bug in kernel (I think): 2.6.15 to 2.6.17 and 3c59x driver
Hi,
First time reporting a bug! I'm unsure whether this is kernel or driver? (or both?)
I've been using Knoppix for some time and they have worked great. The
latest 5.0 and 5.0.1 versions have a problem configuring the network
card on one of my mahcines. I have three machines here (1) PII/400 with
BX motherboard and 3Com 3c900B NIC. (2) PIII/800 with BX motherboard
and 3Com 3c900B NIC. (3) K6-2/400 with VIA motherboard and 3Com 3c900
and 3c905 NICs (firewall). Booting Knoppix 5 on (1) the NIC looks like
it is configured but is unresponsive. (2) and (3) the NICs work fine,
but...
Looking into why (1) didn't work I found "ifconfig -a" showed eth0 with the wrong base address.
I tracked down these lines in dmesg:
PCI: Found IRQ 11 for device 0000:00:0b.0
3c59x: Donald Becker and others. www .scyld.com/network/vortex.html
0000:00:0b.0: 3Com PCI 3c900 Cyclone 10Mbps TPC at d085a000. Vers LK1.1.19
The "at d085a000" used to be "at 0x6800" on previous versions of
Knoppix. In this example ifconfig shows eth0 base address as 0xa000.
Base address of eth0 should be 0x6800 but values have been 0, 0x2000,
0xa000, 0xc000, 0xe000 ie it varies every reboot. Machine (2) shows a
similar behaviour. Machine (3) eth0 is correct (sortof, ifconfig shows
0xe800 but dmesg shows 0001e800), eth1 varies like above (should be
0xec00). Previous versions of Knoppix work as expected.
The address shown looks suspiciously like a memory address (~top 16
bits) and corresponds with the range of addresses shown by "cat
/proc/kallsyms|grep 3c59x". lspci -v and cat /proc/pci show correct
io port values
Kernel version for 5.0 is 2.6.15 and for 5.0.1 is 2.6.17 and it felt
like a kernel issue, so I installed Debian 3.0 on machine (1) from the
Woody DVD and upgraded to Etch kernel 2.6.15-8 and 2.6.16-17. They
failed in exactly the same way as Knoppix 5. So now I'm sure it's
nothing added by Knoppix causing the error. I'm using Sarge with 2.4
kernel on machine (2) which is my main machine and I really didn't want
to experiment on it! So I've only tried knoppix. Ditto for machine (3).
I tried the Etch beta2 (2006-02-04) net installer and latest
(2006-07-31) net installer. They do the same. Booting with the 2.6.16
kernel works all the way up to trying to use the network. Booting with
install24 works like a dream of course.
The error is very repeatable! It looks like it started around 2.6.15
and is still in 2.6.17. 2.6.11 was the last working. Not tried
2.6.12 to 2.6.14 kernels (still looking for live CDs).
Unfortunately, I only have 3com cards for testing so I can't tell if it
is a kernel or driver issue. The three machines are fairly different
(excluding the NICs) so I can't believe it's a hardware issue and I
can't believe I'm the only one to have seen it! They do have *me* in
common! But they are fairly ordinary, not overclocked, no exotic
hardware...
I have reported this on the debian-knoppix mailing list and on the
knoppix forum but it appears to have confounded everyone! This may have
some more important details:
http://www.knoppix.net/forum/viewtopic.php?t=24746
So,
- Is this a bug?
- Has it already been reported?
- Should I report it as kernel or driver?
- Should I file a seperate report about the debian-installer even though it's really the same bug?
- How do I report this in a way that is useful? (has enough detail etc)
Thanks,
Danny.
Reply to: