RE: [RFC v2] arm,x86,fs/resctrl: Generic schema description Proof of Concept

[Luck, Tony]

> It seems that taking the weakest link (minimum) across all sources is a pervasive
> convention in resctrl, which can be used to avoid out-of-bounds 
> CLOS/RMID access.
> If a future platform does not support the CPUID/MSR interfaces, we can only rely on
> the ACPI table to get the max RMID/CLOSID.

The CPUID enumeration will always define the maximum values that can be written
to the IA32_PQR_ASSOC.{CLOS,RMID} fields regardless of whether MSR interfaces
are provided. There should[1] be a #GP fault if the OS attempts to write a larger value
to either field.

If an ACPI table (or VSEC enumeration) supplies some larger values they should be
over-ridden by limits from CPUID.

-Tony

[1] I've a vague recollection that some early RDT supporting Xeon didn't #GP for out
of range RMID ... but that might have only been on a pre-production stepping.