v0.90 driver now available for test

Stephen Mills smills@lan1.com.au
Wed Oct 21 01:05:38 1998


Just thought Id say thanks for this new updated chipset Donald.
Finally my Compex FL100TX 21143PD card works perfectly with 0.90 [at the
moment :)] (.88K would 100% halt with aaic errors at boot)

This means I can use these Compex cards that we sell instead of buying the
3com cards which I see you also do the driver for which I do Linux installs
:)
I noticed the older driver .88H I believe would detect it as 21142/21143,
but this new driver detects as 21143 straight off the mark and works fine at
10 and 100 speeds.

Good work,
Stephen Mills
Sys Admin
Lan 1 Pty Ltd



>I've just made v0.90 of tulip.c available for testing on
>  ftp://cesdis.gsfc.nasa.gov/pub/linux/drivers/test/tulip.c
>
>It makes the following changes:
>
> 21143:
>   Masks upper bits on the Rx packet size to work with the new DS21143-PC
chip.
>
>   Initializes SIA registers on the 21143 that prevent a MII transceiver
>   from being detected with certain chip revs.
>
>   Once again runs the reset and initialization sequence for 21143 boards
>   with MII transceivers.  This is required on a few boards, like the
>   Adaptec 6911, which use GPIO outputs to enable the transceiver power.
>
>   Use full-power mode: the recommended 21143 snooze mode doesn't work with
>   the TDK transceiver which requires a continuously stable transmit clock.
>
> PNIC:
>   Bug fix for reading MII transceiver registers on PNIC-168 chip boards.
>   (The PNIC-169 has working hardware.)
>
> MXIC:
>   Various MXIC 98713 fixes (see below).
>
>I've a bunch of time tracking down the MXIC problems for this version.
>There were several:
>
>  I had thought that the duplicate receive packets were related to my
>  receive descriptor list structure, or some subtle timing issue when
>  manipulating the list.  After putting in zillions of consistency checks
>  I found that what was realling happing was that the 98713 occasionally
>  just dumps its Rx FIFO contents from the beginning.
>  The fix is the one suggested by Steve Chen from Macronix: setting the
>  undocumented CSR16 register to the magic value of 0x0F37****.
>
>  The MXIC actually takes 256 bytes of I/O space, not 128.
>
>  The fix for media selection was to have the MX98713 ignore the EEPROM
>  media table.   While the media table looks exactly like that of a 21140,
>  and the datasheet describes it as such, the '713 actually has a hidden
>  NWay autonegotiation engine that's only confused if the driver tries to
do
>  media selection.
>
>I'll be at the Atlanta Linux Showcase this weekend...
>
>Donald Becker   becker@cesdis.gsfc.nasa.gov
>USRA-CESDIS, Center of Excellence in Space Data and Information Sciences.
>Code 930.5, Goddard Space Flight Center,  Greenbelt, MD.  20771
>301-286-0882      http://cesdis.gsfc.nasa.gov/people/becker/whoiam.html