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

[Arınç ÜNAL]

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 at gmail.com wrote:
>>>> From: Arınç ÜNAL <arinc.unal at 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 at 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().

Arınç