On Mon, Dec 23, 2002 at 12:10:47PM +1100, Rusty Russell wrote:
> > This one is due to the way AGPGART does (or has done for the last 3
> > years) its module locking. It does a MOD_INC_USE_COUNT as soon as
> > someone calls the acquire routines.
> Which is racy under SMP, and under preempt, which is why it's
> deprecated.
Crapola. I've just realised why this is no longer relevant.
I've moved what this was protecting into the per chipset modules.
Right now its possible to load modules, start x (which loads DRM),
then rmmod via_agp from under its feet. result - boom when something
tries to use 3d.
Sure it's unlikely someone would be crazy enough to try and do this,
and they deserve what they get, but it's not exactly clean, or nice.
So where is the documentation describing module locking de jour ?
> > (So you can't unload agpgart whilst you've a 3d using app (like X)
> > open). This seems quite sensible, but these days you can't unload
> > agpgart.ko anyway because the chipset module (via-agp.ko in your
> > case) already has it 'in use', so I'm tempted to drop those bits.
> If this is true (it usually is), you can simply drop them.
I'll need 'something' in the chipset drivers. The first thing that
jumps to mind is to give the chipset drivers an 'acquire' op
which does the locking much like the old agp_backend_acquire() does.
> There are other cases where the caller is not grabbing references, so
> MOD_INC_USE_COUNT is better than nothing (should the warning stay for
> 2.6? Good question).
Why exactly isn't it safe any more? If there's documentation on this,
I'd love to read it. If there isn't, there really should be.
Dave
-- | Dave Jones. http://www.codemonkey.org.uk | SuSE Labs - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Mon Dec 23 2002 - 22:00:32 EST