Re: [PATCH] idle, thermal, acpi: Remove home grown idle implementations

From: Peter Zijlstra
Date: Thu Jun 05 2014 - 02:55:30 EST


On Wed, Jun 04, 2014 at 01:58:12AM -0700, Jacob Pan wrote:
> On Wed, 4 Jun 2014 10:54:18 +0200
> Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> >
> > I'm still sitting on this patch. Jacub you were going to make it play
> > nice with QoS?
> >
> I had a patchset to work through system PM QOS and still maintain the
> idle injection efficiency. When I saw you did not merge the patch
> below, I thought you have abandoned it :)

I was waiting for you to do the QoS bits :-)

> The only issue as per our last discussion is the lack of notification
> when PM QOS cannot be met. But that is intrinsic to PM QOS itself.
>
> I also consulted with Arjan and looked at directly intercept with
> intel_idle since both intel_powerclamp and intel_idle are arch specific
> drivers. But I think that is hard to do at per idle period basis,
> since we should still allow "natural" idle during the forced idle time.
>
> So, I think we can take a two stepped approach,
> 1. integrate your patch with a
> updated version of https://lkml.org/lkml/2013/11/26/534 such that there
> is no performance/efficiency regression.
> 2. add notification mechanism to system qos when constraints cannot be
> met.

That's fine with me; can you respin those bits?

Attachment: pgpJOUdZXpIZI.pgp
Description: PGP signature