As mentioned before, couldn't get numbers due to huge performance jumps between reboots outweighting any gains of this series, so sending just as an RFC. Jens, sorry for pestering, but if you will be doing benchmarking, can you compare how it is with the 1/2? 2/2 clearly slashes some extra cycles. Pavel Begunkov (2): io_uring: use kvmalloc for fixed files io_uring: inline fixed part of io_file_get() fs/io_uring.c | 98 +++++++++++++++++++++++++-------------------------- 1 file changed, 49 insertions(+), 49 deletions(-) -- 2.31.1
