bnx2: kernel panic: killing interrupt handler

[Daniel Topa]

Binary package hint: linux-image-2.6.15-27-amd64-server

Kernel panic on HP BL460c blade server.

The server hangs after high "network" load with bnx2 module (see attachment). No special boot kernel parameters are used.

# cat /proc/cmdline
root=/dev/mapper/vg00-lv_root ro vga=771

Only eth1 is used.

# hwinfo --network
...
49: None 01.0: 10701 Ethernet
  [Created at net.114]
  Unique ID: L2Ua.ndpeucax6V1
  Parent ID: JNkJ.OKRlPBdQgr3
  SysFS ID: /class/net/eth1
  SysFS Device Link: /devices/pci0000:00/0000:00:1c.0/0000:02:00.0/0000:03:00.0
  Hardware Class: network interface
  Model: "Ethernet network interface"
  Driver: "bnx2"
  Device File: eth1
  HW Address: 00:18:fe:34:19:5c
  Link detected: yes
  Config Status: cfg=new, avail=yes, need=no, active=unknown
  Attached to: #29 (Ethernet controller)
...

[Timo Aaltonen (tjaalton)]

Confirming, the same happens on our HP DL380G5 and I'm able to reproduce it with ease. The server is our apt mirror, so every time the clients update themselves the server crashes. We worked around it by distributing the load, and it worked fine for a month until this morning.

Although the call trace is not the same, I suspect the bug is. It should be fixed upstream, here is what I found googling around:

http://bugs.centos.org/view.php?id=2103

that trace is pretty similar to yours.

[Timo Aaltonen (tjaalton)]

I compiled the latest driver (1.6.7b), and it has been in use for five days now without any issues.

[Leann Ogasawara (leannogasawara)]

It looks like Hardy has version 1.6.9. Timo or Daniel, care to confirm this is no longer an issue for Hardy at least?

[Timo Aaltonen (tjaalton)]

It's a production server so I can't test Hardy on it, but I'm sure this issue is fixed at least since Feisty.

[Timo Aaltonen (tjaalton)]

Btw, the backported driver has been in use nearly three months now without any issues..

[Leann Ogasawara (leannogasawara)]

Awesome, thanks.