Re: [PATCH] perf: arm_spe: Support heterogeneous ACPI systems

[Jeremy Linton]

Hi,

On 10/15/26 11:20 PM, Johnny Fan wrote:
> On Thu, Oct 15, 2026 at 10:58:10AM -0500, Jeremy Linton wrote:
> > I don't think the GICC is sufficient, you have to consider actual
> > heterogeneous machines that have differing SPE implementations with the
> > same PMSVer, which mean you have to keep the PPTT hetid check, and set
> > the mask based on both it and the GSIV. Then, like DT create multiple
> > platform devices. Which is true with the current patch as well, the
> > hetid still needs to be there even if you have differing PPI's because
> > it could have more than two differing implementations.
>
> Following that direction, the question on ACPI is how each registered
> SPE device gets its supported_cpus: the platform devices are created
> without an fwnode, so platform_get_irq_affinity() falls back to
> cpu_possible_mask today, regardless of which CPUs actually implement
> SPE. That's the plumbing question I asked Marc about -- would be good
> to know if you have a preference here too.


I may not be fully understanding what your asking, but you could just 
attach a cpumask via platform_device_add_data(), or similar method, then 
retrieve it in the driver(s) and use it to set the affinity.