Re: sys_sched_yield keeps locked irq before call of schedule()

[Ingo Molnar <mingo@elte.hu>]

* Steven Rostedt <rostedt@goodmis.org> wrote:

> 
> On Thu, 6 Nov 2008, Ingo Molnar wrote:
> > 
> > * Zdenek Kabelac <zdenek.kabelac@gmail.com> wrote:
> > 
> > > > is this some special warning you added? The stock kernel certainly 
> > > > does not emit this warning.
> > > 
> > > Yes - it's my personal debug checker that tries to watch wether irq 
> > > & irqsafe are running in pair. So it shows a warning if there is a 
> > > call of spin_lock_irq and irq is already dissabled.
> > 
> > btw., i have added lockdep instrumentation for this two years ago, 
> > it's in the upstream kernel:
> > 
> > /proc/lockdep_stats: redundant hardirq ons:                7744
> > /proc/lockdep_stats: redundant hardirq offs:            1873736
> > /proc/lockdep_stats: redundant softirq ons:                   0
> > /proc/lockdep_stats: redundant softirq offs:                  0
> > 
> > you could extend the code to trace it. See 
> > kernel/lockdep.c:trace_hardirqs_on_caller(). This bit does it:
> > 
> >         if (unlikely(curr->hardirqs_enabled)) {
> >                 debug_atomic_inc(&redundant_hardirqs_on);
> >                 return;
> >         }
> > 
> > you could add a ftrace_printk("redundant hardirq") call right there, 
> > enable ftrace, and enable stack tracing of each trace entry:
> 
> Quick note. Currently ftrace_printk only shows up if it is set in the 
> iter_ctrl.
> 
>   echo trace_printk > /debug/tracing/iter_ctrl
> 
> I'll send out a patch to have it on by default.

ok, agreed - but in that case please make the NOP tracer 
tracing_enabled=0 by default. (we'll get a lot of ftrace_printk 
tracepoints by default otherwise)

	Ingo