Re: [PATCH net-next 1/4] net: dsa: move up phy enabling in core

[Florian Fainelli]

On 09/22/2017 09:32 AM, Andrew Lunn wrote:
> On Fri, Sep 22, 2017 at 12:17:50PM -0400, Vivien Didelot wrote:
>> bcm_sf2 is currently the only driver using the phy argument passed to
>> .port_enable. It resets the state machine if the phy has been hard
>> reset. This check is generic and can be moved to DSA core.
>>  
>>  dsa_port_set_state_now(p->dp, stp_state);
>>  
>> -if (p->phy)
>> -phy_start(p->phy);
>> +if (phy) {
>> +/* If phy_stop() has been called before, phy will be in
>> + * halted state, and phy_start() will call resume.
>> + *
>> + * The resume path does not configure back autoneg
>> + * settings, and since the internal phy may have been
>> + * hard reset, we need to reset the state machine also.
>> + */
>> +phy->state = PHY_READY;
>> +phy_init_hw(phy);
>> +phy_start(phy);
>> +}
> 
> Hi Vivien
> 
> If this is generic, why is it needed at all here? Shouldn't this
> actually by in phylib?

This does not belong in the core logic within net/dsa/slave.c. The
reason why this is necessary here is because we are doing a HW-based
reset of the PHY, as the comment explains this is specific to how the HW
works. There may be a cleaner solution to this problem, but in any case,
I don't think other drivers should inherit that logic.
-- 
Florian

Re: [PATCH net-next 1/4] net: dsa: move up phy enabling in core

[Florian Fainelli]

On 09/22/2017 09:17 AM, Vivien Didelot wrote:
> bcm_sf2 is currently the only driver using the phy argument passed to
> .port_enable. It resets the state machine if the phy has been hard
> reset. This check is generic and can be moved to DSA core.

This is completely specific to bcm_sf2 because it does call
bcm_sf2_gphy_enable_set() which performs a HW reset of the PHY, you
can't move this to the generic portion of net/dsa/slave.c. NACK.

> 
> Signed-off-by: Vivien Didelot 
> ---
>  drivers/net/dsa/bcm_sf2.c | 16 +---
>  net/dsa/slave.c   | 15 +--
>  2 files changed, 14 insertions(+), 17 deletions(-)
> 
> diff --git a/drivers/net/dsa/bcm_sf2.c b/drivers/net/dsa/bcm_sf2.c
> index 898d5642b516..ad96b9725a2c 100644
> --- a/drivers/net/dsa/bcm_sf2.c
> +++ b/drivers/net/dsa/bcm_sf2.c
> @@ -184,22 +184,8 @@ static int bcm_sf2_port_setup(struct dsa_switch *ds, int 
> port,
>   core_writel(priv, reg, CORE_PORT_TC2_QOS_MAP_PORT(port));
>  
>   /* Re-enable the GPHY and re-apply workarounds */
> - if (priv->int_phy_mask & 1 << port && priv->hw_params.num_gphy == 1) {
> + if (priv->int_phy_mask & 1 << port && priv->hw_params.num_gphy == 1)
>   bcm_sf2_gphy_enable_set(ds, true);
> - if (phy) {
> - /* if phy_stop() has been called before, phy
> -  * will be in halted state, and phy_start()
> -  * will call resume.
> -  *
> -  * the resume path does not configure back
> -  * autoneg settings, and since we hard reset
> -  * the phy manually here, we need to reset the
> -  * state machine also.
> -  */
> - phy->state = PHY_READY;
> - phy_init_hw(phy);
> - }
> - }
>  
>   /* Enable MoCA port interrupts to get notified */
>   if (port == priv->moca_port)
> diff --git a/net/dsa/slave.c b/net/dsa/slave.c
> index 02ace7d462c4..606812160fd5 100644
> --- a/net/dsa/slave.c
> +++ b/net/dsa/slave.c
> @@ -72,6 +72,7 @@ static int dsa_slave_get_iflink(const struct net_device 
> *dev)
>  static int dsa_slave_open(struct net_device *dev)
>  {
>   struct dsa_slave_priv *p = netdev_priv(dev);
> + struct phy_device *phy = p->phy;
>   struct dsa_port *dp = p->dp;
>   struct dsa_switch *ds = dp->ds;
>   struct net_device *master = dsa_master_netdev(p);
> @@ -106,8 +107,18 @@ static int dsa_slave_open(struct net_device *dev)
>  
>   dsa_port_set_state_now(p->dp, stp_state);
>  
> - if (p->phy)
> - phy_start(p->phy);
> + if (phy) {
> + /* If phy_stop() has been called before, phy will be in
> +  * halted state, and phy_start() will call resume.
> +  *
> +  * The resume path does not configure back autoneg
> +  * settings, and since the internal phy may have been
> +  * hard reset, we need to reset the state machine also.
> +  */
> + phy->state = PHY_READY;
> + phy_init_hw(phy);
> + phy_start(phy);
> + }
>  
>   return 0;
>  
> 


-- 
Florian

Re: [PATCH net-next 1/4] net: dsa: move up phy enabling in core

[Andrew Lunn]

On Fri, Sep 22, 2017 at 12:17:50PM -0400, Vivien Didelot wrote:
> bcm_sf2 is currently the only driver using the phy argument passed to
> .port_enable. It resets the state machine if the phy has been hard
> reset. This check is generic and can be moved to DSA core.
>  
>   dsa_port_set_state_now(p->dp, stp_state);
>  
> - if (p->phy)
> - phy_start(p->phy);
> + if (phy) {
> + /* If phy_stop() has been called before, phy will be in
> +  * halted state, and phy_start() will call resume.
> +  *
> +  * The resume path does not configure back autoneg
> +  * settings, and since the internal phy may have been
> +  * hard reset, we need to reset the state machine also.
> +  */
> + phy->state = PHY_READY;
> + phy_init_hw(phy);
> + phy_start(phy);
> + }

Hi Vivien

If this is generic, why is it needed at all here? Shouldn't this
actually by in phylib?

Florian ?

   Andrew