
355 unsigned long get_safe_page(gfp_t gfp_mask)
357 return (unsigned long)alloc_image_page(gfp_mask, 1);
361 * alloc_pagedir - Allocate the page directory.
363 * First, determine exactly how many pages we need and
366 * We arrange the pages in a chain: each page is an array of PBES_PER_PAGE
367 * struct pbe elements (pbes) and the last element in the page points
370 * On each page we set up a list of struct_pbe elements.
373 struct pbe *alloc_pagedir(unsigned int nr_pages, gfp_t gfp_mask, int safe_needed)
376 struct pbe *pblist, *pbe;
381 pblist = alloc_image_page(gfp_mask, safe_needed);
382 /* FIXME: rewrite this ugly loop */
383 for (pbe = pblist, num = PBES_PER_PAGE; pbe && num < nr_pages;
384 pbe = pbe->next, num += PBES_PER_PAGE) {
386 pbe->next = alloc_image_page(gfp_mask, safe_needed);
388 if (!pbe) { /* get_zeroed_page() failed */
389 free_pagedir(pblist);
392 create_pbe_list(pblist, nr_pages);
397 * Free pages we allocated for suspend. Suspend pages are alocated
398 * before atomic copy, so we need to free them after resume.
401 void swsusp_free(void)
404 unsigned long zone_pfn;
406 for_each_zone(zone) {
407 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
408 if (pfn_valid(zone_pfn + zone->zone_start_pfn)) {
410 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
411 if (PageNosave(page) && PageNosaveFree(page)) {
412 ClearPageNosave(page);
413 ClearPageNosaveFree(page);
414 free_page((long) page_address(page));
420 pagedir_nosave = NULL;
425 * enough_free_mem - Make sure we enough free memory to snapshot.
427 * Returns TRUE or FALSE after checking the number of available
431 static int enough_free_mem(unsigned int nr_pages)
437 if (!is_highmem(zone))
438 n += zone->free_pages;
439 pr_debug("swsusp: available memory: %u pages\n", n);
440 return n > (nr_pages + PAGES_FOR_IO +
441 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
444 static int alloc_data_pages(struct pbe *pblist, gfp_t gfp_mask, int safe_needed)
448 for_each_pbe (p, pblist) {
449 p->address = (unsigned long)alloc_image_page(gfp_mask, safe_needed);
456 static struct pbe *swsusp_alloc(unsigned int nr_pages)
460 if (!(pblist = alloc_pagedir(nr_pages, GFP_ATOMIC | __GFP_COLD, 0))) {
461 printk(KERN_ERR "suspend: Allocating pagedir failed.\n");
465 if (alloc_data_pages(pblist, GFP_ATOMIC | __GFP_COLD, 0)) {
466 printk(KERN_ERR "suspend: Allocating image pages failed.\n");
474 asmlinkage int swsusp_save(void)
476 unsigned int nr_pages;
478 pr_debug("swsusp: critical section: \n");
481 nr_pages = count_data_pages();
482 printk("swsusp: Need to copy %u pages\n", nr_pages);
484 pr_debug("swsusp: pages needed: %u + %lu + %u, free: %u\n",
486 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE,
487 PAGES_FOR_IO, nr_free_pages());
489 if (!enough_free_mem(nr_pages)) {
490 printk(KERN_ERR "swsusp: Not enough free memory\n");
494 pagedir_nosave = swsusp_alloc(nr_pages);
498 /* During allocating of suspend pagedir, new cold pages may appear.
502 copy_data_pages(pagedir_nosave);
505 * End of critical section. From now on, we can write to memory,
506 * but we should not touch disk. This specially means we must _not_
507 * touch swap space! Except we must write out our image of course.
510 nr_copy_pages = nr_pages;
511 nr_meta_pages = (nr_pages * sizeof(long) + PAGE_SIZE - 1) >> PAGE_SHIFT;
513 printk("swsusp: critical section/: done (%d pages copied)\n", nr_pages);
517 static void init_header(struct swsusp_info *info)
519 memset(info, 0, sizeof(struct swsusp_info));
520 info->version_code = LINUX_VERSION_CODE;
521 info->num_physpages = num_physpages;
522 memcpy(&info->uts, &system_utsname, sizeof(system_utsname));
523 info->cpus = num_online_cpus();
524 info->image_pages = nr_copy_pages;
525 info->pages = nr_copy_pages + nr_meta_pages + 1;
529 * pack_orig_addresses - the .orig_address fields of the PBEs from the
530 * list starting at @pbe are stored in the array @buf[] (1 page)
533 static inline struct pbe *pack_orig_addresses(unsigned long *buf, struct pbe *pbe)
537 for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
538 buf[j] = pbe->orig_address;
542 for (; j < PAGE_SIZE / sizeof(long); j++)
548 * snapshot_read_next - used for reading the system memory snapshot.
550 * On the first call to it @handle should point to a zeroed
551 * snapshot_handle structure. The structure gets updated and a pointer
552 * to it should be passed to this function every next time.
554 * The @count parameter should contain the number of bytes the caller
555 * wants to read from the snapshot. It must not be zero.
557 * On success the function returns a positive number. Then, the caller
558 * is allowed to read up to the returned number of bytes from the memory
559 * location computed by the data_of() macro. The number returned
560 * may be smaller than @count, but this only happens if the read would
561 * cross a page boundary otherwise.
563 * The function returns 0 to indicate the end of data stream condition,
564 * and a negative number is returned on error. In such cases the
565 * structure pointed to by @handle is not updated and should not be used
569 int snapshot_read_next(struct snapshot_handle *handle, size_t count)
571 static unsigned long *buffer;
573 if (handle->page > nr_meta_pages + nr_copy_pages)
576 /* This makes the buffer be freed by swsusp_free() */
577 buffer = alloc_image_page(GFP_ATOMIC, 0);
581 if (!handle->offset) {
582 init_header((struct swsusp_info *)buffer);
583 handle->buffer = buffer;
584 handle->pbe = pagedir_nosave;
586 if (handle->prev < handle->page) {
587 if (handle->page <= nr_meta_pages) {
588 handle->pbe = pack_orig_addresses(buffer, handle->pbe);
590 handle->pbe = pagedir_nosave;
592 handle->buffer = (void *)handle->pbe->address;
593 handle->pbe = handle->pbe->next;
595 handle->prev = handle->page;
597 handle->buf_offset = handle->page_offset;
598 if (handle->page_offset + count >= PAGE_SIZE) {
599 count = PAGE_SIZE - handle->page_offset;
600 handle->page_offset = 0;
603 handle->page_offset += count;
605 handle->offset += count;
610 * mark_unsafe_pages - mark the pages that cannot be used for storing
611 * the image during resume, because they conflict with the pages that
612 * had been used before suspend
615 static int mark_unsafe_pages(struct pbe *pblist)
618 unsigned long zone_pfn;
621 if (!pblist) /* a sanity check */
624 /* Clear page flags */
625 for_each_zone (zone) {
626 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
627 if (pfn_valid(zone_pfn + zone->zone_start_pfn))
628 ClearPageNosaveFree(pfn_to_page(zone_pfn +
629 zone->zone_start_pfn));
632 /* Mark orig addresses */
633 for_each_pbe (p, pblist) {
634 if (virt_addr_valid(p->orig_address))
635 SetPageNosaveFree(virt_to_page(p->orig_address));
643 static void copy_page_backup_list(struct pbe *dst, struct pbe *src)
645 /* We assume both lists contain the same number of elements */
647 dst->orig_address = src->orig_address;
653 static int check_header(struct swsusp_info *info)
657 if (info->version_code != LINUX_VERSION_CODE)
658 reason = "kernel version";
659 if (info->num_physpages != num_physpages)
660 reason = "memory size";
661 if (strcmp(info->uts.sysname,system_utsname.sysname))
662 reason = "system type";
663 if (strcmp(info->uts.release,system_utsname.release))
664 reason = "kernel release";
665 if (strcmp(info->uts.version,system_utsname.version))
667 if (strcmp(info->uts.machine,system_utsname.machine))
670 printk(KERN_ERR "swsusp: Resume mismatch: %s\n", reason);
677 * load header - check the image header and copy data from it
680 static int load_header(struct snapshot_handle *handle,
681 struct swsusp_info *info)
686 error = check_header(info);
688 pblist = alloc_pagedir(info->image_pages, GFP_ATOMIC, 0);
691 pagedir_nosave = pblist;
692 handle->pbe = pblist;
693 nr_copy_pages = info->image_pages;
694 nr_meta_pages = info->pages - info->image_pages - 1;
700 * unpack_orig_addresses - copy the elements of @buf[] (1 page) to
701 * the PBEs in the list starting at @pbe
704 static inline struct pbe *unpack_orig_addresses(unsigned long *buf,
709 for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
710 pbe->orig_address = buf[j];
717 * create_image - use metadata contained in the PBE list
718 * pointed to by pagedir_nosave to mark the pages that will
719 * be overwritten in the process of restoring the system
720 * memory state from the image and allocate memory for
721 * the image avoiding these pages
724 static int create_image(struct snapshot_handle *handle)
727 struct pbe *p, *pblist;
730 error = mark_unsafe_pages(p);
732 pblist = alloc_pagedir(nr_copy_pages, GFP_ATOMIC, 1);
734 copy_page_backup_list(pblist, p);
740 error = alloc_data_pages(pblist, GFP_ATOMIC, 1);
742 release_eaten_pages();
743 pagedir_nosave = pblist;
745 pagedir_nosave = NULL;
754 * snapshot_write_next - used for writing the system memory snapshot.
756 * On the first call to it @handle should point to a zeroed
757 * snapshot_handle structure. The structure gets updated and a pointer
758 * to it should be passed to this function every next time.
760 * The @count parameter should contain the number of bytes the caller
761 * wants to write to the image. It must not be zero.
763 * On success the function returns a positive number. Then, the caller
764 * is allowed to write up to the returned number of bytes to the memory
765 * location computed by the data_of() macro. The number returned
766 * may be smaller than @count, but this only happens if the write would
767 * cross a page boundary otherwise.
769 * The function returns 0 to indicate the "end of file" condition,
770 * and a negative number is returned on error. In such cases the
771 * structure pointed to by @handle is not updated and should not be used
775 int snapshot_write_next(struct snapshot_handle *handle, size_t count)
777 static unsigned long *buffer;
780 if (handle->prev && handle->page > nr_meta_pages + nr_copy_pages)
783 /* This makes the buffer be freed by swsusp_free() */
784 buffer = alloc_image_page(GFP_ATOMIC, 0);
789 handle->buffer = buffer;
790 if (handle->prev < handle->page) {
792 error = load_header(handle, (struct swsusp_info *)buffer);
795 } else if (handle->prev <= nr_meta_pages) {
796 handle->pbe = unpack_orig_addresses(buffer, handle->pbe);
798 error = create_image(handle);
801 handle->pbe = pagedir_nosave;
802 handle->buffer = (void *)handle->pbe->address;
805 handle->pbe = handle->pbe->next;
806 handle->buffer = (void *)handle->pbe->address;
808 handle->prev = handle->page;
810 handle->buf_offset = handle->page_offset;
811 if (handle->page_offset + count >= PAGE_SIZE) {
812 count = PAGE_SIZE - handle->page_offset;
813 handle->page_offset = 0;
816 handle->page_offset += count;
818 handle->offset += count;
822 int snapshot_image_loaded(struct snapshot_handle *handle)
824 return !(!handle->pbe || handle->pbe->next || !nr_copy_pages ||
825 handle->page <= nr_meta_pages + nr_copy_pages);