Consider the following topology:
DIE [ ]
MC [ ][ ]
0 1 2 3
capacity_orig_of(x \in {0-1}) < capacity_orig_of(x \in {2-3})
w/ CPUs 2-3 idle and CPUs 0-1 running CPU hogs (util_avg=1024).
When CPU2 goes through load_balance() (via periodic / NOHZ balance), it
should pull one CPU hog from either CPU0 or CPU1 (this is misfit task
upmigration). However, should a e.g. pcpu kworker awake on CPU0 just before
this load_balance() happens and preempt the CPU hog running there, we would
have, for the [0-1] group at CPU2's DIE level:
o sgs->sum_nr_running > sgs->group_weight
o sgs->group_capacity * 100 < sgs->group_util * imbalance_pct
IOW, this group is group_overloaded.
Considering CPU0 is picked by find_busiest_queue(), we would then visit the
preempted CPU hog in detach_tasks(). However, given it has just been
preempted by this pcpu kworker, task_hot() will prevent it from being
detached. We then leave load_balance() without having done anything.
Long story short, preempted misfit tasks are affected by task_hot(), while
currently running misfit tasks are intentionally preempted by the stopper
task to migrate them over to a higher-capacity CPU.
Align detach_tasks() with the active-balance logic and let it pick a
cache-hot misfit task when the destination CPU can provide a capacity
uplift.
Reviewed-by: Qais Yousef <qais.you...@arm.com>
Signed-off-by: Valentin Schneider <valentin.schnei...@arm.com>
---
kernel/sched/fair.c | 11 +++++++++++
1 file changed, 11 insertions(+)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 41cdda7a8ea6..5454429cea7a 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -7474,6 +7474,17 @@ static int task_hot(struct task_struct *p, struct lb_env
*env)
if (env->sd->flags & SD_SHARE_CPUCAPACITY)
return 0;
+ /*
+ * On a (sane) asymmetric CPU capacity system, the increase in compute
+ * capacity should offset any potential performance hit caused by a
+ * migration.
+ */
+ if (sd_has_asym_cpucapacity(env->sd) &&
+ env->idle != CPU_NOT_IDLE &&
+ !task_fits_capacity(p, capacity_of(env->src_cpu)) &&
+ cpu_capacity_greater(env->dst_cpu, env->src_cpu))
+ return 0;
+
/*
* Buddy candidates are cache hot:
*/
--
2.25.1
