Thread (34 messages) 34 messages, 3 authors, 2025-02-25

Re: [PATCH v7 13/14] iommu/arm-smmu-v3: Report events that belong to devices attached to vIOMMU

From: Jason Gunthorpe <jgg@nvidia.com>
Date: 2025-02-24 23:35:13
Also in: linux-doc, linux-iommu, linux-kselftest, linux-patches, lkml

On Mon, Feb 24, 2025 at 01:56:46PM -0800, Nicolin Chen wrote:
quoted
Just thinking out loud here:
I understand the goal here is to "emulate" an IOMMU. But I'm just
wondering if we could report struct events instead of the raw event?

For example, can't we have something like arm_smmu_event here with the
sid changed to vsid? 

Are we taking the raw event since we want to keep the `u64 event_data[]`
field within `struct iommufd_vevent` generic to all architectures?
The ABIs for vSMMU are defined in the HW languange, e.g. cmd, ste.
Thus, here evt in raw too.
Right, the point is that it gives as a safe uABI that is effectively
being managed by ARM.

If we make our own thing then we have to take the responsiblity to
make it safe and extensible. I don't see a justification to do that..

It is the same discussion we had around the vSTE as input, the raw
invalidation command and the IDRs. Since we've already done 'follow
the SMMU spec' so many times already now we should keep doing it.

Jason
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help