On Sun, Mar 1, 2026 at 11:22 AM Jason Gunthorpe <[email protected]> wrote: > > On Fri, Feb 27, 2026 at 09:04:49AM -0700, Alex Williamson wrote: > > > Not only fabric topology, but also routing. > > Yes > > > ACS overrides on the > > command line would need to be enforced between the original and kexec > > kernel such that IOMMU groups are deterministic. Thanks, > > That's a good point, I think as a reasonable starting point we should > require live update preserved devices to have singleton iommu_groups > on both sides. That is easy to check and enforce, and it doesn't > matter how the administrator makes that happen.
I can add that in the next version. > You also can't change the ACS flags while traffic is flowing because > that changes the fabric routing too. This should take care to enforce > that restriction as well. I'll look into this but will probably defer it to a future series. We need several more pieces in place until we are ready to support traffic to/from preserved devices during the Live Update.
