Re: [PATCH] new irq tracer

[Peter Zijlstra <peterz@infradead.org>]

On Fri, 2009-02-27 at 12:14 +0900, KOSAKI Motohiro wrote:
> > > > Given this scenario :
> > > > 
> > > > A telecommunication system runs, but the client notices 
> > > something wrong.
> > > > They call their service provider. The provider enables tracing 
> > > > _remotely_ on the _production system_ while it's _active in 
> > > the field_.
> > > > 
> > > > Bam, those few milliseconds interrupt latencies become unacceptable.
> > > > 
> > > > Hopefully this scenario makes the use-case clearer. The 
> > > problem is not 
> > > > that interrupt latencies would occur while tracing is on, 
> > > but rather 
> > > > that it would happen on a running production system when switching 
> > > > tracing on. This is what is totally unacceptable for this use-case.
> > > > 
> > > > For more details about such requirements, I'm CCing 
> > > Dominique Toupin 
> > > > from Ericsson who I'm sure would be happy to give more 
> > > details about 
> > > > this if needed.
> > > 
> > > Hmm, so this system in the field is running Linux with the 
> > > Real-Time Patch?  Because if it isn't it will suffer from 
> > > millisecond latencies in normal operation.
> >
> > In many cases we don't use Linux real-time, we have many systems that
> > are soft-real-time an non real-time Linux is good enough.
> >  
> 
> Agreed, rt-patch seems off topics. we discuss to mainline kernel.

Exactly, hence you should not worry about ms irq-off latencies, since
mainline is happy to generate those for you, regardless of
function-trace.

The only way to complain about those is if your base system is better
than that, and currently only preempt-rt provides that.

So unless you're on preempt-rt, complaining about ms-order irq latencies
just isn't a valid argument.