On Thu, Jun 10, 2021 at 04:23:56PM +0200, Krzysztof Kozlowski wrote:

On 10/06/2021 11:19, Thierry Reding wrote:

quoted

On Tue, Jun 08, 2021 at 05:48:51PM +0100, Will Deacon wrote:

quoted

On Tue, Jun 08, 2021 at 04:38:48PM +0200, Thierry Reding wrote:

quoted

On Tue, Jun 08, 2021 at 01:01:29PM +0100, Will Deacon wrote:

quoted

On Mon, Jun 07, 2021 at 10:49:10AM +0200, Krzysztof Kozlowski wrote:

quoted

This is the pull for you to base further SMMU aptches (prevent early SMMU
faults).

This is a tonne of code for me to pull into the SMMU tree given that I only
want one patch!

Thierry, if I just stick:

https://lore.kernel.org/r/20210603164632.1000458-4-thierry.reding@gmail.com (local)

on its own branch, can you stitch together whatever you need?

I'm not sure I understand what you're proposing. For reference, here's
the set of patches that I sent out:

  1. memory: tegra: Implement SID override programming
  2. dt-bindings: arm-smmu: Add Tegra186 compatible string
  3. iommu/arm-smmu: Implement ->probe_finalize()
  4. iommu/arm-smmu: tegra: Detect number of instances at runtime
  5. iommu/arm-smmu: tegra: Implement SID override programming
  6. iommu/arm-smmu: Use Tegra implementation on Tegra186
  7. arm64: tegra: Use correct compatible string for Tegra186 SMMU
  8. arm64: tegra: Hook up memory controller to SMMU on Tegra186
  9. arm64: tegra: Enable SMMU support on Tegra194

Krzysztof already picked up patch 1 and I was assuming that you'd pick
up 2-6 because they are to the ARM SMMU driver. However, if you're
primarily interested in just patch 3, which is more "core" ARM SMMU than
the rest, which are Tegra-specific, then I suppose what we could do is
for you to give an Acked-by on the rest (2, 4-6) and then Krzysztof or I
can pick them up and take them via ARM SoC, based on the stable branch
from your tree that only has patch 3.

I think you previously said that patch 5 depends on patch 1, so I can't
take 2-6 without also pulling in the memory controller queue.

quoted

Patch 6 touches arm-smmu-impl.c, though it's a two-line change that
touches only the Tegra-specific matching bit in arm_smmu_impl_init(), so
the likelihood of that conflicting with anything else is fairly small.

Is that what you were proposing?

I can queue as much or as little of 2-6 as you like, but I would like to
avoid pulling in the memory controller queue into the arm smmu tree. But
yes, whichever of those I take, I can put them on a separate branch so
that you're not blocked for the later patches.

You have a better handle on the dependencies, so please tell me what works
for you. I just want to make sure that at least patch 3 lands in my tree,
so we don't get late conflicts with other driver changes.

Yes, if you could pick up patch 3 and send out a link with the stable
branch, I think Krzysztof or I could pull in that branch and pick up the
remaining patches. It'd be good if you could also ack the remaining SMMU
patches so that ARM SoC knows that they've been sanctioned.

Krzysztof: would you be okay with picking up patches 2 and 4-6 on top of
your memory branch for v5.14?

You mean the iommu patches? Yes, I can take them and later explain to
Arnd/Olof why they come through me.

Okay, great.

Will, can you provide that stable branch? Or would you prefer if I
prepared it and sent you a pull request? We're kind of running out of
time, since for ARM SoC the cut-off point for new material is usually
-rc6 and that's coming up pretty fast.

Thierry