Skip to contentSkip to conversation

Optimization for touch_atime

PATCH In review Andi Kleen 3 participants 5 messages Jul 6, 2009 – Jul 8, 2009 VFS

lore
Share & tools
lore

Andi Kleen

author
#0 lore raw

PATCH 1/2 Optimization for touch_atime

Some benchmark testing shows touch_atime to be high up in profile logs for IO intensive workloads. Most likely that's due to the lock in mnt_want_write(). Unfortunately touch_atime first takes the lock, and then does all the other tests that could avoid atime updates (like noatime or relatime).

Do it the other way round -- first try to avoid the update and only then if that didn't succeed take the lock. That works because none of the atime avoidance tests rely on locking.

This also eliminates a goto.

---
 fs/inode.c |   20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)
fs/inode.c +10 -10
Index: linux-2.6.30-ak/fs/inode.c
===================================================================
@@ -1361,31 +1361,31 @@ void touch_atime(struct vfsmount *mnt, s
 	struct inode *inode = dentry->d_inode;
 	struct timespec now;
 
-	if (mnt_want_write(mnt))
-		return;
 	if (inode->i_flags & S_NOATIME)
-		goto out;
+		return;
 	if (IS_NOATIME(inode))
-		goto out;
+		return;
 	if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
-		goto out;
+		return;
 
 	if (mnt->mnt_flags & MNT_NOATIME)
-		goto out;
+		return;
 	if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
-		goto out;
+		return;
 
 	now = current_fs_time(inode->i_sb);
 
 	if (!relatime_need_update(mnt, inode, now))
-		goto out;
+		return;
 
 	if (timespec_equal(&inode->i_atime, &now))
-		goto out;
+		return;
+
+	if (mnt_want_write(mnt))
+		return;
 
 	inode->i_atime = now;
 	mark_inode_dirty_sync(inode);
-out:
 	mnt_drop_write(mnt);
 }
 EXPORT_SYMBOL(touch_atime);
author
#1 lore raw

PATCH 2/2 Optimize touch_time too

Do a similar optimization as earlier for touch_atime. Getting the lock in mnt_get_write is relatively costly, so try all avenues to avoid it first.

This patch is careful to still only update inode fields inside the lock region.

This didn't show up in benchmarks, but it's easy enough to do.

---
 fs/inode.c |   43 +++++++++++++++++++++++--------------------
 1 file changed, 23 insertions(+), 20 deletions(-)
fs/inode.c +23 -20
Index: linux-2.6.31-rc1-ak/fs/inode.c
===================================================================
@@ -1431,34 +1431,37 @@ void file_update_time(struct file *file)
 {
 	struct inode *inode = file->f_path.dentry->d_inode;
 	struct timespec now;
-	int sync_it = 0;
-	int err;
+	enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
 
+	/* First try to exhause all avenues to not sync */
 	if (IS_NOCMTIME(inode))
 		return;
 
-	err = mnt_want_write_file(file);
-	if (err)
-		return;
-
 	now = current_fs_time(inode->i_sb);
-	if (!timespec_equal(&inode->i_mtime, &now)) {
-		inode->i_mtime = now;
-		sync_it = 1;
-	}
+	if (!timespec_equal(&inode->i_mtime, &now))
+		sync_it = S_MTIME;
 
-	if (!timespec_equal(&inode->i_ctime, &now)) {
-		inode->i_ctime = now;
-		sync_it = 1;
-	}
+	if (!timespec_equal(&inode->i_ctime, &now))
+		sync_it |= S_CTIME;
 
-	if (IS_I_VERSION(inode)) {
-		inode_inc_iversion(inode);
-		sync_it = 1;
-	}
+	if (IS_I_VERSION(inode))
+		sync_it |= S_VERSION;
+
+	if (!sync_it)
+		return;
+
+	/* Finally allowed to write? Takes lock. */
+	if (!mnt_want_write_file(file))
+		return;
 
-	if (sync_it)
-		mark_inode_dirty_sync(inode);
+	/* Only change inode inside the lock region */
+	if (sync_it & S_VERSION)
+		inode_inc_iversion(inode);
+	if (sync_it & S_CTIME)
+		inode->i_ctime = now;
+	if (sync_it & S_MTIME)
+		inode->i_mtime = now;
+	mark_inode_dirty_sync(inode);
 	mnt_drop_write(file->f_path.mnt);
 }
 EXPORT_SYMBOL(file_update_time);
#3 lore raw
In reply to Andi Kleen
Quoted 7 lines · Andi Kleen

Some benchmark testing shows touch_atime to be high up in profile logs for IO intensive workloads. Most likely that's due to the lock in mnt_want_write(). Unfortunately touch_atime first takes the lock, and then does all the other tests that could avoid atime updates (like noatime or relatime).

Quoted 7 lines

Do it the other way round -- first try to avoid the update and only then if that didn't succeed take the lock. That works because none of the atime avoidance tests rely on locking.

This also eliminates a goto.

Looks good to me.

#2 lore raw

Re: [PATCH] [2/2] Optimize touch_time too

In reply to Andi Kleen
Quoted 22 lines · Andi Kleen

Do a similar optimization as earlier for touch_atime. Getting the lock in mnt_get_write is relatively costly, so try all avenues to avoid it first.

This patch is careful to still only update inode fields inside the lock region.

This didn't show up in benchmarks, but it's easy enough to do.

---
 fs/inode.c |   43 +++++++++++++++++++++++--------------------
 1 file changed, 23 insertions(+), 20 deletions(-)

Index: linux-2.6.31-rc1-ak/fs/inode.c
===================================================================

fs/inode.c void file_update_time(struct file *file) · line 1434

Quoted 6 lines
fs/inode.c +1 -2
@@ -1431,34 +1431,37 @@ void file_update_time(struct file *file)
 {
 	struct inode *inode = file->f_path.dentry->d_inode;
 	struct timespec now;
-	int sync_it = 0;
-	int err;
+	enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;

Looks good, and makes sense to keep thise in sync with file_update_atime.

Reviewed-by on v1 1/2
#4 lore raw
In reply to Andi Kleen
Quoted 55 lines · Andi Kleen

Some benchmark testing shows touch_atime to be high up in profile logs for IO intensive workloads. Most likely that's due to the lock in mnt_want_write(). Unfortunately touch_atime first takes the lock, and then does all the other tests that could avoid atime updates (like noatime or relatime).

Do it the other way round -- first try to avoid the update and only then if that didn't succeed take the lock. That works because none of the atime avoidance tests rely on locking.

This also eliminates a goto.

---
 fs/inode.c |   20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)

Index: linux-2.6.30-ak/fs/inode.c
===================================================================

fs/inode.c +7 -9
@@ -1361,31 +1361,31 @@ void touch_atime(struct vfsmount *mnt, s
 	struct inode *inode = dentry->d_inode;
 	struct timespec now;
 
-	if (mnt_want_write(mnt))
-		return;
 	if (inode->i_flags & S_NOATIME)
-		goto out;
+		return;
 	if (IS_NOATIME(inode))
-		goto out;
+		return;
 	if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
-		goto out;
+		return;
 
 	if (mnt->mnt_flags & MNT_NOATIME)
-		goto out;
+		return;
 	if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
-		goto out;
+		return;
 
 	now = current_fs_time(inode->i_sb);
 
 	if (!relatime_need_update(mnt, inode, now))
-		goto out;
+		return;
 
 	if (timespec_equal(&inode->i_atime, &now))
-		goto out;
+		return;

fs/inode.c void touch_atime(struct vfsmount *mnt, s · line 1391

Quoted 10 lines
fs/inode.c +3 -1
+
+	if (mnt_want_write(mnt))
+		return;
 
 	inode->i_atime = now;
 	mark_inode_dirty_sync(inode);
-out:
 	mnt_drop_write(mnt);
 }
 EXPORT_SYMBOL(touch_atime);

Nice!

-VAL