On 5/17/22 00:21, Jens Axboe wrote:
Hi, This series builds to adding support for a different way of doing provided buffers, which is a lot more efficient than the existing scheme for high rate of consumption/provision of buffers. The interesting bits here are patch 3, which also has some performance numbers an an explanation of it. Patch 1 adds NOP support for provided buffers, just so that we can benchmark the last change. Patch 2 just abstracts out the pinning code. Patch 3 adds the actual feature. This passes the full liburing suite, and various test cases I adopted to use ring provided buffers. v6: - Change layout so that 'head' overlaps with reserved field in first buffer, avoiding the weird split of first page having N-1 buffers and the rest N (Dylan) - Rebase on current kernel bits - Fix missing ring unlock on out-of-bufs - Fix issue in io_recv() Can also be found in my git repo, for-5.19/io_uring-pbuf branch: https://git.kernel.dk/cgit/linux-block/log/?h=for-5.19/io_uring-pbuf and there's an associated liburing branch too: https://git.kernel.dk/cgit/liburing/log/?h=huge
should be the buf-ring branch I guess
fs/io_uring.c | 323 +++++++++++++++++++++++++++++----- include/uapi/linux/io_uring.h | 36 ++++ 2 files changed, 319 insertions(+), 40 deletions(-)
