On Tue, Sep 21, 2021 at 10:18 PM Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote: > > On Tue, Sep 21 2021 at 20:05, Rafael J. Wysocki wrote: > > On 9/19/2021 2:14 AM, Thomas Gleixner wrote: > >> What's the proper way to figure out whether PC10 is supported? > > > > I can't say without research. I think it'd be sufficient to check if > > C10 is supported, because asking for it is the only way to get PC10. > > Do we have a common function for that or do I need to implement the > gazillionst CPUID query for that? intel_idle has intel_idle_verify_cstate() that works on MWAIT substates from CPUID. It looks like this could be reused. > > However, even if it is supported, the problem is not there until the > > kernel asks for C10. So instead, I'd disable the TSC watchdog on the > > first attempt to ask the processor for C10 from the cpuidle code and I'd > > do that from the relevant drivers (intel_idle and ACPI idle). > > > > There would be no TSC watchdog for the C10 users, but wouldn't that be a > > fair game? > > Not really because that makes any other HPET usage broken as well and we > can't pull the rug under that. So we are better off to disable it > upfront. Allright.
