Re: [RFC PATCH net-next] net: dsa: mt7530: fix port specifications for MT7988

[Daniel Golle <daniel@makrotopia.org>]

On Fri, Apr 07, 2023 at 01:46:20PM +0300, Arınç ÜNAL wrote:
> On 7.04.2023 12:28, Daniel Golle wrote:
> > On Fri, Apr 07, 2023 at 11:56:08AM +0300, Arınç ÜNAL wrote:
> > > On 7.04.2023 00:57, Daniel Golle wrote:
> > > > On Fri, Apr 07, 2023 at 12:43:41AM +0300, Arınç ÜNAL wrote:
> > > > > On 6.04.2023 14:07, Russell King (Oracle) wrote:
> > > > > > On Thu, Apr 06, 2023 at 01:04:45PM +0300, arinc9.unal@gmail.com wrote:
> > > > > > > From: Arınç ÜNAL <arinc.unal@arinc9.com>
> > > > > > > 
> > > > > > > On the switch on the MT7988 SoC, there are only 4 PHYs. There's only port 6
> > > > > > > as the CPU port, there's no port 5. Split the switch statement with a check
> > > > > > > to enforce these for the switch on the MT7988 SoC. The internal phy-mode is
> > > > > > > specific to MT7988 so put it for MT7988 only.
> > > > > > > 
> > > > > > > Signed-off-by: Arınç ÜNAL <arinc.unal@arinc9.com>
> > > > > > > ---
> > > > > > > 
> > > > > > > Daniel, this is based on the information you provided me about the switch.
> > > > > > > I will add this to my current patch series if it looks good to you.
> > > > > > > 
> > > > > > > Arınç
> > > > > > > 
> > > > > > > ---
> > > > > > >     drivers/net/dsa/mt7530.c | 67 ++++++++++++++++++++++++++--------------
> > > > > > >     1 file changed, 43 insertions(+), 24 deletions(-)
> > > > > > > 
> > > > > > > diff --git a/drivers/net/dsa/mt7530.c b/drivers/net/dsa/mt7530.c
> > > > > > > index 6fbbdcb5987f..f167fa135ef1 100644
> > > > > > > --- a/drivers/net/dsa/mt7530.c
> > > > > > > +++ b/drivers/net/dsa/mt7530.c
> > > > > > > @@ -2548,7 +2548,7 @@ static void mt7988_mac_port_get_caps(struct dsa_switch *ds, int port,
> > > > > > >     	phy_interface_zero(config->supported_interfaces);
> > > > > > >     	switch (port) {
> > > > > > > -	case 0 ... 4: /* Internal phy */
> > > > > > > +	case 0 ... 3: /* Internal phy */
> > > > > > >     		__set_bit(PHY_INTERFACE_MODE_INTERNAL,
> > > > > > >     			  config->supported_interfaces);
> > > > > > >     		break;
> > > > > > > @@ -2710,37 +2710,56 @@ mt753x_phylink_mac_config(struct dsa_switch *ds, int port, unsigned int mode,
> > > > > > >     	struct mt7530_priv *priv = ds->priv;
> > > > > > >     	u32 mcr_cur, mcr_new;
> > > > > > > -	switch (port) {
> > > > > > > -	case 0 ... 4: /* Internal phy */
> > > > > > > -		if (state->interface != PHY_INTERFACE_MODE_GMII &&
> > > > > > > -		    state->interface != PHY_INTERFACE_MODE_INTERNAL)
> > > > > > > -			goto unsupported;
> > > > > > > -		break;
> > > > > > > -	case 5: /* Port 5, a CPU port. */
> > > > > > > -		if (priv->p5_interface == state->interface)
> > > > > > > +	if (priv->id == ID_MT7988) {
> > > > > > > +		switch (port) {
> > > > > > > +		case 0 ... 3: /* Internal phy */
> > > > > > > +			if (state->interface != PHY_INTERFACE_MODE_INTERNAL)
> > > > > > 
> > > > > > How do these end up with PHY_INTERFACE_MODE_INTERNAL ? phylib defaults
> > > > > > to GMII mode without something else being specified in DT.
> > > > > > 
> > > > > > Also note that you should *not* be validating state->interface in the
> > > > > > mac_config() method because it's way too late to reject it - if you get
> > > > > > an unsupported interface here, then that is down to the get_caps()
> > > > > > method being buggy. Only report interfaces in get_caps() that you are
> > > > > > prepared to handle in the rest of the system.
> > > > > 
> > > > > This is already the case for all three get_caps(). The supported interfaces
> > > > > for each port are properly defined.
> > > > > 
> > > > > Though mt7988_mac_port_get_caps() clears the config->supported_interfaces
> > > > > bitmap before reporting the supported interfaces. I don't think this is
> > > > > needed as all bits in the bitmap should already be initialized to zero when
> > > > > the phylink_config structure is allocated.
> > > > > 
> > > > > I'm not sure if your suggestion is to make sure the supported interfaces are
> > > > > properly reported on get_caps(), or validate state->interface somewhere
> > > > > else.
> > > > 
> > > > I think what Russell meant is just there is no point in being overly
> > > > precise about permitted interface modes in mt753x_phylink_mac_config,
> > > > as this function is not meant and called too late to validate the
> > > > validity of the selected interface mode.
> > > > 
> > > > You change to mt7988_mac_port_get_caps looks correct to me and doing
> > > > this will already prevent mt753x_phylink_mac_config from ever being
> > > > called on MT7988 for port == 4 as well as and port == 5.
> > > 
> > > Ah, thanks for pointing this out Daniel. I see ds->ops->phylink_get_caps()
> > > is run right before phylink_create() on dsa_port_phylink_create(), as it
> > > should get the capabilities before creating an instance.
> > > 
> > > Should I remove phy_interface_zero(config->supported_interfaces);
> > > mt7988_mac_port_get_caps()? I'd prefer to do identical operations on each
> > > get_caps(), if there's no apparent reason for this to be on
> > > mt7988_mac_port_get_caps().
> > 
> > Yes, sounds sane to me, please do so.
> > 
> > Also we could make .mac_port_config optional, as for MT7988 we actually
> > won't need it at all:
> > 
> > diff --git a/drivers/net/dsa/mt7530.c b/drivers/net/dsa/mt7530.c
> > index e4bb5037d3525..5efcb9897eb18 100644
> > --- a/drivers/net/dsa/mt7530.c
> > +++ b/drivers/net/dsa/mt7530.c
> > @@ -2653,17 +2653,6 @@ static bool mt753x_is_mac_port(u32 port)
> >   	return (port == 5 || port == 6);
> >   }
> > -static int
> > -mt7988_mac_config(struct dsa_switch *ds, int port, unsigned int mode,
> > -		  phy_interface_t interface)
> > -{
> > -	if (dsa_is_cpu_port(ds, port) &&
> > -	    interface == PHY_INTERFACE_MODE_INTERNAL)
> > -		return 0;
> > -
> > -	return -EINVAL;
> > -}
> > -
> >   static int
> >   mt7531_mac_config(struct dsa_switch *ds, int port, unsigned int mode,
> >   		  phy_interface_t interface)
> > @@ -2704,6 +2693,9 @@ mt753x_mac_config(struct dsa_switch *ds, int port, unsigned int mode,
> >   {
> >   	struct mt7530_priv *priv = ds->priv;
> > +	if (!priv->info->mac_port_config)
> > +		return 0;
> > +
> >   	return priv->info->mac_port_config(ds, port, mode, state->interface);
> >   }
> > @@ -3157,7 +3149,6 @@ const struct mt753x_info mt753x_table[] = {
> >   		.pad_setup = mt7988_pad_setup,
> >   		.cpu_port_config = mt7988_cpu_port_config,
> >   		.mac_port_get_caps = mt7988_mac_port_get_caps,
> > -		.mac_port_config = mt7988_mac_config,
> >   	},
> >   };
> >   EXPORT_SYMBOL_GPL(mt753x_table);
> > @@ -3186,8 +3177,7 @@ mt7530_probe_common(struct mt7530_priv *priv)
> >   	 */
> >   	if (!priv->info->sw_setup || !priv->info->pad_setup ||
> >   	    !priv->info->phy_read_c22 || !priv->info->phy_write_c22 ||
> > -	    !priv->info->mac_port_get_caps ||
> > -	    !priv->info->mac_port_config)
> > +	    !priv->info->mac_port_get_caps)
> 
> Why split the sanity check? Isn't just removing mt7988_mac_config() and
> .mac_port_config = mt7988_mac_config enough?

No, because the sanity check currently requires a pointer to a
mac_port_config function to be non-NULL. If we want to make it optional
then we'll also have to skip that in the santity check which will
otherwise fail.