Re: [PATCH v10 clocksource 1/7] clocksource: Provide module parameters to inject delays in watchdog

From: Thomas Gleixner
Date: Wed Apr 28 2021 - 10:37:46 EST


On Wed, Apr 28 2021 at 22:24, Luming Yu wrote:
> On Wed, Apr 28, 2021 at 9:57 PM Paul E. McKenney <paulmck@xxxxxxxxxx> wrote:
>> Therefore, the code marks the clock unstable if it has four bad reads
>> in a row, as it should.
>
> The hard problem to solve is tsc is still in good shape and it can be verified
> with a quick cross check with other thread/core's tsc counts in the
> injected situation or in real life case
> to prove if it is truly a tsc problem or reference clock's problem of
> the watchdog.
>
> Ideally, we could factor out hard-to-debug unstable tsc problems from
> clock source watchdog problems
> and get less and less tsc sightings caused by clock source watchdog.

The only way to fix this is at the hardware level. Everything else is
wishful thinking.

Thanks,

tglx