On 22 Apr 2000 Thilo Manske wrote:
>> To my surprise and delight, my kernel recognizes the 3Com 3c905 board.
>> Unfortunately, it does not work.
>Well, *my* 3c905B works extremly well in my Umax Apus Mac clone (ex
>driver/netbsd current).
>
>Do you have the exphy driver in your config file as well?
>
>ex*     at pci? dev ? function ?        # 3Com 90x[B]
>exphy*  at mii? phy ?                   # 3Com internal PHYs
>
>BTW: Here's my little survey othe range of 100TX PCI NICs I have:
>
>epic (SMC 9432TX, I think this is the Etherpower II) doesn't work
>     - kernel keeps printing "epic0: receiver queue empty"
>rtl  (RealTek 8139) does not work, it can send but doesn't receive
>     anything
>vr   (D-Link DFE-530TX, not plus!) works, but I had some problems with
>     it on my i386 (see PR kern/7948) and AFAIK that hasn't been fixed
>     yet (at least I haven't seen a change in the sources that might fix
>     it).
>ex   (3Com905B-TX) works like a charm
>
>Tested in a Umax Apus 3000 with a few days old -current.

I went wild and built several kernels with various configurations.
When I added both the ex* and the exphy* devices, my 3c905[b]-TX
card got configured by the kernel, and appeared to work...

... that is until I stressed the card.

It works fine as long as I don't try to move data any faster than about
1 MByte/sec.  If I connect to the NetBSD server with the 10/100
card from another machine with a 10/100 capability, I get errors
on the console: "/netbsd: ex0: device timeout".  These messages are
coming out of the src/sys/dev/ic/elinkxl.c. at about line 1446.

When it works, it's sweet.  I've seen transfers for a few seconds at
a time of about 3.5 Mbytes/sec.

I did some diffs between the current and 1.4.2 sources and did not find
anything different about the sources that appear to affect the
ex(4) deiver.

I've dug in a little, but I don't know enough about the system, or
the chip to be of much use.

Ideas?

Thanks,

-dgl-