2009/4/6 Alan Stern <stern@xxxxxxxxxxxxxxxxxxx>: > On Mon, 6 Apr 2009, Ming Lei wrote: > >> 2009/4/5 Alan Stern <stern@xxxxxxxxxxxxxxxxxxx>: >> > >> > BTW, I'm concerned about that usb-serial.c patch posted earlier. In >> >> what is the purpose of the patch? which fixes? > > The purpose of the patch is to fix the bug reported here: > > http://bugs.freedesktop.org/show_bug.cgi?id=20703 > > The bug is that the usb_serial and usb_serial_port structures are not > reference-counted correctly. > >> > particular, does serial_open() hold disc_mutex long enough? What >> >> IMHO, it is not a kind of open/disconnect race, serial_port_open() is more >> proper than serial_open(). > > What are you talking about? There is no serial_port_open() routine in > drivers/usb/serial, nor anywhere else I know of. I mean that maybe the name of serial_port_open is more proper , after all it is used to open a usb serial port, not a usb serial, but the disconnect event is received by usb_serial, instead of usb_serial_port. > >> Once usb_serial_get_by_index() returns a non NULL >> serial, the port device doesn't disappear even if the usb serial >> device is disconnected. > > First of all, this doesn't make sense. usb_serial_get_by_index() > returns a pointer to the usb_serial structure, not to the port device. > So the usb_serial structure won't disappear, but there's nothing to > prevent the usb_serial_port structure from disappearing. usb_serial_port structure is freed in destroy_serial(), which is only called if the reference count of usb_serial is zero. > > Second, none of this matters since I'm not concerned about whether the > port device disappears. Rather, I'm concerned about whether > serial_open() can call usb_autopm_get_interface() after > usb_serial_disconnect() has returned. That should never happen. If usb_serial_disconnect() has returned, usb_autopm_get_interface() may return a failure, and serial_open can process the case. > >> So, maybe holding disc_mutex in serial_open is not necessary. >> >> > happens if a disconnect occurs while serial_open() is waiting to >> > acquire port->mutex? Do you think the calls to >> >> mutex_lock_interruptible() should return a failed value. > > Why shoult it? The only reason mutex_lock_interruptible() ever returns > a failure is if the task gets interrupted (for example, the user > presses ^C). I guess it is the tty_hangup( called by usb_serial_disconnect),which can cause mutex_lock_interruptible() to return a failure. -- Lei Ming -- 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
