Re: [PATCH] spi: dt-bindings: samsung: make dma properties not required

From: Tudor Ambarus
Date: Mon Mar 04 2024 - 13:15:57 EST


Hi, Rob,

On 3/4/24 16:56, Rob Herring wrote:
> On Sat, Mar 02, 2024 at 10:23:16AM -0600, Sam Protsenko wrote:
>> On Sat, Mar 2, 2024 at 3:36 AM Tudor Ambarus <tudor.ambarus@xxxxxxxxxx> wrote:
>>>
>>>
>>>
>>> On 01.03.2024 22:42, Mark Brown wrote:
>>>> On Fri, Mar 01, 2024 at 01:28:35PM -0600, Sam Protsenko wrote:
>>>>> On Fri, Mar 1, 2024 at 5:55 AM Tudor Ambarus <tudor.ambarus@xxxxxxxxxx> wrote:
>>>>
>>>>>> Since the addition of the driver in 2009, the driver selects between DMA
>>>>>> and polling mode depending on the transfer length - DMA mode for
>>>>>> transfers bigger than the FIFO depth, polling mode otherwise. All
>>>>>> versions of the IP support polling mode, make the dma properties not
>>>>>> required.
>>>>
>>>>> AFAIU, the device tree has nothing to do with drivers, it's about
>>>>> hardware description. Does making DMA properties not required here
>>>
>>> correct
>>>
>>>>> mean that there are some HW out there which doesn't integrate DMA in
>>>
>>> no, to me it means that the IP can work without DMA, only in PIO mode,
>>> regardless if DMA is integrated or not. Not required means that the
>>> property is not mandatory, which is what I'm trying to achieve here.
>>>
>>>>> SPI blocks? Even if this change is ok (I'm not sure), the
>>>>> argumentation doesn't look sound to me.
>>>
>>> switching to PIO mode in the driver for sizes smaller than FIFO depths
>>> in the driver guarantees that all existing compatibles support PIO mode.
>>>
>>> Are you saying that if there is a physical line between an IP and DMA
>>> controller, then the DMA properties must always be specified in dt? I
>>> thought they can be marked as optional in this case, and that's what I
>>> did with this patch.
>>>
>>
>> No, I would wait for maintainers to clarify on that bit. Change itself
>> can be ok. But the commit message shouldn't mention the driver,
>> because the driver uses (depends on) device tree, not vice versa. The
>> device tree can be used in other projects as well (like U-Boot and
>> OP-TEE), so it should be designed to be universal and not depend on
>> kernel drivers. The commit message should be based on particular HW
>> layout features and how the patch makes the bindings describe that HW
>> better. It shouldn't rely on driver implementations.
>
> If the controller is DMA capable then it should have dma properties. The

should have as in required/mandatory?

> compatible should be enough to tell if it is a case of 'can only work

yes, I agree

> with DMA'. Otherwise, it is going to be up to a specific user. Even
> within Linux, you may have a serial port that doesn't use DMA for the
> console, but uses it for the tty or serdev.
>
> Of course, if a new device is added without DMA properties and they
> are added later on, then they are going to be optional even though the
> DMA support is always there. I can't fully understand everyone's h/w.
>

The SPI controller that I'm working with has a dedicated channel to the
DMA controller. It can work without DMA too, just by polling registers
or by interrupts.

I can't get the DMA controller to work correctly yet, and since the SPI
controller can work without DMA, I thought that I can mark the DMA
properties as optional, add the SPI node in dt without DMA, and add the
DMA properties later on, after I have the DMA controller working
correctly. Is this approach wrong?

Thanks,
ta