[realtek] Too much work...

[P.Peti]

Hello,

I've already found a question similar to my one here in the
archives but did not find the answer, so here it goes:

My RTL8139C card eventually stops working (even at light net
loads). The kernel log says:

kernel: NETDEV WATCHDOG: eth0: transmit timed out
kernel: eth0: Tx queue start entry 36231 dirty entry 36227
kernel: eth0: Tx descriptor 0 is 00002000.
kernel: eth0: Tx descriptor 1 is 00002000.
kernel: eth0: Tx descriptor 2 is 00002000.
kernel: eth0: Tx descriptor 3 is 00002000. (queue head)
kernel: eth0: Setting half-duplex based on auto-negotiatied
partner ability 0000.

The above repeated several times. During these times, ping
echoes return in weird ordering, restarting the networking
by init.d solves the problem. 

During the shutdown process, I get several "Too much work at
interrupt"s, with code 0x0001. So I checked the interrupts,
the kernel bootup messages say:

8139too Fast Ethernet driver 0.9.18a
PCI: Found IRQ11 for device 02:01.0
PCI: Sharing IRQ 11 with 00:1f:3
eth0: RealTek 8139 Fast Ethernet at ...., IRQ 11
eth0: Identified 8139 chip type 'RTL-8139C'

Looks like an IRQ conflict, /proc/interrupts confirms:

11: 930994 XT-PIC usb-uhci, usb-uhci, au88xx, eth0

I tried to solve this IRQ conflict in the BIOS, but all the
devices bind to INT pin 3, which is mapped to IRQ 11.


I really couldn't find more information, so if you have any
suggestions, please help.

My box:

Intel i815EP based mainboard
2.4.9-ac9 uniprocessor kernel

Thanx
         PP.