Re: [PATCH 0/5] Share sva domains with all devices bound to a mm

[Tina Zhang <tina.zhang@intel.com>]

Hi,

On 8/9/23 18:51, Baolu Lu wrote:
> On 2023/8/9 17:44, Tian, Kevin wrote:
>>> From: Baolu Lu<baolu.lu@linux.intel.com>
>>> Sent: Wednesday, August 9, 2023 8:18 AM
>>>
>>> On 2023/8/8 15:49, Tina Zhang wrote:
>>>> A sva domain's lifetime begins with binding a device to a mm and ends
>>>> by releasing all the bound devices from that sva domain. Technically,
>>>> there could be more than one sva domain identified by the mm PASID for
>>>> the use of bound devices issuing DMA transactions.
>>>>
>>>> To support mm PASID 1:n with sva domains, each mm needs to keep both
>>> a
>>>> reference list of allocated sva domains and the corresponding PASID.
>>>> However, currently, mm struct only has one pasid field for sva usage,
>>>> which is used to keep the info of an assigned PASID. That pasid field
>>>> cannot provide sufficient info to build up the 1:n mapping between 
>>>> PASID
>>>> and sva domains.
>>> Is it more appropriate to have the same life cycle for sva domain and mm
>>> pasid? I feel that they represent the same thing, that is, the address
>>> space shared by mm to a device.
>>>
>> iirc it's a simplification to free mm pasid at __mmdrop() otherwise the
>> implementation is tricky, but I don't remember all the detail...
> 
> Yeah, probably we could also free the sva domains in __mmdrop()? Remove
> the refcount for sva domain just like what we did for pasid (at the
> beginning we had refcount for each pasid...).

For sva usage, mm->mm_count is increased in iommu_sva_domain_alloc(), 
and gets decreased when the domain has no users (which is checked in 
iommu_sva_unbind_device()).

So, in a mm's life time, there could be multiple sva domains, though 
they are using the same PASID. I think it makes sense to mm. Because it 
makes no sense to keep a sva domain alive when no users are using it, 
even though the mm is alive.

Regards,
-Tina

> 
> Best regards,
> baolu