Re: [PATCH v3] spapr: Add a new level of NUMA for GPUs

[Daniel Henrique Barboza]

On 7/16/20 1:00 PM, Reza Arbab wrote:
> On Thu, Jul 16, 2020 at 06:42:11AM -0300, Daniel Henrique Barboza wrote:
> > Which would translate here to:
> >
> >        uint32_t associativity[] = {
> >            cpu_to_be32(0x4),
> >            cpu_to_be32(nvslot->numa_id),
> >            cpu_to_be32(nvslot->numa_id),
> >            cpu_to_be32(nvslot->numa_id),
> >            cpu_to_be32(nvslot->numa_id),
> >        };
>
> Sure, that's how it originally was in v1 of the patch.

Yeah, Greg commented this in v2 about this chunk:

------------
This is a guest visible change. It should theoretically be controlled
with a compat property of the PHB (look for "static GlobalProperty" in
spapr.c). But since this code is only used for GPU passthrough and we
don't support migration of such devices, I guess it's okay. Maybe just
mention it in the changelog.
--------------

By all means you can mention in changelog/code comment why the associativity
of the GPU is nvslot->numa_id 4 times in a row, but I believe this
format is still clearer for us to understand. It also makes it equal
to skiboot.

And it deprecates the SPAPR_GPU_NUMA_ID macro, allowing us to use its value
(1) for other internal purposes regarding NUMA without collision with the
GPU semantics.



Thanks,


DHB


> I'll send a v4 today. It's been a while so I need to rebase anyway.