On Thu, Jul 12, 2007 at 06:11:32PM +0200, renardier philippe wrote: > > 1) I can't understand relationship between, say, pci driver and > network driver. If my ethernet controler is hosted on a pci bus is > this device a struct net_device or a struct pci_dev ? Both are involved. > If it is a pci_dev it's registered by the pci core before the core > calls the probe callback of my driver. Yes. > If it is a net_device my driver must call register_netdev for the > device. Yes. > What's the right way ? You need to do both. > (if both device structures are needed when to load the module which > holds net_device driver code with respect to the loading of the module > which holds pci_dev driver code ?) Look at the zillion real pci network drivers in the kernel tree for how to do this properly if you are confused. > (I have focused on struct net_device as an example but the same > question happens for other classes of devices whose controler can be > hosted on a pci bus). Exactly, you have to handle both. Your driver is the binding between the hardware layer bus, and the logical layer bus (pci and networking in your example.) > 2) An other (unrelated) question is : why usb driver must register > itself the devices it is bounded to ? (unlike pci driver). There is no difference between the way usb and pci drivers register themselves, other than the function calls and the structures involved. The process is the same. > It would make sens to have the same registration scheme for devices on > both buses. I agree, how do you think they are different? The same person wrote them both :) thanks, greg k-h -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ
