Hi Andy,
Thanks for the review. I have few questions about your remarks below.
On Fri, 2 Sept 2022 at 17:28, Andy Shevchenko <andy.shevchenko@xxxxxxxxx> wrote:
>
> On Fri, Sep 2, 2022 at 4:13 PM <cmo@xxxxxxxxxxx> wrote:
> >
> > From: Crt Mori <cmo@xxxxxxxxxxx>
> >
> > Sensor can operate in lower power modes and even make measurements when
>
> The sensor
> ...or..
> Sensors
>
> > in those lower powered modes. The decision was taken that if measurement
> > is not requested within 2 seconds the sensor will remain in SLEEP_STEP
> > power mode, where measurements are triggered on request with setting the
> > start of measurement bit (SOB). In this mode the measurements are taking
> > a bit longer because we need to start it and complete it. Currently, in
> > continuous mode we read ready data and this mode is activated if sensor
> > measurement is requested within 2 seconds. The suspend timeout is
> > increased to 6 seconds (instead of 3 before), because that enables more
> > measurements in lower power mode (SLEEP_STEP), with the lowest refresh
> > rate (2 seconds).
>
> ...
>
> > #define MLX90632_PWR_STATUS_CONTINUOUS MLX90632_PWR_STATUS(3) /* continuous*/
> >
> > +#define MLX90632_EE_RR(ee_val) (ee_val & GENMASK(10, 8)) /* Only Refresh Rate bits*/
>
> Missed space. Seems like a copy'n'paste from previous comments that
> lacks the space as well.
>
> ...
>
> > + unsigned long interraction_timestamp; /* in jiffies */
>
> _ts for timestamp is a fine abbreviation. Also move comment to the kernel doc.
>
> ...
>
> > +static int mlx90632_wakeup(struct mlx90632_data *data);
>
> Can we avoid forward declaration? (I don't even see how it is used
> among dozens of lines of below code in the patch)
>
This is existing function that I did not want to move upwards. Should
I have just moved it rather?
> > static s32 mlx90632_pwr_set_sleep_step(struct regmap *regmap)
> > {
>
> > + struct iio_dev *indio_dev = i2c_get_clientdata(to_i2c_client(regmap_get_device(regmap)));
>
> Why ping-ponging here and not using dev_get_drvdata()? Ditto for similar cases.
>
> > + struct mlx90632_data *data = iio_priv(indio_dev);
> > + s32 ret = 0;
>
> Assignment is not needed, use 'return 0;' directly. Ditto for all
> cases like this.
>
This is used, because when powerstatus is not equal to sleep_step it
returns, otherwise the ret is changed in the function.
> > + if (data->powerstatus != MLX90632_PWR_STATUS_SLEEP_STEP) {
> > + ret = regmap_write_bits(regmap, MLX90632_REG_CONTROL,
> > + MLX90632_CFG_PWR_MASK,
> > + MLX90632_PWR_STATUS_SLEEP_STEP);
> > + if (ret < 0)
> > + return ret;
> > +
> > + data->powerstatus = MLX90632_PWR_STATUS_SLEEP_STEP;
> > + }
> > + return ret;
> > }
>
> ...
>
> > + reg = MLX90632_EE_RR(reg) >> 8;
>
> This makes it harder to understand the semantics of reg, can we simply
> unite this line with the below?
>
I find it easier to have it split but I can make one long statement.
> > + return MLX90632_MEAS_MAX_TIME >> reg;
>
> ...
>
> > + refresh_time = refresh_time + ret;
>
> += ?
>
> ...
>
> > + refresh_time = refresh_time + ret;
>
> +=
>
> ...
>
> > + refresh_time = refresh_time + ret;
>
> Ditto.
>
> ...
>
> > + unsigned int reg_status;
> > int ret;
>
> Keep the reversed xmas tree order (like here!) elsewhere for the sake
> of consistency.
>
> ...
>
> > + ret = regmap_read_poll_timeout(data->regmap, MLX90632_REG_STATUS,
> > + reg_status,
> > + ((reg_status & MLX90632_STAT_BUSY) == 0),
>
> Too many parentheses
>
I like the outer parentheses it shows that it is a break condition. I
have same in another function in this file.
> > + 10000, 100 * 10000);
> > + if (ret < 0) {
> > + dev_err(&data->client->dev, "data not ready");
> > + return -ETIMEDOUT;
> > + }
>
> ...
>
> > + int current_powerstatus = data->powerstatus;
>
> Please, split the assignment and move it closer to the first user.
>
> ...
>
> > + data->powerstatus = MLX90632_PWR_STATUS_HALT;
> > +
> > + if (current_powerstatus == MLX90632_PWR_STATUS_SLEEP_STEP)
> > + return mlx90632_pwr_set_sleep_step(data->regmap);
>
> > + else
>
> Redundant.
>
No, the powermode changes among the type.
> > + return mlx90632_pwr_continuous(data->regmap);
>
> ...
>
> > + ret = read_poll_timeout(mlx90632_perform_measurement, meas, meas == 19,
> > + 50000, 800000, false, data);
> > + if (ret != 0)
>
> Drop this ' != 0' part. It's useless.
>
> > + goto read_unlock;
>
> > +
>
> Seems redundant blank line.
>
> ...
>
> > + }
> > +
> >
>
> Ditto.
>
> ...
>
> > + int ret = 0;
>
> Redundant assignment. Use return 0; directly.
>
> ...
>
> > + if (time_in_range(now, data->interraction_timestamp,
> > + data->interraction_timestamp +
>
> > + msecs_to_jiffies(MLX90632_MEAS_MAX_TIME + 100))) {
>
> With a local variable you will have better to read code.
>
> > + }
>
>
> ...
>
> > struct iio_dev *indio_dev = i2c_get_clientdata(to_i2c_client(dev));
>
> Maybe a separate patch to drop these here-there dereferences...
>
> ...
>
> > +static int __maybe_unused mlx90632_pm_runtime_suspend(struct device *dev)
>
> No __maybe_unused, use pm_ptr() / pm_sleep_ptr() below.
>
Care to explain a bit more about this? I just followed what other
drivers have...
> > +{
> > + struct iio_dev *indio_dev = i2c_get_clientdata(to_i2c_client(dev));
> > + struct mlx90632_data *data = iio_priv(indio_dev);
> > +
> > + return mlx90632_pwr_set_sleep_step(data->regmap);
> > +}
> > +
> > +const struct dev_pm_ops mlx90632_pm_ops = {
> > + SET_SYSTEM_SLEEP_PM_OPS(mlx90632_pm_suspend, mlx90632_pm_resume)
> > + SET_RUNTIME_PM_OPS(mlx90632_pm_runtime_suspend,
> > + NULL, NULL)
>
> Please, use new macros from pm.h / runtime_pm.h
>
> > +};
> > +EXPORT_SYMBOL_GPL(mlx90632_pm_ops);
>
> Can we use special EXPORT macro from pm.h
>
> --
> With Best Regards,
> Andy Shevchenko
