[PATCH 2/4] staging: vme: make match() driver specific to improve non-VME64x support

[Manohar Vanga]

> > diff --git a/drivers/staging/vme/vme.c b/drivers/staging/vme/vme.c
> > index 76e08f3..9cb6938 100644
> > --- a/drivers/staging/vme/vme.c
> > +++ b/drivers/staging/vme/vme.c
> > @@ -1317,6 +1317,7 @@ static int vme_add_bus(struct vme_bridge *bridge)
> >  		if ((vme_bus_numbers & (1 << i)) == 0) {
> >  			vme_bus_numbers |= (1 << i);
> >  			bridge->num = i;
> > +			INIT_LIST_HEAD(&bridge->devices);
> >  			list_add_tail(&bridge->bus_list, &vme_bus_list);
> 
> Just realised that vme_buses_lock also protects vme_bus_list; appending
> to this list here should be protected by the mutex, otherwise
> vme_add_bus (ie this function) could race with __vme_register_driver,
> which might access a corrupt copy of the list.
> 
> Note that vme_remove_bus does the right thing though; it acquires
> the lock before unpinning the bridge from vme_bus_list.

It _does_ acquire the lock. The patch here doesn't show it:

	mutex_lock(&vme_buses_lock);
	for (i = 0; i < sizeof(vme_bus_numbers) * 8; i++) {
		if ((vme_bus_numbers & (1 << i)) == 0) {
			vme_bus_numbers |= (1 << i);
			bridge->num = i;
			INIT_LIST_HEAD(&bridge->devices);
			list_add_tail(&bridge->bus_list, &vme_bus_list);
			ret = 0;
			break;
		}
	}
	mutex_unlock(&vme_buses_lock);

This was submitted and acknowledged by Martyn in a previous patch:
	https://lkml.org/lkml/2011/8/12/107

Thanks!

-- 
/manohar