diff options
| author |   Pavel Begunkov <asml.silence@gmail.com> | 2021-01-19 13:32:47 +0000 |
|---|---|---|
| committer |   Jens Axboe <axboe@kernel.dk> | 2021-02-01 10:02:42 -0700 |
| commit | e342c807f556dbcee1370ab78af1d8faf497d771 (patch) | |
| tree | ab6f899c1118e520edc267eadc9e47df521913a1 /fs/internal.h | |
| parent | io_uring: don't flush CQEs deep down the stack (diff) | |
| download | linux-dev-e342c807f556dbcee1370ab78af1d8faf497d771.tar.xz linux-dev-e342c807f556dbcee1370ab78af1d8faf497d771.zip | |
io_uring: save atomic dec for inline executed reqs
When a request is completed with comp_state, its completion reference
put is deferred to io_submit_flush_completions(), but the submission
is put not far from there, so do it together to save one atomic dec per
request. That targets requests that complete inline, e.g. buffered rw,
send/recv.
Proper benchmarking haven't been conducted but for nops(batch=32) it was
around 7901 vs 8117 KIOPS (~2.7%), or ~4% per perf profiling.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Diffstat (limited to 'fs/internal.h')
0 files changed, 0 insertions, 0 deletions