On Tue, Jun 08, 2021 at 03:13:37PM +0200, Greg Kroah-Hartman wrote:
> > If multiple drivers are matchable what creates determinism in which
> > will bind?
>
> Magic :)
I suppose you mean something like this code:
/* Probe the USB device with the driver in hand, but only
* defer to a generic driver in case the current USB
* device driver has an id_table or a match function; i.e.,
* when the device driver was explicitly matched against
* a device.
*
* If the device driver does not have either of these,
* then we assume that it can bind to any device and is
* not truly a more specialized/non-generic driver, so a
* return value of -ENODEV should not force the device
* to be handled by the generic USB driver, as there
* can still be another, more specialized, device driver.
*
* This accommodates the usbip driver.
*
* TODO: What if, in the future, there are multiple
* specialized USB device drivers for a particular device?
* In such cases, there is a need to try all matching
* specialised device drivers prior to setting the
* use_generic_driver bit.
*/
error = udriver->probe(udev);
if (error == -ENODEV && udriver != &usb_generic_driver &&
(udriver->id_table || udriver->match)) {
udev->use_generic_driver = 1;
return -EPROBE_DEFER;
}
So it does use EPROBE_DEFER to recycle through another attempt at
driver binding. Yikes. Why didn't it return -ENODEV I wonder?
But OK, I can think of no way to find all the cases like this to even
try to do something else at this point, so this has to be
preserved. I'll try to do that, maybe add a comment or two.
Jason
