Re: [PATCH v3 1/5] iio: ad7949: Fix dummy read cycle placement

[Alexandru Ardelean <ardeleanalex@gmail.com>]

On Mon, May 13, 2019 at 7:18 PM Adam Michaelis
<adam.michaelis@rockwellcollins.com> wrote:
>
> The AD7949 requires two conversion cycles following the first
> configuration change, and one extra cycle following any other
> configuration change (including changing the analog channel being
> sampled). Therefore, adding a dummy read cycle when config is changed
> and removing the extra cycle at initial configuration (the first dummy
> cycle is now performed as part of applying the configuration change).
>

CC-ing my work email.
And Stefan as well.

We'll have to try this driver & changes internally a bit.
Hopefully we have a compatible device around the office.
Since it wasn't written by us, we're also unsure about things just by
looking at the code.

> Signed-off-by: Adam Michaelis <adam.michaelis@rockwellcollins.com>
> ---
>         V2:
>         - Add some defines to reduce use of magic numbers.
>         V3:
>         - Switch back to using a u32 data buffer.
>         - Add-back the second dummy cycle on initialization.
>         - Move to first patch in series.
> ---
>  drivers/iio/adc/ad7949.c | 20 ++++++++++++++++++--
>  1 file changed, 18 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/iio/adc/ad7949.c b/drivers/iio/adc/ad7949.c
> index ac0ffff6c5ae..c7fe27aa2519 100644
> --- a/drivers/iio/adc/ad7949.c
> +++ b/drivers/iio/adc/ad7949.c
> @@ -100,6 +100,23 @@ static int ad7949_spi_write_cfg(struct ad7949_adc_chip *ad7949_adc, u16 val,
>          * send a new command to the device
>          */
>         udelay(2);
> +
> +       /*
> +        * Perform extra read cycle to allow configuration, acquisition,
> +        * and conversion sequences to complete for new configuration.
> +        */
> +       ad7949_adc->buffer = 0;
> +
> +       spi_message_init_with_transfers(&msg, tx, 1);
> +
> +       ret = spi_sync(ad7949_adc->spi, &msg);
> +
> +       /*
> +        * This delay is to avoid a new request before the required time
> +        * to send a new command to the device.
> +        */
> +       udelay(2);
> +

Is this needed if the channel doesn't change ?
If it isn't, maybe add a check in ad7949_spi_read_channel() to skip
the call to ad7949_spi_write_cfg().

We should also take performance into account when doing SPI
transactions to the device, and if we can skip some of them, all the
better.
This change, introduces a performance penalty by doing this extra read + udelay.

>         return ret;
>  }
>
> @@ -229,11 +246,10 @@ static int ad7949_spi_init(struct ad7949_adc_chip *ad7949_adc)
>         ret = ad7949_spi_write_cfg(ad7949_adc, 0x3C79, AD7949_MASK_TOTAL);
>
>         /*
> -        * Do two dummy conversions to apply the first configuration setting.
> +        * Do a dummy conversion to apply the first configuration setting.
>          * Required only after the start up of the device.
>          */
>         ad7949_spi_read_channel(ad7949_adc, &val, ad7949_adc->current_channel);
> -       ad7949_spi_read_channel(ad7949_adc, &val, ad7949_adc->current_channel);

The datasheet mentions that 2 dummy conversions are needed on power-up / init.

The way, this was done here was a bit easier to follow (or
straightforward) with the datasheet.

This isn't to say that this is bad, but if we need to do an extra SPI
read (and skip the SPI write part), then I would just add an SPI read
function, instead of moving it completely into ad7949_spi_write_cfg(),
which would then compact things in ad7949_spi_read_channel().

>
>         return ret;
>  }
> --
> 1.9.1
>