Re: [PATCH] cpufreq: dt: Set default policy->transition_delay_ns

[Rafael J. Wysocki]

On Wednesday, June 28, 2017 09:44:55 AM Viresh Kumar wrote:
> On 27-06-17, 18:08, Rafael J. Wysocki wrote:
> > On Tue, Jun 27, 2017 at 6:20 AM, Viresh Kumar <viresh.kumar@xxxxxxxxxx> wrote:
> > > @Rafael: Will it be fine to lower down the value of LATENCY_MULTIPLIER?
> > 
> > We can do that, but then I think we need to compensate for the change
> > in the old governors code or there may be surprises.
> 
> Why shouldn't we change the value of LATENCY_MULTIPLIER for old
> governors as well? They use the same calculations and the sampling
> rate there is also this bad (like rate_limit_us).

On some systems.  On other systems it isn't.

> If we aren't going to change that for old governors, then we can
> create a local version of LATENCY_MULTIPLIER for schedutil I believe.

OK, so at least for intel_pstate and acpi-cpufreq we want a 10 ms default
which is what we have currently.

If you want to rework all that, make sure you preserve that.

Thanks,
Rafael