[PATCHv3 0/7] dma mapping optimisations
Keith Busch
kbusch at fb.com
Fri Aug 5 09:24:37 PDT 2022
From: Keith Busch <kbusch at kernel.org>
Changes since v2:
Fixed incorrect io_uring io_fixed_file index validit checksy: this should
have been validating the file_ptr (Ammar)
Various micro-optimizations: move up dma in iov type checks, skip
iov_iter_advance on async IO (Jens).
NVMe driver cleanups splitting the fast and slow paths.
NVMe driver prp list setup fixes when using the slow path.
Summary:
A user address undergoes various represenations for a typical read or
write command. Each consumes memory and CPU cycles. When the backing
storage is NVMe, the sequence looks something like the following:
__user void *
struct iov_iter
struct pages[]
struct bio_vec[]
struct scatterlist[]
__le64[]
Applications will often use the same buffer for many IO, so these
potentially costly per-IO transformations to reach the exact same
hardware descriptor can be skipped.
The io_uring interface already provides a way for users to register
buffers to get to 'struct bio_vec[]'. That still leaves the scatterlist
needed for the repeated dma_map_sg(), then transform to nvme's PRP list
format.
This series takes the registered buffers a step further. A block driver
can implement a new .dma_map() callback to reach the hardware's DMA
mapped address format, and return a cookie so a user can reference it
later for any given IO. When used, the block stack can skip significant
amounts of code, improving CPU utilization and IOPs.
The implementation is currently limited to mapping a registered buffer
to a single io_uring fixed file.
Keith Busch (7):
blk-mq: add ops to dma map bvec
file: add ops to dma map bvec
iov_iter: introduce type for preregistered dma tags
block: add dma tag bio type
io_uring: introduce file slot release helper
io_uring: add support for dma pre-mapping
nvme-pci: implement dma_map support
block/bdev.c | 20 +++
block/bio.c | 24 ++-
block/blk-merge.c | 19 ++
block/fops.c | 24 ++-
drivers/nvme/host/pci.c | 314 +++++++++++++++++++++++++++++++--
fs/file.c | 15 ++
include/linux/bio.h | 22 ++-
include/linux/blk-mq.h | 24 +++
include/linux/blk_types.h | 6 +-
include/linux/blkdev.h | 16 ++
include/linux/fs.h | 20 +++
include/linux/io_uring_types.h | 2 +
include/linux/uio.h | 9 +
include/uapi/linux/io_uring.h | 12 ++
io_uring/filetable.c | 34 ++--
io_uring/filetable.h | 10 +-
io_uring/io_uring.c | 139 +++++++++++++++
io_uring/net.c | 2 +-
io_uring/rsrc.c | 27 +--
io_uring/rsrc.h | 10 +-
io_uring/rw.c | 2 +-
lib/iov_iter.c | 27 ++-
22 files changed, 724 insertions(+), 54 deletions(-)
--
2.30.2
More information about the Linux-nvme
mailing list