Hi Marc,
On Thu, May 19, 2016 at 3:13 PM, Marc Zyngier <marc.zyngier@xxxxxxx> wrote:
> On 19/05/16 12:08, Geert Uytterhoeven wrote:
>> On Mon, Apr 11, 2016 at 10:57 AM, Marc Zyngier <marc.zyngier@xxxxxxx> wrote:
>>> In order to prepare the genirq layer for the concept of partitionned
>>> percpu interrupts, let's allow an affinity to be associated with
>>> such an interrupt. We introduce:
>>>
>>> - irq_set_percpu_devid_partition: flag an interrupt as a percpu-devid
>>> interrupt, and associate it with an affinity
>>> - irq_get_percpu_devid_partition: allow the affinity of that interrupt
>>> to be retrieved.
>>>
>>> This will allow a driver to discover which CPUs the per-cpu interrupt
>>> can actually fire on.
>>>
>>> Signed-off-by: Marc Zyngier <marc.zyngier@xxxxxxx>
>>
>>> --- a/include/linux/irqdesc.h
>>> +++ b/include/linux/irqdesc.h
>>> @@ -66,6 +66,7 @@ struct irq_desc {
>>> int threads_handled_last;
>>> raw_spinlock_t lock;
>>> struct cpumask *percpu_enabled;
>>> + const struct cpumask *percpu_affinity;
>>
>> Adding this field showed up on my bloat-o-meter radar...
>
> By how much?
By NR_IRQS * 4 bytes, which ranges from 32 to 1024 bytes on m68k,
depending on the platform.
>> Does it make sense to move it (and percpu_enabled) inside the "#ifdef
>> CONFIG_SMP" below, and rework the code to not need it on UP?
>>
>>> #ifdef CONFIG_SMP
>>> const struct cpumask *affinity_hint;
>>> struct irq_affinity_notify *affinity_notify;
>
> I wonder if we couldn't actually unify affinity_hint and
> percpu_affinity. I'll have a look.
Thanks!
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
