23 Mar
2023
23 Mar
'23
2:35 p.m.
On Thu, Mar 23, 2023 at 08:04:24AM +0000, Tian, Kevin wrote:
From: Jason Gunthorpe jgg@nvidia.com Sent: Wednesday, March 22, 2023 3:15 AM
The implementation is complicated because we have to introduce some per-iommu_group memory in iommufd and redo how we think about multi- device groups to be more explicit. This solves all the locking problems in the prior attempts.
By tracking groups explicitly should we also move DMA ownership claim to be per group?
That looks like it would make things more complicated as we'd have to be alot more careful with the group refcount to avoid bugs.
Jason