[RFC PATCH 1/2] rtc: s3c: Remove unnecessary call to disable already disabled clock

[RFC PATCH 1/2] rtc: s3c: Remove unnecessary call to disable already disabled clock

[Alim Akhtar]

At the end of s3c_rtc_probe(), s3c_rtc_disable_clk() being called with rtc
clock already disabled, which looks extra and unnecessary call.
Lets clean it up.

Signed-off-by: Alim Akhtar <alim.akhtar@samsung.com>
---
 drivers/rtc/rtc-s3c.c |    2 --
 1 file changed, 2 deletions(-)

diff --git a/drivers/rtc/rtc-s3c.c b/drivers/rtc/rtc-s3c.c
index d01ad7e..b083840 100644
--- a/drivers/rtc/rtc-s3c.c
+++ b/drivers/rtc/rtc-s3c.c
@@ -577,8 +577,6 @@ static int s3c_rtc_probe(struct platform_device *pdev)
 
 	s3c_rtc_setfreq(info, 1);
 
-	s3c_rtc_disable_clk(info);
-
 	return 0;
 
  err_nortc:
-- 
1.7.10.4

[RFC PATCH 2/2] rtc: s3c: Add s3c_rtc_{enable/disable}_clk in s3c_rtc_setfreq()

[Alim Akhtar]

As per code flow it is possible that s3c_rtc_setfreq() might get called
with rtc clock disabled and in set_freq we perform h/w registers read/write,
which might results in a kernel crash while probing rtc driver.
Below is one such case:
s3c_rtc_probe()
    clk_prepare_enable(info->rtc_clk) // rtc clock enabled
      s3c_rtc_gettime() // will enable clk if not done, and disable it upon exit
        s3c_rtc_setfreq() //then this will be called with clk disabled

This patch take cares of such issue by adding s3c_rtc_{enable/disable}_clk in
s3c_rtc_setfreq().

Signed-off-by: Alim Akhtar <alim.akhtar@samsung.com>
---
 drivers/rtc/rtc-s3c.c |    2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/rtc/rtc-s3c.c b/drivers/rtc/rtc-s3c.c
index b083840..1168814 100644
--- a/drivers/rtc/rtc-s3c.c
+++ b/drivers/rtc/rtc-s3c.c
@@ -149,12 +149,14 @@ static int s3c_rtc_setfreq(struct s3c_rtc *info, int freq)
 	if (!is_power_of_2(freq))
 		return -EINVAL;
 
+	s3c_rtc_enable_clk(info);
 	spin_lock_irq(&info->pie_lock);
 
 	if (info->data->set_freq)
 		info->data->set_freq(info, freq);
 
 	spin_unlock_irq(&info->pie_lock);
+	s3c_rtc_disable_clk(info);
 
 	return 0;
 }
-- 
1.7.10.4

Re: [RFC PATCH 2/2] rtc: s3c: Add s3c_rtc_{enable/disable}_clk in s3c_rtc_setfreq()

[Krzysztof Kozlowski]

On 07/04/2016 01:03 PM, Alim Akhtar wrote:
> As per code flow it is possible that s3c_rtc_setfreq() might get called
> with rtc clock disabled and in set_freq we perform h/w registers read/write,
> which might results in a kernel crash while probing rtc driver.
> Below is one such case:
> s3c_rtc_probe()
>     clk_prepare_enable(info->rtc_clk) // rtc clock enabled
>       s3c_rtc_gettime() // will enable clk if not done, and disable it upon exit
>         s3c_rtc_setfreq() //then this will be called with clk disabled

The indentation suggests levels of calls (chain) not sequence. This
should be:
s3c_rtc_probe()
  clk_prepare_enable(info->rtc_clk) // rtc clock enabled
  s3c_rtc_gettime() // will enable clk if not done, and disable it upon exit
  s3c_rtc_setfreq() //then this will be called with clk disabled

> 
> This patch take cares of such issue by adding s3c_rtc_{enable/disable}_clk in
> s3c_rtc_setfreq().

What I don't get is that you wrote "it is *possible* that
s3c_rtc_setfreq() *might* get called". From my understanding this will
happen always because src_rtc_gettime() always disables the clocks.

Why it does not happen always?

Best regards,
Krzysztof

Re: [RFC PATCH 2/2] rtc: s3c: Add s3c_rtc_{enable/disable}_clk in s3c_rtc_setfreq()

[Alim Akhtar]

On 07/05/2016 01:46 PM, Krzysztof Kozlowski wrote:
> On 07/04/2016 01:03 PM, Alim Akhtar wrote:
>> As per code flow it is possible that s3c_rtc_setfreq() might get called
>> with rtc clock disabled and in set_freq we perform h/w registers read/write,
>> which might results in a kernel crash while probing rtc driver.
>> Below is one such case:
>> s3c_rtc_probe()
>>      clk_prepare_enable(info->rtc_clk) // rtc clock enabled
>>        s3c_rtc_gettime() // will enable clk if not done, and disable it upon exit
>>          s3c_rtc_setfreq() //then this will be called with clk disabled
>
> The indentation suggests levels of calls (chain) not sequence. This
> should be:
> s3c_rtc_probe()
>    clk_prepare_enable(info->rtc_clk) // rtc clock enabled
>    s3c_rtc_gettime() // will enable clk if not done, and disable it upon exit
>    s3c_rtc_setfreq() //then this will be called with clk disabled
>
>>
>> This patch take cares of such issue by adding s3c_rtc_{enable/disable}_clk in
>> s3c_rtc_setfreq().
>
> What I don't get is that you wrote "it is *possible* that
> s3c_rtc_setfreq() *might* get called". From my understanding this will
> happen always because src_rtc_gettime() always disables the clocks.
>
> Why it does not happen always?
>

Yes, you are right, it is always disabled when reaches s3c_rtc_setfreq().
And I observed a kernel crash while testing on exynos7 platform in 
s3c_rtc_setfreq() because clock was always disabled at this point. And 
this patches fixes this issue.


> Best regards,
> Krzysztof
>

Re: [RFC PATCH 2/2] rtc: s3c: Add s3c_rtc_{enable/disable}_clk in s3c_rtc_setfreq()

[Krzysztof Kozlowski]

On 07/05/2016 10:52 AM, Alim Akhtar wrote:
> 
> 
> On 07/05/2016 01:46 PM, Krzysztof Kozlowski wrote:
>> On 07/04/2016 01:03 PM, Alim Akhtar wrote:
>>> As per code flow it is possible that s3c_rtc_setfreq() might get called
>>> with rtc clock disabled and in set_freq we perform h/w registers
>>> read/write,
>>> which might results in a kernel crash while probing rtc driver.
>>> Below is one such case:
>>> s3c_rtc_probe()
>>>      clk_prepare_enable(info->rtc_clk) // rtc clock enabled
>>>        s3c_rtc_gettime() // will enable clk if not done, and disable
>>> it upon exit
>>>          s3c_rtc_setfreq() //then this will be called with clk disabled
>>
>> The indentation suggests levels of calls (chain) not sequence. This
>> should be:
>> s3c_rtc_probe()
>>    clk_prepare_enable(info->rtc_clk) // rtc clock enabled
>>    s3c_rtc_gettime() // will enable clk if not done, and disable it
>> upon exit
>>    s3c_rtc_setfreq() //then this will be called with clk disabled
>>
>>>
>>> This patch take cares of such issue by adding
>>> s3c_rtc_{enable/disable}_clk in
>>> s3c_rtc_setfreq().
>>
>> What I don't get is that you wrote "it is *possible* that
>> s3c_rtc_setfreq() *might* get called". From my understanding this will
>> happen always because src_rtc_gettime() always disables the clocks.
>>
>> Why it does not happen always?
>>
> 
> Yes, you are right, it is always disabled when reaches s3c_rtc_setfreq().
> And I observed a kernel crash while testing on exynos7 platform in
> s3c_rtc_setfreq() because clock was always disabled at this point. And
> this patches fixes this issue.

Ok, thanks! Could you adjust the commit message mentioning that it
always causes crash on Exynos 7 platform and adding:

Fixes: 24e1455493da ("drivers/rtc/rtc-s3c.c: delete duplicate clock
control")
Cc: <stable@vger.kernel.org>

Reviewed-by: Krzysztof Kozlowski <k.kozlowski@samsung.com>

Best regards,
Krzysztof

Re: [RFC PATCH 2/2] rtc: s3c: Add s3c_rtc_{enable/disable}_clk in s3c_rtc_setfreq()

[pankaj.dubey]

Hi Alim,

On Tuesday 05 July 2016 02:26 PM, Krzysztof Kozlowski wrote:
> On 07/05/2016 10:52 AM, Alim Akhtar wrote:
>>
>>
>> On 07/05/2016 01:46 PM, Krzysztof Kozlowski wrote:
>>> On 07/04/2016 01:03 PM, Alim Akhtar wrote:
>>>> As per code flow it is possible that s3c_rtc_setfreq() might get called
>>>> with rtc clock disabled and in set_freq we perform h/w registers
>>>> read/write,
>>>> which might results in a kernel crash while probing rtc driver.
>>>> Below is one such case:
>>>> s3c_rtc_probe()
>>>>      clk_prepare_enable(info->rtc_clk) // rtc clock enabled
>>>>        s3c_rtc_gettime() // will enable clk if not done, and disable
>>>> it upon exit
>>>>          s3c_rtc_setfreq() //then this will be called with clk disabled
>>>
>>> The indentation suggests levels of calls (chain) not sequence. This
>>> should be:
>>> s3c_rtc_probe()
>>>    clk_prepare_enable(info->rtc_clk) // rtc clock enabled
>>>    s3c_rtc_gettime() // will enable clk if not done, and disable it
>>> upon exit
>>>    s3c_rtc_setfreq() //then this will be called with clk disabled
>>>
>>>>
>>>> This patch take cares of such issue by adding
>>>> s3c_rtc_{enable/disable}_clk in
>>>> s3c_rtc_setfreq().
>>>
>>> What I don't get is that you wrote "it is *possible* that
>>> s3c_rtc_setfreq() *might* get called". From my understanding this will
>>> happen always because src_rtc_gettime() always disables the clocks.
>>>
>>> Why it does not happen always?
>>>
>>
>> Yes, you are right, it is always disabled when reaches s3c_rtc_setfreq().
>> And I observed a kernel crash while testing on exynos7 platform in
>> s3c_rtc_setfreq() because clock was always disabled at this point. And
>> this patches fixes this issue.
> 
> Ok, thanks! Could you adjust the commit message mentioning that it
> always causes crash on Exynos 7 platform and adding:
> 
> Fixes: 24e1455493da ("drivers/rtc/rtc-s3c.c: delete duplicate clock
> control")
> Cc: <stable@vger.kernel.org>
> 
> Reviewed-by: Krzysztof Kozlowski <k.kozlowski@samsung.com>
> 

After addressing Krzysztof's review comments, Please feel free to add

Reviewed-by: Pankaj Dubey <pankaj.dubey@samsung.com>

For testing on Exynos7420
Tested-by: Pankaj Dubey <pankaj.dubey@samsung.com>

Thanks,
Pankaj Dubey

> Best regards,
> Krzysztof
> 
> 

Re: [RFC PATCH 2/2] rtc: s3c: Add s3c_rtc_{enable/disable}_clk in s3c_rtc_setfreq()

[pankaj.dubey]

Hi Alim,

On Tuesday 05 July 2016 02:22 PM, Alim Akhtar wrote:
> 
> 
> On 07/05/2016 01:46 PM, Krzysztof Kozlowski wrote:
>> On 07/04/2016 01:03 PM, Alim Akhtar wrote:
>>> As per code flow it is possible that s3c_rtc_setfreq() might get called
>>> with rtc clock disabled and in set_freq we perform h/w registers
>>> read/write,
>>> which might results in a kernel crash while probing rtc driver.
>>> Below is one such case:
>>> s3c_rtc_probe()
>>>      clk_prepare_enable(info->rtc_clk) // rtc clock enabled
>>>        s3c_rtc_gettime() // will enable clk if not done, and disable
>>> it upon exit
>>>          s3c_rtc_setfreq() //then this will be called with clk disabled
>>
>> The indentation suggests levels of calls (chain) not sequence. This
>> should be:
>> s3c_rtc_probe()
>>    clk_prepare_enable(info->rtc_clk) // rtc clock enabled
>>    s3c_rtc_gettime() // will enable clk if not done, and disable it
>> upon exit
>>    s3c_rtc_setfreq() //then this will be called with clk disabled
>>
>>>
>>> This patch take cares of such issue by adding
>>> s3c_rtc_{enable/disable}_clk in
>>> s3c_rtc_setfreq().
>>
>> What I don't get is that you wrote "it is *possible* that
>> s3c_rtc_setfreq() *might* get called". From my understanding this will
>> happen always because src_rtc_gettime() always disables the clocks.
>>
>> Why it does not happen always?
>>
> 
> Yes, you are right, it is always disabled when reaches s3c_rtc_setfreq().
> And I observed a kernel crash while testing on exynos7 platform in
> s3c_rtc_setfreq() because clock was always disabled at this point. And
> this patches fixes this issue.
> 

After addressing Krzysztof's review comments, Please feel free to add

Reviewed-by: Pankaj Dubey <pankaj.dubey@samsung.com>

For testing on Exynos7420
Tested-by: Pankaj Dubey <pankaj.dubey@samsung.com>

Thanks,
Pankaj Dubey

> 
>> Best regards,
>> Krzysztof
>>
> 

Re: [RFC PATCH 1/2] rtc: s3c: Remove unnecessary call to disable already disabled clock

[Krzysztof Kozlowski]

On 07/04/2016 01:03 PM, Alim Akhtar wrote:
> At the end of s3c_rtc_probe(), s3c_rtc_disable_clk() being called with rtc
> clock already disabled, which looks extra and unnecessary call.
> Lets clean it up.

Does not look right. Till that place, the clocks are enabled. Then
s3c_rtc_setaie() is called which expects that clocks are disabled...
otherwise counters get mixed.

So overall this looks like wrong approach unless I am missing something?

Best regards,
Krzysztof

> 
> Signed-off-by: Alim Akhtar <alim.akhtar@samsung.com>
> ---
>  drivers/rtc/rtc-s3c.c |    2 --
>  1 file changed, 2 deletions(-)
> 
> diff --git a/drivers/rtc/rtc-s3c.c b/drivers/rtc/rtc-s3c.c
> index d01ad7e..b083840 100644
> --- a/drivers/rtc/rtc-s3c.c
> +++ b/drivers/rtc/rtc-s3c.c
> @@ -577,8 +577,6 @@ static int s3c_rtc_probe(struct platform_device *pdev)
>  
>  	s3c_rtc_setfreq(info, 1);
>  
> -	s3c_rtc_disable_clk(info);
> -
>  	return 0;
>  
>   err_nortc:

Re: [RFC PATCH 1/2] rtc: s3c: Remove unnecessary call to disable already disabled clock

[Alim Akhtar]

Hi Krzsztof,

On 07/05/2016 12:48 PM, Krzysztof Kozlowski wrote:
> On 07/04/2016 01:03 PM, Alim Akhtar wrote:
>> At the end of s3c_rtc_probe(), s3c_rtc_disable_clk() being called with rtc
>> clock already disabled, which looks extra and unnecessary call.
>> Lets clean it up.
>
> Does not look right. Till that place, the clocks are enabled. Then
> s3c_rtc_setaie() is called which expects that clocks are disabled...
> otherwise counters get mixed.
>

The clock is always disabled when it reach s3c_rtc_setfreq() in probe(),
because s3c_rtc_gettime() will always disable the clock.

As far as s3c_rtc_setaie() is concern, it enables clock while entering 
and disables it while leaving the function. And in 
s3c_rtc_{enable,disable}_clk() there is check info->clk_disabled flag 
which will make sure clock balancing.

> So overall this looks like wrong approach unless I am missing something?
>
> Best regards,
> Krzysztof
>
>>
>> Signed-off-by: Alim Akhtar <alim.akhtar@samsung.com>
>> ---
>>   drivers/rtc/rtc-s3c.c |    2 --
>>   1 file changed, 2 deletions(-)
>>
>> diff --git a/drivers/rtc/rtc-s3c.c b/drivers/rtc/rtc-s3c.c
>> index d01ad7e..b083840 100644
>> --- a/drivers/rtc/rtc-s3c.c
>> +++ b/drivers/rtc/rtc-s3c.c
>> @@ -577,8 +577,6 @@ static int s3c_rtc_probe(struct platform_device *pdev)
>>
>>   	s3c_rtc_setfreq(info, 1);
>>
>> -	s3c_rtc_disable_clk(info);
>> -
>>   	return 0;
>>
>>    err_nortc:
>
>

Re: [RFC PATCH 1/2] rtc: s3c: Remove unnecessary call to disable already disabled clock

[Krzysztof Kozlowski]

On 07/05/2016 10:46 AM, Alim Akhtar wrote:
> Hi Krzsztof,
> 
> On 07/05/2016 12:48 PM, Krzysztof Kozlowski wrote:
>> On 07/04/2016 01:03 PM, Alim Akhtar wrote:
>>> At the end of s3c_rtc_probe(), s3c_rtc_disable_clk() being called
>>> with rtc
>>> clock already disabled, which looks extra and unnecessary call.
>>> Lets clean it up.
>>
>> Does not look right. Till that place, the clocks are enabled. Then
>> s3c_rtc_setaie() is called which expects that clocks are disabled...
>> otherwise counters get mixed.
>>
> 
> The clock is always disabled when it reach s3c_rtc_setfreq() in probe(),
> because s3c_rtc_gettime() will always disable the clock.
> 
> As far as s3c_rtc_setaie() is concern, it enables clock while entering
> and disables it while leaving the function. And in
> s3c_rtc_{enable,disable}_clk() there is check info->clk_disabled flag
> which will make sure clock balancing.

Ah, you're right. Looks correct although the information that clock is
disabled because of s3c_rtc_gettime() would be useful in the commit message.

Reviewed-by: Krzysztof Kozlowski <k.kozlowski@samsung.com>

Best regards,
Krzysztof