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>] [day] [month] [year] [list]
Message-ID: <CACFaPDVAVuVWQ+tPHCBLeahJdYPY3Te+xKGr_w90TxrhDOV7pA@mail.gmail.com>
Date: Thu, 23 Jan 2025 15:38:49 +0100
From: Marcin Wielgosz <quizac@...il.com>
To: linux-kernel@...r.kernel.org
Subject: High CPU usage of 1H-kblockd kworkers on 6.8.12 compared to 5.4.233
 with isolcpus

Greetings,

I've noticed much higher CPU consumption of 1H-kblockd workers on
isolated cores on kernel 6.8.12 (Ubuntu 24).
Relevant kernel params: isolcpus=nohz,domain,managed_irq,4-23,28-47
skew_tick=1 nohz=on nohz_full=4-23,28-47 rcu_nocb_poll
rcu_nocbs=4-23,28-47 irqaffinity=0-3,24-27.
All workqueues (/sys/devices/virtual/workqueue/*/cpumask) are set to
'0,00040003' which excludes isolated cores. All IRQs including those
for NVMEs are assigned to cores 24-25.
I ran a script to poll /sys/kernel/debug/sched/debug every second for
10 minutes on both kernels and extracted running times for cores
4-6(isolated). 1H-blocked workers accumulated 0.12% CPU compared to
0.02-0.03% on 5.4.233(Ubuntu 18) under the same conditions. There is
only one DPDK thread running on each core. These threads don't make
any calls to block devices, syscalls nor FS.

I expected all kblockd workers to be either offloaded to housekeeping
cores or inactive on isolated cores. Is there a way to remedy this?

Kind regards,
Marcin Wielgosz

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ