[knowingly buggy, don't use] Just a glimpse on the de-atomising request refcounting for benchmarking, dirty and with a bunch of known problems (in [a]poll and iopoll). Haven't got any perf numbers myself yet. v2: wrong patch with inverted an req_ref_sub_and_test() condition Pavel Begunkov (2): io_uring: defer submission ref put io_uring: non-atomic request refs fs/io_uring.c | 99 ++++++++++++++++++++++++++++++--------------------- 1 file changed, 58 insertions(+), 41 deletions(-) -- 2.31.1
