Re: [Libusb-devel] [RFC] Skip sending Set-Interface during driver unbinding

[Greg KH <gr...@kr...>]

On Fri, Nov 28, 2008 at 02:53:37PM -0500, Alan Stern wrote:
> Currently usbcore sends a Set-Interface request to reinstall altsetting 
> 0 whenever a driver unbinds from an interface.  There are three reasons 
> for doing this:
> 
>      1. When another driver comes along and probes the interface, it
> 	will expect to find the interface in altsetting 0.
> 
>      2. When an interface has multiple altsettings, the altsettings
> 	above 0 generally provide more capabilities and therefore use
> 	more power.  Hence installing altsetting 0 can save energy.
> 
>      3. Set-Interface acts as a lightweight reset for the interface.
> 
> Unforunately, there are quite a few devices with deficient firmware 
> that crashes when it receives a Set-Interface request.  Cypress's EZ 
> USB devices are notorious for this.  We have a quirk flag to prevent 
> sending the request, but adding device IDs to a quirks list is rarely 
> very satisfactory.
> 
> Instead, I propose to send the request only when the interface isn't in 
> altsetting 0 already.  This will continue to fulfill goals 1 and 2 
> above.  Goal 3 is relatively unimportant, since Set-Interface isn't 
> much of a reset.  (Mostly it just sets the data toggles for all the 
> endpoints in the interface back to 0 and clears the endpoint FIFOs.)  
> 
> And of course, if a device doesn't implement Set-Interface correctly
> then it will never leave altsetting 0, so we won't end up crashing it.
> 
> Does anybody have any comments on or objections to the patch below?

No objection from me, it makes sense.

thanks,

greg k-h