Joel reported long preempt and irq off sequence in newidle_balance because of a large number of CPU cgroups in use and having to be updated. This patchset moves the update outside newidle_imblance. This enables to early abort during the updates in case of pending irq as an example.
Instead of kicking a normal ILB that will wakes up CPU which is already idle, patch 6 triggers the update of statistics in the idle thread of the CPU before selecting and entering an idle state. Changes on v4: - Add a dedicated bit for updating blocked load when entering idle. This simplifies the management of concurrency with kick_ilb. Changes on v3: - Fixed a compilation error for !CONFIG_SMP && CONFIG_NO_HZ_COMMON reported by kernel test robot <l...@intel.com> - Took advantage of this new version to add a short desciption for nohz_run_idle_balance Changes on v2: - Fixed some typos and updated some comments - Added more cleanup - Changed to way to trigger ILB in idle thread context to remove a possible race condition between the normal softirq ILB and this new mecanism. The cpu can already be set in idle_cpus_mask because even if the cpu is added later when entering idle, it might not have been removed yet from previous idle phase. Vincent Guittot (7): sched/fair: remove update of blocked load from newidle_balance sched/fair: remove unused return of _nohz_idle_balance sched/fair: remove unused parameter of update_nohz_stats sched/fair: merge for each idle cpu loop of ILB sched/fair: reorder newidle_balance pulled_task tests sched/fair: trigger the update of blocked load on newly idle cpu sched/fair: reduce the window for duplicated update kernel/sched/core.c | 2 +- kernel/sched/fair.c | 118 +++++++++++++++++-------------------------- kernel/sched/idle.c | 6 +++ kernel/sched/sched.h | 7 +++ 4 files changed, 60 insertions(+), 73 deletions(-) -- 2.17.1
