Re: [PATCH] drivers/input/joystick/xpad.c: Add rumble support for original xbox controller

[Dmitry Torokhov <dmitry.torokhov@gmail.com>]

On Sun, Jan 10, 2010 at 07:11:12PM +0100, Benjamin Valentin wrote:
> On Sat, 9 Jan 2010 23:56:16 -0800
> Dmitry Torokhov <dmitry.torokhov@gmail.com> wrote:
> 
> > > --- /usr/src/linux-source-2.6.33/drivers/input/joystick/xpad.c
> > > 2010-01-08 02:56:59.365851076 +0100 +++ xpad.c	2010-01-08
> > > 03:13:38.477835651 +0100 @@ -505,7 +505,7 @@
> > >  	struct usb_endpoint_descriptor *ep_irq_out;
> > >  	int error = -ENOMEM;
> > >  
> > > -	if (xpad->xtype != XTYPE_XBOX360)
> > > +	if (xpad->xtype != XTYPE_XBOX360 && xpad->xtype !=
> > > XTYPE_XBOX) return 0;
> > >  
> > >  	xpad->odata = usb_buffer_alloc(xpad->udev, XPAD_PKT_LEN,
> > > @@ -535,13 +535,13 @@
> > >  
> > >  static void xpad_stop_output(struct usb_xpad *xpad)
> > >  {
> > > -	if (xpad->xtype == XTYPE_XBOX360)
> > > +	if (xpad->xtype == XTYPE_XBOX360 || xpad->xtype !=
> > > XTYPE_XBOX)
> > 
> > This should cretainly be "... || xpad->xtype == XTYPE_XBOX)", I'll fix
> > it up locally.
> 
> Thank you, this made me discover another bug that eventually leads to a
> kernel oops when the device is unplugged while an effect is playing
> (or the effect is somehow else interrupted).
> This should be fixed by taking the mutex when modifying xpad->odata as
> well as checking whether it has been freed before writing to it.
> 

No, you may not take mutex in xpad_play_effect because it is called with
interrupts off and dev->event_lock spinlock held. We probably need a
workqueue-based solution to queue play requests and ensure that 2
different requests do not "step" on each other.

-- 
Dmitry