Possible suspend/resume regression in .32-rc?

[Pavel Machek]

On Mon 2009-11-02 05:22:30, Haojian Zhuang wrote:
> On Sun, Nov 1, 2009 at 6:03 PM, Pavel Machek <pavel at ucw.cz> wrote:
> >> Hi!
> >>
> >> > > Is anyone using suspend/resume with a recent git mainline kernel on PXA
> >> > > or other ARM embedded boards? My platform used to suspend and resume
> >> > > just fine on 2.6.31 but now as I rebased it, it fails the resume part.
> >> > >
> >> > > Unfortunately, I can't bisect it as the platform is not mainline yet and
> >> > > so I always have mandatory patches (without my platform won't do
> >> > > anything) on top of the git repository. Which breaks the bisect logic.
> >> > >
> >> > > What puzzles me is that I see the current raising at wakeup time, so at
> >> > > least the processor seems to resume, but I can't see any serial console
> >> > > output, just like if the kernel crashed very early after wakeup.
> >> > > 'no_console_suspend' didn't help either.
> >> >
> >> > Ok, got it. The culprit is commit d2c37068 ("[ARM] pxa: initialize
> >> > default interrupt priority and use ICHP for IRQ handling"). Reverting it
> >> > make suspend/resume work again on my board.
> >> >
> >> > Haojian, Eric, could you have a look at this?
> >>
> >> Okay, patch is this one: I'll test reverting it shortly.
> >>
> >> commit d2c37068429b29d6549cf3486fc84b836689e122
> >> Author: Haojian Zhuang <haojian.zhuang at marvell.com>
> >> Date:   Wed Aug 19 19:49:31 2009 +0800
> >>
> >>     [ARM] pxa: initialize default interrupt priority and use ICHP for IRQ handling
> >>
> >>     Signed-off-by: Haojian Zhuang <haojian.zhuang at marvell.com>
> >>     Signed-off-by: Eric Miao <eric.y.miao at gmail.com>
> >
> > And yes, reverting it _does_ fix suspend on spitz.
> 
> Em, it's not caused by the IRQ patch.
> 
> The kernel is blocked in resume path. When console is resumed, IRQ is
> already disabled and system is blocked. Actually, IRQ shouldn't be
> disabled at here. Up to now, I only find which patch will cause this
> issue. But I can't find the best solution on it. The patch with issue
> is pasted in below.
> 
> So this issue is only occused when console suspend is enabled. If you
> enable no_console_suspend in command, you won't meet this issue. It
> seems that it's caused by removing termios setting in
> uart_resume_port() in the below patch. If I add these code back, the
> issue doesn't occur any more.

Given that it hangs very early, in arch_suspend_enable_irqs() (see my
other mail), I don't trust your analysis.

> Deepak,
> Could you help to investigate this issue also?
> 
> By the way, the suggested quick test command is in below. If we tried
> these commands for 20 times, the issue would occur. I only test it in
> PXA silicon. I don't know the symptom on other silicons.
> $ echo devices > /sys/power/pm_test
> $ echo mem > /sys/power/state

I'm not using serial console on spitz, and I have never had successful
resume with the patch applied.

The original patch is also poorly changelogged. What is the advantage
of ICHP and why do we want default priorities?

								Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html