On 10/24/26 4:57 PM, Anna Maniscalco wrote: > On 10/24/26 1:36 AM, Dmitry Baryshkov wrote: >> On Fri, Oct 23, 2026 at 03:16:27AM +0530, Akhil P Oommen wrote:
[...] >>>> One thing sashiko is pointing out is that new dt should work with hold >>>> kernel so, in general moving sid1 from there would break it since old >>>> kernels don't know how to handle them separately. >> The rule is the opposite: new kernels must continue to work with the old >> DTs. If the new DT breaks with the older kernel, just specify it in the >> cover letter so that Bjorn doesn't merge DT changes before the driver >> changes find the way into his tree. > Ah ok thanks for clarifying that. >> >>> + Bjorn/Konrad >>> >>> Not sure how strict is the requirement for compatibility with older kernel >>> version. An >>> option in your case could be just do an override in the board dts file of >>> the >>> Steam Frame. >> Anyway, would it be a problem? Does GPU normally use SID 1? Or is it >> only used by the LPAC? > It is a problem because, even if nothing touches the LPAC ring, it's firmware > will still initialize and that will cause page faults if sid1 isn't setup. We're slowly coming to terms with the fact that all of our multimedia iommu setup was terribly wrong all the time. Venus/Iris recently had a flurry of breaking changes. We have to follow suit. Konrad
