Re: [PATCH 2/2 resend] iio:temperature:mlx90632: Adding extended calibration option

[Crt Mori]

On Fri, 7 Aug 2020 at 12:29, Andy Shevchenko <andy.shevchenko@xxxxxxxxx> wrote:
>
> On Fri, Aug 7, 2020 at 12:21 PM Crt Mori <cmo@xxxxxxxxxxx> wrote:
>
> Oh yeah, you are right, there will be some comments :-)
>

Told ya. No matter how many times I go through it, I always find
something. I will prepare v3 with fixes, except for some additional
questions below.

> > For some time market wants medical grade accuracy in medical range,
>
> the market
>
> > while still retaining the declared accuracy outside of the medical range
> > within the same sensor. That is why we created extended calibration
> > which is automatically switched to when object temperature is too high.
> >
> > This patch also introduces the object_ambient_temperature variable which
> > is needed for more accurate calculation of the object infra-red
> > footprint as sensor's ambient temperature might be totally different
> > than what the ambient temperature is at object and that is why we can
> > have some more error which can be eliminated. Currently this temperature
>
> errors
>
> > is fixed at 25, but interface to adjust it by user (with external sensor
>
> the interface
>
> > or just IR measurement of the another object which acts as ambient),
>
> 'of another' or 'the other' if we know what it is exactly.
>
> > will be introduced in another commit.
>
> ...
>
> >  struct mlx90632_data {
> >         struct i2c_client *client;
> >         struct mutex lock; /* Multiple reads for single measurement */
> >         struct regmap *regmap;
> >         u16 emissivity;
>
> > +       u8 mtyp; /* measurement type - to enable extended range calculations */
>
> Perhaps better to switch this struct to follow kernel doc in one of
> preparatory patches and add the description of this field accordingly.
>

Can you explain a bit more? I was looking in kernel doc, but could not
find much about how to comment these members.

> > +       u32 object_ambient_temperature;
> >  };
>
> ...
>
> > +static int mlx90632_set_meas_type(struct regmap *regmap, u8 type)
> > +{
> > +       int ret;
> > +
> > +       if ((type != MLX90632_MTYP_MEDICAL) & (type != MLX90632_MTYP_EXTENDED))
> > +               return -EINVAL;
>
> Not sure I understand the point of & vs. && here.
>

Should indeed be &&, if it is needed at all. Both are boolean types.

> > +       ret = regmap_write(regmap, MLX90632_REG_I2C_CMD, MLX90632_RESET_CMD);
> > +       if (ret < 0)
> > +               return ret;
> > +
> > +       ret = regmap_write_bits(regmap, MLX90632_REG_CONTROL,
> > +                                (MLX90632_CFG_MTYP_MASK | MLX90632_CFG_PWR_MASK),
> > +                                (MLX90632_MTYP_STATUS(type) | MLX90632_PWR_STATUS_HALT));
> > +       if (ret < 0)
> > +               return ret;
> > +
> > +       mlx90632_pwr_continuous(regmap);
>
> > +
> > +       return ret;
>
> Since you are using ' < 0' above and below (and I think it doesn't
> worth it, i.o.w. you may drop them) here is something interesting
> might be returned (actually not, see first part of this sentence).
> Should be
>
> return 0;
>
> > +}
>
> ...
>
> > +static int mlx90632_read_ambient_raw_extended(struct regmap *regmap,
> > +                                             s16 *ambient_new_raw, s16 *ambient_old_raw)
> > +{
>
> > +       int ret;
> > +       unsigned int read_tmp;
>
> Please keep them in reversed xmas tree format.
>
> > +
> > +       ret = regmap_read(regmap, MLX90632_RAM_3(17), &read_tmp);
> > +       if (ret < 0)
> > +               return ret;
> > +       *ambient_new_raw = (s16)read_tmp;
> > +
> > +       ret = regmap_read(regmap, MLX90632_RAM_3(18), &read_tmp);
> > +       if (ret < 0)
> > +               return ret;
> > +       *ambient_old_raw = (s16)read_tmp;
>
> > +       return ret;
>
> Same comments as per previous function.
>
> > +}
>
> > +static int mlx90632_read_object_raw_extended(struct regmap *regmap, s16 *object_new_raw)
> > +{
> > +       int ret;
> > +       unsigned int read_tmp;
> > +       s32 read;
>
> Besides all above comments being applicable here...
>
> > +       ret = regmap_read(regmap, MLX90632_RAM_1(17), &read_tmp);
> > +       if (ret < 0)
> > +               return ret;
> > +       read = (s16)read_tmp;
> > +
> > +       ret = regmap_read(regmap, MLX90632_RAM_2(17), &read_tmp);
> > +       if (ret < 0)
> > +               return ret;
> > +       read = read - (s16)read_tmp;
>
> ...I'm wondering if you can use bulk reads of those registers.

I cant, sensor does not support it and single read case did not work
few years back, but maybe regmap now improved...

> Also I'm not sure you need explicit castings.
>
> > +       ret = regmap_read(regmap, MLX90632_RAM_1(18), &read_tmp);
> > +       if (ret < 0)
> > +               return ret;
> > +       read = read - (s16)read_tmp;
> > +
> > +       ret = regmap_read(regmap, MLX90632_RAM_2(18), &read_tmp);
> > +       if (ret < 0)
> > +               return ret;
> > +       read = (read + (s16)read_tmp) / 2;
>
> Ditto.
>
> > +       ret = regmap_read(regmap, MLX90632_RAM_1(19), &read_tmp);
> > +       if (ret < 0)
> > +               return ret;
> > +       read = read + (s16)read_tmp;
> > +
> > +       ret = regmap_read(regmap, MLX90632_RAM_2(19), &read_tmp);
> > +       if (ret < 0)
> > +               return ret;
> > +       read = read + (s16)read_tmp;
>
> > +       if (read > 32767 || read < -32768)
>
> These are defined as S16_MIN and S16_MAX. Use limits.h.
>
> > +               return -EINVAL;
>
> -ERANGE
>
> > +       *object_new_raw = (int16_t)read;
>
> Oh, no. Please avoid user space types in the kernel. And what's the
> point anyway after checking the range?
>
> > +       return ret;
> > +}
>
> ...
>
> > +static int mlx90632_read_all_channel_extended(struct mlx90632_data *data, s16 *object_new_raw,
> > +                                             s16 *ambient_new_raw, s16 *ambient_old_raw)
> > +{
> > +       s32 ret;
> > +       int tries = 4;
> > +
> > +       mutex_lock(&data->lock);
> > +       ret = mlx90632_set_meas_type(data->regmap, MLX90632_MTYP_EXTENDED);
> > +       if (ret < 0)
> > +               goto read_unlock;
>
>
> > +       while (tries-- > 0) {
> > +               ret = mlx90632_perform_measurement(data);
> > +               if (ret < 0)
> > +                       goto read_unlock;
> > +
>
> > +               if (ret == 19)
>
> It's funny. What does this magic mean?
>

That we should break the loop once channels up to 19 are filled (we
read 17 18 and 19 in this case, we read 1 2 in normal case). A comment
maybe here?

> > +                       break;
> > +       }
> > +       if (tries < 0) {
> > +               ret = -ETIMEDOUT;
> > +               goto read_unlock;
> > +       }
>
> Timeout loops are much better in a following style
>
> unsigned int iterations = 4;
>
> do {
>   ...
> } while (--iterations);
> if (!iterations) {
>   ...-ETIMEDOUT...
> }
>
> Besides that consider the iopoll.h APIs, perhaps it may be applied here.
>
> > +       ret = mlx90632_read_object_raw_extended(data->regmap, object_new_raw);
> > +       if (ret < 0)
> > +               goto read_unlock;
> > +
> > +       ret = mlx90632_read_ambient_raw_extended(data->regmap, ambient_new_raw, ambient_old_raw);
> > +
> > +read_unlock:
> > +       (void) mlx90632_set_meas_type(data->regmap, MLX90632_MTYP_MEDICAL);
> > +
> > +       mutex_unlock(&data->lock);
> > +       return ret;
> > +}
>
> ...
>
> > +static s64 mlx90632_preprocess_temp_obj_extended(s16 object_new_raw, s16 ambient_new_raw,
> > +                                                s16 ambient_old_raw, s16 Ka)
> > +{
> > +       s64 VR_IR, kKa, tmp;
> > +
> > +       kKa = ((s64)Ka * 1000LL) >> 10ULL;
> > +       VR_IR = (s64)ambient_old_raw * 1000000LL +
> > +               kKa * div64_s64(((s64)ambient_new_raw * 1000LL),
> > +                       (MLX90632_REF_3));
>
> And the point of using parentheses? It's not a Lisp language :-)
> (Applicable everywhere in your code, the rule of thumb that any
> particular comment given by reviewer should be considered against
> entire code where it's appropriate)
>
> > +       tmp = div64_s64(
> > +                       div64_s64((((s64)object_new_raw) * 1000000000000LL), MLX90632_REF_12),
> > +                       VR_IR);
> > +       return div64_s64((tmp << 19ULL), 1000LL);
> > +}
>
> ...
>
> > +       TAdut = div64_s64(((ambient - kTA0) * 1000000LL), kTA) + 25 * 1000000LL;
> > +       Tr4 = (div64_long(reflected, 10) + 27315) *
> > +               (div64_long(reflected, 10) + 27315) *
> > +               (div64_long(reflected, 10) + 27315) *
> > +               (div64_long(reflected, 10) + 27315);
> > +       TAdut4 = (div64_s64(TAdut, 10000LL) + 27315) *
> > +               (div64_s64(TAdut, 10000LL) + 27315) *
> > +               (div64_s64(TAdut, 10000LL)  + 27315) *
> > +               (div64_s64(TAdut, 10000LL) + 27315);
>
> Okay, looking at this I definitely think that this patch should be
> split into a few smaller logically separated pieces like introducing
> some helpers to calculate above with them.
>
> ...
>
> > +       mlx90632->object_ambient_temperature = 25000; /* 25 degrees Celsius */
>
> Comment is lying. milliCelsius.
>
> --
> With Best Regards,
> Andy Shevchenko