[PATCH] [v2] media: venus: core: Fix runtime PM imbalance in venus_probe

[PATCH] [v2] media: venus: core: Fix runtime PM imbalance in venus_probe

[Dinghao Liu]

pm_runtime_get_sync() increments the runtime PM usage counter even
when it returns an error code. Thus a pairing decrement is needed on
the error handling path to keep the counter balanced. For other error
paths after this call, things are the same.

Signed-off-by: Dinghao Liu <dinghao.liu@zju.edu.cn>
---

Changelog:

v2: - Add pm_runtime_get_noresume() on failure of
      pm_runtime_put_sync() to balance PM counter instead of
      releasing everything here.
---
 drivers/media/platform/qcom/venus/core.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/media/platform/qcom/venus/core.c b/drivers/media/platform/qcom/venus/core.c
index 203c6538044f..b0b932bf8c02 100644
--- a/drivers/media/platform/qcom/venus/core.c
+++ b/drivers/media/platform/qcom/venus/core.c
@@ -287,8 +287,10 @@ static int venus_probe(struct platform_device *pdev)
 		goto err_core_deinit;
 
 	ret = pm_runtime_put_sync(dev);
-	if (ret)
+	if (ret) {
+		pm_runtime_get_noresume(dev);
 		goto err_dev_unregister;
+	}
 
 	return 0;
 
@@ -299,6 +301,7 @@ static int venus_probe(struct platform_device *pdev)
 err_venus_shutdown:
 	venus_shutdown(core);
 err_runtime_disable:
+	pm_runtime_put_noidle(dev);
 	pm_runtime_set_suspended(dev);
 	pm_runtime_disable(dev);
 	hfi_destroy(core);
-- 
2.17.1

Re: [PATCH] [v2] media: venus: core: Fix runtime PM imbalance in venus_probe

[Stanimir Varbanov]

Hi Dinghao,

On 6/24/20 9:30 AM, Dinghao Liu wrote:
> pm_runtime_get_sync() increments the runtime PM usage counter even
> when it returns an error code. Thus a pairing decrement is needed on
> the error handling path to keep the counter balanced. For other error
> paths after this call, things are the same.
> 
> Signed-off-by: Dinghao Liu <dinghao.liu@zju.edu.cn>
> ---
> 
> Changelog:
> 
> v2: - Add pm_runtime_get_noresume() on failure of
>       pm_runtime_put_sync() to balance PM counter instead of
>       releasing everything here.

You are adding pm_runtime_get_noresume in pm_runtime_put_sync error path
but the patch description is referring to pm_runtime_get_sync. I'm confused.

> ---
>  drivers/media/platform/qcom/venus/core.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/media/platform/qcom/venus/core.c b/drivers/media/platform/qcom/venus/core.c
> index 203c6538044f..b0b932bf8c02 100644
> --- a/drivers/media/platform/qcom/venus/core.c
> +++ b/drivers/media/platform/qcom/venus/core.c
> @@ -287,8 +287,10 @@ static int venus_probe(struct platform_device *pdev)
>  		goto err_core_deinit;
>  
>  	ret = pm_runtime_put_sync(dev);
> -	if (ret)
> +	if (ret) {
> +		pm_runtime_get_noresume(dev);
>  		goto err_dev_unregister;
> +	}
>  
>  	return 0;
>  
> @@ -299,6 +301,7 @@ static int venus_probe(struct platform_device *pdev)
>  err_venus_shutdown:
>  	venus_shutdown(core);
>  err_runtime_disable:
> +	pm_runtime_put_noidle(dev);
>  	pm_runtime_set_suspended(dev);
>  	pm_runtime_disable(dev);
>  	hfi_destroy(core);
> 

-- 
regards,
Stan

Re: Re: [PATCH] [v2] media: venus: core: Fix runtime PM imbalance in venus_probe

[dinghao.liu]

> Hi Dinghao,
> 
> On 6/24/20 9:30 AM, Dinghao Liu wrote:
> > pm_runtime_get_sync() increments the runtime PM usage counter even
> > when it returns an error code. Thus a pairing decrement is needed on
> > the error handling path to keep the counter balanced. For other error
> > paths after this call, things are the same.
> > 
> > Signed-off-by: Dinghao Liu <dinghao.liu@zju.edu.cn>
> > ---
> > 
> > Changelog:
> > 
> > v2: - Add pm_runtime_get_noresume() on failure of
> >       pm_runtime_put_sync() to balance PM counter instead of
> >       releasing everything here.
> 
> You are adding pm_runtime_get_noresume in pm_runtime_put_sync error path
> but the patch description is referring to pm_runtime_get_sync. I'm confused.
> 

When pm_runtime_put_sync failed, the control flow will jump to 
err_dev_unregister, where has already been a pm_runtime_put_noidle. 
If we don't add an extra pm_runtime_get_noresume here, we will 
decrease PM usage counter twice. 

Anyway, this may seem a little strange. Do you have a better
strategy to reorder the labels? 

Regards,
Dinghao

> > ---
> >  drivers/media/platform/qcom/venus/core.c | 5 ++++-
> >  1 file changed, 4 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/media/platform/qcom/venus/core.c b/drivers/media/platform/qcom/venus/core.c
> > index 203c6538044f..b0b932bf8c02 100644
> > --- a/drivers/media/platform/qcom/venus/core.c
> > +++ b/drivers/media/platform/qcom/venus/core.c
> > @@ -287,8 +287,10 @@ static int venus_probe(struct platform_device *pdev)
> >  		goto err_core_deinit;
> >  
> >  	ret = pm_runtime_put_sync(dev);
> > -	if (ret)
> > +	if (ret) {
> > +		pm_runtime_get_noresume(dev);
> >  		goto err_dev_unregister;
> > +	}
> >  
> >  	return 0;
> >  
> > @@ -299,6 +301,7 @@ static int venus_probe(struct platform_device *pdev)
> >  err_venus_shutdown:
> >  	venus_shutdown(core);
> >  err_runtime_disable:
> > +	pm_runtime_put_noidle(dev);
> >  	pm_runtime_set_suspended(dev);
> >  	pm_runtime_disable(dev);
> >  	hfi_destroy(core);
> > 
> 
> -- 
> regards,
> Stan

Re: [PATCH] [v2] media: venus: core: Fix runtime PM imbalance in venus_probe

[Stanimir Varbanov]

Hi,

On 6/26/20 3:54 AM, dinghao.liu@zju.edu.cn wrote:
> 
>> Hi Dinghao,
>>
>> On 6/24/20 9:30 AM, Dinghao Liu wrote:
>>> pm_runtime_get_sync() increments the runtime PM usage counter even
>>> when it returns an error code. Thus a pairing decrement is needed on
>>> the error handling path to keep the counter balanced. For other error
>>> paths after this call, things are the same.
>>>
>>> Signed-off-by: Dinghao Liu <dinghao.liu@zju.edu.cn>
>>> ---
>>>
>>> Changelog:
>>>
>>> v2: - Add pm_runtime_get_noresume() on failure of
>>>       pm_runtime_put_sync() to balance PM counter instead of
>>>       releasing everything here.

Could you reword this and add it to the patch description.

>>
>> You are adding pm_runtime_get_noresume in pm_runtime_put_sync error path
>> but the patch description is referring to pm_runtime_get_sync. I'm confused.
>>
> 
> When pm_runtime_put_sync failed, the control flow will jump to 
> err_dev_unregister, where has already been a pm_runtime_put_noidle. 
> If we don't add an extra pm_runtime_get_noresume here, we will 
> decrease PM usage counter twice.
> 
> Anyway, this may seem a little strange. Do you have a better
> strategy to reorder the labels?

No, it looks fine. With above addition to patch description:

Acked-by: Stanimir Varbanov <stanimir.varbanov@linaro.org>

> 
> Regards,
> Dinghao
> 
>>> ---
>>>  drivers/media/platform/qcom/venus/core.c | 5 ++++-
>>>  1 file changed, 4 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/media/platform/qcom/venus/core.c b/drivers/media/platform/qcom/venus/core.c
>>> index 203c6538044f..b0b932bf8c02 100644
>>> --- a/drivers/media/platform/qcom/venus/core.c
>>> +++ b/drivers/media/platform/qcom/venus/core.c
>>> @@ -287,8 +287,10 @@ static int venus_probe(struct platform_device *pdev)
>>>  		goto err_core_deinit;
>>>  
>>>  	ret = pm_runtime_put_sync(dev);
>>> -	if (ret)
>>> +	if (ret) {
>>> +		pm_runtime_get_noresume(dev);
>>>  		goto err_dev_unregister;
>>> +	}
>>>  
>>>  	return 0;
>>>  
>>> @@ -299,6 +301,7 @@ static int venus_probe(struct platform_device *pdev)
>>>  err_venus_shutdown:
>>>  	venus_shutdown(core);
>>>  err_runtime_disable:
>>> +	pm_runtime_put_noidle(dev);
>>>  	pm_runtime_set_suspended(dev);
>>>  	pm_runtime_disable(dev);
>>>  	hfi_destroy(core);
>>>
>>
>> -- 
>> regards,
>> Stan

-- 
regards,
Stan

Re: Re: [PATCH] [v2] media: venus: core: Fix runtime PM imbalance in venus_probe

[dinghao.liu]

> 
> Could you reword this and add it to the patch description.
> 

Fine. I will fix this in the next version of patch.

Regards,
Dinghao

Re: [PATCH v2] media: venus: core: Fix runtime PM imbalance in venus_probe

[Markus Elfring]

I propose to combine two tags in the previous patch subject.


> pm_runtime_get_sync() increments the runtime PM usage counter even
> when it returns an error code. Thus a pairing decrement is needed on
> the error handling path to keep the counter balanced. For other error
> paths after this call, things are the same.

How do you think about a wording variant like the following?

   Change description:
   The PM runtime usage counter is incremented even if a call of
   the function “pm_runtime_get_sync” failed. Thus decrement it also
   in error cases so that the reference counting is kept consistent.

   Add also a call of the function “pm_runtime_get_noresume”
   in one if branch for the completion of the desired exception handling.


Would you like to add the tag “Fixes” to the commit message?

Regards,
Markus