ASoC: davinci-pcm: comments for the conversion to BATCH mode
authorBen Gardiner <bengardiner@nanometrics.ca>
Wed, 25 May 2011 13:27:22 +0000 (09:27 -0400)
committerMark Brown <broonie@opensource.wolfsonmicro.com>
Wed, 25 May 2011 14:59:07 +0000 (22:59 +0800)
In the previous commit 'ASoC: davinci-pcm: convert to BATCH mode', the phase
offset of 2 was mentioned in the commit message but not well commented in the
source.

Add descriptive comments of the phase offset with and without ping-pong
buffers enabled.

Signed-off-by: Ben Gardiner <bengardiner@nanometrics.ca>
Acked-by: Liam Girdwood <lrg@ti.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
sound/soc/davinci/davinci-pcm.c

index fa8fc61..c9e0320 100644 (file)
@@ -605,6 +605,18 @@ static int davinci_pcm_prepare(struct snd_pcm_substream *substream)
                print_buf_info(prtd->asp_link[0], "asp_link[0]");
                print_buf_info(prtd->asp_link[1], "asp_link[1]");
 
+               /*
+                * There is a phase offset of 2 periods between the position
+                * used by dma setup and the position reported in the pointer
+                * function.
+                *
+                * The phase offset, when not using ping-pong buffers, is due to
+                * the two consecutive calls to davinci_pcm_enqueue_dma() below.
+                *
+                * Whereas here, with ping-pong buffers, the phase is due to
+                * there being an entire buffer transfer complete before the
+                * first dma completion event triggers davinci_pcm_dma_irq().
+                */
                davinci_pcm_period_elapsed(substream);
                davinci_pcm_period_elapsed(substream);
 
@@ -631,6 +643,13 @@ davinci_pcm_pointer(struct snd_pcm_substream *substream)
        int asp_count;
        unsigned int period_size = snd_pcm_lib_period_bytes(substream);
 
+       /*
+        * There is a phase offset of 2 periods between the position used by dma
+        * setup and the position reported in the pointer function. Either +2 in
+        * the dma setup or -2 here in the pointer function (with wrapping,
+        * both) accounts for this offset -- choose the latter since it makes
+        * the first-time setup clearer.
+        */
        spin_lock(&prtd->lock);
        asp_count = prtd->period - 2;
        spin_unlock(&prtd->lock);