[PATCH] iomap: fix the logic about poll io in iomap_dio_bio_actor

[PATCH] iomap: fix the logic about poll io in iomap_dio_bio_actor

[yangerkun]

Just set REQ_HIPRI for the last bio in iomap_dio_bio_actor. Because
multi bio created by this function can goto different cpu since this
process can be preempted by other process. And in iomap_dio_rw we will
just poll for the last bio. Fix it by only set polled for the last bio.

Signed-off-by: yangerkun <yangerkun@huawei.com>
---
 fs/iomap/direct-io.c | 22 +++++++++++++++-------
 1 file changed, 15 insertions(+), 7 deletions(-)

diff --git a/fs/iomap/direct-io.c b/fs/iomap/direct-io.c
index 1fc28c2da279..05dee6e7ca64 100644
--- a/fs/iomap/direct-io.c
+++ b/fs/iomap/direct-io.c
@@ -59,15 +59,16 @@ int iomap_dio_iopoll(struct kiocb *kiocb, bool spin)
 EXPORT_SYMBOL_GPL(iomap_dio_iopoll);
 
 static void iomap_dio_submit_bio(struct iomap_dio *dio, struct iomap *iomap,
-		struct bio *bio)
+		struct bio *bio, bool is_poll)
 {
 	atomic_inc(&dio->ref);
 
-	if (dio->iocb->ki_flags & IOCB_HIPRI)
+	if (is_poll) {
 		bio_set_polled(bio, dio->iocb);
-
-	dio->submit.last_queue = bdev_get_queue(iomap->bdev);
-	dio->submit.cookie = submit_bio(bio);
+		dio->submit.last_queue = bdev_get_queue(iomap->bdev);
+		dio->submit.cookie = submit_bio(bio);
+	} else
+		submit_bio(bio);
 }
 
 static ssize_t iomap_dio_complete(struct iomap_dio *dio)
@@ -191,7 +192,7 @@ iomap_dio_zero(struct iomap_dio *dio, struct iomap *iomap, loff_t pos,
 	get_page(page);
 	__bio_add_page(bio, page, len, 0);
 	bio_set_op_attrs(bio, REQ_OP_WRITE, flags);
-	iomap_dio_submit_bio(dio, iomap, bio);
+	iomap_dio_submit_bio(dio, iomap, bio, false);
 }
 
 static loff_t
@@ -255,6 +256,8 @@ iomap_dio_bio_actor(struct inode *inode, loff_t pos, loff_t length,
 
 	do {
 		size_t n;
+		bool is_poll = false;
+
 		if (dio->error) {
 			iov_iter_revert(dio->submit.iter, copied);
 			return 0;
@@ -301,7 +304,12 @@ iomap_dio_bio_actor(struct inode *inode, loff_t pos, loff_t length,
 		copied += n;
 
 		nr_pages = iov_iter_npages(&iter, BIO_MAX_PAGES);
-		iomap_dio_submit_bio(dio, iomap, bio);
+
+		/* Only set poll for the last bio. */
+		if (!nr_pages && dio->iocb->ki_flags & IOCB_HIPRI)
+			is_poll = true;
+
+		iomap_dio_submit_bio(dio, iomap, bio, is_poll);
 	} while (nr_pages);
 
 	/*
-- 
2.17.2

Re: [PATCH] iomap: fix the logic about poll io in iomap_dio_bio_actor

[Christoph Hellwig]

On Mon, Oct 14, 2019 at 10:43:13PM +0800, yangerkun wrote:
> Just set REQ_HIPRI for the last bio in iomap_dio_bio_actor. Because
> multi bio created by this function can goto different cpu since this
> process can be preempted by other process. And in iomap_dio_rw we will
> just poll for the last bio. Fix it by only set polled for the last bio.

I agree that there is a problem with the separate poll queue now.  But
doing partially polled I/O also doesn't seem very useful.  Until we
can find a way to poll for multiple bios from one kiocb I think we need
to limit polling to iocbs with just a single bio.  Can you look into
that?  __blkdev_direct_IO do_blockdev_direct_IO probably have the same
issues.  The former should be just as simple to fix, and for the latter
it might make sense to drop polling support entirely.

Re: [PATCH] iomap: fix the logic about poll io in iomap_dio_bio_actor

[yangerkun]

On 2019/10/15 16:05, Christoph Hellwig wrote:
> On Mon, Oct 14, 2019 at 10:43:13PM +0800, yangerkun wrote:
>> Just set REQ_HIPRI for the last bio in iomap_dio_bio_actor. Because
>> multi bio created by this function can goto different cpu since this
>> process can be preempted by other process. And in iomap_dio_rw we will
>> just poll for the last bio. Fix it by only set polled for the last bio.
> 
> I agree that there is a problem with the separate poll queue now.  But
> doing partially polled I/O also doesn't seem very useful.  Until we
> can find a way to poll for multiple bios from one kiocb I think we need
> to limit polling to iocbs with just a single bio.  Can you look into
> that?  __blkdev_direct_IO do_blockdev_direct_IO probably have the same
> issues.  The former should be just as simple to fix, and for the latter
> it might make sense to drop polling support entirely.
> 
Thanks for your suggestion, i will try to fix it.

Thanks,
Kun.
>