[PATCH 10/11] ARM: vexpress: Add config bus components and clocks to DTs

[robherring2@gmail.com (Rob Herring)]

On 09/04/2012 08:05 AM, Pawel Moll wrote:
> On Tue, 2012-09-04 at 13:58 +0100, Rob Herring wrote:
>>> -			sysctl at 020000 {
>>> +			v2m_sysctl: sysctl at 020000 {
>>>  				compatible = "arm,sp810", "arm,primecell";
>>>  				reg = <0x020000 0x1000>;
>>> +				clocks = <&v2m_refclk32khz>, <&v2m_refclk1mhz>, <&v2m_osc_clk0>;
>>> +				clock-names = "refclk", "timclk", "apb_pclk";
>>
>> See Documentation/devicetree/bindings/arm/primecell.txt
>>
>> apb_pclk should be first in the list.
> 
> Hm. Why, if you don't mind me asking? The amba_get_enable_pclk()
> explicitly asks for "apb_pclk" id:

So apb_pclk is always in the same position in the list and I had to
define something. Clock names are supposed to be optional, so we have to
be able to identify which clock is the bus clock without the name.

> struct clk *pclk = clk_get(&pcdev->dev, "apb_pclk");
> 
> Now, let's have a look at MMCI:
> 
>                         mmci at 050000 {
>                                 compatible = "arm,pl180", "arm,primecell";
>                                 reg = <0x050000 0x1000>;
>                                 interrupts = <9 10>;
> +                               cd-gpios = <&v2m_sysreg 0 0>;
> +                               wp-gpios = <&v2m_sysreg 1 0>;
> +                               max-frequency = <12000000>;
> +                               vmmc-supply = <&v2m_fixed_3v3>;
> +                               clocks = <&v2m_clk_24mhz>, <&v2m_osc_clk0>;
> +                               clock-names = "mclk", "apb_pclk";
>                         };
> 
> and mmci_probe() does:
> 
>         host->clk = clk_get(&dev->dev, NULL);
>         if (IS_ERR(host->clk)) {
>                 ret = PTR_ERR(host->clk);
>                 host->clk = NULL;
>                 goto host_free;
>         }       
> 
> Now, if I put "apb_pclk" first, before "mclk", the driver will get the
> wrong clock.

You're getting lucky with how clk_get is implemented. The driver should
be more specific with which clock it wants if there is more than 1.
Perhaps we need a clk_get_by_index() function?

Rob

> 
> Pawe?
> 
>