Re: [PATCH v3] lib: optimize cpumask_local_spread()

[Michal Hocko]

On Fri 15-11-19 17:09:13, Shaokun Zhang wrote:
[...]
> Oh, my mistake, for the previous instance, I don't list all IRQs and
> just choose one IRQ from one NUMA node. You can see that the IRQ
> number is not consistent :-).
> IRQ from 345 to 368 will be bound to CPU cores which are in NUMA node2
> and each IRQ is corresponding to one core.
> 

This is quite confusing then. I would suggest providing all IRQ used for
the device with the specific node affinity to see the difference in the
setup.

-- 
Michal Hocko
SUSE Labs