Re: [PATCH v2 1/2] staging: iio: frequency: ad9834: Move frequency to standard iio types

From: Jonathan Cameron
Date: Sun Mar 03 2019 - 12:02:52 EST


On Fri, 22 Feb 2019 13:10:26 +0000
"Bia, Beniamin" <Beniamin.Bia@xxxxxxxxxx> wrote:

> Thank you for taking some time reviewing my code.
> Ad9833/ad9834 has two frequency and phase registers and only one
> output. The user can select which register is selected as input with a
> mux.Because I wanted to reduce as much as possible the custom
> attributes, I mapped the frequency 0 from register to channel 0 and
> frequency 1 to channel 1. Same rule applies for phase.
I can see the argument, but really don't think this ends up as an
intuitive mapping. I think we need something new to handle
values that change with 'symbol'.

>
> In order to keep the same structure as before:
> out_altvoltage0_frequency0 and out_altvoltage1_frequency1,
> we would have to create custom attributes for every frequency and phase value.
> This is the only way as far as I know, but I don't have a lot of experience with iio. What is your opinion? Do you have a better solution?

Sorry it took me so long to reply to this.

So, we have had similar issues before, particularly around power
meters where we need a 'broader' concept of index to allow for multiple
related measurements.

However, in this case I think we can sort of think of the values being
aspects controlled by a symbol, which might map onto an 'axis' for
example (stretching the definition).

So maybe define some new modifiers, symbolA, symbolB for example giving
us
out_altvoltage0_symbolA_frequency
out_altvoltage0_symbolB_frequency.

This is a discussion I'd like others to contribute to though as
interface design is always a pain to get right!



>
> Thanks,
> Ben
> > [External]
> >
> >
> > On Thu, 14 Feb 2019 18:41:29 +0200
> > Beniamin Bia <beniamin.bia@xxxxxxxxxx> wrote:
> >
> > > Frequency attribute is added with a standard type from iio
> > > framework
> > > instead of custom attribute. This is a small step towards removing
> > > any
> > > unnecessary custom attribute. Ad9834 will diverge from ad9833 in
> > > the
> > > future, that is why we have two identical arrays for ad9834 and
> > > 9833.
> > >
> > > Signed-off-by: Beniamin Bia <beniamin.bia@xxxxxxxxxx>
> >
> > Hi Beniamin
> >
> > When you make a change like this, please explain in detail how the
> > ABI
> > changes. Here I'm not sure we can actually map it to channels like
> > this.
> > We are controlling the two frequencies of FSK not independent
> > channels.
> >
> > The ABI around this needs very careful thinking out. I would
> > document
> > your proposed ABI first and share that. Once we reach agreement on
> > the
> > ABI, the actual code should be more straight forward!
> >
> > Thanks,
> >
> > Jonathan
> >
> > > ---
> > > Changes in v2:
> > > -the personal email address was replaced by the work email
> > > -separate define for frequency channel
> > > -address field from channel specification was removed
> > > -frequency variables were replaced by an array
> > > -specified in comment why we have differente chan_spec for
> > > ad9834 and ad9833
> > > -enum used for write_frequency function
> > >
> > > drivers/staging/iio/frequency/ad9834.c | 110 ++++++++++++++++++++-
> > > ----
> > > 1 file changed, 91 insertions(+), 19 deletions(-)
> > >
> > > diff --git a/drivers/staging/iio/frequency/ad9834.c
> > > b/drivers/staging/iio/frequency/ad9834.c
> > > index f036f75d1f22..561617046c20 100644
> > > --- a/drivers/staging/iio/frequency/ad9834.c
> > > +++ b/drivers/staging/iio/frequency/ad9834.c
> > > @@ -81,6 +81,8 @@ struct ad9834_state {
> > > struct spi_message freq_msg;
> > > struct mutex lock; /* protect sensor
> > > state */
> > >
> > > + unsigned long frequency[2];
> > > +
> > > /*
> > > * DMA (thus cache coherency maintenance) requires the
> > > * transfer buffers to live in their own cache lines.
> > > @@ -89,6 +91,11 @@ struct ad9834_state {
> > > __be16 freq_data[2];
> > > };
> > >
> > > +enum ad9834_ch_addr {
> > > + AD9834_CHANNEL_ADDRESS0,
> > > + AD9834_CHANNEL_ADDRESS1,
> > > +};
> > > +
> > > /**
> > > * ad9834_supported_device_ids:
> > > */
> > > @@ -100,6 +107,24 @@ enum ad9834_supported_device_ids {
> > > ID_AD9838,
> > > };
> > >
> > > +#define AD9833_CHANNEL(chan)
> > > { \
> > > + .type =
> > > IIO_ALTVOLTAGE, \
> > > + .indexed =
> > > 1, \
> > > + .output =
> > > 1, \
> > > + .channel =
> > > (chan), \
> > > + .info_mask_separate =
> > > BIT(IIO_CHAN_INFO_FREQUENCY) \
> > > +}
> > > +
> > > +static const struct iio_chan_spec ad9833_channels[] = {
> > > + AD9833_CHANNEL(0),
> > > + AD9833_CHANNEL(1),
> > > +};
> > > +
> > > +static const struct iio_chan_spec ad9834_channels[] = {
> > > + AD9833_CHANNEL(0),
> > > + AD9833_CHANNEL(1),
> > > +};
> > > +
> > > static unsigned int ad9834_calc_freqreg(unsigned long mclk,
> > > unsigned long fout)
> > > {
> > > unsigned long long freqreg = (u64)fout *
> > > (u64)BIT(AD9834_FREQ_BITS);
> > > @@ -109,10 +134,12 @@ static unsigned int
> > > ad9834_calc_freqreg(unsigned long mclk, unsigned long fout)
> > > }
> > >
> > > static int ad9834_write_frequency(struct ad9834_state *st,
> > > - unsigned long addr, unsigned long
> > > fout)
> > > + enum ad9834_ch_addr addr,
> > > + unsigned long fout)
> > > {
> > > unsigned long clk_freq;
> > > unsigned long regval;
> > > + int ret;
> > >
> > > clk_freq = clk_get_rate(st->mclk);
> > >
> > > @@ -121,13 +148,27 @@ static int ad9834_write_frequency(struct
> > > ad9834_state *st,
> > >
> > > regval = ad9834_calc_freqreg(clk_freq, fout);
> > >
> > > - st->freq_data[0] = cpu_to_be16(addr | (regval &
> > > - RES_MASK(AD9834_FREQ_BITS /
> > > 2)));
> > > - st->freq_data[1] = cpu_to_be16(addr | ((regval >>
> > > - (AD9834_FREQ_BITS / 2)) &
> > > - RES_MASK(AD9834_FREQ_BITS /
> > > 2)));
> > > + if (addr == AD9834_CHANNEL_ADDRESS0) {
> >
> > So this if statement picks between the two addresses with no other
> > differences?
> > That's fine, but use a local variable for the register address to
> > simplify
> > the code.
> >
> > > + st->freq_data[0] = cpu_to_be16(AD9834_REG_FREQ0 |
> > > (regval &
> > > + RES_MASK(AD9834_FREQ_B
> > > ITS / 2)));
> > > + st->freq_data[1] = cpu_to_be16(AD9834_REG_FREQ0 |
> > > ((regval >>
> > > + (AD9834_FREQ_BITS /
> > > 2)) &
> > > + RES_MASK(AD9834_FREQ_B
> > > ITS / 2)));
> > > + } else {
> > > + st->freq_data[0] = cpu_to_be16(AD9834_REG_FREQ1 |
> > > (regval &
> > > + RES_MASK(AD9834_FREQ_B
> > > ITS / 2)));
> > > + st->freq_data[1] = cpu_to_be16(AD9834_REG_FREQ1 |
> > > ((regval >>
> > > + (AD9834_FREQ_BITS /
> > > 2)) &
> > > + RES_MASK(AD9834_FREQ_B
> > > ITS / 2)));
> > > + }
> > > +
> > > + ret = spi_sync(st->spi, &st->freq_msg);
> > > + if (ret)
> > > + return ret;
> > > +
> > > + st->frequency[(int)addr] = fout;
> > >
> > > - return spi_sync(st->spi, &st->freq_msg);
> > > + return 0;
> > > }
> > >
> > > static int ad9834_write_phase(struct ad9834_state *st,
> > > @@ -140,6 +181,39 @@ static int ad9834_write_phase(struct
> > > ad9834_state *st,
> > > return spi_sync(st->spi, &st->msg);
> > > }
> > >
> > > +static int ad9834_read_raw(struct iio_dev *indio_dev,
> > > + struct iio_chan_spec const *chan,
> > > + int *val, int *val2, long mask)
> > > +{
> > > + struct ad9834_state *st = iio_priv(indio_dev);
> > > +
> > > + switch (mask) {
> > > + case IIO_CHAN_INFO_FREQUENCY:
> > > + *val = st->frequency[chan->channel];
> > > + return IIO_VAL_INT;
> > > + }
> > > +
> > > + return -EINVAL;
> > > +}
> > > +
> > > +static int ad9834_write_raw(struct iio_dev *indio_dev,
> > > + struct iio_chan_spec const *chan,
> > > + int val, int val2, long mask)
> > > +{
> > > + struct ad9834_state *st = iio_priv(indio_dev);
> > > +
> > > + switch (mask) {
> > > + case IIO_CHAN_INFO_FREQUENCY:
> > > + return ad9834_write_frequency(st,
> > > + (enum
> > > ad9834_ch_addr)chan->channel,
> > > + val);
> > > + default:
> > > + return -EINVAL;
> > > + }
> > > +
> > > + return 0;
> > > +}
> > > +
> > > static ssize_t ad9834_write(struct device *dev,
> > > struct device_attribute *attr,
> > > const char *buf,
> > > @@ -157,10 +231,6 @@ static ssize_t ad9834_write(struct device
> > > *dev,
> > >
> > > mutex_lock(&st->lock);
> > > switch ((u32)this_attr->address) {
> > > - case AD9834_REG_FREQ0:
> > > - case AD9834_REG_FREQ1:
> > > - ret = ad9834_write_frequency(st, this_attr->address,
> > > val);
> > > - break;
> > > case AD9834_REG_PHASE0:
> > > case AD9834_REG_PHASE1:
> > > ret = ad9834_write_phase(st, this_attr->address,
> > > val);
> > > @@ -323,8 +393,6 @@ static
> > > IIO_DEVICE_ATTR(out_altvoltage0_out1_wavetype_available, 0444,
> > > * see dds.h for further information
> > > */
> > >
> > > -static IIO_DEV_ATTR_FREQ(0, 0, 0200, NULL, ad9834_write,
> > > AD9834_REG_FREQ0);
> > > -static IIO_DEV_ATTR_FREQ(0, 1, 0200, NULL, ad9834_write,
> > > AD9834_REG_FREQ1);
> > > static IIO_DEV_ATTR_FREQSYMBOL(0, 0200, NULL, ad9834_write,
> > > AD9834_FSEL);
> > > static IIO_CONST_ATTR_FREQ_SCALE(0, "1"); /* 1Hz */
> > >
> > > @@ -342,8 +410,6 @@ static IIO_DEV_ATTR_OUT_WAVETYPE(0, 0,
> > > ad9834_store_wavetype, 0);
> > > static IIO_DEV_ATTR_OUT_WAVETYPE(0, 1, ad9834_store_wavetype, 1);
> > >
> > > static struct attribute *ad9834_attributes[] = {
> > > - &iio_dev_attr_out_altvoltage0_frequency0.dev_attr.attr,
> > > - &iio_dev_attr_out_altvoltage0_frequency1.dev_attr.attr,
> > > &iio_const_attr_out_altvoltage0_frequency_scale.dev_attr.attr
> > > ,
> > > &iio_dev_attr_out_altvoltage0_phase0.dev_attr.attr,
> > > &iio_dev_attr_out_altvoltage0_phase1.dev_attr.attr,
> > > @@ -361,8 +427,6 @@ static struct attribute *ad9834_attributes[] =
> > > {
> > > };
> > >
> > > static struct attribute *ad9833_attributes[] = {
> > > - &iio_dev_attr_out_altvoltage0_frequency0.dev_attr.attr,
> > > - &iio_dev_attr_out_altvoltage0_frequency1.dev_attr.attr,
> >
> > This may be a problem... These don't fit in the existing ABI in that
> > they
> > are (IIRC) two different frequencies for the same channel, not two
> > different
> > channels... This is meant for simple FSK modulation.
> >
> > > &iio_const_attr_out_altvoltage0_frequency_scale.dev_attr.attr
> > > ,
> > > &iio_dev_attr_out_altvoltage0_phase0.dev_attr.attr,
> > > &iio_dev_attr_out_altvoltage0_phase1.dev_attr.attr,
> > > @@ -384,11 +448,15 @@ static const struct attribute_group
> > > ad9833_attribute_group = {
> > > };
> > >
> > > static const struct iio_info ad9834_info = {
> > > + .write_raw = &ad9834_write_raw,
> > > + .read_raw = &ad9834_read_raw,
> > > .attrs = &ad9834_attribute_group,
> > > .driver_module = THIS_MODULE,
> > > };
> > >
> > > static const struct iio_info ad9833_info = {
> > > + .write_raw = &ad9834_write_raw,
> > > + .read_raw = &ad9834_read_raw,
> > > .attrs = &ad9833_attribute_group,
> > > .driver_module = THIS_MODULE,
> > > };
> > > @@ -435,9 +503,13 @@ static int ad9834_probe(struct spi_device
> > > *spi)
> > > switch (st->devid) {
> > > case ID_AD9833:
> > > case ID_AD9837:
> > > + indio_dev->channels = ad9833_channels;
> > > + indio_dev->num_channels =
> > > ARRAY_SIZE(ad9833_channels);
> > > indio_dev->info = &ad9833_info;
> > > break;
> > > default:
> >
> > For future reference, I would definitely like to see the options that
> > lead to here explicitly listed. Semantically it's not a 'default'
> > but rather an alternative equally valid choice.
> > People also tend to grep for their devid ;)
> >
> > > + indio_dev->channels = ad9834_channels;
> > > + indio_dev->num_channels =
> > > ARRAY_SIZE(ad9834_channels);
> > > indio_dev->info = &ad9834_info;
> > > break;
> > > }
> > > @@ -474,11 +546,11 @@ static int ad9834_probe(struct spi_device
> > > *spi)
> > > goto error_clock_unprepare;
> > > }
> > >
> > > - ret = ad9834_write_frequency(st, AD9834_REG_FREQ0, 1000000);
> > > + ret = ad9834_write_frequency(st, AD9834_CHANNEL_ADDRESS0,
> > > 1000000);
> > > if (ret)
> > > goto error_clock_unprepare;
> > >
> > > - ret = ad9834_write_frequency(st, AD9834_REG_FREQ1, 5000000);
> > > + ret = ad9834_write_frequency(st, AD9834_CHANNEL_ADDRESS1,
> > > 5000000);
> > > if (ret)
> > > goto error_clock_unprepare;
> > >
> >
> >