spi: atmel: improve internal vs gpio chip-select choice

Commit Message

Cyrille Pitchen Jan. 11, 2016, 4:33 p.m. UTC

Hi Mans,

Le 11/01/2016 16:26, Nicolas Ferre a écrit :
> Le 08/01/2016 01:11, Mans Rullgard a écrit :
[...]
>> @@ -1569,13 +1576,6 @@ static int atmel_spi_probe(struct platform_device *pdev)
>>  
>>  	atmel_get_caps(as);
>>  
>> -	as->use_cs_gpios = true;
>> -	if (atmel_spi_is_v2(as) &&
> 
> I don't see this atmel_spi_is_v2() test in the resulting code anymore:
> did you make sure that the v1 of the peripheral is protected?
> 
>> -	    !of_get_property(pdev->dev.of_node, "cs-gpios", NULL)) {
>> -		as->use_cs_gpios = false;
>> -		master->num_chipselect = 4;
> 
> This line is pretty important: you mustn't remove it blindly!
> 
> 

Actually, few lines above master->num_chipselect is initialized with:
	master->dev.of_node = pdev->dev.of_node;
[...]
	master->num_chipselect = master->dev.of_node ? 0 : 4;


So 0 in the case of DT support, 4 otherwise.


Now looking at the source code of spi_register_master(), this function:

1 - calls of_spi_register_master(), which initializes master->num_chipselect:
	nb = of_gpio_name_count(np, "cs-gpios");
	master->num_chipselect = max_t(int, nb, master->num_chipselect);

	if (nb == 0 || nb == -ENOENT)
		return 0;

2 - check master->num_chipselect with:
	if (master->num_chipselect == 0)
		return -EINVAL;

It means that in the case of DT support, master->num_chipselect was initialized
according to the value of the "cs-gpios" DT property. Hence this property used
to be mandatory but now we want it to be optional for hardware version 2 and
later.

This is why we added a check "hw version >= 2 and cs-gpios missing". In such
a case the master->num_chipselect was initialized to 4 so spi_register_master()
does not fail with -EINVAL.

I have applied the following fix after your patch and now the driver works
again on sama5d2:


>> -	}
>> -
>>  	as->use_dma = false;
>>  	as->use_pdc = false;
>>  	if (as->caps.has_dma_support) {
>>
> 
> So, sorry but it's a NACK for me.
> 
> Bye,
> 

Best regards,

Cyrille

Comments

Måns Rullgård Jan. 11, 2016, 4:37 p.m. UTC | #1

Cyrille Pitchen <[e-mail]> writes:

> Hi Mans,
>
> Le 11/01/2016 16:26, Nicolas Ferre a écrit :
>> Le 08/01/2016 01:11, Mans Rullgard a écrit :
> [...]
>>> @@ -1569,13 +1576,6 @@ static int atmel_spi_probe(struct platform_device *pdev)
>>>  
>>>  	atmel_get_caps(as);
>>>  
>>> -	as->use_cs_gpios = true;
>>> -	if (atmel_spi_is_v2(as) &&
>> 
>> I don't see this atmel_spi_is_v2() test in the resulting code anymore:
>> did you make sure that the v1 of the peripheral is protected?
>> 
>>> -	    !of_get_property(pdev->dev.of_node, "cs-gpios", NULL)) {
>>> -		as->use_cs_gpios = false;
>>> -		master->num_chipselect = 4;
>> 
>> This line is pretty important: you mustn't remove it blindly!
>> 
>> 
>
> Actually, few lines above master->num_chipselect is initialized with:
> 	master->dev.of_node = pdev->dev.of_node;
> [...]
> 	master->num_chipselect = master->dev.of_node ? 0 : 4;
>
> So 0 in the case of DT support, 4 otherwise.
>
> Now looking at the source code of spi_register_master(), this function:
>
> 1 - calls of_spi_register_master(), which initializes master->num_chipselect:
> 	nb = of_gpio_name_count(np, "cs-gpios");
> 	master->num_chipselect = max_t(int, nb, master->num_chipselect);
>
> 	if (nb == 0 || nb == -ENOENT)
> 		return 0;
>
> 2 - check master->num_chipselect with:
> 	if (master->num_chipselect == 0)
> 		return -EINVAL;
>
> It means that in the case of DT support, master->num_chipselect was
> initialized according to the value of the "cs-gpios" DT
> property. Hence this property used to be mandatory but now we want it
> to be optional for hardware version 2 and later.
>
> This is why we added a check "hw version >= 2 and cs-gpios
> missing". In such a case the master->num_chipselect was initialized to
> 4 so spi_register_master() does not fail with -EINVAL.
>
> I have applied the following fix after your patch and now the driver works
> again on sama5d2:
>
> --- a/drivers/spi/spi-atmel.c
> +++ b/drivers/spi/spi-atmel.c
> @@ -1576,6 +1576,11 @@ static int atmel_spi_probe(struct platform_device *pdev)
>
>         atmel_get_caps(as);
>
> +       if (atmel_spi_is_v2(as) &&
> +           master->dev.of_node &&
> +           !of_get_property(master->dev.of_node, "cs-gpios", NULL))
> +               master->num_chipselect = 4;
> +
>         as->use_dma = false;
>         as->use_pdc = false;
>         if (as->caps.has_dma_support) {

That's pretty much what I did (left the original alone) in v2 of the
patch.

Thanks for testing.

Patch

--- a/drivers/spi/spi-atmel.c
+++ b/drivers/spi/spi-atmel.c
@@ -1576,6 +1576,11 @@  static int atmel_spi_probe(struct platform_device *pdev)
 
        atmel_get_caps(as);
 
+       if (atmel_spi_is_v2(as) &&
+           master->dev.of_node &&
+           !of_get_property(master->dev.of_node, "cs-gpios", NULL))
+               master->num_chipselect = 4;
+
        as->use_dma = false;
        as->use_pdc = false;
        if (as->caps.has_dma_support) {