Re: [PATCH v9 2/2] iio: adc: ad4130: add AD4130 driver

[Cosmin Tanislav]

On 10/9/22 20:31, Jonathan Cameron wrote:
> On Thu,  6 Oct 2022 17:07:37 +0300
> Cosmin Tanislav <demonsingur@xxxxxxxxx> wrote:
>
> > AD4130-8 is an ultra-low power, high precision, measurement solution for
> > low bandwidth battery operated applications.
> >
> > The fully integrated AFE (Analog Front-End) includes a multiplexer for up
> > to 16 single-ended or 8 differential inputs, PGA (Programmable Gain
> > Amplifier), 24-bit Sigma-Delta ADC, on-chip reference and oscillator,
> > selectable filter options, smart sequencer, sensor biasing and excitation
> > options, diagnostics, and a FIFO buffer.
> >
> > Signed-off-by: Cosmin Tanislav <cosmin.tanislav@xxxxxxxxxx>
> > Reviewed-by: Andy Shevchenko <andy.shevchenko@xxxxxxxxx>
> Hi Cosmin,
>
> I've cropped down (mostly) to the clock changes.
> A few minor things in there + this looks like it would suffer from the issue
> with IIO_CONST_ATTR() not being handled correctly for buffer attributes.
>
> Jonathan
>
>
>
> > +static IIO_CONST_ATTR(hwfifo_watermark_min, "1");
> > +static IIO_CONST_ATTR(hwfifo_watermark_max, __stringify(AD4130_FIFO_SIZE));
>
> These look like they'd suffer from same problem
> https://lore.kernel.org/all/cover.1664782676.git.mazziesaccount@xxxxxxxxx/
> tackles.  Short term fix is don't use IIO_CONST_ATTR for buffer attributes.

Right, this only works downstream.

Should I switch to IIO_STATIC_CONST_DEVICE_ATTR?

> > +static IIO_DEVICE_ATTR_RO(hwfifo_watermark, 0);
> > +static IIO_DEVICE_ATTR_RO(hwfifo_enabled, 0);
> > +
> > +static const struct attribute *ad4130_fifo_attributes[] = {
> > +	&iio_const_attr_hwfifo_watermark_min.dev_attr.attr,
> > +	&iio_const_attr_hwfifo_watermark_max.dev_attr.attr,
> > +	&iio_dev_attr_hwfifo_watermark.dev_attr.attr,
> > +	&iio_dev_attr_hwfifo_enabled.dev_attr.attr,
> > +	NULL
> > +};
>
>
> > +static void ad4130_clk_disable_unprepare(void *clk)
> > +{
> > +	clk_disable_unprepare(clk);
> > +}
> > +
> > +static int ad4130_set_mclk_sel(struct ad4130_state *st,
> > +			       enum ad4130_mclk_sel mclk_sel)
> > +{
> > +	return regmap_update_bits(st->regmap, AD4130_ADC_CONTROL_REG,
> > +				 AD4130_ADC_CONTROL_MCLK_SEL_MASK,
> > +				 FIELD_PREP(AD4130_ADC_CONTROL_MCLK_SEL_MASK,
> > +					    mclk_sel));
> > +}
> > +
> > +static unsigned long ad4130_int_clk_recalc_rate(struct clk_hw *hw,
> > +						unsigned long parent_rate)
> > +{
> > +	return AD4130_MCLK_FREQ_76_8KHZ;
> > +}
> > +
> > +static int ad4130_int_clk_is_enabled(struct clk_hw *hw)
> > +{
> > +	struct ad4130_state *st = container_of(hw, struct ad4130_state, int_clk_hw);
> > +
> > +	return st->mclk_sel == AD4130_MCLK_76_8KHZ_OUT;
> > +}
> > +
> > +static int ad4130_int_clk_prepare(struct clk_hw *hw)
> > +{
> > +	struct ad4130_state *st = container_of(hw, struct ad4130_state, int_clk_hw);
> > +	int ret;
> > +
> > +	ret = ad4130_set_mclk_sel(st, AD4130_MCLK_76_8KHZ_OUT);
> > +	if (ret)
> > +		return ret;
> > +
> > +	st->mclk_sel = AD4130_MCLK_76_8KHZ_OUT;
> > +
> > +	return 0;
> > +}
> > +
> > +static void ad4130_int_clk_unprepare(struct clk_hw *hw)
> > +{
> > +	struct ad4130_state *st = container_of(hw, struct ad4130_state, int_clk_hw);
> > +	int ret;
> > +
> > +	ret = ad4130_set_mclk_sel(st, AD4130_MCLK_76_8KHZ);
> > +	if (ret)
> > +		return;
> > +
> > +	st->mclk_sel = AD4130_MCLK_76_8KHZ;
> > +}
> > +
> > +static const struct clk_ops ad4130_int_clk_ops = {
> > +	.recalc_rate = ad4130_int_clk_recalc_rate,
> > +	.is_enabled = ad4130_int_clk_is_enabled,
> > +	.prepare = ad4130_int_clk_prepare,
> > +	.unprepare = ad4130_int_clk_unprepare,
> > +};
> > +
> > +static int ad4130_setup_int_clk(struct ad4130_state *st)
> > +{
> > +	struct device *dev = &st->spi->dev;
> > +	struct device_node *of_node = dev->of_node;
>
> Hmm. There goes our careful use of generic firmware properties.
> I guess there still isn't much we can do about that for clks
> so at least it's contained to this one function.
>
> Also is this code safe to of_node == NULL?

No, I guess it is not. I'll fix it.
Should I just
if (!of_node) return 0;
?

> > +	struct clk_init_data init;
> > +	const char *clk_name;
> > +	struct clk *clk;
> > +
> > +	if (st->int_pin_sel == AD4130_INT_PIN_CLK ||
> > +	    st->mclk_sel != AD4130_MCLK_76_8KHZ)
> > +		return 0;
> > +
> > +	clk_name = of_node->name;
> > +	of_property_read_string(of_node, "clock-output-names", &clk_name);
>
> Probably want to check success of that read before using it.
> I'd also expect that these to be optional + doesn't he dt binding need
> updating to add this stuff?

It does need updating, sorry.
of_node->name is the default clk_name, if clock-output-names is present
then the of_property_read_string() result will be used instead. If not,
there's no trouble, and we don't care about the return value since we
have the default clk_name assigned just above.
I can also switch to device_property_read_string() here to minimize the
damage from using OF.

> > +
> > +	init.name = clk_name;
> > +	init.ops = &ad4130_int_clk_ops;
> > +
> > +	st->int_clk_hw.init = &init;
> > +	clk = devm_clk_register(dev, &st->int_clk_hw);
> > +	if (IS_ERR(clk))
> > +		return PTR_ERR(clk);
> > +
> > +	return of_clk_add_provider(of_node, of_clk_src_simple_get, clk);
> > +}
> > +