[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20191022090334.GA85349@google.com>
Date: Tue, 22 Oct 2019 10:03:34 +0100
From: Quentin Perret <qperret@...gle.com>
To: Thara Gopinath <thara.gopinath@...aro.org>
Cc: mingo@...hat.com, peterz@...radead.org, ionela.voinescu@....com,
vincent.guittot@...aro.org, rui.zhang@...el.com,
edubezval@...il.com, linux-kernel@...r.kernel.org,
amit.kachhap@...il.com, javi.merino@...nel.org,
daniel.lezcano@...aro.org, kernel-team@...roid.com
Subject: Re: [Patch v3 7/7] sched: thermal: Enable tuning of decay period
Hi Thara,
On Monday 21 Oct 2019 at 17:03:56 (-0400), Thara Gopinath wrote:
> On 10/15/2019 06:14 AM, Quentin Perret wrote:
> > Hi Thara,
> >
> > On Sunday 13 Oct 2019 at 20:58:25 (-0400), Thara Gopinath wrote:
> >> diff --git a/kernel/sysctl.c b/kernel/sysctl.c
> >> index 00fcea2..5056c08 100644
> >> --- a/kernel/sysctl.c
> >> +++ b/kernel/sysctl.c
> >> @@ -376,6 +376,13 @@ static struct ctl_table kern_table[] = {
> >> .mode = 0644,
> >> .proc_handler = proc_dointvec,
> >> },
> >> + {
> >> + .procname = "sched_thermal_decay_coeff",
> >> + .data = &sysctl_sched_thermal_decay_coeff,
> >> + .maxlen = sizeof(unsigned int),
> >> + .mode = 0644,
> >> + .proc_handler = proc_dointvec,
> >
> > Perhaps change this for 'sched_proc_update_handler' with min and max
> > values ? Otherwise userspace is allowed to write nonsensical values
> > here. And since sysctl_sched_thermal_decay_coeff is used to shift, this
> > can lead to an undefined behaviour.
> Will do
> >
> > Also, could we take this sysctl out of SCHED_DEBUG ? I expect this to be
> > used/tuned on production devices where SCHED_DEBUG should theoretically
> > be off.
>
> I will take it out of SCHED_DEBUG. I am wondering if this should be
> a runtime control at all. Because this is a shift this changes the
> accumulating window for the thermal pressure signal. A runtime change
> will not guarantee a clean start of the window. May be I should make
> this a config option.
I'd personally prefer if it wan't a Kconfig option. We'd like to make
Android devices (which are going to use this) work with a Generic Kernel
Image, which means there will be a single config for everyone. But I
expect this knob to be tuned to different values depending on the SoC.
If you really don't want a sysctl, perhaps a cmdline option could work ?
Thanks,
Quentin
Powered by blists - more mailing lists