[PATCH 1/2 V1] [PATCH] fs/block-dev.c:fix performance regression in O_DIRECT writes to md block devices

Subsystems: filesystems (vfs and infrastructure), the rest

STALE5116d

5 messages, 3 authors, 2012-08-02 · open the first message on its own page

[PATCH 1/2 V1] [PATCH] fs/block-dev.c:fix performance regression in O_DIRECT writes to md block devices

From: majianpeng <hidden>
Date: 2012-07-16 06:21:03

For regular file, write operaion used blk_plug function.But for block
file,write operation did not use blk_plug.
This patch is also for write-cache mode for block-device.

Signed-off-by: Jianpeng Ma <redacted>
---
 fs/block_dev.c |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)

diff --git a/fs/block_dev.c b/fs/block_dev.c
index c2bbe1f..cf10778 100644
--- a/fs/block_dev.c
+++ b/fs/block_dev.c

@@ -1579,9 +1579,11 @@ ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
 {
 	struct file *file = iocb->ki_filp;
 	ssize_t ret;
+	struct blk_plug plug;
 
 	BUG_ON(iocb->ki_pos != pos);
 
+	blk_start_plug(&plug);
 	ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
 	if (ret > 0 || ret == -EIOCBQUEUED) {
 		ssize_t err;

@@ -1590,6 +1592,7 @@ ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
 		if (err < 0 && ret > 0)
 			ret = err;
 	}
+	blk_finish_plug(&plug);
 	return ret;
 }
 EXPORT_SYMBOL_GPL(blkdev_aio_write);

-- 
1.7.5.4

Re: [PATCH 1/2 V1] [PATCH] fs/block-dev.c:fix performance regression in O_DIRECT writes to md block devices

From: NeilBrown <hidden>
Date: 2012-07-17 03:04:31

On Mon, 16 Jul 2012 14:22:03 +0800 majianpeng [off-list ref] wrote:

For regular file, write operaion used blk_plug function.But for block
file,write operation did not use blk_plug.
This patch is also for write-cache mode for block-device.

Signed-off-by: Jianpeng Ma <redacted>

Reviewed-by: NeilBrown <redacted>

Will you take this Jens?  Though mail to you seem to be bouncing:

[off-list ref]: host mail.fusionio.com[10.101.1.19] said: 554 5.4.6 Hop
    count exceeded - possible mail loop (in reply to end of DATA command)

so I'm not sure you'll even see this :-(

NeilBrown

quoted hunk

---
 fs/block_dev.c |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)

diff --git a/fs/block_dev.c b/fs/block_dev.c
index c2bbe1f..cf10778 100644
--- a/fs/block_dev.c
+++ b/fs/block_dev.c

@@ -1579,9 +1579,11 @@ ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
 {
 	struct file *file = iocb->ki_filp;
 	ssize_t ret;
+	struct blk_plug plug;
 
 	BUG_ON(iocb->ki_pos != pos);
 
+	blk_start_plug(&plug);
 	ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
 	if (ret > 0 || ret == -EIOCBQUEUED) {
 		ssize_t err;

@@ -1590,6 +1592,7 @@ ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
 		if (err < 0 && ret > 0)
 			ret = err;
 	}
+	blk_finish_plug(&plug);
 	return ret;
 }
 EXPORT_SYMBOL_GPL(blkdev_aio_write);

Re: Re: [PATCH 1/2 V1] [PATCH] fs/block-dev.c:fix performance regression in O_DIRECT writes to md block devices

From: majianpeng <hidden>
Date: 2012-07-17 05:07:50

On 2012-07-17 11:04 NeilBrown [off-list ref] Wrote:

On Mon, 16 Jul 2012 14:22:03 +0800 majianpeng [off-list ref] wrote:

quoted

For regular file, write operaion used blk_plug function.But for block
file,write operation did not use blk_plug.
This patch is also for write-cache mode for block-device.

Signed-off-by: Jianpeng Ma <redacted>

Reviewed-by: NeilBrown <redacted>

Will you take this Jens?  Though mail to you seem to be bouncing:

[off-list ref]: host mail.fusionio.com[10.101.1.19] said: 554 5.4.6 Hop
   count exceeded - possible mail loop (in reply to end of DATA command)

so I'm not sure you'll even see this :-(

I see it.

NeilBrown

quoted

---
 fs/block_dev.c |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)

diff --git a/fs/block_dev.c b/fs/block_dev.c
index c2bbe1f..cf10778 100644
--- a/fs/block_dev.c
+++ b/fs/block_dev.c

@@ -1579,9 +1579,11 @@ ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
 {
 	struct file *file = iocb->ki_filp;
 	ssize_t ret;
+	struct blk_plug plug;
 
 	BUG_ON(iocb->ki_pos != pos);
 
+	blk_start_plug(&plug);
 	ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
 	if (ret > 0 || ret == -EIOCBQUEUED) {
 		ssize_t err;

@@ -1590,6 +1592,7 @@ ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
 		if (err < 0 && ret > 0)
 			ret = err;
 	}
+	blk_finish_plug(&plug);
 	return ret;
 }
 EXPORT_SYMBOL_GPL(blkdev_aio_write);

Re: Re: [PATCH 1/2 V1] [PATCH] fs/block-dev.c:fix performance regression in O_DIRECT writes to md block devices

From: majianpeng <hidden>
Date: 2012-08-02 06:48:41

On 2012-07-17 11:04 NeilBrown [off-list ref] Wrote:

On Mon, 16 Jul 2012 14:22:03 +0800 majianpeng [off-list ref] wrote:

quoted

For regular file, write operaion used blk_plug function.But for block
file,write operation did not use blk_plug.
This patch is also for write-cache mode for block-device.

Signed-off-by: Jianpeng Ma <redacted>

Reviewed-by: NeilBrown <redacted>

Will you take this Jens?  Though mail to you seem to be bouncing:

[off-list ref]: host mail.fusionio.com[10.101.1.19] said: 554 5.4.6 Hop
   count exceeded - possible mail loop (in reply to end of DATA command)

so I'm not sure you'll even see this :-(

NeilBrown

quoted

---
 fs/block_dev.c |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)

diff --git a/fs/block_dev.c b/fs/block_dev.c
index c2bbe1f..cf10778 100644
--- a/fs/block_dev.c
+++ b/fs/block_dev.c

@@ -1579,9 +1579,11 @@ ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
 {
 	struct file *file = iocb->ki_filp;
 	ssize_t ret;
+	struct blk_plug plug;
 
 	BUG_ON(iocb->ki_pos != pos);
 
+	blk_start_plug(&plug);
 	ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
 	if (ret > 0 || ret == -EIOCBQUEUED) {
 		ssize_t err;

@@ -1590,6 +1592,7 @@ ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
 		if (err < 0 && ret > 0)
 			ret = err;
 	}
+	blk_finish_plug(&plug);
 	return ret;
 }
 EXPORT_SYMBOL_GPL(blkdev_aio_write);

How about this patch? apply or reject? Thanks!

Re: [PATCH 1/2 V1] [PATCH] fs/block-dev.c:fix performance regression in O_DIRECT writes to md block devices

From: Jens Axboe <axboe@kernel.dk>
Date: 2012-08-02 07:51:40

On 08/02/2012 08:48 AM, majianpeng wrote:

On 2012-07-17 11:04 NeilBrown [off-list ref] Wrote:

quoted

On Mon, 16 Jul 2012 14:22:03 +0800 majianpeng [off-list ref] wrote:

quoted

For regular file, write operaion used blk_plug function.But for block
file,write operation did not use blk_plug.
This patch is also for write-cache mode for block-device.

Signed-off-by: Jianpeng Ma <redacted>

Reviewed-by: NeilBrown <redacted>

Will you take this Jens?  Though mail to you seem to be bouncing:

[off-list ref]: host mail.fusionio.com[10.101.1.19] said: 554 5.4.6 Hop
   count exceeded - possible mail loop (in reply to end of DATA command)

so I'm not sure you'll even see this :-(

NeilBrown

quoted

---
 fs/block_dev.c |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)

diff --git a/fs/block_dev.c b/fs/block_dev.c
index c2bbe1f..cf10778 100644
--- a/fs/block_dev.c
+++ b/fs/block_dev.c

@@ -1579,9 +1579,11 @@ ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
 {
 	struct file *file = iocb->ki_filp;
 	ssize_t ret;
+	struct blk_plug plug;
 
 	BUG_ON(iocb->ki_pos != pos);
 
+	blk_start_plug(&plug);
 	ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
 	if (ret > 0 || ret == -EIOCBQUEUED) {
 		ssize_t err;

@@ -1590,6 +1592,7 @@ ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
 		if (err < 0 && ret > 0)
 			ret = err;
 	}
+	blk_finish_plug(&plug);
 	return ret;
 }
 EXPORT_SYMBOL_GPL(blkdev_aio_write);

How about this patch? apply or reject? Thanks!

I've applied it for a bit of testing, then it'll got into 3.6. Thanks.

-- 
Jens Axboe

`h`	back out one level
`j`	next message in thread
`k`	previous message in thread
`l`	drill in
`Esc`	close help / fold thread tree
`?`	toggle this help