On 18/03/22 02:28, Guenter Roeck wrote:
> On 3/16/22 16:41, Chris Packham wrote:
>> The adt7473, adt7475, adt7476 and adt7490 have pins that can be used for
>> different functions. On the adt7473 and adt7475 this is pins 5 and 9.
>> On the adt7476 and adt7490 this is pins 10 and 14.
>>
>> The first pin can either be PWM2(default) or SMBALERT#. The second pin
>> can be TACH4(default), THERM#, SMBALERT# or GPIO.
>>
>> The adt7475 driver has always been able to detect the configuration if
>> it had been done by an earlier boot stage. Add support for configuring
>> the pins based on the hardware description in the device tree.
>>
>> Signed-off-by: Chris Packham <chris.packham@xxxxxxxxxxxxxxxxxxx>
>> ---
>> drivers/hwmon/adt7475.c | 95 +++++++++++++++++++++++++++++++++++++++++
>> 1 file changed, 95 insertions(+)
>>
>> diff --git a/drivers/hwmon/adt7475.c b/drivers/hwmon/adt7475.c
>> index 9d5b019651f2..ad5e5a7a844b 100644
>> --- a/drivers/hwmon/adt7475.c
>> +++ b/drivers/hwmon/adt7475.c
>> @@ -112,6 +112,8 @@
>> #define CONFIG3_THERM 0x02
>> #define CONFIG4_PINFUNC 0x03
>> +#define CONFIG4_THERM 0x01
>> +#define CONFIG4_SMBALERT 0x02
>> #define CONFIG4_MAXDUTY 0x08
>> #define CONFIG4_ATTN_IN10 0x30
>> #define CONFIG4_ATTN_IN43 0xC0
>> @@ -1460,6 +1462,95 @@ static int adt7475_update_limits(struct
>> i2c_client *client)
>> return 0;
>> }
>> +static int load_pin10_config(const struct i2c_client *client,
>> const char *propname)
>> +{
>
> A better function name would probably be load_config3() or similar.
Yep that'd be a better name.
>
>> + const char *function;
>> + u8 config3;
>> + int err;
>> +
>> + err = of_property_read_string(client->dev.of_node, propname,
>> &function);
>> + if (!err) {
>> + config3 = adt7475_read(REG_CONFIG3);
>
> error check missing (I see the driver is notorious for that, but that
> is not
> a reason to keep doing it).
Ikegami-san and Dan did to some good work to address some of that. The
probe function is still quite careless.
I'll see what I can do to make sure my additions don't make it worse.
>
>> +
>> + if (!strcmp("pwm2", function))
>> + config3 &= ~CONFIG3_SMBALERT;
>> + else if (!strcmp("smbalert#", function))
>> + config3 |= CONFIG3_SMBALERT;
>> + else
>> + return -EINVAL;
>> +
>> + return i2c_smbus_write_byte_data(client, REG_CONFIG3, config3);
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static int load_pin14_config(const struct i2c_client *client, const
>> char *propname)
>> +{
>
> load_config4() ?
>
>> + const char *function;
>> + u8 config4;
>> + int err;
>> +
>> + err = of_property_read_string(client->dev.of_node, propname,
>> &function);
>> + if (!err) {
>> + config4 = adt7475_read(REG_CONFIG4);
>
> error check
>
>> + config4 &= ~CONFIG4_PINFUNC;
>> +
>> + if (!strcmp("tach4", function))
>> + ;
>> + else if (!strcmp("therm#", function))
>> + config4 |= CONFIG4_THERM;
>> + else if (!strcmp("smbalert#", function))
>> + config4 |= CONFIG4_SMBALERT;
>> + else if (!strcmp("gpio", function))
>> + config4 |= CONFIG4_PINFUNC;
>> + else
>> + return -EINVAL;
>> +
>> + return i2c_smbus_write_byte_data(client, REG_CONFIG4, config4);
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static int load_config(const struct i2c_client *client, int chip)
>> +{
>> + int err;
>> + const char *conf_prop1, *conf_prop2;
>
> conf_ prefix is unnecessary.
>
>> +
>> + switch (chip) {
>> + case adt7473:
>> + case adt7475:
>> + conf_prop1 = "adi,pin5-function";
>> + conf_prop2 = "adi,pin9-function";
>> + break;
>> + case adt7476:
>> + case adt7490:
>> + conf_prop1 = "adi,pin10-function";
>> + conf_prop2 = "adi,pin14-function";
>> + break;
>> + default:
>> + return -EINVAL;
>
> It doesn't seem right to return -EINVAL here.
>
Have you got a better suggestion? I was trying to avoid someone
specifying compatible = "adi,adt7476" with "adi,pin5-function". Is your
concern that I should use -ENODEV or that I should just pick more
generic names for the configurable pins (naming things is hard).
Or perhaps just dev_warn() and return 0?
>> + }
>> +
>> + if (chip != adt7476 && chip != adt7490)
>> + return 0;
>> +
>
> Why not check this first, and what is the point of assigning values to
> conf_prop1 and conf_prop2 for the other chips in the case statement above
> only to return 0 here ? It would be much simpler to drop the other chips
> from the case statement and have default: return 0.
>
Sorry that is old. I initially was under the impression that only these
2 had configurable pins but then I read the other datasheets more closely.
>> + err = load_pin10_config(client, conf_prop1);
>> + if (err) {
>> + dev_err(&client->dev, "failed to configure PIN10\n");
>
> The messages are misleading. This isn't always pin 10/14.
>
Now I've got the prop names I can use that instead.
>> + return err;
>> + }
>> +
>> + err = load_pin14_config(client, conf_prop2);
>> + if (err) {
>> + dev_err(&client->dev, "failed to configure PIN14\n");
>> + return err;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> static int set_property_bit(const struct i2c_client *client, char
>> *property,
>> u8 *config, u8 bit_index)
>> {
>> @@ -1585,6 +1676,10 @@ static int adt7475_probe(struct i2c_client
>> *client)
>> revision = adt7475_read(REG_DEVID2) & 0x07;
>> }
>> + ret = load_config(client, chip);
>> + if (ret)
>> + return ret;
>> +
>> config3 = adt7475_read(REG_CONFIG3);
>> /* Pin PWM2 may alternatively be used for ALERT output */
>> if (!(config3 & CONFIG3_SMBALERT))
>
