nfsd: track last inode only in use_wgather case
[safe/jmp/linux-2.6] / kernel / power / swsusp.c
index ae46506..78c3504 100644 (file)
@@ -1,11 +1,10 @@
 /*
  * linux/kernel/power/swsusp.c
  *
- * This file is to realize architecture-independent
- * machine suspend feature using pretty near only high-level routines
+ * This file provides code to write suspend image to swap and read it back.
  *
  * Copyright (C) 1998-2001 Gabor Kuti <seasons@fornax.hu>
- * Copyright (C) 1998,2001-2004 Pavel Machek <pavel@suse.cz>
+ * Copyright (C) 1998,2001-2005 Pavel Machek <pavel@suse.cz>
  *
  * This file is released under the GPLv2.
  *
  * Alex Badea <vampire@go.ro>:
  * Fixed runaway init
  *
- * Andreas Steinmetz <ast@domdv.de>:
- * Added encrypted suspend option
+ * Rafael J. Wysocki <rjw@sisk.pl>
+ * Reworked the freeing of memory and the handling of swap
  *
  * More state savers are welcome. Especially for the scsi layer...
  *
  * For TODOs,FIXMEs also look in Documentation/power/swsusp.txt
  */
 
-#include <linux/module.h>
 #include <linux/mm.h>
 #include <linux/suspend.h>
-#include <linux/smp_lock.h>
-#include <linux/file.h>
-#include <linux/utsname.h>
-#include <linux/version.h>
-#include <linux/delay.h>
-#include <linux/reboot.h>
-#include <linux/bitops.h>
-#include <linux/vt_kern.h>
-#include <linux/kbd_kern.h>
-#include <linux/keyboard.h>
 #include <linux/spinlock.h>
-#include <linux/genhd.h>
 #include <linux/kernel.h>
 #include <linux/major.h>
 #include <linux/swap.h>
 #include <linux/pm.h>
-#include <linux/device.h>
-#include <linux/buffer_head.h>
 #include <linux/swapops.h>
 #include <linux/bootmem.h>
 #include <linux/syscalls.h>
-#include <linux/console.h>
 #include <linux/highmem.h>
-#include <linux/bio.h>
-#include <linux/mount.h>
-
-#include <asm/uaccess.h>
-#include <asm/mmu_context.h>
-#include <asm/pgtable.h>
-#include <asm/tlbflush.h>
-#include <asm/io.h>
-
-#include <linux/random.h>
-#include <linux/crypto.h>
-#include <asm/scatterlist.h>
+#include <linux/time.h>
+#include <linux/rbtree.h>
+#include <linux/io.h>
 
 #include "power.h"
 
-#define CIPHER "aes"
-#define MAXKEY 32
-#define MAXIV  32
-
-/* References to section boundaries */
-extern const void __nosave_begin, __nosave_end;
-
-/* Variables to be preserved over suspend */
-static int nr_copy_pages_check;
-
-extern char resume_file[];
-
-/* Local variables that should not be affected by save */
-static unsigned int nr_copy_pages __nosavedata = 0;
-
-/* Suspend pagedir is allocated before final copy, therefore it
-   must be freed after resume
-
-   Warning: this is evil. There are actually two pagedirs at time of
-   resume. One is "pagedir_save", which is empty frame allocated at
-   time of suspend, that must be freed. Second is "pagedir_nosave",
-   allocated at time of resume, that travels through memory not to
-   collide with anything.
-
-   Warning: this is even more evil than it seems. Pagedirs this file
-   talks about are completely different from page directories used by
-   MMU hardware.
- */
-suspend_pagedir_t *pagedir_nosave __nosavedata = NULL;
-static suspend_pagedir_t *pagedir_save;
-
-#define SWSUSP_SIG     "S1SUSPEND"
-
-static struct swsusp_header {
-       char reserved[PAGE_SIZE - 20 - MAXKEY - MAXIV - sizeof(swp_entry_t)];
-       u8 key_iv[MAXKEY+MAXIV];
-       swp_entry_t swsusp_info;
-       char    orig_sig[10];
-       char    sig[10];
-} __attribute__((packed, aligned(PAGE_SIZE))) swsusp_header;
-
-static struct swsusp_info swsusp_info;
-
 /*
- * XXX: We try to keep some more pages free so that I/O operations succeed
- * without paging. Might this be more?
+ * Preferred image size in bytes (tunable via /sys/power/image_size).
+ * When it is set to N, swsusp will do its best to ensure the image
+ * size will not exceed N bytes, but if that is impossible, it will
+ * try to create the smallest image possible.
  */
-#define PAGES_FOR_IO   512
+unsigned long image_size = 500 * 1024 * 1024;
 
-/*
- * Saving part...
- */
-
-/* We memorize in swapfile_used what swap devices are used for suspension */
-#define SWAPFILE_UNUSED    0
-#define SWAPFILE_SUSPEND   1   /* This is the suspending device */
-#define SWAPFILE_IGNORED   2   /* Those are other swap devices ignored for suspension */
-
-static unsigned short swapfile_used[MAX_SWAPFILES];
-static unsigned short root_swap;
-
-static int write_page(unsigned long addr, swp_entry_t * loc);
-static int bio_read_page(pgoff_t page_off, void * page);
-
-static u8 key_iv[MAXKEY+MAXIV];
-
-#ifdef CONFIG_SWSUSP_ENCRYPT
-
-static int crypto_init(int mode, void **mem)
-{
-       int error = 0;
-       int len;
-       char *modemsg;
-       struct crypto_tfm *tfm;
-
-       modemsg = mode ? "suspend not possible" : "resume not possible";
-
-       tfm = crypto_alloc_tfm(CIPHER, CRYPTO_TFM_MODE_CBC);
-       if(!tfm) {
-               printk(KERN_ERR "swsusp: no tfm, %s\n", modemsg);
-               error = -EINVAL;
-               goto out;
-       }
-
-       if(MAXKEY < crypto_tfm_alg_min_keysize(tfm)) {
-               printk(KERN_ERR "swsusp: key buffer too small, %s\n", modemsg);
-               error = -ENOKEY;
-               goto fail;
-       }
-
-       if (mode)
-               get_random_bytes(key_iv, MAXKEY+MAXIV);
-
-       len = crypto_tfm_alg_max_keysize(tfm);
-       if (len > MAXKEY)
-               len = MAXKEY;
-
-       if (crypto_cipher_setkey(tfm, key_iv, len)) {
-               printk(KERN_ERR "swsusp: key setup failure, %s\n", modemsg);
-               error = -EKEYREJECTED;
-               goto fail;
-       }
-
-       len = crypto_tfm_alg_ivsize(tfm);
-
-       if (MAXIV < len) {
-               printk(KERN_ERR "swsusp: iv buffer too small, %s\n", modemsg);
-               error = -EOVERFLOW;
-               goto fail;
-       }
-
-       crypto_cipher_set_iv(tfm, key_iv+MAXKEY, len);
-
-       *mem=(void *)tfm;
-
-       goto out;
-
-fail:  crypto_free_tfm(tfm);
-out:   return error;
-}
-
-static __inline__ void crypto_exit(void *mem)
-{
-       crypto_free_tfm((struct crypto_tfm *)mem);
-}
-
-static __inline__ int crypto_write(struct pbe *p, void *mem)
-{
-       int error = 0;
-       struct scatterlist src, dst;
-
-       src.page   = virt_to_page(p->address);
-       src.offset = 0;
-       src.length = PAGE_SIZE;
-       dst.page   = virt_to_page((void *)&swsusp_header);
-       dst.offset = 0;
-       dst.length = PAGE_SIZE;
-
-       error = crypto_cipher_encrypt((struct crypto_tfm *)mem, &dst, &src,
-                                       PAGE_SIZE);
-
-       if (!error)
-               error = write_page((unsigned long)&swsusp_header,
-                               &(p->swap_address));
-       return error;
-}
-
-static __inline__ int crypto_read(struct pbe *p, void *mem)
-{
-       int error = 0;
-       struct scatterlist src, dst;
-
-       error = bio_read_page(swp_offset(p->swap_address), (void *)p->address);
-       if (!error) {
-               src.offset = 0;
-               src.length = PAGE_SIZE;
-               dst.offset = 0;
-               dst.length = PAGE_SIZE;
-               src.page = dst.page = virt_to_page((void *)p->address);
-
-               error = crypto_cipher_decrypt((struct crypto_tfm *)mem, &dst,
-                                               &src, PAGE_SIZE);
-       }
-       return error;
-}
-#else
-static __inline__ int crypto_init(int mode, void *mem)
-{
-       return 0;
-}
-
-static __inline__ void crypto_exit(void *mem)
-{
-}
-
-static __inline__ int crypto_write(struct pbe *p, void *mem)
-{
-       return write_page(p->address, &(p->swap_address));
-}
-
-static __inline__ int crypto_read(struct pbe *p, void *mem)
-{
-       return bio_read_page(swp_offset(p->swap_address), (void *)p->address);
-}
-#endif
-
-static int mark_swapfiles(swp_entry_t prev)
-{
-       int error;
-
-       rw_swap_page_sync(READ,
-                         swp_entry(root_swap, 0),
-                         virt_to_page((unsigned long)&swsusp_header));
-       if (!memcmp("SWAP-SPACE",swsusp_header.sig, 10) ||
-           !memcmp("SWAPSPACE2",swsusp_header.sig, 10)) {
-               memcpy(swsusp_header.orig_sig,swsusp_header.sig, 10);
-               memcpy(swsusp_header.sig,SWSUSP_SIG, 10);
-               memcpy(swsusp_header.key_iv, key_iv, MAXKEY+MAXIV);
-               swsusp_header.swsusp_info = prev;
-               error = rw_swap_page_sync(WRITE,
-                                         swp_entry(root_swap, 0),
-                                         virt_to_page((unsigned long)
-                                                      &swsusp_header));
-       } else {
-               pr_debug("swsusp: Partition is not swap space.\n");
-               error = -ENODEV;
-       }
-       return error;
-}
-
-/*
- * Check whether the swap device is the specified resume
- * device, irrespective of whether they are specified by
- * identical names.
- *
- * (Thus, device inode aliasing is allowed.  You can say /dev/hda4
- * instead of /dev/ide/host0/bus0/target0/lun0/part4 [if using devfs]
- * and they'll be considered the same device.  This is *necessary* for
- * devfs, since the resume code can only recognize the form /dev/hda4,
- * but the suspend code would see the long name.)
- */
-static int is_resume_device(const struct swap_info_struct *swap_info)
-{
-       struct file *file = swap_info->swap_file;
-       struct inode *inode = file->f_dentry->d_inode;
-
-       return S_ISBLK(inode->i_mode) &&
-               swsusp_resume_device == MKDEV(imajor(inode), iminor(inode));
-}
-
-static int swsusp_swap_check(void) /* This is called before saving image */
-{
-       int i, len;
-
-       len=strlen(resume_file);
-       root_swap = 0xFFFF;
-
-       spin_lock(&swap_lock);
-       for (i=0; i<MAX_SWAPFILES; i++) {
-               if (!(swap_info[i].flags & SWP_WRITEOK)) {
-                       swapfile_used[i]=SWAPFILE_UNUSED;
-               } else {
-                       if (!len) {
-                               printk(KERN_WARNING "resume= option should be used to set suspend device" );
-                               if (root_swap == 0xFFFF) {
-                                       swapfile_used[i] = SWAPFILE_SUSPEND;
-                                       root_swap = i;
-                               } else
-                                       swapfile_used[i] = SWAPFILE_IGNORED;
-                       } else {
-                               /* we ignore all swap devices that are not the resume_file */
-                               if (is_resume_device(&swap_info[i])) {
-                                       swapfile_used[i] = SWAPFILE_SUSPEND;
-                                       root_swap = i;
-                               } else {
-                                       swapfile_used[i] = SWAPFILE_IGNORED;
-                               }
-                       }
-               }
-       }
-       spin_unlock(&swap_lock);
-       return (root_swap != 0xffff) ? 0 : -ENODEV;
-}
+int in_suspend __nosavedata = 0;
 
 /**
- * This is called after saving image so modification
- * will be lost after resume... and that's what we want.
- * we make the device unusable. A new call to
- * lock_swapdevices can unlock the devices.
+ *     The following functions are used for tracing the allocated
+ *     swap pages, so that they can be freed in case of an error.
  */
-static void lock_swapdevices(void)
-{
-       int i;
-
-       spin_lock(&swap_lock);
-       for (i = 0; i< MAX_SWAPFILES; i++)
-               if (swapfile_used[i] == SWAPFILE_IGNORED) {
-                       swap_info[i].flags ^= SWP_WRITEOK;
-               }
-       spin_unlock(&swap_lock);
-}
-
-/**
- *     write_page - Write one page to a fresh swap location.
- *     @addr:  Address we're writing.
- *     @loc:   Place to store the entry we used.
- *
- *     Allocate a new swap entry and 'sync' it. Note we discard -EIO
- *     errors. That is an artifact left over from swsusp. It did not
- *     check the return of rw_swap_page_sync() at all, since most pages
- *     written back to swap would return -EIO.
- *     This is a partial improvement, since we will at least return other
- *     errors, though we need to eventually fix the damn code.
- */
-static int write_page(unsigned long addr, swp_entry_t * loc)
-{
-       swp_entry_t entry;
-       int error = 0;
-
-       entry = get_swap_page();
-       if (swp_offset(entry) &&
-           swapfile_used[swp_type(entry)] == SWAPFILE_SUSPEND) {
-               error = rw_swap_page_sync(WRITE, entry,
-                                         virt_to_page(addr));
-               if (error == -EIO)
-                       error = 0;
-               if (!error)
-                       *loc = entry;
-       } else
-               error = -ENOSPC;
-       return error;
-}
-
-/**
- *     data_free - Free the swap entries used by the saved image.
- *
- *     Walk the list of used swap entries and free each one.
- *     This is only used for cleanup when suspend fails.
- */
-static void data_free(void)
-{
-       swp_entry_t entry;
-       struct pbe * p;
-
-       for_each_pbe(p, pagedir_nosave) {
-               entry = p->swap_address;
-               if (entry.val)
-                       swap_free(entry);
-               else
-                       break;
-       }
-}
 
-/**
- *     data_write - Write saved image to swap.
- *
- *     Walk the list of pages in the image and sync each one to swap.
- */
-static int data_write(void)
-{
-       int error = 0, i = 0;
-       unsigned int mod = nr_copy_pages / 100;
-       struct pbe *p;
-       void *tfm;
-
-       if ((error = crypto_init(1, &tfm)))
-               return error;
-
-       if (!mod)
-               mod = 1;
-
-       printk( "Writing data to swap (%d pages)...     ", nr_copy_pages );
-       for_each_pbe (p, pagedir_nosave) {
-               if (!(i%mod))
-                       printk( "\b\b\b\b%3d%%", i / mod );
-               if ((error = crypto_write(p, tfm))) {
-                       crypto_exit(tfm);
-                       return error;
-               }
-               i++;
-       }
-       printk("\b\b\b\bdone\n");
-       crypto_exit(tfm);
-       return error;
-}
-
-static void dump_info(void)
-{
-       pr_debug(" swsusp: Version: %u\n",swsusp_info.version_code);
-       pr_debug(" swsusp: Num Pages: %ld\n",swsusp_info.num_physpages);
-       pr_debug(" swsusp: UTS Sys: %s\n",swsusp_info.uts.sysname);
-       pr_debug(" swsusp: UTS Node: %s\n",swsusp_info.uts.nodename);
-       pr_debug(" swsusp: UTS Release: %s\n",swsusp_info.uts.release);
-       pr_debug(" swsusp: UTS Version: %s\n",swsusp_info.uts.version);
-       pr_debug(" swsusp: UTS Machine: %s\n",swsusp_info.uts.machine);
-       pr_debug(" swsusp: UTS Domain: %s\n",swsusp_info.uts.domainname);
-       pr_debug(" swsusp: CPUs: %d\n",swsusp_info.cpus);
-       pr_debug(" swsusp: Image: %ld Pages\n",swsusp_info.image_pages);
-       pr_debug(" swsusp: Pagedir: %ld Pages\n",swsusp_info.pagedir_pages);
-}
-
-static void init_header(void)
-{
-       memset(&swsusp_info, 0, sizeof(swsusp_info));
-       swsusp_info.version_code = LINUX_VERSION_CODE;
-       swsusp_info.num_physpages = num_physpages;
-       memcpy(&swsusp_info.uts, &system_utsname, sizeof(system_utsname));
-
-       swsusp_info.suspend_pagedir = pagedir_nosave;
-       swsusp_info.cpus = num_online_cpus();
-       swsusp_info.image_pages = nr_copy_pages;
-}
-
-static int close_swap(void)
-{
-       swp_entry_t entry;
-       int error;
-
-       dump_info();
-       error = write_page((unsigned long)&swsusp_info, &entry);
-       if (!error) {
-               printk( "S" );
-               error = mark_swapfiles(entry);
-               printk( "|\n" );
-       }
-       return error;
-}
-
-/**
- *     free_pagedir_entries - Free pages used by the page directory.
- *
- *     This is used during suspend for error recovery.
- */
-
-static void free_pagedir_entries(void)
-{
-       int i;
-
-       for (i = 0; i < swsusp_info.pagedir_pages; i++)
-               swap_free(swsusp_info.pagedir[i]);
-}
-
-
-/**
- *     write_pagedir - Write the array of pages holding the page directory.
- *     @last:  Last swap entry we write (needed for header).
- */
-
-static int write_pagedir(void)
-{
-       int error = 0;
-       unsigned n = 0;
-       struct pbe * pbe;
-
-       printk( "Writing pagedir...");
-       for_each_pb_page (pbe, pagedir_nosave) {
-               if ((error = write_page((unsigned long)pbe, &swsusp_info.pagedir[n++])))
-                       return error;
-       }
-
-       swsusp_info.pagedir_pages = n;
-       printk("done (%u pages)\n", n);
-       return error;
-}
-
-/**
- *     write_suspend_image - Write entire image and metadata.
- *
- */
-static int write_suspend_image(void)
-{
-       int error;
-
-       init_header();
-       if ((error = data_write()))
-               goto FreeData;
-
-       if ((error = write_pagedir()))
-               goto FreePagedir;
-
-       if ((error = close_swap()))
-               goto FreePagedir;
- Done:
-       memset(key_iv, 0, MAXKEY+MAXIV);
-       return error;
- FreePagedir:
-       free_pagedir_entries();
- FreeData:
-       data_free();
-       goto Done;
-}
-
-
-#ifdef CONFIG_HIGHMEM
-struct highmem_page {
-       char *data;
-       struct page *page;
-       struct highmem_page *next;
+struct swsusp_extent {
+       struct rb_node node;
+       unsigned long start;
+       unsigned long end;
 };
 
-static struct highmem_page *highmem_copy;
-
-static int save_highmem_zone(struct zone *zone)
-{
-       unsigned long zone_pfn;
-       mark_free_pages(zone);
-       for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
-               struct page *page;
-               struct highmem_page *save;
-               void *kaddr;
-               unsigned long pfn = zone_pfn + zone->zone_start_pfn;
-
-               if (!(pfn%1000))
-                       printk(".");
-               if (!pfn_valid(pfn))
-                       continue;
-               page = pfn_to_page(pfn);
-               /*
-                * PageReserved results from rvmalloc() sans vmalloc_32()
-                * and architectural memory reservations.
-                *
-                * rvmalloc should not cause this, because all implementations
-                * appear to always be using vmalloc_32 on architectures with
-                * highmem. This is a good thing, because we would like to save
-                * rvmalloc pages.
-                *
-                * It appears to be triggered by pages which do not point to
-                * valid memory (see arch/i386/mm/init.c:one_highpage_init(),
-                * which sets PageReserved if the page does not point to valid
-                * RAM.
-                *
-                * XXX: must remove usage of PageReserved!
-                */
-               if (PageReserved(page))
-                       continue;
-               BUG_ON(PageNosave(page));
-               if (PageNosaveFree(page))
-                       continue;
-               save = kmalloc(sizeof(struct highmem_page), GFP_ATOMIC);
-               if (!save)
-                       return -ENOMEM;
-               save->next = highmem_copy;
-               save->page = page;
-               save->data = (void *) get_zeroed_page(GFP_ATOMIC);
-               if (!save->data) {
-                       kfree(save);
-                       return -ENOMEM;
-               }
-               kaddr = kmap_atomic(page, KM_USER0);
-               memcpy(save->data, kaddr, PAGE_SIZE);
-               kunmap_atomic(kaddr, KM_USER0);
-               highmem_copy = save;
-       }
-       return 0;
-}
-#endif /* CONFIG_HIGHMEM */
-
-
-static int save_highmem(void)
-{
-#ifdef CONFIG_HIGHMEM
-       struct zone *zone;
-       int res = 0;
-
-       pr_debug("swsusp: Saving Highmem\n");
-       for_each_zone (zone) {
-               if (is_highmem(zone))
-                       res = save_highmem_zone(zone);
-               if (res)
-                       return res;
-       }
-#endif
-       return 0;
-}
-
-static int restore_highmem(void)
-{
-#ifdef CONFIG_HIGHMEM
-       printk("swsusp: Restoring Highmem\n");
-       while (highmem_copy) {
-               struct highmem_page *save = highmem_copy;
-               void *kaddr;
-               highmem_copy = save->next;
-
-               kaddr = kmap_atomic(save->page, KM_USER0);
-               memcpy(kaddr, save->data, PAGE_SIZE);
-               kunmap_atomic(kaddr, KM_USER0);
-               free_page((long) save->data);
-               kfree(save);
-       }
-#endif
-       return 0;
-}
-
-
-static int pfn_is_nosave(unsigned long pfn)
-{
-       unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
-       unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
-       return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
-}
-
-/**
- *     saveable - Determine whether a page should be cloned or not.
- *     @pfn:   The page
- *
- *     We save a page if it's Reserved, and not in the range of pages
- *     statically defined as 'unsaveable', or if it isn't reserved, and
- *     isn't part of a free chunk of pages.
- */
-
-static int saveable(struct zone * zone, unsigned long * zone_pfn)
-{
-       unsigned long pfn = *zone_pfn + zone->zone_start_pfn;
-       struct page * page;
-
-       if (!pfn_valid(pfn))
-               return 0;
-
-       page = pfn_to_page(pfn);
-       if (PageNosave(page))
-               return 0;
-       if (pfn_is_nosave(pfn)) {
-               pr_debug("[nosave pfn 0x%lx]", pfn);
-               return 0;
-       }
-       if (PageNosaveFree(page))
-               return 0;
-
-       return 1;
-}
+static struct rb_root swsusp_extents = RB_ROOT;
 
-static void count_data_pages(void)
+static int swsusp_extents_insert(unsigned long swap_offset)
 {
-       struct zone *zone;
-       unsigned long zone_pfn;
+       struct rb_node **new = &(swsusp_extents.rb_node);
+       struct rb_node *parent = NULL;
+       struct swsusp_extent *ext;
 
-       nr_copy_pages = 0;
-
-       for_each_zone (zone) {
-               if (is_highmem(zone))
-                       continue;
-               mark_free_pages(zone);
-               for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
-                       nr_copy_pages += saveable(zone, &zone_pfn);
-       }
-}
-
-static void copy_data_pages(void)
-{
-       struct zone *zone;
-       unsigned long zone_pfn;
-       struct pbe *pbe = pagedir_nosave, *p;
-
-       pr_debug("copy_data_pages(): pages to copy: %d\n", nr_copy_pages);
-       for_each_zone (zone) {
-               if (is_highmem(zone))
-                       continue;
-               mark_free_pages(zone);
-               /* This is necessary for swsusp_free() */
-               for_each_pb_page (p, pagedir_nosave)
-                       SetPageNosaveFree(virt_to_page(p));
-               for_each_pbe(p, pagedir_nosave)
-                       SetPageNosaveFree(virt_to_page(p->address));
-               for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
-                       if (saveable(zone, &zone_pfn)) {
-                               struct page * page;
-                               page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
-                               BUG_ON(!pbe);
-                               pbe->orig_address = (unsigned long)page_address(page);
-                               /* copy_page is not usable for copying task structs. */
-                               memcpy((void *)pbe->address, (void *)pbe->orig_address, PAGE_SIZE);
-                               pbe = pbe->next;
+       /* Figure out where to put the new node */
+       while (*new) {
+               ext = container_of(*new, struct swsusp_extent, node);
+               parent = *new;
+               if (swap_offset < ext->start) {
+                       /* Try to merge */
+                       if (swap_offset == ext->start - 1) {
+                               ext->start--;
+                               return 0;
                        }
+                       new = &((*new)->rb_left);
+               } else if (swap_offset > ext->end) {
+                       /* Try to merge */
+                       if (swap_offset == ext->end + 1) {
+                               ext->end++;
+                               return 0;
+                       }
+                       new = &((*new)->rb_right);
+               } else {
+                       /* It already is in the tree */
+                       return -EINVAL;
                }
        }
-       BUG_ON(pbe);
-}
-
-
-/**
- *     free_pagedir - free pages allocated with alloc_pagedir()
- */
-
-static inline void free_pagedir(struct pbe *pblist)
-{
-       struct pbe *pbe;
-
-       while (pblist) {
-               pbe = (pblist + PB_PAGE_SKIP)->next;
-               ClearPageNosave(virt_to_page(pblist));
-               ClearPageNosaveFree(virt_to_page(pblist));
-               free_page((unsigned long)pblist);
-               pblist = pbe;
-       }
-}
-
-/**
- *     fill_pb_page - Create a list of PBEs on a given memory page
- */
-
-static inline void fill_pb_page(struct pbe *pbpage)
-{
-       struct pbe *p;
+       /* Add the new node and rebalance the tree. */
+       ext = kzalloc(sizeof(struct swsusp_extent), GFP_KERNEL);
+       if (!ext)
+               return -ENOMEM;
 
-       p = pbpage;
-       pbpage += PB_PAGE_SKIP;
-       do
-               p->next = p + 1;
-       while (++p < pbpage);
+       ext->start = swap_offset;
+       ext->end = swap_offset;
+       rb_link_node(&ext->node, parent, new);
+       rb_insert_color(&ext->node, &swsusp_extents);
+       return 0;
 }
 
 /**
- *     create_pbe_list - Create a list of PBEs on top of a given chain
- *     of memory pages allocated with alloc_pagedir()
+ *     alloc_swapdev_block - allocate a swap page and register that it has
+ *     been allocated, so that it can be freed in case of an error.
  */
 
-static void create_pbe_list(struct pbe *pblist, unsigned nr_pages)
+sector_t alloc_swapdev_block(int swap)
 {
-       struct pbe *pbpage, *p;
-       unsigned num = PBES_PER_PAGE;
-
-       for_each_pb_page (pbpage, pblist) {
-               if (num >= nr_pages)
-                       break;
+       unsigned long offset;
 
-               fill_pb_page(pbpage);
-               num += PBES_PER_PAGE;
-       }
-       if (pbpage) {
-               for (num -= PBES_PER_PAGE - 1, p = pbpage; num < nr_pages; p++, num++)
-                       p->next = p + 1;
-               p->next = NULL;
-       }
-       pr_debug("create_pbe_list(): initialized %d PBEs\n", num);
-}
-
-static void *alloc_image_page(void)
-{
-       void *res = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_COLD);
-       if (res) {
-               SetPageNosave(virt_to_page(res));
-               SetPageNosaveFree(virt_to_page(res));
+       offset = swp_offset(get_swap_page_of_type(swap));
+       if (offset) {
+               if (swsusp_extents_insert(offset))
+                       swap_free(swp_entry(swap, offset));
+               else
+                       return swapdev_block(swap, offset);
        }
-       return res;
+       return 0;
 }
 
 /**
- *     alloc_pagedir - Allocate the page directory.
- *
- *     First, determine exactly how many pages we need and
- *     allocate them.
- *
- *     We arrange the pages in a chain: each page is an array of PBES_PER_PAGE
- *     struct pbe elements (pbes) and the last element in the page points
- *     to the next page.
- *
- *     On each page we set up a list of struct_pbe elements.
+ *     free_all_swap_pages - free swap pages allocated for saving image data.
+ *     It also frees the extents used to register which swap entres had been
+ *     allocated.
  */
 
-static struct pbe * alloc_pagedir(unsigned nr_pages)
+void free_all_swap_pages(int swap)
 {
-       unsigned num;
-       struct pbe *pblist, *pbe;
-
-       if (!nr_pages)
-               return NULL;
-
-       pr_debug("alloc_pagedir(): nr_pages = %d\n", nr_pages);
-       pblist = (struct pbe *)alloc_image_page();
-       for (pbe = pblist, num = PBES_PER_PAGE; pbe && num < nr_pages;
-                       pbe = pbe->next, num += PBES_PER_PAGE) {
-               pbe += PB_PAGE_SKIP;
-               pbe->next = (struct pbe *)alloc_image_page();
-       }
-       if (!pbe) { /* get_zeroed_page() failed */
-               free_pagedir(pblist);
-               pblist = NULL;
-        }
-       return pblist;
-}
+       struct rb_node *node;
 
-/**
- * Free pages we allocated for suspend. Suspend pages are alocated
- * before atomic copy, so we need to free them after resume.
- */
+       while ((node = swsusp_extents.rb_node)) {
+               struct swsusp_extent *ext;
+               unsigned long offset;
 
-void swsusp_free(void)
-{
-       struct zone *zone;
-       unsigned long zone_pfn;
+               ext = container_of(node, struct swsusp_extent, node);
+               rb_erase(node, &swsusp_extents);
+               for (offset = ext->start; offset <= ext->end; offset++)
+                       swap_free(swp_entry(swap, offset));
 
-       for_each_zone(zone) {
-               for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
-                       if (pfn_valid(zone_pfn + zone->zone_start_pfn)) {
-                               struct page * page;
-                               page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
-                               if (PageNosave(page) && PageNosaveFree(page)) {
-                                       ClearPageNosave(page);
-                                       ClearPageNosaveFree(page);
-                                       free_page((long) page_address(page));
-                               }
-                       }
+               kfree(ext);
        }
 }
 
-/**
- *     enough_free_mem - Make sure we enough free memory to snapshot.
- *
- *     Returns TRUE or FALSE after checking the number of available
- *     free pages.
- */
-
-static int enough_free_mem(void)
+int swsusp_swap_in_use(void)
 {
-       pr_debug("swsusp: available memory: %u pages\n", nr_free_pages());
-       return nr_free_pages() > (nr_copy_pages + PAGES_FOR_IO +
-               nr_copy_pages/PBES_PER_PAGE + !!(nr_copy_pages%PBES_PER_PAGE));
+       return (swsusp_extents.rb_node != NULL);
 }
 
-
 /**
- *     enough_swap - Make sure we have enough swap to save the image.
- *
- *     Returns TRUE or FALSE after checking the total amount of swap
- *     space avaiable.
+ *     swsusp_show_speed - print the time elapsed between two events represented by
+ *     @start and @stop
  *
- *     FIXME: si_swapinfo(&i) returns all swap devices information.
- *     We should only consider resume_device.
+ *     @nr_pages -     number of pages processed between @start and @stop
+ *     @msg -          introductory message to print
  */
 
-static int enough_swap(void)
+void swsusp_show_speed(struct timeval *start, struct timeval *stop,
+                       unsigned nr_pages, char *msg)
 {
-       struct sysinfo i;
+       s64 elapsed_centisecs64;
+       int centisecs;
+       int k;
+       int kps;
 
-       si_swapinfo(&i);
-       pr_debug("swsusp: available swap: %lu pages\n", i.freeswap);
-       return i.freeswap > (nr_copy_pages + PAGES_FOR_IO +
-               nr_copy_pages/PBES_PER_PAGE + !!(nr_copy_pages%PBES_PER_PAGE));
-}
-
-static int swsusp_alloc(void)
-{
-       struct pbe * p;
-
-       pagedir_nosave = NULL;
-
-       if (!(pagedir_save = alloc_pagedir(nr_copy_pages))) {
-               printk(KERN_ERR "suspend: Allocating pagedir failed.\n");
-               return -ENOMEM;
-       }
-       create_pbe_list(pagedir_save, nr_copy_pages);
-       pagedir_nosave = pagedir_save;
-
-       for_each_pbe (p, pagedir_save) {
-               p->address = (unsigned long)alloc_image_page();
-               if (!p->address) {
-                       printk(KERN_ERR "suspend: Allocating image pages failed.\n");
-                       swsusp_free();
-                       return -ENOMEM;
-               }
-       }
-
-       return 0;
-}
-
-static int suspend_prepare_image(void)
-{
-       int error;
-
-       pr_debug("swsusp: critical section: \n");
-       if (save_highmem()) {
-               printk(KERN_CRIT "swsusp: Not enough free pages for highmem\n");
-               restore_highmem();
-               return -ENOMEM;
-       }
-
-       drain_local_pages();
-       count_data_pages();
-       printk("swsusp: Need to copy %u pages\n", nr_copy_pages);
-       nr_copy_pages_check = nr_copy_pages;
-
-       pr_debug("swsusp: pages needed: %u + %lu + %u, free: %u\n",
-                nr_copy_pages,
-                nr_copy_pages/PBES_PER_PAGE + !!(nr_copy_pages%PBES_PER_PAGE),
-                PAGES_FOR_IO, nr_free_pages());
-
-       if (!enough_free_mem()) {
-               printk(KERN_ERR "swsusp: Not enough free memory\n");
-               return -ENOMEM;
-       }
-
-       if (MAX_PBES < nr_copy_pages / PBES_PER_PAGE +
-           !!(nr_copy_pages % PBES_PER_PAGE)) {
-               printk(KERN_ERR "swsusp: Too many image pages\n");
-               return -ENOSPC;
-       }
-
-       if (!enough_swap()) {
-               printk(KERN_ERR "swsusp: Not enough free swap\n");
-               return -ENOSPC;
-       }
-
-       error = swsusp_alloc();
-       if (error)
-               return error;
-
-       /* During allocating of suspend pagedir, new cold pages may appear.
-        * Kill them.
-        */
-       drain_local_pages();
-       copy_data_pages();
-
-       /*
-        * End of critical section. From now on, we can write to memory,
-        * but we should not touch disk. This specially means we must _not_
-        * touch swap space! Except we must write out our image of course.
-        */
-
-       printk("swsusp: critical section/: done (%d pages copied)\n", nr_copy_pages );
-       return 0;
-}
-
-
-/* It is important _NOT_ to umount filesystems at this point. We want
- * them synced (in case something goes wrong) but we DO not want to mark
- * filesystem clean: it is not. (And it does not matter, if we resume
- * correctly, we'll mark system clean, anyway.)
- */
-int swsusp_write(void)
-{
-       int error;
-       device_resume();
-       lock_swapdevices();
-       error = write_suspend_image();
-       /* This will unlock ignored swap devices since writing is finished */
-       lock_swapdevices();
-       return error;
-
-}
-
-
-extern asmlinkage int swsusp_arch_suspend(void);
-extern asmlinkage int swsusp_arch_resume(void);
-
-
-asmlinkage int swsusp_save(void)
-{
-       return suspend_prepare_image();
-}
-
-int swsusp_suspend(void)
-{
-       int error;
-       if ((error = arch_prepare_suspend()))
-               return error;
-       local_irq_disable();
-       /* At this point, device_suspend() has been called, but *not*
-        * device_power_down(). We *must* device_power_down() now.
-        * Otherwise, drivers for some devices (e.g. interrupt controllers)
-        * become desynchronized with the actual state of the hardware
-        * at resume time, and evil weirdness ensues.
-        */
-       if ((error = device_power_down(PMSG_FREEZE))) {
-               printk(KERN_ERR "Some devices failed to power down, aborting suspend\n");
-               local_irq_enable();
-               return error;
-       }
-
-       if ((error = swsusp_swap_check())) {
-               printk(KERN_ERR "swsusp: cannot find swap device, try swapon -a.\n");
-               device_power_up();
-               local_irq_enable();
-               return error;
-       }
-
-       save_processor_state();
-       if ((error = swsusp_arch_suspend()))
-               printk(KERN_ERR "Error %d suspending\n", error);
-       /* Restore control flow magically appears here */
-       restore_processor_state();
-       BUG_ON (nr_copy_pages_check != nr_copy_pages);
-       restore_highmem();
-       device_power_up();
-       local_irq_enable();
-       return error;
-}
-
-int swsusp_resume(void)
-{
-       int error;
-       local_irq_disable();
-       if (device_power_down(PMSG_FREEZE))
-               printk(KERN_ERR "Some devices failed to power down, very bad\n");
-       /* We'll ignore saved state, but this gets preempt count (etc) right */
-       save_processor_state();
-       error = swsusp_arch_resume();
-       /* Code below is only ever reached in case of failure. Otherwise
-        * execution continues at place where swsusp_arch_suspend was called
-         */
-       BUG_ON(!error);
-       restore_processor_state();
-       restore_highmem();
-       touch_softlockup_watchdog();
-       device_power_up();
-       local_irq_enable();
-       return error;
+       elapsed_centisecs64 = timeval_to_ns(stop) - timeval_to_ns(start);
+       do_div(elapsed_centisecs64, NSEC_PER_SEC / 100);
+       centisecs = elapsed_centisecs64;
+       if (centisecs == 0)
+               centisecs = 1;  /* avoid div-by-zero */
+       k = nr_pages * (PAGE_SIZE / 1024);
+       kps = (k * 100) / centisecs;
+       printk(KERN_INFO "PM: %s %d kbytes in %d.%02d seconds (%d.%02d MB/s)\n",
+                       msg, k,
+                       centisecs / 100, centisecs % 100,
+                       kps / 1000, (kps % 1000) / 10);
 }
 
 /**
- *     On resume, for storing the PBE list and the image,
- *     we can only use memory pages that do not conflict with the pages
- *     which had been used before suspend.
+ *     swsusp_shrink_memory -  Try to free as much memory as needed
  *
- *     We don't know which pages are usable until we allocate them.
+ *     ... but do not OOM-kill anyone
  *
- *     Allocated but unusable (ie eaten) memory pages are linked together
- *     to create a list, so that we can free them easily
- *
- *     We could have used a type other than (void *)
- *     for this purpose, but ...
+ *     Notice: all userland should be stopped before it is called, or
+ *     livelock is possible.
  */
-static void **eaten_memory = NULL;
 
-static inline void eat_page(void *page)
+#define SHRINK_BITE    10000
+static inline unsigned long __shrink_memory(long tmp)
 {
-       void **c;
-
-       c = eaten_memory;
-       eaten_memory = page;
-       *eaten_memory = c;
-}
-
-unsigned long get_usable_page(gfp_t gfp_mask)
-{
-       unsigned long m;
-
-       m = get_zeroed_page(gfp_mask);
-       while (!PageNosaveFree(virt_to_page(m))) {
-               eat_page((void *)m);
-               m = get_zeroed_page(gfp_mask);
-               if (!m)
-                       break;
-       }
-       return m;
+       if (tmp > SHRINK_BITE)
+               tmp = SHRINK_BITE;
+       return shrink_all_memory(tmp);
 }
 
-void free_eaten_memory(void)
-{
-       unsigned long m;
-       void **c;
-       int i = 0;
-
-       c = eaten_memory;
-       while (c) {
-               m = (unsigned long)c;
-               c = *c;
-               free_page(m);
-               i++;
-       }
-       eaten_memory = NULL;
-       pr_debug("swsusp: %d unused pages freed\n", i);
-}
-
-/**
- *     check_pagedir - We ensure here that pages that the PBEs point to
- *     won't collide with pages where we're going to restore from the loaded
- *     pages later
- */
-
-static int check_pagedir(struct pbe *pblist)
-{
-       struct pbe *p;
-
-       /* This is necessary, so that we can free allocated pages
-        * in case of failure
-        */
-       for_each_pbe (p, pblist)
-               p->address = 0UL;
-
-       for_each_pbe (p, pblist) {
-               p->address = get_usable_page(GFP_ATOMIC);
-               if (!p->address)
-                       return -ENOMEM;
-       }
-       return 0;
-}
-
-/**
- *     swsusp_pagedir_relocate - It is possible, that some memory pages
- *     occupied by the list of PBEs collide with pages where we're going to
- *     restore from the loaded pages later.  We relocate them here.
- */
-
-static struct pbe * swsusp_pagedir_relocate(struct pbe *pblist)
+int swsusp_shrink_memory(void)
 {
+       long tmp;
        struct zone *zone;
-       unsigned long zone_pfn;
-       struct pbe *pbpage, *tail, *p;
-       void *m;
-       int rel = 0, error = 0;
-
-       if (!pblist) /* a sanity check */
-               return NULL;
-
-       pr_debug("swsusp: Relocating pagedir (%lu pages to check)\n",
-                       swsusp_info.pagedir_pages);
-
-       /* Set page flags */
-
-       for_each_zone (zone) {
-               for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
-                       SetPageNosaveFree(pfn_to_page(zone_pfn +
-                                       zone->zone_start_pfn));
-       }
-
-       /* Clear orig addresses */
-
-       for_each_pbe (p, pblist)
-               ClearPageNosaveFree(virt_to_page(p->orig_address));
-
-       tail = pblist + PB_PAGE_SKIP;
-
-       /* Relocate colliding pages */
-
-       for_each_pb_page (pbpage, pblist) {
-               if (!PageNosaveFree(virt_to_page((unsigned long)pbpage))) {
-                       m = (void *)get_usable_page(GFP_ATOMIC | __GFP_COLD);
-                       if (!m) {
-                               error = -ENOMEM;
-                               break;
+       unsigned long pages = 0;
+       unsigned int i = 0;
+       char *p = "-\\|/";
+       struct timeval start, stop;
+
+       printk(KERN_INFO "PM: Shrinking memory...  ");
+       do_gettimeofday(&start);
+       do {
+               long size, highmem_size;
+
+               highmem_size = count_highmem_pages();
+               size = count_data_pages() + PAGES_FOR_IO + SPARE_PAGES;
+               tmp = size;
+               size += highmem_size;
+               for_each_populated_zone(zone) {
+                       tmp += snapshot_additional_pages(zone);
+                       if (is_highmem(zone)) {
+                               highmem_size -=
+                                       zone_page_state(zone, NR_FREE_PAGES);
+                       } else {
+                               tmp -= zone_page_state(zone, NR_FREE_PAGES);
+                               tmp += zone->lowmem_reserve[ZONE_NORMAL];
                        }
-                       memcpy(m, (void *)pbpage, PAGE_SIZE);
-                       if (pbpage == pblist)
-                               pblist = (struct pbe *)m;
-                       else
-                               tail->next = (struct pbe *)m;
-
-                       eat_page((void *)pbpage);
-                       pbpage = (struct pbe *)m;
-
-                       /* We have to link the PBEs again */
-
-                       for (p = pbpage; p < pbpage + PB_PAGE_SKIP; p++)
-                               if (p->next) /* needed to save the end */
-                                       p->next = p + 1;
-
-                       rel++;
                }
-               tail = pbpage + PB_PAGE_SKIP;
-       }
 
-       if (error) {
-               printk("\nswsusp: Out of memory\n\n");
-               free_pagedir(pblist);
-               free_eaten_memory();
-               pblist = NULL;
-               /* Is this even worth handling? It should never ever happen, and we
-                  have just lost user's state, anyway... */
-       } else
-               printk("swsusp: Relocated %d pages\n", rel);
+               if (highmem_size < 0)
+                       highmem_size = 0;
+
+               tmp += highmem_size;
+               if (tmp > 0) {
+                       tmp = __shrink_memory(tmp);
+                       if (!tmp)
+                               return -ENOMEM;
+                       pages += tmp;
+               } else if (size > image_size / PAGE_SIZE) {
+                       tmp = __shrink_memory(size - (image_size / PAGE_SIZE));
+                       pages += tmp;
+               }
+               printk("\b%c", p[i++%4]);
+       } while (tmp > 0);
+       do_gettimeofday(&stop);
+       printk("\bdone (%lu pages freed)\n", pages);
+       swsusp_show_speed(&start, &stop, pages, "Freed");
 
-       return pblist;
+       return 0;
 }
 
 /*
- *     Using bio to read from swap.
- *     This code requires a bit more work than just using buffer heads
- *     but, it is the recommended way for 2.5/2.6.
- *     The following are to signal the beginning and end of I/O. Bios
- *     finish asynchronously, while we want them to happen synchronously.
- *     A simple atomic_t, and a wait loop take care of this problem.
+ * Platforms, like ACPI, may want us to save some memory used by them during
+ * hibernation and to restore the contents of this memory during the subsequent
+ * resume.  The code below implements a mechanism allowing us to do that.
  */
 
-static atomic_t io_done = ATOMIC_INIT(0);
-
-static int end_io(struct bio * bio, unsigned int num, int err)
-{
-       if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
-               panic("I/O error reading memory image");
-       atomic_set(&io_done, 0);
-       return 0;
-}
+struct nvs_page {
+       unsigned long phys_start;
+       unsigned int size;
+       void *kaddr;
+       void *data;
+       struct list_head node;
+};
 
-static struct block_device * resume_bdev;
+static LIST_HEAD(nvs_list);
 
 /**
- *     submit - submit BIO request.
- *     @rw:    READ or WRITE.
- *     @off    physical offset of page.
- *     @page:  page we're reading or writing.
+ *     hibernate_nvs_register - register platform NVS memory region to save
+ *     @start - physical address of the region
+ *     @size - size of the region
  *
- *     Straight from the textbook - allocate and initialize the bio.
- *     If we're writing, make sure the page is marked as dirty.
- *     Then submit it and wait.
+ *     The NVS region need not be page-aligned (both ends) and we arrange
+ *     things so that the data from page-aligned addresses in this region will
+ *     be copied into separate RAM pages.
  */
-
-static int submit(int rw, pgoff_t page_off, void * page)
+int hibernate_nvs_register(unsigned long start, unsigned long size)
 {
-       int error = 0;
-       struct bio * bio;
-
-       bio = bio_alloc(GFP_ATOMIC, 1);
-       if (!bio)
-               return -ENOMEM;
-       bio->bi_sector = page_off * (PAGE_SIZE >> 9);
-       bio_get(bio);
-       bio->bi_bdev = resume_bdev;
-       bio->bi_end_io = end_io;
-
-       if (bio_add_page(bio, virt_to_page(page), PAGE_SIZE, 0) < PAGE_SIZE) {
-               printk("swsusp: ERROR: adding page to bio at %ld\n",page_off);
-               error = -EFAULT;
-               goto Done;
-       }
+       struct nvs_page *entry, *next;
 
-       if (rw == WRITE)
-               bio_set_pages_dirty(bio);
+       while (size > 0) {
+               unsigned int nr_bytes;
 
-       atomic_set(&io_done, 1);
-       submit_bio(rw | (1 << BIO_RW_SYNC), bio);
-       while (atomic_read(&io_done))
-               yield();
+               entry = kzalloc(sizeof(struct nvs_page), GFP_KERNEL);
+               if (!entry)
+                       goto Error;
 
- Done:
-       bio_put(bio);
-       return error;
-}
-
-static int bio_read_page(pgoff_t page_off, void * page)
-{
-       return submit(READ, page_off, page);
-}
-
-static int bio_write_page(pgoff_t page_off, void * page)
-{
-       return submit(WRITE, page_off, page);
-}
+               list_add_tail(&entry->node, &nvs_list);
+               entry->phys_start = start;
+               nr_bytes = PAGE_SIZE - (start & ~PAGE_MASK);
+               entry->size = (size < nr_bytes) ? size : nr_bytes;
 
-/*
- * Sanity check if this image makes sense with this kernel/swap context
- * I really don't think that it's foolproof but more than nothing..
- */
-
-static const char * sanity_check(void)
-{
-       dump_info();
-       if (swsusp_info.version_code != LINUX_VERSION_CODE)
-               return "kernel version";
-       if (swsusp_info.num_physpages != num_physpages)
-               return "memory size";
-       if (strcmp(swsusp_info.uts.sysname,system_utsname.sysname))
-               return "system type";
-       if (strcmp(swsusp_info.uts.release,system_utsname.release))
-               return "kernel release";
-       if (strcmp(swsusp_info.uts.version,system_utsname.version))
-               return "version";
-       if (strcmp(swsusp_info.uts.machine,system_utsname.machine))
-               return "machine";
-#if 0
-       /* We can't use number of online CPUs when we use hotplug to remove them ;-))) */
-       if (swsusp_info.cpus != num_possible_cpus())
-               return "number of cpus";
-#endif
-       return NULL;
-}
-
-
-static int check_header(void)
-{
-       const char * reason = NULL;
-       int error;
-
-       if ((error = bio_read_page(swp_offset(swsusp_header.swsusp_info), &swsusp_info)))
-               return error;
-
-       /* Is this same machine? */
-       if ((reason = sanity_check())) {
-               printk(KERN_ERR "swsusp: Resume mismatch: %s\n",reason);
-               return -EPERM;
+               start += entry->size;
+               size -= entry->size;
        }
-       nr_copy_pages = swsusp_info.image_pages;
-       return error;
-}
-
-static int check_sig(void)
-{
-       int error;
-
-       memset(&swsusp_header, 0, sizeof(swsusp_header));
-       if ((error = bio_read_page(0, &swsusp_header)))
-               return error;
-       if (!memcmp(SWSUSP_SIG, swsusp_header.sig, 10)) {
-               memcpy(swsusp_header.sig, swsusp_header.orig_sig, 10);
-               memcpy(key_iv, swsusp_header.key_iv, MAXKEY+MAXIV);
-               memset(swsusp_header.key_iv, 0, MAXKEY+MAXIV);
+       return 0;
 
-               /*
-                * Reset swap signature now.
-                */
-               error = bio_write_page(0, &swsusp_header);
-       } else { 
-               return -EINVAL;
+ Error:
+       list_for_each_entry_safe(entry, next, &nvs_list, node) {
+               list_del(&entry->node);
+               kfree(entry);
        }
-       if (!error)
-               pr_debug("swsusp: Signature found, resuming\n");
-       return error;
+       return -ENOMEM;
 }
 
 /**
- *     data_read - Read image pages from swap.
- *
- *     You do not need to check for overlaps, check_pagedir()
- *     already did that.
+ *     hibernate_nvs_free - free data pages allocated for saving NVS regions
  */
-
-static int data_read(struct pbe *pblist)
+void hibernate_nvs_free(void)
 {
-       struct pbe * p;
-       int error = 0;
-       int i = 0;
-       int mod = swsusp_info.image_pages / 100;
-       void *tfm;
-
-       if ((error = crypto_init(0, &tfm)))
-               return error;
-
-       if (!mod)
-               mod = 1;
-
-       printk("swsusp: Reading image data (%lu pages):     ",
-                       swsusp_info.image_pages);
+       struct nvs_page *entry;
 
-       for_each_pbe (p, pblist) {
-               if (!(i % mod))
-                       printk("\b\b\b\b%3d%%", i / mod);
-
-               if ((error = crypto_read(p, tfm))) {
-                       crypto_exit(tfm);
-                       return error;
+       list_for_each_entry(entry, &nvs_list, node)
+               if (entry->data) {
+                       free_page((unsigned long)entry->data);
+                       entry->data = NULL;
+                       if (entry->kaddr) {
+                               iounmap(entry->kaddr);
+                               entry->kaddr = NULL;
+                       }
                }
-
-               i++;
-       }
-       printk("\b\b\b\bdone\n");
-       crypto_exit(tfm);
-       return error;
 }
 
 /**
- *     read_pagedir - Read page backup list pages from swap
+ *     hibernate_nvs_alloc - allocate memory necessary for saving NVS regions
  */
-
-static int read_pagedir(struct pbe *pblist)
+int hibernate_nvs_alloc(void)
 {
-       struct pbe *pbpage, *p;
-       unsigned i = 0;
-       int error;
-
-       if (!pblist)
-               return -EFAULT;
-
-       printk("swsusp: Reading pagedir (%lu pages)\n",
-                       swsusp_info.pagedir_pages);
+       struct nvs_page *entry;
 
-       for_each_pb_page (pbpage, pblist) {
-               unsigned long offset = swp_offset(swsusp_info.pagedir[i++]);
-
-               error = -EFAULT;
-               if (offset) {
-                       p = (pbpage + PB_PAGE_SKIP)->next;
-                       error = bio_read_page(offset, (void *)pbpage);
-                       (pbpage + PB_PAGE_SKIP)->next = p;
+       list_for_each_entry(entry, &nvs_list, node) {
+               entry->data = (void *)__get_free_page(GFP_KERNEL);
+               if (!entry->data) {
+                       hibernate_nvs_free();
+                       return -ENOMEM;
                }
-               if (error)
-                       break;
        }
-
-       if (error)
-               free_pagedir(pblist);
-       else
-               BUG_ON(i != swsusp_info.pagedir_pages);
-
-       return error;
-}
-
-
-static int check_suspend_image(void)
-{
-       int error = 0;
-
-       if ((error = check_sig()))
-               return error;
-
-       if ((error = check_header()))
-               return error;
-
        return 0;
 }
 
-static int read_suspend_image(void)
-{
-       int error = 0;
-       struct pbe *p;
-
-       if (!(p = alloc_pagedir(nr_copy_pages)))
-               return -ENOMEM;
-
-       if ((error = read_pagedir(p)))
-               return error;
-
-       create_pbe_list(p, nr_copy_pages);
-
-       if (!(pagedir_nosave = swsusp_pagedir_relocate(p)))
-               return -ENOMEM;
-
-       /* Allocate memory for the image and read the data from swap */
-
-       error = check_pagedir(pagedir_nosave);
-
-       if (!error)
-               error = data_read(pagedir_nosave);
-
-       if (error) { /* We fail cleanly */
-               free_eaten_memory();
-               for_each_pbe (p, pagedir_nosave)
-                       if (p->address) {
-                               free_page(p->address);
-                               p->address = 0UL;
-                       }
-               free_pagedir(pagedir_nosave);
-       }
-       return error;
-}
-
 /**
- *      swsusp_check - Check for saved image in swap
+ *     hibernate_nvs_save - save NVS memory regions
  */
-
-int swsusp_check(void)
+void hibernate_nvs_save(void)
 {
-       int error;
+       struct nvs_page *entry;
 
-       resume_bdev = open_by_devnum(swsusp_resume_device, FMODE_READ);
-       if (!IS_ERR(resume_bdev)) {
-               set_blocksize(resume_bdev, PAGE_SIZE);
-               error = check_suspend_image();
-               if (error)
-                   blkdev_put(resume_bdev);
-       } else
-               error = PTR_ERR(resume_bdev);
+       printk(KERN_INFO "PM: Saving platform NVS memory\n");
 
-       if (!error)
-               pr_debug("swsusp: resume file found\n");
-       else
-               pr_debug("swsusp: Error %d check for resume file\n", error);
-       return error;
+       list_for_each_entry(entry, &nvs_list, node)
+               if (entry->data) {
+                       entry->kaddr = ioremap(entry->phys_start, entry->size);
+                       memcpy(entry->data, entry->kaddr, entry->size);
+               }
 }
 
 /**
- *     swsusp_read - Read saved image from swap.
+ *     hibernate_nvs_restore - restore NVS memory regions
+ *
+ *     This function is going to be called with interrupts disabled, so it
+ *     cannot iounmap the virtual addresses used to access the NVS region.
  */
-
-int swsusp_read(void)
+void hibernate_nvs_restore(void)
 {
-       int error;
+       struct nvs_page *entry;
 
-       if (IS_ERR(resume_bdev)) {
-               pr_debug("swsusp: block device not initialised\n");
-               return PTR_ERR(resume_bdev);
-       }
-
-       error = read_suspend_image();
-       blkdev_put(resume_bdev);
-       memset(key_iv, 0, MAXKEY+MAXIV);
-
-       if (!error)
-               pr_debug("swsusp: Reading resume file was successful\n");
-       else
-               pr_debug("swsusp: Error %d resuming\n", error);
-       return error;
-}
-
-/**
- *     swsusp_close - close swap device.
- */
-
-void swsusp_close(void)
-{
-       if (IS_ERR(resume_bdev)) {
-               pr_debug("swsusp: block device not initialised\n");
-               return;
-       }
+       printk(KERN_INFO "PM: Restoring platform NVS memory\n");
 
-       blkdev_put(resume_bdev);
+       list_for_each_entry(entry, &nvs_list, node)
+               if (entry->data)
+                       memcpy(entry->kaddr, entry->data, entry->size);
 }