[RFC,v2,1/1] clk: Add notifier support in clk_prepare/clk_unprepare

Add the below two notifier events so drivers which are interested in
knowing the clock status can act accordingly. This is extremely useful
in some of the DVFS (Dynamic Voltage Frequency Scaling) design.

CLK_PREPARED
CLK_UNPREPARED

Signed-off-by: Bill Huang <bilhuang@nvidia.com>
---
 drivers/clk/clk.c   |    3 +++
 include/linux/clk.h |    2 ++
 2 files changed, 5 insertions(+)

On Thu, 2013-03-14 at 17:31 +0800, Bill Huang wrote:
> Add the below two notifier events so drivers which are interested in
> knowing the clock status can act accordingly. This is extremely useful
> in some of the DVFS (Dynamic Voltage Frequency Scaling) design.
> 
> CLK_PREPARED
> CLK_UNPREPARED
> 
> Signed-off-by: Bill Huang <bilhuang@nvidia.com>
> ---
>  drivers/clk/clk.c   |    3 +++
>  include/linux/clk.h |    2 ++
>  2 files changed, 5 insertions(+)
> 
> diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
> index ed87b24..3292cec 100644
> --- a/drivers/clk/clk.c
> +++ b/drivers/clk/clk.c
> @@ -550,6 +550,7 @@ void clk_unprepare(struct clk *clk)
>  {
>  	mutex_lock(&prepare_lock);
>  	__clk_unprepare(clk);
> +	__clk_notify(clk, CLK_UNPREPARED, clk->rate, clk->rate);
>  	mutex_unlock(&prepare_lock);
>  }
>  EXPORT_SYMBOL_GPL(clk_unprepare);
> @@ -598,6 +599,8 @@ int clk_prepare(struct clk *clk)
>  
>  	mutex_lock(&prepare_lock);
>  	ret = __clk_prepare(clk);
> +	if (!ret)
> +		__clk_notify(clk, CLK_PREPARED, clk->rate, clk->rate);
>  	mutex_unlock(&prepare_lock);
>  
>  	return ret;
> diff --git a/include/linux/clk.h b/include/linux/clk.h
> index b3ac22d..16c1d92 100644
> --- a/include/linux/clk.h
> +++ b/include/linux/clk.h
> @@ -43,6 +43,8 @@ struct clk;
>  #define PRE_RATE_CHANGE			BIT(0)
>  #define POST_RATE_CHANGE		BIT(1)
>  #define ABORT_RATE_CHANGE		BIT(2)
> +#define CLK_PREPARED			BIT(3)
> +#define CLK_UNPREPARED			BIT(4)
>  
>  /**
>   * struct clk_notifier - associate a clk with a notifier
Sorry, I'll send v3 since this bread build.

On Thu, Mar 14, 2013 at 02:31:04AM -0700, Bill Huang wrote:
> Add the below two notifier events so drivers which are interested in
> knowing the clock status can act accordingly. This is extremely useful
> in some of the DVFS (Dynamic Voltage Frequency Scaling) design.
> 
> CLK_PREPARED
> CLK_UNPREPARED
> 
> Signed-off-by: Bill Huang <bilhuang@nvidia.com>
> ---
>  drivers/clk/clk.c   |    3 +++
>  include/linux/clk.h |    2 ++
>  2 files changed, 5 insertions(+)
> 
> diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
> index ed87b24..3292cec 100644
> --- a/drivers/clk/clk.c
> +++ b/drivers/clk/clk.c
> @@ -550,6 +550,7 @@ void clk_unprepare(struct clk *clk)
>  {
>  	mutex_lock(&prepare_lock);
>  	__clk_unprepare(clk);
> +	__clk_notify(clk, CLK_UNPREPARED, clk->rate, clk->rate);
>  	mutex_unlock(&prepare_lock);
>  }
>  EXPORT_SYMBOL_GPL(clk_unprepare);
> @@ -598,6 +599,8 @@ int clk_prepare(struct clk *clk)
>  
>  	mutex_lock(&prepare_lock);
>  	ret = __clk_prepare(clk);
> +	if (!ret)
> +		__clk_notify(clk, CLK_PREPARED, clk->rate, clk->rate);

So, on prepare, we notify after we've prepared the clock.  On unprepare,
we notify after the clock has been shut down.  Are you sure that's the
correct ordering?  Would it not be better to view it in a stack-like
fashion, iow:

get
	prepare
		notify-prepare
			enable
			disable
		notify-unprepare
	unprepare
put

?

> diff --git a/include/linux/clk.h b/include/linux/clk.h
> index b3ac22d..16c1d92 100644
> --- a/include/linux/clk.h
> +++ b/include/linux/clk.h
> @@ -43,6 +43,8 @@ struct clk;
>  #define PRE_RATE_CHANGE			BIT(0)
>  #define POST_RATE_CHANGE		BIT(1)
>  #define ABORT_RATE_CHANGE		BIT(2)
> +#define CLK_PREPARED			BIT(3)
> +#define CLK_UNPREPARED			BIT(4)

This implies that we're only going to have a maximum of 32 reason codes
here.  Is that enough?

On 03/15/2013 11:45 AM, Russell King - ARM Linux wrote:
> On Thu, Mar 14, 2013 at 02:31:04AM -0700, Bill Huang wrote:
>> Add the below two notifier events so drivers which are interested in
>> knowing the clock status can act accordingly. This is extremely useful
>> in some of the DVFS (Dynamic Voltage Frequency Scaling) design.
>>
>> CLK_PREPARED
>> CLK_UNPREPARED
>>
>> Signed-off-by: Bill Huang <bilhuang@nvidia.com>
>> ---
>>  drivers/clk/clk.c   |    3 +++
>>  include/linux/clk.h |    2 ++
>>  2 files changed, 5 insertions(+)
>>
>> diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
>> index ed87b24..3292cec 100644
>> --- a/drivers/clk/clk.c
>> +++ b/drivers/clk/clk.c
>> @@ -550,6 +550,7 @@ void clk_unprepare(struct clk *clk)
>>  {
>>  	mutex_lock(&prepare_lock);
>>  	__clk_unprepare(clk);
>> +	__clk_notify(clk, CLK_UNPREPARED, clk->rate, clk->rate);
>>  	mutex_unlock(&prepare_lock);
>>  }
>>  EXPORT_SYMBOL_GPL(clk_unprepare);
>> @@ -598,6 +599,8 @@ int clk_prepare(struct clk *clk)
>>  
>>  	mutex_lock(&prepare_lock);
>>  	ret = __clk_prepare(clk);
>> +	if (!ret)
>> +		__clk_notify(clk, CLK_PREPARED, clk->rate, clk->rate);
> 
> So, on prepare, we notify after we've prepared the clock.  On unprepare,
> we notify after the clock has been shut down.  Are you sure that's the
> correct ordering?  Would it not be better to view it in a stack-like
> fashion, iow:

> get
> 	prepare
> 		notify-prepare
> 			enable
> 			disable
> 		notify-unprepare
> 	unprepare
> put

Yes, these should be stacked/nested better for consistency.

But for DVFS, the voltage needs to be raised before the prepare body is
run, so that if clk_prepare actually enables the clock, the voltage is
already at the safe level required by that clock. Similarly, for
unprepare, you can only lower the voltage after having turned off the
clock, which is guaranteed after the unprepare body. So, I think you
want to move the notifier for prepare in the code above (and rename it
to pre/before_prepare?), rather than the notifier for unprepare.

In order to cover more cases, you might have both
{pre,post}_{un,}prepare notifiers, although I'm not sure when you'd use
the other two options.

>> diff --git a/include/linux/clk.h b/include/linux/clk.h

>> +#define CLK_PREPARED			BIT(3)
>> +#define CLK_UNPREPARED			BIT(4)

On Sat, 2013-03-16 at 03:51 +0800, Stephen Warren wrote:
> On 03/15/2013 11:45 AM, Russell King - ARM Linux wrote:
> > On Thu, Mar 14, 2013 at 02:31:04AM -0700, Bill Huang wrote:
> >> Add the below two notifier events so drivers which are interested in
> >> knowing the clock status can act accordingly. This is extremely useful
> >> in some of the DVFS (Dynamic Voltage Frequency Scaling) design.
> >>
> >> CLK_PREPARED
> >> CLK_UNPREPARED
> >>
> >> Signed-off-by: Bill Huang <bilhuang@nvidia.com>
> >> ---
> >>  drivers/clk/clk.c   |    3 +++
> >>  include/linux/clk.h |    2 ++
> >>  2 files changed, 5 insertions(+)
> >>
> >> diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
> >> index ed87b24..3292cec 100644
> >> --- a/drivers/clk/clk.c
> >> +++ b/drivers/clk/clk.c
> >> @@ -550,6 +550,7 @@ void clk_unprepare(struct clk *clk)
> >>  {
> >>  	mutex_lock(&prepare_lock);
> >>  	__clk_unprepare(clk);
> >> +	__clk_notify(clk, CLK_UNPREPARED, clk->rate, clk->rate);
> >>  	mutex_unlock(&prepare_lock);
> >>  }
> >>  EXPORT_SYMBOL_GPL(clk_unprepare);
> >> @@ -598,6 +599,8 @@ int clk_prepare(struct clk *clk)
> >>  
> >>  	mutex_lock(&prepare_lock);
> >>  	ret = __clk_prepare(clk);
> >> +	if (!ret)
> >> +		__clk_notify(clk, CLK_PREPARED, clk->rate, clk->rate);
> > 
> > So, on prepare, we notify after we've prepared the clock.  On unprepare,
> > we notify after the clock has been shut down.  Are you sure that's the
> > correct ordering?  Would it not be better to view it in a stack-like
> > fashion, iow:
> 
> > get
> > 	prepare
> > 		notify-prepare
> > 			enable
> > 			disable
> > 		notify-unprepare
> > 	unprepare
> > put
> 
> Yes, these should be stacked/nested better for consistency.
> 
> But for DVFS, the voltage needs to be raised before the prepare body is
> run, so that if clk_prepare actually enables the clock, the voltage is
> already at the safe level required by that clock. Similarly, for
> unprepare, you can only lower the voltage after having turned off the
> clock, which is guaranteed after the unprepare body. So, I think you
> want to move the notifier for prepare in the code above (and rename it
> to pre/before_prepare?), rather than the notifier for unprepare.

Oh yes I should raised notify before prepare body is run.
> 
> In order to cover more cases, you might have both
> {pre,post}_{un,}prepare notifiers, although I'm not sure when you'd use
> the other two options.

Right, maybe {pre,post}_{un,}prepare will be useful.
> 
> >> diff --git a/include/linux/clk.h b/include/linux/clk.h
> 
> >> +#define CLK_PREPARED			BIT(3)
> >> +#define CLK_UNPREPARED			BIT(4)
>