Re: [thiscpuops upgrade 10/10] Lockless (and preemptless) fastpathsfor slub

[Christoph Lameter]

On Wed, 24 Nov 2010, Peter Zijlstra wrote:

> On Wed, 2010-11-24 at 10:14 -0600, Christoph Lameter wrote:
> > On Wed, 24 Nov 2010, Peter Zijlstra wrote:
> >
> > > This thing still relies on disabling IRQs in the slow path, which means
> > > its still going to be a lot of work to make it work on -rt.
> >
> > The disabling of irqs is because slab operations are used from interrupt
> > context. If we can avoid slab operations from interrupt contexts then we
> > can drop the interrupt disable in the slab allocators.
>
> That's not so much the point, there's per-cpu assumptions due to that.

Sure the exclusive access to per cpu area is exploited during
irq off sections. That would have to change.

> Not everything is under a proper lock, see for example this bit:
>
>         new = new_slab(s, gfpflags, node);
>
>         if (gfpflags & __GFP_WAIT)
>                 local_irq_disable();
>
>         if (new) {
>                 c = __this_cpu_ptr(s->cpu_slab);
>                 stat(s, ALLOC_SLAB);
>                 if (c->page)
>                         flush_slab(s, c);
>                 slab_lock(new);
>                 __SetPageSlubFrozen(new);
>                 c->page = new;
>                 goto load_freelist;
>         }
>
> There we have the __this_cpu_ptr, c->page deref and flush_slab()->stat()
> call all before we take a lock.

All per cpu data and therefore would have get different treatment
if we wanted to drop the processing in interrupt off mode.

Disabling preempt may be initially sufficient there.



--
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/