Re: [PATCH] dmaengine: qcom: bam_dma: fix runtime PM underflow

[Caleb Connolly <caleb.connolly@linaro.org>]

On 16/06/2022 15:00, Vinod Koul wrote:
> On 09-06-22, 20:50, Caleb Connolly wrote:
>> When PM runtime is disabled, pm_runtime_get() isn't called, but
>> pm_runtime_put() still is. Fix this by creating a matching wrapper
>> on pm_runtime_put_autosuspend().
>>
>> Fixes: dbad41e7bb5f ("dmaengine: qcom: bam_dma: check if the runtime pm enabled")
>> Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>
>> ---
>>   drivers/dma/qcom/bam_dma.c | 18 +++++++++++++-----
>>   1 file changed, 13 insertions(+), 5 deletions(-)
>>
>> diff --git a/drivers/dma/qcom/bam_dma.c b/drivers/dma/qcom/bam_dma.c
>> index 87f6ca1541cf..a36dedee262e 100644
>> --- a/drivers/dma/qcom/bam_dma.c
>> +++ b/drivers/dma/qcom/bam_dma.c
>> @@ -566,6 +566,14 @@ static int bam_pm_runtime_get_sync(struct device *dev)
>>   	return 0;
>>   }
>>   
>> +static int bam_pm_runtime_put_autosuspend(struct device *dev)
>> +{
>> +	if (pm_runtime_enabled(dev))
>> +		return pm_runtime_put_autosuspend(dev);
>> +
>> +	return 0;
>> +}
> 
> should we really do a wrapper to fix this ;-) I would think dropping the
> get wrapper and calling pm_runtime_get() unconditionally would be
> better..?
The original pm_runtime_get() wrapper was added because _get() returns -EACCES 
if PM is disabled for the device, I've read through some of the pm_runtime docs 
and I don't see any way to handle this nicely in the framework. I think the real 
fix will have to be in pm_runtime.

I can rework this to explicitly check for -EACCES on every _get() call, but we 
still need to check if pm_runtime is enabled before calling _put().
> 
>> +
>>   /**
>>    * bam_free_chan - Frees dma resources associated with specific channel
>>    * @chan: specified channel
>> @@ -617,7 +625,7 @@ static void bam_free_chan(struct dma_chan *chan)
>>   
>>   err:
>>   	pm_runtime_mark_last_busy(bdev->dev);
>> -	pm_runtime_put_autosuspend(bdev->dev);
>> +	bam_pm_runtime_put_autosuspend(bdev->dev);
>>   }
>>   
>>   /**
>> @@ -793,7 +801,7 @@ static int bam_pause(struct dma_chan *chan)
>>   	bchan->paused = 1;
>>   	spin_unlock_irqrestore(&bchan->vc.lock, flag);
>>   	pm_runtime_mark_last_busy(bdev->dev);
>> -	pm_runtime_put_autosuspend(bdev->dev);
>> +	bam_pm_runtime_put_autosuspend(bdev->dev);
>>   
>>   	return 0;
>>   }
>> @@ -819,7 +827,7 @@ static int bam_resume(struct dma_chan *chan)
>>   	bchan->paused = 0;
>>   	spin_unlock_irqrestore(&bchan->vc.lock, flag);
>>   	pm_runtime_mark_last_busy(bdev->dev);
>> -	pm_runtime_put_autosuspend(bdev->dev);
>> +	bam_pm_runtime_put_autosuspend(bdev->dev);
>>   
>>   	return 0;
>>   }
>> @@ -936,7 +944,7 @@ static irqreturn_t bam_dma_irq(int irq, void *data)
>>   	}
>>   
>>   	pm_runtime_mark_last_busy(bdev->dev);
>> -	pm_runtime_put_autosuspend(bdev->dev);
>> +	bam_pm_runtime_put_autosuspend(bdev->dev);
>>   
>>   	return IRQ_HANDLED;
>>   }
>> @@ -1111,7 +1119,7 @@ static void bam_start_dma(struct bam_chan *bchan)
>>   			bam_addr(bdev, bchan->id, BAM_P_EVNT_REG));
>>   
>>   	pm_runtime_mark_last_busy(bdev->dev);
>> -	pm_runtime_put_autosuspend(bdev->dev);
>> +	bam_pm_runtime_put_autosuspend(bdev->dev);
>>   }
>>   
>>   /**
>> -- 
>> 2.36.1
> 

-- 
Kind Regards,
Caleb (they/he)