[vortex] re :message when net card crash

[Bogdan Costescu]

On Fri, 19 Apr 2002, Wu Yadong wrote:

> When net card crash , system is ok ,other processes can run without problems, the net card can not send frames any more, and the screen  will continuously display below messages until system shut down :
> 
> eth0:transmit timed out, tx_status 00 status e000.
> Flags;bus-master 1,full 1;dirty 312953(9) current 312969(9).

                     ^^^^^^ - what kernel/driver version is this ?

> Transmit list 012b9290 vs. c12b9290.
> followed by a list of frames

OK, more details this time, better chances to diagnoze. First, check if 
there are no other error messages from the driver - the one liners can 
easily be overlooked through the tens of lines the Tx timeouts write.

A Tx timeout is nothing special on a 10baseT hub. Because of the lower 
speed (compared with a 100baseTx one) and the fact that is a hub (so it's 
affected by collisions), the Tx queue can stay full for some time and the 
Tx timeout condition be triggered more often than for a 100baseTx one.
You can try to increase the Tx timeout value and see what happens.

> Is there something which I can do to avoid this? Should I adjust sending speed while testing, or do something else( wait until all frames in queue have been sent)?

The return value of the dev_start_xmit function can tell if the packet was 
succesfully queued or not, in which case most likely the queue is full. 
But I don't know if your packet generator is so low level to have access 
to this info.

Limiting sending speed might not be a solution, as even with a low one 
there might a situation when packets are just sitting in the queue and 
accumulate and some time later the Tx timeout will occur.

-- 
Bogdan Costescu

IWR - Interdisziplinaeres Zentrum fuer Wissenschaftliches Rechnen
Universitaet Heidelberg, INF 368, D-69120 Heidelberg, GERMANY
Telephone: +49 6221 54 8869, Telefax: +49 6221 54 8868
E-mail: Bogdan.Costescu@IWR.Uni-Heidelberg.De