Re: [PATCH 0/6] Optimize the cpu hotplug locking -v2

[Peter Zijlstra <peterz@infradead.org>]

On Thu, Oct 10, 2013 at 12:16:16PM -0700, Linus Torvalds wrote:
> On Thu, Oct 10, 2013 at 12:04 PM, Steven Rostedt <rostedt@goodmis.org> wrote:
> >
> > I'm wondering if we can have a for_each_cpu() that only disables
> > preemption in the loop.
> 
> I think we'd generally want to have it be something the loop asks for.
> 
> If the loop is just some kind of "gather statistics" thing, I don't
> think it's required. The cost per loop is so low (usually adding up a
> couple of words) that the downside drowns the upside.
> 
> And we could easily look at MAXSMP (or NR_CPUS) at compile-time, and
> not do it for common small values (although it looks like Fedora
> defaults to 128 CPU's for their distro kernels, which seems a bit
> excessive - too many by far for normal people, too few for the crazy
> big ones).

Alternatively we could write it something like:

	rcu_read_lock();
	for_each_online_node(node) {
		for_each_cpu(cpu, cpumask_of_node(node)) {
			...
		}
		cond_resched_rcu();
	}
	rcu_read_unlock();

But yes, I think this pattern (and variations) should work for most
cases.

The one I'm struggling with atm is kernel/cpuset.c:
rebuild_sched_domains_locked(). Although I'm thinking we should do that
seqlock style, just build it and verify if its still valid, if not, try
again -- although I'm sure it'll be 'fun' to get correct.