cifs: don't cap ra_pages at the same level as default_backing_dev_info
authorJeff Layton <jlayton@redhat.com>
Tue, 1 May 2012 21:41:49 +0000 (17:41 -0400)
committerSteve French <sfrench@us.ibm.com>
Wed, 2 May 2012 03:27:54 +0000 (22:27 -0500)
While testing, I've found that even when we are able to negotiate a
much larger rsize with the server, on-the-wire reads often end up being
capped at 128k because of ra_pages being capped at that level.

Lifting this restriction gave almost a twofold increase in sequential
read performance on my craptactular KVM test rig with a 1M rsize.

I think this is safe since the actual ra_pages that the VM requests
is run through max_sane_readahead() prior to submitting the I/O. Under
memory pressure we should end up with large readahead requests being
suppressed anyway.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
fs/cifs/connect.c

index f4d381e..6ce888b 100644 (file)
@@ -3614,22 +3614,6 @@ cifs_get_volume_info(char *mount_data, const char *devname)
        return volume_info;
 }
 
-/* make sure ra_pages is a multiple of rsize */
-static inline unsigned int
-cifs_ra_pages(struct cifs_sb_info *cifs_sb)
-{
-       unsigned int reads;
-       unsigned int rsize_pages = cifs_sb->rsize / PAGE_CACHE_SIZE;
-
-       if (rsize_pages >= default_backing_dev_info.ra_pages)
-               return default_backing_dev_info.ra_pages;
-       else if (rsize_pages == 0)
-               return rsize_pages;
-
-       reads = default_backing_dev_info.ra_pages / rsize_pages;
-       return reads * rsize_pages;
-}
-
 int
 cifs_mount(struct cifs_sb_info *cifs_sb, struct smb_vol *volume_info)
 {
@@ -3717,7 +3701,7 @@ try_mount_again:
        cifs_sb->rsize = cifs_negotiate_rsize(tcon, volume_info);
 
        /* tune readahead according to rsize */
-       cifs_sb->bdi.ra_pages = cifs_ra_pages(cifs_sb);
+       cifs_sb->bdi.ra_pages = cifs_sb->rsize / PAGE_CACHE_SIZE;
 
 remote_path_check:
 #ifdef CONFIG_CIFS_DFS_UPCALL