Re: [PATCH v7 4/8] iio: dac: adi-axi-dac: extend features

[Angelo Dureghello]

On 24.10.2024 09:04, Nuno Sá wrote:
> On Wed, 2024-10-23 at 16:22 +0100, Conor Dooley wrote:
> > On Wed, Oct 23, 2024 at 04:56:39PM +0200, Nuno Sá wrote:
> > > On Tue, 2024-10-22 at 18:21 +0100, Conor Dooley wrote:
> > > > On Tue, Oct 22, 2024 at 02:36:44PM +0200, Nuno Sá wrote:
> > > > > On Mon, 2024-10-21 at 14:40 +0200, Angelo Dureghello wrote:
> > > > > > From: Angelo Dureghello <adureghello@xxxxxxxxxxxx>
> > > > > > 
> > > > > > Extend AXI-DAC backend with new features required to interface
> > > > > > to the ad3552r DAC. Mainly, a new compatible string is added to
> > > > > > support the ad3552r-axi DAC IP, very similar to the generic DAC
> > > > > > IP but with some customizations to work with the ad3552r.
> > > > > > 
> > > > > > Then, a series of generic functions has been added to match with
> > > > > > ad3552r needs. Function names has been kept generic as much as
> > > > > > possible, to allow re-utilization from other frontend drivers.
> > > > > > 
> > > > > > Signed-off-by: Angelo Dureghello <adureghello@xxxxxxxxxxxx>
> > > > > > ---
> > > > > 
> > > > > Looks mostly good,
> > > > > 
> > > > > one minor thing that (I think) could be improved
> > > > > >  drivers/iio/dac/adi-axi-dac.c | 269
> > > > > > +++++++++++++++++++++++++++++++++++++++--
> > > > > > -
> > > > > >  1 file changed, 255 insertions(+), 14 deletions(-)
> > > > > > 
> > > > > > diff --git a/drivers/iio/dac/adi-axi-dac.c b/drivers/iio/dac/adi-axi-dac.c
> > > > > > index 04193a98616e..9d6809fe7a67 100644
> > > > > > --- a/drivers/iio/dac/adi-axi-dac.c
> > > > > > +++ b/drivers/iio/dac/adi-axi-dac.c
> > > > > > @@ -46,9 +46,28 @@
> > > > > >  #define AXI_DAC_CNTRL_1_REG			0x0044
> > > > > >  #define   AXI_DAC_CNTRL_1_SYNC			BIT(0)
> > > > > >  #define AXI_DAC_CNTRL_2_REG			0x0048
> > > > > > +#define   AXI_DAC_CNTRL_2_SDR_DDR_N		BIT(16)
> > > > > > +#define   AXI_DAC_CNTRL_2_SYMB_8B		BIT(14)
> > > > > >  #define   ADI_DAC_CNTRL_2_R1_MODE		BIT(5)
> > > > > > +#define   AXI_DAC_CNTRL_2_UNSIGNED_DATA		BIT(4)
> > > > > > +#define AXI_DAC_STATUS_1_REG			0x0054
> > > > > > +#define AXI_DAC_STATUS_2_REG			0x0058
> > > > > >  #define AXI_DAC_DRP_STATUS_REG			0x0074
> > > > > >  #define   AXI_DAC_DRP_STATUS_DRP_LOCKED		BIT(17)
> > > > > > +#define AXI_DAC_CUSTOM_RD_REG			0x0080
> > > > > > +#define AXI_DAC_CUSTOM_WR_REG			0x0084
> > > > > > +#define   AXI_DAC_CUSTOM_WR_DATA_8		GENMASK(23, 16)
> > > > > > +#define   AXI_DAC_CUSTOM_WR_DATA_16		GENMASK(23, 8)
> > > > > > +#define AXI_DAC_UI_STATUS_REG			0x0088
> > > > > > +#define   AXI_DAC_UI_STATUS_IF_BUSY		BIT(4)
> > > > > > +#define AXI_DAC_CUSTOM_CTRL_REG			0x008C
> > > > > > +#define   AXI_DAC_CUSTOM_CTRL_ADDRESS		GENMASK(31, 24)
> > > > > > +#define   AXI_DAC_CUSTOM_CTRL_SYNCED_TRANSFER	BIT(2)
> > > > > > +#define   AXI_DAC_CUSTOM_CTRL_STREAM		BIT(1)
> > > > > > +#define   AXI_DAC_CUSTOM_CTRL_TRANSFER_DATA	BIT(0)
> > > > > 
> > > > > ...
> > > > >  
> > > > > >  static int axi_dac_probe(struct platform_device *pdev)
> > > > > >  {
> > > > > > -	const unsigned int *expected_ver;
> > > > > >  	struct axi_dac_state *st;
> > > > > >  	void __iomem *base;
> > > > > >  	unsigned int ver;
> > > > > > @@ -566,14 +780,29 @@ static int axi_dac_probe(struct platform_device
> > > > > > *pdev)
> > > > > >  	if (!st)
> > > > > >  		return -ENOMEM;
> > > > > >  
> > > > > > -	expected_ver = device_get_match_data(&pdev->dev);
> > > > > > -	if (!expected_ver)
> > > > > > +	st->info = device_get_match_data(&pdev->dev);
> > > > > > +	if (!st->info)
> > > > > >  		return -ENODEV;
> > > > > > +	clk = devm_clk_get_enabled(&pdev->dev, "s_axi_aclk");
> > > > > > +	if (IS_ERR(clk)) {
> > > > > 
> > > > > If clock-names is not given, then we'll get -EINVAL. Hence we could assume
> > > > > that:
> > > > > 
> > > > > 		if (PTR_ERR(clk) != -EINVAL)
> > > > > 			return dev_err_probe();
> > > > 
> > > > clock-names isn't a required property, but the driver code effectively
> > > > makes it one. Doesn't this lookup need to be by index, unless
> > > > clock-names is made required for this variant?
> > > 
> > > Likely I'm missing something but the driver is not making clock-names mandatory,
> > > is it?
> > 
> > Did you miss the "for this variant"? Maybe I left the comment in not
> 
> I guess so :)
> 
> > exactly the right place, but I don't think the code works correctly for
> > the new variant if clock-names aren't provided:
> > 
> > +	if (st->info->has_dac_clk) {
> > +		struct clk *dac_clk;
> > +		dac_clk = devm_clk_get_enabled(&pdev->dev, "dac_clk");
> > +		if (IS_ERR(dac_clk))
> > +			return dev_err_probe(&pdev->dev, PTR_ERR(dac_clk),
> > +					     "failed to get dac_clk clock\n");
> > +
> > +		/* We only care about the streaming mode rate */
> > +		st->dac_clk_rate = clk_get_rate(dac_clk) / 2;
> > 
> > Isn't this going to cause a probe failure?
> 
> Exactly. And that goes in line with what I wrote about the bindings not describing
> (currently) this. So yes, for the new variant (which has 'has_dac_clk' set to true)
> clock-names is indeed mandatory and probe will fail if it's not given.
> 
> >  
> > > At least for the s_axi_aclk, we first try to get it using clock-names and if
> > > that fails we backup to what we're doing which is passing NULL (which
> > > effectively get's the first clock in the array).
> > > 
> > > The reasoning is that on the generic variant we only need the AXI clk and we
> > > can't now enforce clock-names on it. But to keep things flexible, this was
> > > purposed.
> > 
> > Why not always just get the first clock by index and avoid the
> > complexity?
> 
> And that was also suggested in the previous version but then Jonathan suggested this
> [1]. I agree things now are a bit confusing because we expect clock-names to be
> optional for the generic but mandatory for this new variant and the code is not being
> that explicit about it.
> 
> > 
> > > Another alternative that might have more lines of code (but simpler to
> > > understand the intent) is to have (for example) a callback get_clocks function
> > > that we set depending on the variant. And this also makes me realize that we
> > > could improve the bindings. I mean, for the generic dac variant we do not need
> > > clock-names but for this new variant, clock-names is mandatory and I'm fairly
> > > sure we can express that in the bindings.
> > 
> > Right. You can "edit" required in the if/then/else branch for the new
> > variant.
> 
> Yeah, and IMO that should be set in the bindings (it would help understanding what
> the driver is actually doinfg.
>

ok, thanks, so 

so modified yaml in this way:

  clocks:
    minItems: 1
    maxItems: 2

  clock-names:
    items:
      - const: s_axi_aclk
      - const: dac_clk
    minItems: 1

  '#io-backend-cells':
    const: 0

required:
  - compatible
  - dmas
  - reg
  - clocks

allOf:
  - if:
      properties:
        compatible:
          contains:
            const: adi,axi-ad3552r
    then:
      $ref: /schemas/spi/spi-controller.yaml#
      properties:
        clocks:
          minItems: 2
        clock-names:
          minItems: 2
      required:
        - clock-names
    else:
      properties:
        clocks:
          maxItems: 1
        clock-names:
          maxItems: 1

 
> [1]: https://lore.kernel.org/linux-iio/20241019160817.10c3a2bf@jic23-huawei/
> 
> - Nuno Sá
> 

Regards,
  angelo