From: Keith Busch <kbusch@xxxxxxxxxx>
Handling passthrough metadata ("integrity") today introduces overhead
and complications that we can avoid if we just map user space addresses
directly. This patch series implements that, falling back to a kernel
bounce buffer if necessary.
v1->v2:
Bounce to a kernel buffer if the user buffer fails to map to the
device's integrity constraints. The user address remains pinned for
the duration of the IO, which makes the copy out on completion safe
within interrupt context.
Merged up to current io_uring branch, which moved the driver owned
flags to a different file.
Keith Busch (4):
block: bio-integrity: directly map user buffers
nvme: use bio_integrity_map_user
iouring: remove IORING_URING_CMD_POLLED
io_uring: remove uring_cmd cookie
block/bio-integrity.c | 202 ++++++++++++++++++++++++++++++++++++++
drivers/nvme/host/ioctl.c | 174 +++++---------------------------
include/linux/bio.h | 9 ++
include/linux/io_uring.h | 9 +-
io_uring/uring_cmd.c | 1 -
5 files changed, 240 insertions(+), 155 deletions(-)
--
2.34.1
