On 6/23/22 7:24 AM, Pavel Begunkov wrote: > 1-5 are clean ups, can be considered separately. > > 6 optimises the final atomic_dec() in __io_arm_poll_handler(). Jens > measured almost the same patch imrpoving some of the tests (netbench?) > by ~1-1.5%. > > v2: fix inverted EPOLLET check This passes tests, and I really like the cleanups before the actual change. Thanks! -- Jens Axboe
