CHROMIUM: ASoC: Tegra: Double-buffer DMA in HW

Avoid audio FIFO underflow due to DMA ISR latency.

MODE_ONESHOT works like this:

* CPU tells DMA HW to make a transfer .
* DMA HW performs the transfer, fires an interrupt, and goes idle.
* Interrupt latency (including e.g. LP2 exit time) is seen here.
* DMA driver ISR programs the next transfer, from a SW transfer queue.

In that model, the interrupt latency must be hidden by the I2S FIFO, which
is 16 samples IIRC, i.e. a very short time (~0.3mS @ 48KHz).

MODE_CONTINUOUS_SINGLE works like this:

* CPU tells DMA HW to make a transfer .
* CPU tells DMA HW to make a second transfer.
* DMA HW performs the transfer, fires an interrupt, and starts the second
  transfer immediately.
* DMA driver ISR programs the next (third) transfer, while the second is
  in progress, from a SW transfer queue.

Thus any DMA ISR latency is seen between transfer N completing and transfer
N+2 starting (rather than N+1), since the HW itself already knows how to
perform transfer N+1 before transfer N completes. Hence, the system is
tolerant of much larger latencies.

BUG=chromium-os-partner:3294
TEST=Seaboard clamshell: "aplay sin-500hz-48000-60s.wav" multiple times
  while system idle and power LED flickering indicating CPU power saving.

Signed-off-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Olof Johansson <olofj@chromium.org>

1 files changed, 1 insertions, 1 deletions

diff --git a/sound/soc/tegra/tegra_pcm.c b/sound/soc/tegra/tegra_pcm.c
index 4524b0f88ef8..dde24b42483b 100644
--- a/sound/soc/tegra/tegra_pcm.c
+++ b/sound/soc/tegra/tegra_pcm.c
@@ -161,7 +161,7 @@ static int tegra_pcm_open(struct snd_pcm_substream *substream)
 	prtd->dma_req[1].dev = prtd;
 
 	prtd->dma_chan = tegra_dma_allocate_channel(
-				TEGRA_DMA_MODE_ONESHOT, "pcm");
+				TEGRA_DMA_MODE_CONTINUOUS_SINGLE, "pcm");
 	if (prtd->dma_chan == NULL) {
 		ret = -ENOMEM;
 		goto err;