Messages in this thread | | | From | Frederic Weisbecker <> | Subject | [GIT PULL] nohz: Bunch of fixes | Date | Fri, 3 May 2013 18:47:54 +0200 |
| |
Ingo,
Please pull the timers/nohz-hz1 branch that can be found at:
git://git.kernel.org/pub/scm/linux/kernel/git/frederic/linux-dynticks.git timers/nohz-hz1
HEAD: a7f4d69bb1d5e1f573b680355a2ad51566338887
It fixes a Kconfig dependency issue and forces a minimum of 1 tick every seconds to keep handling the scheduler_tick() duties, even at a very low granularity. This is a workaround until we can handle all these duties through on-demand driven solutions rather than using periodic events, as per your suggestion.
I just noted two things:
* update_cpu_load_active() seem to rely on the fixed periodic tick at the HZ rate. There is certainly something to tweak there to make it really correct with dynamick ticks. I have the feeling that modifying pending_updates won't work as it seems to decay assuming the time to catch up was idle.
* I'll probably have to disable CONFIG_CFS_BANDWIDTH when full dynticks is enabled. First of all it modifies rq->nr_running without using inc/dec_nr_running standard API, which is required for full dynticks. And second, I need to triple check it's safe to use with 1 tick per second.
Ah and please note the merge commit (c032862fba51a3ca504752d3a25186b324c5ce83) that was needed to get latest RCU and sched updates for the Kconfig fix.
Thanks, Frederic ---
Frederic Weisbecker (2): rcu: Fix full dynticks' dependency on wide RCU nocb mode sched: Keep at least 1 tick per second for active dynticks tasks
include/linux/sched.h | 1 + init/Kconfig | 4 ++-- kernel/sched/core.c | 30 ++++++++++++++++++++++++++++++ kernel/sched/idle_task.c | 1 + kernel/sched/sched.h | 10 ++++++++++ kernel/time/Kconfig | 1 - kernel/time/tick-sched.c | 7 +++++++ 7 files changed, 51 insertions(+), 3 deletions(-)
| |