When an hrtick timer event occurs too quickly, we just bail out and don't attempt to set a new hrtick timeout. That means that the time slice for that particular task grows until the next HZ tick occurs. That again may create significant jitter for the respective task, as it will not get scheduled for as long as it executed before, to bring the overall queue's vruntime into balance again.
With this patch, even a too early hrtick timer event will just reconfigure the hrtick to when we expected it to fire, removing overall jitter from the system. Signed-off-by: Alexander Graf <g...@amazon.com> --- kernel/sched/fair.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 0d4ff3ab2572..66e7aae8b15e 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -99,6 +99,8 @@ static int __init setup_sched_thermal_decay_shift(char *str) } __setup("sched_thermal_decay_shift=", setup_sched_thermal_decay_shift); +static void hrtick_update(struct rq *rq); + #ifdef CONFIG_SMP /* * For asym packing, by default the lower numbered CPU has higher priority. @@ -4458,8 +4460,10 @@ check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) * narrow margin doesn't have to wait for a full slice. * This also mitigates buddy induced latencies under load. */ - if (delta_exec < sysctl_sched_min_granularity) + if (delta_exec < sysctl_sched_min_granularity) { + hrtick_update(rq_of(cfs_rq)); return; + } se = __pick_first_entity(cfs_rq); delta = curr->vruntime - se->vruntime; -- 2.26.2 Amazon Development Center Germany GmbH Krausenstr. 38 10117 Berlin Geschaeftsfuehrung: Christian Schlaeger, Jonathan Weiss Eingetragen am Amtsgericht Charlottenburg unter HRB 149173 B Sitz: Berlin Ust-ID: DE 289 237 879