> On Dec 13, 2019, at 2:32 PM, Pavel Begunkov <asml.silence@xxxxxxxxx> wrote: > > On 13/12/2019 21:32, Jens Axboe wrote: >>> On 12/13/19 11:22 AM, Jens Axboe wrote: >>> On 12/13/19 12:51 AM, Pavel Begunkov wrote: >>>> There is no reliable way to submit and wait in a single syscall, as >>>> io_submit_sqes() may under-consume sqes (in case of an early error). >>>> Then it will wait for not-yet-submitted requests, deadlocking the user >>>> in most cases. >>> >>> Why not just cap the wait_nr? If someone does to_submit = 8, wait_nr = 8, >>> and we only submit 4, just wait for 4? Ala: >>> > > Is it worth entangling the code? I don't expect anyone trying to recover, > maybe except full reset/restart. So, failing ASAP seemed to me as the > right thing to do. It may also mean nothing to the user if e.g. > submit(1), submit(1), ..., submit_and_wait(1, n) > > Anyway, this shouldn't even happen in a not buggy code, so I'm fine with > any version as long as it doesn't lock up. I'll resend if you still prefer > to cap it. I like the cap version a lot better, and in fact we did used to have that but lost it early. I like it behaviorally a lot better, too. Can you resend? Thanks!
