Re: 2.6.17-rc5-mm1 - output of lock validator

[Andrew Morton <akpm@osdl.org>]

Ingo Molnar <mingo@elte.hu> wrote:
>
> 
> * Andrew Morton <akpm@osdl.org> wrote:
> 
> > On Wed, 31 May 2006 16:28:59 +0200
> > Arjan van de Ven <arjan@linux.intel.com> wrote:
> > 
> > > --- linux-2.6.17-rc5-mm1.5.orig/drivers/net/8390.c
> > > +++ linux-2.6.17-rc5-mm1.5/drivers/net/8390.c
> > > @@ -299,7 +299,7 @@ static int ei_start_xmit(struct sk_buff 
> > >  	 
> > >  	disable_irq_nosync(dev->irq);
> > >  	
> > > -	spin_lock(&ei_local->page_lock);
> > > +	spin_lock_irqsave(&ei_local->page_lock, flags);
> > 
> > Again, notabug - we did disable_irq().
> > 
> > I think lockdep needs to be taught about this idiom.  Perhaps add a 
> > new disable_irq_tell_lockdep() which assumes that we're in an 
> > equivalent-to-local_irq_disable() state.
> 
> agreed. I'll cook up an API for that. The best would be to disable local 
> irqs if LOCKDEP is enabled - i.e. how about disable_irq_lockdep() that 
> maps to disable_irq() if !LOCKDEP and on LOCKDEP it also disables local 
> interrupts? Likewise there would be an enable_irq_lockdep() which would 
> re-enable local irqs.
> 

That would probably work - we'd have to watch out for people doing
schedule() inside disable_irq_lockdep().