Subject: 10/100 card support
To: None <port-macppc@netbsd.org>
From: Donald Lee <donlee_ppc@icompute.com>
List: port-macppc
Date: 09/17/2001 13:38:50
I've been trying for some time to lay hands on a 10/100 ethernet card
that I could put in production. I have thwo cards now that "work", and one
on the way that supposedly will work *well*.
To wit:
-Report from the field on 10/100 ethernet PCI cards in a PM 7600/132
and NetBSD 1.5.1 kernel:
-the rtk driver (for a D-link DFE-530TX+ el-cheapo) works fine, sees
the 100Base T, but has poor performance, as far as I can tell.
Reportedly this is because it's a sub-marvelous chip.
-I have an Asante 09-00169-01 card (Dec 21140-AF chip). It
works, but only partially. It works fine in 10BaseT mode.
It does not auto-select 100Base T, nor FDX. I can manually select
100baseTX, but it starts getting errors and whines and complains
and gets lots of errors. The result is poor (dismal) performance.
I am using the de driver. I could try the tlp driver, but I'm
not sure exactly how. Is there a How-To somewhere on this?
I have an EtherPower II card coming my way in the next few days.
I'm hoping that will work well.
I may also get wild and buy a 3Com board. They're more expensive, though.
-dgl-
(below - included for context)
>On Wed, Jul 18, 2001 at 11:20:48AM -0700, Henry B. Hotz wrote:
>
> > Is there a discussion of this somewhere? I might want to get a
> > "good" 100BaseT card and how do I know which ones actually are
> > supported with no-copy drivers?
>
>Err, not sure if there's really any good discussion of this... but I
>can list off a few:
>
> - SMC EtherPower II (SMC 9432TX). This is supported by
> the "epic" driver. This is one of my favorite Ethernet
> chips, from a driver-writer's perspective.
>
> - Sundance Tech. ST-201. This is a pretty good chip, and
> is found on the D-Link DFE-550TX.
>
> - 3Com 3c905B, 3c905C, etc. These are pretty good, as well.
>
>Note I have listed here chips that can DMA to arbitrary buffer boundaries
>on the receive side -- many chips, for whatever braindamaged reasons, require
>a 4-byte aligned DMA buffer for receive, which means that the IP payload is
>misaligned, which means on arch's like PowerPC, you have to copy the packet
>in order to adjust alignment.
>
>--
> -- Jason R. Thorpe <thorpej@zembu.com>