Re: [linux-usb-devel] Fw: X is killed when trying to suspend with USB Mouse plugged in

[Dmitry Torokhov <dtor_core@ameritech.net>]

On Tuesday 19 October 2004 10:35 am, Alan Stern wrote:
> On Mon, 18 Oct 2004, Dmitry Torokhov wrote:
> 
> > > I don't know about /dev/input/mouse1.  But the oops isn't a bug... it's a 
> > > weakness in the way Linux implements loadable kernel modules.
> > > 
> > 
> > Ugh, it is not module implementation weakness, it looks like refcounting
> > problem in USB.
> 
> Could you explain that more fully?  Are you talking about a particular 
> refcounting problem in the usbhid subsystem or do you mean a more 
> pervasive problem in the whole USB system?  And why do you say it's a 
> refcounting problem in the first place?
> 

I am not sure it it is HID-specific problem or a wider one but it looks
like usbhid can be unloaded while there are references to objects produced
by this module - hence refcounting problem. You either have to disallow
unloading while there are references (but this path leads to potential
deadlocks) or have a generic release function registered with the core that
pretty much always stays there. Then you can free all device-specific data
at unload time and mark the object as a zombie so anything that tries to
touch it releases it quickly and then the core routine will free skeleton
data at last.

The patch that I sent should hide the problem somewhat as at disconnect
time it will unregister corresponsing class devices thus dropping the
reference that was pinning usbhid structures.

-- 
Dmitry