Public bug reported:
Ubuntu 26.04 LTS systems running the linux-aws 7.0 kernel can
intermittently place systemd PID 1 into uninterruptible D state for
approximately 5 to 10 minutes. During the stall, new SSH sessions fail
because pam_systemd/logind cannot create a session.
The issue has been observed with `linux-aws` `7.0.0-1004.4` and
`7.0.0-1006` on a 192-vCPU c8a.metal-48xl system configured for low-
latency workloads.
The relevant kernel parameters include:
isolcpus=nohz,18-95,108-178
nohz_full=18-95,108-178
rcu_nocbs=18-95,108-178
rcu_nocb_poll
idle=poll
numa_balancing=disable
Busy userspace threads are pinned to the isolated CPUs.
The blocked workqueue stack is:
cpuset_migrate_mm_workfn
do_migrate_pages
lru_cache_disable
__lru_add_drain_all
flush_work
Per-CPU LRU drain work is queued to nohz_full CPUs occupied by busy-
polling userspace threads. The work does not run until those CPUs re-
enter the kernel or otherwise reschedule. cpuset migration therefore
waits in flush_work. The flush_migrate_mm_task_workfn path subsequently
causes PID 1 to wait for the blocked migration work, preventing logind
from creating sessions.
This may be a regression associated with the CONFIG_PREEMPT_LAZY
configuration used by the Ubuntu 26.04 kernel. The behavior has not been
observed with the Ubuntu 24.04 6.x kernel using the equivalent workload.
** Affects: linux-aws (Ubuntu)
Importance: Undecided
Status: New
** Package changed: linux (Ubuntu) => linux-aws (Ubuntu)
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2165410
Title:
Ubuntu LTS 26.04 linux-aws: systemd enters D state and blocks SSH
during cpuset migration on nohz_full CPUs
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux-aws/+bug/2165410/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
