3c59x: recurring transmit errors (solved?)

Utz-Uwe Haus haus@zib.de
Tue Sep 8 12:41:21 1998


[I'm CC:ing linux-smp on this one to distribute my kludge to a wider audience]

> On Tue, 8 Sep 1998, Yeak Nai Siew wrote:
> 
> > Has anyone solved the problem with transmit errors for 3c59x driver?
> > 
> > The transmit erorr appears like this,
> > 
> > Sep  7 18:20:36 lisa kernel: eth0: transmit timed out, tx_status 00 status
> > e000.
> > Sep  7 18:21:07 lisa last message repeated 39 times
> > Sep  7 18:22:08 lisa last message repeated 46 times
> > Sep  7 18:23:10 lisa last message repeated 30 times
> 
> 
> I get these errors also. Sometimes they bring down the machine, sometimes
> not. I have not seen any fix for this problem posted on this list for the
> last few weeks, that I've been on it. I've not gotten any response from
> Donald Becker yet either, but I'm hopeful. I'm sure he is too busy.
I haven't gotten any reply either, however I haven't seen to many postings
from Don either...

I seem to have temporarily fixed the problem here on a SMP PII-system
running 2.1.120 by forcing the interrupt to be an XT-PIC style
interrupt (inserting (1<<10) into the blackout list in 
arch/i386/kernel/somewhere.c where 10 is the IRQ of the card). This forced
the interrupt to not be redirected and handled only by the boot CPU.
This is a hack and probably hides an SMP locking problem, however it
withstood 24h of testing with load averages >=10 and saturated 10Mb 
ethernet [tcpspray].
NOTE: I have the problem with a 3c905B, 21140 dec-based board and a rtl8139
but only the 3c59x and rtl give such 'nice' error reports before crashing
the machine.

Sincerely hoping for a better fix

Utz-Uwe Haus
-- 
Utz-Uwe Haus                                   haus@zib.de
                 Ask for my PGP public key!