Re: Locking Between User Context and Soft IRQs in 2.4.0

[Andi Kleen <ak@suse.de>]

On Sun, Nov 05, 2000 at 12:28:55PM +1100, Andrew Morton wrote:
> Andi Kleen wrote:
> > 
> > On Sat, Nov 04, 2000 at 12:07:34PM -0500, Jeff Garzik wrote:
> > > Andi Kleen wrote:
> > > > All the MOD_INC/DEC_USE_COUNT are done inside the modules themselves. There
> > > > is nothing that would a driver prevent from being unloaded on a different
> > > > CPU while it is already executing in ->open but has not yet executed the add
> > > > yet or after it has executed the _DEC but it is still running in module code
> > > > Normally the windows are pretty small, but very long running interrupt
> > > > on one CPU hitting exactly in the wrong moment can change that.
> > >
> > > Module unload calls unregister_netdev, which grabs rtnl_lock.
> > > dev->open runs under rtnl_lock.
> > >
> > > Given this, how can the driver be unloaded if dev->open is running?
> > 
> > It does not help, because when the semaphore synchronizes it is already
> > too late -- free_module already did the zero module count check and
> > nothing is going to stop it from unloading.
> 
> aaarrrggh!!!

A simple fix at least for this case would be a recheck of module count
after free_module() executed the cleanup function.

-Andi
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/