Re: [PATCH 4/7] phy: rockchip-usb: add missing of_node_put

["Heiko Stübner" <heiko@sntech.de>]

Hi Julia,

Am Montag, 16. November 2015, 12:33:17 schrieb Julia Lawall:
> diff --git a/drivers/phy/phy-rockchip-usb.c b/drivers/phy/phy-rockchip-usb.c
> index 91d6f34..62c43c4 100644
> --- a/drivers/phy/phy-rockchip-usb.c
> +++ b/drivers/phy/phy-rockchip-usb.c
> @@ -108,13 +108,16 @@ static int rockchip_usb_phy_probe(struct
> platform_device *pdev)
> 
>  	for_each_available_child_of_node(dev->of_node, child) {
>  		rk_phy = devm_kzalloc(dev, sizeof(*rk_phy), GFP_KERNEL);
> -		if (!rk_phy)
> -			return -ENOMEM;
> +		if (!rk_phy) {
> +			err = -ENOMEM;
> +			goto put_child;
> +		}
> 
>  		if (of_property_read_u32(child, "reg", &reg_offset)) {
>  			dev_err(dev, "missing reg property in node %s\n",
>  				child->name);
> -			return -EINVAL;
> +			err = -EINVAL;
> +			goto put_child;
>  		}
> 
>  		rk_phy->reg_offset = reg_offset;
> @@ -127,18 +130,22 @@ static int rockchip_usb_phy_probe(struct
> platform_device *pdev) rk_phy->phy = devm_phy_create(dev, child, &ops);
>  		if (IS_ERR(rk_phy->phy)) {
>  			dev_err(dev, "failed to create PHY\n");
> -			return PTR_ERR(rk_phy->phy);
> +			err = PTR_ERR(rk_phy->phy);
> +			goto put_child;
>  		}
>  		phy_set_drvdata(rk_phy->phy, rk_phy);
> 
>  		/* only power up usb phy when it use, so disable it when init*/
>  		err = rockchip_usb_phy_power(rk_phy, 1);
>  		if (err)
> -			return err;
> +			goto put_child;
>  	}
> 
>  	phy_provider = devm_of_phy_provider_register(dev, of_phy_simple_xlate);
>  	return PTR_ERR_OR_ZERO(phy_provider);
> +put_child:
> +	of_node_put(child);
> +	return err;
>  }
> 
>  static const struct of_device_id rockchip_usb_phy_dt_ids[] = {

hmm, while I agree that the rockchip phy has an issue in the node lifecycle, 
I'm not sure that patch fixes it fully.

It currently iterates over each phy, but would only of_node_put the phy it 
handled last. So if an error happens on the 3rd phy, the first 2 are already 
instantiated and would also get removed when the overall probe fails, but 
their of_node would never be "put".

I think this goes together with what Brian described in patch1 having the phy-
core also handle the node-reference. When this is included we could just 
always put the of_node when finishing its loop iteration, as the phy-core will 
have its own reference on the node.


Heiko