RE: [PATCH net] net: libwx: fix to set pause param

["Jiawen Wu" <jiawenwu@trustnetic.com>]

On Fri, Apr 25, 2025 3:38 PM, Russell King (Oracle) wrote:
> On Fri, Apr 25, 2025 at 03:09:42PM +0800, Jiawen Wu wrote:
> > @@ -266,11 +266,20 @@ int wx_set_pauseparam(struct net_device *netdev,
> >  		      struct ethtool_pauseparam *pause)
> >  {
> >  	struct wx *wx = netdev_priv(netdev);
> > +	int err;
> >
> >  	if (wx->mac.type == wx_mac_aml)
> >  		return -EOPNOTSUPP;
> >
> > -	return phylink_ethtool_set_pauseparam(wx->phylink, pause);
> > +	err = phylink_ethtool_set_pauseparam(wx->phylink, pause);
> > +	if (err)
> > +		return err;
> > +
> > +	if (wx->fc.rx_pause != pause->rx_pause ||
> > +	    wx->fc.tx_pause != pause->tx_pause)
> > +		return wx_fc_enable(wx, pause->tx_pause, pause->rx_pause);
> 
> Why? phylink_ethtool_set_pauseparam() will cause mac_link_down() +
> mac_link_up() to be called with the new parameters.
> 
> One of the points of phylink is to stop drivers implementing stuff
> buggily - which is exactly what the above is.
> 
> ->rx_pause and ->tx_pause do not set the pause enables unconditionally.
> Please read the documentation in include/uapi/linux/ethtool.h which
> states how these two flags are interpreted, specifically the last
> paragraph of the struct's documentation.
> 
> I'm guessing your change comes from a misunderstanding how the
> interface is supposed to work and you believe that phylink isn't
> implementing it correctly.

You are right.
I should set autoneg off first, although there has no autoneg bit in this link mode.