MX987xx NIC Linux driver support
stevechen@MXIC.COM.TW
stevechen@MXIC.COM.TW
Tue Sep 8 03:01:53 1998
--0__=EBXAyeNAPoWuz2YQRaWnvMMNobpQLA7gKjzH4SdW7JfZ95kSnOH52k8s
Content-type: text/plain; charset=big5
Content-Disposition: inline
Dear Sir,
Thanks for quick response, and also for the effort of supporting MX987xx
driver.
According to Thomas Sailer & Nicolas Edwards,
Donald Becker's tulip driver already support MX987xx chips, but still
have some problems left.
To make MX987xx driver work correctly with best performance,
I have some suggestion as follows,
1. For normal operation the CSR6,CSR16 should be set as follows,
CSR6 = 01a82202
CSR16 = 0f37xxxx (for MX98713A,PCI revision register = 0x10)
CSR16 = 0b3cxxxx (for MX98715, PCI revision register = 0x2x)
CSR16 = 0b3cxxxx (for MX98725, PCI revision register = 0x3x)
I guess this setting may fix your packet rx issue.
2. For forced media speed setting, the CSR6 & CSR14 should be
special taken card of according to MX98715 application note
(Page 4, section 4.41 SPEED SELECTION)
3. For EEPROM layout issue, Macronix's definition is totally different
from DEC's.
Macronix's driver only get the NetworkID (according to byte offset
0x70) for set up frame
during initialization.
So I think the Linux driver should do the same thing.
(refer to MX98715 application note Appendix A: EEPROM Format)
I am now here working together with you as much as I can to make it
possible
for MX987xx chips as standard support by Linux.
Please feel free to contact me for any MX987xx problems any time.
Best regards,
Steve Chen
Thomas Sailer <sailer@ife.ee.ethz.ch> 於 98/09/07 06:25:13 PM
到： Steve Chen/TAIWAN/MXIC
副本抄送:
主旨： Re: MX987xx NIC driver support
Dear Mr. Chen
Thank you for your interest in making Linux
drivers available for Linux.
The main Linux Ethernet driver author is
Donald Becker (becker@cesdis.gsfc.nasa.gov), and his
tulip driver (originally intended for the DEC 21xxx chips,
but now incorporating support for various compatible chips)
already basically supports the MX chips. There are however
a few problems left. I recently tried to remedy them
(with partial success) since I bought several MX98713A cards.
The problem areas currently are:
1. EEPROM layout. I've recently mailed a patch to Donald Becker
that parses the EEPROM layout documented in 98713app.pdf.
Do you know of manufacturers who do not follow the
specification in section 3.4?
2. Medium selection/Initialisation of certain registers
(CSR16 and above). I've basically taken the source
code of your diagnostic application as documentation
on how to do this.
3. There are a lot of duplicate packets received,
and the packet latency is very high. This is currently
the worst problem, and from databook/code inspection
I didn't find anything that could be the culprit.
The interesting thing is that this routine works with
the DEC chips. Do you know of any crucial difference
in packet rx (and possibly tx) between the Macronix and
the DEC chips?
Thank you again
Thomas Sailer
PS: I'll be away until end of Sept, so I can't respond
before October...
Nick Edwards <devel@nicholas.net> 於 98/09/07 11:58:25 PM
請回應 到 Nick Edwards <devel@nicholas.net>
到： Steve Chen/TAIWAN/MXIC
副本抄送: linux-net@vger.rutgers.edu, Planet Lin/TAIWAN/MXIC,
linux-tulip@cesdis1.gsfc.nasa.gov
主旨： Re: MX987xx NIC driver support
--0__=EBXAyeNAPoWuz2YQRaWnvMMNobpQLA7gKjzH4SdW7JfZ95kSnOH52k8s
Content-type: text/plain; charset=us-ascii
Content-Disposition: inline
Hello,
Glad to hear from you. I have several of your adapters and I am quite
pleased with their performance in windows 95, however, the Linux support
is severely lacking.
Currently, people seem to be using the driver for the DEC Tulip 2X41X
cards. While this *works* in the strickest sense of the word, performance
is not good, packet duplications are common, and it is impossible for the
driver to work in any media mode except 10base half-duplex.
The problem seems not to be with the register compatibility with the DEC
2141x, but rather with the fact that many cards with the MX chipsets do
not have full EEPROM tables. It is possible for the driver to include
vendor/chip specific programming for each card, and this has been done for
some other cards based on 2141x clones, etc.
The maintainer of the Linux 2141x driver is Donald Becker, a research
scientist at NASA. In addition to the Tulip driver, he has written many
other drivers for a myriad of Linux ethernet interfaces and is very highly
respected in the Linux community. I am not sure what his role is in
programming vendor-specific EEPROM information that would help to make the
MX cards more compatible, but you can certainly talk to him. His email
address is:
becker@cesdis.gsfc.nasa.gov
Again, thank you for your interest in the support of your chipset. There
are alot of ethernet vendors out there who are commited to not making
their programming information public for Linux support. I am sure that you
will see that working together with the Linux community will be the
fastest way to getting your products best supported and will benefit both
Linux users and your company.
Regards,
Nicholas Edwards
On Mon, 7 Sep 1998 stevechen@MXIC.COM.TW wrote:
>
> Dear Sir,
> My name is Steve Chen, system manager of Fast Ethernet
> product group of Macronix Internatinal Corp. Taiwan.
>
> We provides many kinds of Fast Ethernet series product including NIC,
> Hub, Bridge and Switch controller.
> Most of them was already volume shiped into the market.
> You can find the details from the web site: http://www.mxic.com.tw.
> As you can find in this web site, Macronix alread provided a complete
> set of high quality drivers (except Linux) for easier and more efficient
> way to
> interface with MX987xx series NIC product. (MX98713, MX98715 & MX98725
> which are basicly 95% register level compatible with DEC21140)
>
> How can I make it possible for these NICs as standard support by Linux ?
> Who will be the contact window for this kind of support ?
>
> My email address: stevechen@mxic.com.tw
> Headquarter address:
> Attention: Steve Chen, Dept 351
> Macronix International Co., Ltd.
> No. 3, Creation Road III
> Science-based industrial Park, HsinChu
> Taiwan
>
> Looging forward to hearing from you soon.
>
> Thanks & best regards,
> Steve Chen
>
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-net" in
> the body of a message to majordomo@vger.rutgers.edu
>
--0__=EBXAyeNAPoWuz2YQRaWnvMMNobpQLA7gKjzH4SdW7JfZ95kSnOH52k8s--