Hi Jonathan, Peter,
Apologies for not getting back to you sooner. I got caught up on other
work and wasn't able to dedicate time to this earlier. Hopefully, this
time around, I'll be able to get this to the finish line :-)
I left out IIO_VAL_INT overflows for now, so that I can focus on getting
the rest of these changes pulled in, but I don't mind adding a patch for
that later on.
This series focuses on adding temperature rescaling support to the IIO
Analog Front End (AFE) driver.
The first few patches address minor bugs in IIO inkernel functions, and
prepare the AFE driver for the additional features.
The main changes to the AFE driver include an initial Kunit test suite,
support for IIO_VAL_INT_PLUS_{NANO,MICRO} scales, and support for RTDs
and temperature transducer sensors.
Thanks for your time,
Liam
Changes since v8:
- reword comment
- fix erroneous 64-bit division
- optimize and use 32-bit divisions when values are know to not overflow
- keep IIO_VAL_FRACTIONAL scale when possible, if not default to fixed
point
- add test cases
- use nano precision in test cases
- simplify offset calculation in rtd_props()
Changes since v7:
- drop gcd() logic in rescale_process_scale()
- use div_s64() instead of do_div() for signed 64-bit divisions
- combine IIO_VAL_FRACTIONAL and IIO_VAL_FRACTIONAL_LOG2 scale cases
- switch to INT_PLUS_NANO when accuracy is lost with FRACTIONAL scales
- rework test logic to allow for small relative error
- rename test variables to align error output messages
Changes since v6:
- rework IIO_VAL_INT_PLUS_{NANO,MICRO} based on Peter's suggestion
- combine IIO_VAL_INT_PLUS_{NANO,MICRO} cases
- add test cases for negative IIO_VAL_INT_PLUS_{NANO,MICRO} corner cases
- force use of positive integers with gcd()
- reduce risk of integer overflow in IIO_VAL_FRACTIONAL_LOG2
- fix duplicate symbol build error
- apply Reviewed-by
Changes since v5:
- add include/linux/iio/afe/rescale.h
- expose functions use to process scale and offset
- add basic iio-rescale kunit test cases
- fix integer overflow case
- improve precision for IIO_VAL_FRACTIONAL_LOG2
Changes since v4:
- only use gcd() when necessary in overflow mitigation
- fix INT_PLUS_{MICRO,NANO} support
- apply Reviewed-by
- fix temperature-transducer bindings
Changes since v3:
- drop unnecessary fallthrough statements
- drop redundant local variables in some calculations
- fix s64 divisions on 32bit platforms by using do_div
- add comment describing iio-rescaler offset calculation
- drop unnecessary MAINTAINERS entry
Changes since v2:
- don't break implicit offset truncations
- make a best effort to get a valid value for fractional types
- drop return value change in iio_convert_raw_to_processed_unlocked()
- don't rely on processed value for offset calculation
- add INT_PLUS_{MICRO,NANO} support in iio-rescale
- revert generic implementation in favor of temperature-sense-rtd and
temperature-transducer
- add separate section to MAINTAINERS file
Changes since v1:
- rebase on latest iio `testing` branch
- also apply consumer scale on integer channel scale types
- don't break implicit truncation in processed channel offset
calculation
- drop temperature AFE flavors in favor of a simpler generic
implementation
Liam Beguin (14):
iio: inkern: apply consumer scale on IIO_VAL_INT cases
iio: inkern: apply consumer scale when no channel scale is available
iio: inkern: make a best effort on offset calculation
iio: afe: rescale: expose scale processing function
iio: afe: rescale: add INT_PLUS_{MICRO,NANO} support
iio: afe: rescale: add offset support
iio: afe: rescale: use s64 for temporary scale calculations
iio: afe: rescale: reduce risk of integer overflow
iio: afe: rescale: fix accuracy for small fractional scales
iio: test: add basic tests for the iio-rescale driver
iio: afe: rescale: add RTD temperature sensor support
iio: afe: rescale: add temperature transducers
dt-bindings: iio: afe: add bindings for temperature-sense-rtd
dt-bindings: iio: afe: add bindings for temperature transducers
.../iio/afe/temperature-sense-rtd.yaml | 101 +++
.../iio/afe/temperature-transducer.yaml | 114 +++
drivers/iio/afe/iio-rescale.c | 271 ++++++-
drivers/iio/inkern.c | 40 +-
drivers/iio/test/Kconfig | 10 +
drivers/iio/test/Makefile | 1 +
drivers/iio/test/iio-test-rescale.c | 705 ++++++++++++++++++
include/linux/iio/afe/rescale.h | 34 +
8 files changed, 1232 insertions(+), 44 deletions(-)
create mode 100644 Documentation/devicetree/bindings/iio/afe/temperature-sense-rtd.yaml
create mode 100644 Documentation/devicetree/bindings/iio/afe/temperature-transducer.yaml
create mode 100644 drivers/iio/test/iio-test-rescale.c
create mode 100644 include/linux/iio/afe/rescale.h
Range-diff against v8:
1: 42a7a1047edc = 1: ae3cc93baee6 iio: inkern: apply consumer scale on IIO_VAL_INT cases
2: a1cd89fdad11 = 2: 06f66e7f7403 iio: inkern: apply consumer scale when no channel scale is available
3: ed0721fb6bd1 = 3: 2dbf6b3bbaeb iio: inkern: make a best effort on offset calculation
4: f8fb78bb1112 = 4: b083cf307268 iio: afe: rescale: expose scale processing function
5: 504b7a3f830b ! 5: a0bde29ecc8c iio: afe: rescale: add INT_PLUS_{MICRO,NANO} support
@@ drivers/iio/afe/iio-rescale.c: int rescale_process_scale(struct rescale *rescale
+ else
+ mult = 1000000LL;
+ /*
-+ * For IIO_VAL_INT_PLUS_{MICRO,NANO} scale types if *val OR
-+ * *val2 is negative the schan scale is negative
++ * For IIO_VAL_INT_PLUS_{MICRO,NANO} scale types if either *val
++ * OR *val2 is negative the schan scale is negative, i.e.
++ * *val = 1 and *val2 = -0.5 yields -1.5 not -0.5.
+ */
+ neg = *val < 0 || *val2 < 0;
+
6: c254e9ae813e = 6: c3d0e6248033 iio: afe: rescale: add offset support
7: ee8814d6abe4 = 7: 2a81fa735103 iio: afe: rescale: use s64 for temporary scale calculations
8: 62cdcfbc9836 = 8: 8315548d0fce iio: afe: rescale: reduce risk of integer overflow
9: 88309a5136ee ! 9: 223ed0569cd2 iio: afe: rescale: fix accuracy for small fractional scales
@@ drivers/iio/afe/iio-rescale.c: int rescale_process_scale(struct rescale *rescale
+
+ tmp = div_s64_rem(tmp, 1000000000LL, &rem);
*val = tmp;
+- return scale_type;
++
++ if (!rem)
++ return scale_type;
+
-+ /*
-+ * For small values, the approximation can be costly,
-+ * change scale type to maintain accuracy.
-+ *
-+ * 100 vs. 10000000 NANO caps the error to about 100 ppm.
-+ */
+ if (scale_type == IIO_VAL_FRACTIONAL)
+ tmp = *val2;
+ else
+ tmp = 1 << *val2;
+
-+ if (abs(rem) > 10000000 && abs(*val / tmp) < 100) {
-+ *val = div_s64_rem(*val, tmp, &rem2);
-+
-+ *val2 = div_s64(rem, tmp);
-+ if (rem2)
-+ *val2 += div_s64(rem2 * 1000000000LL, tmp);
++ rem2 = *val % (int)tmp;
++ *val = *val / (int)tmp;
+
-+ return IIO_VAL_INT_PLUS_NANO;
-+ }
++ *val2 = rem / (int)tmp;
++ if (rem2)
++ *val2 += div_s64((s64)rem2 * 1000000000LL, tmp);
+
- return scale_type;
++ return IIO_VAL_INT_PLUS_NANO;
case IIO_VAL_INT_PLUS_NANO:
case IIO_VAL_INT_PLUS_MICRO:
+ if (scale_type == IIO_VAL_INT_PLUS_NANO)
10: fb505a9f42f1 ! 10: 90044efdf8be iio: test: add basic tests for the iio-rescale driver
@@ drivers/iio/test/Makefile
# Keep in alphabetical order
+obj-$(CONFIG_IIO_RESCALE_KUNIT_TEST) += iio-test-rescale.o ../afe/iio-rescale.o
obj-$(CONFIG_IIO_TEST_FORMAT) += iio-test-format.o
+ CFLAGS_iio-test-format.o += $(DISABLE_STRUCTLEAK_PLUGIN)
## drivers/iio/test/iio-test-rescale.c (new) ##
@@
@@ drivers/iio/test/iio-test-rescale.c (new)
+ * Use cases with small scales involving divisions
+ */
+ {
++ .name = "small IIO_VAL_FRACTIONAL, 261/509 scaled by 90/1373754273",
++ .numerator = 261,
++ .denominator = 509,
++ .schan_scale_type = IIO_VAL_FRACTIONAL,
++ .schan_val = 90,
++ .schan_val2 = 1373754273,
++ .expected = "0.000000033594",
++ },
++ {
++ .name = "small IIO_VAL_FRACTIONAL, 90/1373754273 scaled by 261/509",
++ .numerator = 90,
++ .denominator = 1373754273,
++ .schan_scale_type = IIO_VAL_FRACTIONAL,
++ .schan_val = 261,
++ .schan_val2 = 509,
++ .expected = "0.000000033594",
++ },
++ {
++ .name = "small IIO_VAL_FRACTIONAL, 760/1373754273 scaled by 427/2727",
++ .numerator = 760,
++ .denominator = 1373754273,
++ .schan_scale_type = IIO_VAL_FRACTIONAL,
++ .schan_val = 427,
++ .schan_val2 = 2727,
++ .expected = "0.000000086626",
++ },
++ {
++ .name = "small IIO_VAL_FRACTIONAL, 761/1373754273 scaled by 427/2727",
++ .numerator = 761,
++ .denominator = 1373754273,
++ .schan_scale_type = IIO_VAL_FRACTIONAL,
++ .schan_val = 427,
++ .schan_val2 = 2727,
++ .expected = "0.000000086740",
++ },
++ {
++ .name = "small IIO_VAL_FRACTIONAL, 5/32768 scaled by 3/10000",
++ .numerator = 5,
++ .denominator = 32768,
++ .schan_scale_type = IIO_VAL_FRACTIONAL,
++ .schan_val = 3,
++ .schan_val2 = 10000,
++ .expected = "0.0000000457763671875",
++ },
++ {
+ .name = "small IIO_VAL_FRACTIONAL, 0 < scale < 1",
+ .numerator = 6,
+ .denominator = 6,
@@ drivers/iio/test/iio-test-rescale.c (new)
+ .expected = "-1.3333333333333333",
+ },
+ {
++ .name = "small IIO_VAL_FRACTIONAL_LOG2, 760/32768 scaled by 15/22",
++ .numerator = 760,
++ .denominator = 32768,
++ .schan_scale_type = IIO_VAL_FRACTIONAL_LOG2,
++ .schan_val = 15,
++ .schan_val2 = 22,
++ .expected = "0.000000082946",
++ },
++ {
++ .name = "small IIO_VAL_FRACTIONAL_LOG2, 761/32768 scaled by 15/22",
++ .numerator = 761,
++ .denominator = 32768,
++ .schan_scale_type = IIO_VAL_FRACTIONAL_LOG2,
++ .schan_val = 15,
++ .schan_val2 = 22,
++ .expected = "0.000000083055",
++ },
++ {
+ .name = "small IIO_VAL_FRACTIONAL_LOG2, 0 < scale < 1",
+ .numerator = 16,
+ .denominator = 3,
@@ drivers/iio/test/iio-test-rescale.c (new)
+KUNIT_ARRAY_PARAM(iio_rescale_offset, offset_cases, case_to_desc);
+
+/**
-+ * iio_str_to_micro() - Parse a fixed-point string to get an
-+ * IIO_VAL_INT_PLUS_MICRO value
++ * iio_str_to_nano() - Parse a fixed-point string to get an
++ * IIO_VAL_INT_PLUS_NANO value
+ * @str: The string to parse
-+ * @micro: The number as an integer
++ * @nano: The number as an integer
+ *
+ * Returns 0 on success, or a negative error code if the string cound not be
+ * parsed.
+ */
-+static int iio_str_to_micro(const char *str, s64 *micro)
++static int iio_str_to_nano(const char *str, s64 *nano)
+{
-+ int fract_mult = 100000LL;
++ int fract_mult = 100000000LL;
+ int tmp, tmp2;
+ int ret = 0;
+
@@ drivers/iio/test/iio-test-rescale.c (new)
+ if (tmp < 0)
+ tmp2 *= -1;
+
-+ *micro = (s64)tmp * 10 * fract_mult + tmp2;
++ *nano = (s64)tmp * 10 * fract_mult + tmp2;
+
+ return ret;
+}
+
+/**
-+ * iio_test_relative_error_ppm() - Compute relative error (in ppm) between two
-+ * fixed-point strings
++ * iio_test_relative_error_ppm() - Compute relative error (in parts-per-million)
++ * between two fixed-point strings
+ * @real_str: The real value as a string
+ * @exp_str: The expected value as a string
+ *
+ * Returns a negative error code if the strings cound not be parsed, or the
-+ * relative error in ppm.
++ * relative error in parts-per-million.
+ */
+static int iio_test_relative_error_ppm(const char *real_str, const char *exp_str)
+{
+ s64 real, exp, err;
+ int ret;
+
-+ ret = iio_str_to_micro(real_str, &real);
++ ret = iio_str_to_nano(real_str, &real);
+ if (ret < 0)
+ return ret;
+
-+ ret = iio_str_to_micro(exp_str, &exp);
++ ret = iio_str_to_nano(exp_str, &exp);
+ if (ret < 0)
+ return ret;
+
++ if (!exp) {
++ pr_err("Expected value is null, relative error is undefined\n");
++ return -EINVAL;
++ }
++
+ err = 1000000 * abs(exp - real);
+ err = div64_u64(err, abs(exp));
+ return (int)err;
@@ drivers/iio/test/iio-test-rescale.c (new)
+ rel_ppm = iio_test_relative_error_ppm(buff, t->expected);
+ KUNIT_EXPECT_GE_MSG(test, rel_ppm, 0, "failed to compute ppm\n");
+
-+ KUNIT_EXPECT_LT_MSG(test, rel_ppm, 500,
++ KUNIT_EXPECT_EQ_MSG(test, rel_ppm, 0,
+ "\t real=%s"
+ "\texpected=%s\n",
+ buff, t->expected);
11: 050487186e14 = 11: c4ed463e5fb0 iio: afe: rescale: add RTD temperature sensor support
12: f36a44a5d898 ! 12: ff2f0dc248a7 iio: afe: rescale: add temperature transducers
@@ drivers/iio/afe/iio-rescale.c: static int rescale_temp_sense_rtd_props(struct de
+ s32 offset = 0;
+ s32 sense = 1;
+ s32 alpha;
-+ s64 tmp;
+ int ret;
+
+ device_property_read_u32(dev, "sense-offset-millicelsius", &offset);
@@ drivers/iio/afe/iio-rescale.c: static int rescale_temp_sense_rtd_props(struct de
+ rescale->numerator = 1000000;
+ rescale->denominator = alpha * sense;
+
-+ tmp = (s64)offset * (s64)alpha * (s64)sense;
-+ rescale->offset = div_s64(tmp, (s32)1000000);
++ rescale->offset = div_s64((s64)offset * rescale->denominator,
++ rescale->numerator);
+
+ return 0;
+}
13: 63be647fd110 = 13: 84bc1f7d1ab5 dt-bindings: iio: afe: add bindings for temperature-sense-rtd
14: c2f5c19dece3 = 14: 1b76cfb37e23 dt-bindings: iio: afe: add bindings for temperature transducers
base-commit: 2b6bff0b122785f09cfbdc34b1aa9edceea6e4c1
--
2.32.0.452.g940fe202adcb
From: Liam Beguin <[email protected]>
The approximation caused by integer divisions can be costly on smaller
scale values since the decimal part is significant compared to the
integer part. Switch to an IIO_VAL_INT_PLUS_NANO scale type in such
cases to maintain accuracy.
Signed-off-by: Liam Beguin <[email protected]>
---
drivers/iio/afe/iio-rescale.c | 23 ++++++++++++++++++++---
1 file changed, 20 insertions(+), 3 deletions(-)
diff --git a/drivers/iio/afe/iio-rescale.c b/drivers/iio/afe/iio-rescale.c
index 17036130d364..8a2f1c0ca5a3 100644
--- a/drivers/iio/afe/iio-rescale.c
+++ b/drivers/iio/afe/iio-rescale.c
@@ -22,7 +22,7 @@ int rescale_process_scale(struct rescale *rescale, int scale_type,
int *val, int *val2)
{
s64 tmp;
- s32 rem;
+ s32 rem, rem2;
u32 mult;
u32 neg;
@@ -38,9 +38,26 @@ int rescale_process_scale(struct rescale *rescale, int scale_type,
tmp = (s64)*val * 1000000000LL;
tmp = div_s64(tmp, rescale->denominator);
tmp *= rescale->numerator;
- tmp = div_s64(tmp, 1000000000LL);
+
+ tmp = div_s64_rem(tmp, 1000000000LL, &rem);
*val = tmp;
- return scale_type;
+
+ if (!rem)
+ return scale_type;
+
+ if (scale_type == IIO_VAL_FRACTIONAL)
+ tmp = *val2;
+ else
+ tmp = 1 << *val2;
+
+ rem2 = *val % (int)tmp;
+ *val = *val / (int)tmp;
+
+ *val2 = rem / (int)tmp;
+ if (rem2)
+ *val2 += div_s64((s64)rem2 * 1000000000LL, tmp);
+
+ return IIO_VAL_INT_PLUS_NANO;
case IIO_VAL_INT_PLUS_NANO:
case IIO_VAL_INT_PLUS_MICRO:
if (scale_type == IIO_VAL_INT_PLUS_NANO)
--
2.32.0.452.g940fe202adcb
From: Liam Beguin <[email protected]>
All four scaling coefficients can take signed values.
Make tmp a signed 64-bit integer and switch to div_s64() to preserve
signs during 64-bit divisions.
Signed-off-by: Liam Beguin <[email protected]>
---
drivers/iio/afe/iio-rescale.c | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/drivers/iio/afe/iio-rescale.c b/drivers/iio/afe/iio-rescale.c
index da462e0910ff..394f8b16b29c 100644
--- a/drivers/iio/afe/iio-rescale.c
+++ b/drivers/iio/afe/iio-rescale.c
@@ -21,7 +21,7 @@
int rescale_process_scale(struct rescale *rescale, int scale_type,
int *val, int *val2)
{
- unsigned long long tmp;
+ s64 tmp;
s32 rem;
u32 mult;
u32 neg;
@@ -38,10 +38,10 @@ int rescale_process_scale(struct rescale *rescale, int scale_type,
*val2 = rescale->denominator;
return IIO_VAL_FRACTIONAL;
case IIO_VAL_FRACTIONAL_LOG2:
- tmp = *val * 1000000000LL;
- do_div(tmp, rescale->denominator);
+ tmp = (s64)*val * 1000000000LL;
+ tmp = div_s64(tmp, rescale->denominator);
tmp *= rescale->numerator;
- do_div(tmp, 1000000000LL);
+ tmp = div_s64(tmp, 1000000000LL);
*val = tmp;
return scale_type;
case IIO_VAL_INT_PLUS_NANO:
--
2.32.0.452.g940fe202adcb
From: Liam Beguin <[email protected]>
A temperature transducer is a device that converts a thermal quantity
into any other physical quantity. This patch add support for temperature
to voltage (like the LTC2997) and temperature to current (like the
AD590) linear transducers.
In both cases these are assumed to be connected to a voltage ADC.
Signed-off-by: Liam Beguin <[email protected]>
---
drivers/iio/afe/iio-rescale.c | 32 ++++++++++++++++++++++++++++++++
1 file changed, 32 insertions(+)
diff --git a/drivers/iio/afe/iio-rescale.c b/drivers/iio/afe/iio-rescale.c
index ea162ba2a674..6bdcf6e139c7 100644
--- a/drivers/iio/afe/iio-rescale.c
+++ b/drivers/iio/afe/iio-rescale.c
@@ -422,11 +422,37 @@ static int rescale_temp_sense_rtd_props(struct device *dev,
return 0;
}
+static int rescale_temp_transducer_props(struct device *dev,
+ struct rescale *rescale)
+{
+ s32 offset = 0;
+ s32 sense = 1;
+ s32 alpha;
+ int ret;
+
+ device_property_read_u32(dev, "sense-offset-millicelsius", &offset);
+ device_property_read_u32(dev, "sense-resistor-ohms", &sense);
+ ret = device_property_read_u32(dev, "alpha-ppm-per-celsius", &alpha);
+ if (ret) {
+ dev_err(dev, "failed to read alpha-ppm-per-celsius: %d\n", ret);
+ return ret;
+ }
+
+ rescale->numerator = 1000000;
+ rescale->denominator = alpha * sense;
+
+ rescale->offset = div_s64((s64)offset * rescale->denominator,
+ rescale->numerator);
+
+ return 0;
+}
+
enum rescale_variant {
CURRENT_SENSE_AMPLIFIER,
CURRENT_SENSE_SHUNT,
VOLTAGE_DIVIDER,
TEMP_SENSE_RTD,
+ TEMP_TRANSDUCER,
};
static const struct rescale_cfg rescale_cfg[] = {
@@ -446,6 +472,10 @@ static const struct rescale_cfg rescale_cfg[] = {
.type = IIO_TEMP,
.props = rescale_temp_sense_rtd_props,
},
+ [TEMP_TRANSDUCER] = {
+ .type = IIO_TEMP,
+ .props = rescale_temp_transducer_props,
+ },
};
static const struct of_device_id rescale_match[] = {
@@ -457,6 +487,8 @@ static const struct of_device_id rescale_match[] = {
.data = &rescale_cfg[VOLTAGE_DIVIDER], },
{ .compatible = "temperature-sense-rtd",
.data = &rescale_cfg[TEMP_SENSE_RTD], },
+ { .compatible = "temperature-transducer",
+ .data = &rescale_cfg[TEMP_TRANSDUCER], },
{ /* sentinel */ }
};
MODULE_DEVICE_TABLE(of, rescale_match);
--
2.32.0.452.g940fe202adcb
From: Liam Beguin <[email protected]>
An RTD (Resistance Temperature Detector) is a kind of temperature
sensor used to get a linear voltage to temperature reading within a
give range (usually 0 to 100 degrees Celsius). Common types of RTDs
include PT100, PT500, and PT1000.
Signed-off-by: Liam Beguin <[email protected]>
---
drivers/iio/afe/iio-rescale.c | 48 +++++++++++++++++++++++++++++++++++
1 file changed, 48 insertions(+)
diff --git a/drivers/iio/afe/iio-rescale.c b/drivers/iio/afe/iio-rescale.c
index 8a2f1c0ca5a3..ea162ba2a674 100644
--- a/drivers/iio/afe/iio-rescale.c
+++ b/drivers/iio/afe/iio-rescale.c
@@ -381,10 +381,52 @@ static int rescale_voltage_divider_props(struct device *dev,
return 0;
}
+static int rescale_temp_sense_rtd_props(struct device *dev,
+ struct rescale *rescale)
+{
+ u32 factor;
+ u32 alpha;
+ u32 iexc;
+ u32 tmp;
+ int ret;
+ u32 r0;
+
+ ret = device_property_read_u32(dev, "excitation-current-microamp",
+ &iexc);
+ if (ret) {
+ dev_err(dev, "failed to read excitation-current-microamp: %d\n",
+ ret);
+ return ret;
+ }
+
+ ret = device_property_read_u32(dev, "alpha-ppm-per-celsius", &alpha);
+ if (ret) {
+ dev_err(dev, "failed to read alpha-ppm-per-celsius: %d\n",
+ ret);
+ return ret;
+ }
+
+ ret = device_property_read_u32(dev, "r-naught-ohms", &r0);
+ if (ret) {
+ dev_err(dev, "failed to read r-naught-ohms: %d\n", ret);
+ return ret;
+ }
+
+ tmp = r0 * iexc * alpha / 1000000;
+ factor = gcd(tmp, 1000000);
+ rescale->numerator = 1000000 / factor;
+ rescale->denominator = tmp / factor;
+
+ rescale->offset = -1 * ((r0 * iexc) / 1000);
+
+ return 0;
+}
+
enum rescale_variant {
CURRENT_SENSE_AMPLIFIER,
CURRENT_SENSE_SHUNT,
VOLTAGE_DIVIDER,
+ TEMP_SENSE_RTD,
};
static const struct rescale_cfg rescale_cfg[] = {
@@ -400,6 +442,10 @@ static const struct rescale_cfg rescale_cfg[] = {
.type = IIO_VOLTAGE,
.props = rescale_voltage_divider_props,
},
+ [TEMP_SENSE_RTD] = {
+ .type = IIO_TEMP,
+ .props = rescale_temp_sense_rtd_props,
+ },
};
static const struct of_device_id rescale_match[] = {
@@ -409,6 +455,8 @@ static const struct of_device_id rescale_match[] = {
.data = &rescale_cfg[CURRENT_SENSE_SHUNT], },
{ .compatible = "voltage-divider",
.data = &rescale_cfg[VOLTAGE_DIVIDER], },
+ { .compatible = "temperature-sense-rtd",
+ .data = &rescale_cfg[TEMP_SENSE_RTD], },
{ /* sentinel */ }
};
MODULE_DEVICE_TABLE(of, rescale_match);
--
2.32.0.452.g940fe202adcb
From: Liam Beguin <[email protected]>
An ADC is often used to measure other quantities indirectly. This
binding describe one case, the measurement of a temperature through the
voltage across an RTD resistor such as a PT1000.
Signed-off-by: Liam Beguin <[email protected]>
Reviewed-by: Rob Herring <[email protected]>
---
.../iio/afe/temperature-sense-rtd.yaml | 101 ++++++++++++++++++
1 file changed, 101 insertions(+)
create mode 100644 Documentation/devicetree/bindings/iio/afe/temperature-sense-rtd.yaml
diff --git a/Documentation/devicetree/bindings/iio/afe/temperature-sense-rtd.yaml b/Documentation/devicetree/bindings/iio/afe/temperature-sense-rtd.yaml
new file mode 100644
index 000000000000..336ce96371db
--- /dev/null
+++ b/Documentation/devicetree/bindings/iio/afe/temperature-sense-rtd.yaml
@@ -0,0 +1,101 @@
+# SPDX-License-Identifier: (GPL-2.0 OR BSD-2-Clause)
+%YAML 1.2
+---
+$id: http://devicetree.org/schemas/iio/afe/temperature-sense-rtd.yaml#
+$schema: http://devicetree.org/meta-schemas/core.yaml#
+
+title: Temperature Sense RTD
+
+maintainers:
+ - Liam Beguin <[email protected]>
+
+description: |
+ RTDs (Resistance Temperature Detectors) are a kind of temperature sensors
+ used to get a linear voltage to temperature reading within a give range
+ (usually 0 to 100 degrees Celsius).
+
+ When an io-channel measures the output voltage across an RTD such as a
+ PT1000, the interesting measurement is almost always the corresponding
+ temperature, not the voltage output. This binding describes such a circuit.
+
+ The general transfer function here is (using SI units)
+
+ V = R(T) * iexc
+ R(T) = r0 * (1 + alpha * T)
+ T = 1 / (alpha * r0 * iexc) * (V - r0 * iexc)
+
+ The following circuit matches what's in the examples section.
+
+ 5V0
+ -----
+ |
+ +---+----+
+ | R 5k |
+ +---+----+
+ |
+ V 1mA
+ |
+ +---- Vout
+ |
+ +---+----+
+ | PT1000 |
+ +---+----+
+ |
+ -----
+ GND
+
+properties:
+ compatible:
+ const: temperature-sense-rtd
+
+ io-channels:
+ maxItems: 1
+ description: |
+ Channel node of a voltage io-channel.
+
+ '#io-channel-cells':
+ const: 0
+
+ excitation-current-microamp:
+ description: The current fed through the RTD sensor.
+
+ alpha-ppm-per-celsius:
+ description: |
+ alpha can also be expressed in micro-ohms per ohm Celsius. It's a linear
+ approximation of the resistance versus temperature relationship
+ between 0 and 100 degrees Celsius.
+
+ alpha = (R_100 - R_0) / (100 * R_0)
+
+ Where, R_100 is the resistance of the sensor at 100 degrees Celsius, and
+ R_0 (or r-naught-ohms) is the resistance of the sensor at 0 degrees
+ Celsius.
+
+ Pure platinum has an alpha of 3925. Industry standards such as IEC60751
+ and ASTM E-1137 specify an alpha of 3850.
+
+ r-naught-ohms:
+ description: |
+ Resistance of the sensor at 0 degrees Celsius.
+ Common values are 100 for PT100, 500 for PT500, and 1000 for PT1000
+
+additionalProperties: false
+required:
+ - compatible
+ - io-channels
+ - excitation-current-microamp
+ - alpha-ppm-per-celsius
+ - r-naught-ohms
+
+examples:
+ - |
+ pt1000_1: temperature-sensor0 {
+ compatible = "temperature-sense-rtd";
+ #io-channel-cells = <0>;
+ io-channels = <&temp_adc1 0>;
+
+ excitation-current-microamp = <1000>; /* i = U/R = 5 / 5000 */
+ alpha-ppm-per-celsius = <3908>;
+ r-naught-ohms = <1000>;
+ };
+...
--
2.32.0.452.g940fe202adcb
From: Liam Beguin <[email protected]>
An ADC is often used to measure other quantities indirectly.
This binding describe one case, the measurement of a temperature
through a temperature transducer (either voltage or current).
Signed-off-by: Liam Beguin <[email protected]>
Reviewed-by: Rob Herring <[email protected]>
---
.../iio/afe/temperature-transducer.yaml | 114 ++++++++++++++++++
1 file changed, 114 insertions(+)
create mode 100644 Documentation/devicetree/bindings/iio/afe/temperature-transducer.yaml
diff --git a/Documentation/devicetree/bindings/iio/afe/temperature-transducer.yaml b/Documentation/devicetree/bindings/iio/afe/temperature-transducer.yaml
new file mode 100644
index 000000000000..cfbf5350db27
--- /dev/null
+++ b/Documentation/devicetree/bindings/iio/afe/temperature-transducer.yaml
@@ -0,0 +1,114 @@
+# SPDX-License-Identifier: (GPL-2.0 OR BSD-2-Clause)
+%YAML 1.2
+---
+$id: http://devicetree.org/schemas/iio/afe/temperature-transducer.yaml#
+$schema: http://devicetree.org/meta-schemas/core.yaml#
+
+title: Temperature Transducer
+
+maintainers:
+ - Liam Beguin <[email protected]>
+
+description: |
+ A temperature transducer is a device that converts a thermal quantity
+ into any other physical quantity. This binding applies to temperature to
+ voltage (like the LTC2997), and temperature to current (like the AD590)
+ linear transducers.
+ In both cases these are assumed to be connected to a voltage ADC.
+
+ When an io-channel measures the output voltage of a temperature analog front
+ end such as a temperature transducer, the interesting measurement is almost
+ always the corresponding temperature, not the voltage output. This binding
+ describes such a circuit.
+
+ The general transfer function here is (using SI units)
+ V(T) = Rsense * Isense(T)
+ T = (Isense(T) / alpha) + offset
+ T = 1 / (Rsense * alpha) * (V + offset * Rsense * alpha)
+
+ When using a temperature to voltage transducer, Rsense is set to 1.
+
+ The following circuits show a temperature to current and a temperature to
+ voltage transducer that can be used with this binding.
+
+ VCC
+ -----
+ |
+ +---+---+
+ | AD590 | VCC
+ +---+---+ -----
+ | |
+ V proportional to T +----+----+
+ | D+ --+ |
+ +---- Vout | LTC2997 +--- Vout
+ | D- --+ |
+ +---+----+ +---------+
+ | Rsense | |
+ +---+----+ -----
+ | GND
+ -----
+ GND
+
+properties:
+ compatible:
+ const: temperature-transducer
+
+ io-channels:
+ maxItems: 1
+ description: |
+ Channel node of a voltage io-channel.
+
+ '#io-channel-cells':
+ const: 0
+
+ sense-offset-millicelsius:
+ description: |
+ Temperature offset.
+ This offset is commonly used to convert from Kelvins to degrees Celsius.
+ In that case, sense-offset-millicelsius would be set to <(-273150)>.
+ default: 0
+
+ sense-resistor-ohms:
+ description: |
+ The sense resistor.
+ By default sense-resistor-ohms cancels out the resistor making the
+ circuit behave like a temperature transducer.
+ default: 1
+
+ alpha-ppm-per-celsius:
+ description: |
+ Sometimes referred to as output gain, slope, or temperature coefficient.
+
+ alpha is expressed in parts per million which can be micro-amps per
+ degrees Celsius or micro-volts per degrees Celsius. The is the main
+ characteristic of a temperature transducer and should be stated in the
+ datasheet.
+
+additionalProperties: false
+
+required:
+ - compatible
+ - io-channels
+ - alpha-ppm-per-celsius
+
+examples:
+ - |
+ ad950: temperature-sensor-0 {
+ compatible = "temperature-transducer";
+ #io-channel-cells = <0>;
+ io-channels = <&temp_adc 3>;
+
+ sense-offset-millicelsius = <(-273150)>; /* Kelvin to degrees Celsius */
+ sense-resistor-ohms = <8060>;
+ alpha-ppm-per-celsius = <1>; /* 1 uA/K */
+ };
+ - |
+ znq_tmp: temperature-sensor-1 {
+ compatible = "temperature-transducer";
+ #io-channel-cells = <0>;
+ io-channels = <&temp_adc 2>;
+
+ sense-offset-millicelsius = <(-273150)>; /* Kelvin to degrees Celsius */
+ alpha-ppm-per-celsius = <4000>; /* 4 mV/K */
+ };
+...
--
2.32.0.452.g940fe202adcb
From: Liam Beguin <[email protected]>
iio_convert_raw_to_processed_unlocked() assumes the offset is an
integer. Make a best effort to get a valid offset value for fractional
cases without breaking implicit truncations.
Fixes: 48e44ce0f881 ("iio:inkern: Add function to read the processed value")
Signed-off-by: Liam Beguin <[email protected]>
---
drivers/iio/inkern.c | 32 +++++++++++++++++++++++++++-----
1 file changed, 27 insertions(+), 5 deletions(-)
diff --git a/drivers/iio/inkern.c b/drivers/iio/inkern.c
index dbe13fad3cbb..aff6d2d3447c 100644
--- a/drivers/iio/inkern.c
+++ b/drivers/iio/inkern.c
@@ -595,13 +595,35 @@ EXPORT_SYMBOL_GPL(iio_read_channel_average_raw);
static int iio_convert_raw_to_processed_unlocked(struct iio_channel *chan,
int raw, int *processed, unsigned int scale)
{
- int scale_type, scale_val, scale_val2, offset;
+ int scale_type, scale_val, scale_val2;
+ int offset_type, offset_val, offset_val2;
s64 raw64 = raw;
- int ret;
- ret = iio_channel_read(chan, &offset, NULL, IIO_CHAN_INFO_OFFSET);
- if (ret >= 0)
- raw64 += offset;
+ offset_type = iio_channel_read(chan, &offset_val, &offset_val2,
+ IIO_CHAN_INFO_OFFSET);
+ if (offset_type >= 0) {
+ switch (offset_type) {
+ case IIO_VAL_INT:
+ break;
+ case IIO_VAL_INT_PLUS_MICRO:
+ case IIO_VAL_INT_PLUS_NANO:
+ /*
+ * Both IIO_VAL_INT_PLUS_MICRO and IIO_VAL_INT_PLUS_NANO
+ * implicitely truncate the offset to it's integer form.
+ */
+ break;
+ case IIO_VAL_FRACTIONAL:
+ offset_val /= offset_val2;
+ break;
+ case IIO_VAL_FRACTIONAL_LOG2:
+ offset_val /= (1 << offset_val2);
+ break;
+ default:
+ return -EINVAL;
+ }
+
+ raw64 += offset_val;
+ }
scale_type = iio_channel_read(chan, &scale_val, &scale_val2,
IIO_CHAN_INFO_SCALE);
--
2.32.0.452.g940fe202adcb
From: Liam Beguin <[email protected]>
In preparation for the addition of kunit tests, expose the logic
responsible for combining channel scales.
Signed-off-by: Liam Beguin <[email protected]>
---
drivers/iio/afe/iio-rescale.c | 65 ++++++++++++++-------------------
include/linux/iio/afe/rescale.h | 30 +++++++++++++++
2 files changed, 58 insertions(+), 37 deletions(-)
create mode 100644 include/linux/iio/afe/rescale.h
diff --git a/drivers/iio/afe/iio-rescale.c b/drivers/iio/afe/iio-rescale.c
index 774eb3044edd..d0669fd8eac5 100644
--- a/drivers/iio/afe/iio-rescale.c
+++ b/drivers/iio/afe/iio-rescale.c
@@ -11,35 +11,46 @@
#include <linux/gcd.h>
#include <linux/iio/consumer.h>
#include <linux/iio/iio.h>
+#include <linux/iio/afe/rescale.h>
#include <linux/module.h>
#include <linux/of.h>
#include <linux/of_device.h>
#include <linux/platform_device.h>
#include <linux/property.h>
-struct rescale;
-
-struct rescale_cfg {
- enum iio_chan_type type;
- int (*props)(struct device *dev, struct rescale *rescale);
-};
+int rescale_process_scale(struct rescale *rescale, int scale_type,
+ int *val, int *val2)
+{
+ unsigned long long tmp;
-struct rescale {
- const struct rescale_cfg *cfg;
- struct iio_channel *source;
- struct iio_chan_spec chan;
- struct iio_chan_spec_ext_info *ext_info;
- bool chan_processed;
- s32 numerator;
- s32 denominator;
-};
+ switch (scale_type) {
+ case IIO_VAL_FRACTIONAL:
+ *val *= rescale->numerator;
+ *val2 *= rescale->denominator;
+ return scale_type;
+ case IIO_VAL_INT:
+ *val *= rescale->numerator;
+ if (rescale->denominator == 1)
+ return scale_type;
+ *val2 = rescale->denominator;
+ return IIO_VAL_FRACTIONAL;
+ case IIO_VAL_FRACTIONAL_LOG2:
+ tmp = *val * 1000000000LL;
+ do_div(tmp, rescale->denominator);
+ tmp *= rescale->numerator;
+ do_div(tmp, 1000000000LL);
+ *val = tmp;
+ return scale_type;
+ default:
+ return -EOPNOTSUPP;
+ }
+}
static int rescale_read_raw(struct iio_dev *indio_dev,
struct iio_chan_spec const *chan,
int *val, int *val2, long mask)
{
struct rescale *rescale = iio_priv(indio_dev);
- unsigned long long tmp;
int ret;
switch (mask) {
@@ -65,27 +76,7 @@ static int rescale_read_raw(struct iio_dev *indio_dev,
} else {
ret = iio_read_channel_scale(rescale->source, val, val2);
}
- switch (ret) {
- case IIO_VAL_FRACTIONAL:
- *val *= rescale->numerator;
- *val2 *= rescale->denominator;
- return ret;
- case IIO_VAL_INT:
- *val *= rescale->numerator;
- if (rescale->denominator == 1)
- return ret;
- *val2 = rescale->denominator;
- return IIO_VAL_FRACTIONAL;
- case IIO_VAL_FRACTIONAL_LOG2:
- tmp = *val * 1000000000LL;
- do_div(tmp, rescale->denominator);
- tmp *= rescale->numerator;
- do_div(tmp, 1000000000LL);
- *val = tmp;
- return ret;
- default:
- return -EOPNOTSUPP;
- }
+ return rescale_process_scale(rescale, ret, val, val2);
default:
return -EINVAL;
}
diff --git a/include/linux/iio/afe/rescale.h b/include/linux/iio/afe/rescale.h
new file mode 100644
index 000000000000..14d4ee1227c6
--- /dev/null
+++ b/include/linux/iio/afe/rescale.h
@@ -0,0 +1,30 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2021 Liam Beguin <[email protected]>
+ */
+
+#ifndef __IIO_RESCALE_H__
+#define __IIO_RESCALE_H__
+
+#include <linux/iio/iio.h>
+
+struct rescale;
+
+struct rescale_cfg {
+ enum iio_chan_type type;
+ int (*props)(struct device *dev, struct rescale *rescale);
+};
+
+struct rescale {
+ const struct rescale_cfg *cfg;
+ struct iio_channel *source;
+ struct iio_chan_spec chan;
+ struct iio_chan_spec_ext_info *ext_info;
+ bool chan_processed;
+ s32 numerator;
+ s32 denominator;
+};
+
+int rescale_process_scale(struct rescale *rescale, int scale_type,
+ int *val, int *val2);
+#endif /* __IIO_RESCALE_H__ */
--
2.32.0.452.g940fe202adcb
On Sun, 14 Nov 2021 22:43:20 -0500
Liam Beguin <[email protected]> wrote:
> Hi Jonathan, Peter,
>
> Apologies for not getting back to you sooner. I got caught up on other
> work and wasn't able to dedicate time to this earlier. Hopefully, this
> time around, I'll be able to get this to the finish line :-)
>
> I left out IIO_VAL_INT overflows for now, so that I can focus on getting
> the rest of these changes pulled in, but I don't mind adding a patch for
> that later on.
>
> This series focuses on adding temperature rescaling support to the IIO
> Analog Front End (AFE) driver.
>
> The first few patches address minor bugs in IIO inkernel functions, and
> prepare the AFE driver for the additional features.
>
> The main changes to the AFE driver include an initial Kunit test suite,
> support for IIO_VAL_INT_PLUS_{NANO,MICRO} scales, and support for RTDs
> and temperature transducer sensors.
>
> Thanks for your time,
> Liam
Hi Liam,
I'm fine with these. The comment about using the MICRO etc defines can
be handled as as trivial follow up patch. Hopefully someone else can
figure out the 0-day build issue as I didn't managed to.
However, I've long ago lost track of the various precision discussions
you and Peter were having so would like Peter's input before taking these.
Thanks again for your persistence with this,
Jonathan
>
> Changes since v8:
> - reword comment
> - fix erroneous 64-bit division
> - optimize and use 32-bit divisions when values are know to not overflow
> - keep IIO_VAL_FRACTIONAL scale when possible, if not default to fixed
> point
> - add test cases
> - use nano precision in test cases
> - simplify offset calculation in rtd_props()
>
> Changes since v7:
> - drop gcd() logic in rescale_process_scale()
> - use div_s64() instead of do_div() for signed 64-bit divisions
> - combine IIO_VAL_FRACTIONAL and IIO_VAL_FRACTIONAL_LOG2 scale cases
> - switch to INT_PLUS_NANO when accuracy is lost with FRACTIONAL scales
> - rework test logic to allow for small relative error
> - rename test variables to align error output messages
>
> Changes since v6:
> - rework IIO_VAL_INT_PLUS_{NANO,MICRO} based on Peter's suggestion
> - combine IIO_VAL_INT_PLUS_{NANO,MICRO} cases
> - add test cases for negative IIO_VAL_INT_PLUS_{NANO,MICRO} corner cases
> - force use of positive integers with gcd()
> - reduce risk of integer overflow in IIO_VAL_FRACTIONAL_LOG2
> - fix duplicate symbol build error
> - apply Reviewed-by
>
> Changes since v5:
> - add include/linux/iio/afe/rescale.h
> - expose functions use to process scale and offset
> - add basic iio-rescale kunit test cases
> - fix integer overflow case
> - improve precision for IIO_VAL_FRACTIONAL_LOG2
>
> Changes since v4:
> - only use gcd() when necessary in overflow mitigation
> - fix INT_PLUS_{MICRO,NANO} support
> - apply Reviewed-by
> - fix temperature-transducer bindings
>
> Changes since v3:
> - drop unnecessary fallthrough statements
> - drop redundant local variables in some calculations
> - fix s64 divisions on 32bit platforms by using do_div
> - add comment describing iio-rescaler offset calculation
> - drop unnecessary MAINTAINERS entry
>
> Changes since v2:
> - don't break implicit offset truncations
> - make a best effort to get a valid value for fractional types
> - drop return value change in iio_convert_raw_to_processed_unlocked()
> - don't rely on processed value for offset calculation
> - add INT_PLUS_{MICRO,NANO} support in iio-rescale
> - revert generic implementation in favor of temperature-sense-rtd and
> temperature-transducer
> - add separate section to MAINTAINERS file
>
> Changes since v1:
> - rebase on latest iio `testing` branch
> - also apply consumer scale on integer channel scale types
> - don't break implicit truncation in processed channel offset
> calculation
> - drop temperature AFE flavors in favor of a simpler generic
> implementation
>
> Liam Beguin (14):
> iio: inkern: apply consumer scale on IIO_VAL_INT cases
> iio: inkern: apply consumer scale when no channel scale is available
> iio: inkern: make a best effort on offset calculation
> iio: afe: rescale: expose scale processing function
> iio: afe: rescale: add INT_PLUS_{MICRO,NANO} support
> iio: afe: rescale: add offset support
> iio: afe: rescale: use s64 for temporary scale calculations
> iio: afe: rescale: reduce risk of integer overflow
> iio: afe: rescale: fix accuracy for small fractional scales
> iio: test: add basic tests for the iio-rescale driver
> iio: afe: rescale: add RTD temperature sensor support
> iio: afe: rescale: add temperature transducers
> dt-bindings: iio: afe: add bindings for temperature-sense-rtd
> dt-bindings: iio: afe: add bindings for temperature transducers
>
> .../iio/afe/temperature-sense-rtd.yaml | 101 +++
> .../iio/afe/temperature-transducer.yaml | 114 +++
> drivers/iio/afe/iio-rescale.c | 271 ++++++-
> drivers/iio/inkern.c | 40 +-
> drivers/iio/test/Kconfig | 10 +
> drivers/iio/test/Makefile | 1 +
> drivers/iio/test/iio-test-rescale.c | 705 ++++++++++++++++++
> include/linux/iio/afe/rescale.h | 34 +
> 8 files changed, 1232 insertions(+), 44 deletions(-)
> create mode 100644 Documentation/devicetree/bindings/iio/afe/temperature-sense-rtd.yaml
> create mode 100644 Documentation/devicetree/bindings/iio/afe/temperature-transducer.yaml
> create mode 100644 drivers/iio/test/iio-test-rescale.c
> create mode 100644 include/linux/iio/afe/rescale.h
>
> Range-diff against v8:
> 1: 42a7a1047edc = 1: ae3cc93baee6 iio: inkern: apply consumer scale on IIO_VAL_INT cases
> 2: a1cd89fdad11 = 2: 06f66e7f7403 iio: inkern: apply consumer scale when no channel scale is available
> 3: ed0721fb6bd1 = 3: 2dbf6b3bbaeb iio: inkern: make a best effort on offset calculation
> 4: f8fb78bb1112 = 4: b083cf307268 iio: afe: rescale: expose scale processing function
> 5: 504b7a3f830b ! 5: a0bde29ecc8c iio: afe: rescale: add INT_PLUS_{MICRO,NANO} support
> @@ drivers/iio/afe/iio-rescale.c: int rescale_process_scale(struct rescale *rescale
> + else
> + mult = 1000000LL;
> + /*
> -+ * For IIO_VAL_INT_PLUS_{MICRO,NANO} scale types if *val OR
> -+ * *val2 is negative the schan scale is negative
> ++ * For IIO_VAL_INT_PLUS_{MICRO,NANO} scale types if either *val
> ++ * OR *val2 is negative the schan scale is negative, i.e.
> ++ * *val = 1 and *val2 = -0.5 yields -1.5 not -0.5.
> + */
> + neg = *val < 0 || *val2 < 0;
> +
> 6: c254e9ae813e = 6: c3d0e6248033 iio: afe: rescale: add offset support
> 7: ee8814d6abe4 = 7: 2a81fa735103 iio: afe: rescale: use s64 for temporary scale calculations
> 8: 62cdcfbc9836 = 8: 8315548d0fce iio: afe: rescale: reduce risk of integer overflow
> 9: 88309a5136ee ! 9: 223ed0569cd2 iio: afe: rescale: fix accuracy for small fractional scales
> @@ drivers/iio/afe/iio-rescale.c: int rescale_process_scale(struct rescale *rescale
> +
> + tmp = div_s64_rem(tmp, 1000000000LL, &rem);
> *val = tmp;
> +- return scale_type;
> ++
> ++ if (!rem)
> ++ return scale_type;
> +
> -+ /*
> -+ * For small values, the approximation can be costly,
> -+ * change scale type to maintain accuracy.
> -+ *
> -+ * 100 vs. 10000000 NANO caps the error to about 100 ppm.
> -+ */
> + if (scale_type == IIO_VAL_FRACTIONAL)
> + tmp = *val2;
> + else
> + tmp = 1 << *val2;
> +
> -+ if (abs(rem) > 10000000 && abs(*val / tmp) < 100) {
> -+ *val = div_s64_rem(*val, tmp, &rem2);
> -+
> -+ *val2 = div_s64(rem, tmp);
> -+ if (rem2)
> -+ *val2 += div_s64(rem2 * 1000000000LL, tmp);
> ++ rem2 = *val % (int)tmp;
> ++ *val = *val / (int)tmp;
> +
> -+ return IIO_VAL_INT_PLUS_NANO;
> -+ }
> ++ *val2 = rem / (int)tmp;
> ++ if (rem2)
> ++ *val2 += div_s64((s64)rem2 * 1000000000LL, tmp);
> +
> - return scale_type;
> ++ return IIO_VAL_INT_PLUS_NANO;
> case IIO_VAL_INT_PLUS_NANO:
> case IIO_VAL_INT_PLUS_MICRO:
> + if (scale_type == IIO_VAL_INT_PLUS_NANO)
> 10: fb505a9f42f1 ! 10: 90044efdf8be iio: test: add basic tests for the iio-rescale driver
> @@ drivers/iio/test/Makefile
> # Keep in alphabetical order
> +obj-$(CONFIG_IIO_RESCALE_KUNIT_TEST) += iio-test-rescale.o ../afe/iio-rescale.o
> obj-$(CONFIG_IIO_TEST_FORMAT) += iio-test-format.o
> + CFLAGS_iio-test-format.o += $(DISABLE_STRUCTLEAK_PLUGIN)
>
> ## drivers/iio/test/iio-test-rescale.c (new) ##
> @@
> @@ drivers/iio/test/iio-test-rescale.c (new)
> + * Use cases with small scales involving divisions
> + */
> + {
> ++ .name = "small IIO_VAL_FRACTIONAL, 261/509 scaled by 90/1373754273",
> ++ .numerator = 261,
> ++ .denominator = 509,
> ++ .schan_scale_type = IIO_VAL_FRACTIONAL,
> ++ .schan_val = 90,
> ++ .schan_val2 = 1373754273,
> ++ .expected = "0.000000033594",
> ++ },
> ++ {
> ++ .name = "small IIO_VAL_FRACTIONAL, 90/1373754273 scaled by 261/509",
> ++ .numerator = 90,
> ++ .denominator = 1373754273,
> ++ .schan_scale_type = IIO_VAL_FRACTIONAL,
> ++ .schan_val = 261,
> ++ .schan_val2 = 509,
> ++ .expected = "0.000000033594",
> ++ },
> ++ {
> ++ .name = "small IIO_VAL_FRACTIONAL, 760/1373754273 scaled by 427/2727",
> ++ .numerator = 760,
> ++ .denominator = 1373754273,
> ++ .schan_scale_type = IIO_VAL_FRACTIONAL,
> ++ .schan_val = 427,
> ++ .schan_val2 = 2727,
> ++ .expected = "0.000000086626",
> ++ },
> ++ {
> ++ .name = "small IIO_VAL_FRACTIONAL, 761/1373754273 scaled by 427/2727",
> ++ .numerator = 761,
> ++ .denominator = 1373754273,
> ++ .schan_scale_type = IIO_VAL_FRACTIONAL,
> ++ .schan_val = 427,
> ++ .schan_val2 = 2727,
> ++ .expected = "0.000000086740",
> ++ },
> ++ {
> ++ .name = "small IIO_VAL_FRACTIONAL, 5/32768 scaled by 3/10000",
> ++ .numerator = 5,
> ++ .denominator = 32768,
> ++ .schan_scale_type = IIO_VAL_FRACTIONAL,
> ++ .schan_val = 3,
> ++ .schan_val2 = 10000,
> ++ .expected = "0.0000000457763671875",
> ++ },
> ++ {
> + .name = "small IIO_VAL_FRACTIONAL, 0 < scale < 1",
> + .numerator = 6,
> + .denominator = 6,
> @@ drivers/iio/test/iio-test-rescale.c (new)
> + .expected = "-1.3333333333333333",
> + },
> + {
> ++ .name = "small IIO_VAL_FRACTIONAL_LOG2, 760/32768 scaled by 15/22",
> ++ .numerator = 760,
> ++ .denominator = 32768,
> ++ .schan_scale_type = IIO_VAL_FRACTIONAL_LOG2,
> ++ .schan_val = 15,
> ++ .schan_val2 = 22,
> ++ .expected = "0.000000082946",
> ++ },
> ++ {
> ++ .name = "small IIO_VAL_FRACTIONAL_LOG2, 761/32768 scaled by 15/22",
> ++ .numerator = 761,
> ++ .denominator = 32768,
> ++ .schan_scale_type = IIO_VAL_FRACTIONAL_LOG2,
> ++ .schan_val = 15,
> ++ .schan_val2 = 22,
> ++ .expected = "0.000000083055",
> ++ },
> ++ {
> + .name = "small IIO_VAL_FRACTIONAL_LOG2, 0 < scale < 1",
> + .numerator = 16,
> + .denominator = 3,
> @@ drivers/iio/test/iio-test-rescale.c (new)
> +KUNIT_ARRAY_PARAM(iio_rescale_offset, offset_cases, case_to_desc);
> +
> +/**
> -+ * iio_str_to_micro() - Parse a fixed-point string to get an
> -+ * IIO_VAL_INT_PLUS_MICRO value
> ++ * iio_str_to_nano() - Parse a fixed-point string to get an
> ++ * IIO_VAL_INT_PLUS_NANO value
> + * @str: The string to parse
> -+ * @micro: The number as an integer
> ++ * @nano: The number as an integer
> + *
> + * Returns 0 on success, or a negative error code if the string cound not be
> + * parsed.
> + */
> -+static int iio_str_to_micro(const char *str, s64 *micro)
> ++static int iio_str_to_nano(const char *str, s64 *nano)
> +{
> -+ int fract_mult = 100000LL;
> ++ int fract_mult = 100000000LL;
> + int tmp, tmp2;
> + int ret = 0;
> +
> @@ drivers/iio/test/iio-test-rescale.c (new)
> + if (tmp < 0)
> + tmp2 *= -1;
> +
> -+ *micro = (s64)tmp * 10 * fract_mult + tmp2;
> ++ *nano = (s64)tmp * 10 * fract_mult + tmp2;
> +
> + return ret;
> +}
> +
> +/**
> -+ * iio_test_relative_error_ppm() - Compute relative error (in ppm) between two
> -+ * fixed-point strings
> ++ * iio_test_relative_error_ppm() - Compute relative error (in parts-per-million)
> ++ * between two fixed-point strings
> + * @real_str: The real value as a string
> + * @exp_str: The expected value as a string
> + *
> + * Returns a negative error code if the strings cound not be parsed, or the
> -+ * relative error in ppm.
> ++ * relative error in parts-per-million.
> + */
> +static int iio_test_relative_error_ppm(const char *real_str, const char *exp_str)
> +{
> + s64 real, exp, err;
> + int ret;
> +
> -+ ret = iio_str_to_micro(real_str, &real);
> ++ ret = iio_str_to_nano(real_str, &real);
> + if (ret < 0)
> + return ret;
> +
> -+ ret = iio_str_to_micro(exp_str, &exp);
> ++ ret = iio_str_to_nano(exp_str, &exp);
> + if (ret < 0)
> + return ret;
> +
> ++ if (!exp) {
> ++ pr_err("Expected value is null, relative error is undefined\n");
> ++ return -EINVAL;
> ++ }
> ++
> + err = 1000000 * abs(exp - real);
> + err = div64_u64(err, abs(exp));
> + return (int)err;
> @@ drivers/iio/test/iio-test-rescale.c (new)
> + rel_ppm = iio_test_relative_error_ppm(buff, t->expected);
> + KUNIT_EXPECT_GE_MSG(test, rel_ppm, 0, "failed to compute ppm\n");
> +
> -+ KUNIT_EXPECT_LT_MSG(test, rel_ppm, 500,
> ++ KUNIT_EXPECT_EQ_MSG(test, rel_ppm, 0,
> + "\t real=%s"
> + "\texpected=%s\n",
> + buff, t->expected);
> 11: 050487186e14 = 11: c4ed463e5fb0 iio: afe: rescale: add RTD temperature sensor support
> 12: f36a44a5d898 ! 12: ff2f0dc248a7 iio: afe: rescale: add temperature transducers
> @@ drivers/iio/afe/iio-rescale.c: static int rescale_temp_sense_rtd_props(struct de
> + s32 offset = 0;
> + s32 sense = 1;
> + s32 alpha;
> -+ s64 tmp;
> + int ret;
> +
> + device_property_read_u32(dev, "sense-offset-millicelsius", &offset);
> @@ drivers/iio/afe/iio-rescale.c: static int rescale_temp_sense_rtd_props(struct de
> + rescale->numerator = 1000000;
> + rescale->denominator = alpha * sense;
> +
> -+ tmp = (s64)offset * (s64)alpha * (s64)sense;
> -+ rescale->offset = div_s64(tmp, (s32)1000000);
> ++ rescale->offset = div_s64((s64)offset * rescale->denominator,
> ++ rescale->numerator);
> +
> + return 0;
> +}
> 13: 63be647fd110 = 13: 84bc1f7d1ab5 dt-bindings: iio: afe: add bindings for temperature-sense-rtd
> 14: c2f5c19dece3 = 14: 1b76cfb37e23 dt-bindings: iio: afe: add bindings for temperature transducers
>
> base-commit: 2b6bff0b122785f09cfbdc34b1aa9edceea6e4c1
On Sun, Nov 21, 2021 at 11:25:56AM +0000, Jonathan Cameron wrote:
> On Sun, 14 Nov 2021 22:43:20 -0500
> Liam Beguin <[email protected]> wrote:
>
> > Hi Jonathan, Peter,
> >
> > Apologies for not getting back to you sooner. I got caught up on other
> > work and wasn't able to dedicate time to this earlier. Hopefully, this
> > time around, I'll be able to get this to the finish line :-)
> >
> > I left out IIO_VAL_INT overflows for now, so that I can focus on getting
> > the rest of these changes pulled in, but I don't mind adding a patch for
> > that later on.
> >
> > This series focuses on adding temperature rescaling support to the IIO
> > Analog Front End (AFE) driver.
> >
> > The first few patches address minor bugs in IIO inkernel functions, and
> > prepare the AFE driver for the additional features.
> >
> > The main changes to the AFE driver include an initial Kunit test suite,
> > support for IIO_VAL_INT_PLUS_{NANO,MICRO} scales, and support for RTDs
> > and temperature transducer sensors.
> >
> > Thanks for your time,
> > Liam
>
> Hi Liam,
Hi Jonathan,
> I'm fine with these. The comment about using the MICRO etc defines can
> be handled as as trivial follow up patch. Hopefully someone else can
> figure out the 0-day build issue as I didn't managed to.
I'll prepare the MICRO, NANO changes if we have a v10, otherwise I'll
send them as a follow up patch as you suggest.
I'll also try to find more time to dig more into that 0-day build issue.
> However, I've long ago lost track of the various precision discussions
> you and Peter were having so would like Peter's input before taking these.
That's my fault, apologies for letting this sit for so long.
> Thanks again for your persistence with this,
No worries, thanks for your patience :)
Cheers,
Liam
>
> Jonathan
>
> >
> > Changes since v8:
> > - reword comment
> > - fix erroneous 64-bit division
> > - optimize and use 32-bit divisions when values are know to not overflow
> > - keep IIO_VAL_FRACTIONAL scale when possible, if not default to fixed
> > point
> > - add test cases
> > - use nano precision in test cases
> > - simplify offset calculation in rtd_props()
> >
> > Changes since v7:
> > - drop gcd() logic in rescale_process_scale()
> > - use div_s64() instead of do_div() for signed 64-bit divisions
> > - combine IIO_VAL_FRACTIONAL and IIO_VAL_FRACTIONAL_LOG2 scale cases
> > - switch to INT_PLUS_NANO when accuracy is lost with FRACTIONAL scales
> > - rework test logic to allow for small relative error
> > - rename test variables to align error output messages
> >
> > Changes since v6:
> > - rework IIO_VAL_INT_PLUS_{NANO,MICRO} based on Peter's suggestion
> > - combine IIO_VAL_INT_PLUS_{NANO,MICRO} cases
> > - add test cases for negative IIO_VAL_INT_PLUS_{NANO,MICRO} corner cases
> > - force use of positive integers with gcd()
> > - reduce risk of integer overflow in IIO_VAL_FRACTIONAL_LOG2
> > - fix duplicate symbol build error
> > - apply Reviewed-by
> >
> > Changes since v5:
> > - add include/linux/iio/afe/rescale.h
> > - expose functions use to process scale and offset
> > - add basic iio-rescale kunit test cases
> > - fix integer overflow case
> > - improve precision for IIO_VAL_FRACTIONAL_LOG2
> >
> > Changes since v4:
> > - only use gcd() when necessary in overflow mitigation
> > - fix INT_PLUS_{MICRO,NANO} support
> > - apply Reviewed-by
> > - fix temperature-transducer bindings
> >
> > Changes since v3:
> > - drop unnecessary fallthrough statements
> > - drop redundant local variables in some calculations
> > - fix s64 divisions on 32bit platforms by using do_div
> > - add comment describing iio-rescaler offset calculation
> > - drop unnecessary MAINTAINERS entry
> >
> > Changes since v2:
> > - don't break implicit offset truncations
> > - make a best effort to get a valid value for fractional types
> > - drop return value change in iio_convert_raw_to_processed_unlocked()
> > - don't rely on processed value for offset calculation
> > - add INT_PLUS_{MICRO,NANO} support in iio-rescale
> > - revert generic implementation in favor of temperature-sense-rtd and
> > temperature-transducer
> > - add separate section to MAINTAINERS file
> >
> > Changes since v1:
> > - rebase on latest iio `testing` branch
> > - also apply consumer scale on integer channel scale types
> > - don't break implicit truncation in processed channel offset
> > calculation
> > - drop temperature AFE flavors in favor of a simpler generic
> > implementation
> >
> > Liam Beguin (14):
> > iio: inkern: apply consumer scale on IIO_VAL_INT cases
> > iio: inkern: apply consumer scale when no channel scale is available
> > iio: inkern: make a best effort on offset calculation
> > iio: afe: rescale: expose scale processing function
> > iio: afe: rescale: add INT_PLUS_{MICRO,NANO} support
> > iio: afe: rescale: add offset support
> > iio: afe: rescale: use s64 for temporary scale calculations
> > iio: afe: rescale: reduce risk of integer overflow
> > iio: afe: rescale: fix accuracy for small fractional scales
> > iio: test: add basic tests for the iio-rescale driver
> > iio: afe: rescale: add RTD temperature sensor support
> > iio: afe: rescale: add temperature transducers
> > dt-bindings: iio: afe: add bindings for temperature-sense-rtd
> > dt-bindings: iio: afe: add bindings for temperature transducers
> >
> > .../iio/afe/temperature-sense-rtd.yaml | 101 +++
> > .../iio/afe/temperature-transducer.yaml | 114 +++
> > drivers/iio/afe/iio-rescale.c | 271 ++++++-
> > drivers/iio/inkern.c | 40 +-
> > drivers/iio/test/Kconfig | 10 +
> > drivers/iio/test/Makefile | 1 +
> > drivers/iio/test/iio-test-rescale.c | 705 ++++++++++++++++++
> > include/linux/iio/afe/rescale.h | 34 +
> > 8 files changed, 1232 insertions(+), 44 deletions(-)
> > create mode 100644 Documentation/devicetree/bindings/iio/afe/temperature-sense-rtd.yaml
> > create mode 100644 Documentation/devicetree/bindings/iio/afe/temperature-transducer.yaml
> > create mode 100644 drivers/iio/test/iio-test-rescale.c
> > create mode 100644 include/linux/iio/afe/rescale.h
> >
> > Range-diff against v8:
> > 1: 42a7a1047edc = 1: ae3cc93baee6 iio: inkern: apply consumer scale on IIO_VAL_INT cases
> > 2: a1cd89fdad11 = 2: 06f66e7f7403 iio: inkern: apply consumer scale when no channel scale is available
> > 3: ed0721fb6bd1 = 3: 2dbf6b3bbaeb iio: inkern: make a best effort on offset calculation
> > 4: f8fb78bb1112 = 4: b083cf307268 iio: afe: rescale: expose scale processing function
> > 5: 504b7a3f830b ! 5: a0bde29ecc8c iio: afe: rescale: add INT_PLUS_{MICRO,NANO} support
> > @@ drivers/iio/afe/iio-rescale.c: int rescale_process_scale(struct rescale *rescale
> > + else
> > + mult = 1000000LL;
> > + /*
> > -+ * For IIO_VAL_INT_PLUS_{MICRO,NANO} scale types if *val OR
> > -+ * *val2 is negative the schan scale is negative
> > ++ * For IIO_VAL_INT_PLUS_{MICRO,NANO} scale types if either *val
> > ++ * OR *val2 is negative the schan scale is negative, i.e.
> > ++ * *val = 1 and *val2 = -0.5 yields -1.5 not -0.5.
> > + */
> > + neg = *val < 0 || *val2 < 0;
> > +
> > 6: c254e9ae813e = 6: c3d0e6248033 iio: afe: rescale: add offset support
> > 7: ee8814d6abe4 = 7: 2a81fa735103 iio: afe: rescale: use s64 for temporary scale calculations
> > 8: 62cdcfbc9836 = 8: 8315548d0fce iio: afe: rescale: reduce risk of integer overflow
> > 9: 88309a5136ee ! 9: 223ed0569cd2 iio: afe: rescale: fix accuracy for small fractional scales
> > @@ drivers/iio/afe/iio-rescale.c: int rescale_process_scale(struct rescale *rescale
> > +
> > + tmp = div_s64_rem(tmp, 1000000000LL, &rem);
> > *val = tmp;
> > +- return scale_type;
> > ++
> > ++ if (!rem)
> > ++ return scale_type;
> > +
> > -+ /*
> > -+ * For small values, the approximation can be costly,
> > -+ * change scale type to maintain accuracy.
> > -+ *
> > -+ * 100 vs. 10000000 NANO caps the error to about 100 ppm.
> > -+ */
> > + if (scale_type == IIO_VAL_FRACTIONAL)
> > + tmp = *val2;
> > + else
> > + tmp = 1 << *val2;
> > +
> > -+ if (abs(rem) > 10000000 && abs(*val / tmp) < 100) {
> > -+ *val = div_s64_rem(*val, tmp, &rem2);
> > -+
> > -+ *val2 = div_s64(rem, tmp);
> > -+ if (rem2)
> > -+ *val2 += div_s64(rem2 * 1000000000LL, tmp);
> > ++ rem2 = *val % (int)tmp;
> > ++ *val = *val / (int)tmp;
> > +
> > -+ return IIO_VAL_INT_PLUS_NANO;
> > -+ }
> > ++ *val2 = rem / (int)tmp;
> > ++ if (rem2)
> > ++ *val2 += div_s64((s64)rem2 * 1000000000LL, tmp);
> > +
> > - return scale_type;
> > ++ return IIO_VAL_INT_PLUS_NANO;
> > case IIO_VAL_INT_PLUS_NANO:
> > case IIO_VAL_INT_PLUS_MICRO:
> > + if (scale_type == IIO_VAL_INT_PLUS_NANO)
> > 10: fb505a9f42f1 ! 10: 90044efdf8be iio: test: add basic tests for the iio-rescale driver
> > @@ drivers/iio/test/Makefile
> > # Keep in alphabetical order
> > +obj-$(CONFIG_IIO_RESCALE_KUNIT_TEST) += iio-test-rescale.o ../afe/iio-rescale.o
> > obj-$(CONFIG_IIO_TEST_FORMAT) += iio-test-format.o
> > + CFLAGS_iio-test-format.o += $(DISABLE_STRUCTLEAK_PLUGIN)
> >
> > ## drivers/iio/test/iio-test-rescale.c (new) ##
> > @@
> > @@ drivers/iio/test/iio-test-rescale.c (new)
> > + * Use cases with small scales involving divisions
> > + */
> > + {
> > ++ .name = "small IIO_VAL_FRACTIONAL, 261/509 scaled by 90/1373754273",
> > ++ .numerator = 261,
> > ++ .denominator = 509,
> > ++ .schan_scale_type = IIO_VAL_FRACTIONAL,
> > ++ .schan_val = 90,
> > ++ .schan_val2 = 1373754273,
> > ++ .expected = "0.000000033594",
> > ++ },
> > ++ {
> > ++ .name = "small IIO_VAL_FRACTIONAL, 90/1373754273 scaled by 261/509",
> > ++ .numerator = 90,
> > ++ .denominator = 1373754273,
> > ++ .schan_scale_type = IIO_VAL_FRACTIONAL,
> > ++ .schan_val = 261,
> > ++ .schan_val2 = 509,
> > ++ .expected = "0.000000033594",
> > ++ },
> > ++ {
> > ++ .name = "small IIO_VAL_FRACTIONAL, 760/1373754273 scaled by 427/2727",
> > ++ .numerator = 760,
> > ++ .denominator = 1373754273,
> > ++ .schan_scale_type = IIO_VAL_FRACTIONAL,
> > ++ .schan_val = 427,
> > ++ .schan_val2 = 2727,
> > ++ .expected = "0.000000086626",
> > ++ },
> > ++ {
> > ++ .name = "small IIO_VAL_FRACTIONAL, 761/1373754273 scaled by 427/2727",
> > ++ .numerator = 761,
> > ++ .denominator = 1373754273,
> > ++ .schan_scale_type = IIO_VAL_FRACTIONAL,
> > ++ .schan_val = 427,
> > ++ .schan_val2 = 2727,
> > ++ .expected = "0.000000086740",
> > ++ },
> > ++ {
> > ++ .name = "small IIO_VAL_FRACTIONAL, 5/32768 scaled by 3/10000",
> > ++ .numerator = 5,
> > ++ .denominator = 32768,
> > ++ .schan_scale_type = IIO_VAL_FRACTIONAL,
> > ++ .schan_val = 3,
> > ++ .schan_val2 = 10000,
> > ++ .expected = "0.0000000457763671875",
> > ++ },
> > ++ {
> > + .name = "small IIO_VAL_FRACTIONAL, 0 < scale < 1",
> > + .numerator = 6,
> > + .denominator = 6,
> > @@ drivers/iio/test/iio-test-rescale.c (new)
> > + .expected = "-1.3333333333333333",
> > + },
> > + {
> > ++ .name = "small IIO_VAL_FRACTIONAL_LOG2, 760/32768 scaled by 15/22",
> > ++ .numerator = 760,
> > ++ .denominator = 32768,
> > ++ .schan_scale_type = IIO_VAL_FRACTIONAL_LOG2,
> > ++ .schan_val = 15,
> > ++ .schan_val2 = 22,
> > ++ .expected = "0.000000082946",
> > ++ },
> > ++ {
> > ++ .name = "small IIO_VAL_FRACTIONAL_LOG2, 761/32768 scaled by 15/22",
> > ++ .numerator = 761,
> > ++ .denominator = 32768,
> > ++ .schan_scale_type = IIO_VAL_FRACTIONAL_LOG2,
> > ++ .schan_val = 15,
> > ++ .schan_val2 = 22,
> > ++ .expected = "0.000000083055",
> > ++ },
> > ++ {
> > + .name = "small IIO_VAL_FRACTIONAL_LOG2, 0 < scale < 1",
> > + .numerator = 16,
> > + .denominator = 3,
> > @@ drivers/iio/test/iio-test-rescale.c (new)
> > +KUNIT_ARRAY_PARAM(iio_rescale_offset, offset_cases, case_to_desc);
> > +
> > +/**
> > -+ * iio_str_to_micro() - Parse a fixed-point string to get an
> > -+ * IIO_VAL_INT_PLUS_MICRO value
> > ++ * iio_str_to_nano() - Parse a fixed-point string to get an
> > ++ * IIO_VAL_INT_PLUS_NANO value
> > + * @str: The string to parse
> > -+ * @micro: The number as an integer
> > ++ * @nano: The number as an integer
> > + *
> > + * Returns 0 on success, or a negative error code if the string cound not be
> > + * parsed.
> > + */
> > -+static int iio_str_to_micro(const char *str, s64 *micro)
> > ++static int iio_str_to_nano(const char *str, s64 *nano)
> > +{
> > -+ int fract_mult = 100000LL;
> > ++ int fract_mult = 100000000LL;
> > + int tmp, tmp2;
> > + int ret = 0;
> > +
> > @@ drivers/iio/test/iio-test-rescale.c (new)
> > + if (tmp < 0)
> > + tmp2 *= -1;
> > +
> > -+ *micro = (s64)tmp * 10 * fract_mult + tmp2;
> > ++ *nano = (s64)tmp * 10 * fract_mult + tmp2;
> > +
> > + return ret;
> > +}
> > +
> > +/**
> > -+ * iio_test_relative_error_ppm() - Compute relative error (in ppm) between two
> > -+ * fixed-point strings
> > ++ * iio_test_relative_error_ppm() - Compute relative error (in parts-per-million)
> > ++ * between two fixed-point strings
> > + * @real_str: The real value as a string
> > + * @exp_str: The expected value as a string
> > + *
> > + * Returns a negative error code if the strings cound not be parsed, or the
> > -+ * relative error in ppm.
> > ++ * relative error in parts-per-million.
> > + */
> > +static int iio_test_relative_error_ppm(const char *real_str, const char *exp_str)
> > +{
> > + s64 real, exp, err;
> > + int ret;
> > +
> > -+ ret = iio_str_to_micro(real_str, &real);
> > ++ ret = iio_str_to_nano(real_str, &real);
> > + if (ret < 0)
> > + return ret;
> > +
> > -+ ret = iio_str_to_micro(exp_str, &exp);
> > ++ ret = iio_str_to_nano(exp_str, &exp);
> > + if (ret < 0)
> > + return ret;
> > +
> > ++ if (!exp) {
> > ++ pr_err("Expected value is null, relative error is undefined\n");
> > ++ return -EINVAL;
> > ++ }
> > ++
> > + err = 1000000 * abs(exp - real);
> > + err = div64_u64(err, abs(exp));
> > + return (int)err;
> > @@ drivers/iio/test/iio-test-rescale.c (new)
> > + rel_ppm = iio_test_relative_error_ppm(buff, t->expected);
> > + KUNIT_EXPECT_GE_MSG(test, rel_ppm, 0, "failed to compute ppm\n");
> > +
> > -+ KUNIT_EXPECT_LT_MSG(test, rel_ppm, 500,
> > ++ KUNIT_EXPECT_EQ_MSG(test, rel_ppm, 0,
> > + "\t real=%s"
> > + "\texpected=%s\n",
> > + buff, t->expected);
> > 11: 050487186e14 = 11: c4ed463e5fb0 iio: afe: rescale: add RTD temperature sensor support
> > 12: f36a44a5d898 ! 12: ff2f0dc248a7 iio: afe: rescale: add temperature transducers
> > @@ drivers/iio/afe/iio-rescale.c: static int rescale_temp_sense_rtd_props(struct de
> > + s32 offset = 0;
> > + s32 sense = 1;
> > + s32 alpha;
> > -+ s64 tmp;
> > + int ret;
> > +
> > + device_property_read_u32(dev, "sense-offset-millicelsius", &offset);
> > @@ drivers/iio/afe/iio-rescale.c: static int rescale_temp_sense_rtd_props(struct de
> > + rescale->numerator = 1000000;
> > + rescale->denominator = alpha * sense;
> > +
> > -+ tmp = (s64)offset * (s64)alpha * (s64)sense;
> > -+ rescale->offset = div_s64(tmp, (s32)1000000);
> > ++ rescale->offset = div_s64((s64)offset * rescale->denominator,
> > ++ rescale->numerator);
> > +
> > + return 0;
> > +}
> > 13: 63be647fd110 = 13: 84bc1f7d1ab5 dt-bindings: iio: afe: add bindings for temperature-sense-rtd
> > 14: c2f5c19dece3 = 14: 1b76cfb37e23 dt-bindings: iio: afe: add bindings for temperature transducers
> >
> > base-commit: 2b6bff0b122785f09cfbdc34b1aa9edceea6e4c1
>
Hi Liam!
On 2021-11-15 04:43, Liam Beguin wrote:
> Hi Jonathan, Peter,
>
> Apologies for not getting back to you sooner. I got caught up on other
> work and wasn't able to dedicate time to this earlier. Hopefully, this
> time around, I'll be able to get this to the finish line :-)
>
> I left out IIO_VAL_INT overflows for now, so that I can focus on getting
> the rest of these changes pulled in, but I don't mind adding a patch for
> that later on.
>
> This series focuses on adding temperature rescaling support to the IIO
> Analog Front End (AFE) driver.
>
> The first few patches address minor bugs in IIO inkernel functions, and
> prepare the AFE driver for the additional features.
>
> The main changes to the AFE driver include an initial Kunit test suite,
> support for IIO_VAL_INT_PLUS_{NANO,MICRO} scales, and support for RTDs
> and temperature transducer sensors.
>
> Thanks for your time,
And thanks for yours!
> Liam
>
> Changes since v8:
> - reword comment
> - fix erroneous 64-bit division
> - optimize and use 32-bit divisions when values are know to not overflow
> - keep IIO_VAL_FRACTIONAL scale when possible, if not default to fixed
> point
This is not what is going on. Patch 9/14 will convert all fractional
scales to fixed point. But I would really like if you in the "reduce
risk of integer overflow" patch (8/14) would hold true to the above
and keep the fractional scale when possible and only fall back to
the less precise fractional-log case if any of the multiplications
needed for an exact fractional scale causes overflow.
The v8 discussion concluded that this was a valid approach, right?
I know you also said that the core exposes the scale with nano
precision in sysfs anyway, but that is not true for in-kernel
consumers. They have an easier time reading the "real" scale value
compared to going via the string representation of fixed point
returned from iio_format_value. At least the rescaler itself does so,
which means that chaining rescalers might suffer needless accuracy
degradation.
So, please add the overflow fallback thingy right away, it would make
me feel much better.
> - add test cases
> - use nano precision in test cases
> - simplify offset calculation in rtd_props()
>
> Changes since v7:
> - drop gcd() logic in rescale_process_scale()
> - use div_s64() instead of do_div() for signed 64-bit divisions
> - combine IIO_VAL_FRACTIONAL and IIO_VAL_FRACTIONAL_LOG2 scale cases
> - switch to INT_PLUS_NANO when accuracy is lost with FRACTIONAL scales
> - rework test logic to allow for small relative error
> - rename test variables to align error output messages
>
> Changes since v6:
> - rework IIO_VAL_INT_PLUS_{NANO,MICRO} based on Peter's suggestion
> - combine IIO_VAL_INT_PLUS_{NANO,MICRO} cases
> - add test cases for negative IIO_VAL_INT_PLUS_{NANO,MICRO} corner cases
> - force use of positive integers with gcd()
> - reduce risk of integer overflow in IIO_VAL_FRACTIONAL_LOG2
> - fix duplicate symbol build error
> - apply Reviewed-by
>
> Changes since v5:
> - add include/linux/iio/afe/rescale.h
> - expose functions use to process scale and offset
> - add basic iio-rescale kunit test cases
> - fix integer overflow case
> - improve precision for IIO_VAL_FRACTIONAL_LOG2
>
> Changes since v4:
> - only use gcd() when necessary in overflow mitigation
> - fix INT_PLUS_{MICRO,NANO} support
> - apply Reviewed-by
> - fix temperature-transducer bindings
>
> Changes since v3:
> - drop unnecessary fallthrough statements
> - drop redundant local variables in some calculations
> - fix s64 divisions on 32bit platforms by using do_div
> - add comment describing iio-rescaler offset calculation
> - drop unnecessary MAINTAINERS entry
>
> Changes since v2:
> - don't break implicit offset truncations
> - make a best effort to get a valid value for fractional types
> - drop return value change in iio_convert_raw_to_processed_unlocked()
> - don't rely on processed value for offset calculation
> - add INT_PLUS_{MICRO,NANO} support in iio-rescale
> - revert generic implementation in favor of temperature-sense-rtd and
> temperature-transducer
> - add separate section to MAINTAINERS file
>
> Changes since v1:
> - rebase on latest iio `testing` branch
> - also apply consumer scale on integer channel scale types
> - don't break implicit truncation in processed channel offset
> calculation
> - drop temperature AFE flavors in favor of a simpler generic
> implementation
>
> Liam Beguin (14):
> iio: inkern: apply consumer scale on IIO_VAL_INT cases
> iio: inkern: apply consumer scale when no channel scale is available
> iio: inkern: make a best effort on offset calculation
> iio: afe: rescale: expose scale processing function
> iio: afe: rescale: add INT_PLUS_{MICRO,NANO} support
> iio: afe: rescale: add offset support
> iio: afe: rescale: use s64 for temporary scale calculations
> iio: afe: rescale: reduce risk of integer overflow
> iio: afe: rescale: fix accuracy for small fractional scales
Can you please swap the order of these two patches? (i.e. "reduce
risk..." and "fix accuracy...")
Basically, I think the accuracy of the IIO_VAL_FRACTIONAL_LOG2
case should be improved before the IIO_VAL_FRACTIONAL case is
joined with it. I.e. swap the order of 8/14 and 9/14 (or almost,
you need to also move the addition of the
scale_type == IIO_VAL_FRACTIONAL condition to the other patch in
order for it to make sense).
That's all I'm finding. But then again, I don't know what to do
about the 0day report on 10/14. It does say that it's a W=1
build, maybe we need not worry about it?
Cheers,
Peter
> iio: test: add basic tests for the iio-rescale driver
> iio: afe: rescale: add RTD temperature sensor support
> iio: afe: rescale: add temperature transducers
> dt-bindings: iio: afe: add bindings for temperature-sense-rtd
> dt-bindings: iio: afe: add bindings for temperature transducers
On Mon, 22 Nov 2021 01:53:44 +0100
Peter Rosin <[email protected]> wrote:
> Hi Liam!
>
> On 2021-11-15 04:43, Liam Beguin wrote:
> > Hi Jonathan, Peter,
> >
> > Apologies for not getting back to you sooner. I got caught up on other
> > work and wasn't able to dedicate time to this earlier. Hopefully, this
> > time around, I'll be able to get this to the finish line :-)
> >
> > I left out IIO_VAL_INT overflows for now, so that I can focus on getting
> > the rest of these changes pulled in, but I don't mind adding a patch for
> > that later on.
> >
> > This series focuses on adding temperature rescaling support to the IIO
> > Analog Front End (AFE) driver.
> >
> > The first few patches address minor bugs in IIO inkernel functions, and
> > prepare the AFE driver for the additional features.
> >
> > The main changes to the AFE driver include an initial Kunit test suite,
> > support for IIO_VAL_INT_PLUS_{NANO,MICRO} scales, and support for RTDs
> > and temperature transducer sensors.
> >
> > Thanks for your time,
>
> And thanks for yours!
>
> > Liam
> >
> > Changes since v8:
> > - reword comment
> > - fix erroneous 64-bit division
> > - optimize and use 32-bit divisions when values are know to not overflow
> > - keep IIO_VAL_FRACTIONAL scale when possible, if not default to fixed
> > point
>
> This is not what is going on. Patch 9/14 will convert all fractional
> scales to fixed point. But I would really like if you in the "reduce
> risk of integer overflow" patch (8/14) would hold true to the above
> and keep the fractional scale when possible and only fall back to
> the less precise fractional-log case if any of the multiplications
> needed for an exact fractional scale causes overflow.
>
> The v8 discussion concluded that this was a valid approach, right?
>
> I know you also said that the core exposes the scale with nano
> precision in sysfs anyway, but that is not true for in-kernel
> consumers. They have an easier time reading the "real" scale value
> compared to going via the string representation of fixed point
> returned from iio_format_value. At least the rescaler itself does so,
> which means that chaining rescalers might suffer needless accuracy
> degradation.
>
> So, please add the overflow fallback thingy right away, it would make
> me feel much better.
>
> > - add test cases
> > - use nano precision in test cases
> > - simplify offset calculation in rtd_props()
> >
> > Changes since v7:
> > - drop gcd() logic in rescale_process_scale()
> > - use div_s64() instead of do_div() for signed 64-bit divisions
> > - combine IIO_VAL_FRACTIONAL and IIO_VAL_FRACTIONAL_LOG2 scale cases
> > - switch to INT_PLUS_NANO when accuracy is lost with FRACTIONAL scales
> > - rework test logic to allow for small relative error
> > - rename test variables to align error output messages
> >
> > Changes since v6:
> > - rework IIO_VAL_INT_PLUS_{NANO,MICRO} based on Peter's suggestion
> > - combine IIO_VAL_INT_PLUS_{NANO,MICRO} cases
> > - add test cases for negative IIO_VAL_INT_PLUS_{NANO,MICRO} corner cases
> > - force use of positive integers with gcd()
> > - reduce risk of integer overflow in IIO_VAL_FRACTIONAL_LOG2
> > - fix duplicate symbol build error
> > - apply Reviewed-by
> >
> > Changes since v5:
> > - add include/linux/iio/afe/rescale.h
> > - expose functions use to process scale and offset
> > - add basic iio-rescale kunit test cases
> > - fix integer overflow case
> > - improve precision for IIO_VAL_FRACTIONAL_LOG2
> >
> > Changes since v4:
> > - only use gcd() when necessary in overflow mitigation
> > - fix INT_PLUS_{MICRO,NANO} support
> > - apply Reviewed-by
> > - fix temperature-transducer bindings
> >
> > Changes since v3:
> > - drop unnecessary fallthrough statements
> > - drop redundant local variables in some calculations
> > - fix s64 divisions on 32bit platforms by using do_div
> > - add comment describing iio-rescaler offset calculation
> > - drop unnecessary MAINTAINERS entry
> >
> > Changes since v2:
> > - don't break implicit offset truncations
> > - make a best effort to get a valid value for fractional types
> > - drop return value change in iio_convert_raw_to_processed_unlocked()
> > - don't rely on processed value for offset calculation
> > - add INT_PLUS_{MICRO,NANO} support in iio-rescale
> > - revert generic implementation in favor of temperature-sense-rtd and
> > temperature-transducer
> > - add separate section to MAINTAINERS file
> >
> > Changes since v1:
> > - rebase on latest iio `testing` branch
> > - also apply consumer scale on integer channel scale types
> > - don't break implicit truncation in processed channel offset
> > calculation
> > - drop temperature AFE flavors in favor of a simpler generic
> > implementation
> >
> > Liam Beguin (14):
> > iio: inkern: apply consumer scale on IIO_VAL_INT cases
> > iio: inkern: apply consumer scale when no channel scale is available
> > iio: inkern: make a best effort on offset calculation
> > iio: afe: rescale: expose scale processing function
> > iio: afe: rescale: add INT_PLUS_{MICRO,NANO} support
> > iio: afe: rescale: add offset support
> > iio: afe: rescale: use s64 for temporary scale calculations
> > iio: afe: rescale: reduce risk of integer overflow
> > iio: afe: rescale: fix accuracy for small fractional scales
>
> Can you please swap the order of these two patches? (i.e. "reduce
> risk..." and "fix accuracy...")
>
> Basically, I think the accuracy of the IIO_VAL_FRACTIONAL_LOG2
> case should be improved before the IIO_VAL_FRACTIONAL case is
> joined with it. I.e. swap the order of 8/14 and 9/14 (or almost,
> you need to also move the addition of the
> scale_type == IIO_VAL_FRACTIONAL condition to the other patch in
> order for it to make sense).
>
> That's all I'm finding. But then again, I don't know what to do
> about the 0day report on 10/14. It does say that it's a W=1
> build, maybe we need not worry about it?
W=1 won't affect that undefined symbols error.
I'd be cynical and assume it's a random issue, post a v10 perhaps
with a note in the cover letter on this.
Jonathan
>
> Cheers,
> Peter
>
> > iio: test: add basic tests for the iio-rescale driver
> > iio: afe: rescale: add RTD temperature sensor support
> > iio: afe: rescale: add temperature transducers
> > dt-bindings: iio: afe: add bindings for temperature-sense-rtd
> > dt-bindings: iio: afe: add bindings for temperature transducers
>
Hi Peter,
On Mon, Nov 22, 2021 at 01:53:44AM +0100, Peter Rosin wrote:
> Hi Liam!
>
> On 2021-11-15 04:43, Liam Beguin wrote:
> > Hi Jonathan, Peter,
> >
> > Apologies for not getting back to you sooner. I got caught up on other
> > work and wasn't able to dedicate time to this earlier. Hopefully, this
> > time around, I'll be able to get this to the finish line :-)
> >
> > I left out IIO_VAL_INT overflows for now, so that I can focus on getting
> > the rest of these changes pulled in, but I don't mind adding a patch for
> > that later on.
> >
> > This series focuses on adding temperature rescaling support to the IIO
> > Analog Front End (AFE) driver.
> >
> > The first few patches address minor bugs in IIO inkernel functions, and
> > prepare the AFE driver for the additional features.
> >
> > The main changes to the AFE driver include an initial Kunit test suite,
> > support for IIO_VAL_INT_PLUS_{NANO,MICRO} scales, and support for RTDs
> > and temperature transducer sensors.
> >
> > Thanks for your time,
>
> And thanks for yours!
>
> > Liam
> >
> > Changes since v8:
> > - reword comment
> > - fix erroneous 64-bit division
> > - optimize and use 32-bit divisions when values are know to not overflow
> > - keep IIO_VAL_FRACTIONAL scale when possible, if not default to fixed
> > point
>
> This is not what is going on. Patch 9/14 will convert all fractional
> scales to fixed point. But I would really like if you in the "reduce
> risk of integer overflow" patch (8/14) would hold true to the above
> and keep the fractional scale when possible and only fall back to
> the less precise fractional-log case if any of the multiplications
> needed for an exact fractional scale causes overflow.
Thanks for looking at these patches again.
> The v8 discussion concluded that this was a valid approach, right?
Yes, I remember you saying that you'd be more comfortable keeping the
IIO_VAL_FRACTIONAL.
> I know you also said that the core exposes the scale with nano
> precision in sysfs anyway, but that is not true for in-kernel
> consumers. They have an easier time reading the "real" scale value
> compared to going via the string representation of fixed point
> returned from iio_format_value. At least the rescaler itself does so,
> which means that chaining rescalers might suffer needless accuracy
> degradation.
Agreed, that makes total sense.
If I'm not mistaken, the first condition in the case, if (!rem), will
return IIO_VAL_FRACTIONAL if the division is exact, keeping all the
precision. No?
> So, please add the overflow fallback thingy right away, it would make
> me feel much better.
>
> > - add test cases
> > - use nano precision in test cases
> > - simplify offset calculation in rtd_props()
> >
> > Changes since v7:
> > - drop gcd() logic in rescale_process_scale()
> > - use div_s64() instead of do_div() for signed 64-bit divisions
> > - combine IIO_VAL_FRACTIONAL and IIO_VAL_FRACTIONAL_LOG2 scale cases
> > - switch to INT_PLUS_NANO when accuracy is lost with FRACTIONAL scales
> > - rework test logic to allow for small relative error
> > - rename test variables to align error output messages
> >
> > Changes since v6:
> > - rework IIO_VAL_INT_PLUS_{NANO,MICRO} based on Peter's suggestion
> > - combine IIO_VAL_INT_PLUS_{NANO,MICRO} cases
> > - add test cases for negative IIO_VAL_INT_PLUS_{NANO,MICRO} corner cases
> > - force use of positive integers with gcd()
> > - reduce risk of integer overflow in IIO_VAL_FRACTIONAL_LOG2
> > - fix duplicate symbol build error
> > - apply Reviewed-by
> >
> > Changes since v5:
> > - add include/linux/iio/afe/rescale.h
> > - expose functions use to process scale and offset
> > - add basic iio-rescale kunit test cases
> > - fix integer overflow case
> > - improve precision for IIO_VAL_FRACTIONAL_LOG2
> >
> > Changes since v4:
> > - only use gcd() when necessary in overflow mitigation
> > - fix INT_PLUS_{MICRO,NANO} support
> > - apply Reviewed-by
> > - fix temperature-transducer bindings
> >
> > Changes since v3:
> > - drop unnecessary fallthrough statements
> > - drop redundant local variables in some calculations
> > - fix s64 divisions on 32bit platforms by using do_div
> > - add comment describing iio-rescaler offset calculation
> > - drop unnecessary MAINTAINERS entry
> >
> > Changes since v2:
> > - don't break implicit offset truncations
> > - make a best effort to get a valid value for fractional types
> > - drop return value change in iio_convert_raw_to_processed_unlocked()
> > - don't rely on processed value for offset calculation
> > - add INT_PLUS_{MICRO,NANO} support in iio-rescale
> > - revert generic implementation in favor of temperature-sense-rtd and
> > temperature-transducer
> > - add separate section to MAINTAINERS file
> >
> > Changes since v1:
> > - rebase on latest iio `testing` branch
> > - also apply consumer scale on integer channel scale types
> > - don't break implicit truncation in processed channel offset
> > calculation
> > - drop temperature AFE flavors in favor of a simpler generic
> > implementation
> >
> > Liam Beguin (14):
> > iio: inkern: apply consumer scale on IIO_VAL_INT cases
> > iio: inkern: apply consumer scale when no channel scale is available
> > iio: inkern: make a best effort on offset calculation
> > iio: afe: rescale: expose scale processing function
> > iio: afe: rescale: add INT_PLUS_{MICRO,NANO} support
> > iio: afe: rescale: add offset support
> > iio: afe: rescale: use s64 for temporary scale calculations
> > iio: afe: rescale: reduce risk of integer overflow
> > iio: afe: rescale: fix accuracy for small fractional scales
>
> Can you please swap the order of these two patches? (i.e. "reduce
> risk..." and "fix accuracy...")
>
> Basically, I think the accuracy of the IIO_VAL_FRACTIONAL_LOG2
> case should be improved before the IIO_VAL_FRACTIONAL case is
> joined with it. I.e. swap the order of 8/14 and 9/14 (or almost,
> you need to also move the addition of the
> scale_type == IIO_VAL_FRACTIONAL condition to the other patch in
> order for it to make sense).
Makes sense! I'll swap the order of these commits.
> That's all I'm finding. But then again, I don't know what to do
> about the 0day report on 10/14. It does say that it's a W=1
> build, maybe we need not worry about it?
I didn't have a chance to look into that more, but will now.
Cheers,
Liam
> Cheers,
> Peter
>
> > iio: test: add basic tests for the iio-rescale driver
> > iio: afe: rescale: add RTD temperature sensor support
> > iio: afe: rescale: add temperature transducers
> > dt-bindings: iio: afe: add bindings for temperature-sense-rtd
> > dt-bindings: iio: afe: add bindings for temperature transducers
>
Hi!
On 2021-11-27 21:27, Liam Beguin wrote:
> Hi Peter,
>
> On Mon, Nov 22, 2021 at 01:53:44AM +0100, Peter Rosin wrote:
>> Hi Liam!
>>
>> On 2021-11-15 04:43, Liam Beguin wrote:
>>> Hi Jonathan, Peter,
snip
>>> - keep IIO_VAL_FRACTIONAL scale when possible, if not default to fixed
>>> point
>>
>> This is not what is going on. Patch 9/14 will convert all fractional
>> scales to fixed point. But I would really like if you in the "reduce
>> risk of integer overflow" patch (8/14) would hold true to the above
>> and keep the fractional scale when possible and only fall back to
>> the less precise fractional-log case if any of the multiplications
>> needed for an exact fractional scale causes overflow.
>
> Thanks for looking at these patches again.
>
>> The v8 discussion concluded that this was a valid approach, right?
>
> Yes, I remember you saying that you'd be more comfortable keeping the
> IIO_VAL_FRACTIONAL.
>
>> I know you also said that the core exposes the scale with nano
>> precision in sysfs anyway, but that is not true for in-kernel
>> consumers. They have an easier time reading the "real" scale value
>> compared to going via the string representation of fixed point
>> returned from iio_format_value. At least the rescaler itself does so,
>> which means that chaining rescalers might suffer needless accuracy
>> degradation.
>
> Agreed, that makes total sense.
>
> If I'm not mistaken, the first condition in the case, if (!rem), will
> return IIO_VAL_FRACTIONAL if the division is exact, keeping all the
> precision. No?
Only if the resulting scale fits in nine decimals. That's never the
case if you have primes other than 2 and 5 in the denominator (after
eliminating gcd of course). Which mean that if you chain one rescaler
doing 1/3 and one doing 3/1, you would get a combined scale of
0.999999999 instead of 3/3 if we take the approach of these patches.
So, what I'm after is that - for IIO_VAL_FRACTIONAL - not take the
multiply-by-1e9 code path /unless/ the existing fractional approach
overflows in either numerator or denominator (or both).
Side note: The same could be done for IIO_VAL_INT when the numerator
overflows (since the denominator cannot overflow), but I guess that
can be done later.
Cheers,
Peter
Hi Peter,
On Sun, Nov 28, 2021 at 10:17:50AM +0100, Peter Rosin wrote:
> Hi!
>
> On 2021-11-27 21:27, Liam Beguin wrote:
> > Hi Peter,
> >
> > On Mon, Nov 22, 2021 at 01:53:44AM +0100, Peter Rosin wrote:
> >> Hi Liam!
> >>
> >> On 2021-11-15 04:43, Liam Beguin wrote:
> >>> Hi Jonathan, Peter,
>
> snip
>
> >>> - keep IIO_VAL_FRACTIONAL scale when possible, if not default to fixed
> >>> point
> >>
> >> This is not what is going on. Patch 9/14 will convert all fractional
> >> scales to fixed point. But I would really like if you in the "reduce
> >> risk of integer overflow" patch (8/14) would hold true to the above
> >> and keep the fractional scale when possible and only fall back to
> >> the less precise fractional-log case if any of the multiplications
> >> needed for an exact fractional scale causes overflow.
> >
> > Thanks for looking at these patches again.
> >
> >> The v8 discussion concluded that this was a valid approach, right?
> >
> > Yes, I remember you saying that you'd be more comfortable keeping the
> > IIO_VAL_FRACTIONAL.
> >
> >> I know you also said that the core exposes the scale with nano
> >> precision in sysfs anyway, but that is not true for in-kernel
> >> consumers. They have an easier time reading the "real" scale value
> >> compared to going via the string representation of fixed point
> >> returned from iio_format_value. At least the rescaler itself does so,
> >> which means that chaining rescalers might suffer needless accuracy
> >> degradation.
> >
> > Agreed, that makes total sense.
> >
> > If I'm not mistaken, the first condition in the case, if (!rem), will
> > return IIO_VAL_FRACTIONAL if the division is exact, keeping all the
> > precision. No?
>
> Only if the resulting scale fits in nine decimals. That's never the
> case if you have primes other than 2 and 5 in the denominator (after
> eliminating gcd of course). Which mean that if you chain one rescaler
> doing 1/3 and one doing 3/1, you would get a combined scale of
> 0.999999999 instead of 3/3 if we take the approach of these patches.
>
> So, what I'm after is that - for IIO_VAL_FRACTIONAL - not take the
> multiply-by-1e9 code path /unless/ the existing fractional approach
> overflows in either numerator or denominator (or both).
Understood, I'll update based on this.
> Side note: The same could be done for IIO_VAL_INT when the numerator
> overflows (since the denominator cannot overflow), but I guess that
> can be done later.
Agreed, I don't mind working on this later but I'd like to focus on
getting the current changes in first.
Thanks,
Liam
> Cheers,
> Peter