[PATCH] s5p-fimc: set m2m context to null at the end of fimc_m2m_resume

[PATCH] s5p-fimc: set m2m context to null at the end of fimc_m2m_resume

[Shaik Ameer Basha]

fimc_m2m_job_finish() has to be called with the m2m context for the necessary
cleanup while resume. But currently fimc_m2m_job_finish() always passes
fimc->m2m.ctx as NULL.

This patch changes the order of the calls for proper cleanup while resume.

Signed-off-by: Shaik Ameer Basha <shaik.ameer@samsung.com>
---
 drivers/media/platform/s5p-fimc/fimc-core.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/drivers/media/platform/s5p-fimc/fimc-core.c b/drivers/media/platform/s5p-fimc/fimc-core.c
index 2a1558a..5b11544 100644
--- a/drivers/media/platform/s5p-fimc/fimc-core.c
+++ b/drivers/media/platform/s5p-fimc/fimc-core.c
@@ -868,14 +868,15 @@ static int fimc_m2m_resume(struct fimc_dev *fimc)
 {
 	unsigned long flags;
 
+	if (test_and_clear_bit(ST_M2M_SUSPENDED, &fimc->state))
+		fimc_m2m_job_finish(fimc->m2m.ctx,
+				    VB2_BUF_STATE_ERROR);
+
 	spin_lock_irqsave(&fimc->slock, flags);
 	/* Clear for full H/W setup in first run after resume */
 	fimc->m2m.ctx = NULL;
 	spin_unlock_irqrestore(&fimc->slock, flags);
 
-	if (test_and_clear_bit(ST_M2M_SUSPENDED, &fimc->state))
-		fimc_m2m_job_finish(fimc->m2m.ctx,
-				    VB2_BUF_STATE_ERROR);
 	return 0;
 }
 
-- 
1.8.0

Re: [PATCH] s5p-fimc: set m2m context to null at the end of fimc_m2m_resume

[Sylwester Nawrocki]

Hi Shaik,

On 01/18/2013 11:01 AM, Shaik Ameer Basha wrote:
> fimc_m2m_job_finish() has to be called with the m2m context for the necessary
> cleanup while resume. But currently fimc_m2m_job_finish() always passes
> fimc->m2m.ctx as NULL.
> 
> This patch changes the order of the calls for proper cleanup while resume.
> 
> Signed-off-by: Shaik Ameer Basha <shaik.ameer@samsung.com>
> ---
>  drivers/media/platform/s5p-fimc/fimc-core.c | 7 ++++---
>  1 file changed, 4 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/media/platform/s5p-fimc/fimc-core.c b/drivers/media/platform/s5p-fimc/fimc-core.c
> index 2a1558a..5b11544 100644
> --- a/drivers/media/platform/s5p-fimc/fimc-core.c
> +++ b/drivers/media/platform/s5p-fimc/fimc-core.c
> @@ -868,14 +868,15 @@ static int fimc_m2m_resume(struct fimc_dev *fimc)
>  {
>  	unsigned long flags;
>  
> +	if (test_and_clear_bit(ST_M2M_SUSPENDED, &fimc->state))
> +		fimc_m2m_job_finish(fimc->m2m.ctx,
> +				    VB2_BUF_STATE_ERROR);
> +
>  	spin_lock_irqsave(&fimc->slock, flags);
>  	/* Clear for full H/W setup in first run after resume */
>  	fimc->m2m.ctx = NULL;
>  	spin_unlock_irqrestore(&fimc->slock, flags);
>  
> -	if (test_and_clear_bit(ST_M2M_SUSPENDED, &fimc->state))
> -		fimc_m2m_job_finish(fimc->m2m.ctx,
> -				    VB2_BUF_STATE_ERROR);

Thanks for the patch. Not sure how I managed to miss that...
I'm not convince this is the right fix though. fimc->m2m.ctx should
be reset so the device is properly configured in fimc_dma_run()
callback. Since after suspend/resume cycle all previous registers'
state is lost.
So I think something more like below is needed. Can you check if it
helps ? And what problem exactly are you observing ? Streaming is not
resumed after system resume ?


diff --git a/drivers/media/platform/s5p-fimc/fimc-core.c
b/drivers/media/platform/s5p-fimc/fimc-core.c
index bdb544f..feb8620 100644
--- a/drivers/media/platform/s5p-fimc/fimc-core.c
+++ b/drivers/media/platform/s5p-fimc/fimc-core.c
@@ -869,16 +869,18 @@ static int fimc_m2m_suspend(struct fimc_dev *fimc)

 static int fimc_m2m_resume(struct fimc_dev *fimc)
 {
+       struct fimc_ctx *ctx;
        unsigned long flags;

        spin_lock_irqsave(&fimc->slock, flags);
        /* Clear for full H/W setup in first run after resume */
+       ctx = fimc->m2m.ctx;
        fimc->m2m.ctx = NULL;
        spin_unlock_irqrestore(&fimc->slock, flags);

        if (test_and_clear_bit(ST_M2M_SUSPENDED, &fimc->state))
-               fimc_m2m_job_finish(fimc->m2m.ctx,
-                                   VB2_BUF_STATE_ERROR);
+               fimc_m2m_job_finish(ctx, VB2_BUF_STATE_ERROR);
+
        return 0;
 }

Regards,
Sylwester

-- 
Sylwester Nawrocki
Samsung Poland R&D Center

Re: [PATCH] s5p-fimc: set m2m context to null at the end of fimc_m2m_resume

[Shaik Ameer Basha]

Hi Sylwester,

On Fri, Jan 18, 2013 at 3:58 PM, Sylwester Nawrocki
<s.nawrocki@samsung.com> wrote:
> Hi Shaik,
>
> On 01/18/2013 11:01 AM, Shaik Ameer Basha wrote:
>> fimc_m2m_job_finish() has to be called with the m2m context for the necessary
>> cleanup while resume. But currently fimc_m2m_job_finish() always passes
>> fimc->m2m.ctx as NULL.
>>
>> This patch changes the order of the calls for proper cleanup while resume.
>>
>> Signed-off-by: Shaik Ameer Basha <shaik.ameer@samsung.com>
>> ---
>>  drivers/media/platform/s5p-fimc/fimc-core.c | 7 ++++---
>>  1 file changed, 4 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/media/platform/s5p-fimc/fimc-core.c b/drivers/media/platform/s5p-fimc/fimc-core.c
>> index 2a1558a..5b11544 100644
>> --- a/drivers/media/platform/s5p-fimc/fimc-core.c
>> +++ b/drivers/media/platform/s5p-fimc/fimc-core.c
>> @@ -868,14 +868,15 @@ static int fimc_m2m_resume(struct fimc_dev *fimc)
>>  {
>>       unsigned long flags;
>>
>> +     if (test_and_clear_bit(ST_M2M_SUSPENDED, &fimc->state))
>> +             fimc_m2m_job_finish(fimc->m2m.ctx,
>> +                                 VB2_BUF_STATE_ERROR);
>> +
>>       spin_lock_irqsave(&fimc->slock, flags);
>>       /* Clear for full H/W setup in first run after resume */
>>       fimc->m2m.ctx = NULL;
>>       spin_unlock_irqrestore(&fimc->slock, flags);
>>
>> -     if (test_and_clear_bit(ST_M2M_SUSPENDED, &fimc->state))
>> -             fimc_m2m_job_finish(fimc->m2m.ctx,
>> -                                 VB2_BUF_STATE_ERROR);
>
> Thanks for the patch. Not sure how I managed to miss that...
> I'm not convince this is the right fix though. fimc->m2m.ctx should
> be reset so the device is properly configured in fimc_dma_run()
> callback. Since after suspend/resume cycle all previous registers'
> state is lost.

Yes, you are right. In case, more buffers are queued for the same
context, there can be issues.
I think you can apply this patch with your modified changes.

> So I think something more like below is needed. Can you check if it
> helps ? And what problem exactly are you observing ? Streaming is not
> resumed after system resume ?
>

I was reviewing my gsc-m2m code and found out this issue.
As you know gsc-m2m mostly follows fimc-m2m ;)

Regards,
Shaik Ameer Basha

>
> diff --git a/drivers/media/platform/s5p-fimc/fimc-core.c
> b/drivers/media/platform/s5p-fimc/fimc-core.c
> index bdb544f..feb8620 100644
> --- a/drivers/media/platform/s5p-fimc/fimc-core.c
> +++ b/drivers/media/platform/s5p-fimc/fimc-core.c
> @@ -869,16 +869,18 @@ static int fimc_m2m_suspend(struct fimc_dev *fimc)
>
>  static int fimc_m2m_resume(struct fimc_dev *fimc)
>  {
> +       struct fimc_ctx *ctx;
>         unsigned long flags;
>
>         spin_lock_irqsave(&fimc->slock, flags);
>         /* Clear for full H/W setup in first run after resume */
> +       ctx = fimc->m2m.ctx;
>         fimc->m2m.ctx = NULL;
>         spin_unlock_irqrestore(&fimc->slock, flags);
>
>         if (test_and_clear_bit(ST_M2M_SUSPENDED, &fimc->state))
> -               fimc_m2m_job_finish(fimc->m2m.ctx,
> -                                   VB2_BUF_STATE_ERROR);
> +               fimc_m2m_job_finish(ctx, VB2_BUF_STATE_ERROR);
> +
>         return 0;
>  }
>
> Regards,
> Sylwester
>
> --
> Sylwester Nawrocki
> Samsung Poland R&D Center
> --
> To unsubscribe from this list: send the line "unsubscribe linux-media" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html