[RFC 2/2] clocksource: don't suspend/resume when unused

[linux@arm.linux.org.uk (Russell King - ARM Linux)]

On Fri, Jan 16, 2015 at 11:35:30AM +0100, Alexandre Belloni wrote:
> On 16/01/2015 at 11:23:32 +0100, Thomas Gleixner wrote :
> > On Fri, 16 Jan 2015, Alexandre Belloni wrote:
> > 
> > > There is no point in calling suspend/resume for unused
> > > clocksources.
> > > 
> > > Signed-off-by: Alexandre Belloni <alexandre.belloni@free-electrons.com>
> > > ---
> > >  kernel/time/clocksource.c | 4 ++--
> > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/kernel/time/clocksource.c b/kernel/time/clocksource.c
> > > index 920a4da58eb0..baea4e42ae90 100644
> > > --- a/kernel/time/clocksource.c
> > > +++ b/kernel/time/clocksource.c
> > > @@ -493,7 +493,7 @@ void clocksource_suspend(void)
> > >  	struct clocksource *cs;
> > >  
> > >  	list_for_each_entry_reverse(cs, &clocksource_list, list)
> > > -		if (cs->suspend)
> > > +		if (cs->suspend && (cs->flags & CLOCK_SOURCE_USED))
> > >  			cs->suspend(cs);
> > 
> > This might be dangerous. If the clocksource has no enable/disable
> > callbacks, but suspend/resume, then you might leave it enabled across
> > suspend.
> > 
> 
> Isn't that already the case?
> Right now, if you call clocksource_suspend, it doesn't matter whether
> the clocksource has an enable or not, it will be suspended. Maybe I'm
> mistaken but my patch doesn't seem to change that behaviour.

You change an "always suspend" problem to a "never suspend" problem
since those clocksources which are used, but do not have an ->enable
callback will not be marked with CLOCK_SOURCE_USED.

Look at patch 1:

-               if (!new->enable || new->enable(new) == 0) {
+               if (!new->enable || clocksource_enable(new) == 0) {

If new->enable is NULL, clocksource_enable() won't be called, which
means there's nothing which sets CLOCK_SOURCE_USED, which in turn
means that ->suspend() will never be called.

-- 
FTTC broadband for 0.8mile line: currently at 10.5Mbps down 400kbps up
according to speedtest.net.