[PATCH 0/6] ublk: io_desc optimizations

[Caleb Sander Mateos]

ublk passes the parameters of incoming I/O in memory shared between the
kernel ublk driver and userspace ublk server in struct ublksrv_io_desc.

The thread submitting the ublk I/O writes to the io_desc, while the ublk
server thread handling the I/O reads the io_desc. This basically
guarantees a cache miss on both threads for each ublk I/O. Avoid the
cache misses by writing the io_desc on the server thread in the kernel
before dispatching the I/O to userspace.

The size of each io_desc is currently fixed to 24 bytes, which has been
an obstacle to extending it with additional fields [1]. Additionally,
with multiple ublk server threads handling I/Os from the same ublk queue
(possible with UBLK_F_PER_IO_DAEMON or UBLK_F_BATCH_IO), false sharing
results from adjacent io_descs sharing the same cache line. Add a
UBLK_F_IO_DESC_SIZE feature allowing the ublk server to increase the
size of the io_descs for its ublk devices.

[1]: https://lore.kernel.org/linux-block/aV8QfvaNO5P6vOs6@fedora/

Caleb Sander Mateos (6):
  ublk: consistently use u16 for queue and tag numbers
  ublk: remove struct ublk_zoned_report_desc's operation field
  ublk: split request validation from io_desc init
  ublk: initialize io_desc on daemon task
  ublk: add UBLK_F_IO_DESC_SIZE
  ublk: lift need_map check out of ublk_{,un}map_io()

 drivers/block/ublk_drv.c      | 213 +++++++++++++++++++---------------
 include/uapi/linux/ublk_cmd.h |   5 +-
 2 files changed, 124 insertions(+), 94 deletions(-)

-- 
2.54.0