Re: [PATCH net-next 05/12] net: ftgmac100: Convert using devm_clk_get_enabled() in ftgmac100_setup_clk()

[Li Zetao <lizetao1@huawei.com>]

Hi,

在 2024/9/3 16:09, Uwe Kleine-König 写道:
> Hello,
> 
> On Sat, Aug 31, 2024 at 10:13:27AM +0800, Li Zetao wrote:
>> Use devm_clk_get_enabled() instead of devm_clk_get() +
>> clk_prepare_enable(), which can make the clk consistent with the device
>> life cycle and reduce the risk of unreleased clk resources. Since the
>> device framework has automatically released the clk resource, there is
>> no need to execute clk_disable_unprepare(clk) on the error path, drop
>> the cleanup_clk label, and the original error process can return directly.
>>
>> Signed-off-by: Li Zetao <lizetao1@huawei.com>
>> ---
>>   drivers/net/ethernet/faraday/ftgmac100.c | 27 ++++++------------------
>>   1 file changed, 7 insertions(+), 20 deletions(-)
>>
>> diff --git a/drivers/net/ethernet/faraday/ftgmac100.c b/drivers/net/ethernet/faraday/ftgmac100.c
>> index 4c546c3aef0f..eb57c822c5ac 100644
>> --- a/drivers/net/ethernet/faraday/ftgmac100.c
>> +++ b/drivers/net/ethernet/faraday/ftgmac100.c
>> @@ -1752,13 +1752,10 @@ static int ftgmac100_setup_clk(struct ftgmac100 *priv)
>>   	struct clk *clk;
>>   	int rc;
>>   
>> -	clk = devm_clk_get(priv->dev, NULL /* MACCLK */);
>> +	clk = devm_clk_get_enabled(priv->dev, NULL /* MACCLK */);
>>   	if (IS_ERR(clk))
>>   		return PTR_ERR(clk);
>>   	priv->clk = clk;
>> -	rc = clk_prepare_enable(priv->clk);
>> -	if (rc)
>> -		return rc;
>>   
>>   	/* Aspeed specifies a 100MHz clock is required for up to
>>   	 * 1000Mbit link speeds. As NCSI is limited to 100Mbit, 25MHz
>> @@ -1767,21 +1764,17 @@ static int ftgmac100_setup_clk(struct ftgmac100 *priv)
>>   	rc = clk_set_rate(priv->clk, priv->use_ncsi ? FTGMAC_25MHZ :
>>   			  FTGMAC_100MHZ);
>>   	if (rc)
>> -		goto cleanup_clk;
>> +		return rc;
>>   
>>   	/* RCLK is for RMII, typically used for NCSI. Optional because it's not
>>   	 * necessary if it's the AST2400 MAC, or the MAC is configured for
>>   	 * RGMII, or the controller is not an ASPEED-based controller.
>>   	 */
>> -	priv->rclk = devm_clk_get_optional(priv->dev, "RCLK");
>> -	rc = clk_prepare_enable(priv->rclk);
>> -	if (!rc)
>> -		return 0;
>> +	priv->rclk = devm_clk_get_optional_enabled(priv->dev, "RCLK");
>> +	if (IS_ERR(priv->rclk))
>> +		return PTR_ERR(priv->rclk);
>>   
>> -cleanup_clk:
>> -	clk_disable_unprepare(priv->clk);
>> -
>> -	return rc;
>> +	return 0;
> 
> You're changing semantics here. Before your patch ftgmac100_setup_clk()
> was left with priv->clk disabled; now you keep it enabled.
Before my patch, ftgmac100_setup_clk() was only left with priv->clk 
disabled when error occurs, and was left with priv->clk enabled when no 
error occurs because when enabling priv->rclk successfully, it will 
return 0 directly, and when enabling priv->rclk failed, it will disable 
priv->clk.

It turns out that the code logic is a bit counter-intuitive, but the 
readability has been improved after adjustments.
> 
> Further note that there is a bug here, because in ftgmac100_probe()
> (i.e. the caller of ftgmac100_setup_clk())
> clk_disable_unprepare(priv->clk) is called in the error path.
> (I only looked quickly, so I might have missed a detail.)
I have considered the case that clk_disable_unprepare will not be 
executed on the wrong path in ftgmac100_probe(). So I understand that 
the problem you mentioned should not exist.
> 
> So while your patch is an improvement for clock enable/disable
> balancing, it might regress on power consumption.
> 
>>   }
>>   
>>   static bool ftgmac100_has_child_node(struct device_node *np, const char *name)
>> @@ -1996,16 +1989,13 @@ static int ftgmac100_probe(struct platform_device *pdev)
>>   	err = register_netdev(netdev);
>>   	if (err) {
>>   		dev_err(&pdev->dev, "Failed to register netdev\n");
>> -		goto err_register_netdev;
>> +		goto err_phy_connect;
>>   	}
>>   
>>   	netdev_info(netdev, "irq %d, mapped at %p\n", netdev->irq, priv->base);
>>   
>>   	return 0;
>>   
>> -err_register_netdev:
>> -	clk_disable_unprepare(priv->rclk);
>> -	clk_disable_unprepare(priv->clk);
>>   err_phy_connect:
>>   	ftgmac100_phy_disconnect(netdev);
>>   err_ncsi_dev:
>> @@ -2034,9 +2024,6 @@ static void ftgmac100_remove(struct platform_device *pdev)
>>   		ncsi_unregister_dev(priv->ndev);
>>   	unregister_netdev(netdev);
>>   
>> -	clk_disable_unprepare(priv->rclk);
>> -	clk_disable_unprepare(priv->clk);
>> -
>>   	/* There's a small chance the reset task will have been re-queued,
>>   	 * during stop, make sure it's gone before we free the structure.
>>   	 */
> 
> Best regards
> Uwe

Thanks,
Li Zetao.