On Thu, Oct 10, 2002, Oliver Neukum <[EMAIL PROTECTED]> wrote: > Am Donnerstag, 10. Oktober 2002 01:31 schrieb Johannes Erdfelt: > > On Thu, Oct 10, 2002, Oliver Neukum <[EMAIL PROTECTED]> wrote: > > > I've started looking at doing configurations correctly. > > > There are several questions. > > > > > > 1. Must it be legal for a driver to set configuration during probe ? > > > > This is tricky. There is a certain bit of coordination that is implied > > with selecting a configuration since it can change the interfaces > > available. > > > > We don't want to switch a configuration out from under another driver, > > nor do we want to switch a configuration before another interface is > > offered it. > > > > I'm not sure what the best answer here is and I'm sure how we choose the > > configuration will determine the answer to this question :) > > I hope the answer is no. Else this'll be very painful.
I hope so too. There's some tricky situations in that case. I think the big remaining question right now is determining how a configuration should be elected. In my devfs patch a long time ago, I had created an election process where drivers could specify the priority of their support. The highest priority won. Here's the thought process I originally used: - Multiple configurations are rare - Devices won't change substantially between configurations (ie won't go from a keyboard to a cdrom) - Used typically to provide different levels of support depending on bus or host limitations (ie config for lower power, config for generic printer class, but another one for a superset ala uss720) If that's the case, a priority would probably work well. Can anyone think of a reason to use different configurations outside of what I was thinking? > > > 4. May I split usb_disconnect() into an interface and a usb_device > > > part ? > > > > Depends on the intent. If it's to notify a driver that the configuration > > changed out from underneath it and the interface no longer exists, sure. > > > > Other than that, you don't really "disconnect" an interface, but that's > > just wording. > > Yes, I need infrastructure to "disconnect" the interface without disconnecting > the usb_interface. Understood. This also brings up the question of if we should allow the configuration to change miduse (meaning after drivers have bound already to any interfaces)? JE ------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf _______________________________________________ [EMAIL PROTECTED] To unsubscribe, use the last form field at: https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
