Re: tip.today - scheduler bam boom crash (cpu hotplug)

[Peter Zijlstra <peterz@infradead.org>]

On Wed, Mar 01, 2017 at 09:39:18PM +0800, Wanpeng Li wrote:
> 2017-02-28 16:11 GMT+08:00 Wanpeng Li <kernellwp@gmail.com>:
> > 2017-02-28 16:08 GMT+08:00 Peter Zijlstra <peterz@infradead.org>:
> >> On Tue, Feb 28, 2017 at 09:51:07AM +0800, Wanpeng Li wrote:
> >>> diff --git a/arch/x86/kernel/cpu/amd.c b/arch/x86/kernel/cpu/amd.c
> >>> index 4e95b2e..ed8eda4 100644
> >>> --- a/arch/x86/kernel/cpu/amd.c
> >>> +++ b/arch/x86/kernel/cpu/amd.c
> >>> @@ -557,7 +557,7 @@ static void early_init_amd(struct cpuinfo_x86 *c)
> >>>          set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
> >>>          if (check_tsc_unstable())
> >>>              clear_sched_clock_stable();
> >>> -    } else {
> >>> +    } else if (!boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
> >>>          clear_sched_clock_stable();
> >>>      }
> >>
> >> That's wrong, you can have HYPERVISOR and still use
> >> native_sched_clock() (lguest does that for one).
> >
> > My posting delay to check this in tsc_init().
> 
> Does it make sense?

It makes confusion from where I'm sitting. Please look at that latest
patch I send. I think it should work, but in case you have doubts, try
and find the holes.

  http://lkml.kernel.org/r/20170228080855.GV6515@twins.programming.kicks-ass.net

That marries the sched_clock stable to the existing TSC stable machinery
and removes a bunch of quirks.