gnome locks up with cgroup-bin installed

Bug #598335 reported by Douglas Bagnall
22
This bug affects 3 people
Affects Status Importance Assigned to Milestone
libcgroup (Ubuntu)
Expired
Undecided
Unassigned
Nominated for Maverick by agent 8131

Bug Description

I installed cgroup-bin, and its dependency libcgroup1.

Over the next few boots, Network Manager failed to associate with the usual AP (which was functioning fine), and asked for the pass phrase. Whether I entered it or cancelled, I got no further response from Gnome. The mouse pointer was still mobile, but neither clicking nor keyboard shortcuts had any effect. I could however switch to virtual terminals with ctrl-alt-F[1-6], and there I tried sending increasingly blunt signals, culminating in the likes of

sudo killall -9 NetworkManager
sudo killall -9 wpa_supplicant

but not only did this not kill anything, but the killall process itself would hang and ignore ctrl-C, which disuaded me from this approach as I ran out of virtual terminals.

The problem disappeared when I uninstalled cgroup-bin. Incidentally, aptitude would also hang until I booted into single user mode.

I didn't record dmesg, but I'm attaching a sample syslog cycle. The obvious things to me are the tracebacks preceded by messages like "INFO: task gnome-panel:1944 blocked for more than 120 seconds".

I'm using karmic 32 bit x86 with the 2.6.31-22-generic kernel. All packages are up to date.

Revision history for this message
Douglas Bagnall (douglasbagnall) wrote :
Revision history for this message
agent 8131 (agent-8131) wrote :

I just discovered that this was my problem as well though I am using amd64 and lucid. When cgroup-bin is installed all operations involving adding or removing modules fail (modprobe, rmmod). There may be other factors at play but for my system once installing cgroup-bin I can reproduce 100% of the time by running "sudo modprobe aes". Removing the packages fixes the problem.

Changed in libcgroup (Ubuntu):
status: New → Confirmed
Revision history for this message
agent 8131 (agent-8131) wrote :

I detailed some of my systems when investigating this problem here:
http://ubuntuforums.org/showthread.php?t=1518092

Revision history for this message
Jon Bernard (jbernard) wrote :

Hey guys, thanks for the bug report and detailed explanation, I will take a close look at this as soon as my workload eases up; In the mean time, I've prepared a PPA (ppa:jbernard/libcgroup) that contains backports of the latest stable version of libcgroup. If we could verify that this behaviour exists against the latest release, that would be a huge help. If you get a chance, try the packages in the PPA:

    https://launchpad.net/~jbernard/+archive/libcgroup/+packages

Thanks!

--
Jon

Revision history for this message
Douglas Bagnall (douglasbagnall) wrote :

hey Jon,

It is just the same with version 0.36.2-1~9.10prevu1_i386 of cgroup-bin and libcgroup1. The dmesg output and kern.log are attached.

I didn't install the pam module (either this time or the first time, IIRC).

(Sorry about the delay: I got sick and then I forgot).

Douglas

Revision history for this message
Douglas Bagnall (douglasbagnall) wrote :
Revision history for this message
Balbir Singh (bsingharora) wrote :

I took a quick look and found all rtnl paths waiting on mutex_lock. The stuck tasks were gnome-panel, ureadahead and workqueues (modprobe, waiting on a lock).. I also found wpa_supplicant waiting on

[ 240.393172] [<c05739d5>] schedule_timeout+0x185/0x200
[ 240.393181] [<c012e7a0>] ? __wake_up_common+0x40/0x70
[ 240.393191] [<c0132f90>] ? __wake_up+0x40/0x50
[ 240.393200] [<c05736f2>] wait_for_common+0xa2/0x120
[ 240.393209] [<c013c4a0>] ? default_wake_function+0x0/0x10
[ 240.393218] [<c0573802>] wait_for_completion+0x12/0x20
[ 240.393228] [<c01568b7>] call_usermodehelper_exec+0xc7/0xd0

The rtnl_lock() is held in wext_ioctl_dispatch() (from what I can see).

Observations

1. I don't see cgroup showing up anywhere in the stack

Questions

1. What is the user mode helper? It seems like it is holding/blocking the rest of the system from making progress.

Can we please get the output of cat /proc/mounts and the cgroup filesystem output, please?

Revision history for this message
Serge Hallyn (serge-hallyn) wrote :

If anyone is still having this problem, can you please supply the info requested in comment #7?

Changed in libcgroup (Ubuntu):
status: Confirmed → Incomplete
Revision history for this message
Launchpad Janitor (janitor) wrote :

[Expired for libcgroup (Ubuntu) because there has been no activity for 60 days.]

Changed in libcgroup (Ubuntu):
status: Incomplete → Expired
Revision history for this message
Douglas Bagnall (douglasbagnall) wrote :

FWIW (i.e. not much), I replaced the originally troubled laptop when its backlight died, and my intention to provide more info foundered on a shortage of means and motivation.

Revision history for this message
John Clarke (jrc61) wrote :

Could this be a duplicate of https://bugs.launchpad.net/ubuntu/+source/libcgroup/+bug/693594?

I had problems with "modprobe -r" hanging, and after that any attempt to do anything with modules also hung. After much testing I discovered that cgroups was to blame, and setting cpu.rt_runtime_us in the sysdefault cgroup as suggested in comment 13 in that bug fixed the problem.

To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.