We've also seen this or a similar one on 6.17.0-1018.19-gcp; AI-assisted
RCA of that one:
The serial console (watchdog capture) shows the kernel deadlocked. 19
tasks wedged in two interlocking groups, both on the global events
workqueue:
18× events / bpf_prog_free_deferred → stuck in _vm_unmap_aliases →
__mutex_lock, waiting on vmap_purge_lock
1× events / drain_vmap_area_work → in __purge_vmap_area_lazy → __flush_work;
holds vmap_purge_lock, flush-waiting on the events wq
ABBA deadlock: drain_vmap_area_work holds vmap_purge_lock and flush-waits on
the events workqueue; the 18 bpf_prog_free_deferred workers occupy that same
workqueue and wait for vmap_purge_lock. The whole events wq jams → box stops
doing kernel work.
** Also affects: linux-gcp-6.17 (Ubuntu)
Importance: Undecided
Status: New
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2134841
Title:
Node drops off network; serial console shows hung task in
bpf_prog_free_deferred
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux-gcp-6.14/+bug/2134841/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
