accel: adxl380: Improve data reading from FIFO

[PATCH v2 2/2] iio: accel: adxl380: Optimize reading of FIFO entries in interrupt handler

Posted by Francesco Lavra 2 weeks, 6 days ago

In order to minimize the time required for transferring FIFO data from the
sensor to the host machine, perform the read from the FIFO in a single call
to regmap_bulk_read().
This allows reading acceleration data for all 3 axes at 16 kHz
sampling frequency using a 1MHz I2C bus frequency.

Signed-off-by: Francesco Lavra <flavra@baylibre.com>
---
 drivers/iio/accel/adxl380.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)

diff --git a/drivers/iio/accel/adxl380.c b/drivers/iio/accel/adxl380.c
index 650cdbffd4a7..e3e6b182eb3d 100644
--- a/drivers/iio/accel/adxl380.c
+++ b/drivers/iio/accel/adxl380.c
@@ -965,14 +965,12 @@ static irqreturn_t adxl380_irq_handler(int irq, void  *p)
 		return IRQ_HANDLED;
 
 	fifo_entries = rounddown(fifo_entries, st->fifo_set_size);
-	for (i = 0; i < fifo_entries; i += st->fifo_set_size) {
-		ret = regmap_noinc_read(st->regmap, ADXL380_FIFO_DATA,
-					&st->fifo_buf[i],
-					2 * st->fifo_set_size);
-		if (ret)
-			return IRQ_HANDLED;
+	ret = regmap_noinc_read(st->regmap, ADXL380_FIFO_DATA, &st->fifo_buf,
+				sizeof(*st->fifo_buf) * fifo_entries);
+	if (ret)
+		return IRQ_HANDLED;
+	for (i = 0; i < fifo_entries; i += st->fifo_set_size)
 		iio_push_to_buffers(indio_dev, &st->fifo_buf[i]);
-	}
 
 	return IRQ_HANDLED;
 }
-- 
2.39.5

Re: [PATCH v2 2/2] iio: accel: adxl380: Optimize reading of FIFO entries in interrupt handler

Posted by Jonathan Cameron 2 weeks, 3 days ago

On Mon, 19 Jan 2026 11:23:17 +0100
Francesco Lavra <flavra@baylibre.com> wrote:

> In order to minimize the time required for transferring FIFO data from the
> sensor to the host machine, perform the read from the FIFO in a single call
> to regmap_bulk_read().
It's a call to regmap_noinc_read()  not regmap_bulk_read()
I'll fix it.

> This allows reading acceleration data for all 3 axes at 16 kHz
> sampling frequency using a 1MHz I2C bus frequency.
> 
> Signed-off-by: Francesco Lavra <flavra@baylibre.com>
> ---
>  drivers/iio/accel/adxl380.c | 12 +++++-------
>  1 file changed, 5 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/iio/accel/adxl380.c b/drivers/iio/accel/adxl380.c
> index 650cdbffd4a7..e3e6b182eb3d 100644
> --- a/drivers/iio/accel/adxl380.c
> +++ b/drivers/iio/accel/adxl380.c
> @@ -965,14 +965,12 @@ static irqreturn_t adxl380_irq_handler(int irq, void  *p)
>  		return IRQ_HANDLED;
>  
>  	fifo_entries = rounddown(fifo_entries, st->fifo_set_size);
> -	for (i = 0; i < fifo_entries; i += st->fifo_set_size) {
> -		ret = regmap_noinc_read(st->regmap, ADXL380_FIFO_DATA,
> -					&st->fifo_buf[i],
> -					2 * st->fifo_set_size);
> -		if (ret)
> -			return IRQ_HANDLED;
> +	ret = regmap_noinc_read(st->regmap, ADXL380_FIFO_DATA, &st->fifo_buf,
> +				sizeof(*st->fifo_buf) * fifo_entries);
> +	if (ret)
> +		return IRQ_HANDLED;
> +	for (i = 0; i < fifo_entries; i += st->fifo_set_size)
>  		iio_push_to_buffers(indio_dev, &st->fifo_buf[i]);
> -	}
>  
>  	return IRQ_HANDLED;
>  }

Re: [PATCH v2 2/2] iio: accel: adxl380: Optimize reading of FIFO entries in interrupt handler

Posted by Andy Shevchenko 2 weeks, 2 days ago

On Thu, Jan 22, 2026 at 07:53:13PM +0000, Jonathan Cameron wrote:
> On Mon, 19 Jan 2026 11:23:17 +0100
> Francesco Lavra <flavra@baylibre.com> wrote:
> 
> > In order to minimize the time required for transferring FIFO data from the
> > sensor to the host machine, perform the read from the FIFO in a single call
> > to regmap_bulk_read().

> It's a call to regmap_noinc_read()  not regmap_bulk_read()
> I'll fix it.

This is interesting, does it mean the patch was never tested on the real HW?
Or maybe noninc is intentional?

-- 
With Best Regards,
Andy Shevchenko

Re: [PATCH v2 2/2] iio: accel: adxl380: Optimize reading of FIFO entries in interrupt handler

Posted by Jonathan Cameron 2 weeks, 2 days ago

On Fri, 23 Jan 2026 10:13:16 +0200
Andy Shevchenko <andriy.shevchenko@intel.com> wrote:

> On Thu, Jan 22, 2026 at 07:53:13PM +0000, Jonathan Cameron wrote:
> > On Mon, 19 Jan 2026 11:23:17 +0100
> > Francesco Lavra <flavra@baylibre.com> wrote:
> >   
> > > In order to minimize the time required for transferring FIFO data from the
> > > sensor to the host machine, perform the read from the FIFO in a single call
> > > to regmap_bulk_read().  
> 
> > It's a call to regmap_noinc_read()  not regmap_bulk_read()
> > I'll fix it.  
> 
> This is interesting, does it mean the patch was never tested on the real HW?
> Or maybe noninc is intentional?
> 
It's absolutely intentional.  The device presents a single register address
that has the fifo behind it (which is what we added noinc for years ago).
So patch should definitely work.  Goes from reading that register N times in
each call within a loop of M to reading it N x M.

If regcache is in use this makes a difference as avoids corrupting registers
after this point as the address increment expected in the hardware for
a bulk read doesn't happen.  Without regcache it's just "documentation" as
at least for protocols used here its the same as a bulk read on the bus.

So patch is fine, it just mentioned the wrong call in the description
I'd guess this is because someone is working with a tree that has the wrong call in
it but might be wrong.

Jonathan

Re: [PATCH v2 2/2] iio: accel: adxl380: Optimize reading of FIFO entries in interrupt handler

Posted by Francesco Lavra 2 weeks, 2 days ago

On Fri, 2026-01-23 at 09:54 +0000, Jonathan Cameron wrote:
> On Fri, 23 Jan 2026 10:13:16 +0200
> Andy Shevchenko <andriy.shevchenko@intel.com> wrote:
> 
> > On Thu, Jan 22, 2026 at 07:53:13PM +0000, Jonathan Cameron wrote:
> > > On Mon, 19 Jan 2026 11:23:17 +0100
> > > Francesco Lavra <flavra@baylibre.com> wrote:
> > >   
> > > > In order to minimize the time required for transferring FIFO data
> > > > from the
> > > > sensor to the host machine, perform the read from the FIFO in a
> > > > single call
> > > > to regmap_bulk_read().  
> > 
> > > It's a call to regmap_noinc_read()  not regmap_bulk_read()
> > > I'll fix it.  
> > 
> > This is interesting, does it mean the patch was never tested on the
> > real HW?

The driver has been tested (by me) on real hardware, both before and after
this patch.

> > Or maybe noninc is intentional?
> > 
> It's absolutely intentional.  The device presents a single register
> address
> that has the fifo behind it (which is what we added noinc for years ago).
> So patch should definitely work.  Goes from reading that register N times
> in
> each call within a loop of M to reading it N x M.
> 
> If regcache is in use this makes a difference as avoids corrupting
> registers
> after this point as the address increment expected in the hardware for
> a bulk read doesn't happen.  Without regcache it's just "documentation"
> as
> at least for protocols used here its the same as a bulk read on the bus.
> 
> So patch is fine, it just mentioned the wrong call in the description
> I'd guess this is because someone is working with a tree that has the
> wrong call in
> it but might be wrong.

I have this patch applied also to older kernel versions where
regmap_noinc_read() doesn't exist, so this is just a copy-paste error in
the commit message. Sorry about that.

Re: [PATCH v2 2/2] iio: accel: adxl380: Optimize reading of FIFO entries in interrupt handler

Posted by Andy Shevchenko 2 weeks, 2 days ago

On Fri, Jan 23, 2026 at 12:09:36PM +0100, Francesco Lavra wrote:
> On Fri, 2026-01-23 at 09:54 +0000, Jonathan Cameron wrote:
> > On Fri, 23 Jan 2026 10:13:16 +0200
> > Andy Shevchenko <andriy.shevchenko@intel.com> wrote:
> > > On Thu, Jan 22, 2026 at 07:53:13PM +0000, Jonathan Cameron wrote:
> > > > On Mon, 19 Jan 2026 11:23:17 +0100
> > > > Francesco Lavra <flavra@baylibre.com> wrote:
> > > >   
> > > > > In order to minimize the time required for transferring FIFO data
> > > > > from the
> > > > > sensor to the host machine, perform the read from the FIFO in a
> > > > > single call
> > > > > to regmap_bulk_read().  
> > > 
> > > > It's a call to regmap_noinc_read()  not regmap_bulk_read()
> > > > I'll fix it.  
> > > 
> > > This is interesting, does it mean the patch was never tested on the
> > > real HW?
> 
> The driver has been tested (by me) on real hardware, both before and after
> this patch.
> 
> > > Or maybe noninc is intentional?
> > > 
> > It's absolutely intentional.  The device presents a single register
> > address
> > that has the fifo behind it (which is what we added noinc for years ago).
> > So patch should definitely work.  Goes from reading that register N times
> > in
> > each call within a loop of M to reading it N x M.
> > 
> > If regcache is in use this makes a difference as avoids corrupting
> > registers
> > after this point as the address increment expected in the hardware for
> > a bulk read doesn't happen.  Without regcache it's just "documentation"
> > as
> > at least for protocols used here its the same as a bulk read on the bus.
> > 
> > So patch is fine, it just mentioned the wrong call in the description
> > I'd guess this is because someone is working with a tree that has the
> > wrong call in
> > it but might be wrong.
> 
> I have this patch applied also to older kernel versions where
> regmap_noinc_read() doesn't exist, so this is just a copy-paste error in
> the commit message. Sorry about that.

Thanks for clarifying this!

-- 
With Best Regards,
Andy Shevchenko