On Thu, Jun 24, 2004 at 03:08:01PM -0400, Alan Stern wrote: > Greg: > > This patch creates a special locking routine intended specifically for use > when calling usb_reset_device(). A special routine like this is needed to > avoid deadlocks with disconnect(). Symbolically: > > Khubd Driver > ----- ------ > Wants to reset a device, so calls > its own do_reset() function > Receives connect-change > notification and calls > usb_disconnect() > > Locks the device and calls > driver->disconnect() > do_reset() tries to lock the > device for the reset, and blocks > disconnect() can't return > until do_reset() is finished > > One simple-minded way out would be for the driver's do_reset routine to > use usb_trylock_device(). I don't like this because it will fail if the > device happens to be locked for some innocuous reason, like somebody > reading its entry in /proc/bus/usb/devices. > > Instead, usb_lock_device_for_reset() calls usb_trylock_device() > repeatedly, in a polling loop. It is careful to check that the device > state isn't NOTATTACHED and that the driver's interface isn't being > unbound; this will allow the lock-and-reset sequence to fail in scenarios > like the one above. > > For the new routine to be able to tell when an interface is being unbound, > I had to add an additional field to struct usb_interface. It's a simple > enumeration with four possible values: > > USB_INTERFACE_UNBOUND, USB_INTERFACE_BINDING, > USB_INTERFACE_BOUND, USB_INTERFACE_UNBINDING > > The extra overhead of the new field is not large enough to matter. > > A quick summary: > > Rename __usb_reset_device to usb_reset_device and get rid of > the alternate entry point. > > Notify the HCD that endpoint-0's maxpacket size may change > during a device reset (should have been written earlier). > > Add the new usb_interface_condition enumeration field to > struct usb_interface. > > Set the new field appropriately as interfaces are bound, > unbound, claimed, and released. > > Add usb_lock_device_for_reset(). > > Change the usb-storage, microtek, and stir4200 IRDA drivers > to make them use the new locking routine. > > With this patch applied, and using an altered gadget driver that forces > usb-storage to request a device reset, I've been able to go through a > large number of different tests successfully. Reset during probe(), > reset after probe(), disconnect during reset(), reset with device > descriptor changing... They all work beautifully. > > Please apply.
Applied, thanks. greg k-h ------------------------------------------------------- This SF.Net email sponsored by Black Hat Briefings & Training. Attend Black Hat Briefings & Training, Las Vegas July 24-29 - digital self defense, top technical experts, no vendor pitches, unmatched networking opportunities. Visit www.blackhat.com _______________________________________________ [EMAIL PROTECTED] To unsubscribe, use the last form field at: https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
