Re: [PATCH 3/4] workqueue: Create low-level unbound workqueues cpumask

[Kevin Hilman <khilman@kernel.org>]

Frederic Weisbecker <fweisbec@gmail.com> writes:

> On Mon, Mar 16, 2015 at 10:12:12AM -0700, Kevin Hilman wrote:
>> Lai Jiangshan <laijs@cn.fujitsu.com> writes:
>> 
>> > On 03/14/2015 07:49 AM, Kevin Hilman wrote:

[...]

>> >> 
>> >> As I mentioned in an earlier discussion[1], I still think this could
>> >> default too the housekeeping CPUs in the NO_HZ_FULL case:
>> >> 
>> >> #ifdef CONFIG_NO_HZ_FULL
>> >> 	cpumask_complement(wq_unbound_cpumask, tick_nohz_full_mask);
>> >
>> >
>> > No, the default/booted wq_unbound_cpumask should be cpu_possible_mask.
>> >
>> 
>> Even for NO_HZ_FULL?  
>> 
>> IMO, for NO_HZ_FULL, we want the unbound workqueues to be on the
>> housekeeping CPU(s).
>
> If it should be the default on NO_HZ_FULL, maybe we should do this from the
> tick nohz code. Some late or fs initcall that will do the workqueue affinity,
> timer affinity, etc...

Sure, I'd be fine with that too.

Kevin