Re: [PATCH v3 10/13] iommu/amd: Refactor helper function for attaching / detaching device

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

Jason,


On 12/12/2023 8:29 PM, Jason Gunthorpe wrote:
> On Tue, Dec 12, 2023 at 11:11:24AM +0530, Vasant Hegde wrote:
>> Jason,
>>
>>
>> On 11/30/2023 11:25 PM, Jason Gunthorpe wrote:
>>> On Thu, Nov 23, 2023 at 11:09:20PM +0530, Vasant Hegde wrote:
>>>>
>>>>
>>>> On 11/7/2023 6:58 PM, Jason Gunthorpe wrote:
>>>>> On Tue, Nov 07, 2023 at 11:25:49AM +0530, Vasant Hegde wrote:
>>>>>
>>>>>>>  RID domain=BLOCKED & PASID - ?? [V=1, TV=0, GV=1, GIOV=1 with GCR3 entry 0 being non-valid] 
>>>>>>
>>>>>> I haven't thought this scenario. Why do we even need this case?
>>>>>
>>>>> It is triggerable by iommufd, it has to be supported.
>>>>
>>>> You mean PASID support with BLOCKED domain?
>>>
>>> Yes, that is API possible.
>>
>> This will be tricky as we set DTE flags per device, not per PASID. This needs
>> more investigation. We will implement BLOCKING domain first then we can look
>> into PASID with BLOCKED domain support.
> 
> You can look at how the smmu stuff ended up for some inspiration, it
> is quite similar.
> 
> It is what I keep saying about properly structuring things so the DTE
> content is decided high up and then flowed down to programming
> logic. Then you don't get stuck with "we set DTE flags based on the
> device" which is not actually workable.

Agreed. That's the ultimiate goal. Just that we want to do it in stages so that
its easy to manage. I have posted v4 last week. Please take a look.

[1]
https://lore.kernel.org/linux-iommu/20231212085224.6985-1-vasant.hegde@amd.com/]


-Vasant