Re: [PATCH] net: phy: add driver for MaxLinear MxL86110 PHY

[Stefano Radaelli <stefano.radaelli21@gmail.com>]

Hi Andrew,

Thanks for the review and for pointing this out.

> Why are these two special and use the locked variant, when all the
> others don't?
> Please think about locking, when can unlocked versions be used? When
> should they not be used?
> Are you testing this with CONFIG_PROVE_LOCKING enabled?

You're right, the use of the _locked variants was confusing and inconsistent.
After enabling CONFIG_PROVE_LOCKING and testing, I identified where
locking is necessary.
Functions like get_wol, set_wol, config_init, and LED controls can be called
concurrently from userspace or asynchronous contexts, so they require locking
the MDIO bus. Without proper locking, I got warnings like:
WARNING: ... at drivers/net/phy/mdio_bus.c:920 __mdiobus_write
WARNING: ... at drivers/net/phy/mdio_bus.c:891 __mdiobus_read

To clean up, I removed the locked helpers and instead wrapped extended
register accesses with phy_lock_mdio_bus() where needed.
Now, no warnings appear with CONFIG_PROVE_LOCKING enabled
after testing every implemented functionality.

> Please also take a read of:
> https://www.kernel.org/doc/html/latest/process/maintainer-netdev.html
> https://docs.kernel.org/process/submitting-patches.html

Thanks, I have reread those documents, updated the code accordingly,
and plan to send my patch today with the [PATCH net-next] tag.

Best Regards,

Stefano

Il giorno gio 15 mag 2025 alle ore 21:03 Andrew Lunn <andrew@lunn.ch>
ha scritto:
>
>         > +static void mxl86110_get_wol(struct phy_device *phydev, struct ethtool_wolinfo *wol)
> > +{
> > +     int value;
> > +
> > +     wol->supported = WAKE_MAGIC;
> > +     wol->wolopts = 0;
> > +     value = mxl86110_locked_read_extended_reg(phydev, MXL86110_EXT_WOL_CFG_REG);
>
>
> > +static int mxl86110_led_hw_control_set(struct phy_device *phydev, u8 index,
> > +                                    unsigned long rules)
> ..
>
> > +     ret = mxl86110_locked_write_extended_reg(phydev, MXL86110_LED0_CFG_REG + index, val);
>
> Why are these two special and use the _locked_ variant, when all the
> others don't?
>
> Please think about locking, when can unlocked versions be used? When
> should they not be used?
>
> Are you testing this with CONFIG_PROVE_LOCKING enabled?
>
> Please also take a read of:
>
> https://www.kernel.org/doc/html/latest/process/maintainer-netdev.html
> https://docs.kernel.org/process/submitting-patches.html
>
>     Andrew
>
> ---
> pw-bot: cr
>
>