[PATCHv9,0/7] write hints with nvme fdp, scsi streams

Message

Keith Busch Oct. 25, 2024, 9:36 p.m. UTC

From: Keith Busch <[e-mail]>

A little something for everyone here.

Upfront, I really didn't get the feedback about setting different flags
for stream vs. temperature support. Who wants to use it, and where and
how is that information used?

Changes from v8:

  Added reviews.

  Removed an unused header.

  Changed "hint" to "streams" in the commit logs.

  Ability to split available hints that a partition can use.

  Dropped all the generic filesystem changes that were defaulting to the
  kiocb write_hint. They are unchanged, which having no functional
  change was really the intention anyway, so let's just not change them.

  The above means we don't need a special fop flag to indicate support
  for the kiocb write_hint. Those filesystems that don't support it
  simply don't read it.

  Added the SCSI support since I had to read the spec anyway, and it is
  just a one-line change.

Kanchan Joshi (2):
  io_uring: enable per-io hinting capability
  nvme: enable FDP support

Keith Busch (5):
  block: use generic u16 for write hints
  block: introduce max_write_hints queue limit
  block: allow ability to limit partition write hints
  block, fs: add write hint to kiocb
  scsi: set permanent stream count in block limits

 Documentation/ABI/stable/sysfs-block |  7 +++
 block/bdev.c                         | 15 +++++
 block/blk-settings.c                 |  3 +
 block/blk-sysfs.c                    |  3 +
 block/fops.c                         | 26 ++++++++-
 block/partitions/core.c              | 46 +++++++++++++++-
 drivers/nvme/host/core.c             | 82 ++++++++++++++++++++++++++++
 drivers/nvme/host/nvme.h             |  5 ++
 drivers/scsi/sd.c                    |  2 +
 include/linux/blk-mq.h               |  3 +-
 include/linux/blk_types.h            |  4 +-
 include/linux/blkdev.h               | 12 ++++
 include/linux/fs.h                   |  1 +
 include/linux/nvme.h                 | 19 +++++++
 include/uapi/linux/io_uring.h        |  4 ++
 io_uring/rw.c                        |  3 +-
 16 files changed, 225 insertions(+), 10 deletions(-)

Comments

Christoph Hellwig Oct. 28, 2024, 11:49 a.m. UTC | #1

On Fri, Oct 25, 2024 at 02:36:38PM -0700, Keith Busch wrote:
> Upfront, I really didn't get the feedback about setting different flags
> for stream vs. temperature support. Who wants to use it, and where and
> how is that information used?

The SBC permanent streams have a 'relative lifetime' associated with
them.  So if you expose them as plain write streams you violate the
contract with the device.