can QEMU's arm smmu model handle non-PCI devices?

can QEMU's arm smmu model handle non-PCI devices?

[Peter Maydell]

Hi; in real hardware an SMMUv3 might be sat in front of any
DMA-capable device. Putting one in front of a PCI bus is common
but not the only use case. Does QEMU's SMMUv3 model handle that,
or is the implementation currently restricted to only handling DMA
from PCI devices?

I ask because for the Realm Management Extension (aka Confidential
Compute Architecture) we're going to need to put SMMUs in front of
other devices (such as the GIC), so that their DMA is checked against
the RME Granule Protection Tables. So if we don't currently handle
non-PCI devices then that's a bit of extra preliminary work that we'll
want to do :-)

thanks
-- PMM

Re: can QEMU's arm smmu model handle non-PCI devices?

[Eric Auger]

Hi Peter,
On 5/23/22 16:52, Peter Maydell wrote:
> Hi; in real hardware an SMMUv3 might be sat in front of any
> DMA-capable device. Putting one in front of a PCI bus is common
> but not the only use case. Does QEMU's SMMUv3 model handle that,
> or is the implementation currently restricted to only handling DMA
> from PCI devices?
only PCI is supported at the moment. IOMMU MR regions are only created
for PCI devices
(smmu_find_add_as() in hw/arm/smmu-common.c)

As a reminder, Chunming Li submitted several iterations of a series dedicated to that. See

https://patchew.org/QEMU/1629878922-173270-1-git-send-email-chunming_li1234@163.com/

It is not upstream though.

Thanks

Eric




>
> I ask because for the Realm Management Extension (aka Confidential
> Compute Architecture) we're going to need to put SMMUs in front of
> other devices (such as the GIC), so that their DMA is checked against
> the RME Granule Protection Tables. So if we don't currently handle
> non-PCI devices then that's a bit of extra preliminary work that we'll
> want to do :-)
>
> thanks
> -- PMM
>