Re: [Q]: Linux and real device drivers

Nathan Hand (nathanh@chirp.com.au)
Thu, 30 Sep 1999 12:49:54 +1000

On Tue, Sep 28, 1999 at 02:14:57PM -0400, tytso@mit.edu wrote:
> Date: Mon, 27 Sep 1999 15:48:11 +0100 (BST)
> From: Alan Cox <alan@lxorguk.ukuu.org.uk>
>
> > User buys Banavision capture card. Oops! Not supported in 2.2. User
> > has to wait for the 2.4 kernel to come out (months and months
> > away).... or use a development kernel. Whoops! 2.3 kernel has
>
> Come off it Ted, you know perfectly well we back port drivers as they become
> stable - aggressively so.
>
> Yes, but (1) it takes time before they show up in the 2.2 releases, and
> (2) does this really scale? Consider how many speciality devices are
> out there --- from business card scanners to TV digitizing boards, etc.
> Can you really keep up with all of the new devices? Remember, today
> people can purchase these devices and they come with device driver disks
> for Windows. Are we saying that there's something Windows can do that
> we can't? That's embarassing.

Many drivers are appearing in user space anyway, often with truly token
kernel support. Examples

- XFree86 (uses mmap and user space dma)

- SANE (uses scsi generic, perhaps ide-scsi emulation)

- cdrecord (uses scsi generic, perhaps ide-scsi emulation)

- ghostscript (de-facto printer drivers)

Locking the kernel interfaces down won't help any company coding binary
only drivers for any of these projects.

My point: the "problem" is much broader than just the kernel. You can't
save the world by locking down the Linux kernel driver interfaces.

> OK, so we don't want to lock down interfaces forever. I can accept
> that. But do we really need to allow interfaces to arbitrarily change
> during a stable kernel series? (Most of the times when we've need to
> make such major changes to the stable kernel series, particularly during
> the 2.0 series, stability has suffered.) I'd claim that it really isn't
> that difficult to lock down internal interfaces during a stable kernel
> release, and allow manufacturers to release device drivers that work
> against 2.2 kernels. If they need to rewrite their drivers for the 2.4
> kernels, fine.

Every stability issue I've ever had with Linux has been traced to a bug
in a driver. Binary only drivers would have meant adding kludges to the
kernel to work around bugs in the drivers. No thanks. I want source for
my driver, and if I have source, there's no problem changing interfaces
and recompiling (someone will know how it's done).

My point: there's no such thing as a "stable kernel release" if drivers
are available only in binary form. Drivers Needs Source.

ObReligiousArgument: whether everything needs source or not isn't worth
arguing about, but drivers are too valuable to be binary only.

-- 
Nathan Hand - Chirp Web Design - http://www.chirp.com.au/ - $e^{i\pi}+1 = 0$
Phone: +61 2 6230 1871   Fax: +61 2 6230 1515   E-mail: nathanh@chirp.com.au


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/