[tulip-bug] ADMtek Comet bug

Donald Becker becker@scyld.com
Wed, 18 Oct 2000 00:54:10 -0400 (EDT)


On Tue, 17 Oct 2000, Dan Hollis wrote:

> > >  Port selection is 100mbps-SYM/PCS 100baseTx scrambler, half-duplex.
>
> The claim of "half-duplex" is *definitely* wrong btw. Its full duplex for
> sure.

The Comet/Centaur are a special case -- the duplex is set and read only from
the transceiver register.  It's a cleaner design than other Tulip chips, but
it means that the standard CSR6 register report is inaccurate.

> > >   The transmit unit is set to store-and-forward.
> > This setting is a little curious.  What is the setting of CSR6 when you
> > first start the interface and it's working?  (CSR6 is ffb7e117, the sixth
> > entry above.)
> 
> Here is the card immediately after startup:
> 
> ADMtek AL985 Centaur-P chip registers at 0xe000:
>   fff98000 ffffffff ffffffff 01165000 01165200 fc664010 ff972117 ffffebff
...
>   The transmit threshold is 128.

OK, we that's a little curious.  The chip's Tx FIFO underflowed enough times
for the chip to switch to store-and-forward mode!  That's what the driver is
supposed to do, but it indicates a PCI bandwidth problem.  What else are you
running on the PCI bus?


> After a minute or two:
> ADMtek AL985 Centaur-P chip registers at 0xe000:
>   fff98000 ffffffff ffffffff 01165000 01165200 fc664010 ff97e117 ffffebff
...
>   The transmit threshold is 1024.

The driver has already increased the Tx threshold!  This isn't the problem,
but it is suspicious.

Donald Becker				becker@scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Second Generation Beowulf Clusters
Annapolis MD 21403			410-990-9993