Re: [PATCH 0/6] Add latency_nice priority

[Qais Yousef]

+CC Wei

On 03/28/22 14:56, Vincent Guittot wrote:
> Hi Dietmar,
> 
> 
> On Mon, 28 Mar 2022 at 11:24, Dietmar Eggemann <dietmar.eggemann@xxxxxxx> wrote:
> >
> > On 11/03/2022 17:14, Vincent Guittot wrote:
> > > This patchset restarts the work about adding a latency nice priority to
> > > describe the latency tolerance of cfs tasks.
> > >
> > > The patches [1-4] have been done by Parth:
> > > https://lore.kernel.org/lkml/20200228090755.22829-1-parth@xxxxxxxxxxxxx/
> > >
> > > I have just rebased and moved the set of latency priority outside the
> > > priority update. I have removed the reviewed tag because the patches
> > > are 2 years old.
> > >
> > > The patches [5-6] use latency nice priority to decide if a cfs task can
> > > preempt the current running task. Patch 5 gives some tests results with
> > > cyclictests and hackbench to highlight the benefit of latency nice
> > > priority for short interactive task or long intensive tasks.
> >
> > The Android specific `latency_nice` (in Android `latency_sensitive`
> > [latency_nice < 0]) use case `Skip energy aware task placement` favors
> > an idle CPU over the EAS search path for a `latency_sensitive` task.
> >
> > https://lkml.kernel.org/r/2aa4b838-c298-ec7d-08f3-caa50cc87dc2@xxxxxxx
> >
> > This is Android proprietary code similar to what we have in
> > find_idlest_group_cpu() in mainline.
> > We talked to the Android folks last week and IMHO they are not convinced
> > that they can switch this to the proposed `latency_nice->tweak
> > preemption` use case.
> 
> Thanks for discussing this with Android folks. It's not always easy to
> change the behavior of a product and I would be interested to discuss
> this with them. Sometimes you need a PoC to get convinced

I think it's good to clarify for me at least here whether you intend this as
a replacement for disable EAS and revert to CAS or you see this as an
additional thing? As I understood from the discussion we had on the cover
letter, this is an additional improvement and not intended to replace any of
the previous use cases we brought up before.

Wei/Quentin, any thoughts on this?

Thanks

--
Qais Yousef