lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <20250701024549.40166-1-adamli@os.amperecomputing.com>
Date: Tue,  1 Jul 2025 02:45:47 +0000
From: Adam Li <adamli@...amperecomputing.com>
To: mingo@...hat.com,
	peterz@...radead.org,
	juri.lelli@...hat.com,
	vincent.guittot@...aro.org
Cc: dietmar.eggemann@....com,
	rostedt@...dmis.org,
	bsegall@...gle.com,
	mgorman@...e.de,
	vschneid@...hat.com,
	cl@...ux.com,
	linux-kernel@...r.kernel.org,
	patches@...erecomputing.com,
	Adam Li <adamli@...amperecomputing.com>
Subject: [PATCH 0/2] Fix imbalance issue when balancing fork

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. So CPU 8-11 can be
crowded with newly forked threads, at the same time CPU 0-7 can be idle.

The first patch 'Only update stats of allowed CPUs when looking for dst
group' *alone* can fix this imbalance issue.

And I think the second patch also makes sense in this scenario. If group
weight includes CPUs a task cannot use, group classification can be
incorrect. Please comment.

Adam Li (2):
  sched/fair: Only update stats of allowed CPUs when looking for dst
    group
  sched/fair: Only count group weight for allowed CPUs when looking for
    dst group

 kernel/sched/fair.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

-- 
2.34.1


Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ