[PATCH 0/3][SRU Focal] UBUNTU: SAUCE: Add IB peer memory interface

[dann frazier]

Hi Stefan, Kamal,

  Combining replies to both of you here since they seem to be tightly
coupled. Before that I want to clarify current status. Based on Tim's
NAK, I assumed consideration of this backport was paused on your end
until we are able to get a split-out version of [PATCH 3/3] from
from upstream. It is also on pause from my end due to a recently
discovered practical issue: while the goal here is to provide users
with this feature in a way that does not require overriding our
drivers with the MOFED driver stack, we've learned recently that it is
possible that the target users might still need the MOFED stack for
other reasons. I think we should pause further reviews until that is
resolved. But more inline...
  
On Thu, Nov 04, 2021 at 03:52:32PM +0100, Stefan Bader wrote:
> I have beeen thinking over this a while. One think I am not feeling really
> happy with changing the API of a public function (one never knows which
> external users are around).

On Thu, Nov 04, 2021 at 07:56:16AM -0700, Kamal Mostafa wrote:
> We discussed this in the Kernel Team meeting today -- we've got some
> significant concerns about applying this to Focal.
> 
> It's awfully huge, and appears to change lots of API-level stuff which
> could be a problem for other users of the subsystem (as you noted -- "Of
> course, we can't rule out other users").  We are also very reluctant to
> apply such a massive change to our Focal kernel source for fear that it
> will be highly likely to result in merge collisions with future upstream
> v5.4-stable updates.

Good feedback and, while I don't disagree, I'll note that it would've
saved some time and effort to have received it during the RFC phase...

Can you confirm that the only concerning API-level issue you're noting
is the change to the exported ib_umem_get() symbol? And can you
confirm that this would be another blocker to merge?

On Thu, Nov 04, 2021 at 03:52:32PM +0100, Stefan Bader wrote:
> Also I am wondering why having this in the HWE kernels is not
> enough. It just seems like a lot of change in a LTS kernel to
> be "safe".

The target deployments are users who currently use the LTS kernel, and
who are not comfortable with a track that requires a new upstream
base every 6 months for support.

On Thu, Nov 04, 2021 at 07:56:16AM -0700, Kamal Mostafa wrote:
> The team would also like to know: How large and invasive is the
> Hirsute/Impish SAUCE delta for this, compared to this Focal backport?

The biggest difference is the initial patch, "IB: Allow
calls to ib_umem_get from kernel ULPs", a backport from upstream that
is only needed for focal. [PATCH 2/3] is the equivalent patch to what
we're already shipping in H/I, and is very similar wrt size and
invasiveness:

  https://kernel.ubuntu.com/git/ubuntu/ubuntu-impish.git/commit/?id=25555f7881b75d149a69f4ca8970a2da0ac73049

[PATCH 3/3] is basically the same for all 3 trees, and has yet to be
applied on any of them due to Tim's request for a split out.

Cheers!

  -dann