Skip to contentSkip to conversation

iio: imu: st_lsm6dsx: Decouple sensor ODR from FIFO batch data rate

PATCH Superseded Francesco Lavra 6 participants 18 messages Oct 9, 2025 – Oct 18, 2025

Subsystems (2)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

cover iio: imu: st_lsm6dsx: Decouple sensor ODR from FIFO batch data rate

The rate at which accelerometer or gyroscope sensor samples are fed to the hardware FIFO (batch data rate, or BDR) does not have to coincide with the sensor sampling frequency (output data rate, or ODR); the only requirement is for the BDR to not be greater than the ODR. Having a BDR lower than the ODR is useful in cases where an application requires a high sampling rate for accurate detection of motion events (e.g. wakeup events), but wants to read sensor sample values from the device buffer at a lower data rate. This change set amends the st_lsm6dsx IIO driver to support the above use case. The first commit is a trivial fix to the source code comment that indicates what measurement unit is used for the odr field of struct st_lsm6ds_sensor, while the second commit introduces a new bdr field in the same struct to implement the new functionality.

Francesco Lavra (2):
iio: imu: st_lsm6dsx: Fix measurement unit for odr struct member
iio: imu: st_lsm6dsx: Decouple sensor ODR from FIFO batch data rate

 drivers/iio/imu/st_lsm6dsx/st_lsm6dsx.h       |  4 +-
 .../iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c    | 64 ++++++++++++++++---
 drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_core.c  |  9 ++-
 drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_shub.c  |  4 +-
 4 files changed, 67 insertions(+), 14 deletions(-)
Signature · 1 line
2.39.5
#1 lore raw

PATCH 1/2 iio: imu: st_lsm6dsx: Fix measurement unit for odr struct member

The odr field in struct st_lsm6dsx_sensor contains a data rate value expressed in mHz, not in Hz.

---
 drivers/iio/imu/st_lsm6dsx/st_lsm6dsx.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx.h +1 -1
index c225b246c8a5..bd366c6e282a 100644
@@ -365,7 +365,7 @@ enum st_lsm6dsx_fifo_mode {
  * @id: Sensor identifier.
  * @hw: Pointer to instance of struct st_lsm6dsx_hw.
  * @gain: Configured sensor sensitivity.
- * @odr: Output data rate of the sensor [Hz].
+ * @odr: Output data rate of the sensor [mHz].
  * @samples_to_discard: Number of samples to discard for filters settling time.
  * @watermark: Sensor watermark level.
  * @decimator: Sensor decimation factor.
Signature · 1 line
2.39.5
maintainer
#2 lore raw

Re: [PATCH 1/2] iio: imu: st_lsm6dsx: Fix measurement unit for odr struct member

Quoted 2 lines

The odr field in struct st_lsm6dsx_sensor contains a data rate value expressed in mHz, not in Hz.

ack, can you please the proper Fixes tag?

Regards,
Lorenzo

Quoted 22 lines
---
 drivers/iio/imu/st_lsm6dsx/st_lsm6dsx.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx.h +1 -1
index c225b246c8a5..bd366c6e282a 100644
@@ -365,7 +365,7 @@ enum st_lsm6dsx_fifo_mode {
  * @id: Sensor identifier.
  * @hw: Pointer to instance of struct st_lsm6dsx_hw.
  * @gain: Configured sensor sensitivity.
- * @odr: Output data rate of the sensor [Hz].
+ * @odr: Output data rate of the sensor [mHz].
  * @samples_to_discard: Number of samples to discard for filters settling time.
  * @watermark: Sensor watermark level.
  * @decimator: Sensor decimation factor.
Signature · 1 line
2.39.5
#3 lore raw

PATCH 2/2 iio: imu: st_lsm6dsx: Decouple sensor ODR from FIFO batch data rate

In reply to Francesco Lavra

The rate at which accelerometer or gyroscope sensor samples are fed to the hardware FIFO (batch data rate, or BDR) does not have to coincide with the sensor sampling frequency (output data rate, or ODR); the only requirement is for the BDR to not be greater than the ODR. Having a BDR lower than the ODR is useful in cases where an application requires a high sampling rate for accurate detection of motion events (e.g. wakeup events), but wants to read sensor sample values from the device buffer at a lower data rate. To support the above use case, add a sampling_frequency sysfs attribute to the buffer directory of st_lsm6dsx IIO devices, which controls the BDR for a given sensor independently from the "main" sampling_frequency attribute (which controls the ODR); introduce a new bdr field in struct st_lsm6dsx_sensor to keep track of the current BDR value, and use this field instead of the odr field in the code that deals with the FIFO data rate. In the sensor hub driver, make the bdr value always mirror the odr value, since there is no separate configuration setting to control the BDR for data produced by the sensor hub functionality.

---
 drivers/iio/imu/st_lsm6dsx/st_lsm6dsx.h       |  2 +
 .../iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c    | 64 ++++++++++++++++---
 drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_core.c  |  9 ++-
 drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_shub.c  |  4 +-
 4 files changed, 66 insertions(+), 13 deletions(-)
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx.h +2 -0
index bd366c6e282a..dc4aeea3a3b8 100644
@@ -366,6 +366,7 @@ enum st_lsm6dsx_fifo_mode {
  * @hw: Pointer to instance of struct st_lsm6dsx_hw.
  * @gain: Configured sensor sensitivity.
  * @odr: Output data rate of the sensor [mHz].
+ * @bdr: Batch data rate [mHz]
  * @samples_to_discard: Number of samples to discard for filters settling time.
  * @watermark: Sensor watermark level.
  * @decimator: Sensor decimation factor.
@@ -380,6 +381,7 @@ struct st_lsm6dsx_sensor {
 
 	u32 gain;
 	u32 odr;
+	u32 bdr;
 
 	u16 samples_to_discard;
 	u16 watermark;
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +56 -8

111 lines not on this page · raw message

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_core.c +6 -3
index c65ad49829e7..e4922578329e 100644
@@ -1847,10 +1847,13 @@ static int st_lsm6dsx_write_raw(struct iio_dev *iio_dev,
 
 		val = val * 1000 + val2 / 1000;
 		val = st_lsm6dsx_check_odr(sensor, val, &data);
-		if (val < 0)
+		if (val < 0) {
 			err = val;
-		else
+		} else {
 			sensor->odr = val;
+			/* the batch data rate must not exceed the sensor ODR */
+			sensor->bdr = min_t(u32, sensor->bdr, sensor->odr);
+		}
 		break;
 	}
 	default:
@@ -2383,7 +2386,7 @@ static struct iio_dev *st_lsm6dsx_alloc_iiodev(struct st_lsm6dsx_hw *hw,
 	sensor = iio_priv(iio_dev);
 	sensor->id = id;
 	sensor->hw = hw;
-	sensor->odr = hw->settings->odr_table[id].odr_avl[0].milli_hz;
+	sensor->odr = sensor->bdr = hw->settings->odr_table[id].odr_avl[0].milli_hz;
 	sensor->gain = hw->settings->fs_table[id].fs_avl[0].gain;
 	sensor->watermark = 1;
 
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_shub.c +2 -2
index 3c5e65dc0f97..01d73002e888 100644
@@ -639,7 +639,7 @@ __st_lsm6dsx_shub_write_raw(struct iio_dev *iio_dev,
 			return odr;
 
 		sensor->ext_info.slv_odr = val;
-		sensor->odr = odr;
+		sensor->odr = sensor->bdr = odr;
 		return 0;
 	}
 	case IIO_CHAN_INFO_SCALE:
@@ -745,7 +745,7 @@ st_lsm6dsx_shub_alloc_iiodev(struct st_lsm6dsx_hw *hw,
 	sensor = iio_priv(iio_dev);
 	sensor->id = id;
 	sensor->hw = hw;
-	sensor->odr = hw->settings->odr_table[ref_id].odr_avl[0].milli_hz;
+	sensor->odr = sensor->bdr = hw->settings->odr_table[ref_id].odr_avl[0].milli_hz;
 	sensor->ext_info.slv_odr = info->odr_table.odr_avl[0].milli_hz;
 	sensor->gain = info->fs_table.fs_avl[0].gain;
 	sensor->ext_info.settings = info;
Signature · 1 line
2.39.5
maintainer
#4 lore raw
Quoted 8 lines

The rate at which accelerometer or gyroscope sensor samples are fed to the hardware FIFO (batch data rate, or BDR) does not have to coincide with the sensor sampling frequency (output data rate, or ODR); the only requirement is for the BDR to not be greater than the ODR. Having a BDR lower than the ODR is useful in cases where an application requires a high sampling rate for accurate detection of motion events (e.g. wakeup events), but wants to read sensor sample values from the device buffer at a lower data rate.

can you please provide more details here? Are you using the hw fifo to read data? If we configure the hw fifo according to the BDR (even assuming the watermark is set 1) the hw will generate interrupts according to the BDR (bdr < odr).

Quoted 113 lines

To support the above use case, add a sampling_frequency sysfs attribute to the buffer directory of st_lsm6dsx IIO devices, which controls the BDR for a given sensor independently from the "main" sampling_frequency attribute (which controls the ODR); introduce a new bdr field in struct st_lsm6dsx_sensor to keep track of the current BDR value, and use this field instead of the odr field in the code that deals with the FIFO data rate. In the sensor hub driver, make the bdr value always mirror the odr value, since there is no separate configuration setting to control the BDR for data produced by the sensor hub functionality.

---
 drivers/iio/imu/st_lsm6dsx/st_lsm6dsx.h       |  2 +
 .../iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c    | 64 ++++++++++++++++---
 drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_core.c  |  9 ++-
 drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_shub.c  |  4 +-
 4 files changed, 66 insertions(+), 13 deletions(-)
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx.h +2 -0
index bd366c6e282a..dc4aeea3a3b8 100644
@@ -366,6 +366,7 @@ enum st_lsm6dsx_fifo_mode {
  * @hw: Pointer to instance of struct st_lsm6dsx_hw.
  * @gain: Configured sensor sensitivity.
  * @odr: Output data rate of the sensor [mHz].
+ * @bdr: Batch data rate [mHz]
  * @samples_to_discard: Number of samples to discard for filters settling time.
  * @watermark: Sensor watermark level.
  * @decimator: Sensor decimation factor.
@@ -380,6 +381,7 @@ struct st_lsm6dsx_sensor {
 
 	u32 gain;
 	u32 odr;
+	u32 bdr;
 
 	u16 samples_to_discard;
 	u16 watermark;
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +20 -6
index 8a9d2593576a..5912ea76d493 100644
@@ -56,6 +56,7 @@
 #include <linux/iio/kfifo_buf.h>
 #include <linux/iio/iio.h>
 #include <linux/iio/buffer.h>
+#include <linux/iio/sysfs.h>
 #include <linux/regmap.h>
 #include <linux/bitfield.h>
 
@@ -105,7 +106,7 @@ static int
 st_lsm6dsx_get_decimator_val(struct st_lsm6dsx_sensor *sensor, u32 max_odr)
 {
 	const int max_size = ARRAY_SIZE(st_lsm6dsx_decimator_table);
-	u32 decimator =  max_odr / sensor->odr;
+	u32 decimator =  max_odr / sensor->bdr;
 	int i;
 
 	if (decimator > 1)
@@ -136,14 +137,14 @@ static void st_lsm6dsx_get_max_min_odr(struct st_lsm6dsx_hw *hw,
 		if (!(hw->enable_mask & BIT(sensor->id)))
 			continue;
 
-		*max_odr = max_t(u32, *max_odr, sensor->odr);
-		*min_odr = min_t(u32, *min_odr, sensor->odr);
+		*max_odr = max_t(u32, *max_odr, sensor->bdr);
+		*min_odr = min_t(u32, *min_odr, sensor->bdr);
 	}
 }
 
 static u8 st_lsm6dsx_get_sip(struct st_lsm6dsx_sensor *sensor, u32 min_odr)
 {
-	u8 sip = sensor->odr / min_odr;
+	u8 sip = sensor->bdr / min_odr;
 
 	return sip > 1 ? round_down(sip, 2) : sip;
 }
@@ -231,7 +232,7 @@ static int st_lsm6dsx_set_fifo_odr(struct st_lsm6dsx_sensor *sensor,
 		if (enable) {
 			int err;
 
-			err = st_lsm6dsx_check_odr(sensor, sensor->odr,
+			err = st_lsm6dsx_check_odr(sensor, sensor->bdr,
 						   &data);
 			if (err < 0)
 				return err;
@@ -713,7 +714,7 @@ st_lsm6dsx_update_samples_to_discard(struct st_lsm6dsx_sensor *sensor)
 
 	data = &hw->settings->samples_to_discard[sensor->id];
 	for (i = 0; i < ST_LSM6DSX_ODR_LIST_SIZE; i++) {
-		if (data->val[i].milli_hz == sensor->odr) {
+		if (data->val[i].milli_hz == sensor->bdr) {
 			sensor->samples_to_discard = data->val[i].samples;
 			return;
 		}
@@ -799,6 +800,52 @@ static const struct iio_buffer_setup_ops st_lsm6dsx_buffer_ops = {
 	.postdisable = st_lsm6dsx_buffer_postdisable,
 };
 
+static ssize_t st_lsm6dsx_bdr_show(struct device *dev,
+				   struct device_attribute *attr, char *buf)
+{
+	struct st_lsm6dsx_sensor *sensor = iio_priv(dev_to_iio_dev(dev));
+	u32 bdr = sensor->bdr;
+
+	return sysfs_emit(buf, "%d.%03d\n", bdr / 1000, bdr % 1000);
+}
+
+static ssize_t st_lsm6dsx_bdr_store(struct device *dev,
+				    struct device_attribute *attr,
+				    const char *buf, size_t len)
+{

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c static const struct iio_buffer_setup_ops st_lsm6dsx_buffer_ops = { · line 825

Quoted 10 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +10 -0
+	struct iio_dev *iio_dev = dev_to_iio_dev(dev);
+	struct st_lsm6dsx_sensor *sensor = iio_priv(iio_dev);
+	int integer, fract;
+	int ret;
+	u32 bdr;
+	u8 data;
+
+	ret = iio_str_to_fixpoint(buf, 100, &integer, &fract);
+	if (ret)
+		return ret;

nit: new line here.

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c static const struct iio_buffer_setup_ops st_lsm6dsx_buffer_ops = { · line 829

Quoted 4 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +4 -0
+	bdr = integer * 1000 + fract;
+	ret = st_lsm6dsx_check_odr(sensor, bdr, &data);
+	if (ret < 0)
+		return ret;

nit: new line here.

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c static const struct iio_buffer_setup_ops st_lsm6dsx_buffer_ops = { · line 832

Quoted 3 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +3 -0
+	bdr = ret;
+	if (!iio_device_claim_direct(iio_dev))
+		return -EBUSY;

I guess you can check it at the beginning of the routine.

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c static const struct iio_buffer_setup_ops st_lsm6dsx_buffer_ops = { · line 837

Quoted 5 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +5 -0
+	/* the batch data rate must not exceed the sensor output data rate */
+	if (bdr <= sensor->odr)
+		sensor->bdr = bdr;
+	else
+		ret = -EINVAL;

nit: new line here.

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c static const struct iio_buffer_setup_ops st_lsm6dsx_buffer_ops = { · line 838

Quoted 1 line
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +1 -0
+	iio_device_release_direct(iio_dev);

nit: new line here.

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c static const struct iio_buffer_setup_ops st_lsm6dsx_buffer_ops = { · line 839

Quoted 1 line
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +1 -0
+	return (ret < 0) ? ret : len;

nit: we do not need brackets here.

Quoted 45 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +12 -2
+}
+
+static IIO_DEV_ATTR_SAMP_FREQ(0664, st_lsm6dsx_bdr_show, st_lsm6dsx_bdr_store);
+
+static const struct iio_dev_attr *st_lsm6dsx_buffer_attrs[] = {
+	&iio_dev_attr_sampling_frequency,
+	NULL
+};
+
 int st_lsm6dsx_fifo_setup(struct st_lsm6dsx_hw *hw)
 {
 	int i, ret;
@@ -807,8 +854,9 @@ int st_lsm6dsx_fifo_setup(struct st_lsm6dsx_hw *hw)
 		if (!hw->iio_devs[i])
 			continue;
 
-		ret = devm_iio_kfifo_buffer_setup(hw->dev, hw->iio_devs[i],
-						  &st_lsm6dsx_buffer_ops);
+		ret = devm_iio_kfifo_buffer_setup_ext(hw->dev, hw->iio_devs[i],
+						      &st_lsm6dsx_buffer_ops,
+						      st_lsm6dsx_buffer_attrs);
 		if (ret)
 			return ret;
 	}
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_core.c +5 -2
index c65ad49829e7..e4922578329e 100644
@@ -1847,10 +1847,13 @@ static int st_lsm6dsx_write_raw(struct iio_dev *iio_dev,
 
 		val = val * 1000 + val2 / 1000;
 		val = st_lsm6dsx_check_odr(sensor, val, &data);
-		if (val < 0)
+		if (val < 0) {
 			err = val;
-		else
+		} else {
 			sensor->odr = val;
+			/* the batch data rate must not exceed the sensor ODR */
+			sensor->bdr = min_t(u32, sensor->bdr, sensor->odr);
+		}
 		break;
 	}
 	default:

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_core.c static struct iio_dev *st_lsm6dsx_alloc_iiodev(struct st_lsm6dsx_hw *hw, · line 2389

Quoted 5 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_core.c +1 -1
@@ -2383,7 +2386,7 @@ static struct iio_dev *st_lsm6dsx_alloc_iiodev(struct st_lsm6dsx_hw *hw,
 	sensor = iio_priv(iio_dev);
 	sensor->id = id;
 	sensor->hw = hw;
-	sensor->odr = hw->settings->odr_table[id].odr_avl[0].milli_hz;
+	sensor->odr = sensor->bdr = hw->settings->odr_table[id].odr_avl[0].milli_hz;

please add a new line to set sensor->bdr here.

Quoted 17 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_core.c +0 -0
 	sensor->gain = hw->settings->fs_table[id].fs_avl[0].gain;
 	sensor->watermark = 1;
 
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_shub.c +1 -1
index 3c5e65dc0f97..01d73002e888 100644
@@ -639,7 +639,7 @@ __st_lsm6dsx_shub_write_raw(struct iio_dev *iio_dev,
 			return odr;
 
 		sensor->ext_info.slv_odr = val;
-		sensor->odr = odr;
+		sensor->odr = sensor->bdr = odr;
 		return 0;
 	}
 	case IIO_CHAN_INFO_SCALE:

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_shub.c st_lsm6dsx_shub_alloc_iiodev(struct st_lsm6dsx_hw *hw, · line 748

Quoted 5 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_shub.c +1 -1
@@ -745,7 +745,7 @@ st_lsm6dsx_shub_alloc_iiodev(struct st_lsm6dsx_hw *hw,
 	sensor = iio_priv(iio_dev);
 	sensor->id = id;
 	sensor->hw = hw;
-	sensor->odr = hw->settings->odr_table[ref_id].odr_avl[0].milli_hz;
+	sensor->odr = sensor->bdr = hw->settings->odr_table[ref_id].odr_avl[0].milli_hz;

please add a new line to set sensor->bdr here.

Quoted 6 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_shub.c +0 -0
 	sensor->ext_info.slv_odr = info->odr_table.odr_avl[0].milli_hz;
 	sensor->gain = info->fs_table.fs_avl[0].gain;
 	sensor->ext_info.settings = info;
Signature · 1 line
2.39.5
#5 lore raw
Quoted 9 lines · Lorenzo Bianconi
Quoted 8 lines

The rate at which accelerometer or gyroscope sensor samples are fed to the hardware FIFO (batch data rate, or BDR) does not have to coincide with the sensor sampling frequency (output data rate, or ODR); the only requirement is for the BDR to not be greater than the ODR. Having a BDR lower than the ODR is useful in cases where an application requires a high sampling rate for accurate detection of motion events (e.g. wakeup events), but wants to read sensor sample values from the device buffer at a lower data rate.

Quoted 5 lines

can you please provide more details here? Are you using the hw fifo to read
data? If we configure the hw fifo according to the BDR (even assuming the watermark is set 1) the hw will generate interrupts according to the BDR (bdr < odr).

Yes, I'm using the hw fifo to read data. The use case is to enable event detection (which works best at high sampling rates) and sensor data streaming at the same time, without requiring the data stream to be at the same rate as the sensor sampling rate. So the amount of I2C (or SPI) traffic (as well as the rate of periodic interrupts) required by the data stream is kept to a minimum without sacrificing the accuracy of event detection.

maintainer
#6 lore raw
Quoted 16 lines · Lorenzo Bianconi
Quoted 9 lines
Quoted 8 lines

The rate at which accelerometer or gyroscope sensor samples are fed to the hardware FIFO (batch data rate, or BDR) does not have to coincide with the sensor sampling frequency (output data rate, or ODR); the only requirement is for the BDR to not be greater than the ODR. Having a BDR lower than the ODR is useful in cases where an application requires a high sampling rate for accurate detection of motion events (e.g. wakeup events), but wants to read sensor sample values from the device buffer at a lower data rate.

Quoted 5 lines

can you please provide more details here? Are you using the hw fifo to read
data? If we configure the hw fifo according to the BDR (even assuming the watermark is set 1) the hw will generate interrupts according to the BDR (bdr < odr).

Quoted 7 lines

Yes, I'm using the hw fifo to read data. The use case is to enable event detection (which works best at high sampling rates) and sensor data streaming at the same time, without requiring the data stream to be at the same rate as the sensor sampling rate. So the amount of I2C (or SPI) traffic (as well as the rate of periodic interrupts) required by the data stream is kept to a minimum without sacrificing the accuracy of event detection.

I guess you can get the same result (reduce sensor data interrupt rate keeping high odr value) configuring the hw fifo watermark. Does it work for you?

Regards,
Lorenzo

Quoted 2 lines
#7 lore raw
Quoted 30 lines · Lorenzo Bianconi
Quoted 19 lines · Lorenzo Bianconi
Quoted 9 lines
Quoted 8 lines

The rate at which accelerometer or gyroscope sensor samples are fed to the hardware FIFO (batch data rate, or BDR) does not have to coincide with the sensor sampling frequency (output data rate, or ODR); the only requirement is for the BDR to not be greater than the ODR. Having a BDR lower than the ODR is useful in cases where an application requires a high sampling rate for accurate detection of motion events (e.g. wakeup events), but wants to read sensor sample values from the device buffer at a lower data rate.

Quoted 8 lines

can you please provide more details here? Are you using the hw fifo to
read
data? If we configure the hw fifo according to the BDR (even assuming the
watermark is set 1) the hw will generate interrupts according to the BDR
(bdr < odr).

Quoted 10 lines

Yes, I'm using the hw fifo to read data. The use case is to enable event
detection (which works best at high sampling rates) and sensor data streaming at the same time, without requiring the data stream to be at the
same rate as the sensor sampling rate. So the amount of I2C (or SPI) traffic (as well as the rate of periodic interrupts) required by the data
stream is kept to a minimum without sacrificing the accuracy of event detection.

Quoted 3 lines

I guess you can get the same result (reduce sensor data interrupt rate keeping high odr value) configuring the hw fifo watermark. Does it work for you?

Setting the hw fifo watermark to a high value reduces the rate of interrupts, but doesn't do much to reduce the amount of I2C traffic, so the issue would still be there.

maintainer
#8 lore raw
Quoted 31 lines · Francesco Lavra
Quoted 30 lines · Lorenzo Bianconi
Quoted 29 lines · Lorenzo Bianconi
Quoted 9 lines
Quoted 8 lines

The rate at which accelerometer or gyroscope sensor samples are fed to the hardware FIFO (batch data rate, or BDR) does not have to coincide with the sensor sampling frequency (output data rate, or ODR); the only requirement is for the BDR to not be greater than the ODR. Having a BDR lower than the ODR is useful in cases where an application requires a high sampling rate for accurate detection of motion events (e.g. wakeup events), but wants to read sensor sample values from the device buffer at a lower data rate.

Quoted 8 lines

can you please provide more details here? Are you using the hw fifo to
read
data? If we configure the hw fifo according to the BDR (even assuming the
watermark is set 1) the hw will generate interrupts according to the BDR
(bdr < odr).

Yes, I'm using the hw fifo to read data. The use case is to enable event
detection (which works best at high sampling rates) and sensor data streaming at the same time, without requiring the data stream to be at the
same rate as the sensor sampling rate. So the amount of I2C (or SPI) traffic (as well as the rate of periodic interrupts) required by the data
stream is kept to a minimum without sacrificing the accuracy of event detection.

Quoted 7 lines
Quoted 3 lines

I guess you can get the same result (reduce sensor data interrupt rate keeping high odr value) configuring the hw fifo watermark. Does it work for you?

Setting the hw fifo watermark to a high value reduces the rate of interrupts, but doesn't do much to reduce the amount of I2C traffic, so the issue would still be there.

ack, now I got the goal of the series. I think the series is mostly fine. I guess hwfifo_odr instead of bdr is more meaningful, what do you think? Naming is always hard.

Regards,
Lorenzo

reviewer
#9 lore raw
Quoted 40 lines · Lorenzo Bianconi
Quoted 31 lines · Francesco Lavra
Quoted 30 lines · Lorenzo Bianconi
Quoted 29 lines · Lorenzo Bianconi
Quoted 9 lines
Quoted 8 lines

The rate at which accelerometer or gyroscope sensor samples are fed to the hardware FIFO (batch data rate, or BDR) does not have to coincide with the sensor sampling frequency (output data rate, or ODR); the only requirement is for the BDR to not be greater than the ODR. Having a BDR lower than the ODR is useful in cases where an application requires a high sampling rate for accurate detection of motion events (e.g. wakeup events), but wants to read sensor sample values from the device buffer at a lower data rate.

Quoted 8 lines

can you please provide more details here? Are you using the hw fifo to
read
data? If we configure the hw fifo according to the BDR (even assuming the
watermark is set 1) the hw will generate interrupts according to the BDR
(bdr < odr).

Yes, I'm using the hw fifo to read data. The use case is to enable event
detection (which works best at high sampling rates) and sensor data streaming at the same time, without requiring the data stream to be at the
same rate as the sensor sampling rate. So the amount of I2C (or SPI) traffic (as well as the rate of periodic interrupts) required by the data
stream is kept to a minimum without sacrificing the accuracy of event detection.

Quoted 7 lines
Quoted 3 lines

I guess you can get the same result (reduce sensor data interrupt rate keeping high odr value) configuring the hw fifo watermark. Does it work for you?

Setting the hw fifo watermark to a high value reduces the rate of interrupts, but doesn't do much to reduce the amount of I2C traffic, so the issue would still be there.

Quoted 6 lines

ack, now I got the goal of the series. I think the series is mostly fine. I guess hwfifo_odr instead of bdr is more meaningful, what do you think? Naming is always hard.

Regards,
Lorenzo

In the IIO subsystem, we prefer to include the units in the variable/ field name as well, e.g. hw_fifo_odr_mHz.

maintainer
#10 lore raw
Quoted 45 lines · David Lechner
Quoted 44 lines · Lorenzo Bianconi
Quoted 31 lines · Francesco Lavra
Quoted 30 lines · Lorenzo Bianconi
Quoted 29 lines · Lorenzo Bianconi
Quoted 9 lines
Quoted 8 lines

The rate at which accelerometer or gyroscope sensor samples are fed to the hardware FIFO (batch data rate, or BDR) does not have to coincide with the sensor sampling frequency (output data rate, or ODR); the only requirement is for the BDR to not be greater than the ODR. Having a BDR lower than the ODR is useful in cases where an application requires a high sampling rate for accurate detection of motion events (e.g. wakeup events), but wants to read sensor sample values from the device buffer at a lower data rate.

Quoted 8 lines

can you please provide more details here? Are you using the hw fifo to
read
data? If we configure the hw fifo according to the BDR (even assuming the
watermark is set 1) the hw will generate interrupts according to the BDR
(bdr < odr).

Yes, I'm using the hw fifo to read data. The use case is to enable event
detection (which works best at high sampling rates) and sensor data streaming at the same time, without requiring the data stream to be at the
same rate as the sensor sampling rate. So the amount of I2C (or SPI) traffic (as well as the rate of periodic interrupts) required by the data
stream is kept to a minimum without sacrificing the accuracy of event detection.

Quoted 7 lines
Quoted 3 lines

I guess you can get the same result (reduce sensor data interrupt rate keeping high odr value) configuring the hw fifo watermark. Does it work for you?

Setting the hw fifo watermark to a high value reduces the rate of interrupts, but doesn't do much to reduce the amount of I2C traffic, so the issue would still be there.

ack, now I got the goal of the series. I think the series is mostly fine. I guess hwfifo_odr instead of bdr is more meaningful, what do you think? Naming is always hard.

Quoted 5 lines
Quoted 2 lines

Regards,
Lorenzo

In the IIO subsystem, we prefer to include the units in the variable/ field name as well, e.g. hw_fifo_odr_mHz.

ack, but please avoid camel case :)

#11 lore raw
Quoted 2 lines · Lorenzo Bianconi
Quoted 1 line · David Lechner

On 10/10/25 8:15 AM, Lorenzo Bianconi wrote:

...

Quoted 4 lines
Quoted 2 lines

In the IIO subsystem, we prefer to include the units in the variable/ field name as well, e.g. hw_fifo_odr_mHz.

ack, but please avoid camel case :)

Don't mix up a camel case with units.
Or do you have a way to distinguish milli from mega, please?

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#12 lore raw
Quoted 7 lines · Andy Shevchenko
Quoted 2 lines · Lorenzo Bianconi
Quoted 1 line · David Lechner

On 10/10/25 8:15 AM, Lorenzo Bianconi wrote:

...

Quoted 7 lines
Quoted 4 lines
Quoted 2 lines

In the IIO subsystem, we prefer to include the units in the variable/ field name as well, e.g. hw_fifo_odr_mHz.

ack, but please avoid camel case :)

Don't mix up a camel case with units.
Or do you have a way to distinguish milli from mega, please?

OK, I will use hwfifo_odr_mHz

#13 lore raw
In reply to Francesco Lavra
Quoted 19 lines · Francesco Lavra

The rate at which accelerometer or gyroscope sensor samples are fed to the hardware FIFO (batch data rate, or BDR) does not have to coincide with the sensor sampling frequency (output data rate, or ODR); the only requirement is for the BDR to not be greater than the ODR. Having a BDR lower than the ODR is useful in cases where an application requires a high sampling rate for accurate detection of motion events (e.g. wakeup events), but wants to read sensor sample values from the device buffer at a lower data rate. To support the above use case, add a sampling_frequency sysfs attribute to the buffer directory of st_lsm6dsx IIO devices, which controls the BDR for a given sensor independently from the "main" sampling_frequency attribute (which controls the ODR); introduce a new bdr field in struct st_lsm6dsx_sensor to keep track of the current BDR value, and use this field instead of the odr field in the code that deals with the FIFO data rate. In the sensor hub driver, make the bdr value always mirror the odr value, since there is no separate configuration setting to control the BDR for data produced by the sensor hub functionality.

...

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c static void st_lsm6dsx_get_max_min_odr(struct st_lsm6dsx_hw *hw, · line 141

Quoted 4 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +2 -2
-               *max_odr = max_t(u32, *max_odr, sensor->odr);
-               *min_odr = min_t(u32, *min_odr, sensor->odr);
+               *max_odr = max_t(u32, *max_odr, sensor->bdr);
+               *min_odr = min_t(u32, *min_odr, sensor->bdr);

Can we get rid of '_t' parts at some point? Or IOW what is the good justification for typed macros here?

...

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c static const struct iio_buffer_setup_ops st_lsm6dsx_buffer_ops = { · line 826

Quoted 4 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +4 -0
+       ret = iio_str_to_fixpoint(buf, 100, &integer, &fract);
+       if (ret)
+               return ret;
+       bdr = integer * 1000 + fract;

MILLI?

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#14 lore raw
Quoted 24 lines · Andy Shevchenko
Quoted 19 lines · Francesco Lavra

The rate at which accelerometer or gyroscope sensor samples are fed to the hardware FIFO (batch data rate, or BDR) does not have to coincide with the sensor sampling frequency (output data rate, or ODR); the only requirement is for the BDR to not be greater than the ODR. Having a BDR lower than the ODR is useful in cases where an application requires a high sampling rate for accurate detection of motion events (e.g. wakeup events), but wants to read sensor sample values from the device buffer at a lower data rate. To support the above use case, add a sampling_frequency sysfs attribute to the buffer directory of st_lsm6dsx IIO devices, which controls the BDR for a given sensor independently from the "main" sampling_frequency attribute (which controls the ODR); introduce a new bdr field in struct st_lsm6dsx_sensor to keep track of the current BDR value, and use this field instead of the odr field in the code that deals with the FIFO data rate. In the sensor hub driver, make the bdr value always mirror the odr value, since there is no separate configuration setting to control the BDR for data produced by the sensor hub functionality.

...

Quoted 7 lines
Quoted 4 lines

-               *max_odr = max_t(u32, *max_odr, sensor->odr); -               *min_odr = min_t(u32, *min_odr, sensor->odr); +               *max_odr = max_t(u32, *max_odr, sensor->bdr); +               *min_odr = min_t(u32, *min_odr, sensor->bdr);

Can we get rid of '_t' parts at some point? Or IOW what is the good justification for typed macros here?

I think they are not justified here, I will get take this opportunity to get rid of them.

Quoted 8 lines

...

Quoted 4 lines

+       ret = iio_str_to_fixpoint(buf, 100, &integer, &fract); +       if (ret)
+               return ret;
+       bdr = integer * 1000 + fract;

MILLI?

If you mean replacing fract with milli, will do.

#15 lore raw
Quoted 3 lines · Francesco Lavra
Quoted 2 lines · Andy Shevchenko

On Thu, Oct 9, 2025 at 8:36 PM Francesco Lavra <[email protected]> wrote:

...

Quoted 8 lines
Quoted 6 lines
Quoted 4 lines

+       ret = iio_str_to_fixpoint(buf, 100, &integer, &fract); +       if (ret)
+               return ret;
+       bdr = integer * 1000 + fract;

MILLI?

If you mean replacing fract with milli, will do.

I meant 1000 --> MILLI(?) or whatever SI prefix suits better?

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#16 lore raw
In reply to Francesco Lavra
Quoted 19 lines · On Thu, 9 Oct 2025 19:36:09 +0200 Francesco Lavra <[email protected]> wrote:

The rate at which accelerometer or gyroscope sensor samples are fed to the hardware FIFO (batch data rate, or BDR) does not have to coincide with the sensor sampling frequency (output data rate, or ODR); the only requirement is for the BDR to not be greater than the ODR. Having a BDR lower than the ODR is useful in cases where an application requires a high sampling rate for accurate detection of motion events (e.g. wakeup events), but wants to read sensor sample values from the device buffer at a lower data rate. To support the above use case, add a sampling_frequency sysfs attribute to the buffer directory of st_lsm6dsx IIO devices, which controls the BDR for a given sensor independently from the "main" sampling_frequency attribute (which controls the ODR); introduce a new bdr field in struct st_lsm6dsx_sensor to keep track of the current BDR value, and use this field instead of the odr field in the code that deals with the FIFO data rate. In the sensor hub driver, make the bdr value always mirror the odr value, since there is no separate configuration setting to control the BDR for data produced by the sensor hub functionality.

Quoted 1 line

A few additional trivial things from me. In general this looks fine. Whilst that buffer/sampling_frequency isn't common it's been part of the ABI for a while for this sort of thing.

My only slight concern is backwards compatibility.
Perhaps you can add something on what happens if main sampling_frequency is modified by a user who doesn't know anything about buffer/sampling_frequency?

Given that's a new interface and the ABI always allows a write to one value to change any other maybe we have to say the main sampling frequency write updates the buffer one and a write to the buffer one after that is needed to set it to a different value?

That is a bit ugly but it is backwards compatible I think.

Quoted 13 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +1 -0
index 8a9d2593576a..5912ea76d493 100644
@@ -56,6 +56,7 @@
 #include <linux/iio/kfifo_buf.h>
 #include <linux/iio/iio.h>
 #include <linux/iio/buffer.h>
+#include <linux/iio/sysfs.h>
 #include <linux/regmap.h>
 #include <linux/bitfield.h>
 

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c static int · line 109

Quoted 5 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +1 -1
@@ -105,7 +106,7 @@ static int
 st_lsm6dsx_get_decimator_val(struct st_lsm6dsx_sensor *sensor, u32 max_odr)
 {
 	const int max_size = ARRAY_SIZE(st_lsm6dsx_decimator_table);
-	u32 decimator =  max_odr / sensor->odr;
+	u32 decimator =  max_odr / sensor->bdr;

No idea why there is a bonus space after = but good to cleanup whilst you are here.

Quoted 6 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +4 -0
 	int i;
 
+static ssize_t st_lsm6dsx_bdr_store(struct device *dev,
+				    struct device_attribute *attr,
+				    const char *buf, size_t len)
+{

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c static const struct iio_buffer_setup_ops st_lsm6dsx_buffer_ops = { · line 825

Quoted 10 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +10 -0
+	struct iio_dev *iio_dev = dev_to_iio_dev(dev);
+	struct st_lsm6dsx_sensor *sensor = iio_priv(iio_dev);
+	int integer, fract;
+	int ret;
+	u32 bdr;
+	u8 data;
+
+	ret = iio_str_to_fixpoint(buf, 100, &integer, &fract);
+	if (ret)
+		return ret;

Add blank line after this sort of error handling return. Slightly helps with readability.

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c static const struct iio_buffer_setup_ops st_lsm6dsx_buffer_ops = { · line 829

Quoted 4 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +4 -0
+	bdr = integer * 1000 + fract;
+	ret = st_lsm6dsx_check_odr(sensor, bdr, &data);
+	if (ret < 0)
+		return ret;

Here as well.

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c static const struct iio_buffer_setup_ops st_lsm6dsx_buffer_ops = { · line 830

Quoted 1 line
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +1 -0
+	bdr = ret;

Probably here as well.

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c static const struct iio_buffer_setup_ops st_lsm6dsx_buffer_ops = { · line 838

Quoted 8 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +8 -0
+	if (!iio_device_claim_direct(iio_dev))
+		return -EBUSY;
+	/* the batch data rate must not exceed the sensor output data rate */
+	if (bdr <= sensor->odr)
+		sensor->bdr = bdr;
+	else
+		ret = -EINVAL;
+	iio_device_release_direct(iio_dev);

Add one before the final return.

Quoted 2 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +2 -0
+	return (ret < 0) ? ret : len;
+}
#17 lore raw
Quoted 33 lines · Jonathan Cameron
Quoted 19 lines · On Thu,  9 Oct 2025 19:36:09 +0200 Francesco Lavra <[email protected]> wrote:

The rate at which accelerometer or gyroscope sensor samples are fed to the hardware FIFO (batch data rate, or BDR) does not have to coincide with the sensor sampling frequency (output data rate, or ODR); the only requirement is for the BDR to not be greater than the ODR. Having a BDR lower than the ODR is useful in cases where an application requires a high sampling rate for accurate detection of motion events (e.g. wakeup events), but wants to read sensor sample values from the device buffer at a lower data rate. To support the above use case, add a sampling_frequency sysfs attribute to the buffer directory of st_lsm6dsx IIO devices, which controls the BDR for a given sensor independently from the "main" sampling_frequency attribute (which controls the ODR); introduce a new bdr field in struct st_lsm6dsx_sensor to keep track of the current BDR value, and use this field instead of the odr field in the code that deals with the FIFO data rate. In the sensor hub driver, make the bdr value always mirror the odr value, since there is no separate configuration setting to control the BDR for data produced by the sensor hub functionality.

Quoted 1 line

A few additional trivial things from me.  In general this looks fine. Whilst that buffer/sampling_frequency isn't common it's been part of the ABI for a while for this sort of thing.

My only slight concern is backwards compatibility.
Perhaps you can add something on what happens if main sampling_frequency is modified by a user who doesn't know anything about buffer/sampling_frequency?

Quoted 8 lines

Given that's a new interface and the ABI always allows a write to one value to change any other maybe we have to say the main sampling frequency
write updates the buffer one and a write to the buffer one after that is needed
to set it to a different value?

That is a bit ugly but it is backwards compatible I think.

Yes, for backwards compatibility it makes sense to update the buffer frequency whenever the main frequency is set. Will do.

OK also for the cosmetic changes below.

Quoted 66 lines
Quoted 14 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +1 -0
index 8a9d2593576a..5912ea76d493 100644
@@ -56,6 +56,7 @@
  #include <linux/iio/kfifo_buf.h>
  #include <linux/iio/iio.h>
  #include <linux/iio/buffer.h>
+#include <linux/iio/sysfs.h>
  #include <linux/regmap.h>
  #include <linux/bitfield.h>

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +0 -0
@@ -105,7 +106,7 @@ static int
Quoted 6 lines

 st_lsm6dsx_get_decimator_val(struct st_lsm6dsx_sensor *sensor, u32 max_odr)
 {
        const int max_size = ARRAY_SIZE(st_lsm6dsx_decimator_table); -       u32 decimator =  max_odr / sensor->odr;
+       u32 decimator =  max_odr / sensor->bdr;

No idea why there is a bonus space after = but good to cleanup whilst you are
here.

Quoted 6 lines

        int i;

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +4 -0
+static ssize_t st_lsm6dsx_bdr_store(struct device *dev,
+                                   struct device_attribute *attr,
+                                   const char *buf, size_t len)
+{

drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c static const struct iio_buffer_setup_ops st_lsm6dsx_buffer_ops = { · line 825

Quoted 10 lines
drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c +10 -0
+       struct iio_dev *iio_dev = dev_to_iio_dev(dev);
+       struct st_lsm6dsx_sensor *sensor = iio_priv(iio_dev);
+       int integer, fract;
+       int ret;
+       u32 bdr;
+       u8 data;
+
+       ret = iio_str_to_fixpoint(buf, 100, &integer, &fract);
+       if (ret)
+               return ret;

Add blank line after this sort of error handling return.  Slightly helps with readability.

Quoted 4 lines

+       bdr = integer * 1000 + fract;
+       ret = st_lsm6dsx_check_odr(sensor, bdr, &data); +       if (ret < 0)
+               return ret;

Here as well.

Quoted 1 line

+       bdr = ret;

Probably here as well.

Quoted 9 lines

+       if (!iio_device_claim_direct(iio_dev))
+               return -EBUSY;
+       /* the batch data rate must not exceed the sensor output data rate */
+       if (bdr <= sensor->odr)
+               sensor->bdr = bdr;
+       else
+               ret = -EINVAL;
+       iio_device_release_direct(iio_dev);

Add one before the final return.

Quoted 2 lines

+       return (ret < 0) ? ret : len;
+}