MX 98715 AEC

[Donald Becker]

On Sun, 31 Oct 1999, IronCode wrote:

> Recently I bought a CNet PRO120 NIC, which is 98715 based. tulip detects
> it, everything's ok. But when the driver loads, it prints:
> 
> eth0: Macronix 98715 PMAC rev 32 at 0x6800, 00:80:AD:7A:5D:A7, IRQ 12.
> eth0: EEPROM default media type 10baseT.

What driver version?

> So why it's not autodetect (as should be the default of my NIC), but it's
> 10baseT?

Your EEPROM-set media type is 10baseT, half duplex.
The driver should obey this forced setting.

The tulip-diag program can be used to change this.

> I downloaded some docs from the Macronix' site about the MX 98715. That's
> what I found:
> 
> Also, the MX98715 is equipped with intelligent IEEE802.3u-
> compliant Nway auto-negotiation capability allowing a
> single RJ-45 connector to link with the other IEEE802.3u-
> compliant device without re-configuration.

When you force the media type, you implicitly disable autonegotiation.

> To optimize
> operating bandwidth, network data integrity and through-
> put, the proprietary Adaptive Network Throughput Con-
> trol (ANTC) function is implemented.

The technical name for that is "bullshit".  They mean dynamic threshold
setting, which Linux drivers have implemented for years.  (Although I
deliberately never reduce the Tx threshold, just increase it on underruns.)

A similar situation was using ping-pong buffers on the 8390.  I believe that
my drivers were the first to do this (1992).  A few years later bunches
of NE2000 clones makers were touting their "proprietary" technique to
dramatically improve transmit performance.


Donald Becker
Scyld Computing Corporation, and
USRA-CESDIS,   becker@cesdis.gsfc.nasa.gov