[PATCH] clk: bulk: call of_clk_get() when id is NULL

[PATCH] clk: bulk: call of_clk_get() when id is NULL

[Shawn Guo]

From: Shawn Guo <shawn.guo@linaro.org>

Most of clk API users have their clocks defined in device tree, and
client drivers will have to parse clk ids from DT 'clock-names'
property before using clk_bulk_get().  This is a burden for client
driver code.  And 'clock-names' being an optional DT property makes
it even worse.  The client driver will have no way to provide clock
id.

The patch makes a little improvement on clk_bulk_get() to call
of_clk_get() with index for DT users, if clock id is not available,
so that client drivers working with DT can use clk_bulk_get() to
retrieve clocks more easily.

Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
---
 drivers/clk/clk-bulk.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/clk/clk-bulk.c b/drivers/clk/clk-bulk.c
index c834f5abfc49..65cee595a67e 100644
--- a/drivers/clk/clk-bulk.c
+++ b/drivers/clk/clk-bulk.c
@@ -39,7 +39,10 @@ int __must_check clk_bulk_get(struct device *dev, int num_clks,
 		clks[i].clk = NULL;
 
 	for (i = 0; i < num_clks; i++) {
-		clks[i].clk = clk_get(dev, clks[i].id);
+		if (clks[i].id)
+			clks[i].clk = clk_get(dev, clks[i].id);
+		else if (dev->of_node)
+			clks[i].clk = of_clk_get(dev->of_node, i);
 		if (IS_ERR(clks[i].clk)) {
 			ret = PTR_ERR(clks[i].clk);
 			dev_err(dev, "Failed to get clk '%s': %d\n",
-- 
1.9.1

[PATCH] clk: bulk: call of_clk_get() when id is NULL

[Stephen Boyd]

On 08/09, Shawn Guo wrote:
> From: Shawn Guo <shawn.guo@linaro.org>
> 
> Most of clk API users have their clocks defined in device tree, and
> client drivers will have to parse clk ids from DT 'clock-names'
> property before using clk_bulk_get().  This is a burden for client
> driver code.  And 'clock-names' being an optional DT property makes
> it even worse.  The client driver will have no way to provide clock
> id.
> 
> The patch makes a little improvement on clk_bulk_get() to call
> of_clk_get() with index for DT users, if clock id is not available,
> so that client drivers working with DT can use clk_bulk_get() to
> retrieve clocks more easily.
> 
> Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
> ---
>  drivers/clk/clk-bulk.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/clk/clk-bulk.c b/drivers/clk/clk-bulk.c
> index c834f5abfc49..65cee595a67e 100644
> --- a/drivers/clk/clk-bulk.c
> +++ b/drivers/clk/clk-bulk.c
> @@ -39,7 +39,10 @@ int __must_check clk_bulk_get(struct device *dev, int num_clks,
>  		clks[i].clk = NULL;
>  
>  	for (i = 0; i < num_clks; i++) {
> -		clks[i].clk = clk_get(dev, clks[i].id);
> +		if (clks[i].id)
> +			clks[i].clk = clk_get(dev, clks[i].id);
> +		else if (dev->of_node)
> +			clks[i].clk = of_clk_get(dev->of_node, i);

This seems a little too magical. The omission of an id in an
array of clks would mean that only that one clk is acquired
through of_clk_get(). We could have a mixture of ids and no ids
for some device, and then do very odd things.

How about we add a flag to clk_bulk_data that indicates we want
it to use of_clk_get() instead of clk_get() for all of the clks?
Then the id is ignored for the entire function.

Also, this patch needs to document the new behavior somewhere in
the kernel-doc for this function.

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project

[PATCH] clk: bulk: call of_clk_get() when id is NULL

[Shawn Guo]

Hi Stephen,

Thanks for the review comments.

On Wed, Aug 09, 2017 at 10:33:18AM -0700, Stephen Boyd wrote:
> On 08/09, Shawn Guo wrote:
> > From: Shawn Guo <shawn.guo@linaro.org>
> > 
> > Most of clk API users have their clocks defined in device tree, and
> > client drivers will have to parse clk ids from DT 'clock-names'
> > property before using clk_bulk_get().  This is a burden for client
> > driver code.  And 'clock-names' being an optional DT property makes
> > it even worse.  The client driver will have no way to provide clock
> > id.
> > 
> > The patch makes a little improvement on clk_bulk_get() to call
> > of_clk_get() with index for DT users, if clock id is not available,
> > so that client drivers working with DT can use clk_bulk_get() to
> > retrieve clocks more easily.
> > 
> > Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
> > ---
> >  drivers/clk/clk-bulk.c | 5 ++++-
> >  1 file changed, 4 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/clk/clk-bulk.c b/drivers/clk/clk-bulk.c
> > index c834f5abfc49..65cee595a67e 100644
> > --- a/drivers/clk/clk-bulk.c
> > +++ b/drivers/clk/clk-bulk.c
> > @@ -39,7 +39,10 @@ int __must_check clk_bulk_get(struct device *dev, int num_clks,
> >  		clks[i].clk = NULL;
> >  
> >  	for (i = 0; i < num_clks; i++) {
> > -		clks[i].clk = clk_get(dev, clks[i].id);
> > +		if (clks[i].id)
> > +			clks[i].clk = clk_get(dev, clks[i].id);
> > +		else if (dev->of_node)
> > +			clks[i].clk = of_clk_get(dev->of_node, i);
> 
> This seems a little too magical. The omission of an id in an
> array of clks would mean that only that one clk is acquired
> through of_clk_get(). We could have a mixture of ids and no ids
> for some device, and then do very odd things.

Yes, I agree.

> How about we add a flag to clk_bulk_data that indicates we want
> it to use of_clk_get() instead of clk_get() for all of the clks?
> Then the id is ignored for the entire function.

Good suggestion.

> Also, this patch needs to document the new behavior somewhere in
> the kernel-doc for this function.

Okay, I will take care of it with the new version.

Shawn

[PATCH] clk: bulk: call of_clk_get() when id is NULL

[Shawn Guo]

Hi Stephen,

On Wed, Aug 09, 2017 at 10:33:18AM -0700, Stephen Boyd wrote:
> > diff --git a/drivers/clk/clk-bulk.c b/drivers/clk/clk-bulk.c
> > index c834f5abfc49..65cee595a67e 100644
> > --- a/drivers/clk/clk-bulk.c
> > +++ b/drivers/clk/clk-bulk.c
> > @@ -39,7 +39,10 @@ int __must_check clk_bulk_get(struct device *dev, int num_clks,
> >  		clks[i].clk = NULL;
> >  
> >  	for (i = 0; i < num_clks; i++) {
> > -		clks[i].clk = clk_get(dev, clks[i].id);
> > +		if (clks[i].id)
> > +			clks[i].clk = clk_get(dev, clks[i].id);
> > +		else if (dev->of_node)
> > +			clks[i].clk = of_clk_get(dev->of_node, i);
> 
> This seems a little too magical. The omission of an id in an
> array of clks would mean that only that one clk is acquired
> through of_clk_get(). We could have a mixture of ids and no ids
> for some device, and then do very odd things.
> 
> How about we add a flag to clk_bulk_data that indicates we want
> it to use of_clk_get() instead of clk_get() for all of the clks?
> Then the id is ignored for the entire function.

I have sent v2 to add a flag to clk_bulk_data to address your comments.
But I just found it might not be what you want exactly, because the
current bulk clk APIs defines clk_bulk_data in a different way from what
you are asking here.

It seems that you think clk_bulk_data represents all the clocks that
bulk APIs manage, but actually it only represents one of the multiple
clocks.  Callers need to prepare an array of clk_bulk_data to represent
multiple clocks.  That said, the flag we add to clk_bulk_data only
affects one clock instead of all clocks.

I feel this is something we can improve.  clk_bulk_data can be used to
represent all clocks, something like below.  In this case, bulk clk APIs
do not need to take num_clks as a separate parameter, and callers do not
need to maintain multiple clk_bulk_data instances.  Since we do not have
any in tree users of bulk clk APIs yet, we can make this semantic change
quickly.  Thoughts?

diff --git a/include/linux/clk.h b/include/linux/clk.h
index 31fee2e8d4a2..ddc4812758f6 100644
--- a/include/linux/clk.h
+++ b/include/linux/clk.h
@@ -80,8 +80,10 @@ struct clk_notifier_data {
 /**
  * struct clk_bulk_data - Data used for bulk clk operations.
  *
- * @id: clock consumer ID
- * @clk: struct clk * to store the associated clock
+ * @ids: array of clock consumer IDs which should be provided by caller.
+ * @clks: array of struct clk * to store the associated clocks, which will be
+ *     allocated by clk_bulk_get() and freed by clk_bulk_put().
+ * @num_clks: number of clocks that caller wants bulk API to manage.
  * @flags: bulk clk operation flags
  *
  * The CLK APIs provide a series of clk_bulk_() API calls as
@@ -93,8 +95,9 @@ struct clk_notifier_data {
  *     clk_bulk_get() API will call of_clk_get() to find clock by index.
  */
 struct clk_bulk_data {
-       const char              *id;
-       struct clk              *clk;
+       const char              **ids;
+       struct clk              **clks;
+       unsigned int            num_clks;
        unsigned long           flags;
 };

[PATCH] clk: bulk: call of_clk_get() when id is NULL

[A.s. Dong]

> -----Original Message-----
> From: Shawn Guo [mailto:shawnguo at kernel.org]
> Sent: Sunday, August 13, 2017 10:19 PM
> To: Stephen Boyd
> Cc: Michael Turquette; A.s. Dong; linux-clk at vger.kernel.org; linux-arm-
> kernel at lists.infradead.org; Shawn Guo
> Subject: Re: [PATCH] clk: bulk: call of_clk_get() when id is NULL
> 
> Hi Stephen,
> 
> On Wed, Aug 09, 2017 at 10:33:18AM -0700, Stephen Boyd wrote:
> > > diff --git a/drivers/clk/clk-bulk.c b/drivers/clk/clk-bulk.c index
> > > c834f5abfc49..65cee595a67e 100644
> > > --- a/drivers/clk/clk-bulk.c
> > > +++ b/drivers/clk/clk-bulk.c
> > > @@ -39,7 +39,10 @@ int __must_check clk_bulk_get(struct device *dev,
> int num_clks,
> > >  		clks[i].clk = NULL;
> > >
> > >  	for (i = 0; i < num_clks; i++) {
> > > -		clks[i].clk = clk_get(dev, clks[i].id);
> > > +		if (clks[i].id)
> > > +			clks[i].clk = clk_get(dev, clks[i].id);
> > > +		else if (dev->of_node)
> > > +			clks[i].clk = of_clk_get(dev->of_node, i);
> >
> > This seems a little too magical. The omission of an id in an array of
> > clks would mean that only that one clk is acquired through
> > of_clk_get(). We could have a mixture of ids and no ids for some
> > device, and then do very odd things.
> >
> > How about we add a flag to clk_bulk_data that indicates we want it to
> > use of_clk_get() instead of clk_get() for all of the clks?
> > Then the id is ignored for the entire function.
> 
> I have sent v2 to add a flag to clk_bulk_data to address your comments.
> But I just found it might not be what you want exactly, because the
> current bulk clk APIs defines clk_bulk_data in a different way from what
> you are asking here.
> 
> It seems that you think clk_bulk_data represents all the clocks that bulk
> APIs manage, but actually it only represents one of the multiple clocks.
> Callers need to prepare an array of clk_bulk_data to represent multiple
> clocks.  That said, the flag we add to clk_bulk_data only affects one
> clock instead of all clocks.
> 
> I feel this is something we can improve.  clk_bulk_data can be used to
> represent all clocks, something like below.  In this case, bulk clk APIs
> do not need to take num_clks as a separate parameter, and callers do not
> need to maintain multiple clk_bulk_data instances.  Since we do not have
> any in tree users of bulk clk APIs yet, we can make this semantic change
> quickly.  Thoughts?
> 

Flags may be better used for extending per clk feature support in the future.

IMHO how about implement a of_clk_bulk_get() to handle this issue as we already
have a of_clk_get?
e.g.

int __must_check of_clk_bulk_get(struct device_node *np, int num_clks,
                              struct clk_bulk_data *clks)
{
        for (i = 0; i < num_clks; i++) {
                clks[i].clk = of_clk_get(dev, clks[i].id);
...

}

Then it's less intrusive.

Regards
Dong Aisheng

> diff --git a/include/linux/clk.h b/include/linux/clk.h index
> 31fee2e8d4a2..ddc4812758f6 100644
> --- a/include/linux/clk.h
> +++ b/include/linux/clk.h
> @@ -80,8 +80,10 @@ struct clk_notifier_data {
>  /**
>   * struct clk_bulk_data - Data used for bulk clk operations.
>   *
> - * @id: clock consumer ID
> - * @clk: struct clk * to store the associated clock
> + * @ids: array of clock consumer IDs which should be provided by caller.
> + * @clks: array of struct clk * to store the associated clocks, which
> will be
> + *     allocated by clk_bulk_get() and freed by clk_bulk_put().
> + * @num_clks: number of clocks that caller wants bulk API to manage.
>   * @flags: bulk clk operation flags
>   *
>   * The CLK APIs provide a series of clk_bulk_() API calls as @@ -93,8
> +95,9 @@ struct clk_notifier_data {
>   *     clk_bulk_get() API will call of_clk_get() to find clock by index.
>   */
>  struct clk_bulk_data {
> -       const char              *id;
> -       struct clk              *clk;
> +       const char              **ids;
> +       struct clk              **clks;
> +       unsigned int            num_clks;
>         unsigned long           flags;
>  };

[PATCH] clk: bulk: call of_clk_get() when id is NULL

[Stephen Boyd]

On 08/30, A.s. Dong wrote:
> > -----Original Message-----
> > From: Shawn Guo [mailto:shawnguo at kernel.org]
> > Sent: Sunday, August 13, 2017 10:19 PM
> > To: Stephen Boyd
> > Cc: Michael Turquette; A.s. Dong; linux-clk at vger.kernel.org; linux-arm-
> > kernel at lists.infradead.org; Shawn Guo
> > Subject: Re: [PATCH] clk: bulk: call of_clk_get() when id is NULL
> > 
> > Hi Stephen,
> > 
> > On Wed, Aug 09, 2017 at 10:33:18AM -0700, Stephen Boyd wrote:
> > > > diff --git a/drivers/clk/clk-bulk.c b/drivers/clk/clk-bulk.c index
> > > > c834f5abfc49..65cee595a67e 100644
> > > > --- a/drivers/clk/clk-bulk.c
> > > > +++ b/drivers/clk/clk-bulk.c
> > > > @@ -39,7 +39,10 @@ int __must_check clk_bulk_get(struct device *dev,
> > int num_clks,
> > > >  		clks[i].clk = NULL;
> > > >
> > > >  	for (i = 0; i < num_clks; i++) {
> > > > -		clks[i].clk = clk_get(dev, clks[i].id);
> > > > +		if (clks[i].id)
> > > > +			clks[i].clk = clk_get(dev, clks[i].id);
> > > > +		else if (dev->of_node)
> > > > +			clks[i].clk = of_clk_get(dev->of_node, i);
> > >
> > > This seems a little too magical. The omission of an id in an array of
> > > clks would mean that only that one clk is acquired through
> > > of_clk_get(). We could have a mixture of ids and no ids for some
> > > device, and then do very odd things.
> > >
> > > How about we add a flag to clk_bulk_data that indicates we want it to
> > > use of_clk_get() instead of clk_get() for all of the clks?
> > > Then the id is ignored for the entire function.
> > 
> > I have sent v2 to add a flag to clk_bulk_data to address your comments.
> > But I just found it might not be what you want exactly, because the
> > current bulk clk APIs defines clk_bulk_data in a different way from what
> > you are asking here.
> > 
> > It seems that you think clk_bulk_data represents all the clocks that bulk
> > APIs manage, but actually it only represents one of the multiple clocks.
> > Callers need to prepare an array of clk_bulk_data to represent multiple
> > clocks.  That said, the flag we add to clk_bulk_data only affects one
> > clock instead of all clocks.
> > 
> > I feel this is something we can improve.  clk_bulk_data can be used to
> > represent all clocks, something like below.  In this case, bulk clk APIs
> > do not need to take num_clks as a separate parameter, and callers do not
> > need to maintain multiple clk_bulk_data instances.  Since we do not have
> > any in tree users of bulk clk APIs yet, we can make this semantic change
> > quickly.  Thoughts?
> > 
> 
> Flags may be better used for extending per clk feature support in the future.
> 
> IMHO how about implement a of_clk_bulk_get() to handle this issue as we already
> have a of_clk_get?
> e.g.
> 
> int __must_check of_clk_bulk_get(struct device_node *np, int num_clks,
>                               struct clk_bulk_data *clks)
> {
>         for (i = 0; i < num_clks; i++) {
>                 clks[i].clk = of_clk_get(dev, clks[i].id);
> ...
> 
> }
> 
> Then it's less intrusive.
> 

This looks ok too. Go ahead and send this version instead.

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project

[PATCH] clk: bulk: call of_clk_get() when id is NULL

[A.s. Dong]

Hi Shawn,

> -----Original Message-----
> From: Stephen Boyd [mailto:sboyd at codeaurora.org]
> Sent: Thursday, August 31, 2017 1:27 PM
> To: A.s. Dong
> Cc: Shawn Guo; Michael Turquette; linux-clk at vger.kernel.org; linux-arm-
> kernel at lists.infradead.org; Shawn Guo
> Subject: Re: [PATCH] clk: bulk: call of_clk_get() when id is NULL
> 
> On 08/30, A.s. Dong wrote:

[...]

> > Flags may be better used for extending per clk feature support in the
> future.
> >
> > IMHO how about implement a of_clk_bulk_get() to handle this issue as
> > we already have a of_clk_get?
> > e.g.
> >
> > int __must_check of_clk_bulk_get(struct device_node *np, int num_clks,
> >                               struct clk_bulk_data *clks) {
> >         for (i = 0; i < num_clks; i++) {
> >                 clks[i].clk = of_clk_get(dev, clks[i].id); ...
> >
> > }
> >
> > Then it's less intrusive.
> >
> 
> This looks ok too. Go ahead and send this version instead.
> 

I did not see you send a update version.
If you will not do it, would you mind me to help send it with a Reported-by: tag from you?

Regards
Dong Aisheng

> --
> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
> a Linux Foundation Collaborative Project

[PATCH] clk: bulk: call of_clk_get() when id is NULL

[Shawn Guo]

On Mon, Sep 11, 2017 at 04:51:28AM +0000, A.s. Dong wrote:
> Hi Shawn,
> 
> > -----Original Message-----
> > From: Stephen Boyd [mailto:sboyd at codeaurora.org]
> > Sent: Thursday, August 31, 2017 1:27 PM
> > To: A.s. Dong
> > Cc: Shawn Guo; Michael Turquette; linux-clk at vger.kernel.org; linux-arm-
> > kernel at lists.infradead.org; Shawn Guo
> > Subject: Re: [PATCH] clk: bulk: call of_clk_get() when id is NULL
> > 
> > On 08/30, A.s. Dong wrote:
> 
> [...]
> 
> > > Flags may be better used for extending per clk feature support in the
> > future.
> > >
> > > IMHO how about implement a of_clk_bulk_get() to handle this issue as
> > > we already have a of_clk_get?
> > > e.g.
> > >
> > > int __must_check of_clk_bulk_get(struct device_node *np, int num_clks,
> > >                               struct clk_bulk_data *clks) {
> > >         for (i = 0; i < num_clks; i++) {
> > >                 clks[i].clk = of_clk_get(dev, clks[i].id); ...
> > >
> > > }
> > >
> > > Then it's less intrusive.
> > >
> > 
> > This looks ok too. Go ahead and send this version instead.
> > 
> 
> I did not see you send a update version.
> If you will not do it, would you mind me to help send it with a Reported-by: tag from you?

Sure, go ahead.

Shawn

[PATCH] clk: bulk: call of_clk_get() when id is NULL

[Shawn Guo]

On Wed, Aug 30, 2017 at 03:31:38PM +0000, A.s. Dong wrote:
> IMHO how about implement a of_clk_bulk_get() to handle this issue as we already
> have a of_clk_get?
> e.g.
> 
> int __must_check of_clk_bulk_get(struct device_node *np, int num_clks,
>                               struct clk_bulk_data *clks)
> {
>         for (i = 0; i < num_clks; i++) {
>                 clks[i].clk = of_clk_get(dev, clks[i].id);

The of_clk_get() takes clock index rather than name/id as the second
parameter.

Shawn

> ...
> 
> }

[PATCH] clk: bulk: call of_clk_get() when id is NULL

[A.s. Dong]

> -----Original Message-----
> From: Shawn Guo [mailto:shawnguo at kernel.org]
> Sent: Thursday, August 31, 2017 3:01 PM
> To: A.s. Dong
> Cc: Stephen Boyd; Michael Turquette; linux-clk at vger.kernel.org; linux-arm-
> kernel at lists.infradead.org; Shawn Guo
> Subject: Re: [PATCH] clk: bulk: call of_clk_get() when id is NULL
> 
> On Wed, Aug 30, 2017 at 03:31:38PM +0000, A.s. Dong wrote:
> > IMHO how about implement a of_clk_bulk_get() to handle this issue as
> > we already have a of_clk_get?
> > e.g.
> >
> > int __must_check of_clk_bulk_get(struct device_node *np, int num_clks,
> >                               struct clk_bulk_data *clks) {
> >         for (i = 0; i < num_clks; i++) {
> >                 clks[i].clk = of_clk_get(dev, clks[i].id);
> 
> The of_clk_get() takes clock index rather than name/id as the second
> parameter.
> 

You're right.
It's for only an idea demonstration, so sent too quick! :-)

Regards
Dong Aisheng

> Shawn
> 
> > ...
> >
> > }