Re: [RFC PATCH net-next v3 19/20] net: dsa: qca8k: pass switch_revision info to phy dev_flags

[Vladimir Oltean <olteanv@gmail.com>]

On Sat, May 08, 2021 at 12:33:53AM +0100, Russell King - ARM Linux admin wrote:
> On Sat, May 08, 2021 at 01:26:02AM +0200, Ansuel Smith wrote:
> > On Thu, May 06, 2021 at 02:24:58PM +0300, Vladimir Oltean wrote:
> > > On Wed, May 05, 2021 at 12:29:13AM +0200, Ansuel Smith wrote:
> > > > Define get_phy_flags to pass switch_Revision needed to tweak the
> > > > internal PHY with debug values based on the revision.
> > > > 
> > > > Signed-off-by: Ansuel Smith <ansuelsmth@gmail.com>
> > > > ---
> > > >  drivers/net/dsa/qca8k.c | 19 +++++++++++++++++++
> > > >  1 file changed, 19 insertions(+)
> > > > 
> > > > diff --git a/drivers/net/dsa/qca8k.c b/drivers/net/dsa/qca8k.c
> > > > index b4cd891ad35d..237e09bb1425 100644
> > > > --- a/drivers/net/dsa/qca8k.c
> > > > +++ b/drivers/net/dsa/qca8k.c
> > > > @@ -1654,6 +1654,24 @@ qca8k_port_vlan_del(struct dsa_switch *ds, int port,
> > > >  	return ret;
> > > >  }
> > > >  
> > > > +static u32 qca8k_get_phy_flags(struct dsa_switch *ds, int port)
> > > > +{
> > > > +	struct qca8k_priv *priv = ds->priv;
> > > > +
> > > > +	pr_info("revision from phy %d", priv->switch_revision);
> > > 
> > > Log spam.
> > > 
> > > > +	/* Communicate to the phy internal driver the switch revision.
> > > > +	 * Based on the switch revision different values needs to be
> > > > +	 * set to the dbg and mmd reg on the phy.
> > > > +	 * The first 2 bit are used to communicate the switch revision
> > > > +	 * to the phy driver.
> > > > +	 */
> > > > +	if (port > 0 && port < 6)
> > > > +		return priv->switch_revision;
> > > > +
> > > > +	return 0;
> > > > +}
> > > > +
> > > >  static enum dsa_tag_protocol
> > > >  qca8k_get_tag_protocol(struct dsa_switch *ds, int port,
> > > >  		       enum dsa_tag_protocol mp)
> > > > @@ -1687,6 +1705,7 @@ static const struct dsa_switch_ops qca8k_switch_ops = {
> > > >  	.phylink_mac_config	= qca8k_phylink_mac_config,
> > > >  	.phylink_mac_link_down	= qca8k_phylink_mac_link_down,
> > > >  	.phylink_mac_link_up	= qca8k_phylink_mac_link_up,
> > > > +	.get_phy_flags		= qca8k_get_phy_flags,
> > > >  };
> > > >  
> > > >  static int qca8k_read_switch_id(struct qca8k_priv *priv)
> > > > -- 
> > > > 2.30.2
> > > > 
> > > 
> > > Florian, I think at one point you said that a correct user of
> > > phydev->dev_flags should first check the PHY revision and not apply
> > > dev_flags in blind, since they are namespaced to each PHY driver?
> > > It sounds a bit circular to pass the PHY revision to the PHY through
> > > phydev->dev_flags, either that or I'm missing some piece.
> > 
> > Just to make sure. This is the SWITCH revision not the PHY revision. It
> > was pointed out in old version that I should get this value from the PHY
> > regs but they are different values. This is why the dsa driver needs to
> > use the dev_flags to pass the SWITCH revision to the phy driver. Am I
> > implementing this in the wrong way and I should declare something to
> > pass this value in a more standard way? (anyway i'm pushing v4 so i
> > don't know if we should continue that there)
> 
> Vladimir is confused - it is not PHY revision at all, but the PHY
> identifiers.
> 
> What was actually suggested was checking the PHY identifiers before
> passing PHY-driver specific flags, so that we didn't end up setting
> driver private flags that are intending for one driver, but end up
> actually binding a different driver, and mis-interpreting the flags.
> 
> This is one of the problems of the current scheme: it's just a
> meaningless opaque u32 variable with no defined structure to it that
> the various PHY drivers themselves use in whatever way they see fit.
> That is only fine to use _if_ you know for certain which driver is
> going to bind ahead of time.
> 
> As I mentioned in direct reply to your patch, there was discussions
> about this back in February, but they seem to have stalled.

Yes, I was indeed confused. My problem was mixing up the PHY OUI/device ID
and revision concepts in one big fuzzy notion. I remembered Heiner's
suggestion to do something similar to mv88e6xxx_mdio_read from here:
https://patchwork.kernel.org/project/netdevbpf/patch/20210423014741.11858-12-ansuelsmth@gmail.com/
(where the problem is that some internal PHYs are lacking a device
identifier) and thought that the problem here is the same.

Nonetheless, now it is clear to me that with care (don't set dev_flags
except for internal PHYs which are statically known), it is possible for
the PHY driver to have a larger identifier (PHY ID concatenated with
switch revision passed through dev_flags) based on which it can
configure the hardware.

Sorry.