Re: [PATCH 6.7 427/641] io_uring: dont check iopoll if request completes

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 23. 01. 24, 0:55, Greg Kroah-Hartman wrote:
6.7-stable review patch.  If anyone has any objections, please let me know.

Hi,

6.7.2 fails in liburing tests (both x64 and x86-32 lib on x64 kernel):
[ 115s] Tests failed (5): <fd-pass.t> <msg-ring-overflow.t> <pipe-bug.t> <poll-race-mshot.t> <reg-hint.t>

I cannot reproduce locally, that happens only in openSUSE build machinery (the errors are transient, the links might not be valid in the future):
https://build.opensuse.org/package/live_build_log/openSUSE:Factory:Staging:H/liburing/standard/i586
https://build.opensuse.org/package/live_build_log/openSUSE:Factory:Staging:H/liburing/standard/x86_64

So I cannot tell if 6.8-rc is affected.

I suspect one of the 6.7.2 uring changes:
e24bf5b47a57 io_uring: adjust defer tw counting
22eed9134509 io_uring: ensure local task_work is run on wait timeout
ba8d8a8a36b2 io_uring/rw: ensure io->bytes_done is always initialized
d413a342275d io_uring: don't check iopoll if request completes

It looks like EINVAL is received unexpectedly (see below). Any ideas?

The tests' failures:

[ 53s] Running test fd-pass.t msg_ring failed -22
[   53s] test failed 0 2
[   53s] msg_ring failed -22
[   53s] test failed 1 1
[   53s] msg_ring failed -22
[   53s] test failed 1 0
[   53s] io_uring_register_file_alloc_range -22
[   53s] test failed 1 ALLOC
[   53s] Test fd-pass.t failed with ret 1



[ 65s] Running test msg-ring-overflow.t Destination ring create failed -22
[   65s] test defer failed
[   65s] Test msg-ring-overflow.t failed with ret 1



[ 75s] Running test pipe-bug.t pipe-bug.c:73 io_uring_wait_cqe_timeout(&ring, &cqe, &to) == 0 failed
[   75s] Test pipe-bug.t failed with ret 1



[ 76s] Running test poll-race-mshot.t Bad cqe res 0
[   76s] Bad cqe res 0
[   76s] Bad cqe res 0
[   76s] Bad cqe res 0
[   76s] Bad cqe res 0
... a lot of these
[   76s] Bad cqe res 0
[   76s] Bad cqe res 0
[   76s] Bad cqe res 0
[   76s] Bad cqe res 0
[   76s] Only got 1 requests
[   76s] Test mshot failed loop 0
[   76s] Test poll-race-mshot.t failed with ret 1



[ 84s] Running test reg-hint.t Bad CQE res: -22
[   84s] Test reg-hint.t failed with ret 1

From: Pavel Begunkov <asml.silence@xxxxxxxxx>

commit 9b43ef3d52532a0175ed6654618f7db61d390d2e upstream.

IOPOLL request should never return IOU_OK, so the following iopoll
queueing check in io_issue_sqe() after getting IOU_OK doesn't make any
sense as would never turn true. Let's optimise on that and return a bit
earlier. It's also much more resilient to potential bugs from
mischieving iopoll implementations.

Cc:  <stable@xxxxxxxxxxxxxxx>
Signed-off-by: Pavel Begunkov <asml.silence@xxxxxxxxx>
Link: https://lore.kernel.org/r/2f8690e2fa5213a2ff292fac29a7143c036cdd60.1701390926.git.asml.silence@xxxxxxxxx
Signed-off-by: Jens Axboe <axboe@xxxxxxxxx>
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
---
  io_uring/io_uring.c |    6 +++++-
  1 file changed, 5 insertions(+), 1 deletion(-)

--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -1898,7 +1898,11 @@ static int io_issue_sqe(struct io_kiocb
  			io_req_complete_defer(req);
  		else
  			io_req_complete_post(req, issue_flags);
-	} else if (ret != IOU_ISSUE_SKIP_COMPLETE)
+
+		return 0;
+	}
+
+	if (ret != IOU_ISSUE_SKIP_COMPLETE)
  		return ret;
/* If the op doesn't have a file, we're not polling for it */




--
js
suse labs





[Index of Archives]     [Linux Kernel]     [Kernel Development Newbies]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite Hiking]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux