From: Jay Vosburgh <fubar@us...> - 2003-09-18 00:18:34
>Using the bonding driver in kernel 2.4.22 and the Intel Contributions 2.4
>patch sets released on August 26, 2003, I am able to consistently cause a
>kernel panic. I have not yet tested this with any other kernels, bonding
>drivers, or hardware. The environment consists of two servers each with
>two Intel Pro/1000 ethernet interfaces, and the servers are interconnected
>with crossover cables. The steps to produce the panic are as follows:
>1. modprobe e1000
>2. insmod bonding miimon=100 mode=0
>3. ifconfig eth0 up
>4. ifconfig bond0 192.168.1.1
>5. ifenslave bond0 eth0 eth1
>6. ifconfig bond0 down
I tried this out on 2.6.0-test5-bk2-netdrvr1, and it panics if
I use the stock RH 8.0 ifenslave, although I didn't get the panic
stuff saved. Curiously, if I use the correct 2.6 ifenslave it doesn't
panic, but does go into some sort of heinous infinite loop. Pretty
icky all around; my initial suspicion is that these are two unrelated
I'm testing this on a 2 processor SMP, with Intel e100s (not
e1000s). Will report back as soon as anything interesting happens.
-Jay Vosburgh, IBM Linux Technology Center, fubar@...
From: Shmulik Hen <shmulik.hen@in...> - 2003-09-18 09:43:34
On Wednesday 17 September 2003 11:44 pm, Ned Bass wrote:
> Note that the key to make the panic happen is step #3, bringing up
> the slave interface. I realize that the slave interface is not
> supposed to be brought up prior to bringing up the bonding
> interface, but should probably not cause a kernel panic either :).
> The panic occurs immediately after entering 'ifconfig bond0 down'.
> Also, the panic does not occur if I only enslave eth0 but not eth1.
There was a bug in patch 8 of the cleanup set that left a dangling
pointer in bond_release_all(). Definitely may bad for not announcing
that (and removing the patch set from SFG ?). I was relying on the
fact that people won't test it after I announced the set needs to be
re-done after finding a potential bug in patch 5.
I'm currently in the final stages of re-creating the set after
implementing the bug fixes and doing some improvements resulting from
internal code reviews. Perhaps this time I should publish it as an
"experimental" patch set and give everyone a chance to try it out
while our QA group grinds it to dust.
| Shmulik Hen Advanced Network Services |
| Israel Design Center, Jerusalem |
| LAN Access Division, Platform Networking |
| Intel Communications Group, Intel corp. |