Re: 2.6.17-rc6-rt3

["Sébastien Dugué" <sebastien.dugue@bull.net>]

On Mon, 2006-06-12 at 11:20 +0200, Ingo Molnar wrote:
> * Sébastien Dugué <sebastien.dugue@bull.net> wrote:
> 
> > > I think all of the regressions reported against rt1 are fixed, please 
> > > re-report if any of them is still unfixed.
> > 
> >   Great, boots fine on my dual Xeon and solves the ping problem I was 
> > having.
> > 
> >   Thomas, any hint at what was going on?
> 
> the problem was caused by a mismerge of the __raise_softirq_irqoff() 
> changes of preempt-softirqs. In PREEMPT_SOFTIRQS, softirq activation 
> means a wakeup of the softirq thread - hence __raise_softirq_irqoff() 
> must wake up the softirq thead too. This didnt happen in -rt1 so the 
> network softirq (which processes things like ping reply packets) got 
> delayed to the natural softirq event - the next timer interrupt in the 
> usual case. Hence depending on HZ you got a delay of 1-4-10 msecs 
> (divided into two parts).
> 

  Thanks.

  Sébastien.