[3/4] is splitting io_issue_sqe() as someone once proposed. (I can't
find who it was and the thread). Hopefully, it doesn't add much
overhead.
Apart from massive deduplication, this also reduces sqe propagation
depth, that's a good thing.
Pavel Begunkov (4):
io_uring: set/clear IOCB_NOWAIT into io_read/write
io_uring: remove nonblock arg from io_{rw}_prep()
io_uring: decouple issuing and req preparation
io_uring: move req preps out of io_issue_sqe()
fs/io_uring.c | 316 ++++++++++++--------------------------------------
1 file changed, 77 insertions(+), 239 deletions(-)
--
2.24.0
