ASoC: SOF: ipc4/intel: Add support for chained DMA

Add logic for setting up and tearing down chained DMA connections.

Since pipelines are not used, all the logic to set the pipeline states
can be bypassed, with only the DMA programming sequences remaining. In
addition the same format needs to be used for host- and link-DMA,
without the usual fixup to use the S32_LE format on the link.

Note however that for convenience and compatibility with existing
definitions, the topology relies on the concept of pipelines with a
'USE_CHAIN_DMA' token indicating that all the logic shall be bypassed.

Unlike 'normal' ALSA sequences, the chain DMA is not programmed in
hw_params/hw_free. The IPC message to set-up and tear-down chained DMA
are sent in sof_ipc4_trigger_pipelines(), but the contents prepared
earlier.

Chained DMA is only supported by the Intel HDA DAI for now, and only
S16_LE and S32_LE formats are supported for now.

Signed-off-by: Jyri Sarha <[email protected]>
Reviewed-by: Rander Wang <[email protected]>
Signed-off-by: Peter Ujfalusi <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Mark Brown <[email protected]>

Author: jsarha

include/uapi/sound/sof/tokens.h

@@ -54,6 +54,7 @@
 #define SOF_TKN_SCHED_DYNAMIC_PIPELINE		206
 #define SOF_TKN_SCHED_LP_MODE			207
 #define SOF_TKN_SCHED_MEM_USAGE			208
+#define SOF_TKN_SCHED_USE_CHAIN_DMA		209
 
 /* volume */
 #define SOF_TKN_VOLUME_RAMP_STEP_TYPE		250

sound/soc/sof/intel/hda-dai-ops.c

@@ -277,6 +277,15 @@ static const struct hda_dai_widget_dma_ops hda_ipc4_dma_ops = {
 	.post_trigger = hda_ipc4_post_trigger
 };
 
+static const struct hda_dai_widget_dma_ops hda_ipc4_chain_dma_ops = {
+	.get_hext_stream = hda_get_hext_stream,
+	.assign_hext_stream = hda_assign_hext_stream,
+	.release_hext_stream = hda_release_hext_stream,
+	.setup_hext_stream = hda_setup_hext_stream,
+	.reset_hext_stream = hda_reset_hext_stream,
+	.trigger = hda_trigger,
+};
+
 static int hda_ipc3_post_trigger(struct snd_sof_dev *sdev, struct snd_soc_dai *cpu_dai,
 				 struct snd_pcm_substream *substream, int cmd)
 {
@@ -331,8 +340,15 @@ hda_select_dai_widget_ops(struct snd_sof_dev *sdev, struct snd_sof_widget *swidg
 	{
 		struct sof_ipc4_copier *ipc4_copier = sdai->private;
 
-		if (ipc4_copier->dai_type == SOF_DAI_INTEL_HDA)
+		if (ipc4_copier->dai_type == SOF_DAI_INTEL_HDA) {
+			struct snd_sof_widget *pipe_widget = swidget->spipe->pipe_widget;
+			struct sof_ipc4_pipeline *pipeline = pipe_widget->private;
+
+			if (pipeline->use_chain_dma)
+				return &hda_ipc4_chain_dma_ops;
+
 			return &hda_ipc4_dma_ops;
+		}
 		break;
 	}
 	default:

sound/soc/sof/ipc4-pcm.c

@@ -193,6 +193,88 @@ sof_ipc4_update_pipeline_state(struct snd_sof_dev *sdev, int state, int cmd,
  * prepare ioctl before the START trigger.
  */
 
+/*
+ * Chained DMA is a special case where there is no processing on
+ * DSP. The samples are just moved over by host side DMA to a single
+ * buffer on DSP and directly from there to link DMA. However, the
+ * model on SOF driver has two notional pipelines, one at host DAI,
+ * and another at link DAI. They both shall have the use_chain_dma
+ * attribute.
+ */
+
+static int sof_ipc4_chain_dma_trigger(struct snd_sof_dev *sdev,
+				      struct snd_sof_pcm_stream_pipeline_list *pipeline_list,
+				      int state, int cmd)
+{
+	bool allocate, enable, set_fifo_size;
+	struct sof_ipc4_msg msg = {{ 0 }};
+	int i;
+
+	switch (state) {
+	case SOF_IPC4_PIPE_RUNNING: /* Allocate and start chained dma */
+		allocate = true;
+		enable = true;
+		/*
+		 * SOF assumes creation of a new stream from the presence of fifo_size
+		 * in the message, so we must leave it out in pause release case.
+		 */
+		if (cmd == SNDRV_PCM_TRIGGER_PAUSE_RELEASE)
+			set_fifo_size = false;
+		else
+			set_fifo_size = true;
+		break;
+	case SOF_IPC4_PIPE_PAUSED: /* Disable chained DMA. */
+		allocate = true;
+		enable = false;
+		set_fifo_size = false;
+		break;
+	case SOF_IPC4_PIPE_RESET: /* Disable and free chained DMA. */
+		allocate = false;
+		enable = false;
+		set_fifo_size = false;
+		break;
+	default:
+		dev_err(sdev->dev, "Unexpected state %d", state);
+		return -EINVAL;
+	}
+
+	msg.primary = SOF_IPC4_MSG_TYPE_SET(SOF_IPC4_GLB_CHAIN_DMA);
+	msg.primary |= SOF_IPC4_MSG_DIR(SOF_IPC4_MSG_REQUEST);
+	msg.primary |= SOF_IPC4_MSG_TARGET(SOF_IPC4_FW_GEN_MSG);
+
+	/*
+	 * To set-up the DMA chain, the host DMA ID and SCS setting
+	 * are retrieved from the host pipeline configuration. Likewise
+	 * the link DMA ID and fifo_size are retrieved from the link
+	 * pipeline configuration.
+	 */
+	for (i = 0; i < pipeline_list->count; i++) {
+		struct snd_sof_pipeline *spipe = pipeline_list->pipelines[i];
+		struct snd_sof_widget *pipe_widget = spipe->pipe_widget;
+		struct sof_ipc4_pipeline *pipeline = pipe_widget->private;
+
+		if (!pipeline->use_chain_dma) {
+			dev_err(sdev->dev,
+				"All pipelines in chained DMA stream should have use_chain_dma attribute set.");
+			return -EINVAL;
+		}
+
+		msg.primary |= pipeline->msg.primary;
+
+		/* Add fifo_size (actually DMA buffer size) field to the message */
+		if (set_fifo_size)
+			msg.extension |= pipeline->msg.extension;
+	}
+
+	if (allocate)
+		msg.primary |= SOF_IPC4_GLB_CHAIN_DMA_ALLOCATE_MASK;
+
+	if (enable)
+		msg.primary |= SOF_IPC4_GLB_CHAIN_DMA_ENABLE_MASK;
+
+	return sof_ipc_tx_message(sdev->ipc, &msg, 0, NULL, 0);
+}
+
 static int sof_ipc4_trigger_pipelines(struct snd_soc_component *component,
 				      struct snd_pcm_substream *substream, int state, int cmd)
 {
@@ -201,6 +283,8 @@ static int sof_ipc4_trigger_pipelines(struct snd_soc_component *component,
 	struct snd_sof_pcm_stream_pipeline_list *pipeline_list;
 	struct sof_ipc4_fw_data *ipc4_data = sdev->private;
 	struct ipc4_pipeline_set_state_data *trigger_list;
+	struct snd_sof_widget *pipe_widget;
+	struct sof_ipc4_pipeline *pipeline;
 	struct snd_sof_pipeline *spipe;
 	struct snd_sof_pcm *spcm;
 	int ret;
@@ -218,6 +302,17 @@ static int sof_ipc4_trigger_pipelines(struct snd_soc_component *component,
 	if (!pipeline_list->pipelines || !pipeline_list->count)
 		return 0;
 
+	spipe = pipeline_list->pipelines[0];
+	pipe_widget = spipe->pipe_widget;
+	pipeline = pipe_widget->private;
+
+	/*
+	 * If use_chain_dma attribute is set we proceed to chained DMA
+	 * trigger function that handles the rest for the substream.
+	 */
+	if (pipeline->use_chain_dma)
+		return sof_ipc4_chain_dma_trigger(sdev, pipeline_list, state, cmd);
+
 	/* allocate memory for the pipeline data */
 	trigger_list = kzalloc(struct_size(trigger_list, pipeline_ids, pipeline_list->count),
 			       GFP_KERNEL);
@@ -422,8 +517,10 @@ static int sof_ipc4_pcm_dai_link_fixup(struct snd_soc_pcm_runtime *rtd,
 	struct snd_soc_component *component = snd_soc_rtdcom_lookup(rtd, SOF_AUDIO_PCM_DRV_NAME);
 	struct snd_sof_dai *dai = snd_sof_find_dai(component, rtd->dai_link->name);
 	struct snd_sof_dev *sdev = snd_soc_component_get_drvdata(component);
+	struct snd_soc_dai *cpu_dai = asoc_rtd_to_cpu(rtd, 0);
 	struct sof_ipc4_copier *ipc4_copier;
-	int ret;
+	bool use_chain_dma = false;
+	int dir;
 
 	if (!dai) {
 		dev_err(component->dev, "%s: No DAI found with name %s\n", __func__,
@@ -438,9 +535,26 @@ static int sof_ipc4_pcm_dai_link_fixup(struct snd_soc_pcm_runtime *rtd,
 		return -EINVAL;
 	}
 
-	ret = sof_ipc4_pcm_dai_link_fixup_rate(sdev, params, ipc4_copier);
-	if (ret)
-		return ret;
+	for_each_pcm_streams(dir) {
+		struct snd_soc_dapm_widget *w = snd_soc_dai_get_widget(cpu_dai, dir);
+
+		if (w) {
+			struct snd_sof_widget *swidget = w->dobj.private;
+			struct snd_sof_widget *pipe_widget = swidget->spipe->pipe_widget;
+			struct sof_ipc4_pipeline *pipeline = pipe_widget->private;
+
+			if (pipeline->use_chain_dma)
+				use_chain_dma = true;
+		}
+	}
+
+	/* Chain DMA does not use copiers, so no fixup needed */
+	if (!use_chain_dma) {
+		int ret = sof_ipc4_pcm_dai_link_fixup_rate(sdev, params, ipc4_copier);
+
+		if (ret)
+			return ret;
+	}
 
 	switch (ipc4_copier->dai_type) {
 	case SOF_DAI_INTEL_SSP: