On Fri, May 02, 2014 at 12:35:21AM +0200, Arnd Bergmann wrote:
On Thursday 01 May 2014 12:11:25 Mark Brown wrote:
Why is the fix for this not to ensure that s3c24xx always enables the DMA controller - how likely is it that it would be sane to build a kernel without DMA after all?
S3C24XX_DMA is only needed for ASoC and for s3cmci. The latter uses 'depends on S3C24XX_DMA'. arch/arm/configs/tct_hammer_defconfig is an example of a configuration that does not include DMA because it uses neither of the two drivers.
I'm having a hard time caring about that, the arch code can always do an "if whatever". If that board is what I think it actually does have audio on it, either the defconfig was never updated or the driver wasn't merged into mainline.
How about the patch below?
I guess (modulo the issues below). Killing the wrapper and making the existing selection of DMA support actually select DMA support does avoid the redundancy.
diff --git a/sound/soc/samsung/ac97.c b/sound/soc/samsung/ac97.c index 76b072b..28fe097 100644 --- a/sound/soc/samsung/ac97.c +++ b/sound/soc/samsung/ac97.c @@ -253,10 +253,7 @@ static int s3c_ac97_trigger(struct snd_pcm_substream *substream, int cmd,
writel(ac_glbctrl, s3c_ac97.regs + S3C_AC97_GLBCTRL);
- if (!dma_data->ops)
dma_data->ops = samsung_dma_get_ops();- dma_data->ops->started(dma_data->channel);
s3c2410_dma_ctrl(dma_data->channel, S3C2410_DMAOP_STARTED);
return 0;
}
This code is broken in general - at least s3c64xx has AC'97. That's broken anyway though.