[Intel-gfx] [PATCH] drm/i915: Use per device iommu check

[Tvrtko Ursulin]

On 26/11/2021 08:26, Lu Baolu wrote:
> On 2021/11/25 19:47, Robin Murphy wrote:
>> On 2021-11-25 10:42, Tvrtko Ursulin wrote:
>>> From: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
>>>
>>> With both integrated and discrete Intel GPUs in a system, the current
>>> global check of intel_iommu_gfx_mapped, as done from intel_vtd_active()
>>> may not be completely accurate.
>>>
>>> In this patch we add i915 parameter to intel_vtd_active() in order to
>>> prepare it for multiple GPUs and we also change the check away from 
>>> Intel
>>> specific intel_iommu_gfx_mapped (global exported by the Intel IOMMU
>>> driver) to probing the presence of IOMMU domain on a specific device
>>> using iommu_get_domain_for_dev().
>>
>> FWIW the way you have it now is functionally equivalent to using 
>> device_iommu_mapped(), which I think might be slightly clearer for the 
>> current intent, but I don't have a significantly strong preference 
>> (after all, this *was* the de-facto way of checking before 
>> device_iommu_mapped() was introduced, and there are still other 
>> examples of it around). So from the IOMMU perspective,
>>
>> Acked-by: Robin Murphy <robin.murphy at arm.com>
>>
>> Perhaps the AGP driver could also be tweaked and 
>> intel_iommu_gfx_mapped cleaned away entirely, but I'll leave that for 
>> Baolu to think about :)
> 
> I fully agreed with Robin.
> 
> I prefer device_iommu_mapped() more than iommu_get_domain_for_dev().
> 
> "iommu_get_domain_for_dev(dev) == NULL" simply means that the device
> does not have any domain attached. Although at present, it is equivalent
> to device DMAing without IOMMU translation. But I'm sure it will change
> in the future.
> 
> With device_iommu_mapped() replaced,
> 
> Reviewed-by: Lu Baolu <baolu.lu at linux.intel.com>

I have empirically established device_iommu_mapped() matches 1:1 with 
intel_iommu_gfx_mapped and a simple non-NULL check against 
iommu_get_domain_for_dev(), so okay, I will respin with that change.

We leave the problem of being able to distinguish between pass-through 
and address translation from the drivers for later. Thanks both!

Regards,

Tvrtko