Vojtech Pavlik wrote:
> I'd have no problem with the per-interface one, if it is OK to allocate
> other interfaces than the one for which the probe was called and if the
> probe will be called even when the default configuration isn't CDC (many
> modems have primary configuration vendor-specific). I'm not sure about
> either and can't test it right now.
The first issue (allocating interfaces other than the one for which probe() is
called) is the same with either id table entry. Remember that the id table
says "give me any interface where these properties are true". It doesn't
matter whether you choose to test every interface where the device class is ==
2, or only a single interface where the interface class is ==2 && the subclass
is == 2; because you are still going to have to claim an interface other than
the one for which probe() is called. As long as we probe() on a single
interface, and some devices need more than one, then this problem exists.
Plus, the API supports this behaviour - usb_driver_claim_interface() exists
for this very reason :)

The second issue is important though. We need to ensure that we are calling
acm_probe() on every appropriate device, even if we occasionally call it on a
non-appropriate device.

> As for the per-device approach, it's OK, but slighly unelegant and means
> more modules than needed will be loaded.
Since the per-device approach is safer (doesn't break things that used to
work), and this is supposed to be a _stable_ kernel, I think I'd like to go
back to per-device for a while.

In the longer term (aka Mythical 2.5 Timeframe), perhaps we can sort out those
modems that have weird configurations, simplify acm_probe() and perhaps even
share some of the CDC parsing code.

Vojtech: does this approach suit you?

Brad

_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel

Reply via email to