Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> writes: > The biggest bug may not be an obvious one. Suppose the lvstest driver > has been built into the kernel. When the kernel boots and the root > hubs are registered, what will prevent them all from binding to lvstest > instead of the normal hub driver? I don't know if this is possible to implement in any sane way, but I believe it would be useful for the driver's probe to know that it is called as a result of a write to 'bind'. Similar issues will apply to all cases where multiple drivers are matching and where the driver selection is a question about policy. The system should then have a well defined default, but still allow the user to override this default on a device-by-device basis by e.g. manually binding a device to the non-default driver. NCM 1.0 backwards compatible functions (aka "NCM/MBIM functions") are an example that is on my mind (naturally :-). We currently let the user select NCM or MBIM using a module parameter, but we should ideally let the user override this system wide default per device by writing to 'bind'. Not because I see any use case for mixed NCM/MBIM priorities, but because it doesn't seem logical that e.g. the cdc_ncm driver returns -ENODEV when you try to manually bind a perfectly fine NCM/MBIM function to it with the system default set to MBIM. I am sure there are other similar cases, where the exact choice of driver really is a policy question for the user. Bjørn -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
