Re: [PATCH] cpuset: Rework sched domains and CPU hotplug handling (2.6.27-rc1)

[Max Krasnyansky <maxk@qualcomm.com>]

Paul Jackson wrote:
> Max wrote:
>> Actually it is appropriate, and there is one more user of the
>> arch_reinit_sched_domains() which is S390 topology updates.
>> Those things (mc_power and topology updates) have to update domain flags based
>> on the mc/smt power and current topology settings.
> 
> Hmmm ... ok I suppose.
> 
> Could we have the kernel/sched.c code, in this case, call the
> kernel/cpuset.c routine async_rebuild_sched_domains(), rather
> than the synchronous rebuild_sched_domains() call (in your naming)
> which required details of the get_online_cpus() and put_online_cpus()
> wrapping to leak into kernel/sched.c:arch_reinit_sched_domains()
> routine?

It could I guess. But the questions is why ?
I mean the only reason we've introduced workqueue is because lock 
nesting got too complicated. Otherwise in all those paths we're already 
in a process context and there is no need to schedule a workqueue.

I see your point about get_online_cpus() thing. But it is similar to 
partition_sched_domains() which is an external (from the sched pov) 
interface and must be called within get_online_cpus() ... put_online_cpus().

Max