Subject: Re: bce data error with BCM4401-B0
To: Brook Milligan <brook@biology.nmsu.edu>
From: Daniel Carosone <dan@geek.com.au>
List: current-users
Date: 08/01/2006 08:25:23
--lIC76ItX9S6XOZ/S
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Mon, Jul 31, 2006 at 11:17:13AM -0600, Brook Milligan wrote:
> I am trying to install NetBSD (2006.07.07.00.00.00 for now) on a Dell
> XPS M1210 laptop.=20

Cool. I have one of these on order to replace a decrepit and rapidly
deteriorating older laptop.  It's taking longer to arrive than I'd
expected.  I made a reasonable effort to check that all the devices i
could find specifications for would be supported.  Could you post, or
mail me privately, the dmesg?

> The kernel detects the ethernet device as:
>=20
>     bce0 at pci5 dev 0 function 0: Broadcom BCM4401-B0 10/100 Ethernet
>=20
> [... and it sucks]

Yes, this is a known hardware problem, the chip can't DMA above
1G. This one crappy device caused me a lot of angst in deciding
whether to get this laptop or not, as mine will have 2G (and I would
have preferred a gigE, though not essential - perhaps I'll get a gigE
expresscard one day).

My intended workaround for this problem is to run Xen, and to limit
Dom0 to smaller memory; I hope that Xen's memory allocation
corresponds to physical layout as much as it does to sizing. This will
allow other domains to use the rest of the memory, rather than leaving
it unused as would be the case for the workaround suggested in bce(4)
when running native.  I specifically waited for laptops with the
virtualisation extensions to appear before replacing the older laptop,
because I want to run Xen regardless, so assuming this works I will be
no worse off.

Now, ways to work around the bce hardware limitation have been
discussed, and at least partially tried - though not yet successfully.
They're similar to the bounce-buffers that were needed for ISA devices
with addressing limitations.  The motivation to pursue these was
limited, in part, by an assumption that the crappy chip would
disappear, but now its use seems to be continuing in new systems,
motivation will increase. I guess, at least, mine will if the
alternative winds up with me needing a linux dom0.

> By the way, for testing I would like to boot this machine from a USB
> stick.  Given that I can compile a system, how should I format the USB
> stick to be bootable?

Just treat the usb stick as a hard disk: fdisk, disklabel, newfs and
installboot as normal.

--
Dan.

--lIC76ItX9S6XOZ/S
Content-Type: application/pgp-signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.4 (NetBSD)

iD8DBQFEzoNTEAVxvV4N66cRAn9TAJwKhuJ09qojCX8GIMfl7qVqq51qsACdEZW6
ZJDoh1D9OfBIhMXgrra7Bs0=
=z8xq
-----END PGP SIGNATURE-----

--lIC76ItX9S6XOZ/S--