[PATCH v1 0/2] NVIDIA Tegra serial driver improvements

[PATCH v1 0/2] NVIDIA Tegra serial driver improvements

[Dmitry Osipenko]

Hello,

This series is a complement to [1], which changed APB DMA driver's runtime
PM to make DMA clock enabled only during of actual transfer activity.
Unfortunately turned out that it wasn't enough because serial driver keeps
RX DMA active all the time in practice and that's what the first patch of
this series addresses. Secondly, I noticed that DMA synchronization could
be improved, it is also addressed now. Please review and apply, thanks in
advance!

Dmitry Osipenko (2):
  tty: serial: tegra: Activate RX DMA transfer by request
  tty: serial: tegra: Optimize DMA buffer synchronization

 drivers/tty/serial/serial-tegra.c | 94 ++++++++++++++++++-------------
 1 file changed, 55 insertions(+), 39 deletions(-)

-- 
2.24.0

[PATCH v1 1/2] tty: serial: tegra: Activate RX DMA transfer by request

[Dmitry Osipenko]

This allows DMA engine to go into runtime-suspended mode whenever there is
no data to receive, instead of keeping DMA active all the time while TTY
is opened (i.e. permanently active in practice, like in the case of UART
Bluetooth).

Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/tty/serial/serial-tegra.c | 78 ++++++++++++++++++-------------
 1 file changed, 46 insertions(+), 32 deletions(-)

diff --git a/drivers/tty/serial/serial-tegra.c b/drivers/tty/serial/serial-tegra.c
index b6ace6290e23..3b495e7c9534 100644
--- a/drivers/tty/serial/serial-tegra.c
+++ b/drivers/tty/serial/serial-tegra.c
@@ -141,6 +141,7 @@ struct tegra_uart_port {
 	int					configured_rate;
 	bool					use_rx_pio;
 	bool					use_tx_pio;
+	bool					rx_dma_active;
 };
 
 static void tegra_uart_start_next_tx(struct tegra_uart_port *tup);
@@ -731,6 +732,7 @@ static void tegra_uart_rx_dma_complete(void *args)
 	if (tup->rts_active)
 		set_rts(tup, false);
 
+	tup->rx_dma_active = false;
 	tegra_uart_rx_buffer_push(tup, 0);
 	tegra_uart_start_rx_dma(tup);
 
@@ -742,18 +744,27 @@ static void tegra_uart_rx_dma_complete(void *args)
 	spin_unlock_irqrestore(&u->lock, flags);
 }
 
-static void tegra_uart_handle_rx_dma(struct tegra_uart_port *tup)
+static void tegra_uart_terminate_rx_dma(struct tegra_uart_port *tup)
 {
 	struct dma_tx_state state;
 
-	/* Deactivate flow control to stop sender */
-	if (tup->rts_active)
-		set_rts(tup, false);
+	if (!tup->rx_dma_active)
+		return;
 
 	dmaengine_terminate_all(tup->rx_dma_chan);
 	dmaengine_tx_status(tup->rx_dma_chan, tup->rx_cookie, &state);
+
 	tegra_uart_rx_buffer_push(tup, state.residue);
-	tegra_uart_start_rx_dma(tup);
+	tup->rx_dma_active = false;
+}
+
+static void tegra_uart_handle_rx_dma(struct tegra_uart_port *tup)
+{
+	/* Deactivate flow control to stop sender */
+	if (tup->rts_active)
+		set_rts(tup, false);
+
+	tegra_uart_terminate_rx_dma(tup);
 
 	if (tup->rts_active)
 		set_rts(tup, true);
@@ -763,6 +774,9 @@ static int tegra_uart_start_rx_dma(struct tegra_uart_port *tup)
 {
 	unsigned int count = TEGRA_UART_RX_DMA_BUFFER_SIZE;
 
+	if (tup->rx_dma_active)
+		return 0;
+
 	tup->rx_dma_desc = dmaengine_prep_slave_single(tup->rx_dma_chan,
 				tup->rx_dma_buf_phys, count, DMA_DEV_TO_MEM,
 				DMA_PREP_INTERRUPT);
@@ -771,6 +785,7 @@ static int tegra_uart_start_rx_dma(struct tegra_uart_port *tup)
 		return -EIO;
 	}
 
+	tup->rx_dma_active = true;
 	tup->rx_dma_desc->callback = tegra_uart_rx_dma_complete;
 	tup->rx_dma_desc->callback_param = tup;
 	dma_sync_single_for_device(tup->uport.dev, tup->rx_dma_buf_phys,
@@ -820,6 +835,7 @@ static irqreturn_t tegra_uart_isr(int irq, void *data)
 	struct uart_port *u = &tup->uport;
 	unsigned long iir;
 	unsigned long ier;
+	bool is_rx_start = false;
 	bool is_rx_int = false;
 	unsigned long flags;
 
@@ -832,10 +848,12 @@ static irqreturn_t tegra_uart_isr(int irq, void *data)
 				if (tup->rx_in_progress) {
 					ier = tup->ier_shadow;
 					ier |= (UART_IER_RLSI | UART_IER_RTOIE |
-						TEGRA_UART_IER_EORD);
+						TEGRA_UART_IER_EORD | UART_IER_RDI);
 					tup->ier_shadow = ier;
 					tegra_uart_write(tup, ier, UART_IER);
 				}
+			} else if (is_rx_start) {
+				tegra_uart_start_rx_dma(tup);
 			}
 			spin_unlock_irqrestore(&u->lock, flags);
 			return IRQ_HANDLED;
@@ -854,17 +872,23 @@ static irqreturn_t tegra_uart_isr(int irq, void *data)
 
 		case 4: /* End of data */
 		case 6: /* Rx timeout */
-		case 2: /* Receive */
-			if (!tup->use_rx_pio && !is_rx_int) {
-				is_rx_int = true;
+			if (!tup->use_rx_pio) {
+				is_rx_int = tup->rx_in_progress;
 				/* Disable Rx interrupts */
 				ier = tup->ier_shadow;
-				ier |= UART_IER_RDI;
-				tegra_uart_write(tup, ier, UART_IER);
 				ier &= ~(UART_IER_RDI | UART_IER_RLSI |
 					UART_IER_RTOIE | TEGRA_UART_IER_EORD);
 				tup->ier_shadow = ier;
 				tegra_uart_write(tup, ier, UART_IER);
+				break;
+			}
+			/* Fall through */
+		case 2: /* Receive */
+			if (!tup->use_rx_pio) {
+				is_rx_start = tup->rx_in_progress;
+				tup->ier_shadow  &= ~UART_IER_RDI;
+				tegra_uart_write(tup, tup->ier_shadow,
+						 UART_IER);
 			} else {
 				do_handle_rx_pio(tup);
 			}
@@ -886,7 +910,6 @@ static void tegra_uart_stop_rx(struct uart_port *u)
 {
 	struct tegra_uart_port *tup = to_tegra_uport(u);
 	struct tty_port *port = &tup->uport.state->port;
-	struct dma_tx_state state;
 	unsigned long ier;
 
 	if (tup->rts_active)
@@ -903,13 +926,11 @@ static void tegra_uart_stop_rx(struct uart_port *u)
 	tup->ier_shadow = ier;
 	tegra_uart_write(tup, ier, UART_IER);
 	tup->rx_in_progress = 0;
-	if (tup->rx_dma_chan && !tup->use_rx_pio) {
-		dmaengine_terminate_all(tup->rx_dma_chan);
-		dmaengine_tx_status(tup->rx_dma_chan, tup->rx_cookie, &state);
-		tegra_uart_rx_buffer_push(tup, state.residue);
-	} else {
+
+	if (!tup->use_rx_pio)
+		tegra_uart_terminate_rx_dma(tup);
+	else
 		tegra_uart_handle_rx_pio(tup, port);
-	}
 }
 
 static void tegra_uart_hw_deinit(struct tegra_uart_port *tup)
@@ -1052,12 +1073,6 @@ static int tegra_uart_hw_init(struct tegra_uart_port *tup)
 		tup->lcr_shadow = TEGRA_UART_DEFAULT_LSR;
 		tup->fcr_shadow |= UART_FCR_DMA_SELECT;
 		tegra_uart_write(tup, tup->fcr_shadow, UART_FCR);
-
-		ret = tegra_uart_start_rx_dma(tup);
-		if (ret < 0) {
-			dev_err(tup->uport.dev, "Not able to start Rx DMA\n");
-			return ret;
-		}
 	} else {
 		tegra_uart_write(tup, tup->fcr_shadow, UART_FCR);
 	}
@@ -1067,10 +1082,6 @@ static int tegra_uart_hw_init(struct tegra_uart_port *tup)
 	 * Enable IE_RXS for the receive status interrupts like line errros.
 	 * Enable IE_RX_TIMEOUT to get the bytes which cannot be DMA'd.
 	 *
-	 * If using DMA mode, enable EORD instead of receive interrupt which
-	 * will interrupt after the UART is done with the receive instead of
-	 * the interrupt when the FIFO "threshold" is reached.
-	 *
 	 * EORD is different interrupt than RX_TIMEOUT - RX_TIMEOUT occurs when
 	 * the DATA is sitting in the FIFO and couldn't be transferred to the
 	 * DMA as the DMA size alignment (4 bytes) is not met. EORD will be
@@ -1081,11 +1092,14 @@ static int tegra_uart_hw_init(struct tegra_uart_port *tup)
 	 * both the EORD as well as RX_TIMEOUT - SW sees RX_TIMEOUT first
 	 * then the EORD.
 	 */
+	tup->ier_shadow = UART_IER_RLSI | UART_IER_RTOIE | UART_IER_RDI;
+
+	/*
+	 * If using DMA mode, enable EORD interrupt to notify about RX
+	 * completion.
+	 */
 	if (!tup->use_rx_pio)
-		tup->ier_shadow = UART_IER_RLSI | UART_IER_RTOIE |
-			TEGRA_UART_IER_EORD;
-	else
-		tup->ier_shadow = UART_IER_RLSI | UART_IER_RTOIE | UART_IER_RDI;
+		tup->ier_shadow |= TEGRA_UART_IER_EORD;
 
 	tegra_uart_write(tup, tup->ier_shadow, UART_IER);
 	return 0;
-- 
2.24.0

[PATCH v1 2/2] tty: serial: tegra: Optimize DMA buffer synchronization

[Dmitry Osipenko]

Synchronize only the dirty part of DMA buffer in order to avoid
unnecessary overhead of syncing of the clean part, which is the case
of every serial DMA transfer in practice.

Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/tty/serial/serial-tegra.c | 16 +++++++++-------
 1 file changed, 9 insertions(+), 7 deletions(-)

diff --git a/drivers/tty/serial/serial-tegra.c b/drivers/tty/serial/serial-tegra.c
index 3b495e7c9534..33034b852a51 100644
--- a/drivers/tty/serial/serial-tegra.c
+++ b/drivers/tty/serial/serial-tegra.c
@@ -534,11 +534,12 @@ static int tegra_uart_start_tx_dma(struct tegra_uart_port *tup,
 	struct circ_buf *xmit = &tup->uport.state->xmit;
 	dma_addr_t tx_phys_addr;
 
-	dma_sync_single_for_device(tup->uport.dev, tup->tx_dma_buf_phys,
-				UART_XMIT_SIZE, DMA_TO_DEVICE);
-
 	tup->tx_bytes = count & ~(0xF);
 	tx_phys_addr = tup->tx_dma_buf_phys + xmit->tail;
+
+	dma_sync_single_for_device(tup->uport.dev, tx_phys_addr,
+				   tup->tx_bytes, DMA_TO_DEVICE);
+
 	tup->tx_dma_desc = dmaengine_prep_slave_single(tup->tx_dma_chan,
 				tx_phys_addr, tup->tx_bytes, DMA_MEM_TO_DEV,
 				DMA_PREP_INTERRUPT);
@@ -680,7 +681,7 @@ static void tegra_uart_copy_rx_to_tty(struct tegra_uart_port *tup,
 		return;
 
 	dma_sync_single_for_cpu(tup->uport.dev, tup->rx_dma_buf_phys,
-				TEGRA_UART_RX_DMA_BUFFER_SIZE, DMA_FROM_DEVICE);
+				count, DMA_FROM_DEVICE);
 	copied = tty_insert_flip_string(tty,
 			((unsigned char *)(tup->rx_dma_buf_virt)), count);
 	if (copied != count) {
@@ -688,7 +689,7 @@ static void tegra_uart_copy_rx_to_tty(struct tegra_uart_port *tup,
 		dev_err(tup->uport.dev, "RxData copy to tty layer failed\n");
 	}
 	dma_sync_single_for_device(tup->uport.dev, tup->rx_dma_buf_phys,
-				TEGRA_UART_RX_DMA_BUFFER_SIZE, DMA_TO_DEVICE);
+				   count, DMA_TO_DEVICE);
 }
 
 static void tegra_uart_rx_buffer_push(struct tegra_uart_port *tup,
@@ -788,8 +789,6 @@ static int tegra_uart_start_rx_dma(struct tegra_uart_port *tup)
 	tup->rx_dma_active = true;
 	tup->rx_dma_desc->callback = tegra_uart_rx_dma_complete;
 	tup->rx_dma_desc->callback_param = tup;
-	dma_sync_single_for_device(tup->uport.dev, tup->rx_dma_buf_phys,
-				count, DMA_TO_DEVICE);
 	tup->rx_bytes_requested = count;
 	tup->rx_cookie = dmaengine_submit(tup->rx_dma_desc);
 	dma_async_issue_pending(tup->rx_dma_chan);
@@ -1154,6 +1153,9 @@ static int tegra_uart_dma_channel_allocate(struct tegra_uart_port *tup,
 			dma_release_channel(dma_chan);
 			return -ENOMEM;
 		}
+		dma_sync_single_for_device(tup->uport.dev, dma_phys,
+					   TEGRA_UART_RX_DMA_BUFFER_SIZE,
+					   DMA_TO_DEVICE);
 		dma_sconfig.src_addr = tup->uport.mapbase;
 		dma_sconfig.src_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
 		dma_sconfig.src_maxburst = tup->cdata->max_dma_burst_bytes;
-- 
2.24.0