On 01/07, David Lechner wrote:

On 1/7/25 9:26 AM, Jonathan Santos wrote:

quoted

When the device is configured to Sinc5 filter and decimation x8,
output data is reduced to 16-bits in order to support 1 MHz of
sampling frequency due to clock limitation.

We aren't going to get a 1 MHz sample rate without SPI offload support so maybe
we should save this patch until then?

In this patch, we are still reading 24-bits per sample, so we aren't really
getting any benefit. It is probably fine for now to leave it as 24-bit even if
the last 8 bits are all 0 or just noise.

Indeed we cannot achieve 1 MHz yet, but I believe it is good have this
now so it is more mature for the time SPI offload is supported. Also, will
allow us to backport this patch to other repos.

Also, the datasheet says:

	this path allows viewing of wider bandwidth; however, it is quantization
	noise limited so that output data is reduced to 16 bits

So this doesn't actually seem related to higher sample rates. There is a CONVLEN
bit in the INTERFACE_FORMAT register that globally reduces the output size to
16-bit, which I suspect would be what we will need for achieving the highest
sample rate when we add SPI offload support.

Right, that is true, but the reason we did this patch was to fix the
output size when we configure the filter to sinc5 decx8. The datasheet
says:

	To configure the sinc5 filter for 1.024 MSPS output data rate,
	write 001 to the FILTER bits [6:4] of the DIGITAL_FILTER register
	(Register 0x19). The ADAQ7768-1 automatically changes the decimation
	rate to 8 and output data length is reduced to 16 bits from 24 bits 
	due to the maximum speed limitation of the digital serial interface.

In this case we don't even need to change the value of CONVLEN

quoted

Use multiple scan types feature to enable the driver to switch
scan type in runtime, making possible to support both 24-bit and
16-bit resolution.

Signed-off-by: Jonathan Santos <redacted>
---
 drivers/iio/adc/ad7768-1.c | 65 ++++++++++++++++++++++++++++++++------
 1 file changed, 56 insertions(+), 9 deletions(-)

diff --git a/drivers/iio/adc/ad7768-1.c b/drivers/iio/adc/ad7768-1.c
index 9741a6d47942..5e4e7d387f9a 100644
--- a/drivers/iio/adc/ad7768-1.c
+++ b/drivers/iio/adc/ad7768-1.c

@@ -134,6 +134,11 @@ struct ad7768_clk_configuration {
 	enum ad7768_pwrmode pwrmode;
 };
 
+enum ad7768_scan_type {
+	AD7768_SCAN_TYPE_NORMAL,
+	AD7768_SCAN_TYPE_HIGH_SPEED,
+};
+
 static const char * const ad7768_vcm_modes[] = {
 	"(AVDD1-AVSS)/2",
 	"2V5",

@@ -145,6 +150,10 @@ static const char * const ad7768_vcm_modes[] = {
 	"OFF",
 };
 
+static const int ad7768_mclk_div_rates[4] = {
+	16, 8, 4, 2,
+};
+
 static const struct ad7768_clk_configuration ad7768_clk_config[] = {
 	{ AD7768_MCLK_DIV_2, AD7768_DEC_RATE_8, 16,  AD7768_FAST_MODE },
 	{ AD7768_MCLK_DIV_2, AD7768_DEC_RATE_16, 32,  AD7768_FAST_MODE },

@@ -159,6 +168,21 @@ static const struct ad7768_clk_configuration ad7768_clk_config[] = {
 	{ AD7768_MCLK_DIV_16, AD7768_DEC_RATE_1024, 16384, AD7768_ECO_MODE },
 };
 
+static const struct iio_scan_type ad7768_scan_type[] = {
+	[AD7768_SCAN_TYPE_NORMAL] = {
+		.sign = 's',
+		.realbits = 24,
+		.storagebits = 32,

What happened to .shift = 8, ? If there is a reason for removing it, please add
that to the commit description.

Sorry, will fix this

quoted

+		.endianness = IIO_BE,
+	},
+	[AD7768_SCAN_TYPE_HIGH_SPEED] = {
+		.sign = 's',
+		.realbits = 16,
+		.storagebits = 32,

I guess it doesn't matter much since we are reading one sample at a time, but
I would expect storagebits to be 16 instead of 32. Or if it really needs to be
32, does it need shift = 16?

This is because the hw is configured to return the samples in a 32 bits
format, so if storage is 16 we will get wrong data.

quoted

+		.endianness = IIO_BE,
+	},
+};
+
 static int ad7768_get_vcm(struct iio_dev *dev, const struct iio_chan_spec *chan);
 static int ad7768_set_vcm(struct iio_dev *dev, const struct iio_chan_spec *chan,
 			  unsigned int mode);

...

quoted

@@ -308,6 +329,15 @@ static int ad7768_scan_direct(struct iio_dev *indio_dev)
 	ret = ad7768_spi_reg_read(st, AD7768_REG_ADC_DATA, &readval, 3);
 	if (ret < 0)
 		return ret;
+
+	/*
+	 * When the decimation rate is set to x8, the ADC data precision is reduced
+	 * from 24 bits to 16 bits. Since the AD7768_REG_ADC_DATA register provides
+	 * 24-bit data, the precision is reduced by right-shifting the read value
+	 * by 8 bits.
+	 */
+	if (st->dec_rate == 8)
+		readval = readval >> 8;

Why not change size of ad7768_spi_reg_read() instead of reading 3 bytes and
throwing one away?

Right, i will check this and fix in the next version

quoted

 	/*
 	 * Any SPI configuration of the AD7768-1 can only be
 	 * performed in continuous conversion mode.