Hi Joanne, >>> Thanks, applied and pushed with some cleanups including Luis's clamp idea. >> >> Hi Miklos, >> >> I don't think the timeouts do work with io-uring yet, I'm not sure >> yet if I have time to work on that today or tomorrow (on something >> else right now, I can try, but no promises). > > Hi Bernd, > > What are your thoughts on what is missing on the io-uring side for > timeouts? If a request times out, it will abort the connection and > AFAICT, the abort logic should already be fine for io-uring, as users > can currently abort the connection through the sysfs interface and > there's no internal difference in aborting through sysfs vs timeouts. > in fuse_check_timeout() it iterates over each fud and then fpq. In dev_uring.c fpq is is per queue but unrelated to fud. In current fuse-io-uring fud is not cloned anymore - using fud won't work. And Requests are also not queued at all on the other list fuse_check_timeout() is currently checking. Also, with a ring per core, maybe better to use a per queue check that is core bound? I.e. zero locking overhead? And I think we can also avoid iterating over hash lists (queue->fpq), but can use the 'ent_in_userspace' list. We need to iterate over these other entry queues anyway: ent_w_req_queue fuse_req_bg_queue ent_commit_queue And we also need to iterate over fuse_req_queue fuse_req_bg_queue Thanks, Bernd
