[EMAIL PROTECTED] wrote on Fri, Aug 16, 2002 at 01:05:57PM +0200 :
> > More than likely, it's a firewall of some kind, or your card didn't load
> > the correct model, or you just haven't connected it to the network yet.
> > Either way, it's doubtful it's a bug.
> > I've got the same model of card you've got set up, and I've had no problems
> > with getting anywhere, though I haven't tried things in b3 (yet).
> I forgot a thing:
> My Realtek 8139 is integrated on my Leadtek7350 KDA Motherboard.
> In /etc/modules.conf, the 8139too module is loaded. I think it's the good 
> module.

I was finally able to reproduce this behavior on a machine (turns out my
personal machine, which I had never tried a Beta or RC on, I just try
them here on machines at work).

First, my mobo is an MSI mobo, and I had the same symptoms, so it's
related to the integrated 8139 chipset, not necessarily the mobo.

The really odd thing that struck me was that the packet counters never
incremented.  I figured that it seemed remarkably like an interrupt
problem.  So I took a look at ifconfig and saw that the card was getting
IRQ 17.  That struck me as odd, but it also means (I think) that is
using APIC to get IRQ's above 15.  Then I looked through dmesg and saw
that it was indeed using APIC and IRQ 17 was mapped to 0:17 (I'm sure
Juan could explain what that means).  I figured I didn't have anything
to lose, so I rebooted and passed noapic to the kernel.

Lo and behold, it started working.  I rebooted without it and it didn't
work again.  I rebooted with it and it worked again.  So the problem
seems to be that the 8139too driver seems to have problems with this
particular board's implmentation of APIC, which would make me think that
it could be a fundamentally lower level within the kernel.

Here are messages that appeared in the /var/log/kernel/info:
Sep 30 18:41:43 trip kernel: NETDEV WATCHDOG: eth0: transmit timed out
Sep 30 18:41:43 trip kernel: eth0: Tx queue start entry 4  dirty entry 0.
Sep 30 18:41:43 trip kernel: eth0:  Tx descriptor 0 is 00002000. (queue head)
Sep 30 18:41:43 trip kernel: eth0:  Tx descriptor 1 is 00002000.
Sep 30 18:41:43 trip kernel: eth0:  Tx descriptor 2 is 00002000.
Sep 30 18:41:43 trip kernel: eth0:  Tx descriptor 3 is 00002000.
Sep 30 18:41:43 trip kernel: eth0: Setting 100mbps full-duplex based on
auto-negotiated partner ability 41e1.

So the short-term fix is to turn off apic, the long term fix will be
some modification to the kernel.  Presumably, 8.2 didn't have this
problem, involved some less conflicting apic code.

Note:  When I say apic, I mean APIC, not to be confused with ACPI which
is a power management interface.  I'm talking about the Advanced
Programmable Interrupt Controller => APIC.

Blue skies...                   Todd
-- 
| MandrakeSoft USA             | Security is like an onion.  It's made |
| http://www.mandrakesoft.com  | made up of several layers and makes   |
| http://www.mandrakelinux.com | you cry.              --Howard Chu    |
   Cooker Version mandrake-release-9.0-0.3mdk Kernel 2.4.19-16mdk

Attachment: msg77612/pgp00000.pgp
Description: PGP signature

Reply via email to