[PATCH net-next 07/15] net: dsa: mt7530: do not run mt7530_setup_port5() if port 5 is disabled

[Vladimir Oltean]

On Fri, Dec 08, 2023 at 07:23:38AM +0300, Dan Carpenter wrote:
> On Thu, Dec 07, 2023 at 08:40:15PM +0200, Vladimir Oltean wrote:
> > 
> > We could be more pragmatic about this whole sparse false positive warning,
> > and just move the "if" block which calls mt7530_setup_port5() right
> > after the priv->p5_intf_sel assignments, instead of waiting to "break;"
> > from the for_each_child_of_node() loop.
> > 
> > for_each_child_of_node(dn, mac_np) {
> > 	if (!of_device_is_compatible(mac_np,
> > 				     "mediatek,eth-mac"))
> > 		continue;
> > 
> > 	ret = of_property_read_u32(mac_np, "reg", &id);
> > 	if (ret < 0 || id != 1)
> > 		continue;
> > 
> > 	phy_node = of_parse_phandle(mac_np, "phy-handle", 0);
> > 	if (!phy_node)
> > 		continue;
> > 
> > 	if (phy_node->parent == priv->dev->of_node->parent) {
> > 		ret = of_get_phy_mode(mac_np, &interface);
> > 		if (ret && ret != -ENODEV) {
> > 			of_node_put(mac_np);
> > 			of_node_put(phy_node);
> > 			return ret;
> > 		}
> > 		id = of_mdio_parse_addr(ds->dev, phy_node);
> > 		if (id == 0)
> > 			priv->p5_intf_sel = P5_INTF_SEL_PHY_P0;
> > 		if (id == 4)
> > 			priv->p5_intf_sel = P5_INTF_SEL_PHY_P4;
> > 
> > 		if (priv->p5_intf_sel == P5_INTF_SEL_PHY_P0 || <---- here
> > 		    priv->p5_intf_sel == P5_INTF_SEL_PHY_P4)
> > 			mt7530_setup_port5(ds, interface);
> 
> This doesn't solve the problem that Smatch doesn't know what the
> original value of priv->p5_intf_sel.  And also I don't like this code
> because now we call mt7530_setup_port5() on every iteration after
> we find the first P5_INTF_SEL_PHY_P0.

You seem to have not parsed the "break" from 4 lines below. There is at
most one iteration through for_each_child_of_node().

And why would the "original" value of priv->p5_intf_sel matter? Original
or modified by the "if (id == 0)" and "if (id == 4)" blocks, the code
has already executed the of_get_phy_mode(&interface) call, by the time
we reach the "if" that calls mt7530_setup_port5().

Hmm, maybe the problem, all along, was that we let the -ENODEV return
code from of_get_phy_mode() pass through? "interface" will really be
uninitialized in that case. It's not a false positive.

Instead of:

	ret = of_get_phy_mode(mac_np, &interface);
	if (ret && ret != -ENODEV) {
		...
		return ret;
	}

it should have been like this, to not complain:

	ret = of_get_phy_mode(mac_np, &interface);
	if (ret) {
		...
		return ret;
	}

> > 	}
> > 	of_node_put(mac_np);
> > 	of_node_put(phy_node);
> > 	break;
> > }