Re: PNP patch into kernel when?

Andrew E. Mileski (
Wed, 4 Dec 1996 09:58:49 -0500 (EST)

> Before we (noticeable Linus) add yet another interface for device
> allocation/registration to the kernel, I think we should try to look
> at the needs for all the various architectures and try to specify a
> common interface that is similar for all (bus)architectures. How does
> the new PnP interface compare to the existing Zorro, SBUS and PCI
> config interfaces?

I have no idea, as I've never used them. The resource management API
is very low level; it deals with IRQ, DMA, I/O, address, etc., allocation.
Sitting on top of this is the _optional_ Plug-and-Play API, which is
intended to handle any kind of device that requires resources (all
devices require some sort of hardware resource).

Hooks are already in place for: PnP-ISA, PnP-BIOS, PCI, PCMCIA,
EISA, legacy (non-identifiable and/or non-configurable). Adding hooks
for new device types is easily done. Also, stuff like devices
on docking stations are supported.

At the moment, the interface favours PnP-ISA/PnP-BIOS, but this is
soon to change.

> I see no reason to have N fundamentally different interfaces for this
> if we can have one common interface. This will also be usefull if/when
> we try to integrate/merge various drivers (for instance I dont see
> much of a reason for having 7+ different Lance Ethernet drivers in the
> kernel).

This is exactly what the PnP API will hopefully achieve. A driver
will ask for what devices are present, determine which ones it can handle,
and request that the devices be configured and activated.

Andrew E. Mileski
Linux Plug-and-Play Kernel Project
XFree86 Matrox Team