Re: [PATCH] clk: fix compile for OF && !COMMON_CLK

Re: [PATCH] clk: fix compile for OF && !COMMON_CLK

[Rajendra Nayak]

Rob, Mike,

On Tuesday 17 July 2012 07:38 AM, Rob Herring wrote:
> On 07/16/2012 07:12 PM, Mike Turquette wrote:
>> On 20120716-16:46, Rob Herring wrote:
>>> From: Rob Herring<rob.herring@calxeda.com>
>>>
>>> With commit 766e6a4ec602d0c107 (clk: add DT clock binding support),
>>> compiling with OF&&  !COMMON_CLK is broken.
>>>
>>
>> Hi Rob,
>>
>> Thanks for sending this quickly.
>>
>> <snip>
>>> @@ -313,19 +314,19 @@ int clk_add_alias(const char *alias, const char *alias_dev_name, char *id,
>>>   struct device_node;
>>>   struct of_phandle_args;
>>>
>>> -#ifdef CONFIG_OF
>>> +#if defined(CONFIG_OF)&&  defined(CONFIG_COMMON_CLK)
>>>   struct clk *of_clk_get(struct device_node *np, int index);
>>>   struct clk *of_clk_get_by_name(struct device_node *np, const char *name);
>>>   struct clk *of_clk_get_from_provider(struct of_phandle_args *clkspec);
>>>   #else
>>>   static inline struct clk *of_clk_get(struct device_node *np, int index)
>>>   {
>>> -	return NULL;
>>> +	return ERR_PTR(-EINVAL);

So how is this expected to work on platforms (like OMAP) which have 
CONFIG_OF enabled but not CONFIG_COMMON_CLK?

Archit has been seeing issues with failed clk_get's in the omap dss
driver on linux-next.
The clk_gets pass a valid dev pointer and an alias/con-id.

With the $Subject patch, the of_clk_get_by_name() for our builds always
returns a ERR_PTR(-EINVAL).

Even if we do get the right of_clk_get_by_name() built in,
there is another issue on OMAP where in we have CONFIG_OF
enabled/selected by default for all OMAP2+ builds, even when we
*do not* pass a dt blob to the kernel.

So would the below code fail in such cases because it expects a
valid of_node to be populated for a device (which also has clock
information in it)? if CONFIG_OF is set.

struct clk *clk_get(struct device *dev, const char *con_id)
{
         const char *dev_id = dev ? dev_name(dev) : NULL;
         struct clk *clk;

         if (dev) {
---->
	Any reason why this isn't
	if (dev->of_node) {
---->
                 clk = of_clk_get_by_name(dev->of_node, con_id);
                 if (clk && __clk_get(clk))
                         return clk;
         }

         return clk_get_sys(dev_id, con_id);
}

regards,
Rajendra

Re: [PATCH] clk: fix compile for OF && !COMMON_CLK

[Rajendra Nayak]

On Tuesday 17 July 2012 06:49 PM, Rajendra Nayak wrote:
> struct clk *clk_get(struct device *dev, const char *con_id)
> {
>          const char *dev_id = dev ? dev_name(dev) : NULL;
>          struct clk *clk;
>
>          if (dev) {
> ---->
>      Any reason why this isn't
>      if (dev->of_node) {
> ---->

Or rather,
	if (dev && dev->of_node) {
as Archit just pointed out to me.

Re: [PATCH] clk: fix compile for OF && !COMMON_CLK

[Rob Herring]

On 07/17/2012 08:19 AM, Rajendra Nayak wrote:
> Rob, Mike,
> 
> On Tuesday 17 July 2012 07:38 AM, Rob Herring wrote:
>> On 07/16/2012 07:12 PM, Mike Turquette wrote:
>>> On 20120716-16:46, Rob Herring wrote:
>>>> From: Rob Herring<rob.herring@calxeda.com>
>>>>
>>>> With commit 766e6a4ec602d0c107 (clk: add DT clock binding support),
>>>> compiling with OF&&  !COMMON_CLK is broken.
>>>>
>>>
>>> Hi Rob,
>>>
>>> Thanks for sending this quickly.
>>>
>>> <snip>
>>>> @@ -313,19 +314,19 @@ int clk_add_alias(const char *alias, const
>>>> char *alias_dev_name, char *id,
>>>>   struct device_node;
>>>>   struct of_phandle_args;
>>>>
>>>> -#ifdef CONFIG_OF
>>>> +#if defined(CONFIG_OF)&&  defined(CONFIG_COMMON_CLK)
>>>>   struct clk *of_clk_get(struct device_node *np, int index);
>>>>   struct clk *of_clk_get_by_name(struct device_node *np, const char
>>>> *name);
>>>>   struct clk *of_clk_get_from_provider(struct of_phandle_args
>>>> *clkspec);
>>>>   #else
>>>>   static inline struct clk *of_clk_get(struct device_node *np, int
>>>> index)
>>>>   {
>>>> -    return NULL;
>>>> +    return ERR_PTR(-EINVAL);
> 
> So how is this expected to work on platforms (like OMAP) which have
> CONFIG_OF enabled but not CONFIG_COMMON_CLK?
> 

As I mentioned in my other reply, this really belongs with Shawn's patch
that changes the return value checking from NULL to err values.

> Archit has been seeing issues with failed clk_get's in the omap dss
> driver on linux-next.
> The clk_gets pass a valid dev pointer and an alias/con-id.
> 
> With the $Subject patch, the of_clk_get_by_name() for our builds always
> returns a ERR_PTR(-EINVAL).
> 
> Even if we do get the right of_clk_get_by_name() built in,
> there is another issue on OMAP where in we have CONFIG_OF
> enabled/selected by default for all OMAP2+ builds, even when we
> *do not* pass a dt blob to the kernel.
> 
> So would the below code fail in such cases because it expects a
> valid of_node to be populated for a device (which also has clock
> information in it)? if CONFIG_OF is set.
> 
> struct clk *clk_get(struct device *dev, const char *con_id)
> {
>         const char *dev_id = dev ? dev_name(dev) : NULL;
>         struct clk *clk;
> 
>         if (dev) {
> ---->
>     Any reason why this isn't
>     if (dev->of_node) {
> ---->

dev may be null, we could do (dev && dev->of_node) though. Either way
the error handling here needs to do the right thing.

>                 clk = of_clk_get_by_name(dev->of_node, con_id);
>                 if (clk && __clk_get(clk))
>                         return clk;
>         }
> 
>         return clk_get_sys(dev_id, con_id);
> }
> 
> regards,
> Rajendra
> 
> 
> 
> 
> 

Re: [PATCH] clk: fix compile for OF && !COMMON_CLK

[Rajendra Nayak]

On Tuesday 17 July 2012 07:16 PM, Rob Herring wrote:
>> So how is this expected to work on platforms (like OMAP) which have
>> >  CONFIG_OF enabled but not CONFIG_COMMON_CLK?
>> >
> As I mentioned in my other reply, this really belongs with Shawn's patch
> that changes the return value checking from NULL to err values.
>
ah, I seemed to have missed that. Just had a look at Shawn's patch and
that should work for OMAP too. thanks.