On Wed, Jul 26, 2000 at 02:56:35PM -0500, Erich T. Enke wrote:
> Sorry for the vicious cross-posting, but it concerns all three
> groups.

I guess this could concern more than this - alpha for sure, maybe sparc, 
cobalt, ... well, anything with a PCI bus :)

> 
> johnh and I found a fix for our nsphy problem today.  The problem had to
> do with a commented out, undocumented mystery bit, removed during the mii
> changes from 1.3.3 to 1.4.  The fix uncomments these bits and fixes
> autonegotiation and nastiness in general:

Maybe you could send a PR about this, so it doesn't get lost ?

> 
> --- nsphy.c.orig        Wed Jul 26 13:03:43 2000
> +++ nsphy.c     Wed Jul 26 14:38:06 2000
> @@ -218,14 +218,17 @@
>                  */
>                 reg &= ~PCR_FLINK100;
>  
> -#if 0
>                 /*
>                  * Mystery bits which are supposedly `reserved',
>                  * but we seem to need to set them when the PHY
> -                * is connected to some interfaces!
> +                * is connected to some interfaces:
> +                *
> +                * 0x0400 is needed for fxp
> +                *        (Intel EtherExpress Pro 10+/100B, 82557 chip)
> +                *        (nsphy with a DP83840 chip)
> +                * 0x0100 may be needed for some other card
>                  */
>                 reg |= 0x0100 | 0x0400;

Do you know if the chip revision was different on these boards ?

--
Manuel Bouyer <bouyer@antioche.eu.org>
--