Re: [PATCH RFC 3/8] iio: backend adi-axi-dac: backend features

[Angelo Dureghello <adureghello@baylibre.com>]

Hi,

On 05/09/24 12:49 PM, Nuno Sá wrote:
> On Tue, 2024-09-03 at 20:16 +0100, Jonathan Cameron wrote:
>> On Mon, 2 Sep 2024 18:04:51 +0200
>> Angelo Dureghello <adureghello@baylibre.com> wrote:
>>
>>> On 31/08/24 1:34 PM, Jonathan Cameron wrote:
>>>> On Thu, 29 Aug 2024 14:32:01 +0200
>>>> Angelo Dureghello <adureghello@baylibre.com> wrote:
>>>>   
>>>>> From: Angelo Dureghello <adureghello@baylibre.com>
>>>>>
>>>>> Extend DAC backend with new features required for the AXI driver
>>>>> version for the a3552r DAC.
>>>>>
>>>>> Signed-off-by: Angelo Dureghello <adureghello@baylibre.com>
>>>> Hi Angelo
>>>> Minor comments inline.
>>>>>    
>>>>>    static int axi_dac_enable(struct iio_backend *back)
>>>>> @@ -460,7 +493,13 @@ static int axi_dac_data_source_set(struct
>>>>> iio_backend *back, unsigned int chan,
>>>>>    	case IIO_BACKEND_EXTERNAL:
>>>>>    		return regmap_update_bits(st->regmap,
>>>>>    					
>>>>> AXI_DAC_REG_CHAN_CNTRL_7(chan),
>>>>> -					  AXI_DAC_DATA_SEL,
>>>>> AXI_DAC_DATA_DMA);
>>>>> +					  AXI_DAC_DATA_SEL,
>>>>> +					  AXI_DAC_DATA_DMA);
>>>> Unrelated change.   If you want to change this, separate patch.
>>> Thanks, fixed.
>>>>   
>>>>> +	case IIO_BACKEND_INTERNAL_RAMP_16:
>>>>> +		return regmap_update_bits(st->regmap,
>>>>> +					
>>>>> AXI_DAC_REG_CHAN_CNTRL_7(chan),
>>>>> +					  AXI_DAC_DATA_SEL,
>>>>> +					
>>>>> AXI_DAC_DATA_INTERNAL_RAMP_16);
>>>>>    	default:
>>>>>    		return -EINVAL;
>>>>>    	}
>>>>> @@ -518,9 +557,204 @@ static int axi_dac_reg_access(struct iio_backend
>>>>> *back, unsigned int reg,
>>>>>    	return regmap_write(st->regmap, reg, writeval);
>>>>>    }
>>>>>    
>>>>> +
>>>>> +static int axi_dac_bus_reg_write(struct iio_backend *back,
>>>>> +				 u32 reg, void *val, size_t size)
>>>> Maybe just pass an unsigned int for val?
>>>> So follow what regmap does? You will still need the size, but it
>>>> will just be configuration related rather than affecting the type
>>>> of val.
>>>>   
>>> void * was used since data size in the future may vary depending
>>> on the bus physical interface.
>>>
>> I doubt it will get bigger than u64.  Passing void * is always
>> nasty if we can do something else and this is a register writing
>> operation.  I'm yet to meet an ADC or similar with > 64 bit registers
>> (or even one with 64 bit ones!)
> I think the original thinking was to support thinks like appending crc to the
> register read/write. But even in that case, u32 for val might be enough. Not
> sure. Anyways, as you often say with the backend stuff, this is all in the
> kernel so I guess we can change it to unsigned int and change it in the future
> if we need to.
>
> Since you mentioned regmap, I also want to bring something that was discussed
> before the RFC. Basically we talked about having the backend registering it's
> own regmap_bus. Then we would either:
>
> 1) Have a specific get_regmap_bus() callback for the frontend to initialize a
> regmap on;
> 2) Pass this bus into the core and have a new frontend API like
> devm_iio_backend_regmap_init().
>
> Then, on top of the API already provided by regmap (like _update_bit()), the
> frontend could just use regmap independent of having a backend or not.
>
> The current API is likely more generic but tbh (and David and Angelo are aware
> of it) my preferred approach it to use the regmap_bus stuff. I just don't feel
> that strong about it :)
>
>>> Actually, a reg bus write involves several AXI regmap operations.
>>>>   
>>>>> +{
>>>>> +	struct axi_dac_state *st = iio_backend_get_priv(back);
>>>>> +
>>>>> +	if (!st->bus_type)
>>>>> +		return -EOPNOTSUPP;
>>>>> +
>>>>> +	if (st->bus_type == AXI_DAC_BUS_TYPE_QSPI) {
>>>> As below, I'd use a switch and factor out this block as a separate
>>>> bus specific function.
>>> Ok, changed.
>>>>   
>>>>> +		int ret;
>>>>> +		u32 ival;
>>>>> +
>>>>> +		if (size != 1 && size != 2)
>>>>> +			return -EINVAL;
>>>>> +
>>>>> +		switch (size) {
>>>>> +		case 1:
>>>>> +			ival = FIELD_PREP(AXI_DAC_DATA_WR_8, *(u8
>>>>> *)val);
>>>>> +			break;
>>>>> +		case 2:
>>>>> +			ival =  FIELD_PREP(AXI_DAC_DATA_WR_16, *(u16
>>>>> *)val);
>>>>> +			break;
>>>>> +		default:
>>>>> +			return  -EINVAL;
>>>> Hopefully compiler won't need this and the above. I'd drop the size != 1..
>>>> check in favour of just doing it in this switch.
>>>>   
>>> sure, done.
>>>
>>>
>>>>> +		}
>>>>> +
>>>>> +		ret = regmap_write(st->regmap, AXI_DAC_CNTRL_DATA_WR,
>>>>> ival);
>>>>> +		if (ret)
>>>>> +			return ret;
>>>>> +
>>>>> +		/*
>>>>> +		 * Both REG_CNTRL_2 and AXI_DAC_CNTRL_DATA_WR need to
>>>>> know
>>>>> +		 * the data size. So keeping data size control here
>>>>> only,
>>>>> +		 * since data size is mandatory for to the current
>>>>> transfer.
>>>>> +		 * DDR state handled separately by specific backend
>>>>> calls,
>>>>> +		 * generally all raw register writes are SDR.
>>>>> +		 */
>>>>> +		if (size == 1)
>>>>> +			ret = regmap_set_bits(st->regmap,
>>>>> AXI_DAC_REG_CNTRL_2,
>>>>> +					      AXI_DAC_SYMB_8B);
>>>>> +		else
>>>>> +			ret = regmap_clear_bits(st->regmap,
>>>>> AXI_DAC_REG_CNTRL_2,
>>>>> +						AXI_DAC_SYMB_8B);
>>>>> +		if (ret)
>>>>> +			return ret;
>>>>> +
>>>>> +		ret = regmap_update_bits(st->regmap,
>>>>> AXI_DAC_REG_CUSTOM_CTRL,
>>>>> +					 AXI_DAC_ADDRESS,
>>>>> +					 FIELD_PREP(AXI_DAC_ADDRESS,
>>>>> reg));
>>>>> +		if (ret)
>>>>> +			return ret;
>>>>> +
>>>>> +		ret = regmap_update_bits(st->regmap,
>>>>> AXI_DAC_REG_CUSTOM_CTRL,
>>>>> +					 AXI_DAC_TRANSFER_DATA,
>>>>> +					 AXI_DAC_TRANSFER_DATA);
>>>>> +		if (ret)
>>>>> +			return ret;
>>>>> +
>>>>> +		ret = regmap_read_poll_timeout(st->regmap,
>>>>> +					       AXI_DAC_REG_CUSTOM_CTRL,
>>>>> ival,
>>>>> +					       ival &
>>>>> AXI_DAC_TRANSFER_DATA,
>>>>> +					       10, 100 * KILO);
>>>>> +		if (ret)
>>>>> +			return ret;
>>>>> +
>>>>> +		return regmap_clear_bits(st->regmap,
>>>>> AXI_DAC_REG_CUSTOM_CTRL,
>>>>> +					  AXI_DAC_TRANSFER_DATA);
>>>>> +	}
>>>>> +
>>>>> +	return -EINVAL;
>>>>> +}
>>>>> +
>>>>> +static int axi_dac_bus_reg_read(struct iio_backend *back,
>>>>> +				u32 reg, void *val, size_t size)
>>>> As for write, I'd just use an unsigned int * for val like
>>>> regmap does.
>>> Ok, so initial choice was unsigned int, further thinking of
>>> possible future busses drive the choice to void *.
>>>
>>> Let me know, i can switch to unsigned int in case.
>> I would just go with unsigned int or at a push u64 *
>>
>>>
>>>>   
>>>>> +{
>>>>> +	struct axi_dac_state *st = iio_backend_get_priv(back);
>>>>> +
>>>>> +	if (!st->bus_type)
>>>>> +		return -EOPNOTSUPP;
>>>>> +
>>>>> +	if (st->bus_type == AXI_DAC_BUS_TYPE_QSPI) {
>>>> It got mentioned in binding review but if this isn't QSPI, even
>>>> if similar don't call it that.
>>> It's a bit difficult to find a different name, physically,
>>> it is a QSPI, 4 lanes + clock + cs, and datasheet is naming it Quad SPI.
>>> But looking the data protocol, it's a bit different.
>> is QSPI actually defined anywhere? I assumed it would be like
>> SPI for which everything is so flexible you can build whatever you like.
>>
>>> QSPI has instruction, address and data.
>>> Here we have just ADDR and DATA.
>>>
> I'm not sure the instruction is really relevant for this. From a quick look, it
> feels like something used for accessing external flash memory like spi-nors. So,
> I would not be surprised if things are just like Jonathan said and this is just
> flexible as spi (being that extra instruction field a protocol defined for flash
> memory - where one typically sees this interface)

Ok, so QSPI is the hardware, and the protocol on it may vary for the target
chip/application.

Looks like DDR makes the 33MUPS rate reachable, and not all the controllers
have DDR mode. Also some controllers are supposed to work with a QSPI flash
(so with instructions), and likely this reason driven the need to use a 
custom IP.


Regards,
Angelo


> - Nuno Sá

-- 
  ,,,      Angelo Dureghello
:: :.     BayLibre -runtime team- Developer
:`___:
  `____: