Re: [PATCH] iio: adc: at91-sama5d2_adc: fix DMA conversion crash

[Jonathan Cameron]

On Wed, 16 Sep 2020 10:08:21 +0300
Eugen Hristev <eugen.hristev@xxxxxxxxxxxxx> wrote:

> After the move of the postenable code to preenable, the DMA start was
> done before the DMA init, which is not correct.
> The DMA is initialized in set_watermark. Because of this, we need to call
> the DMA start functions in set_watermark, after the DMA init, instead of
> preenable hook, when the DMA is not properly setup yet.
> 
> Fixes: f3c034f61775 ("iio: at91-sama5d2_adc: adjust iio_triggered_buffer_{predisable,postenable} positions")
> Signed-off-by: Eugen Hristev <eugen.hristev@xxxxxxxxxxxxx>

Basic approach looks fine, but if we were to get an error from  the
buffer_prepare() function, should we be looking to disable dma if we
enabled it earlier in the set_watermark callback?

> ---
> 
> Hi,
> 
> This crash is in the kernel since 5.8-rc1
> 
> Please have a look at discussion here:
> https://lore.kernel.org/linux-iio/CA+U=DsqRUtjjoe5nevP_wNxTgr27+O2V1h9w7d3QijBQ+5f3XA@xxxxxxxxxxxxxx/T/#t
> 
> Thanks !
> 
>  drivers/iio/adc/at91-sama5d2_adc.c | 11 +++++++----
>  1 file changed, 7 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/iio/adc/at91-sama5d2_adc.c b/drivers/iio/adc/at91-sama5d2_adc.c
> index de9583d6cddd..b5196797dcb8 100644
> --- a/drivers/iio/adc/at91-sama5d2_adc.c
> +++ b/drivers/iio/adc/at91-sama5d2_adc.c
> @@ -884,7 +884,7 @@ static bool at91_adc_current_chan_is_touch(struct iio_dev *indio_dev)
>  			       AT91_SAMA5D2_MAX_CHAN_IDX + 1);
>  }
>  
> -static int at91_adc_buffer_preenable(struct iio_dev *indio_dev)
> +static int at91_adc_buffer_prepare(struct iio_dev *indio_dev)
>  {
>  	int ret;
>  	u8 bit;
> @@ -901,7 +901,7 @@ static int at91_adc_buffer_preenable(struct iio_dev *indio_dev)
>  	/* we continue with the triggered buffer */
>  	ret = at91_adc_dma_start(indio_dev);
>  	if (ret) {
> -		dev_err(&indio_dev->dev, "buffer postenable failed\n");
> +		dev_err(&indio_dev->dev, "buffer prepare failed\n");
>  		return ret;
>  	}
>  
> @@ -989,7 +989,6 @@ static int at91_adc_buffer_postdisable(struct iio_dev *indio_dev)
>  }
>  
>  static const struct iio_buffer_setup_ops at91_buffer_setup_ops = {
> -	.preenable = &at91_adc_buffer_preenable,
>  	.postdisable = &at91_adc_buffer_postdisable,
>  };
>  
> @@ -1586,7 +1585,11 @@ static int at91_adc_set_watermark(struct iio_dev *indio_dev, unsigned int val)
>  	else if (val > 1)
>  		at91_adc_dma_init(to_platform_device(&indio_dev->dev));
>  
> -	return 0;
> +	/*
> +	 * We can start the DMA only after setting the watermark and
> +	 * having the DMA initialization completed
> +	 */

Looks superficially to me like we may need to unwind some stuff if this fails.

> +	return at91_adc_buffer_prepare(indio_dev);
>  }
>  
>  static int at91_adc_update_scan_mode(struct iio_dev *indio_dev,