haoxu has quit [Remote host closed the connection]
sgarzare_ has joined #io_uring
sgarzare has quit [Ping timeout: 480 seconds]
sgarzare__ has joined #io_uring
sgarzare__ has quit [Remote host closed the connection]
sgarzare__ has joined #io_uring
sgarzare_ has quit [Ping timeout: 480 seconds]
sgarzare_ has joined #io_uring
sgarzare__ has quit [Ping timeout: 480 seconds]
sgarzare_ has quit [Quit: Leaving]
erwan_taff has joined #io_uring
erwan_taf has quit [Read error: Connection reset by peer]
erwan_taf has joined #io_uring
erwan_taff has quit [Read error: Connection reset by peer]
erwan_taff has joined #io_uring
erwan_taf has quit [Read error: Connection reset by peer]
bnbajwa has joined #io_uring
bnbajwa has quit [Remote host closed the connection]
JoshTriplett has joined #io_uring
<JoshTriplett>
axboe: So, it's working, and performing reasonably well. But if I leave the VM running, about five minutes after the benchmark, I hit the first WARN_ON_ONCE in io_ring_exit_work.
<JoshTriplett>
I'm *guessing* that I've broken some io_uring assumption, which doesn't explode immediately but explodes later on.
* JoshTriplett
is currently trying to figure out what assumption that might be.
<JoshTriplett>
Current hypothesis: in taking the sqes off to another task to issue them, I'm failing to clean up after them in some way that makes io_uring think they haven't actually completed.
<JoshTriplett>
Oh. It would probably help if I called the completion machinery...at all.
<JoshTriplett>
Hmmm, wait; I *am*, sorta. I don't call io_req_complete_post or similar, but that seems to be io_issue_sqe's job iff returning IOU_OK. But I've been returning IOU_ISSUE_SKIP_COMPLETE, for simplicity's sake. So I'm not sure what's not getting called and should be...