On 20/11/2020 19:13, Jens Axboe wrote:

On 11/20/20 10:10 AM, Pavel Begunkov wrote:

quoted

io_uring's direct nowait requests end up waiting on io_schedule() in
sbitmap, that's seems to be so because blkdev_direct_IO() fails to
propagate IOCB_NOWAIT to a bio and hence to blk-mq.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---
 fs/block_dev.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/fs/block_dev.c b/fs/block_dev.c
index 9e84b1928b94..e7e860c78d93 100644
--- a/fs/block_dev.c
+++ b/fs/block_dev.c

@@ -263,6 +263,8 @@ __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
 		bio.bi_opf = dio_bio_write_op(iocb);
 		task_io_account_write(ret);
 	}
+	if (iocb->ki_flags & IOCB_NOWAIT)
+		bio.bi_opf |= REQ_NOWAIT;
 	if (iocb->ki_flags & IOCB_HIPRI)
 		bio_set_polled(&bio, iocb);

Was thinking this wasn't needed, but I guess that users could do sync && NOWAIT
and get -EAGAIN if using preadv2/pwritev2.

quoted

@@ -416,6 +418,8 @@ __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
 			bio->bi_opf = dio_bio_write_op(iocb);
 			task_io_account_write(bio->bi_iter.bi_size);
 		}
+		if (iocb->ki_flags & IOCB_NOWAIT)
+			bio->bi_opf |= REQ_NOWAIT;
 
 		dio->size += bio->bi_iter.bi_size;
 		pos += bio->bi_iter.bi_size;

Looks fine to me, we definitely should not be waiting on tags for IOCB_NOWAIT
IO. Will run some shakedown and test for 5.11.

up

-- 
Pavel Begunkov