Re: [PATCH 9/9] iommu: Use fault cookie to store iopf_param

[Baolu Lu <baolu.lu@linux.intel.com>]

On 2023/7/13 16:01, Tian, Kevin wrote:
>> From: Baolu Lu <baolu.lu@linux.intel.com>
>> Sent: Thursday, July 13, 2023 11:44 AM
>>
>> On 2023/7/13 11:24, Tian, Kevin wrote:
>>>> From: Baolu Lu <baolu.lu@linux.intel.com>
>>>> Sent: Wednesday, July 12, 2023 11:13 AM
>>>>
>>>> On 2023/7/12 6:02, Jacob Pan wrote:
>>>>> On Tue, 11 Jul 2023 09:06:42 +0800, Lu Baolu<baolu.lu@linux.intel.com>
>>>>> wrote:
>>>>>
>>>>>> @@ -158,7 +158,7 @@ int iommu_queue_iopf(struct iommu_fault
>> *fault,
>>>>>> struct device *dev)
>>>>>>     	 * As long as we're holding param->lock, the queue can't be
>>>>>> unlinked
>>>>>>     	 * from the device and therefore cannot disappear.
>>>>>>     	 */
>>>>>> -	iopf_param = param->iopf_param;
>>>>>> +	iopf_param = iommu_get_device_fault_cookie(dev, 0);
>>>>> I am not sure I understand how does it know the cookie type is
>> iopf_param
>>>>> for PASID 0?
>>>>>
>>>>> Between IOPF and IOMMUFD use of the cookie, cookie types are
>> different,
>>>>> right?
>>>>>
>>>>
>>>> The fault cookie is managed by the code that delivers or handles the
>>>> faults. The sva and IOMMUFD paths are exclusive.
>>>>
>>>
>>> what about siov? A siov-capable device can support sva and iommufd
>>> simultaneously.
>>
>> For siov case, the pasid should be global. RID and each pasid are still
>> exclusive, so I don't see any problem. Did I overlook anything?
>>
> 
> they are exclusive but it's weird to see some pasids (for sva) on this
> device are tracked by slot#0 while other pasids (for iommufd) occupies
> per-pasid slot.
> 
> why not generalizing them given you name it as "per-pasid fault cookie"?

Yeah! Get your point now. At least the partial list should be per-pasid.

Let me invest more time on this.

Best regards,
baolu