On Mon, Apr 29, 2019 at 11:50:58AM +0200, Oliver Neukum wrote: > On Do, 2019-04-25 at 18:05 +0200, Johan Hovold wrote: > > @@ -484,6 +503,12 @@ void usb_serial_generic_unthrottle(struct tty_struct > > *tty) > > port->throttled = port->throttle_req = 0; > > spin_unlock_irq(&port->lock); > > > > + /* > > + * Matches the smp_mb__after_atomic() in > > + * usb_serial_generic_read_bulk_callback(). > > + */ > > + smp_mb(); > > + > > if (was_throttled) > > usb_serial_generic_submit_read_urbs(port, GFP_KERNEL); > > > Doesn't the spin_unlock_irq() imply smp_mb()? > Otherwise it looks correct to me.
No, spin_unlock_irq() is only a one-way barrier, and doesn't prevent later accesses from "moving" into the locked section. Johan
