Message-ID: <5368e89c391e1329ed2be6496655f1bb50acfd19.camel@pengutronix.de>
Subject: Re: [PATCH v6 3/4] clk: ralink: make system controller node a reset
 provider
From:   Philipp Zabel <p.zabel@pengutronix.de>
To:     Sergio Paracuellos <sergio.paracuellos@gmail.com>,
        linux-clk@vger.kernel.org
Cc:     john@phrozen.org, linux-staging@lists.linux.dev,
        gregkh@linuxfoundation.org, neil@brown.name,
        linux-kernel@vger.kernel.org, sboyd@kernel.org
Date:   Mon, 10 Jan 2022 10:13:21 +0100
In-Reply-To: <20220110083910.1351020-4-sergio.paracuellos@gmail.com>
References: <20220110083910.1351020-1-sergio.paracuellos@gmail.com>
         <20220110083910.1351020-4-sergio.paracuellos@gmail.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 8BIT
User-Agent: Evolution 3.38.3-1 
MIME-Version: 1.0
Precedence: bulk

Hi Sergio,

On Mon, 2022-01-10 at 09:39 +0100, Sergio Paracuellos wrote:
> +static int mt7621_assert_device(struct reset_controller_dev *rcdev,
> +				unsigned long id)
> +{
> +	struct mt7621_rst *data = to_mt7621_rst(rcdev);
> +	struct regmap *sysc = data->sysc;
> +
> +	if (id == MT7621_RST_SYS)
> +		return -EINVAL;

Better implement the .of_xlate callback and check there instead.

That way it would fail on reset_control_get() rather than handing out a
valid reset controller that just doesn't work.

> +
> +	return regmap_update_bits(sysc, SYSC_REG_RESET_CTRL, BIT(id), BIT(id));
> +}
> +
> +static int mt7621_deassert_device(struct reset_controller_dev *rcdev,
> +				  unsigned long id)
> +{
> +	struct mt7621_rst *data = to_mt7621_rst(rcdev);
> +	struct regmap *sysc = data->sysc;
> +
> +	if (id == MT7621_RST_SYS)
> +		return -EINVAL;

Same as above.

> +	return regmap_update_bits(sysc, SYSC_REG_RESET_CTRL, BIT(id), 0);
> +}
> +
> +static int mt7621_reset_device(struct reset_controller_dev *rcdev,
> +			       unsigned long id)
> +{
> +	int ret;
> +
> +	ret = mt7621_assert_device(rcdev, id);
> +	if (ret < 0)
> +		return ret;
> +
> +	return mt7621_deassert_device(rcdev, id);
> +}

Is this known to work for all possible users, without delay between
assert and deassert?

Are there any users of the reset_control_reset() API at all? This API
was added for self-clearing reset bits, so if there are no users that
need to pretend this is a reset pulse at the hardware level (there may
be), I'd prefer if this was just left out.

Apart from this, this looks good to me.

regards
Philipp