Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / kernel / power / swap.c
diff --git a/kernel/kernel/power/swap.c b/kernel/kernel/power/swap.c
new file mode 100644 (file)
index 0000000..570aff8
--- /dev/null
@@ -0,0 +1,1512 @@
+/*
+ * linux/kernel/power/swap.c
+ *
+ * This file provides functions for reading the suspend image from
+ * and writing it to a swap partition.
+ *
+ * Copyright (C) 1998,2001-2005 Pavel Machek <pavel@ucw.cz>
+ * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
+ * Copyright (C) 2010-2012 Bojan Smojver <bojan@rexursive.com>
+ *
+ * This file is released under the GPLv2.
+ *
+ */
+
+#include <linux/module.h>
+#include <linux/file.h>
+#include <linux/delay.h>
+#include <linux/bitops.h>
+#include <linux/genhd.h>
+#include <linux/device.h>
+#include <linux/bio.h>
+#include <linux/blkdev.h>
+#include <linux/swap.h>
+#include <linux/swapops.h>
+#include <linux/pm.h>
+#include <linux/slab.h>
+#include <linux/lzo.h>
+#include <linux/vmalloc.h>
+#include <linux/cpumask.h>
+#include <linux/atomic.h>
+#include <linux/kthread.h>
+#include <linux/crc32.h>
+#include <linux/ktime.h>
+
+#include "power.h"
+
+#define HIBERNATE_SIG  "S1SUSPEND"
+
+/*
+ *     The swap map is a data structure used for keeping track of each page
+ *     written to a swap partition.  It consists of many swap_map_page
+ *     structures that contain each an array of MAP_PAGE_ENTRIES swap entries.
+ *     These structures are stored on the swap and linked together with the
+ *     help of the .next_swap member.
+ *
+ *     The swap map is created during suspend.  The swap map pages are
+ *     allocated and populated one at a time, so we only need one memory
+ *     page to set up the entire structure.
+ *
+ *     During resume we pick up all swap_map_page structures into a list.
+ */
+
+#define MAP_PAGE_ENTRIES       (PAGE_SIZE / sizeof(sector_t) - 1)
+
+/*
+ * Number of free pages that are not high.
+ */
+static inline unsigned long low_free_pages(void)
+{
+       return nr_free_pages() - nr_free_highpages();
+}
+
+/*
+ * Number of pages required to be kept free while writing the image. Always
+ * half of all available low pages before the writing starts.
+ */
+static inline unsigned long reqd_free_pages(void)
+{
+       return low_free_pages() / 2;
+}
+
+struct swap_map_page {
+       sector_t entries[MAP_PAGE_ENTRIES];
+       sector_t next_swap;
+};
+
+struct swap_map_page_list {
+       struct swap_map_page *map;
+       struct swap_map_page_list *next;
+};
+
+/**
+ *     The swap_map_handle structure is used for handling swap in
+ *     a file-alike way
+ */
+
+struct swap_map_handle {
+       struct swap_map_page *cur;
+       struct swap_map_page_list *maps;
+       sector_t cur_swap;
+       sector_t first_sector;
+       unsigned int k;
+       unsigned long reqd_free_pages;
+       u32 crc32;
+};
+
+struct swsusp_header {
+       char reserved[PAGE_SIZE - 20 - sizeof(sector_t) - sizeof(int) -
+                     sizeof(u32)];
+       u32     crc32;
+       sector_t image;
+       unsigned int flags;     /* Flags to pass to the "boot" kernel */
+       char    orig_sig[10];
+       char    sig[10];
+} __packed;
+
+static struct swsusp_header *swsusp_header;
+
+/**
+ *     The following functions are used for tracing the allocated
+ *     swap pages, so that they can be freed in case of an error.
+ */
+
+struct swsusp_extent {
+       struct rb_node node;
+       unsigned long start;
+       unsigned long end;
+};
+
+static struct rb_root swsusp_extents = RB_ROOT;
+
+static int swsusp_extents_insert(unsigned long swap_offset)
+{
+       struct rb_node **new = &(swsusp_extents.rb_node);
+       struct rb_node *parent = NULL;
+       struct swsusp_extent *ext;
+
+       /* Figure out where to put the new node */
+       while (*new) {
+               ext = rb_entry(*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;
+               }
+       }
+       /* Add the new node and rebalance the tree. */
+       ext = kzalloc(sizeof(struct swsusp_extent), GFP_KERNEL);
+       if (!ext)
+               return -ENOMEM;
+
+       ext->start = swap_offset;
+       ext->end = swap_offset;
+       rb_link_node(&ext->node, parent, new);
+       rb_insert_color(&ext->node, &swsusp_extents);
+       return 0;
+}
+
+/**
+ *     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.
+ */
+
+sector_t alloc_swapdev_block(int swap)
+{
+       unsigned long offset;
+
+       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 0;
+}
+
+/**
+ *     free_all_swap_pages - free swap pages allocated for saving image data.
+ *     It also frees the extents used to register which swap entries had been
+ *     allocated.
+ */
+
+void free_all_swap_pages(int swap)
+{
+       struct rb_node *node;
+
+       while ((node = swsusp_extents.rb_node)) {
+               struct swsusp_extent *ext;
+               unsigned long offset;
+
+               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));
+
+               kfree(ext);
+       }
+}
+
+int swsusp_swap_in_use(void)
+{
+       return (swsusp_extents.rb_node != NULL);
+}
+
+/*
+ * General things
+ */
+
+static unsigned short root_swap = 0xffff;
+struct block_device *hib_resume_bdev;
+
+/*
+ * Saving part
+ */
+
+static int mark_swapfiles(struct swap_map_handle *handle, unsigned int flags)
+{
+       int error;
+
+       hib_bio_read_page(swsusp_resume_block, swsusp_header, NULL);
+       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, HIBERNATE_SIG, 10);
+               swsusp_header->image = handle->first_sector;
+               swsusp_header->flags = flags;
+               if (flags & SF_CRC32_MODE)
+                       swsusp_header->crc32 = handle->crc32;
+               error = hib_bio_write_page(swsusp_resume_block,
+                                       swsusp_header, NULL);
+       } else {
+               printk(KERN_ERR "PM: Swap header not found!\n");
+               error = -ENODEV;
+       }
+       return error;
+}
+
+/**
+ *     swsusp_swap_check - check if the resume device is a swap device
+ *     and get its index (if so)
+ *
+ *     This is called before saving image
+ */
+static int swsusp_swap_check(void)
+{
+       int res;
+
+       res = swap_type_of(swsusp_resume_device, swsusp_resume_block,
+                       &hib_resume_bdev);
+       if (res < 0)
+               return res;
+
+       root_swap = res;
+       res = blkdev_get(hib_resume_bdev, FMODE_WRITE, NULL);
+       if (res)
+               return res;
+
+       res = set_blocksize(hib_resume_bdev, PAGE_SIZE);
+       if (res < 0)
+               blkdev_put(hib_resume_bdev, FMODE_WRITE);
+
+       return res;
+}
+
+/**
+ *     write_page - Write one page to given swap location.
+ *     @buf:           Address we're writing.
+ *     @offset:        Offset of the swap page we're writing to.
+ *     @bio_chain:     Link the next write BIO here
+ */
+
+static int write_page(void *buf, sector_t offset, struct bio **bio_chain)
+{
+       void *src;
+       int ret;
+
+       if (!offset)
+               return -ENOSPC;
+
+       if (bio_chain) {
+               src = (void *)__get_free_page(__GFP_WAIT | __GFP_NOWARN |
+                                             __GFP_NORETRY);
+               if (src) {
+                       copy_page(src, buf);
+               } else {
+                       ret = hib_wait_on_bio_chain(bio_chain); /* Free pages */
+                       if (ret)
+                               return ret;
+                       src = (void *)__get_free_page(__GFP_WAIT |
+                                                     __GFP_NOWARN |
+                                                     __GFP_NORETRY);
+                       if (src) {
+                               copy_page(src, buf);
+                       } else {
+                               WARN_ON_ONCE(1);
+                               bio_chain = NULL;       /* Go synchronous */
+                               src = buf;
+                       }
+               }
+       } else {
+               src = buf;
+       }
+       return hib_bio_write_page(offset, src, bio_chain);
+}
+
+static void release_swap_writer(struct swap_map_handle *handle)
+{
+       if (handle->cur)
+               free_page((unsigned long)handle->cur);
+       handle->cur = NULL;
+}
+
+static int get_swap_writer(struct swap_map_handle *handle)
+{
+       int ret;
+
+       ret = swsusp_swap_check();
+       if (ret) {
+               if (ret != -ENOSPC)
+                       printk(KERN_ERR "PM: Cannot find swap device, try "
+                                       "swapon -a.\n");
+               return ret;
+       }
+       handle->cur = (struct swap_map_page *)get_zeroed_page(GFP_KERNEL);
+       if (!handle->cur) {
+               ret = -ENOMEM;
+               goto err_close;
+       }
+       handle->cur_swap = alloc_swapdev_block(root_swap);
+       if (!handle->cur_swap) {
+               ret = -ENOSPC;
+               goto err_rel;
+       }
+       handle->k = 0;
+       handle->reqd_free_pages = reqd_free_pages();
+       handle->first_sector = handle->cur_swap;
+       return 0;
+err_rel:
+       release_swap_writer(handle);
+err_close:
+       swsusp_close(FMODE_WRITE);
+       return ret;
+}
+
+static int swap_write_page(struct swap_map_handle *handle, void *buf,
+                               struct bio **bio_chain)
+{
+       int error = 0;
+       sector_t offset;
+
+       if (!handle->cur)
+               return -EINVAL;
+       offset = alloc_swapdev_block(root_swap);
+       error = write_page(buf, offset, bio_chain);
+       if (error)
+               return error;
+       handle->cur->entries[handle->k++] = offset;
+       if (handle->k >= MAP_PAGE_ENTRIES) {
+               offset = alloc_swapdev_block(root_swap);
+               if (!offset)
+                       return -ENOSPC;
+               handle->cur->next_swap = offset;
+               error = write_page(handle->cur, handle->cur_swap, bio_chain);
+               if (error)
+                       goto out;
+               clear_page(handle->cur);
+               handle->cur_swap = offset;
+               handle->k = 0;
+
+               if (bio_chain && low_free_pages() <= handle->reqd_free_pages) {
+                       error = hib_wait_on_bio_chain(bio_chain);
+                       if (error)
+                               goto out;
+                       /*
+                        * Recalculate the number of required free pages, to
+                        * make sure we never take more than half.
+                        */
+                       handle->reqd_free_pages = reqd_free_pages();
+               }
+       }
+ out:
+       return error;
+}
+
+static int flush_swap_writer(struct swap_map_handle *handle)
+{
+       if (handle->cur && handle->cur_swap)
+               return write_page(handle->cur, handle->cur_swap, NULL);
+       else
+               return -EINVAL;
+}
+
+static int swap_writer_finish(struct swap_map_handle *handle,
+               unsigned int flags, int error)
+{
+       if (!error) {
+               flush_swap_writer(handle);
+               printk(KERN_INFO "PM: S");
+               error = mark_swapfiles(handle, flags);
+               printk("|\n");
+       }
+
+       if (error)
+               free_all_swap_pages(root_swap);
+       release_swap_writer(handle);
+       swsusp_close(FMODE_WRITE);
+
+       return error;
+}
+
+/* We need to remember how much compressed data we need to read. */
+#define LZO_HEADER     sizeof(size_t)
+
+/* Number of pages/bytes we'll compress at one time. */
+#define LZO_UNC_PAGES  32
+#define LZO_UNC_SIZE   (LZO_UNC_PAGES * PAGE_SIZE)
+
+/* Number of pages/bytes we need for compressed data (worst case). */
+#define LZO_CMP_PAGES  DIV_ROUND_UP(lzo1x_worst_compress(LZO_UNC_SIZE) + \
+                                    LZO_HEADER, PAGE_SIZE)
+#define LZO_CMP_SIZE   (LZO_CMP_PAGES * PAGE_SIZE)
+
+/* Maximum number of threads for compression/decompression. */
+#define LZO_THREADS    3
+
+/* Minimum/maximum number of pages for read buffering. */
+#define LZO_MIN_RD_PAGES       1024
+#define LZO_MAX_RD_PAGES       8192
+
+
+/**
+ *     save_image - save the suspend image data
+ */
+
+static int save_image(struct swap_map_handle *handle,
+                      struct snapshot_handle *snapshot,
+                      unsigned int nr_to_write)
+{
+       unsigned int m;
+       int ret;
+       int nr_pages;
+       int err2;
+       struct bio *bio;
+       ktime_t start;
+       ktime_t stop;
+
+       printk(KERN_INFO "PM: Saving image data pages (%u pages)...\n",
+               nr_to_write);
+       m = nr_to_write / 10;
+       if (!m)
+               m = 1;
+       nr_pages = 0;
+       bio = NULL;
+       start = ktime_get();
+       while (1) {
+               ret = snapshot_read_next(snapshot);
+               if (ret <= 0)
+                       break;
+               ret = swap_write_page(handle, data_of(*snapshot), &bio);
+               if (ret)
+                       break;
+               if (!(nr_pages % m))
+                       printk(KERN_INFO "PM: Image saving progress: %3d%%\n",
+                              nr_pages / m * 10);
+               nr_pages++;
+       }
+       err2 = hib_wait_on_bio_chain(&bio);
+       stop = ktime_get();
+       if (!ret)
+               ret = err2;
+       if (!ret)
+               printk(KERN_INFO "PM: Image saving done.\n");
+       swsusp_show_speed(start, stop, nr_to_write, "Wrote");
+       return ret;
+}
+
+/**
+ * Structure used for CRC32.
+ */
+struct crc_data {
+       struct task_struct *thr;                  /* thread */
+       atomic_t ready;                           /* ready to start flag */
+       atomic_t stop;                            /* ready to stop flag */
+       unsigned run_threads;                     /* nr current threads */
+       wait_queue_head_t go;                     /* start crc update */
+       wait_queue_head_t done;                   /* crc update done */
+       u32 *crc32;                               /* points to handle's crc32 */
+       size_t *unc_len[LZO_THREADS];             /* uncompressed lengths */
+       unsigned char *unc[LZO_THREADS];          /* uncompressed data */
+};
+
+/**
+ * CRC32 update function that runs in its own thread.
+ */
+static int crc32_threadfn(void *data)
+{
+       struct crc_data *d = data;
+       unsigned i;
+
+       while (1) {
+               wait_event(d->go, atomic_read(&d->ready) ||
+                                 kthread_should_stop());
+               if (kthread_should_stop()) {
+                       d->thr = NULL;
+                       atomic_set(&d->stop, 1);
+                       wake_up(&d->done);
+                       break;
+               }
+               atomic_set(&d->ready, 0);
+
+               for (i = 0; i < d->run_threads; i++)
+                       *d->crc32 = crc32_le(*d->crc32,
+                                            d->unc[i], *d->unc_len[i]);
+               atomic_set(&d->stop, 1);
+               wake_up(&d->done);
+       }
+       return 0;
+}
+/**
+ * Structure used for LZO data compression.
+ */
+struct cmp_data {
+       struct task_struct *thr;                  /* thread */
+       atomic_t ready;                           /* ready to start flag */
+       atomic_t stop;                            /* ready to stop flag */
+       int ret;                                  /* return code */
+       wait_queue_head_t go;                     /* start compression */
+       wait_queue_head_t done;                   /* compression done */
+       size_t unc_len;                           /* uncompressed length */
+       size_t cmp_len;                           /* compressed length */
+       unsigned char unc[LZO_UNC_SIZE];          /* uncompressed buffer */
+       unsigned char cmp[LZO_CMP_SIZE];          /* compressed buffer */
+       unsigned char wrk[LZO1X_1_MEM_COMPRESS];  /* compression workspace */
+};
+
+/**
+ * Compression function that runs in its own thread.
+ */
+static int lzo_compress_threadfn(void *data)
+{
+       struct cmp_data *d = data;
+
+       while (1) {
+               wait_event(d->go, atomic_read(&d->ready) ||
+                                 kthread_should_stop());
+               if (kthread_should_stop()) {
+                       d->thr = NULL;
+                       d->ret = -1;
+                       atomic_set(&d->stop, 1);
+                       wake_up(&d->done);
+                       break;
+               }
+               atomic_set(&d->ready, 0);
+
+               d->ret = lzo1x_1_compress(d->unc, d->unc_len,
+                                         d->cmp + LZO_HEADER, &d->cmp_len,
+                                         d->wrk);
+               atomic_set(&d->stop, 1);
+               wake_up(&d->done);
+       }
+       return 0;
+}
+
+/**
+ * save_image_lzo - Save the suspend image data compressed with LZO.
+ * @handle: Swap map handle to use for saving the image.
+ * @snapshot: Image to read data from.
+ * @nr_to_write: Number of pages to save.
+ */
+static int save_image_lzo(struct swap_map_handle *handle,
+                          struct snapshot_handle *snapshot,
+                          unsigned int nr_to_write)
+{
+       unsigned int m;
+       int ret = 0;
+       int nr_pages;
+       int err2;
+       struct bio *bio;
+       ktime_t start;
+       ktime_t stop;
+       size_t off;
+       unsigned thr, run_threads, nr_threads;
+       unsigned char *page = NULL;
+       struct cmp_data *data = NULL;
+       struct crc_data *crc = NULL;
+
+       /*
+        * We'll limit the number of threads for compression to limit memory
+        * footprint.
+        */
+       nr_threads = num_online_cpus() - 1;
+       nr_threads = clamp_val(nr_threads, 1, LZO_THREADS);
+
+       page = (void *)__get_free_page(__GFP_WAIT | __GFP_HIGH);
+       if (!page) {
+               printk(KERN_ERR "PM: Failed to allocate LZO page\n");
+               ret = -ENOMEM;
+               goto out_clean;
+       }
+
+       data = vmalloc(sizeof(*data) * nr_threads);
+       if (!data) {
+               printk(KERN_ERR "PM: Failed to allocate LZO data\n");
+               ret = -ENOMEM;
+               goto out_clean;
+       }
+       for (thr = 0; thr < nr_threads; thr++)
+               memset(&data[thr], 0, offsetof(struct cmp_data, go));
+
+       crc = kmalloc(sizeof(*crc), GFP_KERNEL);
+       if (!crc) {
+               printk(KERN_ERR "PM: Failed to allocate crc\n");
+               ret = -ENOMEM;
+               goto out_clean;
+       }
+       memset(crc, 0, offsetof(struct crc_data, go));
+
+       /*
+        * Start the compression threads.
+        */
+       for (thr = 0; thr < nr_threads; thr++) {
+               init_waitqueue_head(&data[thr].go);
+               init_waitqueue_head(&data[thr].done);
+
+               data[thr].thr = kthread_run(lzo_compress_threadfn,
+                                           &data[thr],
+                                           "image_compress/%u", thr);
+               if (IS_ERR(data[thr].thr)) {
+                       data[thr].thr = NULL;
+                       printk(KERN_ERR
+                              "PM: Cannot start compression threads\n");
+                       ret = -ENOMEM;
+                       goto out_clean;
+               }
+       }
+
+       /*
+        * Start the CRC32 thread.
+        */
+       init_waitqueue_head(&crc->go);
+       init_waitqueue_head(&crc->done);
+
+       handle->crc32 = 0;
+       crc->crc32 = &handle->crc32;
+       for (thr = 0; thr < nr_threads; thr++) {
+               crc->unc[thr] = data[thr].unc;
+               crc->unc_len[thr] = &data[thr].unc_len;
+       }
+
+       crc->thr = kthread_run(crc32_threadfn, crc, "image_crc32");
+       if (IS_ERR(crc->thr)) {
+               crc->thr = NULL;
+               printk(KERN_ERR "PM: Cannot start CRC32 thread\n");
+               ret = -ENOMEM;
+               goto out_clean;
+       }
+
+       /*
+        * Adjust the number of required free pages after all allocations have
+        * been done. We don't want to run out of pages when writing.
+        */
+       handle->reqd_free_pages = reqd_free_pages();
+
+       printk(KERN_INFO
+               "PM: Using %u thread(s) for compression.\n"
+               "PM: Compressing and saving image data (%u pages)...\n",
+               nr_threads, nr_to_write);
+       m = nr_to_write / 10;
+       if (!m)
+               m = 1;
+       nr_pages = 0;
+       bio = NULL;
+       start = ktime_get();
+       for (;;) {
+               for (thr = 0; thr < nr_threads; thr++) {
+                       for (off = 0; off < LZO_UNC_SIZE; off += PAGE_SIZE) {
+                               ret = snapshot_read_next(snapshot);
+                               if (ret < 0)
+                                       goto out_finish;
+
+                               if (!ret)
+                                       break;
+
+                               memcpy(data[thr].unc + off,
+                                      data_of(*snapshot), PAGE_SIZE);
+
+                               if (!(nr_pages % m))
+                                       printk(KERN_INFO
+                                              "PM: Image saving progress: "
+                                              "%3d%%\n",
+                                              nr_pages / m * 10);
+                               nr_pages++;
+                       }
+                       if (!off)
+                               break;
+
+                       data[thr].unc_len = off;
+
+                       atomic_set(&data[thr].ready, 1);
+                       wake_up(&data[thr].go);
+               }
+
+               if (!thr)
+                       break;
+
+               crc->run_threads = thr;
+               atomic_set(&crc->ready, 1);
+               wake_up(&crc->go);
+
+               for (run_threads = thr, thr = 0; thr < run_threads; thr++) {
+                       wait_event(data[thr].done,
+                                  atomic_read(&data[thr].stop));
+                       atomic_set(&data[thr].stop, 0);
+
+                       ret = data[thr].ret;
+
+                       if (ret < 0) {
+                               printk(KERN_ERR "PM: LZO compression failed\n");
+                               goto out_finish;
+                       }
+
+                       if (unlikely(!data[thr].cmp_len ||
+                                    data[thr].cmp_len >
+                                    lzo1x_worst_compress(data[thr].unc_len))) {
+                               printk(KERN_ERR
+                                      "PM: Invalid LZO compressed length\n");
+                               ret = -1;
+                               goto out_finish;
+                       }
+
+                       *(size_t *)data[thr].cmp = data[thr].cmp_len;
+
+                       /*
+                        * Given we are writing one page at a time to disk, we
+                        * copy that much from the buffer, although the last
+                        * bit will likely be smaller than full page. This is
+                        * OK - we saved the length of the compressed data, so
+                        * any garbage at the end will be discarded when we
+                        * read it.
+                        */
+                       for (off = 0;
+                            off < LZO_HEADER + data[thr].cmp_len;
+                            off += PAGE_SIZE) {
+                               memcpy(page, data[thr].cmp + off, PAGE_SIZE);
+
+                               ret = swap_write_page(handle, page, &bio);
+                               if (ret)
+                                       goto out_finish;
+                       }
+               }
+
+               wait_event(crc->done, atomic_read(&crc->stop));
+               atomic_set(&crc->stop, 0);
+       }
+
+out_finish:
+       err2 = hib_wait_on_bio_chain(&bio);
+       stop = ktime_get();
+       if (!ret)
+               ret = err2;
+       if (!ret)
+               printk(KERN_INFO "PM: Image saving done.\n");
+       swsusp_show_speed(start, stop, nr_to_write, "Wrote");
+out_clean:
+       if (crc) {
+               if (crc->thr)
+                       kthread_stop(crc->thr);
+               kfree(crc);
+       }
+       if (data) {
+               for (thr = 0; thr < nr_threads; thr++)
+                       if (data[thr].thr)
+                               kthread_stop(data[thr].thr);
+               vfree(data);
+       }
+       if (page) free_page((unsigned long)page);
+
+       return ret;
+}
+
+/**
+ *     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 from the resume partition.
+ */
+
+static int enough_swap(unsigned int nr_pages, unsigned int flags)
+{
+       unsigned int free_swap = count_swap_pages(root_swap, 1);
+       unsigned int required;
+
+       pr_debug("PM: Free swap pages: %u\n", free_swap);
+
+       required = PAGES_FOR_IO + nr_pages;
+       return free_swap > required;
+}
+
+/**
+ *     swsusp_write - Write entire image and metadata.
+ *     @flags: flags to pass to the "boot" kernel in the image header
+ *
+ *     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(unsigned int flags)
+{
+       struct swap_map_handle handle;
+       struct snapshot_handle snapshot;
+       struct swsusp_info *header;
+       unsigned long pages;
+       int error;
+
+       pages = snapshot_get_image_size();
+       error = get_swap_writer(&handle);
+       if (error) {
+               printk(KERN_ERR "PM: Cannot get swap writer\n");
+               return error;
+       }
+       if (flags & SF_NOCOMPRESS_MODE) {
+               if (!enough_swap(pages, flags)) {
+                       printk(KERN_ERR "PM: Not enough free swap\n");
+                       error = -ENOSPC;
+                       goto out_finish;
+               }
+       }
+       memset(&snapshot, 0, sizeof(struct snapshot_handle));
+       error = snapshot_read_next(&snapshot);
+       if (error < PAGE_SIZE) {
+               if (error >= 0)
+                       error = -EFAULT;
+
+               goto out_finish;
+       }
+       header = (struct swsusp_info *)data_of(snapshot);
+       error = swap_write_page(&handle, header, NULL);
+       if (!error) {
+               error = (flags & SF_NOCOMPRESS_MODE) ?
+                       save_image(&handle, &snapshot, pages - 1) :
+                       save_image_lzo(&handle, &snapshot, pages - 1);
+       }
+out_finish:
+       error = swap_writer_finish(&handle, flags, error);
+       return error;
+}
+
+/**
+ *     The following functions allow us to read data using a swap map
+ *     in a file-alike way
+ */
+
+static void release_swap_reader(struct swap_map_handle *handle)
+{
+       struct swap_map_page_list *tmp;
+
+       while (handle->maps) {
+               if (handle->maps->map)
+                       free_page((unsigned long)handle->maps->map);
+               tmp = handle->maps;
+               handle->maps = handle->maps->next;
+               kfree(tmp);
+       }
+       handle->cur = NULL;
+}
+
+static int get_swap_reader(struct swap_map_handle *handle,
+               unsigned int *flags_p)
+{
+       int error;
+       struct swap_map_page_list *tmp, *last;
+       sector_t offset;
+
+       *flags_p = swsusp_header->flags;
+
+       if (!swsusp_header->image) /* how can this happen? */
+               return -EINVAL;
+
+       handle->cur = NULL;
+       last = handle->maps = NULL;
+       offset = swsusp_header->image;
+       while (offset) {
+               tmp = kmalloc(sizeof(*handle->maps), GFP_KERNEL);
+               if (!tmp) {
+                       release_swap_reader(handle);
+                       return -ENOMEM;
+               }
+               memset(tmp, 0, sizeof(*tmp));
+               if (!handle->maps)
+                       handle->maps = tmp;
+               if (last)
+                       last->next = tmp;
+               last = tmp;
+
+               tmp->map = (struct swap_map_page *)
+                          __get_free_page(__GFP_WAIT | __GFP_HIGH);
+               if (!tmp->map) {
+                       release_swap_reader(handle);
+                       return -ENOMEM;
+               }
+
+               error = hib_bio_read_page(offset, tmp->map, NULL);
+               if (error) {
+                       release_swap_reader(handle);
+                       return error;
+               }
+               offset = tmp->map->next_swap;
+       }
+       handle->k = 0;
+       handle->cur = handle->maps->map;
+       return 0;
+}
+
+static int swap_read_page(struct swap_map_handle *handle, void *buf,
+                               struct bio **bio_chain)
+{
+       sector_t offset;
+       int error;
+       struct swap_map_page_list *tmp;
+
+       if (!handle->cur)
+               return -EINVAL;
+       offset = handle->cur->entries[handle->k];
+       if (!offset)
+               return -EFAULT;
+       error = hib_bio_read_page(offset, buf, bio_chain);
+       if (error)
+               return error;
+       if (++handle->k >= MAP_PAGE_ENTRIES) {
+               handle->k = 0;
+               free_page((unsigned long)handle->maps->map);
+               tmp = handle->maps;
+               handle->maps = handle->maps->next;
+               kfree(tmp);
+               if (!handle->maps)
+                       release_swap_reader(handle);
+               else
+                       handle->cur = handle->maps->map;
+       }
+       return error;
+}
+
+static int swap_reader_finish(struct swap_map_handle *handle)
+{
+       release_swap_reader(handle);
+
+       return 0;
+}
+
+/**
+ *     load_image - load the image using the swap map handle
+ *     @handle and the snapshot handle @snapshot
+ *     (assume there are @nr_pages pages to load)
+ */
+
+static int load_image(struct swap_map_handle *handle,
+                      struct snapshot_handle *snapshot,
+                      unsigned int nr_to_read)
+{
+       unsigned int m;
+       int ret = 0;
+       ktime_t start;
+       ktime_t stop;
+       struct bio *bio;
+       int err2;
+       unsigned nr_pages;
+
+       printk(KERN_INFO "PM: Loading image data pages (%u pages)...\n",
+               nr_to_read);
+       m = nr_to_read / 10;
+       if (!m)
+               m = 1;
+       nr_pages = 0;
+       bio = NULL;
+       start = ktime_get();
+       for ( ; ; ) {
+               ret = snapshot_write_next(snapshot);
+               if (ret <= 0)
+                       break;
+               ret = swap_read_page(handle, data_of(*snapshot), &bio);
+               if (ret)
+                       break;
+               if (snapshot->sync_read)
+                       ret = hib_wait_on_bio_chain(&bio);
+               if (ret)
+                       break;
+               if (!(nr_pages % m))
+                       printk(KERN_INFO "PM: Image loading progress: %3d%%\n",
+                              nr_pages / m * 10);
+               nr_pages++;
+       }
+       err2 = hib_wait_on_bio_chain(&bio);
+       stop = ktime_get();
+       if (!ret)
+               ret = err2;
+       if (!ret) {
+               printk(KERN_INFO "PM: Image loading done.\n");
+               snapshot_write_finalize(snapshot);
+               if (!snapshot_image_loaded(snapshot))
+                       ret = -ENODATA;
+       }
+       swsusp_show_speed(start, stop, nr_to_read, "Read");
+       return ret;
+}
+
+/**
+ * Structure used for LZO data decompression.
+ */
+struct dec_data {
+       struct task_struct *thr;                  /* thread */
+       atomic_t ready;                           /* ready to start flag */
+       atomic_t stop;                            /* ready to stop flag */
+       int ret;                                  /* return code */
+       wait_queue_head_t go;                     /* start decompression */
+       wait_queue_head_t done;                   /* decompression done */
+       size_t unc_len;                           /* uncompressed length */
+       size_t cmp_len;                           /* compressed length */
+       unsigned char unc[LZO_UNC_SIZE];          /* uncompressed buffer */
+       unsigned char cmp[LZO_CMP_SIZE];          /* compressed buffer */
+};
+
+/**
+ * Deompression function that runs in its own thread.
+ */
+static int lzo_decompress_threadfn(void *data)
+{
+       struct dec_data *d = data;
+
+       while (1) {
+               wait_event(d->go, atomic_read(&d->ready) ||
+                                 kthread_should_stop());
+               if (kthread_should_stop()) {
+                       d->thr = NULL;
+                       d->ret = -1;
+                       atomic_set(&d->stop, 1);
+                       wake_up(&d->done);
+                       break;
+               }
+               atomic_set(&d->ready, 0);
+
+               d->unc_len = LZO_UNC_SIZE;
+               d->ret = lzo1x_decompress_safe(d->cmp + LZO_HEADER, d->cmp_len,
+                                              d->unc, &d->unc_len);
+               atomic_set(&d->stop, 1);
+               wake_up(&d->done);
+       }
+       return 0;
+}
+
+/**
+ * load_image_lzo - Load compressed image data and decompress them with LZO.
+ * @handle: Swap map handle to use for loading data.
+ * @snapshot: Image to copy uncompressed data into.
+ * @nr_to_read: Number of pages to load.
+ */
+static int load_image_lzo(struct swap_map_handle *handle,
+                          struct snapshot_handle *snapshot,
+                          unsigned int nr_to_read)
+{
+       unsigned int m;
+       int ret = 0;
+       int eof = 0;
+       struct bio *bio;
+       ktime_t start;
+       ktime_t stop;
+       unsigned nr_pages;
+       size_t off;
+       unsigned i, thr, run_threads, nr_threads;
+       unsigned ring = 0, pg = 0, ring_size = 0,
+                have = 0, want, need, asked = 0;
+       unsigned long read_pages = 0;
+       unsigned char **page = NULL;
+       struct dec_data *data = NULL;
+       struct crc_data *crc = NULL;
+
+       /*
+        * We'll limit the number of threads for decompression to limit memory
+        * footprint.
+        */
+       nr_threads = num_online_cpus() - 1;
+       nr_threads = clamp_val(nr_threads, 1, LZO_THREADS);
+
+       page = vmalloc(sizeof(*page) * LZO_MAX_RD_PAGES);
+       if (!page) {
+               printk(KERN_ERR "PM: Failed to allocate LZO page\n");
+               ret = -ENOMEM;
+               goto out_clean;
+       }
+
+       data = vmalloc(sizeof(*data) * nr_threads);
+       if (!data) {
+               printk(KERN_ERR "PM: Failed to allocate LZO data\n");
+               ret = -ENOMEM;
+               goto out_clean;
+       }
+       for (thr = 0; thr < nr_threads; thr++)
+               memset(&data[thr], 0, offsetof(struct dec_data, go));
+
+       crc = kmalloc(sizeof(*crc), GFP_KERNEL);
+       if (!crc) {
+               printk(KERN_ERR "PM: Failed to allocate crc\n");
+               ret = -ENOMEM;
+               goto out_clean;
+       }
+       memset(crc, 0, offsetof(struct crc_data, go));
+
+       /*
+        * Start the decompression threads.
+        */
+       for (thr = 0; thr < nr_threads; thr++) {
+               init_waitqueue_head(&data[thr].go);
+               init_waitqueue_head(&data[thr].done);
+
+               data[thr].thr = kthread_run(lzo_decompress_threadfn,
+                                           &data[thr],
+                                           "image_decompress/%u", thr);
+               if (IS_ERR(data[thr].thr)) {
+                       data[thr].thr = NULL;
+                       printk(KERN_ERR
+                              "PM: Cannot start decompression threads\n");
+                       ret = -ENOMEM;
+                       goto out_clean;
+               }
+       }
+
+       /*
+        * Start the CRC32 thread.
+        */
+       init_waitqueue_head(&crc->go);
+       init_waitqueue_head(&crc->done);
+
+       handle->crc32 = 0;
+       crc->crc32 = &handle->crc32;
+       for (thr = 0; thr < nr_threads; thr++) {
+               crc->unc[thr] = data[thr].unc;
+               crc->unc_len[thr] = &data[thr].unc_len;
+       }
+
+       crc->thr = kthread_run(crc32_threadfn, crc, "image_crc32");
+       if (IS_ERR(crc->thr)) {
+               crc->thr = NULL;
+               printk(KERN_ERR "PM: Cannot start CRC32 thread\n");
+               ret = -ENOMEM;
+               goto out_clean;
+       }
+
+       /*
+        * Set the number of pages for read buffering.
+        * This is complete guesswork, because we'll only know the real
+        * picture once prepare_image() is called, which is much later on
+        * during the image load phase. We'll assume the worst case and
+        * say that none of the image pages are from high memory.
+        */
+       if (low_free_pages() > snapshot_get_image_size())
+               read_pages = (low_free_pages() - snapshot_get_image_size()) / 2;
+       read_pages = clamp_val(read_pages, LZO_MIN_RD_PAGES, LZO_MAX_RD_PAGES);
+
+       for (i = 0; i < read_pages; i++) {
+               page[i] = (void *)__get_free_page(i < LZO_CMP_PAGES ?
+                                                 __GFP_WAIT | __GFP_HIGH :
+                                                 __GFP_WAIT | __GFP_NOWARN |
+                                                 __GFP_NORETRY);
+
+               if (!page[i]) {
+                       if (i < LZO_CMP_PAGES) {
+                               ring_size = i;
+                               printk(KERN_ERR
+                                      "PM: Failed to allocate LZO pages\n");
+                               ret = -ENOMEM;
+                               goto out_clean;
+                       } else {
+                               break;
+                       }
+               }
+       }
+       want = ring_size = i;
+
+       printk(KERN_INFO
+               "PM: Using %u thread(s) for decompression.\n"
+               "PM: Loading and decompressing image data (%u pages)...\n",
+               nr_threads, nr_to_read);
+       m = nr_to_read / 10;
+       if (!m)
+               m = 1;
+       nr_pages = 0;
+       bio = NULL;
+       start = ktime_get();
+
+       ret = snapshot_write_next(snapshot);
+       if (ret <= 0)
+               goto out_finish;
+
+       for(;;) {
+               for (i = 0; !eof && i < want; i++) {
+                       ret = swap_read_page(handle, page[ring], &bio);
+                       if (ret) {
+                               /*
+                                * On real read error, finish. On end of data,
+                                * set EOF flag and just exit the read loop.
+                                */
+                               if (handle->cur &&
+                                   handle->cur->entries[handle->k]) {
+                                       goto out_finish;
+                               } else {
+                                       eof = 1;
+                                       break;
+                               }
+                       }
+                       if (++ring >= ring_size)
+                               ring = 0;
+               }
+               asked += i;
+               want -= i;
+
+               /*
+                * We are out of data, wait for some more.
+                */
+               if (!have) {
+                       if (!asked)
+                               break;
+
+                       ret = hib_wait_on_bio_chain(&bio);
+                       if (ret)
+                               goto out_finish;
+                       have += asked;
+                       asked = 0;
+                       if (eof)
+                               eof = 2;
+               }
+
+               if (crc->run_threads) {
+                       wait_event(crc->done, atomic_read(&crc->stop));
+                       atomic_set(&crc->stop, 0);
+                       crc->run_threads = 0;
+               }
+
+               for (thr = 0; have && thr < nr_threads; thr++) {
+                       data[thr].cmp_len = *(size_t *)page[pg];
+                       if (unlikely(!data[thr].cmp_len ||
+                                    data[thr].cmp_len >
+                                    lzo1x_worst_compress(LZO_UNC_SIZE))) {
+                               printk(KERN_ERR
+                                      "PM: Invalid LZO compressed length\n");
+                               ret = -1;
+                               goto out_finish;
+                       }
+
+                       need = DIV_ROUND_UP(data[thr].cmp_len + LZO_HEADER,
+                                           PAGE_SIZE);
+                       if (need > have) {
+                               if (eof > 1) {
+                                       ret = -1;
+                                       goto out_finish;
+                               }
+                               break;
+                       }
+
+                       for (off = 0;
+                            off < LZO_HEADER + data[thr].cmp_len;
+                            off += PAGE_SIZE) {
+                               memcpy(data[thr].cmp + off,
+                                      page[pg], PAGE_SIZE);
+                               have--;
+                               want++;
+                               if (++pg >= ring_size)
+                                       pg = 0;
+                       }
+
+                       atomic_set(&data[thr].ready, 1);
+                       wake_up(&data[thr].go);
+               }
+
+               /*
+                * Wait for more data while we are decompressing.
+                */
+               if (have < LZO_CMP_PAGES && asked) {
+                       ret = hib_wait_on_bio_chain(&bio);
+                       if (ret)
+                               goto out_finish;
+                       have += asked;
+                       asked = 0;
+                       if (eof)
+                               eof = 2;
+               }
+
+               for (run_threads = thr, thr = 0; thr < run_threads; thr++) {
+                       wait_event(data[thr].done,
+                                  atomic_read(&data[thr].stop));
+                       atomic_set(&data[thr].stop, 0);
+
+                       ret = data[thr].ret;
+
+                       if (ret < 0) {
+                               printk(KERN_ERR
+                                      "PM: LZO decompression failed\n");
+                               goto out_finish;
+                       }
+
+                       if (unlikely(!data[thr].unc_len ||
+                                    data[thr].unc_len > LZO_UNC_SIZE ||
+                                    data[thr].unc_len & (PAGE_SIZE - 1))) {
+                               printk(KERN_ERR
+                                      "PM: Invalid LZO uncompressed length\n");
+                               ret = -1;
+                               goto out_finish;
+                       }
+
+                       for (off = 0;
+                            off < data[thr].unc_len; off += PAGE_SIZE) {
+                               memcpy(data_of(*snapshot),
+                                      data[thr].unc + off, PAGE_SIZE);
+
+                               if (!(nr_pages % m))
+                                       printk(KERN_INFO
+                                              "PM: Image loading progress: "
+                                              "%3d%%\n",
+                                              nr_pages / m * 10);
+                               nr_pages++;
+
+                               ret = snapshot_write_next(snapshot);
+                               if (ret <= 0) {
+                                       crc->run_threads = thr + 1;
+                                       atomic_set(&crc->ready, 1);
+                                       wake_up(&crc->go);
+                                       goto out_finish;
+                               }
+                       }
+               }
+
+               crc->run_threads = thr;
+               atomic_set(&crc->ready, 1);
+               wake_up(&crc->go);
+       }
+
+out_finish:
+       if (crc->run_threads) {
+               wait_event(crc->done, atomic_read(&crc->stop));
+               atomic_set(&crc->stop, 0);
+       }
+       stop = ktime_get();
+       if (!ret) {
+               printk(KERN_INFO "PM: Image loading done.\n");
+               snapshot_write_finalize(snapshot);
+               if (!snapshot_image_loaded(snapshot))
+                       ret = -ENODATA;
+               if (!ret) {
+                       if (swsusp_header->flags & SF_CRC32_MODE) {
+                               if(handle->crc32 != swsusp_header->crc32) {
+                                       printk(KERN_ERR
+                                              "PM: Invalid image CRC32!\n");
+                                       ret = -ENODATA;
+                               }
+                       }
+               }
+       }
+       swsusp_show_speed(start, stop, nr_to_read, "Read");
+out_clean:
+       for (i = 0; i < ring_size; i++)
+               free_page((unsigned long)page[i]);
+       if (crc) {
+               if (crc->thr)
+                       kthread_stop(crc->thr);
+               kfree(crc);
+       }
+       if (data) {
+               for (thr = 0; thr < nr_threads; thr++)
+                       if (data[thr].thr)
+                               kthread_stop(data[thr].thr);
+               vfree(data);
+       }
+       vfree(page);
+
+       return ret;
+}
+
+/**
+ *     swsusp_read - read the hibernation image.
+ *     @flags_p: flags passed by the "frozen" kernel in the image header should
+ *               be written into this memory location
+ */
+
+int swsusp_read(unsigned int *flags_p)
+{
+       int error;
+       struct swap_map_handle handle;
+       struct snapshot_handle snapshot;
+       struct swsusp_info *header;
+
+       memset(&snapshot, 0, sizeof(struct snapshot_handle));
+       error = snapshot_write_next(&snapshot);
+       if (error < PAGE_SIZE)
+               return error < 0 ? error : -EFAULT;
+       header = (struct swsusp_info *)data_of(snapshot);
+       error = get_swap_reader(&handle, flags_p);
+       if (error)
+               goto end;
+       if (!error)
+               error = swap_read_page(&handle, header, NULL);
+       if (!error) {
+               error = (*flags_p & SF_NOCOMPRESS_MODE) ?
+                       load_image(&handle, &snapshot, header->pages - 1) :
+                       load_image_lzo(&handle, &snapshot, header->pages - 1);
+       }
+       swap_reader_finish(&handle);
+end:
+       if (!error)
+               pr_debug("PM: Image successfully loaded\n");
+       else
+               pr_debug("PM: Error %d resuming\n", error);
+       return error;
+}
+
+/**
+ *      swsusp_check - Check for swsusp signature in the resume device
+ */
+
+int swsusp_check(void)
+{
+       int error;
+
+       hib_resume_bdev = blkdev_get_by_dev(swsusp_resume_device,
+                                           FMODE_READ, NULL);
+       if (!IS_ERR(hib_resume_bdev)) {
+               set_blocksize(hib_resume_bdev, PAGE_SIZE);
+               clear_page(swsusp_header);
+               error = hib_bio_read_page(swsusp_resume_block,
+                                       swsusp_header, NULL);
+               if (error)
+                       goto put;
+
+               if (!memcmp(HIBERNATE_SIG, swsusp_header->sig, 10)) {
+                       memcpy(swsusp_header->sig, swsusp_header->orig_sig, 10);
+                       /* Reset swap signature now */
+                       error = hib_bio_write_page(swsusp_resume_block,
+                                               swsusp_header, NULL);
+               } else {
+                       error = -EINVAL;
+               }
+
+put:
+               if (error)
+                       blkdev_put(hib_resume_bdev, FMODE_READ);
+               else
+                       pr_debug("PM: Image signature found, resuming\n");
+       } else {
+               error = PTR_ERR(hib_resume_bdev);
+       }
+
+       if (error)
+               pr_debug("PM: Image not found (code %d)\n", error);
+
+       return error;
+}
+
+/**
+ *     swsusp_close - close swap device.
+ */
+
+void swsusp_close(fmode_t mode)
+{
+       if (IS_ERR(hib_resume_bdev)) {
+               pr_debug("PM: Image device not initialised\n");
+               return;
+       }
+
+       blkdev_put(hib_resume_bdev, mode);
+}
+
+/**
+ *      swsusp_unmark - Unmark swsusp signature in the resume device
+ */
+
+#ifdef CONFIG_SUSPEND
+int swsusp_unmark(void)
+{
+       int error;
+
+       hib_bio_read_page(swsusp_resume_block, swsusp_header, NULL);
+       if (!memcmp(HIBERNATE_SIG,swsusp_header->sig, 10)) {
+               memcpy(swsusp_header->sig,swsusp_header->orig_sig, 10);
+               error = hib_bio_write_page(swsusp_resume_block,
+                                       swsusp_header, NULL);
+       } else {
+               printk(KERN_ERR "PM: Cannot find swsusp signature!\n");
+               error = -ENODEV;
+       }
+
+       /*
+        * We just returned from suspend, we don't need the image any more.
+        */
+       free_all_swap_pages(root_swap);
+
+       return error;
+}
+#endif
+
+static int swsusp_header_init(void)
+{
+       swsusp_header = (struct swsusp_header*) __get_free_page(GFP_KERNEL);
+       if (!swsusp_header)
+               panic("Could not allocate memory for swsusp_header\n");
+       return 0;
+}
+
+core_initcall(swsusp_header_init);