RE: [PATCH net-next] net: emaclite: Add error handling for of_address_ and phy read functions

[Radhey Shyam Pandey]

> -----Original Message-----
> From: Jakub Kicinski <kuba@xxxxxxxxxx>
> Sent: Sunday, November 8, 2020 1:05 AM
> To: Radhey Shyam Pandey <radheys@xxxxxxxxxx>
> Cc: davem@xxxxxxxxxxxxx; Michal Simek <michals@xxxxxxxxxx>;
> netdev@xxxxxxxxxxxxxxx; linux-arm-kernel@xxxxxxxxxxxxxxxxxxx; linux-
> kernel@xxxxxxxxxxxxxxx; git <git@xxxxxxxxxx>; Shravya Kumbham
> <shravyak@xxxxxxxxxx>
> Subject: Re: [PATCH net-next] net: emaclite: Add error handling for
> of_address_ and phy read functions
> 
> On Tue, 3 Nov 2020 19:01:05 +0530 Radhey Shyam Pandey wrote:
> > From: Shravya Kumbham <shravya.kumbham@xxxxxxxxxx>
> >
> > Add ret variable, conditions to check the return value and it's error
> > path for of_address_to_resource() and phy_read() functions.
> >
> > Addresses-Coverity: Event check_return value.
> > Signed-off-by: Shravya Kumbham <shravya.kumbham@xxxxxxxxxx>
> > Signed-off-by: Radhey Shyam Pandey <radhey.shyam.pandey@xxxxxxxxxx>
> 
> Any reason not to apply this to net as a fix?
Yes, it can be applied to net as a fix. 
> 
> > diff --git a/drivers/net/ethernet/xilinx/xilinx_emaclite.c
> > b/drivers/net/ethernet/xilinx/xilinx_emaclite.c
> > index 0c26f5b..fc5ccd1 100644
> > --- a/drivers/net/ethernet/xilinx/xilinx_emaclite.c
> > +++ b/drivers/net/ethernet/xilinx/xilinx_emaclite.c
> > @@ -820,7 +820,7 @@ static int xemaclite_mdio_write(struct mii_bus
> > *bus, int phy_id, int reg,  static int xemaclite_mdio_setup(struct
> > net_local *lp, struct device *dev)  {
> >  	struct mii_bus *bus;
> > -	int rc;
> > +	int rc, ret;
> >  	struct resource res;
> >  	struct device_node *np = of_get_parent(lp->phy_node);
> >  	struct device_node *npp;
> > @@ -834,7 +834,13 @@ static int xemaclite_mdio_setup(struct net_local
> *lp, struct device *dev)
> >  	}
> >  	npp = of_get_parent(np);
> >
> > -	of_address_to_resource(npp, 0, &res);
> > +	ret = of_address_to_resource(npp, 0, &res);
> > +	if (ret) {
> > +		dev_err(dev, "%s resource error!\n",
> > +			dev->of_node->full_name);
> > +		of_node_put(lp->phy_node);
> 
> I'm always confused by the of_* refcounting. Why do you need to put
> phy_node here, and nowhere else in this function?

Initially, we added of_node_put(phy_node) thinking about this 
particular coverity change. But agree it has to be added for
all error path i.e better place would be in xemaclite_of_probe()
error label.

> 
> > +		return ret;
> > +	}
> 
> >  		/* Restart auto negotiation */
> >  		bmcr = phy_read(lp->phy_dev, MII_BMCR);
> > +		if (bmcr < 0) {
> > +			dev_err(&lp->ndev->dev, "phy_read failed\n");
> > +			phy_disconnect(lp->phy_dev);
> > +			lp->phy_dev = NULL;
> > +
> > +			return bmcr;
> > +		}
> >  		bmcr |= (BMCR_ANENABLE | BMCR_ANRESTART);
> >  		phy_write(lp->phy_dev, MII_BMCR, bmcr);
> 
> Does it really make much sense to validate the return value of
> phy_read() but not check any errors from phy_write()s?
Error handling was added for phy_read as it was using return value
and reported by coverity. But yes we in a follow-up patch we
can extend error handling for phy_write as well.