AMD IOMMU v2 Page Tables Getting Wired Up For Linux 6.1

[phoronix]

Phoronix: AMD IOMMU v2 Page Tables Getting Wired Up For Linux 6.1

For more than a year AMD engineers have been working on IOMMU v2 page table support and with the in-development Linux 6.1 kernel the initial patches are finally being merged...

https://www.phoronix.com/news/AMD-IOMMU-v2-Linux-6.1

[Danny3]

Cool!
But I still hate AMD for refusing to enable SR-IOV on their GPUs!

[ms178]

With all the buzz about AMD in the data center, it is surprising that their software support for some of these hardware features is lacking behind a couple of generations.

[luno]

Cool!
But I still hate AMD for refusing to enable SR-IOV on their GPUs!

I wish they enable it and restrict to 1 VM :/

[Jabberwocky]

Cool!
But I still hate AMD for refusing to enable SR-IOV on their GPUs!

Context?

I mean AFAIK there are no consumer cards that support virtualization like SR-IOV. There are "workarounds" to unlock some Nvidia cards (see vgpu_unlock_hooks.c).

[edwaleni]

About time. AMD IOMMU issues, especially on the mobile Ryzens have been an irritant for a few years. First it was the 3500, recently the 4900H. Mostk Linux versions can handle it OK, but many Ubuntu based distros couldn't without hanging the PC.

[Times Two]

EPYC is specifically referenced. Are these mods specific to EPYC only?

[intelfx]

About time. AMD IOMMU issues, especially on the mobile Ryzens have been an irritant for a few years. First it was the 3500, recently the 4900H. Mostk Linux versions can handle it OK, but many Ubuntu based distros couldn't without hanging the PC.

There's nothing relevant to "AMD IOMMU issues", and nothing to hold your breath against. The patches in question are about IOMMU virtualization (i. e. nested virt).

[agd5f]

There's nothing relevant to "AMD IOMMU issues", and nothing to hold your breath against. The patches in question are about IOMMU virtualization (i. e. nested virt).

The problems with IOMMU on consumer hardware tends to stem from the fact that MS did not enable the IOMMU on windows until recently so very few OEMs actually validated the IOMMU on laptops from that time period if they were windows only. There were often sbios issues or platform features that assumed 1:1 physical memory access from devices.

[edwaleni]

There's nothing relevant to "AMD IOMMU issues", and nothing to hold your breath against. The patches in question are about IOMMU virtualization (i. e. nested virt).

While this patch may not do anything relevant to what I have experienced, the "about time" is the fact that there are patches coming through for the IOMMU which will require regression testing and certification, and more hardware testing in that area is welcome. They may find other things that *are* relevant.