Re: [PATCH 3/3] intel_pstate: Clean up get_target_pstate_use_performance()

[Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>]

On Sat, 2016-05-07 at 01:47 +0200, Rafael J. Wysocki wrote:
> From: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> 
> The way the code in get_target_pstate_use_performance() is arranged
> and the comments in there are totally confusing, so modify them to
> reflect what's going on.
> 
> The results of the computations should be the same as before.
> 
> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>

> ---
>  drivers/cpufreq/intel_pstate.c |   32 +++++++++++++-----------------
> --
>  1 file changed, 13 insertions(+), 19 deletions(-)
> 
> Index: linux-pm/drivers/cpufreq/intel_pstate.c
> ===================================================================
> --- linux-pm.orig/drivers/cpufreq/intel_pstate.c
> +++ linux-pm/drivers/cpufreq/intel_pstate.c
> @@ -1241,43 +1241,37 @@ static inline int32_t get_target_pstate_
>  
>  static inline int32_t get_target_pstate_use_performance(struct
> cpudata *cpu)
>  {
> -	int32_t core_busy, max_pstate, current_pstate, sample_ratio;
> +	int32_t perf_scaled, sample_ratio;
>  	u64 duration_ns;
>  
>  	/*
> -	 * core_busy is the ratio of actual performance to max
> -	 * max_pstate is the max non turbo pstate available
> -	 * current_pstate was the pstate that was requested during
> -	 * 	the last sample period.
> -	 *
> -	 * We normalize core_busy, which was our actual percent
> -	 * performance to what we requested during the last sample
> -	 * period. The result will be a percentage of busy at a
> -	 * specified pstate.
> +	 * perf_scaled is the average performance during the last
> sampling
> +	 * period (in percent) scaled by the ratio of the P-state
> requested
> +	 * last time to the maximum P-state.  That measures the
> system's
> +	 * response to the previous P-state selection.
>  	 */
> -	core_busy = 100 * cpu->sample.core_avg_perf;
> -	max_pstate = cpu->pstate.max_pstate_physical;
> -	current_pstate = cpu->pstate.current_pstate;
> -	core_busy = mul_fp(core_busy, div_fp(max_pstate,
> current_pstate));
> +	perf_scaled = div_fp(cpu->pstate.max_pstate_physical,
> +			     cpu->pstate.current_pstate);
> +	perf_scaled = mul_fp(perf_scaled, 100 * cpu-
> >sample.core_avg_perf);
>  
>  	/*
>  	 * Since our utilization update callback will not run unless
> we are
>  	 * in C0, check if the actual elapsed time is significantly
> greater (3x)
>  	 * than our sample interval.  If it is, then we were idle
> for a long
> -	 * enough period of time to adjust our busyness.
> +	 * enough period of time to adjust our performance metric.
>  	 */
>  	duration_ns = cpu->sample.time - cpu->last_sample_time;
>  	if ((s64)duration_ns > pid_params.sample_rate_ns * 3) {
>  		sample_ratio = div_fp(pid_params.sample_rate_ns,
> duration_ns);
> -		core_busy = mul_fp(core_busy, sample_ratio);
> +		perf_scaled = mul_fp(perf_scaled, sample_ratio);
>  	} else {
>  		sample_ratio = div_fp(100 * cpu->sample.mperf, cpu-
> >sample.tsc);
>  		if (sample_ratio < int_tofp(1))
> -			core_busy = 0;
> +			perf_scaled = 0;
>  	}
>  
> -	cpu->sample.busy_scaled = core_busy;
> -	return cpu->pstate.current_pstate - pid_calc(&cpu->pid,
> core_busy);
> +	cpu->sample.busy_scaled = perf_scaled;
> +	return cpu->pstate.current_pstate - pid_calc(&cpu->pid,
> perf_scaled);
>  }
>  
>  static inline void intel_pstate_update_pstate(struct cpudata *cpu,
> int pstate)
>