[1320] in linux-net channel archive
Re: Network Drivers as Module
daemon@ATHENA.MIT.EDU (Administrateur)
Fri Nov 3 21:33:55 1995
Date: Fri, 3 Nov 1995 18:45:53 +0100 (MET)
From: Administrateur <root@Achaz.saclay.cea.fr>
To: Paul Gortmaker <gpg109@rsphy4.anu.edu.au>
cc: NIIBE Yutaka <gniibe@mri.co.jp>, linux-net@vger.rutgers.edu
In-Reply-To: <9510260834.AA02114@rsphy4.anu.edu.au>
On/Le Thu/Je, 26 Oct 1995, Paul Gortmaker wrote/=E9crit:
> >From "NIIBE Yutaka" Oct 25, 95 01:52:47 pm
>=20
> > Currently (as of 1.3.35), most network drivers cannot handle two (or
> > more) cards simulteneously as module. That is, we don't use module
> > for multi-homed host with same network cards.
>=20
> [...]
>=20
> Unfortunately the two things that usually get overlooked when people
> do device driver work is consideration for multiple card systems,
> and safe probing techniques.
> [...]
>
> I will do the same for the other 8390 cards, one at a time, when=20
> I get a free moment. Somebody else can do the non-8390 ones. If=20
> somebody uses the following patch as a pattern, it is then just an=20
> editing job. Also, I dscovered that none of the 8390 cards would have=20
> ever worked as modules with the 8390 core compiled in, due to missing=20
> ksyms. That is fixed as well. And I got rid of "loading device %s...",
> as each device has its own initialization banner/message anyway.
>=20
I happened to run against this problem last night. I put 2 smc cards in
the same box (one smc ultra, one etherez - they use 8390 and smc-ultra=20
modules). Although I didn't do very extensive testing I think I was=20
successful using a different approach. I ran modprobe twice, and changed=20
the module name the second time as follows:
modprobe smc-ultra io=3D0x220
modprobe smc-ultra -o etherez io=3D0x240
(or maybe it was insmod the second time ?).
Both cards were successfully recognized as eth0 and eth1 respectively.
The 8390 module was loaded once, and the smc-ultra module was loaded once
under "smc-ultra" and once under "etherez". Both seemed to work properly
(but again I didn't do extensive testing).
Now regarding ksyms, I've actually found a large number of symbols useful t=
o
modules that were not exported in recent 1.3.X kernels. I guess it's the
price you have to pay for being on the bleeding edge :).
Jean-Marc Verbavatz
