[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <176060722202.709179.10324554366547920524.tip-bot2@tip-bot2>
Date: Thu, 16 Oct 2025 09:33:42 -0000
From: "tip-bot2 for Adam Li" <tip-bot2@...utronix.de>
To: linux-tip-commits@...r.kernel.org
Cc: Adam Li <adamli@...amperecomputing.com>,
"Peter Zijlstra (Intel)" <peterz@...radead.org>, x86@...nel.org,
linux-kernel@...r.kernel.org
Subject: [tip: sched/core] sched/fair: Only update stats for allowed CPUs when
looking for dst group
The following commit has been merged into the sched/core branch of tip:
Commit-ID: 82d6e01a0699800efd8b048eb584c907ccb47b7a
Gitweb: https://git.kernel.org/tip/82d6e01a0699800efd8b048eb584c907ccb47b7a
Author: Adam Li <adamli@...amperecomputing.com>
AuthorDate: Sat, 11 Oct 2025 06:43:22
Committer: Peter Zijlstra <peterz@...radead.org>
CommitterDate: Thu, 16 Oct 2025 11:13:50 +02:00
sched/fair: Only update stats for allowed CPUs when looking for dst group
Load imbalance is observed when the workload frequently forks new threads.
Due to CPU affinity, the workload can run on CPU 0-7 in the first
group, and only on CPU 8-11 in the second group. CPU 12-15 are always idle.
{ 0 1 2 3 4 5 6 7 } {8 9 10 11 12 13 14 15}
* * * * * * * * * * * *
When looking for dst group for newly forked threads, in many times
update_sg_wakeup_stats() reports the second group has more idle CPUs
than the first group. The scheduler thinks the second group is less
busy. Then it selects least busy CPUs among CPU 8-11. Therefore CPU 8-11
can be crowded with newly forked threads, at the same time CPU 0-7
can be idle.
A task may not use all the CPUs in a schedule group due to CPU affinity.
Only update schedule group statistics for allowed CPUs.
Signed-off-by: Adam Li <adamli@...amperecomputing.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@...radead.org>
---
kernel/sched/fair.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 00f9d6c..ac881df 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -10683,7 +10683,7 @@ static inline void update_sg_wakeup_stats(struct sched_domain *sd,
if (sd->flags & SD_ASYM_CPUCAPACITY)
sgs->group_misfit_task_load = 1;
- for_each_cpu(i, sched_group_span(group)) {
+ for_each_cpu_and(i, sched_group_span(group), p->cpus_ptr) {
struct rq *rq = cpu_rq(i);
unsigned int local;
Powered by blists - more mailing lists