Re: 2.3 wish: integrate pcmcia into mainstream kernel

tytso@mit.edu
Thu, 3 Jun 1999 12:23:30 -0400


A couple of observations here. First of all, I'm told that Linus has a
Sony Vaio laptop (one of the SLimtops, which I have as well), and yes,
that *does* have the BIOS initializing the CD-ROM. I suspect it only
works for the laptop which is specially made for that laptop, but that's
one of the uglinesses.

Note that one of the reasons why this can be done is precisely to
support booting off of El Torito CD-ROM's. If you're going to do this,
you really do need to have the BIOS initialize the card for you.

So I think the main change which the PCMCIA drivers need to make is one
where they can notice if the CD-ROM was plugged in during the initial
power-on (and so was initialized by the BIOS), and leave well enough
alone in that case. That's really the only change that needs to be
made; I don't think there is a requirement to solve this particular
problem that all of the PCMCIA support be moved into the kernel.

Secondly, there are still a number of very good reasons why the PCMCIA
support should remain separate. Hardware manufacturers in this space
are like hardware manufacturers of video cards. They make changes and
tweaks to the hardware frequently. In fact, sometimes it feels like
it's about roughly the same change frequency as new Linux development
kernels come out. :-)

By having the PCMCIA support be done outside of kernel space, it means
that David can release new PCMCIA drivers much more frequently without
trying to get changes into the stable kernel series, *OR* forcing users
to use a development kernel just so their devices are supported. Even
if the necessary work is done to allow the first option to work, it (a)
would require a number of additional stable releases, with the
temptation of putting other destablizing new features into the stable
kernel reelase, and (b) it's more work to maintain two branches of
device drivers, one in the stable and one in the development kernel.

There is also, as David mentioned, the additional difficulty of keeping
the user space and kernel space drivers in sync if they are distributed
separately. As the modultils packages have shown, this isn't
impossible, but it does add to the support burden.

So before we overreact and assume that we have to put PCMCIA support
into the kernel, I think we need to be a little bit more careful. There
will certainly be some cases --- for example, if you need your boot
device be a PCMCIA device while you are bootstrapping a new port, but in
general, I think our existing implementations have a number of
advantages to recommend it, and we need to be careful that we keep
those advantages when going to a new device architecture.

- Ted

-
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/