ASoC: fsi: reserve prefetch period on DMA transferring

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@igel.co.jp>
Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Signed-off-by: Mark Brown <broonie@linaro.org>
diff --git a/sound/soc/sh/fsi.c b/sound/soc/sh/fsi.c
index 3039026..b33ca7c 100644
--- a/sound/soc/sh/fsi.c
+++ b/sound/soc/sh/fsi.c
@@ -235,6 +235,8 @@
 	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 {
@@ -1289,6 +1291,8 @@
 	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;
@@ -1305,11 +1309,15 @@
 	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)
@@ -1321,7 +1329,7 @@
 	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;
@@ -1347,7 +1355,7 @@
 	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))
@@ -1357,26 +1365,33 @@
 	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 = 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;
+		}
+
+		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;
 	}
 
-	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->loop_cnt = 1;
 
 	/*
 	 * FIXME