[ltp] MWave-Linux-Petition

[Valery Trifonov]

On Thu, 9 Sep 1999 22:17:57 -0700 (PDT) Paul Rubin <phr@netcom.com> wrote:

>         Maybe, but I believe the rest of your post makes an excellent
>         point in favor of Mwave support for Linux.
> 
> No.  See below.

> The issue isn't just CPU load.

Sorry, that was the impression I got from your reference to "the advocates of
using the main cpu for these things".

> The Mwave is a separate cpu whose
> firmware has to be loaded from the main cpu, which is exactly the
> wrong thing in this situation.  Otherwise, why stop with the Mwave?
> Why not ask for more hair in the Linux kernel to load firmware into
> the keyboard controller, the battery controller, the different
> processors on the disk drive, etc.--all of these things using
> different special Thinkpad specific interfaces that all had to be
> programmed?

I don't see any reason why the kernel should be loading the firmware, and even
if necessary it's not much of an issue with a modularized kernel. Aside from
that, apparently in some cases downloadable firmware turns out to be a more
cost-efficient path to providing upgrades, and you're fighting market forces.

> Nobody would ever have bothered to get Linux working on
> Thinkpads if that was all required.

True, but it's for the manufacturer to decide if they'll go with a standard
but less flexible interface supported by the whole market, or with a more
flexible and perhaps cheaper solution for a pretty big chunk of the
market. Considering that, as with the video cards, opening the specs to the
public often makes up for the non-standard interface.

> Generic PC's have been enormously
> successful because all the different subsystems have standardized
> interfaces to the software.  In the case of modems, that means the AT
> command set, not some crazy machine specific Mwave weirdness whether
> the weirdness is documented or not.  The same goes for the
> soundblaster functions, which also should not be done on the mwave.
> They should use standard modem and sound chips that respond properly
> to the normal AT commands etc. without needing any special nonstandard
> kernel attention at initialization or at any other time.

There has always been friction between standardization and innovation. New
interfaces appear all the time, different modems have different extensions to
the basic command set, even the Crystal 423[2-9] sound chips are slightly
different. As Bill said there is already an impressive number of kernel
modules dealing with all these interfaces. Sure it would be easier for the
software development if all devices used the same interface, but it would be a
bit hard to make progress in hardware design. I don't see anything inherently
wrong with another interface, as long as it's open.

OTOH having programmable peripherals can help both the software and hardware
development, if the interface is open and at the right level of
abstraction. To borrow an example from the previous round of expansion of
embedded processing, there are quite a few models of PostScript printers out
there, and people are basically happy downloading programs to them. Or
consider OpenGL. Of course a widely-supported standard helps, but sound
processing is still young.

Is this thread way off-topic yet?

Valery