Re: [PATCH v2 3/4] hwmon: (pmbus/core): Add interrupt support

From: Guenter Roeck <linux@roeck-us.net>
To: Naresh Solanki <naresh.solanki@9elements.com>
Cc: Jean Delvare <jdelvare@suse.com>,
	linux-hwmon@vger.kernel.org,
	Patrick Rudolph <patrick.rudolph@9elements.com>,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH v2 3/4] hwmon: (pmbus/core): Add interrupt support
Date: Sat, 11 Feb 2023 07:37:29 -0800	[thread overview]
Message-ID: <20230211153729.GA203244@roeck-us.net> (raw)
In-Reply-To: <20230207120241.2800662-3-Naresh.Solanki@9elements.com>

On Tue, Feb 07, 2023 at 01:02:40PM +0100, Naresh Solanki wrote:
> From: Patrick Rudolph <patrick.rudolph@9elements.com>
> 
> Implement PMBUS irq handler.
> 
> Signed-off-by: Patrick Rudolph <patrick.rudolph@9elements.com>
> Signed-off-by: Naresh Solanki <Naresh.Solanki@9elements.com>
> ---
>  drivers/hwmon/pmbus/pmbus.h      |  2 +-
>  drivers/hwmon/pmbus/pmbus_core.c | 85 ++++++++++++++++++++++++++++++++
>  2 files changed, 86 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/hwmon/pmbus/pmbus.h b/drivers/hwmon/pmbus/pmbus.h
> index 713ea7915425..11e84e141126 100644
> --- a/drivers/hwmon/pmbus/pmbus.h
> +++ b/drivers/hwmon/pmbus/pmbus.h
> @@ -26,7 +26,7 @@ enum pmbus_regs {
>  
>  	PMBUS_CAPABILITY		= 0x19,
>  	PMBUS_QUERY			= 0x1A,
> -
> +	PMBUS_SMBALERT_MASK		= 0x1B,
>  	PMBUS_VOUT_MODE			= 0x20,
>  	PMBUS_VOUT_COMMAND		= 0x21,
>  	PMBUS_VOUT_TRIM			= 0x22,
> diff --git a/drivers/hwmon/pmbus/pmbus_core.c b/drivers/hwmon/pmbus/pmbus_core.c
> index 5ccae8126a56..d5403baad60a 100644
> --- a/drivers/hwmon/pmbus/pmbus_core.c
> +++ b/drivers/hwmon/pmbus/pmbus_core.c
> @@ -3093,6 +3093,85 @@ static int pmbus_regulator_register(struct pmbus_data *data)
>  }
>  #endif
>  
> +static int pmbus_write_smbalert_mask(struct i2c_client *client, u8 page, u8 reg, u8 val)
> +{
> +	int err;
> +
> +	err = pmbus_check_word_register(client, page, reg | (val << 8));
> +	if (err)
> +		return err;

I am not convinced that this is necessary. The next command will return an
error anyway if the register or the specific mask is not supported, so what
is the point ?

> +
> +	return pmbus_write_word_data(client, page, PMBUS_SMBALERT_MASK, reg | (val << 8));
> +}
> +
> +static irqreturn_t pmbus_fault_handler(int irq, void *pdata)
> +{
> +	struct pmbus_data *data = pdata;
> +	struct i2c_client *client = to_i2c_client(data->dev);
> +	int i, status;
> +
> +	mutex_lock(&data->update_lock);
> +	for (i = 0; i < data->info->pages; i++) {
> +		status = pmbus_read_status_word(client, i);
> +		if (status < 0) {
> +			mutex_unlock(&data->update_lock);
> +			return status;
> +		}
> +
> +		if (status & ~(PB_STATUS_OFF | PB_STATUS_BUSY | PB_STATUS_POWER_GOOD_N))
> +			pmbus_clear_fault_page(client, i);
> +	}
> +	mutex_unlock(&data->update_lock);
> +
> +	return IRQ_HANDLED;
> +}
> +
> +static int pmbus_irq_setup(struct i2c_client *client, struct pmbus_data *data)
> +{
> +	struct device *dev = &client->dev;
> +	const struct pmbus_status_category *cat;
> +	const struct pmbus_status_assoc *bit;
> +	int i, j, err, ret, func;
> +	u8 mask;
> +	u8 misc_status[] = {PMBUS_STATUS_CML, PMBUS_STATUS_OTHER, PMBUS_STATUS_MFR_SPECIFIC,
> +			    PMBUS_STATUS_FAN_12, PMBUS_STATUS_FAN_34};

static const u8 ... 

> +
> +	for (i = 0; i < data->info->pages; i++) {
> +		func = data->info->func[i];
> +
> +		for (j = 0; j < ARRAY_SIZE(pmbus_status_flag_map); j++) {
> +			cat = &pmbus_status_flag_map[j];
> +			if (!(func & cat->func))
> +				continue;
> +			mask = 0;
> +			for (bit = cat->bits; bit->pflag; bit++)
> +				mask |= bit->pflag;
> +
> +			err = pmbus_write_smbalert_mask(client, i, cat->reg, ~mask);
> +			if (err)
> +				dev_err_once(dev, "Failed to set smbalert for reg 0x%02x\n",
> +					     cat->reg);

dev_err implies an error. This is ignored and thus not an error. On top of that,
not all chips support PMBUS_SMBALERT_MASK. All of those would see this message.
We can't have that. At best make it a dev_dbg.

> +		}
> +
> +		for (j = 0; j < ARRAY_SIZE(misc_status); j++) {
> +			err = pmbus_write_smbalert_mask(client, i, misc_status[j], 0xff);
> +			if (err)
> +				dev_err_once(dev, "Failed to set smbalert for reg 0x%02x\n",
> +					     misc_status[j]);

We definitely can't have a message here; that would fire for almost
every chip.

> +		}
> +	}
> +
> +	/* Register notifiers - can fail if IRQ is not given */

If there is no irq, what is the point of executing this code in the first
place ? No, wait, in that case the function isn't called in the first place.
I think the comment doesn't add any value and is just confusing.

> +	ret = devm_request_threaded_irq(dev, client->irq, NULL, pmbus_fault_handler, 0,
> +					"pmbus-irq", data);
> +	if (ret) {

Why both "err" and "ret" ?

> +		dev_warn(dev, "IRQ disabled %d\n", ret);

The calling code aborts, so this should be dev_err() and say something
better than "IRQ disabled"; It should be something like "failed to
request irq".

> +		return ret;
> +	}
> +
> +	return 0;
> +}
> +
>  static struct dentry *pmbus_debugfs_dir;	/* pmbus debugfs directory */
>  
>  #if IS_ENABLED(CONFIG_DEBUG_FS)
> @@ -3455,6 +3534,12 @@ int pmbus_do_probe(struct i2c_client *client, struct pmbus_driver_info *info)
>  	if (ret)
>  		return ret;
>  
> +	if (client->irq > 0) {
> +		ret = pmbus_irq_setup(client, data);
> +		if (ret)
> +			return ret;
> +	}
> +

I think it would be better to have the check in pmbus_irq_setup():

pmbus_irq_setup()
{
	if (!client->irq)
		return 0;

	...
}

and here
	ret = pmbus_irq_setup(client, data);
	if (ret)
		return ret;

>  	ret = pmbus_init_debugfs(client, data);
>  	if (ret)
>  		dev_warn(dev, "Failed to register debugfs\n");