The problem appeared when i moved from 2.0.33 to 2.1.105-2.1.108 in
'pcnet32.c' (same with 'lance.c') and makes the ethernet card quite unusable.
First I got a lot of kernel messages like
Jul 9 03:44:40 nomad kernel: Socket destroy delayed (r=0 w=12544)
and
Jul 9 03:45:49 nomad kernel: eth0: transmit timed out, status 0063, resetting.
from 'pcnet32.c'.
I made some modifications to 'pcnet32.c' (1) to discover the driver resets
the ethernet card when transmission wasn't possible within 20 ticks. But
since most transmissions succeded in a range from 1 to 100 ticks, it spent
most of the time with reinitialisation.
When changing the limit to 500 ticks the problem was gone.
Now what makes me curious is that 2.0.33 gave no trouble at all and
there 'lance.c' has the same limit of 20 ticks.
The machine i'm using:
- 2.1.105 uniprocessor kernel
- no kernel profiling
- IBM PC Server 325
- 1 PPro @ 180MHz
- 64MB RAM
- Ethernet controller: PCnet/PCI II 79C970A
(10Mbit, only two machines on the net)
- SCSI: Adaptec AIC-7880U, UW Controller
Bye,
Mark
(1)
see http://www.informatik.uni-rostock.de/~hopf/linux-patches/
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu