Re: [PATCH v2 1/3] firmware loader: Introduce new API - request_firmware_abort()

[Greg Kroah-Hartman <gregkh-hQyY1W1yCW8ekmWlsbkhG0B+6BGkLq7r@public.gmane.org>]

On Tue, Nov 18, 2014 at 06:31:38AM +0000, Kweh, Hock Leong wrote:
> > -----Original Message-----
> > From: Matt Fleming [mailto:matt-HNK1S37rvNbeXh+fF434Mdi2O/JbrIOy@public.gmane.org]
> > Sent: Monday, November 17, 2014 11:12 PM
> > >
> > > - Only doing module unload is required to be aware of this synchronization
> > > 	-> Ensuring the call back does not fall into unloaded code which may
> > cause
> > > 	     undefined behavior.
> > > 	-> Ensuring the put_device() & module_put() code have finished in
> > firmware_class.c
> > > 	     function request_firmware_work_func() before the device is
> > unregistered
> > > 	     and module unloaded happen.
> > 
> > Shouldn't the existing module_{put,get}() and {put,get}_device() calls
> > provide all the necessary synchronisation?
> > 
> > Module unload should not be possible while other code is using the
> > module (and the module refcnt has been incremented accordindly).
> > 
> > Right?
> > 
> > --
> > Matt Fleming, Intel Open Source Technology Center
> 
> Hi Matt,
> 
> Yes, you are right. If the module refcount is not zero, you will get error
> message and returned while you do "rmmod". But I strongly believe if we
> have the capability in our code to take care of it by doing synchronization,
> we should take care of it in case people are doing "rmmod -f". Don't
> you think so?

If you do 'rmmod -f' you get to keep all of the broken pieces of your
kernel, no need to try to help out with crazy things like that.

greg k-h