Re: [PATCH 1/4] xfs: reduce context switches for synchronous buffered I/O

["Darrick J. Wong" <djwong@kernel.org>]

On Mon, Feb 24, 2025 at 07:11:35AM -0800, Christoph Hellwig wrote:
> Currently all metadata I/O completions happen in the m_buf_workqueue
> workqueue.  But for synchronous I/O (i.e. all buffer reads) there is no
> need for that, as there always is a called in process context that is
> waiting for the I/O.  Factor out the guts of xfs_buf_ioend into a
> separate helper and call it from xfs_buf_iowait to avoid a double
> an extra context switch to the workqueue.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  fs/xfs/xfs_buf.c | 43 +++++++++++++++++++++++++++----------------
>  1 file changed, 27 insertions(+), 16 deletions(-)
> 
> diff --git a/fs/xfs/xfs_buf.c b/fs/xfs/xfs_buf.c
> index 15bb790359f8..821aa85e2ce5 100644
> --- a/fs/xfs/xfs_buf.c
> +++ b/fs/xfs/xfs_buf.c
> @@ -1345,6 +1345,7 @@ xfs_buf_ioend_handle_error(
>  resubmit:
>  	xfs_buf_ioerror(bp, 0);
>  	bp->b_flags |= (XBF_DONE | XBF_WRITE_FAIL);
> +	reinit_completion(&bp->b_iowait);
>  	xfs_buf_submit(bp);
>  	return true;
>  out_stale:
> @@ -1355,8 +1356,9 @@ xfs_buf_ioend_handle_error(
>  	return false;
>  }
>  
> -static void
> -xfs_buf_ioend(
> +/* returns false if the caller needs to resubmit the I/O, else false */

"...else true" ?

--D

> +static bool
> +__xfs_buf_ioend(
>  	struct xfs_buf	*bp)
>  {
>  	trace_xfs_buf_iodone(bp, _RET_IP_);
> @@ -1376,7 +1378,7 @@ xfs_buf_ioend(
>  		}
>  
>  		if (unlikely(bp->b_error) && xfs_buf_ioend_handle_error(bp))
> -			return;
> +			return false;
>  
>  		/* clear the retry state */
>  		bp->b_last_error = 0;
> @@ -1397,7 +1399,15 @@ xfs_buf_ioend(
>  
>  	bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD |
>  			 _XBF_LOGRECOVERY);
> +	return true;
> +}
>  
> +static void
> +xfs_buf_ioend(
> +	struct xfs_buf	*bp)
> +{
> +	if (!__xfs_buf_ioend(bp))
> +		return;
>  	if (bp->b_flags & XBF_ASYNC)
>  		xfs_buf_relse(bp);
>  	else
> @@ -1411,15 +1421,8 @@ xfs_buf_ioend_work(
>  	struct xfs_buf		*bp =
>  		container_of(work, struct xfs_buf, b_ioend_work);
>  
> -	xfs_buf_ioend(bp);
> -}
> -
> -static void
> -xfs_buf_ioend_async(
> -	struct xfs_buf	*bp)
> -{
> -	INIT_WORK(&bp->b_ioend_work, xfs_buf_ioend_work);
> -	queue_work(bp->b_mount->m_buf_workqueue, &bp->b_ioend_work);
> +	if (__xfs_buf_ioend(bp))
> +		xfs_buf_relse(bp);
>  }
>  
>  void
> @@ -1491,7 +1494,13 @@ xfs_buf_bio_end_io(
>  		 XFS_TEST_ERROR(false, bp->b_mount, XFS_ERRTAG_BUF_IOERROR))
>  		xfs_buf_ioerror(bp, -EIO);
>  
> -	xfs_buf_ioend_async(bp);
> +	if (bp->b_flags & XBF_ASYNC) {
> +		INIT_WORK(&bp->b_ioend_work, xfs_buf_ioend_work);
> +		queue_work(bp->b_mount->m_buf_workqueue, &bp->b_ioend_work);
> +	} else {
> +		complete(&bp->b_iowait);
> +	}
> +
>  	bio_put(bio);
>  }
>  
> @@ -1568,9 +1577,11 @@ xfs_buf_iowait(
>  {
>  	ASSERT(!(bp->b_flags & XBF_ASYNC));
>  
> -	trace_xfs_buf_iowait(bp, _RET_IP_);
> -	wait_for_completion(&bp->b_iowait);
> -	trace_xfs_buf_iowait_done(bp, _RET_IP_);
> +	do {
> +		trace_xfs_buf_iowait(bp, _RET_IP_);
> +		wait_for_completion(&bp->b_iowait);
> +		trace_xfs_buf_iowait_done(bp, _RET_IP_);
> +	} while (!__xfs_buf_ioend(bp));
>  
>  	return bp->b_error;
>  }
> -- 
> 2.45.2
> 
>