On 9/5/2018 3:34 PM, Rob Clark wrote:

On Wed, Sep 5, 2018 at 5:22 AM Vivek Gautam [off-list ref] wrote:

quoted

On 8/14/2018 5:54 PM, Vivek Gautam wrote:

quoted

Hi Will,


On 8/14/2018 5:10 PM, Will Deacon wrote:

quoted

Hi Vivek,

On Tue, Aug 14, 2018 at 04:25:23PM +0530, Vivek Gautam wrote:

quoted

Qcom's implementation of arm,mmu-500 on sdm845 has a
functional/performance
errata [1] because of which the TCU cache look ups are stalled during
invalidation cycle. This is mitigated by serializing all the
invalidation
requests coming to the smmu.

How does this implementation differ from the one supported by
qcom_iommu.c?
I notice you're adding firmware hooks here, which we avoided by
having the
extra driver. Please help me understand which devices exist, how they
differ, and which drivers are intended to support them!

IIRC, the qcom_iommu driver was intended to support the static context
bank - SID
mapping, and is very specific to the smmu-v2 version present on
msm8916 soc.
However, this is the qcom's mmu-500 implementation specific errata.
qcom_iommu
will not be able to support mmu-500 configurations.
Rob Clark can add more.
Let you know what you suggest.

Rob, can you please comment about how qcom-smmu driver has different
implementation
from arm-smmu driver?

sorry, I missed this thread earlier.  But yeah, as you mentioned, the
purpose for qcom_iommu.c was to deal with the static context/SID
mapping.

(I guess it is all just software, and we could make qcom_iommu.c
support dynamic mapping as well, but I think then it starts to
duplicate most of arm_smmu.c, so that doesn't seem like the right
direction)

Thanks Rob for the response. I will wait for Will's response on how would he
like this support be implemented.

Best regards
Vivek

BR,
-R

quoted

Will, in case we would want to use arm-smmu driver, what would you
suggest for
having the firmware hooks?
Thanks.

Best regards
Vivek