Re: s2r badness

[Thomas Gleixner]

On Mon, 20 Jul 2009, Steven Rostedt wrote:
> On Tue, 21 Jul 2009, Thomas Gleixner wrote:
> 
> > On Mon, 20 Jul 2009, Steven Rostedt wrote:
> > > >  
> > > > Just to make the list more complete. If tracing is enabled across
> > > > suspend/resume you'll hit that one as well:
> > > > 
> > > > WARNING: at /home/tglx/work/kernel/git/linux-2.6/kernel/trace/ring_buffer.c:1392 rb_reserve_next_event+0x133/0x27c()
> > > > 
> > > > Negative timestamp delta which is probably due to sched_clock not yet
> > > > adjusted to the TSC which got reset.
> > > 
> > > Perhaps we need to prevent tracing during a "blackout period" of suspend 
> > > to ram?
> > 
> > Perhaps we need to figure out why this is happening and how we best
> > deal with it. Disabling functionality just because we can not deal
> > with it right now is not a solution.
> 
> Heh, suspend to ram is a black magic art. There's voodoo there that causes 
> ulcers when you look the wrong way. For example, there's times that simply 
> calling smp_processor_id() will reboot the box. Hence, the tracer is very 
> intrusive, and we need to prevent it from doing things at certain areas. 
> I'm not saying disabling functionality per say, I'm just saying that we 
> need to make sure the tracer is not doing something in the guts of 
> bringing the CPU back on line when it is not ready.
> 
> Are you saying that we need to move the initialization of sched_clock up, 
> just to satisfy tracing? If that happens to be the issues here?

No, we need to figure out why the WARN_ON happens and what we can/must
do either to fixup the trace clock early enough or to prevent tracing
until the clock is usable again.

Thanks,

	tglx
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/