On Fri, May 14, 2021 at 06:49:58PM +0100, Mark Brown wrote: > On Fri, May 14, 2021 at 12:19:38PM -0500, Chris Morgan wrote: > > > +static int rk817_codec_parse_dt_property(struct device *dev, > > + struct rk817_codec_priv *rk817) > > +{ > > + struct device_node *node = dev->parent->of_node; > > + > > + if (!node) { > > + dev_err(dev, "%s() dev->parent->of_node is NULL\n", > > + __func__); > > + return -ENODEV; > > + } > > There's no need to fail the probe here, you won't be able to read any DT > properties but that shouldn't stop the driver binding. If I'm not mistaken this is actually telling us to fail if the parent device (the PMIC itself) isn't present. I think I'll remove this as not necessary since if the parent node isn't present the mfd driver will never load, meaning this driver will never load either. Below this line however we're failing if the codec node isn't present. Are you telling me we want to bind the driver if the node isn't present (such as in the edge case where the driver is present and the PMIC is a rk817, but the CODEC is not in use)? I will remove the return code if you think that is what needs to be done. My concern there though is if we do that we'll either be in a position to again report a bunch of errors for the edge case of users who want to use the PMIC but not the codec (in this case missing clocks and whatnot) if we try to bind the driver and the user doesn't want it. I can also set those errors to debug level, but I think that they might be important enough for users who DO want to use the codec to keep them as dev_err. Let me know what you think. Thank you.