Re: [PATCH net-next] mdio: mdiobus: setup of_node for the MDIO device

["Russell King (Oracle)" <linux@armlinux.org.uk>]

On Tue, Jun 15, 2021 at 08:24:44PM +0300, Ioana Ciornei wrote:
> On Tue, Jun 15, 2021 at 06:13:31PM +0100, Russell King (Oracle) wrote:
> > On Tue, Jun 15, 2021 at 06:44:01PM +0300, Ioana Ciornei wrote:
> > > From: Ioana Ciornei <ioana.ciornei@nxp.com>
> > > 
> > > By mistake, the of_node of the MDIO device was not setup in the patch
> > > linked below. As a consequence, any PHY driver that depends on the
> > > of_node in its probe callback was not be able to successfully finish its
> > > probe on a PHY, thus the Generic PHY driver was used instead.
> > > 
> > > Fix this by actually setting up the of_node.
> > > 
> > > Fixes: bc1bee3b87ee ("net: mdiobus: Introduce fwnode_mdiobus_register_phy()")
> > > Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
> > > ---
> > >  drivers/net/mdio/fwnode_mdio.c | 1 +
> > >  1 file changed, 1 insertion(+)
> > > 
> > > diff --git a/drivers/net/mdio/fwnode_mdio.c b/drivers/net/mdio/fwnode_mdio.c
> > > index e96766da8de4..283ddb1185bd 100644
> > > --- a/drivers/net/mdio/fwnode_mdio.c
> > > +++ b/drivers/net/mdio/fwnode_mdio.c
> > > @@ -65,6 +65,7 @@ int fwnode_mdiobus_phy_device_register(struct mii_bus *mdio,
> > >  	 * can be looked up later
> > >  	 */
> > >  	fwnode_handle_get(child);
> > > +	phy->mdio.dev.of_node = to_of_node(child);
> > >  	phy->mdio.dev.fwnode = child;
> > 
> > Yes, this is something that was missed, but let's first look at what
> > other places to when setting up a device:
> > 
> >         pdev->dev.fwnode = pdevinfo->fwnode;
> >         pdev->dev.of_node = of_node_get(to_of_node(pdev->dev.fwnode));
> >         pdev->dev.of_node_reused = pdevinfo->of_node_reused;
> > 
> >         dev->dev.of_node = of_node_get(np);
> >         dev->dev.fwnode = &np->fwnode;
> > 
> >         dev->dev.of_node = of_node_get(node);
> >         dev->dev.fwnode = &node->fwnode;
> > 
> > That seems to be pretty clear that an of_node_get() is also needed.
> > 
> 
> I'm not convinced that an of_node_get() is needed besides the
> fwnode_handle_get() call that's already there.
> 
> The fwnode_handle_get() will call the get callback for that particular
> fwnode_handle. When we are in the OF case, the of_fwnode_get() will be
> invoked which in turn does of_node_get().
> 
> Am I missing something here?

Hmm, I think you're actually correct - the other places increase the
OF node's refcount and then assign the fwnode, which is effectively
what will be happening here (since fwnode_handle_get() will do that
for us.)

However, there's definitely horrid stuff going on in this file with
refcounting:

fwnode_mdiobus_register_phy():

			phy_device_free(phy);
			fwnode_handle_put(phy->mdio.dev.fwnode);

phy_device_free() drops the refcount on the embedded struct device - it
_could_ free it, so we should be assuming that "phy" is dead at that
point - we should not be dereferencing it after the call.

fwnode_mdiobus_phy_device_register():

	fwnode_handle_get(child);
	phy->mdio.dev.fwnode = child;

	rc = phy_device_register(phy);
	if (rc) {
		fwnode_handle_put(child);
		return rc;

Here, we leave this function having dropped the fwnode refcount, but
we have left a dangling pointer to the fwnode in place. Hopefully,
no one will use that dangling pointer, but this is sloppy.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!