On 06/06/2017 12:11 PM, Jean-Philippe Brucker wrote:
> On 05/06/17 18:20, Rob Herring wrote:
>> pci-host-{e,}cam-generic is a special case. I'm okay with having a
>> property for that I suppose. We should not require this property though
>> and allow for it to be implied by the SoC specific compatible as well.
>
> The property isn't useful for host-generic since a host like Qemu can
> easily disable ATS in the IOMMU. It would only be here for consistency.
> Changing IOMMU registers wouldn't be as simple for a firmware running on
> real hardware.
Sorry I was mistaken here. AMD Seattle hardware, for instance, implements
a pure host-ecam-generic. So it's not exclusively used by Qemu. A quick
search in the kernel also brings up alpine and thunder2. So I guess I'll
move ats-supported documentation to host-generic-pci.txt.
Thanks,
Jean
