Re: [RFC] Input: Remove unsafe device module references

[Greg KH <gregkh@suse.de>]

On Wed, Nov 02, 2011 at 02:45:58PM +0100, David Herrmann wrote:
> On Tue, Nov 1, 2011 at 7:18 PM, Dmitry Torokhov
> <dmitry.torokhov@gmail.com> wrote:
> > On Tue, Nov 01, 2011 at 07:09:27PM +0100, David Herrmann wrote:
> >> On Tue, Nov 1, 2011 at 7:00 PM, Greg KH <gregkh@suse.de> wrote:
> >> > On Tue, Nov 01, 2011 at 06:52:11PM +0100, David Herrmann wrote:
> >> >> My solution: Some parent subsystem of us must take and release this
> >> >> module-refcnt instead of us, so this bug doesn't occur.
> >> >
> >> > Yes, that is the ultimate solution for something like this.
> >> >
> >> > But, in reality, we don't care about module unloading races as there are
> >> > plenty of other issues involved there where things can go bad, so we
> >> > just try the best we can :)
> >>
> >> Ah, I am kind of relieved that I got this right. I almost started
> >> thinking I am insane.. ;)
> >>
> >> So your answer is that this is so unlikely that it won't be fixed? I
> >> am fine with that, even though I wonder why stuff like "struct
> >> file_operations" include "owner" fields to protect callbacks but
> >> "struct device_type" does *not* include any protection of it's
> >> "release" callback.
> >
> > I think adding owner to device_type might not be a bad idea at all...
> 
> Exactly. But Greg does not seem to be very amused by that idea :-/

Actually that might work, but again, is it worth it?

Patches, as always, are gladly accepted, if you think this would resolve
the issue.

thanks,

greg k-h