Date:   Tue, 3 Jan 2023 20:56:59 +0530
From:   Naresh Solanki <naresh.solanki@...ements.com>
To:     Guenter Roeck <linux@...ck-us.net>
Cc:     devicetree@...r.kernel.org, Jean Delvare <jdelvare@...e.com>,
        Liam Girdwood <lgirdwood@...il.com>,
        Mark Brown <broonie@...nel.org>, linux-kernel@...r.kernel.org,
        linux-hwmon@...r.kernel.org,
        Patrick Rudolph <patrick.rudolph@...ements.com>
Subject: Re: [PATCH RESEND v6 1/5] hwmon: (pmbus/core): Add interrupt support

Hi Guenter

On 03-01-2023 05:56 pm, Guenter Roeck wrote:
> On Tue, Jan 03, 2023 at 12:18:49PM +0530, Naresh Solanki wrote:
>> Hi Guenter,
>>
>> On 29-12-2022 08:10 pm, Guenter Roeck wrote:
>>> On Wed, Dec 14, 2022 at 09:07:11AM +0100, Naresh Solanki wrote:
>>>> From: Patrick Rudolph <patrick.rudolph@...ements.com>
>>>>
>>>> Implement PMBUS irq handler.
>>>>
>>>> Signed-off-by: Patrick Rudolph <patrick.rudolph@...ements.com>
>>>> Signed-off-by: Naresh Solanki <Naresh.Solanki@...ements.com>
>>>
>>> $ scripts/checkpatch.pl --strict index.html
>>> CHECK: Blank lines aren't necessary after an open brace '{'
>>> #131: FILE: drivers/hwmon/pmbus/pmbus_core.c:3088:
>>> +	for (i = 0; i < data->info->pages; i++) {
>>> +
>>>
>>> CHECK: Alignment should match open parenthesis
>>> #183: FILE: drivers/hwmon/pmbus/pmbus_core.c:3140:
>>> +	ret = devm_request_threaded_irq(dev, client->irq, NULL, pmbus_fault_handler,
>>> +			      0, "pmbus-irq", data);
>>>
>>> CHECK: Please use a blank line after function/struct/union/enum declarations
>>> #197: FILE: drivers/hwmon/pmbus/pmbus_core.c:3154:
>>>    }
>>> +static int pmbus_irq_setup(struct i2c_client *client, struct pmbus_data *data)
>>>
>>> total: 0 errors, 0 warnings, 3 checks, 109 lines checked
>>>
>>> NOTE: For some of the reported defects, checkpatch may be able to
>>>         mechanically convert to the typical style using --fix or --fix-inplace.
>>>
>>> index.html has style problems, please review.
>>>
>>> Please run checkpatch --strict on your patches.
>>> Also see Documentation/hwmon/submitting-patches.rst.
>> I will take care of these errors in the updated version.
>>>
>>>> ---
>>>>    drivers/hwmon/pmbus/pmbus.h      |  2 +-
>>>>    drivers/hwmon/pmbus/pmbus_core.c | 84 ++++++++++++++++++++++++++++++++
>>>>    2 files changed, 85 insertions(+), 1 deletion(-)
>>>>
>>>>
>>>> base-commit: 364ffd2537c44cb6914ff5669153f4a86fffad29
>>>>
>>>> diff --git a/drivers/hwmon/pmbus/pmbus.h b/drivers/hwmon/pmbus/pmbus.h
>>>> index 10fb17879f8e..6b2e6cf93b19 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 95e95783972a..244fd2597252 100644
>>>> --- a/drivers/hwmon/pmbus/pmbus_core.c
>>>> +++ b/drivers/hwmon/pmbus/pmbus_core.c
>>>> @@ -3072,11 +3072,89 @@ static int pmbus_regulator_register(struct pmbus_data *data)
>>>>    	return 0;
>>>>    }
>>>> +
>>>> +static int pmbus_write_smbalert_mask(struct i2c_client *client, u8 page, u8 reg, u8 val)
>>>> +{
>>>> +	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;
>>>> +
>>>> +	for (i = 0; i < data->info->pages; i++) {
>>>> +
>>>> +		mutex_lock(&data->update_lock);
>>>> +		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_regulator_status_category *cat;
>>>> +	const struct pmbus_regulator_status_assoc *bit;
>>>> +	int i, j, err, ret, func;
>>>> +	u8 mask;
>>>> +
>>>> +	for (i = 0; i < data->info->pages; i++) {
>>>> +		func = data->info->func[i];
>>>> +
>>>> +		for (j = 0; j < ARRAY_SIZE(pmbus_regulator_flag_map); j++) {
>>>> +			cat = &pmbus_regulator_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(dev, "Failed to set smbalert for reg 0x%02x\n",	cat->reg);
>>>
>>> This concerns me. It might mean that the chip does not support
>>> PMBUS_SMBALERT_MASK. If so, there would be lots of error messages.
>> After going through the PMBus specification, it appears that this should not
>> be an issue unless there is a violation of the specification.
> 
> PMBus chips have lots of issues which violate the specification.
> Have a look at the various drivers and the workarounds implemented there.
> You'll need to check if the command/register is supported before using it.
> Also, if you want to keep the error message, make it dev_err_once().
> 
> Either case, an error is an error, not to be ignored. An error here
> should result in an error abort.
Yes, I agree that PMBus chips can have issues that violate the 
specification, and that it is important to check whether a command or 
register is supported before using it.
I have noticed that many drivers use the PMBUS_HAVE_* flags to expose 
the presence of specific registers, and I think it would be a good idea 
to add a PMBUS_HAVE_SMBALERT flag as well, so that drivers for supported 
chips can use it to determine whether they should set up an IRQ handler 
or not. If PMBUS_HAVE_SMBALERT is set, then the IRQ handler should be 
set up, otherwise it should be ignored.
Will this approach be right?
> 
>>>
>>>> +		}
>>>> +
>>>> +		pmbus_write_smbalert_mask(client, i, PMBUS_STATUS_CML, 0xff);
>>>> +		pmbus_write_smbalert_mask(client, i, PMBUS_STATUS_OTHER, 0xff);
>>>> +		pmbus_write_smbalert_mask(client, i, PMBUS_STATUS_MFR_SPECIFIC, 0xff);
>>>
>>> Why check the return value from pmbus_write_smbalert_mask above but not here ?
>> Thank you for pointing out the oversight. I will make sure to include an
>> error check at this point.
>>>
>>>> +		if (data->info->func[i] & PMBUS_HAVE_FAN12)
>>>> +			pmbus_write_smbalert_mask(client, i, PMBUS_STATUS_FAN_12, 0xff);
>>>> +		if (data->info->func[i] & PMBUS_HAVE_FAN34)
>>>> +			pmbus_write_smbalert_mask(client, i, PMBUS_STATUS_FAN_34, 0xff);
>>>> +	}
>>>> +
>>>> +	/* Register notifiers - can fail if IRQ is not given */
>>>
>>> The comment does not make sense. pmbus_irq_setup() is not called
>>> if the interrupt "is not given".
>> Yes. The comment here is not relevant and will be removed.
>>>
>>>> +	ret = devm_request_threaded_irq(dev, client->irq, NULL, pmbus_fault_handler,
>>>> +			      0, "pmbus-irq", data);
>>>> +	if (ret) {
>>>> +		dev_warn(dev, "IRQ disabled %d\n", ret);
>>>
>>> This is not a warning, it is an error.
>> Thank you for bringing this to my attention. I will make sure to update the
>> code to reflect that this is an error.
>>>
>>>> +		return ret;
>>>> +	}
>>>> +
>>>> +	return 0;
>>>> +}
>>>> +
>>>>    #else
>>>
>>> This is still in regulator code. I said several times that this is not
>>> acceptable.
>> Thank you for pointing out the mistake. I will make sure to correct this in
>> the next revision.
>>>
>>>>    static int pmbus_regulator_register(struct pmbus_data *data)
>>>>    {
>>>>    	return 0;
>>>>    }
>>>> +static int pmbus_irq_setup(struct i2c_client *client, struct pmbus_data *data)
>>>> +{
>>>> +	return 0;
>>>> +}
>>>>    #endif
>>>>    static struct dentry *pmbus_debugfs_dir;	/* pmbus debugfs directory */
>>>> @@ -3441,6 +3519,12 @@ 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;
>>>> +	}
>>>> +
>>>>    	ret = pmbus_init_debugfs(client, data);
>>>>    	if (ret)
>>>>    		dev_warn(dev, "Failed to register debugfs\n");
>>
>> Thanks,
>> Naresh

Powered by blists - more mailing lists