Re: [PATCH] cpuset: Make rebuild_sched_domains() usable from anycontext

[Max Krasnyansky]

Paul Menage wrote:
> On Tue, Jul 15, 2008 at 4:44 AM, Max Krasnyansky <maxk@xxxxxxxxxxxx> wrote:
> > From: Max Krasnyanskiy <maxk@xxxxxxxxxxxx>
> >
> > I do not really like the current solution of dropping cgroup lock
> > but it shows what I have in mind in general.
>
> I think that dropping the cgroup lock will open up races for cpusets.
> The idea of a separate workqueue/thread to do the sched domain
> rebuilding is simplest.

Actually I audited (to the best of my knowledge) all the paths in 
cpusets and rebuild_sched_domains() is the last action. ie We drop the 
lock right after it anyways. It's just it's embedded deep in the call 
stack and therefor I cannot drop it at the higher level.

The only path where I think it's not safe is the cgroup destroy thing 
where we do
	
cgroup.c
	cgroup_lock();
	for_each_cgroups(...)
		cg->destroy();
	cgroup_unlock();

So in theory it's just that one patch that really needs the workqueue 
trick. But I do agree that it'll make it less tricky across the board.
So I'll pick up you work queue based patch, convert it to single 
threaded, bang on a bit later today and send a patch on top of this one.

Max


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/