Re: [PATCH v3 2/4] drivers: thermal: tsens: Add generic support for TSENS v1 IP

From: Amit Kucheria
Date: Tue Dec 04 2018 - 06:24:40 EST


On Thu, Nov 29, 2018 at 10:26 PM Eduardo Valentin <edubezval@xxxxxxxxx> wrote:
>
> On Tue, Nov 27, 2018 at 09:59:05PM +0530, Amit Kucheria wrote:

<snip>

> > + qfprom_cdata = (u32 *)qfprom_read(tmdev->dev, "calib");
> > + if (IS_ERR(qfprom_cdata))
> > + return PTR_ERR(qfprom_cdata);
> > +
> > + mode = (qfprom_cdata[4] & CAL_SEL_MASK) >> CAL_SEL_SHIFT;
> > + dev_dbg(tmdev->dev, "calibration mode is %d\n", mode);
> > +
> > + switch (mode) {
> > + case TWO_PT_CALIB:
> > + base1 = (qfprom_cdata[4] & BASE1_MASK) >> BASE1_SHIFT;
> > + p2[0] = (qfprom_cdata[0] & S0_P2_MASK) >> S0_P2_SHIFT;
> > + p2[1] = (qfprom_cdata[0] & S1_P2_MASK) >> S1_P2_SHIFT;
> > + /* This value is split over two registers, 2 bits and 4 bits */
> > + lsb = (qfprom_cdata[0] & S2_P2_MASK_1_0) >> S2_P2_SHIFT_1_0;
> > + msb = (qfprom_cdata[1] & S2_P2_MASK_5_2) >> S2_P2_SHIFT_5_2;
> > + p2[2] = msb << 2 | lsb;
> > + p2[3] = (qfprom_cdata[1] & S3_P2_MASK) >> S3_P2_SHIFT;
> > + p2[4] = (qfprom_cdata[1] & S4_P2_MASK) >> S4_P2_SHIFT;
> > + p2[5] = (qfprom_cdata[2] & S5_P2_MASK) >> S5_P2_SHIFT;
> > + p2[6] = (qfprom_cdata[2] & S6_P2_MASK) >> S6_P2_SHIFT;
> > + /* This value is split over two registers, 2 bits and 4 bits */
> > + lsb = (qfprom_cdata[2] & S7_P2_MASK_1_0) >> S7_P2_SHIFT_1_0;
> > + msb = (qfprom_cdata[3] & S7_P2_MASK_5_2) >> S7_P2_SHIFT_5_2;
> > + p2[7] = msb << 2 | lsb;
> > + p2[8] = (qfprom_cdata[3] & S8_P2_MASK) >> S8_P2_SHIFT;
> > + p2[9] = (qfprom_cdata[3] & S9_P2_MASK) >> S9_P2_SHIFT;
> > + for (i = 0; i < tmdev->num_sensors; i++)
> > + p2[i] = ((base1 + p2[i]) << 2);
> > + /* Fall through */
> > + case ONE_PT_CALIB2:
> > + base0 = (qfprom_cdata[4] & BASE0_MASK) >> BASE0_SHIFT;
> > + p1[0] = (qfprom_cdata[0] & S0_P1_MASK) >> S0_P1_SHIFT;
> > + p1[1] = (qfprom_cdata[0] & S1_P1_MASK) >> S1_P1_SHIFT;
> > + p1[2] = (qfprom_cdata[0] & S2_P1_MASK) >> S2_P1_SHIFT;
> > + p1[3] = (qfprom_cdata[1] & S3_P1_MASK) >> S3_P1_SHIFT;
> > + p1[4] = (qfprom_cdata[1] & S4_P1_MASK) >> S4_P1_SHIFT;
> > + p1[5] = (qfprom_cdata[2] & S5_P1_MASK) >> S5_P1_SHIFT;
> > + p1[6] = (qfprom_cdata[2] & S6_P1_MASK) >> S6_P1_SHIFT;
> > + p1[7] = (qfprom_cdata[2] & S7_P1_MASK) >> S7_P1_SHIFT;
> > + p1[8] = (qfprom_cdata[3] & S8_P1_MASK) >> S8_P1_SHIFT;
> > + p1[9] = (qfprom_cdata[3] & S9_P1_MASK) >> S9_P1_SHIFT;
> > + for (i = 0; i < tmdev->num_sensors; i++)
> > + p1[i] = (((base0) + p1[i]) << 2);
> > + break;
> > + default:
> > + for (i = 0; i < tmdev->num_sensors; i++) {
> > + p1[i] = 500;
> > + p2[i] = 780;
> ... Wouldn't be nice to know what 500 or 780 stands for here?

Indeed, however I haven't found a suitable explanations in the
downstream trees or documentation yet. tsens-8974.c and tsens-8916.c
seem to have similar magic numbers.

> Why these defaults? Do we expect to have further patches to keep
> updating these?

No we don't expect them to change. I wish I had a better answer, but
honestly I don't. I'll keep looking for a explanation to replace these
magic numbers.

> > + }
> > + break;
> > + }
> > +
> > + compute_intercept_slope(tmdev, p1, p2, mode);
> > +
> > + return 0;
> > +}
> > +
> > +#define STATUS_OFFSET 0x44
> > +#define LAST_TEMP_MASK 0x3ff
> > +#define STATUS_VALID_BIT BIT(14)
> > +
> > +static int get_temp_tsens_v1(struct tsens_device *tmdev, int id, int *temp)
> > +{
> > + struct tsens_sensor *s = &tmdev->sensor[id];
> > + u32 code;
> > + unsigned int status_reg;
> > + u32 last_temp = 0, last_temp2 = 0, last_temp3 = 0;
> > + int ret;
> > +
> > + status_reg = tmdev->tm_offset + STATUS_OFFSET + s->hw_id * 4;
> > + ret = regmap_read(tmdev->tm_map, status_reg, &code);
> > + if (ret)
> > + return ret;
> > + last_temp = code & LAST_TEMP_MASK;
> > + if (code & STATUS_VALID_BIT)
> > + goto done;
> > +
> > + /* Try a second time */
> > + ret = regmap_read(tmdev->tm_map, status_reg, &code);
> > + if (ret)
> > + return ret;
> > + if (code & STATUS_VALID_BIT) {
> > + last_temp = code & LAST_TEMP_MASK;
> > + goto done;
> > + } else {
> > + last_temp2 = code & LAST_TEMP_MASK;
> > + }
> > +
> > + /* Try a third/last time */
> > + ret = regmap_read(tmdev->tm_map, status_reg, &code);
> > + if (ret)
> > + return ret;
> > + if (code & STATUS_VALID_BIT) {
> > + last_temp = code & LAST_TEMP_MASK;
> > + goto done;
> > + } else {
> > + last_temp3 = code & LAST_TEMP_MASK;
> > + }
> > +
> > + if (last_temp == last_temp2)
> > + last_temp = last_temp2;
> > + else if (last_temp2 == last_temp3)
> > + last_temp = last_temp3;
> > +done:
> > + /* Convert temperature from ADC code to milliCelsius */
> > + *temp = code_to_degc(last_temp, s) * 1000;
>
> This three strikes/read approach seams awkward. Is this a broken
> sensor or are you missing the programming steps? Maybe you need to
> either wait for a IRQ on temp ready, or maybe you need simply wait
> some amount of cycles before the sensor is ready with new temp/ADC
> conversion, no?

This "algorithm" for reading data is specified in the Tsens IP HW
documentation almost word for word. The status_reg contains both the
VALID_BIT and the temperature, so we don't know if this is valid data
until we read the register. I'll be posting IRQ support soon.

Again, this algorithm is used for tsens-v2.c and tsens-8916.c so I
haven't tried to clean this up, yet. I have an experimental branch to
hide IP-specific register changes behind the regmap interface. This
should get rid of copies of these functions across the various files
but that isn't yet ready to post yet.

> Also, the goto's dont really help, as we dont really have any
> resource to rollback here..

Agreed. :-) This function could be made much clearer.

If I fix this, should I fix it in the other files right away, or can
we wait a bit more to clean this up as part of the move to regmap? I
think it'd be logically simpier to clean this up in a separate series
than introducing changes to the known working function as part of new
machine enablement.

Thanks for the review.

Regards,
Amit

> > +
> > + return 0;
> > +}
> > +
> > +static const struct tsens_ops ops_generic_v1 = {
> > + .init = init_common,
> > + .calibrate = calibrate_v1,
> > + .get_temp = get_temp_tsens_v1,
> > +};
> > +
> > +const struct tsens_data data_tsens_v1 = {
> > + .ops = &ops_generic_v1,
> > + .reg_offsets = { [SROT_CTRL_OFFSET] = 0x4 },
> > +};
> > diff --git a/drivers/thermal/qcom/tsens.c b/drivers/thermal/qcom/tsens.c
> > index f1ec9bbe4717..d0cc0c09894a 100644
> > --- a/drivers/thermal/qcom/tsens.c
> > +++ b/drivers/thermal/qcom/tsens.c
> > @@ -63,6 +63,9 @@ static const struct of_device_id tsens_table[] = {
> > }, {
> > .compatible = "qcom,msm8996-tsens",
> > .data = &data_8996,
> > + }, {
> > + .compatible = "qcom,tsens-v1",
> > + .data = &data_tsens_v1,
> > }, {
> > .compatible = "qcom,tsens-v2",
> > .data = &data_tsens_v2,
> > diff --git a/drivers/thermal/qcom/tsens.h b/drivers/thermal/qcom/tsens.h
> > index 7b7feee5dc46..f8dc96c42b94 100644
> > --- a/drivers/thermal/qcom/tsens.h
> > +++ b/drivers/thermal/qcom/tsens.h
> > @@ -90,9 +90,10 @@ char *qfprom_read(struct device *, const char *);
> > void compute_intercept_slope(struct tsens_device *, u32 *, u32 *, u32);
> > int init_common(struct tsens_device *);
> > int get_temp_common(struct tsens_device *, int, int *);
> > +int code_to_degc(u32 adc_code, const struct tsens_sensor *s);
> >
> > /* TSENS v1 targets */
> > -extern const struct tsens_data data_8916, data_8974, data_8960;
> > +extern const struct tsens_data data_8916, data_8974, data_8960, data_tsens_v1;
> > /* TSENS v2 targets */
> > extern const struct tsens_data data_8996, data_tsens_v2;
> >
> > --
> > 2.17.1
> >