[PATCH v4 00/12] Initial support for SMMUv3 nested translation

[Baolu Lu]

On 11/14/24 00:43, Jason Gunthorpe wrote:
> On Wed, Nov 13, 2024 at 10:55:41AM +0800, Baolu Lu wrote:
>> On 11/13/24 09:23, Jason Gunthorpe wrote:
>>>> https://github.com/Linaro/linux-kernel-uadk/tree/6.12-wip
>>>> https://github.com/Linaro/qemu/tree/6.12-wip
>>>>
>>>> Still need this hack
>>>> https://github.com/Linaro/linux-kernel-uadk/commit/
>>>> eaa194d954112cad4da7852e29343e546baf8683
>>>>
>>>> One is adding iommu_dev_enable/disable_feature IOMMU_DEV_FEAT_SVA,
>>>> which you have patchset before.
>>> Yes, I have a more complete version of that here someplace. Need some
>>> help on vt-d but hope to get that done next cycle.
>>
>> Can you please elaborate this a bit more? Are you talking about below
>> change
> 
> I need your help to remove IOMMU_DEV_FEAT_IOPF from the intel
> driver. I have a patch series that eliminates it from all the other
> drivers, and I wrote a patch to remove FEAT_SVA from intel..

Yes, sure. Let's make this happen in the next cycle.

FEAT_IOPF could be removed. IOPF manipulation can be handled in the
domain attachment path. A per-device refcount can be implemented. This
count increments with each iopf-capable domain attachment and decrements
with each detachment. PCI PRI is enabled for the first iopf-capable
domain and disabled when the last one is removed. Probably we can also
solve the PF/VF sharing PRI issue.

With iopf moved to the domain attach path and hardware capability checks
to the SVA domain allocation path, FEAT_SVA becomes essentially a no-op.

> 
>> +	ret = iommu_dev_enable_feature(idev->dev, IOMMU_DEV_FEAT_SVA);
>> +	if (ret)
>> +		return ret;
>>
>> in iommufd_fault_iopf_enable()?
>>
>> I have no idea about why SVA is affected when enabling iopf.
> 
> It is ARM not implementing the API correctly. Only SVA turns on the
> page fault reporting mechanism.
> 
> In the new world the page fault reporting should be managed during
> domain attachment. If the domain is fault capable then faults should
> be delivered to that domain. That is the correct time to setup the
> iopf mechanism as well.
> 
> So I fixed that and now ARM and AMD both have no-op implementations of
> IOMMU_DEV_FEAT_IOPF and IOMMU_DEV_FEAT_SVA. Thus I'd like to remove it
> entirely.

Thank you for the explanation.

--
baolu