Re: [intel-pstate driver regression] processor frequency very high even if in idle

[Srinivas Pandruvada]

On Thu, 2016-03-31 at 17:43 +0200, Rafael J. Wysocki wrote:
> On Thursday, March 31, 2016 05:25:18 PM JÃrg Otte wrote:
> > 2016-03-31 13:42 GMT+02:00 Rafael J. Wysocki <rjw@xxxxxxxxxxxxx>:
> > > On Thursday, March 31, 2016 11:05:56 AM JÃrg Otte wrote:
> > > 
> > > [cut]
> > > 
> > > > > 
> > > > 
> > > > Yes, works for me.
> > > > 
> > > > CPUID(7): No-SGX
> > > > ÂÂÂÂÂCPU Avg_MHzÂÂÂBusy% Bzy_MHz TSC_MHz
> > > > ÂÂÂÂÂÂÂ-ÂÂÂÂÂÂ11ÂÂÂÂ0.66 1682 2494
> > > > ÂÂÂÂÂÂÂ0ÂÂÂÂÂÂ11ÂÂÂÂ0.60 1856 2494
> > > > ÂÂÂÂÂÂÂ1ÂÂÂÂÂÂÂ6ÂÂÂÂ0.34ÂÂÂÂ1898ÂÂÂÂ2494
> > > > ÂÂÂÂÂÂÂ2ÂÂÂÂÂÂ13ÂÂÂÂ0.82ÂÂÂÂ1628ÂÂÂÂ2494
> > > > ÂÂÂÂÂÂÂ3ÂÂÂÂÂÂ13ÂÂÂÂ0.87ÂÂÂÂ1528ÂÂÂÂ2494
> > > > ÂÂÂÂÂCPU Avg_MHzÂÂÂBusy% Bzy_MHz TSC_MHz
> > > > ÂÂÂÂÂÂÂ-ÂÂÂÂÂÂÂ6ÂÂÂÂ0.58ÂÂÂÂÂ963ÂÂÂÂ2494
> > > > ÂÂÂÂÂÂÂ0ÂÂÂÂÂÂÂ8ÂÂÂÂ0.83ÂÂÂÂÂ957ÂÂÂÂ2494
> > > > ÂÂÂÂÂÂÂ1ÂÂÂÂÂÂÂ1ÂÂÂÂ0.08ÂÂÂÂÂ984ÂÂÂÂ2494
> > > > ÂÂÂÂÂÂÂ2ÂÂÂÂÂÂ10ÂÂÂÂ1.04ÂÂÂÂÂ975ÂÂÂÂ2494
> > > > ÂÂÂÂÂÂÂ3ÂÂÂÂÂÂÂ3ÂÂÂÂ0.35ÂÂÂÂÂ934ÂÂÂÂ2494
> > > > 
> > > 
> 
> [cut]
> 
> > > 
> > 
> > No, this patch doesn't help.
> 
> Well, more work to do then.
> 
> I've just noticed a bug in this patch, which is not relevant for the
> results,
> but below is a new version.
> 
> > CPUID(7): No-SGX
> > ÂÂÂÂÂÂCPU Avg_MHzÂÂÂBusy% Bzy_MHz TSC_MHz
> > ÂÂÂÂÂÂÂ-ÂÂÂÂÂÂÂ8ÂÂÂÂ0.32ÂÂÂÂ2507ÂÂÂÂ2495
> > ÂÂÂÂÂÂÂ0ÂÂÂÂÂÂ13ÂÂÂÂ0.53ÂÂÂÂ2505ÂÂÂÂ2495
> > ÂÂÂÂÂÂÂ1ÂÂÂÂÂÂÂ3ÂÂÂÂ0.11ÂÂÂÂ2523ÂÂÂÂ2495
> > ÂÂÂÂÂÂÂ2ÂÂÂÂÂÂÂ1ÂÂÂÂ0.06ÂÂÂÂ2555ÂÂÂÂ2495
> > ÂÂÂÂÂÂÂ3ÂÂÂÂÂÂ15ÂÂÂÂ0.59ÂÂÂÂ2500ÂÂÂÂ2495
> > ÂÂÂÂÂCPU Avg_MHzÂÂÂBusy% Bzy_MHz TSC_MHz
> > ÂÂÂÂÂÂÂ-ÂÂÂÂÂÂÂ8ÂÂÂÂ0.33ÂÂÂÂ2486ÂÂÂÂ2495
> > ÂÂÂÂÂÂÂ0ÂÂÂÂÂÂ12ÂÂÂÂ0.50ÂÂÂÂ2482ÂÂÂÂ2495
> > ÂÂÂÂÂÂÂ1ÂÂÂÂÂÂÂ5ÂÂÂÂ0.22ÂÂÂÂ2489ÂÂÂÂ2495
> > ÂÂÂÂÂÂÂ2ÂÂÂÂÂÂÂ1ÂÂÂÂ0.04ÂÂÂÂ2492ÂÂÂÂ2495
> > ÂÂÂÂÂÂÂ3ÂÂÂÂÂÂ15ÂÂÂÂ0.59ÂÂÂÂ2487ÂÂÂÂ2495
> 
> Please apply the patch below and take a (1s or so) trace from the
> pstate_sample
> tracepoint (under /sys/kernel/debug/tracing/events/power/ on my
> systems).
> 

Jorg,

If you want to know how to trace
# cd /sys/kernel/debug/tracing/
# echo 1 > events/power/pstate_sample/enable
# echo 1 > events/power/cpu_frequency/enable
# cat trace


> Then please apply the revert instead of it and take a trace from that
> tracepoint
> again and send both of the traces to me.
> 
> ---
> From: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
> Subject: [PATCH] intel_pstate: Do not set utilization update hook too
> early
> 
> The utilization update hook in the intel_pstate driver is set too
> early, as it only should be set after the policy has been fully
> initialized by the core.ÂÂThat may cause intel_pstate_update_util()
> to use incorrect data and put the CPUs into incorrect P-states as
> a result.
> 
> To prevent that from happening, make intel_pstate_set_policy() set
> the utilization update hook instead of intel_pstate_init_cpu() so
> intel_pstate_update_util() only runs when all things have been
> initialized as appropriate.
> 
> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
> ---
> Âdrivers/cpufreq/intel_pstate.c |ÂÂÂ27 +++++++++++++++++++--------
> Â1 file changed, 19 insertions(+), 8 deletions(-)
> 
> Index: linux-pm/drivers/cpufreq/intel_pstate.c
> ===================================================================
> --- linux-pm.orig/drivers/cpufreq/intel_pstate.c
> +++ linux-pm/drivers/cpufreq/intel_pstate.c
> @@ -1103,7 +1103,6 @@ static int intel_pstate_init_cpu(unsigne
> Â	intel_pstate_sample(cpu, 0);
> Â
> Â	cpu->update_util.func = intel_pstate_update_util;
> -	cpufreq_set_update_util_data(cpunum, &cpu->update_util);
> Â
> Â	pr_debug("intel_pstate: controlling: cpu %d\n", cpunum);
> Â
> @@ -1122,18 +1121,29 @@ static unsigned int intel_pstate_get(uns
> Â	return get_avg_frequency(cpu);
> Â}
> Â
> +static void intel_pstate_set_update_util_hook(unsigned int cpu)
> +{
> +	cpufreq_set_update_util_data(cpu, &all_cpu_data[cpu]-
> >update_util);
> +}
> +
> +static void intel_pstate_clear_update_util_hook(unsigned int cpu)
> +{
> +	cpufreq_set_update_util_data(cpu, NULL);
> +	synchronize_sched();
> +}
> +
> Âstatic int intel_pstate_set_policy(struct cpufreq_policy *policy)
> Â{
> Â	if (!policy->cpuinfo.max_freq)
> Â		return -ENODEV;
> Â
> +	intel_pstate_clear_update_util_hook(policy->cpu);
> +
> Â	if (policy->policy == CPUFREQ_POLICY_PERFORMANCE &&
> Â	ÂÂÂÂpolicy->max >= policy->cpuinfo.max_freq) {
> Â		pr_debug("intel_pstate: set performance\n");
> Â		limits = &performance_limits;
> -		if (hwp_active)
> -			intel_pstate_hwp_set(policy->cpus);
> -		return 0;
> +		goto out;
> Â	}
> Â
> Â	pr_debug("intel_pstate: set powersave\n");
> @@ -1163,6 +1173,9 @@ static int intel_pstate_set_policy(struc
> Â	limits->max_perf = div_fp(int_tofp(limits->max_perf_pct),
> Â				ÂÂint_tofp(100));
> Â
> + out:
> +	intel_pstate_set_update_util_hook(policy->cpu);
> +
> Â	if (hwp_active)
> Â		intel_pstate_hwp_set(policy->cpus);
> Â
> @@ -1187,8 +1200,7 @@ static void intel_pstate_stop_cpu(struct
> Â
> Â	pr_debug("intel_pstate: CPU %d exiting\n", cpu_num);
> Â
> -	cpufreq_set_update_util_data(cpu_num, NULL);
> -	synchronize_sched();
> +	intel_pstate_clear_update_util_hook(cpu_num);
> Â
> Â	if (hwp_active)
> Â		return;
> @@ -1455,8 +1467,7 @@ out:
> Â	get_online_cpus();
> Â	for_each_online_cpu(cpu) {
> Â		if (all_cpu_data[cpu]) {
> -			cpufreq_set_update_util_data(cpu, NULL);
> -			synchronize_sched();
> +			intel_pstate_clear_update_util_hook(cpu);
> Â			kfree(all_cpu_data[cpu]);
> Â		}
> Â	}
>