Re: 3com 905 driver still missing from 2.0.31 prepatch

Torbjorn Lindgren (tl@funcom.no)
Wed, 6 Aug 1997 10:25:33 +0200 (MDT)


Message CC'd to several people who have talked about 3c90x drivers and
problems, and isn't using 3c59x.c v0.42[mn].

On Tue, 5 Aug 1997, Christof Damian wrote:
> I just checked RedHat in kernel-2.0.30-3.src.rpm (and upto
> kernel-2.0.30-10.src.rpm) they use 3c59x.c 0.40 (3c900.c:v0.40 4/16/97
> becker@cesdis.gsfc.nasa.gov) with this patch:
>
> --- 3c59x.c 1997/04/17 04:13:13 1.5
> +++ 3c59x.c 1997/04/19 16:50:30
> @@ -722,7 +722,7 @@
> vp->info2 = eeprom[15];
> vp->capabilities = eeprom[16];
> if (vp->capabilities & 0x20) {
> - vp->full_bus_master_tx = 1;
> + vp->full_bus_master_tx = 0;
> if (vp->info2 & 1) {
> printk(" Rx Pacing bug fixed, enabling bus-master
> receives.\n")
> ;
> vp->full_bus_master_rx = 1;
>
>
> (basicly what I am using too, only with 0.42)

Is that really necessary. That patch was for 0.40 and *earlier* which had
problems if you enabled busmastering....

I have several machines that required that for 0.37 and 0.40, but works
perfect with 0.42m. The plain 0.41 and 0.42 works too, except that they
sometimes get into trouble with collisions.

In short, I belive that patch is totally useless now, and it certainly
harms the throughput. With that patch you won't get anywhere near full
performance out of the card (in excess of 10+10 MB/s in full duplex mode
according to reports).

> > Before that I used many other versions, including v0.39 and v0.41. All of them
> > had random Tx timeout problems, which could be solved by doing an ifconfig
> > down/up.
>
> 0.42 can recover from timeout problems without ifconfig down/up, but it
> still take a long time to recover

Which is why you should use 0.42n, not 0.42. It can be retreived from
URL:http://www.desy.de:8888/~sieversm/boomerang.html
as reported on the Vortex mailing list.

This problem has been solved for a while, the original patch was for
0.41. The 0.42n patch is much cleaner and neater, and perhaps it will be
in 0.43.

I don't know why it isn't in the base driver yet, nor why most people who
posting on the Vortex mailing-list evidently doesn't read it :-)
(New members? All of them?)

Personally I think I concur with the patch's author (Michael Sivert), in
short:

Since Donald Becker will probably get all the complaints (and flames)
anyway, he should have a say in which version goes into the kernel!

Having said that, 0.42n *is* the version I would prefer to get included in
the kernel, unless 0.43 is released soon with similar enhancments.

-- 
Torbjörn Lindgren
E-mail: tl@funcom.com
If Santa ever DID deliver presents on Christmas Eve, he's dead now.