On Fri, 16 Jan 2004, Oliver Neukum wrote: > > > This lock should not see much contention. It will be held only very > > briefly in any path where a state gets changed. Something like this: > > > > spin_lock_irq(&global_state_lock); > > if (usbdev->state != USB_STATE_NOTATTACHED) > > usbdev->state = USB_STATE_CONFIGURED; > > else > > notattached = 1; > > spin_unlock_irq(&global_state_lock); > > if (notattached) ... > > Ehm. You have to check the state in usb_submit_urb().
True, but that doesn't require grabbing the spinlock since all you want to know is whether the state is NOTATTACHED or not. You're not changing the state. > > > > struct usb_device->serialize will become a read-write semaphore. > > The only reason to go for an rw-semaphore is contention. If contention is > low, there's no point in going for the extra overhead of such a lock. No, there can be other reasons too. For example, suppose one routine that gets a read lock calls another that does the same thing. That works well with rw-semaphores but not with regular ones. That's not contention, it's something else (poor design perhaps -- but Duncan has his reasons). [talking about blocking URB submissions and unlinking URBs in flight...] > > Ah -- we do it when the device is unplugged. So far as I can see, we > > don't do it when usb_deregister() is called. Nor do we set the interfaces > > Nor must we ever do this, unless it's in the interval between unbinding > and reprobe. But from a viewpoint of correctness a blockages that exists > only for an arbitrarily low amount of time doesn't exist at all. This wouldn't be for an arbitrarily short period of time; it would last during the entire call to disconnect(). And yes, certainly it would stop before re-probe(). Alan Stern ------------------------------------------------------- The SF.Net email is sponsored by EclipseCon 2004 Premiere Conference on Open Tools Development and Integration See the breadth of Eclipse activity. February 3-5 in Anaheim, CA. http://www.eclipsecon.org/osdn _______________________________________________ [EMAIL PROTECTED] To unsubscribe, use the last form field at: https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
