[alsa-devel] [PATCH] ASoC: fsi: reserve prefetch period on DMA transferring

[Kuninori Morimoto]

Current FSI is supporting DMAEngine transfer,
but, it needs to use work queue.
Therefore, DMA transfer settings might be late if there is heavy task.
This patch reserves next period beforehand on DMA transfer function.
Android sound will be breaking up without this patch.

Tested-by: Tomohito Esaki <etom at igel.co.jp>
Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx at renesas.com>
---
 sound/soc/sh/fsi.c |   51 +++++++++++++++++++++++++++++++++------------------
 1 file changed, 33 insertions(+), 18 deletions(-)

diff --git a/sound/soc/sh/fsi.c b/sound/soc/sh/fsi.c
index 4fc2a2a..c608c22 100644
--- a/sound/soc/sh/fsi.c
+++ b/sound/soc/sh/fsi.c
@@ -235,6 +235,8 @@ struct fsi_stream {
 	struct sh_dmae_slave	slave; /* see fsi_handler_init() */
 	struct work_struct	work;
 	dma_addr_t		dma;
+	int			loop_cnt;
+	int			additional_pos;
 };
 
 struct fsi_clk {
@@ -1383,6 +1385,8 @@ static int fsi_dma_init(struct fsi_priv *fsi, struct fsi_stream *io)
 	io->bus_option = BUSOP_SET(24, PACKAGE_24BITBUS_BACK) |
 			 BUSOP_SET(16, PACKAGE_16BITBUS_STREAM);
 
+	io->loop_cnt = 2; /* push 1st, 2nd period first, then 3rd, 4th... */
+	io->additional_pos = 0;
 	io->dma = dma_map_single(dai->dev, runtime->dma_area,
 				 snd_pcm_lib_buffer_bytes(io->substream), dir);
 	return 0;
@@ -1399,11 +1403,15 @@ static int fsi_dma_quit(struct fsi_priv *fsi, struct fsi_stream *io)
 	return 0;
 }
 
-static dma_addr_t fsi_dma_get_area(struct fsi_stream *io)
+static dma_addr_t fsi_dma_get_area(struct fsi_stream *io, int additional)
 {
 	struct snd_pcm_runtime *runtime = io->substream->runtime;
+	int period = io->period_pos + additional;
 
-	return io->dma + samples_to_bytes(runtime, io->buff_sample_pos);
+	if (period >= runtime->periods)
+		period = 0;
+
+	return io->dma + samples_to_bytes(runtime, period * io->period_samples);
 }
 
 static void fsi_dma_complete(void *data)
@@ -1415,7 +1423,7 @@ static void fsi_dma_complete(void *data)
 	enum dma_data_direction dir = fsi_stream_is_play(fsi, io) ?
 		DMA_TO_DEVICE : DMA_FROM_DEVICE;
 
-	dma_sync_single_for_cpu(dai->dev, fsi_dma_get_area(io),
+	dma_sync_single_for_cpu(dai->dev, fsi_dma_get_area(io, 0),
 			samples_to_bytes(runtime, io->period_samples), dir);
 
 	io->buff_sample_pos += io->period_samples;
@@ -1441,7 +1449,7 @@ static void fsi_dma_do_work(struct work_struct *work)
 	struct snd_pcm_runtime *runtime;
 	enum dma_data_direction dir;
 	int is_play = fsi_stream_is_play(fsi, io);
-	int len;
+	int len, i;
 	dma_addr_t buf;
 
 	if (!fsi_stream_is_working(fsi, io))
@@ -1451,26 +1459,33 @@ static void fsi_dma_do_work(struct work_struct *work)
 	runtime	= io->substream->runtime;
 	dir	= is_play ? DMA_TO_DEVICE : DMA_FROM_DEVICE;
 	len	= samples_to_bytes(runtime, io->period_samples);
-	buf	= fsi_dma_get_area(io);
 
-	dma_sync_single_for_device(dai->dev, buf, len, dir);
+	for (i = 0; i < io->loop_cnt; i++) {
+		buf	= fsi_dma_get_area(io, io->additional_pos);
 
-	desc = dmaengine_prep_slave_single(io->chan, buf, len, dir,
-					   DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
-	if (!desc) {
-		dev_err(dai->dev, "dmaengine_prep_slave_sg() fail\n");
-		return;
-	}
+		dma_sync_single_for_device(dai->dev, buf, len, dir);
 
-	desc->callback		= fsi_dma_complete;
-	desc->callback_param	= io;
+		desc = dmaengine_prep_slave_single(io->chan, buf, len, dir,
+					DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+		if (!desc) {
+			dev_err(dai->dev, "dmaengine_prep_slave_sg() fail\n");
+			return;
+		}
 
-	if (dmaengine_submit(desc) < 0) {
-		dev_err(dai->dev, "tx_submit() fail\n");
-		return;
+		desc->callback		= fsi_dma_complete;
+		desc->callback_param	= io;
+
+		if (dmaengine_submit(desc) < 0) {
+			dev_err(dai->dev, "tx_submit() fail\n");
+			return;
+		}
+
+		dma_async_issue_pending(io->chan);
+
+		io->additional_pos = 1;
 	}
 
-	dma_async_issue_pending(io->chan);
+	io->loop_cnt = 1;
 
 	/*
 	 * FIXME
-- 
1.7.9.5