[PATCH 0/3] clk: sunxi: error checking on clock setup

[PATCH 0/3] clk: sunxi: error checking on clock setup

[Andre Przywara]

Since I promised to do this some days ago:

Setting up the clocks properly is quite critical to the system's
operation, but currently our error handling is not very verbose.
This series adds error handling and reporting to the sunxi clocks,
so that any errors are correctly detected and reported. Also previous
actions are rolled back in case something went wrong.
This proves to be helpful in debugging clock tree issues,
especially when adding support for new SoCs.
I tested this on a BananaPi by deliberately misspelling
"clock-output-names". The resulting kernel crash in
sun4i_timer_interrupt is totally misleading, but now there is a line
in the dmesg before saying:
=======
sunxi_divider_clk_setup: could not read clock-output-names for "apb0"
=======

This applies on top of Maxime's sunxi/for-next branch.

Please have a look (and apply, if you like it).

Cheers,
Andre.

Andre Przywara (3):
  clk: sunxi: improve mux_clk error handling and reporting
  clk: sunxi: improve divider_clk error handling and reporting
  clk: sunxi: Improve divs_clk error handling and reporting

 drivers/clk/sunxi/clk-sunxi.c | 68 +++++++++++++++++++++++++++++++++++--------
 1 file changed, 56 insertions(+), 12 deletions(-)

-- 
2.6.4

[PATCH 1/3] clk: sunxi: improve mux_clk error handling and reporting

[Andre Przywara]

We now catch and report a failing ioremap, also a failure in the final
step of the clock registration is now handled and reported.
Also warnings are turned into errors.

Signed-off-by: Andre Przywara <andre.przywara@arm.com>
---
 drivers/clk/sunxi/clk-sunxi.c | 19 +++++++++++++------
 1 file changed, 13 insertions(+), 6 deletions(-)

diff --git a/drivers/clk/sunxi/clk-sunxi.c b/drivers/clk/sunxi/clk-sunxi.c
index 49ce283..361f396 100644
--- a/drivers/clk/sunxi/clk-sunxi.c
+++ b/drivers/clk/sunxi/clk-sunxi.c
@@ -690,11 +690,15 @@ static struct clk * __init sunxi_mux_clk_setup(struct device_node *node,
 	int i;
 
 	reg = of_iomap(node, 0);
+	if (!reg) {
+		pr_err("Could not map registers for mux-clk: %s\n", node->name);
+		return NULL;
+	}
 
 	i = of_clk_parent_fill(node, parents, SUNXI_MAX_PARENTS);
 	if (of_property_read_string(node, "clock-output-names", &clk_name)) {
-		pr_warn("%s: could not read clock-output-names for \"%s\"\n",
-			__func__, clk_name);
+		pr_err("%s: could not read clock-output-names for \"%s\"\n",
+		       __func__, clk_name);
 		goto out_unmap;
 	}
 
@@ -704,14 +708,17 @@ static struct clk * __init sunxi_mux_clk_setup(struct device_node *node,
 			       0, &clk_lock);
 
 	if (IS_ERR(clk)) {
-		pr_warn("%s: failed to register mux clock %s: %ld\n", __func__,
-			clk_name, PTR_ERR(clk));
+		pr_err("%s: failed to register mux clock %s: %ld\n", __func__,
+		       clk_name, PTR_ERR(clk));
 		goto out_unmap;
 	}
 
-	of_clk_add_provider(node, of_clk_src_simple_get, clk);
+	if (!of_clk_add_provider(node, of_clk_src_simple_get, clk))
+		return clk;
+
+	pr_err("%s: failed to add clock provider for %s\n", __func__, clk_name);
 
-	return clk;
+	clk_unregister_divider(clk);
 
 out_unmap:
 	iounmap(reg);
-- 
2.6.4

[PATCH 1/3] clk: sunxi: improve mux_clk error handling and reporting

[Chen-Yu Tsai]

Hi,

On Fri, Feb 12, 2016 at 11:11 PM, Andre Przywara <andre.przywara@arm.com> wrote:
> We now catch and report a failing ioremap, also a failure in the final
> step of the clock registration is now handled and reported.
> Also warnings are turned into errors.
>
> Signed-off-by: Andre Przywara <andre.przywara@arm.com>
> ---
>  drivers/clk/sunxi/clk-sunxi.c | 19 +++++++++++++------
>  1 file changed, 13 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/clk/sunxi/clk-sunxi.c b/drivers/clk/sunxi/clk-sunxi.c
> index 49ce283..361f396 100644
> --- a/drivers/clk/sunxi/clk-sunxi.c
> +++ b/drivers/clk/sunxi/clk-sunxi.c
> @@ -690,11 +690,15 @@ static struct clk * __init sunxi_mux_clk_setup(struct device_node *node,
>         int i;
>
>         reg = of_iomap(node, 0);
> +       if (!reg) {
> +               pr_err("Could not map registers for mux-clk: %s\n", node->name);

of_node_full_name(node->name) is better. node->name is almost always "clk",
which is useless. (Unless someone goes through the dts files to replace all
of them.)

> +               return NULL;
> +       }
>
>         i = of_clk_parent_fill(node, parents, SUNXI_MAX_PARENTS);
>         if (of_property_read_string(node, "clock-output-names", &clk_name)) {
> -               pr_warn("%s: could not read clock-output-names for \"%s\"\n",
> -                       __func__, clk_name);
> +               pr_err("%s: could not read clock-output-names for \"%s\"\n",
> +                      __func__, clk_name);

Same here. clk_name defaults to node->name. If you could, please replace it.

>                 goto out_unmap;
>         }
>
> @@ -704,14 +708,17 @@ static struct clk * __init sunxi_mux_clk_setup(struct device_node *node,
>                                0, &clk_lock);
>
>         if (IS_ERR(clk)) {
> -               pr_warn("%s: failed to register mux clock %s: %ld\n", __func__,
> -                       clk_name, PTR_ERR(clk));
> +               pr_err("%s: failed to register mux clock %s: %ld\n", __func__,
> +                      clk_name, PTR_ERR(clk));
>                 goto out_unmap;
>         }
>
> -       of_clk_add_provider(node, of_clk_src_simple_get, clk);
> +       if (!of_clk_add_provider(node, of_clk_src_simple_get, clk))
> +               return clk;
> +
> +       pr_err("%s: failed to add clock provider for %s\n", __func__, clk_name);
>
> -       return clk;

I'd have all the error path in ifs, and have the normal return path
here for consistency.
But I guess this works as well.

Thanks!
ChenYu

> +       clk_unregister_divider(clk);
>
>  out_unmap:
>         iounmap(reg);
> --
> 2.6.4
>

[PATCH 1/3] clk: sunxi: improve mux_clk error handling and reporting

[Andre Przywara]

Hi Chen-Yu,

On 13/02/16 02:44, Chen-Yu Tsai wrote:
> Hi,
> 
> On Fri, Feb 12, 2016 at 11:11 PM, Andre Przywara <andre.przywara@arm.com> wrote:
>> We now catch and report a failing ioremap, also a failure in the final
>> step of the clock registration is now handled and reported.
>> Also warnings are turned into errors.
>>
>> Signed-off-by: Andre Przywara <andre.przywara@arm.com>
>> ---
>>  drivers/clk/sunxi/clk-sunxi.c | 19 +++++++++++++------
>>  1 file changed, 13 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/clk/sunxi/clk-sunxi.c b/drivers/clk/sunxi/clk-sunxi.c
>> index 49ce283..361f396 100644
>> --- a/drivers/clk/sunxi/clk-sunxi.c
>> +++ b/drivers/clk/sunxi/clk-sunxi.c
>> @@ -690,11 +690,15 @@ static struct clk * __init sunxi_mux_clk_setup(struct device_node *node,
>>         int i;
>>
>>         reg = of_iomap(node, 0);
>> +       if (!reg) {
>> +               pr_err("Could not map registers for mux-clk: %s\n", node->name);
> 
> of_node_full_name(node->name) is better. node->name is almost always "clk",
> which is useless. (Unless someone goes through the dts files to replace all
> of them.)

Good point. I both fixed the code here as well as amended the node names
in the A64 DT.

> 
>> +               return NULL;
>> +       }
>>
>>         i = of_clk_parent_fill(node, parents, SUNXI_MAX_PARENTS);
>>         if (of_property_read_string(node, "clock-output-names", &clk_name)) {
>> -               pr_warn("%s: could not read clock-output-names for \"%s\"\n",
>> -                       __func__, clk_name);
>> +               pr_err("%s: could not read clock-output-names for \"%s\"\n",
>> +                      __func__, clk_name);
> 
> Same here. clk_name defaults to node->name. If you could, please replace it.

Really? Isn't clk_name directly from the DT clock-output-names strings here?

>>                 goto out_unmap;
>>         }
>>
>> @@ -704,14 +708,17 @@ static struct clk * __init sunxi_mux_clk_setup(struct device_node *node,
>>                                0, &clk_lock);
>>
>>         if (IS_ERR(clk)) {
>> -               pr_warn("%s: failed to register mux clock %s: %ld\n", __func__,
>> -                       clk_name, PTR_ERR(clk));
>> +               pr_err("%s: failed to register mux clock %s: %ld\n", __func__,
>> +                      clk_name, PTR_ERR(clk));
>>                 goto out_unmap;
>>         }
>>
>> -       of_clk_add_provider(node, of_clk_src_simple_get, clk);
>> +       if (!of_clk_add_provider(node, of_clk_src_simple_get, clk))
>> +               return clk;
>> +
>> +       pr_err("%s: failed to add clock provider for %s\n", __func__, clk_name);
>>
>> -       return clk;
> 
> I'd have all the error path in ifs, and have the normal return path
> here for consistency.
> But I guess this works as well.

Yeah, I wanted to skip that single goto ;-)
But I changed it now as you said and it indeed looks more reasonable.

I took the freedom of applying the comments to the other patches too.

Will send out a v2 in a minute.

Thanks for looking!
Andre.

> 
> Thanks!
> ChenYu
> 
>> +       clk_unregister_divider(clk);
>>
>>  out_unmap:
>>         iounmap(reg);
>> --
>> 2.6.4
>>
> 

[PATCH 1/3] clk: sunxi: improve mux_clk error handling and reporting

[Chen-Yu Tsai]

On Tue, Feb 16, 2016 at 5:45 PM, Andre Przywara <andre.przywara@arm.com> wrote:
> Hi Chen-Yu,
>
> On 13/02/16 02:44, Chen-Yu Tsai wrote:
>> Hi,
>>
>> On Fri, Feb 12, 2016 at 11:11 PM, Andre Przywara <andre.przywara@arm.com> wrote:
>>> We now catch and report a failing ioremap, also a failure in the final
>>> step of the clock registration is now handled and reported.
>>> Also warnings are turned into errors.
>>>
>>> Signed-off-by: Andre Przywara <andre.przywara@arm.com>
>>> ---
>>>  drivers/clk/sunxi/clk-sunxi.c | 19 +++++++++++++------
>>>  1 file changed, 13 insertions(+), 6 deletions(-)
>>>
>>> diff --git a/drivers/clk/sunxi/clk-sunxi.c b/drivers/clk/sunxi/clk-sunxi.c
>>> index 49ce283..361f396 100644
>>> --- a/drivers/clk/sunxi/clk-sunxi.c
>>> +++ b/drivers/clk/sunxi/clk-sunxi.c
>>> @@ -690,11 +690,15 @@ static struct clk * __init sunxi_mux_clk_setup(struct device_node *node,
>>>         int i;
>>>
>>>         reg = of_iomap(node, 0);
>>> +       if (!reg) {
>>> +               pr_err("Could not map registers for mux-clk: %s\n", node->name);
>>
>> of_node_full_name(node->name) is better. node->name is almost always "clk",
>> which is useless. (Unless someone goes through the dts files to replace all
>> of them.)
>
> Good point. I both fixed the code here as well as amended the node names
> in the A64 DT.
>
>>
>>> +               return NULL;
>>> +       }
>>>
>>>         i = of_clk_parent_fill(node, parents, SUNXI_MAX_PARENTS);
>>>         if (of_property_read_string(node, "clock-output-names", &clk_name)) {
>>> -               pr_warn("%s: could not read clock-output-names for \"%s\"\n",
>>> -                       __func__, clk_name);
>>> +               pr_err("%s: could not read clock-output-names for \"%s\"\n",
>>> +                      __func__, clk_name);
>>
>> Same here. clk_name defaults to node->name. If you could, please replace it.
>
> Really? Isn't clk_name directly from the DT clock-output-names strings here?

It fills it from the DT clock-output-names in the of_property_read_string()
call in the if here. If that failed, obviously something was wrong with
the DT clock-output-names strings. :)

>
>>>                 goto out_unmap;
>>>         }
>>>
>>> @@ -704,14 +708,17 @@ static struct clk * __init sunxi_mux_clk_setup(struct device_node *node,
>>>                                0, &clk_lock);
>>>
>>>         if (IS_ERR(clk)) {
>>> -               pr_warn("%s: failed to register mux clock %s: %ld\n", __func__,
>>> -                       clk_name, PTR_ERR(clk));
>>> +               pr_err("%s: failed to register mux clock %s: %ld\n", __func__,
>>> +                      clk_name, PTR_ERR(clk));
>>>                 goto out_unmap;
>>>         }
>>>
>>> -       of_clk_add_provider(node, of_clk_src_simple_get, clk);
>>> +       if (!of_clk_add_provider(node, of_clk_src_simple_get, clk))
>>> +               return clk;
>>> +
>>> +       pr_err("%s: failed to add clock provider for %s\n", __func__, clk_name);
>>>
>>> -       return clk;
>>
>> I'd have all the error path in ifs, and have the normal return path
>> here for consistency.
>> But I guess this works as well.
>
> Yeah, I wanted to skip that single goto ;-)
> But I changed it now as you said and it indeed looks more reasonable.
>
> I took the freedom of applying the comments to the other patches too.

Thanks!
ChenYu

>
> Will send out a v2 in a minute.
>
> Thanks for looking!
> Andre.
>
>>
>> Thanks!
>> ChenYu
>>
>>> +       clk_unregister_divider(clk);
>>>
>>>  out_unmap:
>>>         iounmap(reg);
>>> --
>>> 2.6.4
>>>
>>

[PATCH 1/3] clk: sunxi: improve mux_clk error handling and reporting

[Andre Przywara]

Hi,

On 16/02/16 09:59, Chen-Yu Tsai wrote:
> On Tue, Feb 16, 2016 at 5:45 PM, Andre Przywara <andre.przywara@arm.com> wrote:
>> Hi Chen-Yu,
>>
>> On 13/02/16 02:44, Chen-Yu Tsai wrote:
>>> Hi,
>>>
>>> On Fri, Feb 12, 2016 at 11:11 PM, Andre Przywara <andre.przywara@arm.com> wrote:
>>>> We now catch and report a failing ioremap, also a failure in the final
>>>> step of the clock registration is now handled and reported.
>>>> Also warnings are turned into errors.
>>>>
>>>> Signed-off-by: Andre Przywara <andre.przywara@arm.com>
>>>> ---
>>>>  drivers/clk/sunxi/clk-sunxi.c | 19 +++++++++++++------
>>>>  1 file changed, 13 insertions(+), 6 deletions(-)
>>>>
>>>> diff --git a/drivers/clk/sunxi/clk-sunxi.c b/drivers/clk/sunxi/clk-sunxi.c
>>>> index 49ce283..361f396 100644
>>>> --- a/drivers/clk/sunxi/clk-sunxi.c
>>>> +++ b/drivers/clk/sunxi/clk-sunxi.c
>>>> @@ -690,11 +690,15 @@ static struct clk * __init sunxi_mux_clk_setup(struct device_node *node,
>>>>         int i;
>>>>
>>>>         reg = of_iomap(node, 0);
>>>> +       if (!reg) {
>>>> +               pr_err("Could not map registers for mux-clk: %s\n", node->name);
>>>
>>> of_node_full_name(node->name) is better. node->name is almost always "clk",
>>> which is useless. (Unless someone goes through the dts files to replace all
>>> of them.)
>>
>> Good point. I both fixed the code here as well as amended the node names
>> in the A64 DT.
>>
>>>
>>>> +               return NULL;
>>>> +       }
>>>>
>>>>         i = of_clk_parent_fill(node, parents, SUNXI_MAX_PARENTS);
>>>>         if (of_property_read_string(node, "clock-output-names", &clk_name)) {
>>>> -               pr_warn("%s: could not read clock-output-names for \"%s\"\n",
>>>> -                       __func__, clk_name);
>>>> +               pr_err("%s: could not read clock-output-names for \"%s\"\n",
>>>> +                      __func__, clk_name);
>>>
>>> Same here. clk_name defaults to node->name. If you could, please replace it.
>>
>> Really? Isn't clk_name directly from the DT clock-output-names strings here?
> 
> It fills it from the DT clock-output-names in the of_property_read_string()
> call in the if here. If that failed, obviously something was wrong with
> the DT clock-output-names strings. :)

Of course! I shouldn't write stuff that early in the morning ...

Cheers,
Andre.

>>
>>>>                 goto out_unmap;
>>>>         }
>>>>
>>>> @@ -704,14 +708,17 @@ static struct clk * __init sunxi_mux_clk_setup(struct device_node *node,
>>>>                                0, &clk_lock);
>>>>
>>>>         if (IS_ERR(clk)) {
>>>> -               pr_warn("%s: failed to register mux clock %s: %ld\n", __func__,
>>>> -                       clk_name, PTR_ERR(clk));
>>>> +               pr_err("%s: failed to register mux clock %s: %ld\n", __func__,
>>>> +                      clk_name, PTR_ERR(clk));
>>>>                 goto out_unmap;
>>>>         }
>>>>
>>>> -       of_clk_add_provider(node, of_clk_src_simple_get, clk);
>>>> +       if (!of_clk_add_provider(node, of_clk_src_simple_get, clk))
>>>> +               return clk;
>>>> +
>>>> +       pr_err("%s: failed to add clock provider for %s\n", __func__, clk_name);
>>>>
>>>> -       return clk;
>>>
>>> I'd have all the error path in ifs, and have the normal return path
>>> here for consistency.
>>> But I guess this works as well.
>>
>> Yeah, I wanted to skip that single goto ;-)
>> But I changed it now as you said and it indeed looks more reasonable.
>>
>> I took the freedom of applying the comments to the other patches too.
> 
> Thanks!
> ChenYu
> 
>>
>> Will send out a v2 in a minute.
>>
>> Thanks for looking!
>> Andre.
>>
>>>
>>> Thanks!
>>> ChenYu
>>>
>>>> +       clk_unregister_divider(clk);
>>>>
>>>>  out_unmap:
>>>>         iounmap(reg);
>>>> --
>>>> 2.6.4
>>>>
>>>
> 

[PATCH 1/3] clk: sunxi: improve mux_clk error handling and reporting

[Maxime Ripard]

On Tue, Feb 16, 2016 at 09:45:57AM +0000, Andre Przywara wrote:
> Hi Chen-Yu,
> 
> On 13/02/16 02:44, Chen-Yu Tsai wrote:
> > Hi,
> > 
> > On Fri, Feb 12, 2016 at 11:11 PM, Andre Przywara <andre.przywara@arm.com> wrote:
> >> We now catch and report a failing ioremap, also a failure in the final
> >> step of the clock registration is now handled and reported.
> >> Also warnings are turned into errors.
> >>
> >> Signed-off-by: Andre Przywara <andre.przywara@arm.com>
> >> ---
> >>  drivers/clk/sunxi/clk-sunxi.c | 19 +++++++++++++------
> >>  1 file changed, 13 insertions(+), 6 deletions(-)
> >>
> >> diff --git a/drivers/clk/sunxi/clk-sunxi.c b/drivers/clk/sunxi/clk-sunxi.c
> >> index 49ce283..361f396 100644
> >> --- a/drivers/clk/sunxi/clk-sunxi.c
> >> +++ b/drivers/clk/sunxi/clk-sunxi.c
> >> @@ -690,11 +690,15 @@ static struct clk * __init sunxi_mux_clk_setup(struct device_node *node,
> >>         int i;
> >>
> >>         reg = of_iomap(node, 0);
> >> +       if (!reg) {
> >> +               pr_err("Could not map registers for mux-clk: %s\n", node->name);
> > 
> > of_node_full_name(node->name) is better. node->name is almost always "clk",
> > which is useless. (Unless someone goes through the dts files to replace all
> > of them.)
> 
> Good point. I both fixed the code here as well as amended the node names
> in the A64 DT.

I'm not sure what there is to amend in the DTSI. The node name is
defined as the class or function of the device, so all clocks should
be called clk@<something>.

This doesn't work for clocks that don't have a meaningful unit address
(like the oscillators), but it should be the exception, rather than
the norm.

> 
> > 
> >> +               return NULL;
> >> +       }
> >>
> >>         i = of_clk_parent_fill(node, parents, SUNXI_MAX_PARENTS);
> >>         if (of_property_read_string(node, "clock-output-names", &clk_name)) {
> >> -               pr_warn("%s: could not read clock-output-names for \"%s\"\n",
> >> -                       __func__, clk_name);
> >> +               pr_err("%s: could not read clock-output-names for \"%s\"\n",
> >> +                      __func__, clk_name);
> > 
> > Same here. clk_name defaults to node->name. If you could, please replace it.
> 
> Really? Isn't clk_name directly from the DT clock-output-names strings here?

Only if there is a clock-output-names property in the node. If not,
the string will not be modified by of_property_read_string, and you'll
default to the initial value of clk_name, in this case node->name or
node->full_name.

Maxime

-- 
Maxime Ripard, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20160216/bda9826e/attachment.sig>

[PATCH 2/3] clk: sunxi: improve divider_clk error handling and reporting

[Andre Przywara]

We now report a failing ioremap, failing output names parsing,
failures in table registration and in the final step.
Also there was a bug where clk_register_divider_table() would return
an ERR_PTR value instead of NULL, which we were checking for.
We now implement proper rollback in case of an error.

Signed-off-by: Andre Przywara <andre.przywara@arm.com>
---
 drivers/clk/sunxi/clk-sunxi.c | 33 ++++++++++++++++++++++++++++++---
 1 file changed, 30 insertions(+), 3 deletions(-)

diff --git a/drivers/clk/sunxi/clk-sunxi.c b/drivers/clk/sunxi/clk-sunxi.c
index 361f396..1ac6c9e0 100644
--- a/drivers/clk/sunxi/clk-sunxi.c
+++ b/drivers/clk/sunxi/clk-sunxi.c
@@ -819,17 +819,44 @@ static void __init sunxi_divider_clk_setup(struct device_node *node,
 	void __iomem *reg;
 
 	reg = of_iomap(node, 0);
+	if (!reg) {
+		pr_err("Could not map registers for mux-clk: %s\n", node->name);
+		return;
+	}
 
 	clk_parent = of_clk_get_parent_name(node, 0);
 
-	of_property_read_string(node, "clock-output-names", &clk_name);
+	if (of_property_read_string(node, "clock-output-names", &clk_name)) {
+		pr_err("%s: could not read clock-output-names for \"%s\"\n",
+		       __func__, clk_name);
+		goto out_unmap;
+	}
 
 	clk = clk_register_divider_table(NULL, clk_name, clk_parent, 0,
 					 reg, data->shift, data->width,
 					 data->pow ? CLK_DIVIDER_POWER_OF_TWO : 0,
 					 data->table, &clk_lock);
-	if (clk)
-		of_clk_add_provider(node, of_clk_src_simple_get, clk);
+	if (IS_ERR(clk)) {
+		pr_err("%s: failed to register divider clock %s: %ld\n",
+		       __func__, clk_name, PTR_ERR(clk));
+		goto out_unmap;
+	}
+
+	if (of_clk_add_provider(node, of_clk_src_simple_get, clk)) {
+		pr_err("%s: failed to add clock provider for %s\n",
+		       __func__, clk_name);
+		goto out_unregister;
+	}
+
+	if (!clk_register_clkdev(clk, clk_name, NULL))
+		return;
+
+	of_clk_del_provider(node);
+out_unregister:
+	clk_unregister_divider(clk);
+
+out_unmap:
+	iounmap(reg);
 }
 
 static void __init sun4i_ahb_clk_setup(struct device_node *node)
-- 
2.6.4

[PATCH 3/3] clk: sunxi: Improve divs_clk error handling and reporting

[Andre Przywara]

We catch errors in the base clock registration, failure to ioremap
and failures in the final of_clk_add_provider() call.
Also we unmap the registers when we need to rollback.

Signed-off-by: Andre Przywara <andre.przywara@arm.com>
---
 drivers/clk/sunxi/clk-sunxi.c | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)

diff --git a/drivers/clk/sunxi/clk-sunxi.c b/drivers/clk/sunxi/clk-sunxi.c
index 1ac6c9e0..48ed81d 100644
--- a/drivers/clk/sunxi/clk-sunxi.c
+++ b/drivers/clk/sunxi/clk-sunxi.c
@@ -987,13 +987,20 @@ static struct clk ** __init sunxi_divs_clk_setup(struct device_node *node,
 
 	/* Set up factor clock that we will be dividing */
 	pclk = sunxi_factors_clk_setup(node, data->factors);
+	if (!pclk)
+		return NULL;
 	parent = __clk_get_name(pclk);
 
 	reg = of_iomap(node, 0);
+	if (!reg) {
+		pr_err("Could not map registers for divs-clk: %s\n",
+		       node->name);
+		return NULL;
+	}
 
 	clk_data = kmalloc(sizeof(struct clk_onecell_data), GFP_KERNEL);
 	if (!clk_data)
-		return NULL;
+		goto out_unmap;
 
 	clks = kcalloc(ndivs, sizeof(*clks), GFP_KERNEL);
 	if (!clks)
@@ -1076,9 +1083,10 @@ static struct clk ** __init sunxi_divs_clk_setup(struct device_node *node,
 	/* Adjust to the real max */
 	clk_data->clk_num = i;
 
-	of_clk_add_provider(node, of_clk_src_onecell_get, clk_data);
+	if (!of_clk_add_provider(node, of_clk_src_onecell_get, clk_data))
+		return clks;
 
-	return clks;
+	pr_err("%s: failed to add clock provider for %s\n", __func__, clk_name);
 
 free_gate:
 	kfree(gate);
@@ -1086,6 +1094,8 @@ free_clks:
 	kfree(clks);
 free_clkdata:
 	kfree(clk_data);
+out_unmap:
+	iounmap(reg);
 	return NULL;
 }
 
-- 
2.6.4