Re: [PATCH v1] iio: adc: add buffering support to the TI LMP92064 ADC driver

[Leonard Göhrs]

On 17.06.23 20:27, Jonathan Cameron wrote:
> On Wed, 14 Jun 2023 09:55:36 +0200
> Leonard Göhrs <l.goehrs@xxxxxxxxxxxxxx> wrote:
>
> > Enable buffered reading of samples from the LMP92064 ADC.
> > The main benefit of this change is being able to read out current and
> > voltage measurements in a single transfer, allowing instantaneous power
> > measurements.
> >
> > Reads into the buffer can be triggered by any software triggers, e.g.
> > the iio-trig-hrtimer:
> >
> >      $ mkdir /sys/kernel/config/iio/triggers/hrtimer/my-trigger
> >      $ cat /sys/bus/iio/devices/iio\:device3/name
> >      lmp92064
> >      $ iio_readdev -t my-trigger -b 16 iio:device3 | hexdump
> >      WARNING: High-speed mode not enabled
> >      0000000 0000 0176 0101 0001 5507 abd5 7645 1768
> >      0000010 0000 016d 0101 0001 ee1e ac6b 7645 1768
> >      ...
> >
> > Signed-off-by: Leonard Göhrs <l.goehrs@xxxxxxxxxxxxxx>
> > ---
> >   drivers/iio/adc/ti-lmp92064.c | 54 +++++++++++++++++++++++++++++++++++
> >   1 file changed, 54 insertions(+)
> >
> > diff --git a/drivers/iio/adc/ti-lmp92064.c b/drivers/iio/adc/ti-lmp92064.c
> > index c30ed824924f3..03765c4057dda 100644
> > --- a/drivers/iio/adc/ti-lmp92064.c
> > +++ b/drivers/iio/adc/ti-lmp92064.c
> > @@ -16,7 +16,10 @@
> >   #include <linux/spi/spi.h>
> >   
> >   #include <linux/iio/iio.h>
> > +#include <linux/iio/buffer.h>
> >   #include <linux/iio/driver.h>
> > +#include <linux/iio/triggered_buffer.h>
> > +#include <linux/iio/trigger_consumer.h>
> >   
> >   #define TI_LMP92064_REG_CONFIG_A 0x0000
> >   #define TI_LMP92064_REG_CONFIG_B 0x0001
> > @@ -91,6 +94,13 @@ static const struct iio_chan_spec lmp92064_adc_channels[] = {
> >   		.address = TI_LMP92064_CHAN_INC,
> >   		.info_mask_separate =
> >   			BIT(IIO_CHAN_INFO_RAW) | BIT(IIO_CHAN_INFO_SCALE),
> > +		.scan_index = 0,
> > +		.scan_type = {
> > +			.sign = 'u',
> > +			.realbits = 12,
> > +			.storagebits = 16,
> > +			.shift = 0,
>
> As zero is the 'obvious' default for shift (do nothing case) and c will set it to 0 for
> you anyway, no need to set it explicitly like this.
>
> > +		},
> >   		.datasheet_name = "INC",
> >   	},
> >   	{
> > @@ -98,8 +108,16 @@ static const struct iio_chan_spec lmp92064_adc_channels[] = {
> >   		.address = TI_LMP92064_CHAN_INV,
> >   		.info_mask_separate =
> >   			BIT(IIO_CHAN_INFO_RAW) | BIT(IIO_CHAN_INFO_SCALE),
> > +		.scan_index = 1,
> > +		.scan_type = {
> > +			.sign = 'u',
> > +			.realbits = 12,
> > +			.storagebits = 16,
> > +			.shift = 0,
>
> Same here. No need for this last line.
>
> > +		},
> >   		.datasheet_name = "INV",
> >   	},
> > +	IIO_CHAN_SOFT_TIMESTAMP(2),
> >   };
> >   
> >   static int lmp92064_read_meas(struct lmp92064_adc_priv *priv, u16 *res)
> > @@ -171,6 +189,37 @@ static int lmp92064_read_raw(struct iio_dev *indio_dev,
> >   	}
> >   }
> >   
> > +static irqreturn_t lmp92064_trigger_handler(int irq, void *p)
> > +{
> > +	struct iio_poll_func *pf = p;
> > +	struct iio_dev *indio_dev = pf->indio_dev;
> > +	struct lmp92064_adc_priv *priv = iio_priv(indio_dev);
> > +	int i = 0, j, ret;
> > +	u16 raw[2];
> > +	u16 *data;
> > +
> > +	ret = lmp92064_read_meas(priv, raw);
> > +	if (ret < 0)
> > +		goto done;
> > +
> > +	data = kmalloc(indio_dev->scan_bytes, GFP_KERNEL);
>
> A driver shouldn't be directly accessing scan_bytes.
> You will see iio.h that it is marked [INTERN] to show it should only be accessed
> by the IIO core code.
>
> The reasoning is closely related to Lars' point. This is almost always a small
> structure with a known maximum size (across all possible channel configurations)
> so allocate that space once not every scan (will be 16 bytes I think)

Sounds reasonable. I've more or less copied the trigger handler from
iio_simple_dummy_buffer.c assuming it contained the best practice when
it comes to buffer handling w.r.t. size and alignment (which I was not quite
sure about how to do right).

I've sent a V2 now that uses a stack allocated structure as buffer. This makes
the whole trigger handler a lot more concise.

> > +	if (!data)
> > +		goto done;
> > +
> > +	for_each_set_bit(j, indio_dev->active_scan_mask, indio_dev->masklength)
> > +		data[i++] = raw[j];
> When a sensor 'always' reads the two channels, we can leave the necessary data
> mangling up to the IIO core.
>
> Provide available_scan_masks and then always push the lot every time.
> Mostly people will want both channels anyway (as they paid for them ;)
> so that core code that moves the data around will do nothing.  In the cases
> where they only want one channel it will handle the complexity for you.

Thanks for the tip! This simplifies the trigger handler even further.

> > +
> > +	iio_push_to_buffers_with_timestamp(indio_dev, data,
> > +					   iio_get_time_ns(indio_dev));
> > +
> > +	kfree(data);
> > +
> > +done:
> > +	iio_trigger_notify_done(indio_dev->trig);
> > +
> > +	return IRQ_HANDLED;
> > +}
> > +
> >   static int lmp92064_reset(struct lmp92064_adc_priv *priv,
> >   			  struct gpio_desc *gpio_reset)
> >   {
> > @@ -302,6 +351,11 @@ static int lmp92064_adc_probe(struct spi_device *spi)
> >   	indio_dev->num_channels = ARRAY_SIZE(lmp92064_adc_channels);
> >   	indio_dev->info = &lmp92064_adc_info;
> >   
> > +	ret = devm_iio_triggered_buffer_setup(dev, indio_dev, NULL,
> > +					      lmp92064_trigger_handler, NULL);
> > +	if (ret)
> > +		return dev_err_probe(dev, ret, "Failed to setup buffered read\n");
> > +
> >   	return devm_iio_device_register(dev, indio_dev);
> >   }
> >   
> >
> > base-commit: 9561de3a55bed6bdd44a12820ba81ec416e705a7