On Fri, 31 Jul 2009, Alan Stern wrote: > > > The BKL gets used in several places within usbcore. As far as I can > > > tell, all but one of them are unnecessary -- the exception being > > > usb_device_poll() in devices.c. And that should be replaced with a > > > private mutex, or perhaps we could use usbfs_mutex. > > With respect to poll I agree. > > With respect to open() I dimly remember that hiddev needs BKL. Maybe > > jkosina remembers. > Jiri, do you know the answer? We're talking about usb_open() in > drivers/usb/core/file.c. I don't see why that should need the BKL. [ sorry for late reply, I have been offline for two weeks ] Actually, I remember having discussed this with some of you guys some time ago, and I thought that the conclusion was that a) disconnect() used to do lock_kernel() b) nobody has actually done the hard work of checking all open() methods for dependency on BKL In hiddev we need BKL simply because we need a lock which usb_open() also takes in the right order, so that we avoid race between usb_register_dev() and open(). We can't use simple mutex to guard hiddev_table[], because usb_open() and usb_register_dev() lock minor_rwsem, and we would end up with ABBA deadlock. -- Jiri Kosina SUSE Labs -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
