1.3.62: tcpdump causes Aiee: scheduling in interrupt

Zoltan Hidvegi (hzoli@cs.elte.hu)
Tue, 13 Feb 1996 22:29:17 +0100 (MET)

tcpdump crashed my Linux box with 1.3.61 and 1.3.62 kernels. I did not tested
1.3.60. It first happened with 1.3.61 and then after compiling 1.3.62 I tried
it deliberately. Itried tcpdump -vvttNl. It printed a few pages then the
system froze with the Aiee message. The foreze was complete, no backtrace, no
console switching. But the most intreresting thing that I tried the same on
an other machine and I was unable to crash that. Both machines had IP
forwarding enabled. The machine which crashed has an NCR 53c810 card with two
SCSI drives using 10 MHz FAST SCSI-II and IPX support were enabled in the
kernel (but when I run tcpdump no other program used IPX).

The machine which did not crash did not have IPX but it had PC/TCP
compatibility enabled. Also it uses an IDE harddisk.

The two machines has SiS 496/7 PCI motherboard S3868 VGA, Intel 486 DX4-S/100
CPU and SMC Ultra network card. Both kernels were compiled with 486
optimizations. The one which crashed had 32 MB of memory while the other had
16 MB. Both kernels were compiled without kerneld support.

The network here is quite busy (mainly NFS traffic). The machine
which crased works as an NFS server and font server for several other
machines but the test were made late in the evening when the net
traffic was quite low. Both machines mounts several remote
filesystems from Solaris 2.4 and Ultrix 4.2 NFS servers.

Note that several other people reported this Aiee bug and only one of them
mentiond tcpdump. All of these reports were related to 1.3.60 or later
kernels. 1.3.59 + mark6 did not have this bug.

Cheers,

Zoltan