Runtime driver selection for a certain device
Grant Likely
grant.likely at secretlab.ca
Wed Apr 20 03:25:35 EST 2011
On Tue, Apr 19, 2011 at 06:14:12PM +0200, Wolfgang Grandegger wrote:
> On 04/18/2011 09:22 PM, Grant Likely wrote:
> > On Mon, Apr 18, 2011 at 1:22 PM, Wolfgang Grandegger <wg at grandegger.com> wrote:
> >> Hallo,
> >>
> >> I'm looking for a solution for the following problem:
> >>
> >> A device may be served by more than one driver depending on the
> >> application. Therefore the driver to use should be selectable at
> >> run-time. Lets imaging 4 serial line devices normally handled by the
> >> Linux serial driver. One of those needs eventually be handled by another
> >> driver. How could that be realized, especially together with the device
> >> tree, which is pretty static. Is there a generic way? The bind/unbind
> >> functions look promising but it's not possible to register more than one
> >> driver for a device.
> >
> > Depends on the bus type. I would look at registering the device in
> > such a way that no driver will bind against it, and then change the
> > configuration from userspace so the one of the drivers will get bound.
>
> OK, I got a bit further. I think the following will work:
>
> # echo f0002c00.serial > \
> /sys/devices/f0000000.soc5200/f0002c00.serial/driver/unbind
>
> will unbind the driver for PSC6 and when I then load another serial
> driver module, it will grap the unbound devices, e.g. PSC6 in this
> example. But how can I avoid that the driver probes all relevant devices
> when it's registered, or even better, how can I avoid automatic driver
> binding for certain devices. I think that's not forseen, at least not
> for the platform bus... but maybe I have missed something.
You're venturing into new territory. You'll need to come up with a
mechanism for doing so and propose it. It would be nice to have an
explicit method for specifying the driver that should be bound to a
driver. The following /might/ work:
echo f0002c00.serial > /sys/bus/platform/drivers/<your-driver>/bind
g.
More information about the devicetree-discuss
mailing list