Re: [PATCH v3 1/5] iommu/amd: Remove iommu_v2 module

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

On 2023/9/22 19:59, Jason Gunthorpe wrote:
> On Fri, Sep 22, 2023 at 10:23:26AM +0800, Baolu Lu wrote:
> 
>> 1) PCI device supports ATS but*no*  PASID
>>     In this case, ATS only means a TLB cache in the device, and it does
>>     not make much sense to cache the 1:1 mappings in the device. Even
>>     worse, the ATS translation requests could probably cause I/O
>>     congestion in corner cases. Therefore, the best choice is probably to
>>     disable ATS on the device.
> Nope. We have important use cases where ATS must always be on. 🙁

Oh, I didn't know that. Thank you for letting me know.

> 
> You might make this argument if ACS is also non-isolating though..
> 
> Regardless I think we need to get into a position where the iommu core
> is deciding if PRI or ATS is enabled for a device, not the iommu
> driver.

Agreed. I ever had a series to achieve this and it may be time to
revisit them.

One additional concern is how the core knows whether ATS should be
enabled. In my previous design, the IOMMU core turns on ATS by default
if the device is capable of it, but the driver could enable/disable it
from its driver probe() callback.

Does it make sense?

Best regards,
baolu