Firstly...hope everyone enjoyed (or is still enjoying) their holidays.
Just joined the mailing list. Apologies on the longish post, but I wanted
to provide as much information as possible.
I've got a couple of Cobalt boxes (a RaQ2 and a Qube2) that are successfully
running Debian (as well as Gentoo) running the 2.6.9 branch of CVS plus
Peter Horton's patches. The machines are quite stable and I really haven't
run into any issues until I start stress-testing networking.
Under high network loads when connected at 100TX FDX, I can always get the
Cobalts to lock up. Under anything slower -- 100TX HDX, 10TX FDX, 10TX HDX
-- everything works fine. This problem is specific to:
1. Lots of data transfer to/from the Cobalt.
2. Running at 100TX FDX.
The test scenario is:
Cobalt running 2.6.9CVS + PH's patches. Cobalt is running an NFS server.
From another machine on the network (also running at 100TX FDX), mount the
NFS export and copy something huge (in my case a directory that has about
2GB worth of files in it).
It'll usually get anywhere from 300MB to 800MB of data before the Cobalt
just locks up -- no kernel panic, just a hard lockup that necessitates
cycling power manually.
At first I thought it might be NFS, so I tried something way less kernel
dependent like FTP and still had the same problems. I also recompiled the
as well as set to yes to no avail.
As a last resort I turned on lots of debugging output (I set tulip_debug to
99) and finally I got something usable from the kernel:
eth0: MII status 782d, Link partner report 45e1.
eth0: 21143 negotiation status 000000c6, MII.
Badness in local_bh_enable at kernel/softirq.c:141
Call Trace: [<800b32c8>] [<80084e28>] [<80397ee8>] [<80397f08>]
[<80398af4>] [<8029a374>] [<800b87ac>] [<800ad20c>] [<8029ccbc>]
[<802bc8b0>] [<802bc918>] [<802bcba8>] [<802575bc>] [<8027b370>]
[<800abe34>] [<800abe34>] [<800b3168>] [<800abebc>] [<800abf80>]
[<800ac6b8>] [<8022e900>] [<800ac34c>] [<8022e900>] [<800ac278>]
[<800ac174>] [<80279dec>] [<80279980>] [<800b8954>] [<802b9458>]
[<800b3168>] [<80084808>] [<800b3208>] [<80084e18>] [<80082908>]
[<802dc1bc>] [<80083180>] [<802d89d8>] [<8030ffdc>] [<80303260>]
[<802da298>] [<80084e28>] [<8029afb0>] [<8029b330>] [<80138718>]
[<802dad80>] [<80134538>] [<800a4198>] [<802d07c0>] [<803031ec>]
[<8030ffdc>] [<80214364>] [<80295864>] [<800a7440>] [<8030ffdc>]
[<80214364>] [<80295894>] [<80295864>] [<8029b7b4>] [<80303ab8>]
[<80303aa0>] [<8020e0d4>] [<800a7440>] [<802120f4>] [<800a40c8>]
[<80310070>] [<80398698>] [<80398840>] [<8029ca50>] [<801d6f8c>]
[<8029cd50>] [<8039918c>] [<801cab18>] [<8039b554>] [<8039c8a4>]
[<801d8068>] [<80397740>] [<800bdbe8>] [<801c7398>] [<801c7274>]
[<801c70cc>] [<80086070>] [<80086060>]
I'd already deduced that it was probably a problem related to interrupts
(seems we have a lot of those issues on our lovely blue boxes). Looking at
the relevant line in kernel/softirq.c yields:
if (unlikely(!in_interrupt() &&
So it's clear that something's calling local_bh_enable while interrupts are
disabled, which they shouldn't be. I can recreate this problem at will --
so it's definitely replicable. I've really taken this as far as I can in
terms of debugging the problem on my own. I'd appreciate any/all
assistance/direction in how to track down the culprit here and fix the