From mboxrd@z Thu Jan  1 00:00:00 1970
From: Peter Zijlstra <peterz@infradead.org>
Subject: Re: [PATCH v2] cpuidle: Add 'above' and 'below' idle state metrics
Date: Mon, 10 Dec 2018 13:21:04 +0100
Message-ID: <20181210122104.GL5289@hirez.programming.kicks-ass.net>
References: <3514439.dzOWKx1Cjx@aspire.rjw.lan>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Return-path: <linux-kernel-owner@vger.kernel.org>
Content-Disposition: inline
In-Reply-To: <3514439.dzOWKx1Cjx@aspire.rjw.lan>
Sender: linux-kernel-owner@vger.kernel.org
To: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: Linux PM <linux-pm@vger.kernel.org>, Doug Smythies <dsmythies@telus.net>, LKML <linux-kernel@vger.kernel.org>, Linux Documentation <linux-doc@vger.kernel.org>, Daniel Lezcano <daniel.lezcano@linaro.org>, Giovanni Gherdovich <ggherdovich@suse.cz>, Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
List-Id: linux-pm@vger.kernel.org

On Mon, Dec 10, 2018 at 12:30:23PM +0100, Rafael J. Wysocki wrote:
> From: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> 
> Add two new metrics for CPU idle states, "above" and "below", to count
> the number of times the given state had been asked for (or entered
> from the kernel's perspective), but the observed idle duration turned
> out to be too short or too long for it (respectively).
> 
> These metrics help to estimate the quality of the CPU idle governor
> in use.
> 
> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

> @@ -260,6 +262,33 @@ int cpuidle_enter_state(struct cpuidle_d
>  		dev->last_residency = (int)diff;
>  		dev->states_usage[entered_state].time += dev->last_residency;
>  		dev->states_usage[entered_state].usage++;
> +
> +		if (diff < drv->states[entered_state].target_residency) {
> +			for (i = entered_state - 1; i >= 0; i--) {
> +				if (drv->states[i].disabled ||
> +				    dev->states_usage[i].disable)
> +					continue;
> +
> +				/* Shallower states are enabled, so update. */
> +				dev->states_usage[entered_state].above++;
> +				break;
> +			}
> +		} else if (diff > delay) {
> +			for (i = entered_state + 1; i < drv->state_count; i++) {
> +				if (drv->states[i].disabled ||
> +				    dev->states_usage[i].disable)
> +					continue;
> +
> +				/*
> +				 * Update if a deeper state would have been a
> +				 * better match for the observed idle duration.
> +				 */
> +				if (diff - delay >= drv->states[i].target_residency)
> +					dev->states_usage[entered_state].below++;
> +
> +				break;
> +			}
> +		}

One question on this; why is this tracked unconditionally?

Would not a tracepoint be better?; then there is no overhead in the
normal case where nobody gives a crap about these here numbers.