Re: [RFC v1 1/3] net: dsa: mt753x: make reset optional

[Vladimir Oltean]

On Tue, Apr 26, 2022 at 03:49:22PM +0200, Frank Wunderlich wrote:
> From: Frank Wunderlich <frank-w@xxxxxxxxxxxxxxx>
> 
> Currently a reset line is required, but on BPI-R2-Pro board
> this reset is shared with the gmac and prevents the switch to
> be initialized because mdio is not ready fast enough after
> the reset.
> 
> So make the reset optional to allow shared reset lines.

What does it mean "to allow shared reset lines"? Allow as in "allow them
to sit there, unused"?

> Signed-off-by: Frank Wunderlich <frank-w@xxxxxxxxxxxxxxx>
> ---
>  drivers/net/dsa/mt7530.c | 7 +++----
>  1 file changed, 3 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/net/dsa/mt7530.c b/drivers/net/dsa/mt7530.c
> index 19f0035d4410..ccf4cb944167 100644
> --- a/drivers/net/dsa/mt7530.c
> +++ b/drivers/net/dsa/mt7530.c
> @@ -2134,7 +2134,7 @@ mt7530_setup(struct dsa_switch *ds)
>  		reset_control_assert(priv->rstc);
>  		usleep_range(1000, 1100);
>  		reset_control_deassert(priv->rstc);
> -	} else {
> +	} else if (priv->reset) {

I don't really understand this patch. gpiod_set_value_cansleep() can
tolerate NULL GPIO descriptors.

>  		gpiod_set_value_cansleep(priv->reset, 0);
>  		usleep_range(1000, 1100);
>  		gpiod_set_value_cansleep(priv->reset, 1);
> @@ -2276,7 +2276,7 @@ mt7531_setup(struct dsa_switch *ds)
>  		reset_control_assert(priv->rstc);
>  		usleep_range(1000, 1100);
>  		reset_control_deassert(priv->rstc);
> -	} else {
> +	} else if (priv->reset) {
>  		gpiod_set_value_cansleep(priv->reset, 0);
>  		usleep_range(1000, 1100);
>  		gpiod_set_value_cansleep(priv->reset, 1);
> @@ -3272,8 +3272,7 @@ mt7530_probe(struct mdio_device *mdiodev)
>  		priv->reset = devm_gpiod_get_optional(&mdiodev->dev, "reset",
>  						      GPIOD_OUT_LOW);
>  		if (IS_ERR(priv->reset)) {
> -			dev_err(&mdiodev->dev, "Couldn't get our reset line\n");
> -			return PTR_ERR(priv->reset);
> +			dev_warn(&mdiodev->dev, "Couldn't get our reset line\n");

I certainly don't understand why you're suppressing the pointer-encoded
errors here. The function used is devm_gpiod_get_optional(), which
returns NULL for a missing reset-gpios, not IS_ERR(something). The
IS_ERR(something) is actually important to not ignore, maybe it's
IS_ERR(-EPROBE_DEFER). And this change breaks waiting for the descriptor
to become available.

>  		}
>  	}
>  
> -- 
> 2.25.1
> 

So what doesn't work without this patch, exactly?