On Fri, Oct 18, 2024 at 07:46:44AM +0200, Christoph Hellwig wrote:

On Thu, Oct 17, 2024 at 09:09:33AM -0700, Keith Busch wrote:

quoted

From: Keith Busch <kbusch@kernel.org>

This is still backwards compatible with lifetime hints. It just doesn't
constrain the hints to that definition.

So in the end we'll end up with two uses of it - the existing 5
temperature hints and the new stream separation.  I think it
would be cleaner to make it a union, but I don't care that
strongly.

But we probably want a way to distinguish which one is supported.

E.g. for SCSI we set a net BLK_FEAT_WRITE_HINTS, for NVMe we'll set
BLK_FEAT_STREAM_SEPARATION.

Either way this should probably be the first patch in the series.

I'm not sure I follow this feedback. The SCSI feature is defined as a
lifetime stream association in SBC-5. So it's still a stream for SCSI,
but you want to call it "WRITE_HINT", which is not a term used in the
SCSI spec for this feature. But, you want to call it STREAM_SEPARATION
for NVMe only, even though the FDP spec doesn't use that term? What's
wrong with just calling it a generic hint support feature?

I also don't see why SCSI couldn't use per-io hints just like this
enables for NVMe. The spec doesn't limit SCSI to just 5 streams, so this
provides a way to access them all through the raw block device.