Re: [RFC PATCH net-next 1/3] net: phy: add concept of shared storage for PHYs

[Michael Walle]

Am 2020-04-21 21:30, schrieb Andrew Lunn:
> > Speaking of it. Does anyone have an idea how I could create the hwmon
> > device without the PHY device? At the moment it is attached to the
> > first PHY device and is removed when the PHY is removed, although
> > there might be still other PHYs in this package. Its unlikely to
> > happen though, but if someone has a good idea how to handle that,
> > I'd give it a try.
>
> There is a somewhat similar problem with Marvell Ethernet switches and
> their internal PHYs. The PHYs are the same as the discrete PHYs, and
> the usual Marvell PHY driver is used. But there is only one
> temperature sensor for the whole switch, and it is mapped into all the
> PHYs. So we end up creating multiple hwmon devices for the one
> temperature sensor, one per PHY.
>
> You could take the same approach here. Each PHY exposes a hwmon
> device?
>
> Looking at
> static struct device *
> __hwmon_device_register(struct device *dev, const char *name, void 
> *drvdata,
>                         const struct hwmon_chip_info *chip,
>                         const struct attribute_group **groups)
>
> I think it is O.K. to pass dev as NULL. You don't have to associate it
> to a device. So you could create the hwmon device as part of package
> initialisation and put it into shared->priv.

I actually tried that before writing my mail. Have a look at commit
59df4f4e8e0b ("hwmon: (core) check parent dev != NULL when chip != 
NULL")

and the corresponding discussion here:
  https://patchwork.kernel.org/patch/10381759/

And if I'd had to choose, I'd prefer having one hwmon device on the
first PHY (with its drawback) rather than having it four times.

-michael