On Fri, Jul 06, 2012 at 11:32:51AM +0100, Martyn Welch wrote:
> On 06/07/12 09:15, Dan Carpenter wrote:
> > We don't use the "bus" array or the "bus_num" variable which store the
> > number of elements in the array. The only user was removed in
> > 5d6abf379d ('staging: vme: make match() driver specific to improve
> > non-VME64x support').
> >
>
> Hmm, without a parameter to specify the bus (i.e. a specific vme bridge) to
> which vme_user is to bind, vme_user will bind to all bridges it finds, hogging
> resources on them all.
>
> Ok, so the current implementation of vme_user can only bind to one bus, so
> it's worse than that, it will only every bind to the first bridge when more
> than one are available.
>
> It appears that the real problem here is that vme_user_match is wrong.
>
> Rather than:
>
> static int vme_user_match(struct vme_dev *vdev)
> {
> if (vdev->num >= VME_USER_BUS_MAX)
> return 0;
> return 1;
> }
>
> It should be something like:
>
> static int vme_user_match(struct vme_dev *vdev)
> {
> int i;
>
> for (i = 0; i < VME_USER_BUS_MAX; i++)
> if (vdev->num == bus[i])
> return 1;
> return 0;
> }
>
So the default behavior way should be to bind to the first bus or to
all?
Obviously, we should do your fix. The problem is that I don't want
to be the one to sign off on something that changes how the code
works when I can't test it. Can you send for a patch for this?
regards,
dan carpenter
_______________________________________________
devel mailing list
devel@xxxxxxxxxxxxxxxxxxxxxx
http://driverdev.linuxdriverproject.org/mailman/listinfo/devel
