[eepro100] Transmitter Timeout -- addednum

Donald Becker becker@scyld.com
Tue, 1 Aug 2000 15:52:36 -0400 (EDT)


On Tue, 1 Aug 2000, Kallol Biswas wrote:

> > > the resume the card executes the old cmd word and stalls. We would
> > > not have this problem if all the cmd words in the ring were only
> > > TX commands, as a prefetched command would always be a TX cmd.
> 
> You could have two rings, one for Tx cmds another for non-tx cmds,

That's almost what the current code does.  Long command are linked to by
NoOps inserted in the regular Tx ring.

> what I am trying to say is prefetching of a NOP does not cause any
> harm, it wastes a few cpu cycles, I have logged the command
> q length for my driver and found that a NOP would be executed after
> 33 Tx commands on the average under heavy load condition.

I'm not certain what you mean by this: when it's time to do a command, it
must be done immediately, not after the command unit reaches the end of the
command ring.  For instance, a multicast list change on a machine that isn't
transmitting should still take place immediately.

> BTW, how frequently do you send the multicast workaround for receiver lockup
> bug? If the receiver is idle for two seconds and you have rx_bug set, right?

Correct.  This bug is presumably fixed in the current hardware, the EEPROM
bits indicating the error exists are still set.

Donald Becker				becker@scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Beowulf Clusters / Linux Installations
Annapolis MD 21403