Re: RT patch acceptance

[Esben Nielsen]

On Wed, 1 Jun 2005, Andrea Arcangeli wrote:

> On Wed, Jun 01, 2005 at 04:19:19PM +0200, Ingo Molnar wrote:
> > 
> > * Andrea Arcangeli <andrea@xxxxxxx> wrote:
> > 
> > > I've a bug in my queue that definitely would break preempt-RT:
> > >
> > >       BUG xxx : spends excessive time with interrupts disabled on large memory>
> >        systems
> > >
> > > workaround:
> > >
> > >       #define MAX_ITERATION 100000
> > >       if ((nr_pages > MAX_ITERATION) && !(nr_pages % MAX_ITERATION)) {
> > >               spin_unlock_irq(&zone->lru_lock);
> > >               spin_lock_irq(&zone->lru_lock);
> > >       }
> > 
> > you are wrong. This codepath is not running with interrupts disabled on 
> > PREEMPT_RT. irqs-off spinlocks dont turn off interrupts on PREEMPT_RT.  
> 
> Then I'm afraid preempt-RT infringe on the patent that they take after
> years of doing that in linux. I'm not a lawyer but you may want to
> check before investing too much on this for the next 15 years. The
> nanokernel thing has happened exactly because they couldn't wrap the cli
> calls to do something different than a cli AFIK. Nanokernel was a nice
> workaround to avoid having to us the patented irq disable redefine.
> 
> I assumed you weren't infringing on the patent and in turn disabling irq
> locally would actually do that, sorry.
> 
I haven't seen the patent - it isn't valid here in Europe yet anyway :-) -
but don't worry: disable_irq is _not_ redefined, see below.


> > (there are still some ways to introduce latencies into PREEMPT_RT, but 
> > they are not common and we are working on ways to cover them all.)
> 
> How can you schedule a task while a spinlock is held? Ok irqs will keep
> going, but how can you reschedule risking to deadlock? As long as there
> are regular spinlocks in any driver out there (i.e. all of them) then
> you can still introduce latencies.

I'll explain it again: All spinlocks are made into mutexes. Everything
runs in threads under the normal scheduler with interrupt disabled. Only
the implementation of the former spinlock, now a mutex, is using a
raw_spin_lock, which disables interrupts.

>  It doesn't seem too uncommon to me to
> take a spinlock. Preempt-rt reliability cannot be remotely compared to
> RTAI.

It can come close, but not all the way, no.

Esben


-
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/