[lkml]   [2018]   [Oct]   [3]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH RFC] cpuidle/drivers/menu: Remove get_loadavg in the performance multiplier
On 03/10/2018 10:58, Rafael J. Wysocki wrote:
> On Tuesday, September 25, 2018 11:38:44 AM CEST Daniel Lezcano wrote:
>> The function get_loadavg() returns almost always zero. To be more
>> precise, statistically speaking for a total of 1023379 times passing
>> to the function, the load is equal to zero 1020728 times, greater than
>> 100, 610 times, the remaining is between 0 and 5.
>> I'm putting in question this metric. Is it worth to keep it?
> The honest answer is that I'm not sure.
> Using the CPU load for driving idle state selection is questionable in general
> (there need not be any connection between the two in principle) and this
> particular function is questionable either. If it really makes no difference
> in addition to that, then yes it would be good to get rid of it.
> That said it looks like that may be quite workload-dependent, so maybe there
> are workloads where it does make a difference? Like networking or similar?

Perhaps initially the load was not the same:

commit 69d25870f20c4b2563304f2b79c5300dd60a067e
Author: Arjan van de Ven <>
Date: Mon Sep 21 17:04:08 2009 -0700

cpuidle: fix the menu governor to boost IO performance

The function get_loadavg() was:

static int get_loadavg(void)
unsigned long this = this_cpu_load();

return LOAD_INT(this) * 10 + LOAD_FRAC(this) / 10;


unsigned long this_cpu_load(void)
struct rq *this = this_rq();
return this->cpu_load[0];

Now we pass the load we got from get_iowait_load() which is:

void get_iowait_load(unsigned long *nr_waiters, unsigned long *load)
struct rq *rq = this_rq();
*nr_waiters = atomic_read(&rq->nr_iowait);
*load = rq->load.weight;

Colin Cross did some measurements a few years ago and concluded it is
zero most of the time, otherwise the number is so high it has no
meaning. The get_loadavg() call is removed from Android since 2011 [1].

I can hack my server (bi Xeon - 6 cores), add statistics regarding this
metric and let it run during several days, sure a lot of different
workloads will happen on it (compilation, network, throughput computing,
...) in a real use case manner. That is what I did with an ARM64
evaluation board.

However I won't be able to do that right now and that will take a bit of
time. If you think it is really useful, I can plan to do that in the
next weeks.

-- Daniel


<> │ Open source software for ARM SoCs

Follow Linaro: <> Facebook |
<!/linaroorg> Twitter |
<> Blog

 \ /
  Last update: 2018-10-03 12:26    [W:0.052 / U:1.508 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site