Re: [PROPOSED PATCH] ATM refcount + firestream

[Philipp Rumpf <prumpf@parcelfarce.linux.theplanet.co.uk>]

On Sat, Oct 28, 2000 at 11:50:22AM -0400, Brian Gerst wrote:
> Philipp Rumpf wrote:
> > 
> > On Sat, Oct 28, 2000 at 09:55:21AM -0400, Brian Gerst wrote:
> > > Yes, but they can be called (and sleep) with module refcount == 0.  This
> > > is because the file descripter used to perform the ioctl isn't directly
> > > associated with the network device, thereby not incrementing the
> > > refcount on open.
> > 
> > According to my proposal, it is perfectly safe to call a function in a module
> > while the module's use count is 0.  This function would typically look like this:
> > 
> > foo()
> > {
> >         MOD_INC_USE_COUNT;
> > 
> >         copy_*_user() (or anything else that sleeps);
> > 
> >         MOD_DEC_USE_COUNT;
> > 
> >         return bar;
> > }
> > 
> > The only difference to the "old" module scheme is that the above currently isn't
> > safe on SMP systems.
> 
> This will only work while the kernel is not preemptable.  Once the
> kernel thread can be rescheduled, all bets are off.

The implementation will need adjusting for preemptable kernel threads.  The
concept still works fine.	

> With or without your patch, the network ioctls are unsafe, since they
> don't currently do refcounting at all.

I was under the impression most network ioctls didn't sleep.

> Adding it in the layer above thTe
> driver is the easier and cleaner solution.

I disagree.  I dislike special-casing inter-module calls (and that's not even
taking into account that the current implementation of an inter-module call is
quite ugly).
-
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/