On 13 February 2013 15:08, Damien Wyart damien.wyart@gmail.com wrote:
- Vincent Guittot vincent.guittot@linaro.org [2013-02-13 13:08]:
Damien, Regarding your sched_domain config and especially the flags field, you should not be impacted by my patch because
- need_active_balance is the only new place that use env->src_cpu in
the load_balance function
- and your machine will never test the condition: "env->src_cpu >
env->dst_cpu" in need_active_balance because SD_ASYM_PACKING is not set in your config
Have you tested the patch with others that could have modified the load_balance function ?
Yes, sorry, I should have been more precise in my initial report: your patch was not applied on top of vanilla 3.8-rc7, but a few other patches were also present. Seems the ones impacting load_balance are from Frederic's nohz work (http://git.kernel.org/?p=linux/kernel/git/frederic/linux-dynticks.git%3Ba=sh...)
ok thanks for the pointer. i'm going to have a look
The patches from there modifying load_balance are:
- "sched: Update nohz rq clock before searching busiest group on load balancing"
- "sched: Update clock of nohz busiest rq before balancing"
In this test, I did not use any kernel parameter related to this patchset (full_nohz, etc.).
I am adding Frederic in Cc, not sure if the breakage is to be investigated on your side or his...
probably both
Thanks for your explanations,
Damien