On 2/4/21 6:52 AM, Pavel Begunkov wrote: > Saving one lock/unlock for io-wq is not super important, but adds some > ugliness in the code. More important, atomic decs not turning it to zero > for some archs won't give the right ordering/barriers so the > io_steal_work() may pretty easily get subtly and completely broken. > > Return back 2-step io-wq work exchange and clean it up. IIRC, this wasn't done to skip the lock/unlock exchange, which I agree doesn't matter, but to ensure that a link would not need another io-wq punt. And that is a big deal, it's much faster to run it from that same thread, rather than needing a new async queue and new thread grab to get there. Just want to make sure that's on your mind... Maybe it's still fine as-is, didn't look too closely yet or test it. -- Jens Axboe
