On Tue Jun 23, 2026 at 7:28 AM JST, John Hubbard wrote: > On 6/22/26 12:10 AM, Alexandre Courbot wrote: >> The `Fsp` instance was only used in the Hopper+ boot path, and >> consequently built locally (and immediately dropped) in it. >> >> This worked well as a temporary measure, but the FSP is also needed in >> other parts of the driver, for instance vGPU. > > Or: "will be needed, in future patchsets", but not in this one. May I > suggest that we put this patch in one of those future patchsets, > instead of trying to do it here, in advance of anything using > it?
So we have two patchsets (from Zhi and Eliot) that eventually depend on this, and I am not sure which one would take this patch. :) Besides, I'd argue that this change also stands on its own; FSP is a sub-device of the GPU, we ephemerally instanciated it in the GH100 HAL because that's where we needed it, but from an architectural perspective its scope should be larger.
