Re: [REGRESSION 3.10.1] System does not wake up from suspend

[SÃren Brinkmann]

On Tue, Jul 16, 2013 at 10:38:20PM -0700, SÃren Brinkmann wrote:
> On Wed, Jul 17, 2013 at 10:27:15AM +0530, Srivatsa S. Bhat wrote:
> > On 07/17/2013 08:27 AM, SÃren Brinkmann wrote:
> > > On Mon, Jul 15, 2013 at 12:20:19PM +0530, Srivatsa S. Bhat wrote:
> > >> On 07/15/2013 10:24 AM, Willy Tarreau wrote:
> > >>> On Sun, Jul 14, 2013 at 09:36:25PM -0700, SÃren Brinkmann wrote:
> > >>>> Hi Greg,
> > >>>>
> > >>>> since upgrading to 3.10.1 I find my system not waking up from suspend
> > >>>> from time to time. Unfortunately, I couldn't figure out how to reliably
> > >>>> reproduce the issue. Sometimes it works just fine.
> > >>>>
> > >>>> All I know so far:
> > >>>> On 3.10 I had my system completely freezing a few times. But that did
> > >>>> not seem to be related to suspend/resume. I never experienced the
> > >>>> behavior described below on 3.10.
> > >>>>
> > >>>> On 3.10.1 I hit this bug a few times, that my system did not wake up.
> > >>>> I.e. the screen stayed black and the power LED kept flashing. But
> > >>>> nothing brought the system back up and I had to longpress power to reset
> > >>>> the system.
> > >>>> I had none of the system freezes like the ones described above anymore.
> > >>>
> > >>> Fortunately, 3.10.1 is very small. The only patch it contains which is
> > >>> related to suspend is 18/19 (cpufreq ondemand governor). Maybe you're
> > >>> having a problem with cpufreq on this machine in fact.
> > >>>
> > >>> You may want to try to force it to performance mode or powersave mode
> > >>> and try for a while, it is possible that both 3.10 and 3.10.1 will work
> > >>> fine.
> > >>>
> > >>
> > >> Or, you could try applying the patch shown below on top of 3.10.1 and
> > >> see if it fixes the suspend/resume regression for you.
> > >>
> > >> https://lkml.org/lkml/2013/7/11/661
> > >>
> > >> (Just apply patch 1, not the entire patchset).
> > > So, this didn't end well. With that patch applied I end up in a kernel
> > > panic on suspend, reproducible - well two out of two is a small test set,
> > > but anyway. Unfortunately too late to get any information about it
> > > and my logs don't show anything either. But it's seems worse than the
> > > rather occasional hangs on resume on the vanilla 3.10.1 kernel.
> > > 
> > 
> > Kernel panic? Weird.. I suspect something _else_ is wrong. I suggest trying
> > out the following:
> > 
> > 1. My patches only touch cpufreq. So completely turn off cpufreq by setting
> >    CONFIG_CPU_FREQ=n. Then try suspend/resume and see how it goes. If it
> >    fails, that shows that something _else_ is wrong and you can try a git
> >    bisect for that.
> I changed my config like shown below. That one has a two out of two
> success rate for suspend/resume so far.
> 
> > 
> >    However, if suspend/resume succeeds with CONFIG_CPU_FREQ=n, then try out
> >    step 2 below.
> > 
> > 2. Apply patch 1 (not the entire patchset)
> >    https://lkml.org/lkml/2013/7/11/661
> That's the one I already have in my tree
> 
> > 
> >    On top of that, apply this patch:
> >    https://patchwork.kernel.org/patch/2827512/
> This one does not apply. I'll see if I can make it apply.
Got it ported.
So, with that additional patch and going back to my original config
(with cpufreq included), I'm currently at 3 out of 3 successful
suspend/resume cycles.
Unless the odds are playing an ill game with me, you should be close to
the actual culprit.

	SÃren

Attachment: pgp00000.pgp
Description: PGP signature