Re: [PATCH] [PNP] 'modalias' sysfs export

[Kay Sievers <kay.sievers@vrfy.org>]

On Sat, Mar 11, 2006 at 11:29:57PM -0500, Adam Belay wrote:
> On Sat, Mar 11, 2006 at 05:38:47PM -0800, Andrew Morton wrote:
> > Pierre Ossman <drzeus-list@drzeus.cx> wrote:
> > >
> > >  Here is a patch for doing multi line modalias for PNP devices. This will
> > >  break udev, so that needs to be updated first.
> > > 
> > >  I had a longer look at the card part and it seems that module aliases
> > >  cannot be reliably used for it. Not without restructuring the system at
> > >  least. The possible combinations explode when you notice that the driver
> > >  ids needs to be just at subset of the card, without any ordering.
> > > 
> > >  If I got my calculations right, a PNP card would have to have roughly
> > >  2^(2n) aliases, where n is the number of device ids. So right now, I
> > >  lean towards only adding modalias support for the non-card part of the
> > >  PNP layer.
> > > 
> > >  Andrew, do you want a fix for the patch in -mm or can you remove the
> > >  part of it that modifies drivers/pnp/card.c by yourself?
> > 
> > I assume you mean that the drivers/pnp/card.c patch of
> > pnp-modalias-sysfs-export.patch needs to be removed and this patch applies
> > on top of the result.
> > 
> > But I don't want to break udev.
> 
> I think supporting multiple IDs per node is a reasonable expectation to
> have from udev (even for subsystems beyond PnP).  Kay, would this be
> difficult to add?

Udev does not care about $MODALIAS at all about the string, it just
runs configured programs when a device is added. It's unlikely,
that we will ever need/have a built-in MODALIAS handling in udev.
Distros handle that all differently, most just do "modprobe $MODALIAS"
with the device event.

> However, I'm a little confused as to why we're exporting these "modalias"
> strings from a kernel interface in the first place.  Wouldn't it be better
> from an abstraction barrier standpoint to have userspace generate such
> strings from information it gathered through bus-specific interfaces?

Well, the modalias match strings are native part of the kernel modules, so
it is just convenient to have the kernel to create the modalias to match
against these strings too. Userspace and modprobe usually doesn't care about
the actual content of the strings, as both come from the kernel and just need
to match each other. In theory, there is no need to keep userspace updated,
if a new bus is added, or the format is changed, which is nice.

> As
> can be seen from this example, when the modalias (or essentially a device
> identification key) is imposed at the kernel level driver model interface,
> one has to make policy decisions (e.g. what legacy features such as isapnp
> are we going to only psuedo-support).

What "policy"? We only need a way to export the multiple matches in a
sane way and using a multiline value is not really nice, especially when
added to the environment.

Thanks,
Kay