Re: [PATCH net-next v2 1/3] net: ethernet: ti: am65-cpsw: Move mode specific config to mac_config()

[Siddharth Vadapalli <s-vadapalli@ti.com>]

On 03-04-2023 18:43, Russell King (Oracle) wrote:
> On Mon, Apr 03, 2023 at 06:31:52PM +0530, Siddharth Vadapalli wrote:
>>
>>
>> On 03-04-2023 16:38, Russell King (Oracle) wrote:
>>> On Mon, Apr 03, 2023 at 04:31:04PM +0530, Siddharth Vadapalli wrote:
>>>> Move the interface mode specific configuration to the mac_config()
>>>> callback am65_cpsw_nuss_mac_config().
>>>>
>>>> Signed-off-by: Siddharth Vadapalli <s-vadapalli@ti.com>
>>>> ---
>>>>  drivers/net/ethernet/ti/am65-cpsw-nuss.c | 10 +++++++---
>>>>  1 file changed, 7 insertions(+), 3 deletions(-)
>>>>
>>>> diff --git a/drivers/net/ethernet/ti/am65-cpsw-nuss.c b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
>>>> index d17757ecbf42..74e099828978 100644
>>>> --- a/drivers/net/ethernet/ti/am65-cpsw-nuss.c
>>>> +++ b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
>>>> @@ -1504,12 +1504,17 @@ static void am65_cpsw_nuss_mac_config(struct phylink_config *config, unsigned in
>>>>  							  phylink_config);
>>>>  	struct am65_cpsw_port *port = container_of(slave, struct am65_cpsw_port, slave);
>>>>  	struct am65_cpsw_common *common = port->common;
>>>> +	u32 mac_control = 0;
>>>>  
>>>>  	if (common->pdata.extra_modes & BIT(state->interface)) {
>>>> -		if (state->interface == PHY_INTERFACE_MODE_SGMII)
>>>> +		if (state->interface == PHY_INTERFACE_MODE_SGMII) {
>>>> +			mac_control |= CPSW_SL_CTL_EXT_EN;
>>>>  			writel(ADVERTISE_SGMII,
>>>>  			       port->sgmii_base + AM65_CPSW_SGMII_MR_ADV_ABILITY_REG);
>>>> +		}
>>>>  
>>>> +		if (mac_control)
>>>> +			cpsw_sl_ctl_set(port->slave.mac_sl, mac_control);
>>>>  		writel(AM65_CPSW_SGMII_CONTROL_MR_AN_ENABLE,
>>>>  		       port->sgmii_base + AM65_CPSW_SGMII_CONTROL_REG);
>>>>  	}
>>>> @@ -1553,8 +1558,7 @@ static void am65_cpsw_nuss_mac_link_up(struct phylink_config *config, struct phy
>>>>  
>>>>  	if (speed == SPEED_1000)
>>>>  		mac_control |= CPSW_SL_CTL_GIG;
>>>> -	if (interface == PHY_INTERFACE_MODE_SGMII)
>>>> -		mac_control |= CPSW_SL_CTL_EXT_EN;
>>>> +	/* TODO: Verify whether in-band is necessary for 10 Mbps RGMII */
>>>>  	if (speed == SPEED_10 && phy_interface_mode_is_rgmii(interface))
>>>>  		/* Can be used with in band mode only */
>>>>  		mac_control |= CPSW_SL_CTL_EXT_EN;
>>>
>>> I'm afraid I can see you haven't thought this patch through properly.
>>>
>>> am65_cpsw_nuss_mac_link_down() will call
>>> cpsw_sl_ctl_reset(port->slave.mac_sl); which has the effect of clearing
>>> to zero the entire MAC control register. This will clear
>>> CPSW_SL_CTL_EXT_EN that was set in am65_cpsw_nuss_mac_config() which is
>>> not what you want to be doing.
>>>
>>> Given that we have the 10Mbps issue with RGMII, I think what you want
>>> to be doing is:
>>>
>>> 1. Set CPSW_SL_CTL_EXT_EN in am65_cpsw_nuss_mac_config() if in SGMII
>>>    mode, otherwise clear this bit.
>>>
>>> 2. Clear the mac_control register in am65_cpsw_nuss_mac_link_down()
>>>    if in RMGII mode, otherwise preserve the state of
>>>    CPSW_SL_CTL_EXT_EN but clear all other bits.
>>>
>>> 3. Set CPSW_SL_CTL_EXT_EN in am65_cpsw_nuss_mac_link_up() if in
>>>    RGMII mode and 10Mbps.
>>
>> I plan to implement it as follows:
>> 1. Add a member "u32 mode_config" to "struct am65_cpsw_slave_data" in
>> "am65-cpsw-nuss.h".
>> 2. In am65_cpsw_nuss_mac_config(), store the value of mac_control in
>> "port->slave.mode_config".
>> 3. In am65_cpsw_nuss_mac_link_down(), after the reset via
>> cpsw_sl_ctl_reset(), execute:
>> cpsw_sl_ctl_set(port->slave.mac_sl, port->slave.mode_config) in order to
>> restore the configuration performed in am65_cpsw_nuss_mac_config().
>>
>> Please let me know in case of any suggestions to implement it in a
>> better manner.
> 
> Do you think this complexity is really worth it?
> 
> Let's look at what's available:
> 
> cpsw_sl_ctl_set() - sets bits in the mac control register
> cpsw_sl_ctl_clr() - clears bits in the mac control register
> cpsw_sl_ctl_reset() - sets the mac control register to zero
> 
> So, in mac_config(), we can do:
> 
> 	if (interface == SGMII)
> 		cpsw_sl_ctl_set(CPSW_SL_CTL_EXT_EN);
> 	else
> 		cpsw_sl_ctl_clr(CPSW_SL_CTL_EXT_EN);

While this will work for patch 1/3, once I add support for USXGMII mode
as in patch 3/3, I believe that I have to invert it, beginning by
invoking a cpsw_sl_ctl_clr(CPSW_SL_CTL_EXT_EN) at the start in
mac_config() followed by switching through the modes. If the mode is
SGMII, then I invoke cpsw_sl_ctl_set(CPSW_SL_CTL_EXT_EN), along with the
write to the MR_ADV_ABILITY_REG register to advertise SGMII. If the mode
is USXGMII, then I invoke:
cpsw_sl_ctl_set(CPSW_SL_CTL_XGIG | CPSW_SL_CTL_XGMII_EN)

> 
> 
> in mac_link_down():
> 
> 	u32 ctl;
> 
> 	ctl = CPSW_SL_CTL_GMII_EN | CPSW_SL_CTL_GIG |
> 	      CPSW_SL_CTL_IFCTL_A | CPSW_SL_CTL_FULLDUPLEX |
> 	      CPSW_SL_CTL_RX_FLOW_EN | CPSW_SL_CTL_TX_FLOW_EN;
> 	if (phy_interface_mode_is_rgmii(interface))
> 		ctl |= CPSW_SL_CTL_EXT_EN;
> 
> 	cpsw_sl_ctl_clr(ctl);
> 
> This ensures that we don't touch any bits in mac_link_down() which we
> aren't modifying in the corresponding mac_link_up() implementation.

This is very helpful. Thank you for the suggestion :)

> 
> Q: do we really need to clear the mac control register on link down?
> If we don't, then we can do better, but we need an additional helper
> which allows read-modify-write of the mac control register using a
> mask value and a value of bits to set. Then we can have mac_link_up()
> setting and clearing the bits as necessary - but I would still keep
> the clearing of CPSW_SL_CTL_EXT_EN for RGMII modes in mac_link_down()
> for now.

Clearing the entire register is not necessary. As long as those bits
that were set during mac_link_up() are cleared, it should be sufficient.

For now, I will implement the changes assuming that the mac control
register will be cleared on link down, based on the above suggestions
provided by you.

Regards,
Siddharth.