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/module.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>
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>
28 #include <asm/uaccess.h>
29 #include <asm/mmu_context.h>
30 #include <asm/pgtable.h>
31 #include <asm/tlbflush.h>
40 struct highmem_page *next;
43 static struct highmem_page *highmem_copy;
45 static int save_highmem_zone(struct zone *zone)
47 unsigned long zone_pfn;
48 mark_free_pages(zone);
49 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
51 struct highmem_page *save;
53 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
59 page = pfn_to_page(pfn);
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.
66 if (PageReserved(page)) {
67 printk("highmem reserved page?!\n");
70 BUG_ON(PageNosave(page));
71 if (PageNosaveFree(page))
73 save = kmalloc(sizeof(struct highmem_page), GFP_ATOMIC);
76 save->next = highmem_copy;
78 save->data = (void *) get_zeroed_page(GFP_ATOMIC);
83 kaddr = kmap_atomic(page, KM_USER0);
84 memcpy(save->data, kaddr, PAGE_SIZE);
85 kunmap_atomic(kaddr, KM_USER0);
92 static int save_highmem(void)
97 pr_debug("swsusp: Saving Highmem\n");
98 for_each_zone (zone) {
100 res = save_highmem_zone(zone);
107 int restore_highmem(void)
109 printk("swsusp: Restoring Highmem\n");
110 while (highmem_copy) {
111 struct highmem_page *save = highmem_copy;
113 highmem_copy = save->next;
115 kaddr = kmap_atomic(save->page, KM_USER0);
116 memcpy(kaddr, save->data, PAGE_SIZE);
117 kunmap_atomic(kaddr, KM_USER0);
118 free_page((long) save->data);
124 static int save_highmem(void) { return 0; }
125 int restore_highmem(void) { return 0; }
126 #endif /* CONFIG_HIGHMEM */
129 static int pfn_is_nosave(unsigned long pfn)
131 unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
132 unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
133 return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
137 * saveable - Determine whether a page should be cloned or not.
140 * We save a page if it's Reserved, and not in the range of pages
141 * statically defined as 'unsaveable', or if it isn't reserved, and
142 * isn't part of a free chunk of pages.
145 static int saveable(struct zone *zone, unsigned long *zone_pfn)
147 unsigned long pfn = *zone_pfn + zone->zone_start_pfn;
153 page = pfn_to_page(pfn);
154 BUG_ON(PageReserved(page) && PageNosave(page));
155 if (PageNosave(page))
157 if (PageReserved(page) && pfn_is_nosave(pfn)) {
158 pr_debug("[nosave pfn 0x%lx]", pfn);
161 if (PageNosaveFree(page))
167 static unsigned count_data_pages(void)
170 unsigned long zone_pfn;
173 for_each_zone (zone) {
174 if (is_highmem(zone))
176 mark_free_pages(zone);
177 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
178 n += saveable(zone, &zone_pfn);
183 static void copy_data_pages(struct pbe *pblist)
186 unsigned long zone_pfn;
190 for_each_zone (zone) {
191 if (is_highmem(zone))
193 mark_free_pages(zone);
194 /* This is necessary for swsusp_free() */
195 for_each_pb_page (p, pblist)
196 SetPageNosaveFree(virt_to_page(p));
197 for_each_pbe (p, pblist)
198 SetPageNosaveFree(virt_to_page(p->address));
199 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
200 if (saveable(zone, &zone_pfn)) {
202 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
204 pbe->orig_address = (unsigned long)page_address(page);
205 /* copy_page is not usable for copying task structs. */
206 memcpy((void *)pbe->address, (void *)pbe->orig_address, PAGE_SIZE);
216 * free_pagedir - free pages allocated with alloc_pagedir()
219 void free_pagedir(struct pbe *pblist)
224 pbe = (pblist + PB_PAGE_SKIP)->next;
225 ClearPageNosave(virt_to_page(pblist));
226 ClearPageNosaveFree(virt_to_page(pblist));
227 free_page((unsigned long)pblist);
233 * fill_pb_page - Create a list of PBEs on a given memory page
236 static inline void fill_pb_page(struct pbe *pbpage)
241 pbpage += PB_PAGE_SKIP;
244 while (++p < pbpage);
248 * create_pbe_list - Create a list of PBEs on top of a given chain
249 * of memory pages allocated with alloc_pagedir()
252 void create_pbe_list(struct pbe *pblist, unsigned int nr_pages)
254 struct pbe *pbpage, *p;
255 unsigned int num = PBES_PER_PAGE;
257 for_each_pb_page (pbpage, pblist) {
261 fill_pb_page(pbpage);
262 num += PBES_PER_PAGE;
265 for (num -= PBES_PER_PAGE - 1, p = pbpage; num < nr_pages; p++, num++)
269 pr_debug("create_pbe_list(): initialized %d PBEs\n", num);
273 * @safe_needed - on resume, for storing the PBE list and the image,
274 * we can only use memory pages that do not conflict with the pages
275 * which had been used before suspend.
277 * The unsafe pages are marked with the PG_nosave_free flag
279 * Allocated but unusable (ie eaten) memory pages should be marked
280 * so that swsusp_free() can release them
283 static inline void *alloc_image_page(gfp_t gfp_mask, int safe_needed)
289 res = (void *)get_zeroed_page(gfp_mask);
290 if (res && PageNosaveFree(virt_to_page(res)))
291 /* This is for swsusp_free() */
292 SetPageNosave(virt_to_page(res));
293 } while (res && PageNosaveFree(virt_to_page(res)));
295 res = (void *)get_zeroed_page(gfp_mask);
297 SetPageNosave(virt_to_page(res));
298 SetPageNosaveFree(virt_to_page(res));
303 unsigned long get_safe_page(gfp_t gfp_mask)
305 return (unsigned long)alloc_image_page(gfp_mask, 1);
309 * alloc_pagedir - Allocate the page directory.
311 * First, determine exactly how many pages we need and
314 * We arrange the pages in a chain: each page is an array of PBES_PER_PAGE
315 * struct pbe elements (pbes) and the last element in the page points
318 * On each page we set up a list of struct_pbe elements.
321 struct pbe *alloc_pagedir(unsigned int nr_pages, gfp_t gfp_mask, int safe_needed)
324 struct pbe *pblist, *pbe;
329 pr_debug("alloc_pagedir(): nr_pages = %d\n", nr_pages);
330 pblist = alloc_image_page(gfp_mask, safe_needed);
331 /* FIXME: rewrite this ugly loop */
332 for (pbe = pblist, num = PBES_PER_PAGE; pbe && num < nr_pages;
333 pbe = pbe->next, num += PBES_PER_PAGE) {
335 pbe->next = alloc_image_page(gfp_mask, safe_needed);
337 if (!pbe) { /* get_zeroed_page() failed */
338 free_pagedir(pblist);
345 * Free pages we allocated for suspend. Suspend pages are alocated
346 * before atomic copy, so we need to free them after resume.
349 void swsusp_free(void)
352 unsigned long zone_pfn;
354 for_each_zone(zone) {
355 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
356 if (pfn_valid(zone_pfn + zone->zone_start_pfn)) {
358 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
359 if (PageNosave(page) && PageNosaveFree(page)) {
360 ClearPageNosave(page);
361 ClearPageNosaveFree(page);
362 free_page((long) page_address(page));
370 * enough_free_mem - Make sure we enough free memory to snapshot.
372 * Returns TRUE or FALSE after checking the number of available
376 static int enough_free_mem(unsigned int nr_pages)
378 pr_debug("swsusp: available memory: %u pages\n", nr_free_pages());
379 return nr_free_pages() > (nr_pages + PAGES_FOR_IO +
380 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
383 int alloc_data_pages(struct pbe *pblist, gfp_t gfp_mask, int safe_needed)
387 for_each_pbe (p, pblist) {
388 p->address = (unsigned long)alloc_image_page(gfp_mask, safe_needed);
395 static struct pbe *swsusp_alloc(unsigned int nr_pages)
399 if (!(pblist = alloc_pagedir(nr_pages, GFP_ATOMIC | __GFP_COLD, 0))) {
400 printk(KERN_ERR "suspend: Allocating pagedir failed.\n");
403 create_pbe_list(pblist, nr_pages);
405 if (alloc_data_pages(pblist, GFP_ATOMIC | __GFP_COLD, 0)) {
406 printk(KERN_ERR "suspend: Allocating image pages failed.\n");
414 asmlinkage int swsusp_save(void)
416 unsigned int nr_pages;
418 pr_debug("swsusp: critical section: \n");
419 if (save_highmem()) {
420 printk(KERN_CRIT "swsusp: Not enough free pages for highmem\n");
426 nr_pages = count_data_pages();
427 printk("swsusp: Need to copy %u pages\n", nr_pages);
429 pr_debug("swsusp: pages needed: %u + %lu + %u, free: %u\n",
431 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE,
432 PAGES_FOR_IO, nr_free_pages());
434 /* This is needed because of the fixed size of swsusp_info */
435 if (MAX_PBES < (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE)
438 if (!enough_free_mem(nr_pages)) {
439 printk(KERN_ERR "swsusp: Not enough free memory\n");
443 if (!enough_swap(nr_pages)) {
444 printk(KERN_ERR "swsusp: Not enough free swap\n");
448 pagedir_nosave = swsusp_alloc(nr_pages);
452 /* During allocating of suspend pagedir, new cold pages may appear.
456 copy_data_pages(pagedir_nosave);
459 * End of critical section. From now on, we can write to memory,
460 * but we should not touch disk. This specially means we must _not_
461 * touch swap space! Except we must write out our image of course.
464 nr_copy_pages = nr_pages;
466 printk("swsusp: critical section/: done (%d pages copied)\n", nr_pages);