Optimise percpu_ref_tryget() by not calling it for each request, but
batching it. This gave a measurable ~5% performance boost for large QD.
v2: fix uncommited plug (Jens Axboe)
better comments for percpu_ref_tryget_many (Dennis Zhou)
amortise across io_uring_enter() boundary
v3: drop "batching across syscalls"
remove error handling in io_submit_sqes() from common path
v4: fix error handling
Pavel Begunkov (2):
pcpu_ref: add percpu_ref_tryget_many()
io_uring: batch getting pcpu references
fs/io_uring.c | 26 +++++++++++++++++---------
include/linux/percpu-refcount.h | 26 +++++++++++++++++++++-----
2 files changed, 38 insertions(+), 14 deletions(-)
--
2.24.0
