On Thu, Jul 10, 2014 at 04:03:51PM +0200, Vincent Guittot wrote:

On 10 July 2014 13:18, Peter Zijlstra [off-list ref] wrote:

quoted

On Mon, Jun 30, 2014 at 06:05:39PM +0200, Vincent Guittot wrote:

quoted

+             /*
+              * If the CPUs share their cache and the src_cpu's capacity is
+              * reduced because of other sched_class or IRQs, we trig an
+              * active balance to move the task
+              */
+             if ((sd->flags & SD_SHARE_PKG_RESOURCES)
+              && ((capacity_orig_of(src_cpu) * 100) > (capacity_of(src_cpu) *
+                                                     sd->imbalance_pct)))
                      return 1;

Why is this tied to shared caches?

It's just to limit the change of the policy to a level that can have
benefit without performance regression. I'm not sure that we can do
that at any level without risk

Similar to the other change; so both details _should_ have been in the
changelogs etc..

In any case, its feels rather arbitrary to me. What about machines where
there's no cache sharing at all (the traditional SMP systems). This
thing you're trying to do still seems to make sense there.