Re: [RFC] iommu_ops->domain_alloc_paging() enhancement to support AMD IOMMU driver

[Vasant Hegde <vasant.hegde@amd.com>]

Hi Robin,


On 6/28/2024 11:28 PM, Robin Murphy wrote:
> On 28/06/2024 6:08 pm, Vasant Hegde wrote:
>> Hi Robin,
>>
>> On 6/28/2024 6:36 PM, Robin Murphy wrote:
>>> On 28/06/2024 1:23 pm, Baolu Lu wrote:
>>>> On 2024/6/28 14:43, Vasant Hegde wrote:
>>>>> We are working on adding domain_alloc_paging() support in AMD driver and came
>>>>> across below issue.
>>>>>
>>
>> .../...
>>
>>>> Add an allocation flag seems to work. It doesn't indicate any type of
>>>> domain. Instead, it indicates the required iommu feature.
>>>>
>>>> Something like
>>>>
>>>> #define DOMAIN_ALLOC_FLAG_PASID        0x1
>>>>
>>>> means the allocated domain requires PASID to support on both device and
>>>> IOMMU. If the hardware lacks this support, it should return failure.
>>>
>>> Yeah, we could do with something sufficiently general, as we'll also have a need
>>> for passing pagetable-format options through to drivers to be able to replace
>>> iommu_set_pgtable_quirks().
>>
>> Sorry. I doubt I understood it completely. You mean to say:
>> Enhance iommu_domain_alloc() such that caller can pass any special quirks they
>> need and then pass those quirk flag along with the other flags we need (like
>> PASID support or not) to indivisual driver? So that at the end we are remove
>> ops->set_pgtable_quirks().
>>
>> Is that what you are thinking or something else?
> 
> Basically, what set_pgtable_quirks does currently will need to be known at
> domain_alloc_paging time in order to convert arm-smmu to the latter properly.
> You don't need to worry about the public interface for it just yet, but if we're
> going to start extending the driver-level interface already then it would be
> good to plan ahead for what other domain properties we may want to accommodate
> callers requesting. The things we have as io-pgtable quirks today are definite
> ones (and those would happen to be OK as simple flags), but it's always seemed
> like a good idea for callers to be able to indicate their preference for other
> things like VA size (or even full domain geometry), even perhaps specific
> io-pgtable formats, such that the IOMMU driver can make the most effective
> choice about what it actually allocates.

Thanks for explaining. It does make sense to pass enough details to individual
drivers.

-Vasant