[Pm-utils] upower/pm-utils deadlock when pm-suspend called by upowerd

[Victor Lowther]

On Tue, 2010-06-22 at 18:33 -0500, Victor Lowther wrote:
> On Tue, 2010-06-22 at 12:54 +0200, Vincent Untz wrote:
> > Hi,
> > 
> > (I'm not subscribed, please cc me if you need input from me :-))
> > 
> > There's a nice deadlock when pm-suspend is called by upowerd:
> > 
> >  + upowerd calls pm-suspend with g_spawn_sync(). That's because it wants
> >    to know if the call failed to return an error via dbus if needed.
> > 
> >  + pm-suspend suspends, everything is happy. Then it resumes.
> > 
> >  + the /usr/lib/pm-utils/sleep.d/00powersave hook calls pm-powersave
> >    with no argument
> > 
> >  + pm-powersave needs to know if the computer is on battery or not. So
> >    it calls on_ac_power.
> > 
> >  + on_ac_power knows that upower is cool, so it asks upower via dbus if
> >    the computer is on battery... except that upowerd is still blocked in
> >    g_spawn_sync()
> > 
> > The results of this is that the calling process is blocked for some time
> > (a dbus timeout), which can result in ugly ghost windows. See
> > https://bugzilla.novell.com/show_bug.cgi?id=615646 for example.
> > 
> > Richard is thinking that the pm-powersave part might be better handled
> > by upowerd itself instead of a hook for pm-suspend...
> 
> pm-utils is supposed to be lower level than upowerd.  Since we no longer
> rely on hal, we should just teach on_ac_power how to get what we need
> straight from sysfs.
> 
> It looks pretty strightforward, I will have a patch shortly.

OK, there is a rewritten on_ac_power in pm-utils git master.  Check it
out and see if it works for you.

> > Cheers,
> > 
> > Vincent
> > 
> 

-- 
Victor Lowther
LPIC2 UCP RHCE