Re: [PATCH] Add Silicom Platform Driver

[Guenter Roeck <linux@roeck-us.net>]

On Sat, Aug 19, 2023 at 02:20:32PM +0000, Huibin Shi wrote:
> Hi Guenter,
> 
> Thanks for your comments. Probably, I should not resubmit patch too rushed. I will add version number to subject and change log in cover letter for next resubmission.
> 
> See my comments below. Please let me know whether you accept my explanation.
> 
> Henry
> -----Original Message-----
[ ... ]

> > +
> > +static u32 temp_get(void)
> > +{
> > +     u32 reg;
> > +
> > +     mutex_lock(&mec_io_mutex);
> > +     /* Select memory region */
> > +     outb(IO_REG_BANK, EC_ADDR_MSB);
> > +     outb(0xc, EC_ADDR_LSB);
> > +     /* Get current data from the address */
> > +     reg = inl(MEC_DATA(DEFAULT_CHAN_LO));
> > +     mutex_unlock(&mec_io_mutex);
> > +
> > +     return (reg >> 16) / 10;
> 
> The hwmon ABI expects temperatures to be reported in milli-degrees C.
> The above sets the maximum temperature to 65,535 / 10 = 6,553 milli-degrees or 6.553 degrees C. It is very unlikely that this is correct.
> 
> Again, I commented on this before.
> 
> Henry: this is due to an internal implementation of MIcor-controller firmware, instead of putting real temperature to the register, it put (real temperature * 10 ) to the register. So, in order to report correct temperature to user space application, the read value is divided by 10, then report to user space. 
> 
> Please let me know if you accept this. If not, I can change the code, but let user space application to do adjustment. 

No, I do not accept this. I do not believe that the maximum temperature
reported by the microcontroller is 6.553 degrees C. I suspect it reports
10th of degrees C. In that case, the number reported should be multiplied
by 100 to make it milli-degrees C as expected by the ABI.

[ ... ]

> > +static int silicom_fan_control_read_labels(struct device *dev, enum hwmon_sensor_types type,
> > +                                        u32 attr, int channel, const 
> > +char **str) {
> > +     switch (type) {
> > +     case hwmon_fan:
> > +             *str = "Fan Speed";
> > +             return 0;
> > +     case hwmon_temp:
> > +             *str = "Thermostat Sensor";
> > +             return 0;
> 
> Those labels have no practical value.
> 
> Henry: Those labels will be used by user space code to identify Silicom_platform driver. 
> 

The driver is identified by the driver name, not by the name of
a temperature sensor or fan speed attribute. Any other driver
could return "Fan Speed" or "Thermostat Sensor" here. Userspace
relying on such values to identify the driver are simply broken.

> > +     default:
> > +             return -EOPNOTSUPP;
> > +     }
> > +}
> > +
> > +static const struct hwmon_ops silicom_fan_control_hwmon_ops = {
> > +     .is_visible = silicom_fan_control_is_visible,
> > +     .read = silicom_fan_control_read,
> > +     .write = NULL,
> 
> Unnecessary.
> 
> Henry: OK, will be removed silicom_fan_control_is_visible.

The NULL pointer assignment is unnecessary. I have no idea
what that has to do with silicom_fan_control_is_visible(),
or why you would want to remove that function.

Guenter