On 2022/5/6 03:27, Jason Gunthorpe wrote:
On Thu, May 05, 2022 at 07:56:59PM +0100, Robin Murphy wrote:
Ack to that, there are certainly further improvements to consider once we've
got known-working code into a released kernel, but let's not get ahead of
ourselves just now.
Yes please
(I'm pretty sure we could get away with a single blocking domain per IOMMU
instance, rather than per group, but I deliberately saved that one for later
- right now one per group to match default domains is simpler to reason
about and less churny in the context of the current ownership patches)
I noticed this too..
But I thought the driver can do a better job of this. There is no
reason it has to allocate memory to return a IOMMU_DOMAIN_BLOCKED
domain - this struct could be a globally allocated singleton for the
entire driver and that would be even better memory savings.
For instance, here is a sketch for the Intel driver based on Baolu's
remark that intel_iommu_detach_device() establishes a blocking
behavior already on detach_dev (Baolu if you like it can you make a
proper patch?):
Yes, I will.
The same scheme could also be applied to identity/sva/block domains.
Perhaps make it after v5.19.
Best regards,
baolu
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu
