Re: [PATCH v2 1/1] net: phy: micrel: Add workaround for incomplete autonegotiation

[Heiner Kallweit <hkallweit1@gmail.com>]

On 28.12.2023 00:16, Asmaa Mnebhi wrote:
> Very rarely, the KSZ9031 fails to complete autonegotiation although it was
> initiated via phy_start(). As a result, the link stays down. Restarting
> autonegotiation when in this state solves the issue.
> 
The patch isn't addressed to all relevant maintainers. Please use the
get_maintainers script.

You should use the net/net-next annotation to make clear whether this should
be treated as a fix (in this case add a Fixes tag) or net-next material.

> Signed-off-by: Asmaa Mnebhi <asmaa@nvidia.com>
> ---
> v1->v2:
> - Use msleep() instead of mdelay()
> 
>  drivers/net/phy/micrel.c | 17 +++++++++++++++++
>  1 file changed, 17 insertions(+)
> 
> diff --git a/drivers/net/phy/micrel.c b/drivers/net/phy/micrel.c
> index 08e3915001c3..9952a073413f 100644
> --- a/drivers/net/phy/micrel.c
> +++ b/drivers/net/phy/micrel.c
> @@ -1475,6 +1475,7 @@ static int ksz9031_get_features(struct phy_device *phydev)
>  
>  static int ksz9031_read_status(struct phy_device *phydev)
>  {
> +	u8 timeout = 10;
>  	int err;
>  	int regval;
>  
> @@ -1494,6 +1495,22 @@ static int ksz9031_read_status(struct phy_device *phydev)
>  		return genphy_config_aneg(phydev);
>  	}
>  
> +	/* KSZ9031's autonegotiation takes normally 4-5 seconds to complete.
> +	 * Occasionally it fails to complete autonegotiation. The workaround is
> +	 * to restart it.
> +	 */
> +        if (phydev->autoneg == AUTONEG_ENABLE) {
> +		while (timeout) {
> +			if (phy_aneg_done(phydev))
> +				break;
> +			msleep(1000);
> +			timeout--;

It's not too nice to do this synchronously. Even in the non-problem case this
will block the phylib state machine for seconds. Better find a way to do it
asynchronously.

> +		};
> +
> +		if (timeout == 0)
> +			phy_restart_aneg(phydev);
> +	}
> +
>  	return 0;
>  }
>