[PATCH 10/29] ALSA: dice: Add support for duplex streams with synchronization

[Takashi Sakamoto <o-takashi@sakamocchi.jp>]

This commit adds support for AMDTP in-stream. As a result, Dice driver
supports full duplex streams with synchronization.

AMDTP can transfer timestamps in its packets. By handling the timestamp,
devices can synchronize to the other devices or drivers on the same bus.

When Dice chipset is 'enabled', it starts streams with correct settings.
This 'enable' register is global, thus, when a stream is started to run,
an opposite stream can't start unless turning off 'enable'. Therefore
a pair of streams must be running. This causes a loss of CPU usage when
single stream is needed for neither playbacking or capturing.

This commit assumes that playback-only models also have a functionality
to transmit stream for delivering timestamps.

Currently, sampling clock source is restricted to SYT-Match mode. This is
improved in followed commit. I note that at SYT-Match mode, Dice can select
from 4 streams for synchronization but this driver uses the 1st stream only
for simplicity.

Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
---
 sound/firewire/dice/dice-pcm.c    |   4 +-
 sound/firewire/dice/dice-stream.c | 230 ++++++++++++++++++++++++++++----------
 sound/firewire/dice/dice.c        |  29 +++--
 sound/firewire/dice/dice.h        |  26 +++--
 4 files changed, 202 insertions(+), 87 deletions(-)

diff --git a/sound/firewire/dice/dice-pcm.c b/sound/firewire/dice/dice-pcm.c
index 904cb80..78b5408 100644
--- a/sound/firewire/dice/dice-pcm.c
+++ b/sound/firewire/dice/dice-pcm.c
@@ -180,7 +180,7 @@ static int playback_hw_free(struct snd_pcm_substream *substream)
 {
 	struct snd_dice *dice = substream->private_data;
 
-	snd_dice_stream_stop(dice);
+	snd_dice_stream_stop_duplex(dice);
 
 	return snd_pcm_lib_free_vmalloc_buffer(substream);
 }
@@ -190,7 +190,7 @@ static int playback_prepare(struct snd_pcm_substream *substream)
 	struct snd_dice *dice = substream->private_data;
 	int err;
 
-	err = snd_dice_stream_start(dice, substream->runtime->rate);
+	err = snd_dice_stream_start_duplex(dice, substream->runtime->rate);
 	if (err >= 0)
 		amdtp_stream_pcm_prepare(&dice->rx_stream);
 
diff --git a/sound/firewire/dice/dice-stream.c b/sound/firewire/dice/dice-stream.c
index 6bb340d..edd6be3 100644
--- a/sound/firewire/dice/dice-stream.c
+++ b/sound/firewire/dice/dice-stream.c
@@ -41,55 +41,79 @@ int snd_dice_stream_get_rate_mode(struct snd_dice *dice, unsigned int rate,
 	return -EINVAL;
 }
 
-static void release_resources(struct snd_dice *dice)
+static void release_resources(struct snd_dice *dice,
+			      struct fw_iso_resources *resources)
 {
 	unsigned int channel;
 
 	/* Reset channel number */
 	channel = cpu_to_be32((u32)-1);
-	snd_dice_transaction_write_rx(dice, RX_ISOCHRONOUS, &channel, 4);
-
-	fw_iso_resources_free(&dice->rx_resources);
+	if (resources == &dice->tx_resources)
+		snd_dice_transaction_write_tx(dice, TX_ISOCHRONOUS,
+					      &channel, 4);
+	else
+		snd_dice_transaction_write_rx(dice, RX_ISOCHRONOUS,
+					      &channel, 4);
+
+	fw_iso_resources_free(resources);
 }
 
-static int keep_resources(struct snd_dice *dice, unsigned int max_payload_bytes)
+static int keep_resources(struct snd_dice *dice,
+			  struct fw_iso_resources *resources,
+			  unsigned int max_payload_bytes)
 {
 	unsigned int channel;
 	int err;
 
-	err = fw_iso_resources_allocate(&dice->rx_resources, max_payload_bytes,
+	err = fw_iso_resources_allocate(resources, max_payload_bytes,
 				fw_parent_device(dice->unit)->max_speed);
 	if (err < 0)
 		goto end;
 
 	/* Set channel number */
-	channel = cpu_to_be32(dice->rx_resources.channel);
-	err = snd_dice_transaction_write_rx(dice, RX_ISOCHRONOUS,
-					    &channel, 4);
+	channel = cpu_to_be32(resources->channel);
+	if (resources == &dice->tx_resources)
+		err = snd_dice_transaction_write_tx(dice, TX_ISOCHRONOUS,
+						    &channel, 4);
+	else
+		err = snd_dice_transaction_write_rx(dice, RX_ISOCHRONOUS,
+						    &channel, 4);
 	if (err < 0)
-		release_resources(dice);
+		release_resources(dice, resources);
 end:
 	return err;
 }
 
-static void stop_stream(struct snd_dice *dice)
+static void stop_stream(struct snd_dice *dice, struct amdtp_stream *stream)
 {
-	if (!amdtp_stream_running(&dice->rx_stream))
-		return;
+	amdtp_stream_pcm_abort(stream);
+	amdtp_stream_stop(stream);
 
-	amdtp_stream_pcm_abort(&dice->rx_stream);
-	amdtp_stream_stop(&dice->rx_stream);
-	release_resources(dice);
+	if (stream == &dice->tx_stream)
+		release_resources(dice, &dice->tx_resources);
+	else
+		release_resources(dice, &dice->rx_resources);
 }
 
-static int start_stream(struct snd_dice *dice, unsigned int rate)
+static int start_stream(struct snd_dice *dice, struct amdtp_stream *stream,
+			unsigned int rate)
 {
+	struct fw_iso_resources *resources;
 	unsigned int i, mode, pcm_chs, midi_ports;
 	int err;
 
 	err = snd_dice_stream_get_rate_mode(dice, rate, &mode);
 	if (err < 0)
 		goto end;
+	if (stream == &dice->tx_stream) {
+		resources = &dice->tx_resources;
+		pcm_chs = dice->tx_channels[mode];
+		midi_ports = dice->tx_midi_ports[mode];
+	} else {
+		resources = &dice->rx_resources;
+		pcm_chs = dice->rx_channels[mode];
+		midi_ports = dice->rx_midi_ports[mode];
+	}
 
 	/*
 	 * At 176.4/192.0 kHz, Dice has a quirk to transfer two PCM frames in
@@ -101,53 +125,73 @@ static int start_stream(struct snd_dice *dice, unsigned int rate)
 	 * For this quirk, blocking mode is required and PCM buffer size should
 	 * be aligned to SYT_INTERVAL.
 	 */
-	pcm_chs = dice->rx_channels[mode];
-	midi_ports = dice->rx_midi_ports[mode];
 	if (mode > 1) {
 		rate /= 2;
 		pcm_chs *= 2;
-		dice->rx_stream.double_pcm_frames = true;
+		stream->double_pcm_frames = true;
 	} else {
-		dice->rx_stream.double_pcm_frames = false;
+		stream->double_pcm_frames = false;
 	}
 
-	amdtp_stream_set_parameters(&dice->rx_stream, rate,
-				    pcm_chs, midi_ports);
+	amdtp_stream_set_parameters(stream, rate, pcm_chs, midi_ports);
 	if (mode > 1) {
 		pcm_chs /= 2;
 
 		for (i = 0; i < pcm_chs; i++) {
-			dice->rx_stream.pcm_positions[i] = i * 2;
-			dice->rx_stream.pcm_positions[i + pcm_chs] = i * 2 + 1;
+			stream->pcm_positions[i] = i * 2;
+			stream->pcm_positions[i + pcm_chs] = i * 2 + 1;
 		}
 	}
 
-	err = keep_resources(dice,
-			     amdtp_stream_get_max_payload(&dice->rx_stream));
+	err = keep_resources(dice, resources,
+			     amdtp_stream_get_max_payload(stream));
 	if (err < 0) {
 		dev_err(&dice->unit->device,
 			"fail to keep isochronous resources\n");
 		goto end;
 	}
 
-	err = amdtp_stream_start(&dice->rx_stream, dice->rx_resources.channel,
+	err = amdtp_stream_start(stream, resources->channel,
 				 fw_parent_device(dice->unit)->max_speed);
 	if (err < 0)
-		release_resources(dice);
+		release_resources(dice, resources);
 end:
 	return err;
 }
 
-int snd_dice_stream_start(struct snd_dice *dice, unsigned int rate)
+static int get_sync_mode(struct snd_dice *dice, enum cip_flags *sync_mode)
+{
+	/* Currently, clock source is fixed at SYT-Match mode. */
+	*sync_mode = 0;
+	return 0;
+}
+
+int snd_dice_stream_start_duplex(struct snd_dice *dice, unsigned int rate)
 {
+	struct amdtp_stream *master, *slave;
 	unsigned int curr_rate;
-	int err;
+	enum cip_flags sync_mode;
+	int err = 0;
+
+	if (atomic_read(&dice->substreams_counter) == 0)
+		goto end;
 
 	mutex_lock(&dice->mutex);
 
+	err = get_sync_mode(dice, &sync_mode);
+	if (err < 0)
+		goto end;
+	if (sync_mode == CIP_SYNC_TO_DEVICE) {
+		master = &dice->tx_stream;
+		slave  = &dice->rx_stream;
+	} else {
+		master = &dice->rx_stream;
+		slave  = &dice->tx_stream;
+	}
+
 	/* Some packet queueing errors. */
-	if (amdtp_streaming_error(&dice->rx_stream))
-		stop_stream(dice);
+	if (amdtp_streaming_error(master) || amdtp_streaming_error(slave))
+		stop_stream(dice, master);
 
 	/* Stop stream if rate is different. */
 	err = snd_dice_transaction_get_rate(dice, &curr_rate);
@@ -157,11 +201,14 @@ int snd_dice_stream_start(struct snd_dice *dice, unsigned int rate)
 		goto end;
 	}
 	if (rate != curr_rate)
-		stop_stream(dice);
+		stop_stream(dice, master);
 
-	if (!amdtp_stream_running(&dice->rx_stream)) {
+	if (!amdtp_stream_running(master)) {
+		stop_stream(dice, slave);
 		snd_dice_transaction_clear_enable(dice);
 
+		amdtp_stream_set_sync(sync_mode, master, slave);
+
 		err = snd_dice_transaction_set_rate(dice, rate);
 		if (err < 0) {
 			dev_err(&dice->unit->device,
@@ -169,25 +216,35 @@ int snd_dice_stream_start(struct snd_dice *dice, unsigned int rate)
 			goto end;
 		}
 
-		/* Start stream. */
-		err = start_stream(dice, rate);
+		/* Start both streams. */
+		err = start_stream(dice, master, rate);
+		if (err < 0) {
+			dev_err(&dice->unit->device,
+				"fail to start AMDTP master stream\n");
+			goto end;
+		}
+		err = start_stream(dice, slave, rate);
 		if (err < 0) {
 			dev_err(&dice->unit->device,
-				"fail to start AMDTP stream\n");
+				"fail to start AMDTP slave stream\n");
+			stop_stream(dice, master);
 			goto end;
 		}
 		err = snd_dice_transaction_set_enable(dice);
 		if (err < 0) {
 			dev_err(&dice->unit->device,
 				"fail to enable interface\n");
-			stop_stream(dice);
+			stop_stream(dice, master);
+			stop_stream(dice, slave);
 			goto end;
 		}
 
-		if (!amdtp_stream_wait_callback(&dice->rx_stream,
-						CALLBACK_TIMEOUT)) {
+		/* Wait first callbacks */
+		if (!amdtp_stream_wait_callback(master, CALLBACK_TIMEOUT) ||
+		    !amdtp_stream_wait_callback(slave, CALLBACK_TIMEOUT)) {
 			snd_dice_transaction_clear_enable(dice);
-			stop_stream(dice);
+			stop_stream(dice, master);
+			stop_stream(dice, slave);
 			err = -ETIMEDOUT;
 		}
 	}
@@ -196,54 +253,101 @@ end:
 	return err;
 }
 
-void snd_dice_stream_stop(struct snd_dice *dice)
+void snd_dice_stream_stop_duplex(struct snd_dice *dice)
 {
+	if (atomic_read(&dice->substreams_counter) > 0)
+		return;
+
 	mutex_lock(&dice->mutex);
 
 	snd_dice_transaction_clear_enable(dice);
-	stop_stream(dice);
+
+	stop_stream(dice, &dice->tx_stream);
+	stop_stream(dice, &dice->rx_stream);
 
 	mutex_unlock(&dice->mutex);
 }
 
-int snd_dice_stream_init(struct snd_dice *dice)
+static int init_stream(struct snd_dice *dice, struct amdtp_stream *stream)
 {
 	int err;
+	struct fw_iso_resources *resources;
+	enum amdtp_stream_direction dir;
+
+	if (stream == &dice->tx_stream) {
+		resources = &dice->tx_resources;
+		dir = AMDTP_IN_STREAM;
+	} else {
+		resources = &dice->rx_resources;
+		dir = AMDTP_OUT_STREAM;
+	}
 
-	err = fw_iso_resources_init(&dice->rx_resources, dice->unit);
+	err = fw_iso_resources_init(resources, dice->unit);
 	if (err < 0)
 		goto end;
-	dice->rx_resources.channels_mask = 0x00000000ffffffffuLL;
+	resources->channels_mask = 0x00000000ffffffffuLL;
 
-	err = amdtp_stream_init(&dice->rx_stream, dice->unit, AMDTP_OUT_STREAM,
-				CIP_BLOCKING);
+	err = amdtp_stream_init(stream, dice->unit, dir, CIP_BLOCKING);
+	if (err < 0) {
+		amdtp_stream_destroy(stream);
+		fw_iso_resources_destroy(resources);
+	}
+end:
+	return err;
+}
+
+static void destroy_stream(struct snd_dice *dice, struct amdtp_stream *stream)
+{
+	amdtp_stream_destroy(stream);
+
+	if (stream == &dice->tx_stream)
+		fw_iso_resources_destroy(&dice->tx_resources);
+	else
+		fw_iso_resources_destroy(&dice->rx_resources);
+}
+
+int snd_dice_stream_init_duplex(struct snd_dice *dice)
+{
+	int err;
+
+	atomic_set(&dice->substreams_counter, 0);
+
+	err = init_stream(dice, &dice->tx_stream);
 	if (err < 0)
-		goto error;
+		goto end;
 
-	err = snd_dice_transaction_set_clock_source(dice, CLOCK_SOURCE_ARX1);
+	err = init_stream(dice, &dice->rx_stream);
 	if (err < 0)
-		goto error;
+		goto end;
+
+	/* Currently, clock source is fixed at SYT-Match mode. */
+	err = snd_dice_transaction_set_clock_source(dice, CLOCK_SOURCE_ARX1);
+	if (err < 0) {
+		destroy_stream(dice, &dice->rx_stream);
+		destroy_stream(dice, &dice->tx_stream);
+	}
 end:
 	return err;
-error:
-	amdtp_stream_destroy(&dice->rx_stream);
-	fw_iso_resources_destroy(&dice->rx_resources);
-	return err;
 }
 
-void snd_dice_stream_destroy(struct snd_dice *dice)
+void snd_dice_stream_destroy_duplex(struct snd_dice *dice)
 {
 	mutex_lock(&dice->mutex);
 
 	snd_dice_transaction_clear_enable(dice);
-	stop_stream(dice);
-	amdtp_stream_destroy(&dice->rx_stream);
-	fw_iso_resources_destroy(&dice->rx_resources);
+
+	stop_stream(dice, &dice->tx_stream);
+	destroy_stream(dice, &dice->tx_stream);
+
+	stop_stream(dice, &dice->rx_stream);
+	destroy_stream(dice, &dice->rx_stream);
+
+	atomic_set(&dice->substreams_counter, 0);
 
 	mutex_unlock(&dice->mutex);
 }
 
-void snd_dice_stream_update(struct snd_dice *dice)
+void snd_dice_stream_update_duplex(struct snd_dice *dice)
 {
 	/*
 	 * On a bus reset, the DICE firmware disables streaming and then goes
@@ -258,9 +362,11 @@ void snd_dice_stream_update(struct snd_dice *dice)
 	/* The enable register becomes initialized, then streams are stopped. */
 	dice->global_enabled = false;
 
-	stop_stream(dice);
+	stop_stream(dice, &dice->rx_stream);
+	stop_stream(dice, &dice->tx_stream);
 
 	fw_iso_resources_update(&dice->rx_resources);
+	fw_iso_resources_update(&dice->tx_resources);
 
 	mutex_unlock(&dice->mutex);
 }
diff --git a/sound/firewire/dice/dice.c b/sound/firewire/dice/dice.c
index 7bddd29..2390cc6 100644
--- a/sound/firewire/dice/dice.c
+++ b/sound/firewire/dice/dice.c
@@ -30,7 +30,6 @@ static int dice_interface_check(struct fw_unit *unit)
 	int key, val, vendor = -1, model = -1, err;
 	unsigned int category, i;
 	__be32 *pointers, value;
-	__be32 tx_data[4];
 	__be32 version;
 
 	pointers = kmalloc_array(ARRAY_SIZE(min_values), sizeof(__be32),
@@ -85,16 +84,6 @@ static int dice_interface_check(struct fw_unit *unit)
 		}
 	}
 
-	/* We support playback only. Let capture devices be handled by FFADO. */
-	err = snd_fw_transaction(unit, TCODE_READ_BLOCK_REQUEST,
-				 DICE_PRIVATE_SPACE +
-				 be32_to_cpu(pointers[2]) * 4,
-				 tx_data, sizeof(tx_data), 0);
-	if (err < 0 || (tx_data[0] && tx_data[3])) {
-		err = -ENODEV;
-		goto end;
-	}
-
 	/*
 	 * Check that the implemented DICE driver specification major version
 	 * number matches.
@@ -142,6 +131,8 @@ static int dice_read_mode_params(struct snd_dice *dice, unsigned int mode)
 	int err;
 
 	if (highest_supported_mode_rate(dice, mode, &rate) < 0) {
+		dice->tx_channels[mode] = 0;
+		dice->tx_midi_ports[mode] = 0;
 		dice->rx_channels[mode] = 0;
 		dice->rx_midi_ports[mode] = 0;
 		return 0;
@@ -151,6 +142,14 @@ static int dice_read_mode_params(struct snd_dice *dice, unsigned int mode)
 	if (err < 0)
 		return err;
 
+	err = snd_dice_transaction_read_tx(dice, TX_NUMBER_AUDIO,
+					   values, sizeof(values));
+	if (err < 0)
+		return err;
+
+	dice->tx_channels[mode]   = be32_to_cpu(values[0]);
+	dice->tx_midi_ports[mode] = be32_to_cpu(values[1]);
+
 	err = snd_dice_transaction_read_rx(dice, RX_NUMBER_AUDIO,
 					   values, sizeof(values));
 	if (err < 0)
@@ -280,13 +279,13 @@ static int dice_probe(struct fw_unit *unit, const struct ieee1394_device_id *id)
 
 	snd_dice_create_proc(dice);
 
-	err = snd_dice_stream_init(dice);
+	err = snd_dice_stream_init_duplex(dice);
 	if (err < 0)
 		goto error;
 
 	err = snd_card_register(card);
 	if (err < 0) {
-		snd_dice_stream_destroy(dice);
+		snd_dice_stream_destroy_duplex(dice);
 		goto error;
 	}
 
@@ -304,7 +303,7 @@ static void dice_remove(struct fw_unit *unit)
 
 	snd_card_disconnect(dice->card);
 
-	snd_dice_stream_destroy(dice);
+	snd_dice_stream_destroy_duplex(dice);
 
 	snd_card_free_when_closed(dice->card);
 }
@@ -316,7 +315,7 @@ static void dice_update(struct fw_unit *unit)
 	/* The handler address register becomes initialized. */
 	snd_dice_transaction_reinit(dice);
 
-	snd_dice_stream_update(dice);
+	snd_dice_stream_update_duplex(dice);
 }
 
 #define DICE_INTERFACE	0x000001
diff --git a/sound/firewire/dice/dice.h b/sound/firewire/dice/dice.h
index 8be530f..a868485 100644
--- a/sound/firewire/dice/dice.h
+++ b/sound/firewire/dice/dice.h
@@ -52,18 +52,28 @@ struct snd_dice {
 	unsigned int rsrv_offset;
 
 	unsigned int clock_caps;
+	unsigned int tx_channels[3];
 	unsigned int rx_channels[3];
+	unsigned int tx_midi_ports[3];
 	unsigned int rx_midi_ports[3];
+
 	struct fw_address_handler notification_handler;
 	int owner_generation;
+	u32 notification_bits;
+
+	/* For uapi */
 	int dev_lock_count; /* > 0 driver, < 0 userspace */
 	bool dev_lock_changed;
-	bool global_enabled;
-	struct completion clock_accepted;
 	wait_queue_head_t hwdep_wait;
-	u32 notification_bits;
+
+	/* For streaming */
+	struct fw_iso_resources tx_resources;
 	struct fw_iso_resources rx_resources;
+	struct amdtp_stream tx_stream;
 	struct amdtp_stream rx_stream;
+	bool global_enabled;
+	struct completion clock_accepted;
+	atomic_t substreams_counter;
 };
 
 enum snd_dice_addr_type {
@@ -160,11 +170,11 @@ extern const unsigned int snd_dice_rates[SND_DICE_RATES_COUNT];
 int snd_dice_stream_get_rate_mode(struct snd_dice *dice,
 				  unsigned int rate, unsigned int *mode);
 
-int snd_dice_stream_start(struct snd_dice *dice, unsigned int rate);
-void snd_dice_stream_stop(struct snd_dice *dice);
-int snd_dice_stream_init(struct snd_dice *dice);
-void snd_dice_stream_destroy(struct snd_dice *dice);
-void snd_dice_stream_update(struct snd_dice *dice);
+int snd_dice_stream_start_duplex(struct snd_dice *dice, unsigned int rate);
+void snd_dice_stream_stop_duplex(struct snd_dice *dice);
+int snd_dice_stream_init_duplex(struct snd_dice *dice);
+void snd_dice_stream_destroy_duplex(struct snd_dice *dice);
+void snd_dice_stream_update_duplex(struct snd_dice *dice);
 
 int snd_dice_stream_lock_try(struct snd_dice *dice);
 void snd_dice_stream_lock_release(struct snd_dice *dice);
-- 
1.9.1