block_dev: implement readpages() to optimize sequential read
authorAkinobu Mita <akinobu.mita@gmail.com>
Thu, 9 Oct 2014 22:26:58 +0000 (15:26 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 10 Oct 2014 02:25:53 +0000 (22:25 -0400)
Sequential read from a block device is expected to be equal or faster than
from the file on a filesystem.  But it is not correct due to the lack of
effective readpages() in the address space operations for block device.

This implements readpages() operation for block device by using
mpage_readpages() which can create multipage BIOs instead of BIOs for each
page and reduce system CPU time consumption.

Install 1GB of RAM disk storage:

# modprobe scsi_debug dev_size_mb=1024 delay=0

Sequential read from file on a filesystem:

# mkfs.ext4 /dev/$DEV
# mount /dev/$DEV /mnt
# fio --name=t --size=512m --rw=read --filename=/mnt/file
...
  read : io=524288KB, bw=2133.4MB/s, iops=546133, runt=   240msec

Sequential read from a block device:
# fio --name=t --size=512m --rw=read --filename=/dev/$DEV
...
(Without this commit)
  read : io=524288KB, bw=1700.2MB/s, iops=435455, runt=   301msec

(With this commit)
  read : io=524288KB, bw=2160.4MB/s, iops=553046, runt=   237msec

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/block_dev.c

index 6d7274619bf916c2dcf0d7744ba8d888d948d711..e2f3ad0879ce9165f60d6e5c8cd4a6b4e523d87e 100644 (file)
@@ -304,6 +304,12 @@ static int blkdev_readpage(struct file * file, struct page * page)
        return block_read_full_page(page, blkdev_get_block);
 }
 
+static int blkdev_readpages(struct file *file, struct address_space *mapping,
+                       struct list_head *pages, unsigned nr_pages)
+{
+       return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
+}
+
 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
                        loff_t pos, unsigned len, unsigned flags,
                        struct page **pagep, void **fsdata)
@@ -1622,6 +1628,7 @@ static int blkdev_releasepage(struct page *page, gfp_t wait)
 
 static const struct address_space_operations def_blk_aops = {
        .readpage       = blkdev_readpage,
+       .readpages      = blkdev_readpages,
        .writepage      = blkdev_writepage,
        .write_begin    = blkdev_write_begin,
        .write_end      = blkdev_write_end,