Re: [PATCH] cpuset: Rework sched domains and CPU hotplug handling

[Max Krasnyansky <maxk@qualcomm.com>]

Peter Zijlstra wrote:
> On Mon, 2008-07-28 at 16:07 +0200, Ingo Molnar wrote:
>> * Max Krasnyansky <maxk@qualcomm.com> wrote:
>>
>>> Max Krasnyansky wrote:
>>>> This is an updated version of my previous cpuset patch:
>>>>   "Make rebuild_sched_domains() usable from any context (take 2)"
>>> Folks,
>>>
>>> Any comments on this patch ? We need this to complete sched domain 
>>> handling fixes/improvements that we started with the cpu_active_map, 
>>> and to avoid circular locking issues in the cpu hotplug -> 
>>> rebuild_sched_domains path.
>> Paul, Peter, any comments?
> 
> I'm really not that at home with all that cgroup fiddling, so I'd like a
> word from the two Pauls..
> 
> Questions I have at the moment:
> 
>  -  do we really need a new workqueue for this? Can't we use the regular
> keventd stuff, now that Oleg fixed the get_online_cpus() thing?
> (3da1c84c00c7e5fa8348336bd8c342f9128b0f14)
No we do not. I was not sure when that will workqueue fix would go in.
I'll send a delta patch on top.

> - aren't there funny races with the async_rebuild_sched_domains()
> stuff? 
Rebuilds via cpu hotplug path are synchronous, which I believe is 
important.
Rebuilds caused by writes into /dev/cpuset do not have to be.  I cannot 
think of a scenario where race in that path would be an issue.

Max