linux-kernel - Re: [Resend][PATCH V2] cpufreq: intel_pstate: allow trace in passive mode

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <1513639857.31113.12.camel@linux.intel.com>
Date:   Mon, 18 Dec 2017 15:30:57 -0800
From:   Srinivas Pandruvada <srinivas.pandruvada@...ux.intel.com>
To:     Doug Smythies <doug.smythies@...il.com>, rjw@...ysocki.net
Cc:     dsmythies@...us.net, linux-kernel@...r.kernel.org,
        linux-pm@...r.kernel.org
Subject: Re: [Resend][PATCH V2] cpufreq: intel_pstate: allow trace in
 passive mode

On Fri, 2017-12-15 at 16:43 -0800, Doug Smythies wrote:
> Allow use of the trace_pstate_sample trace function
> when the intel_pstate driver is in passive mode.
> Since the core_busy and scaled_busy fields are not
> used, and it might be desirable to know which path
> through the driver was used, either intel_cpufreq_target
> or intel_cpufreq_fast_switch, re-task the core_busy
> field as a flag indicator.
> 
> The user can then use the intel_pstate_tracer.py utility
> to summarize and plot the trace.
> 
> In Passive mode the driver is only called if there is
> a need to change the target frequency, so durations
> (time gaps between calls) can be very very long. The user
> needs to understand, and not be confused by, this limitation.
> 
> V2: prepare for resend. Rebase to current kernel, 4.15-rc3.
> Signed-off-by: Doug Smythies <dsmythies@...us.net>
> ---
>  drivers/cpufreq/intel_pstate.c | 50
> ++++++++++++++++++++++++++++++++++++++++--
>  1 file changed, 48 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/cpufreq/intel_pstate.c
> b/drivers/cpufreq/intel_pstate.c
> index 93a0e88..fe25d69 100644
> --- a/drivers/cpufreq/intel_pstate.c
> +++ b/drivers/cpufreq/intel_pstate.c
> @@ -1949,7 +1949,10 @@ static int intel_cpufreq_target(struct
> cpufreq_policy *policy,
>  {
>  	struct cpudata *cpu = all_cpu_data[policy->cpu];
>  	struct cpufreq_freqs freqs;
> -	int target_pstate;
> +	struct sample *sample;
> +	int target_pstate, from;
> +	u64 time;
> +	bool sample_taken;
>  
>  	update_turbo_state();
>  
> @@ -1969,12 +1972,32 @@ static int intel_cpufreq_target(struct
> cpufreq_policy *policy,
>  		break;
>  	}
>  	target_pstate = intel_pstate_prepare_request(cpu,
> target_pstate);
> +
> +	from = cpu->pstate.current_pstate;
> +	time = ktime_get();
> +	sample_taken = intel_pstate_sample(cpu, time);
> +
This is quite a bit of overhead for tracing. Why not fold the above two
statements in the next if () with conditional tracing?

>  	if (target_pstate != cpu->pstate.current_pstate) {
>  		cpu->pstate.current_pstate = target_pstate;
>  		wrmsrl_on_cpu(policy->cpu, MSR_IA32_PERF_CTL,
>  			      pstate_funcs.get_val(cpu,
> target_pstate));
>  	}
>  	freqs.new = target_pstate * cpu->pstate.scaling;
> +
> +	if (sample_taken) {
if (trace_pstate_sample_enabled() && sample_taken) {

> +		intel_pstate_calc_avg_perf(cpu);
> +		sample = &cpu->sample;
> +		trace_pstate_sample(0,
Not sure they are statement below are aligned correctly.

> +		0,
> +		from,
> +		cpu->pstate.current_pstate,
> +		sample->mperf,
> +		sample->aperf,
> +		sample->tsc,
> +		get_avg_frequency(cpu),
> +		fp_toint(cpu->iowait_boost * 100));
> +	}
> +
Same below in the intel_cpufreq_fast_switch().

Thanks,
Srinivas