> Date: Mon, 3 Dec 2001 13:33:53 +0100 (CET) > From: Martin Diehl <[EMAIL PROTECTED]>
> > > - the ioctl blocks (no, not sleep_on) on the completion (i.e. until > > > urb->status becomes != EINPROGRESS). > > > > BTW, This is broken. The ioctl must block until a callback > > is triggered. It is safe to examing urb->status only after > > the callback awakens the requesting thread or inside the callback. > Sure, urb->status is modified from interrupt context. But the read for > comparison should be atomic (although not explicitly). >[...] > The only thing I'm wondering is whether urb->status would need to be > qualified volatile to prevent the compiler from bad assumptions. You mentally focused in wrong place, it's not about atomicity at all. Suppose that you have a compiled with low optimizations and SMP. The thread that checks for the status may get a signal and be awaken before callback is called, but after status was set. In that case something oopses: either the thread, or the HCA interrupt. It is very easy to trigger with ISOs by hittinc ^C until the box oopses, not so easy with normal transfers. But still, better be safe. -- Pete _______________________________________________ [EMAIL PROTECTED] To unsubscribe, use the last form field at: https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
