Re: Can multiple device drivers *share* a PCI bridge?

[H. Peter Anvin]

Followup to:  <[e-mail]>
By author:    Friedrich Steven E CONT CNIN <[e-mail]>
In newsgroup: linux.dev.kernel
>
> I have 5 IP modules (Industry Pak I/O) that plug onto an IP carrier.  The
> carrier has a bridge that gets found via vendor ID/device ID, but the *sub*
> devices don't show up as distinct pci devices.  I'm using the *new*
> approach, i.e., defining a pci_device_id struct that has been initialized
> with vendirID/deviceID pairs I'm supporting.
> 
> When my module loads, the kernel calls my probe routine.  If my probe
> routine returns 0, then this pci device is essentially locked to my device
> driver.  How can I share that pci device with multiple drivers?  My current
> thoughts are to simply make a *unified* driver that supports the various IP
> modules.  That unified driver is not a general solution, but it would be ok
> for this project.  I'm curious about how to develop a general solution to
> this problem.  I believe any user of these IP modules would want to be able
> to mix-n-match IP modules at will, merely adding device drivers, not having
> a unified driver.
> 

A properly designed device should have a separate PCI function (with
its own VID/DID) for each of the subdevices.  That's what the PCI
functions are all about.  Your device is doing something nonstandard,
so you need a shim device to handle its nonstandard decoding.

	-hpa
-- 
<[e-mail]> at work, <[e-mail]> in private!
"Unix gives you enough rope to shoot yourself in the foot."
http://www.zytor.com/~hpa/puzzle.txt