Re: [PATCH v2 3/3] net: stmmac: Bring down the clocks to lower frequencies when mac link goes down

[Sagar Cheluvegowda <quic_scheluve@quicinc.com>]

On 6/28/2024 2:50 PM, Sagar Cheluvegowda wrote:
> 
> 
> On 6/26/2024 7:58 AM, Andrew Halaney wrote:
>> On Tue, Jun 25, 2024 at 04:49:30PM GMT, Sagar Cheluvegowda wrote:
>>> When mac link goes down we don't need to mainitain the clocks to operate
>>> at higher frequencies, as an optimized solution to save power when
>>> the link goes down we are trying to bring down the clocks to the
>>> frequencies corresponding to the lowest speed possible.
>>>
>>> Signed-off-by: Sagar Cheluvegowda <quic_scheluve@quicinc.com>
>>> ---
>>>  drivers/net/ethernet/stmicro/stmmac/stmmac_main.c | 8 ++++++++
>>>  1 file changed, 8 insertions(+)
>>>
>>> diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
>>> index ec7c61ee44d4..f0166f0bc25f 100644
>>> --- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
>>> +++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
>>> @@ -996,6 +996,9 @@ static void stmmac_mac_link_down(struct phylink_config *config,
>>>  {
>>>  	struct stmmac_priv *priv = netdev_priv(to_net_dev(config->dev));
>>>  
>>> +	if (priv->plat->fix_mac_speed)
>>> +		priv->plat->fix_mac_speed(priv->plat->bsp_priv, SPEED_10, mode);
>>> +
The above fix_mac_speed needs to be removed, i lately realized this mistake.
>>>  	stmmac_mac_set(priv, priv->ioaddr, false);
>>>  	priv->eee_active = false;
>>>  	priv->tx_lpi_enabled = false;
>>> @@ -1004,6 +1007,11 @@ static void stmmac_mac_link_down(struct phylink_config *config,
>>>  
>>>  	if (priv->dma_cap.fpesel)
>>>  		stmmac_fpe_link_state_handle(priv, false);
>>> +
>>> +	stmmac_set_icc_bw(priv, SPEED_10);
>>> +
>>> +	if (priv->plat->fix_mac_speed)
>>> +		priv->plat->fix_mac_speed(priv->plat->bsp_priv, SPEED_10, mode);
>>
>>
>> I think you're doing this at the beginning and end of
>> stmmac_mac_link_down(), is that intentional?
>>
>>
> 
> I realised that bringing down the clock to 10Mbps should be the last operation
> of the link down process, the reason being if we bring down the clocks first it will
> deprive essential internal clocks to DMA/MTL modules which are required for
> Cleanup operations this might cause excessive delays in stopping DMA
> or flusing MTL queues.
>  
>>