On Tue, May 02, 2000 at 04:13:29PM -0400, Johannes Erdfelt wrote:
> I didn't buy that solution for the simple reason that there are too many
> different types of busses (ISA, PCI, USB, IEEE1394, etc) to require user
> space applications to know about them all and know enough to probe the
> device to find the distinguishing characteristics.
>
> I argued that abstraction layers are there for a reason, to abstract
> away all of that.
>
> Unfortunately, some devices don't have any way to distinguish between
> them. The best we can do on USB for devices without serial numbers is
> mapping the bus topology to see which port the device is plugged into.
> This is far from fool proof tho.
In my opinion the way to identify a device should be configurable. If
the device doesn't have any unique ID, then the only way to identify it
is by bus topology. This will be true for many non-USB devices, too,
imagine PCI vga cards, they can be distinguished by slot # only. But if
there are some more ways than this, it should be possible to specify any
of them.
And because this would be done in userspace, this shouldn't be a big
problem to make the solution clever enough to handle all the cases.
> However, the eventual solution we came with was creating symlinks in
> /dev from a userspace daemon (devfsd or something else).
Yes, this is a very reasonable solution. But kernel support is still
needed to let the userspace daemon (and other applications - for example
X) know what devices are being connected and disconnected, preferably in
a uniform way for all busses. I believe that the very same daemon could
be responsible for loading any needed driver modules, too ...
--
Vojtech Pavlik
SuSE Labs
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
