Messages in this thread | | | Subject | Re: [Resend][PATCH V2] cpufreq: intel_pstate: allow trace in passive mode | From | Srinivas Pandruvada <> | Date | Fri, 05 Jan 2018 13:06:46 -0800 |
| |
On Fri, 2018-01-05 at 12:15 -0800, Doug Smythies wrote: > >
[...]
> On 2017.12.18 16:25 Rafael J. Wysocki wrote: > > > > + from = cpu->pstate.current_pstate; > > > > + time = ktime_get(); > > > > + sample_taken = intel_pstate_sample(cpu, time); > > > > + > > > This is quite a bit of overhead for tracing. > Yes, it is a bit of added code, but without tracing abilities I > do not know how to investigate passive operation. > > > > > > > > > Why not fold the above two > > > statements in the next if () with conditional tracing? > No, I specifically want to do a trace sample, even if the target > is the same as last time. Why? Because we want to know the time > between calls to the driver, i.e. the duration. That information > is incredibly useful.
I am not saying you don't need trace. But you can do all processing when just trace is enabled. Which can be done by if (trace_pstate_sample_enabled())
The above function should return true when trace is enabled.
So in your v3,
in intel_cpufreq_trace() you can simply return if trace_pstate_sample_enabled() is false, without calling intel_pstate_sample().
Thanks, Srinivas
| |