[PATCH v2 0/9] Improve Rx/Tx DMA implementation

[PATCH v2 0/9] Improve Rx/Tx DMA implementation

[Bhuvanchandra DV]

Changes since v1:
Split suspend/resume patch.

Bhuvanchandra DV (5):
  tty: serial: fsl_lpuart: Fix broken 8m/s1 support
  tty: serial: fsl-lpuart: Use cyclic DMA for Rx
  tty: serial: fsl-lpuart: Use scatter/gather DMA for Tx
  tty: serial: fsl-lpuart: Update suspend/resume for DMA mode
  tty: serial: fsl_lpuart: Add support for RS-485

Stefan Agner (4):
  tty: serial: fsl_lpuart: consider TX FIFO too in tx_empty
  clk: imx: vf610: Disable automatic clock gating for lpuart in LPSTOP mode
  tty: serial: fsl_lpuart: support suspend/resume
  tty: serial: fsl_lpuart: fix clearing of receive flag

 drivers/clk/imx/clk-vf610.c     |  12 +-
 drivers/tty/serial/fsl_lpuart.c | 810 +++++++++++++++++++++++-----------------
 2 files changed, 482 insertions(+), 340 deletions(-)

-- 
2.9.0

[PATCH v2 1/9] tty: serial: fsl_lpuart: consider TX FIFO too in tx_empty

[Bhuvanchandra DV]

From: Stefan Agner <stefan@agner.ch>

Currently the tx_empty callback only considers the Transmit Complete Flag (TC).
The reference manual is not quite clear if the TC flag covers the TX FIFO too.
Debug prints on real hardware have shown that from time to time the TC flag is
asserted (indicating Transmitter idle) while there are still data in the
TX FIFO. Hence, in this case the serial core will call the shutdown callback
even though there are data remaining in the TX FIFO buffers.

Avoid early shutdowns by considering the TX FIFO empty flag too. Also avoid
theoretical race conditions between DMA and the driver by checking whether the
TX DMA is in progress too.

Signed-off-by: Stefan Agner <stefan@agner.ch>
Signed-off-by: Bhuvanchandra DV <bhuvanchandra.dv@toradex.com>
---
 drivers/tty/serial/fsl_lpuart.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)

diff --git a/drivers/tty/serial/fsl_lpuart.c b/drivers/tty/serial/fsl_lpuart.c
index 3d79003..fabfa7e 100644
--- a/drivers/tty/serial/fsl_lpuart.c
+++ b/drivers/tty/serial/fsl_lpuart.c
@@ -810,8 +810,18 @@ static irqreturn_t lpuart32_int(int irq, void *dev_id)
 /* return TIOCSER_TEMT when transmitter is not busy */
 static unsigned int lpuart_tx_empty(struct uart_port *port)
 {
-	return (readb(port->membase + UARTSR1) & UARTSR1_TC) ?
-		TIOCSER_TEMT : 0;
+	struct lpuart_port *sport = container_of(port,
+			struct lpuart_port, port);
+	unsigned char sr1 = readb(port->membase + UARTSR1);
+	unsigned char sfifo = readb(port->membase + UARTSFIFO);
+
+	if (sport->dma_tx_in_progress)
+		return 0;
+
+	if (sr1 & UARTSR1_TC && sfifo & UARTSFIFO_TXEMPT)
+		return TIOCSER_TEMT;
+
+	return 0;
 }
 
 static unsigned int lpuart32_tx_empty(struct uart_port *port)
-- 
2.9.0

[PATCH v2 2/9] clk: imx: vf610: Disable automatic clock gating for lpuart in LPSTOP mode

[Bhuvanchandra DV]

From: Stefan Agner <stefan@agner.ch>

In order to allow wake support in STOP sleep mode, clocks are needed. Use
imx_clk_gate2_cgr to disable automatic clock gating in low power mode STOP.
This allows to enable wake by UART using:
echo enabled > /sys/class/tty/ttyLP0/power/wakeup

However, if wake is not enabled, the driver should disable the clocks explicitly
to save power.

Signed-off-by: Stefan Agner <stefan@agner.ch>
Signed-off-by: Bhuvanchandra DV <bhuvanchandra.dv@toradex.com>
---
 drivers/clk/imx/clk-vf610.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/drivers/clk/imx/clk-vf610.c b/drivers/clk/imx/clk-vf610.c
index 3a1f244..0476353 100644
--- a/drivers/clk/imx/clk-vf610.c
+++ b/drivers/clk/imx/clk-vf610.c
@@ -315,12 +315,12 @@ static void __init vf610_clocks_init(struct device_node *ccm_node)
 
 	clk[VF610_CLK_PIT] = imx_clk_gate2("pit", "ipg_bus", CCM_CCGR1, CCM_CCGRx_CGn(7));
 
-	clk[VF610_CLK_UART0] = imx_clk_gate2("uart0", "ipg_bus", CCM_CCGR0, CCM_CCGRx_CGn(7));
-	clk[VF610_CLK_UART1] = imx_clk_gate2("uart1", "ipg_bus", CCM_CCGR0, CCM_CCGRx_CGn(8));
-	clk[VF610_CLK_UART2] = imx_clk_gate2("uart2", "ipg_bus", CCM_CCGR0, CCM_CCGRx_CGn(9));
-	clk[VF610_CLK_UART3] = imx_clk_gate2("uart3", "ipg_bus", CCM_CCGR0, CCM_CCGRx_CGn(10));
-	clk[VF610_CLK_UART4] = imx_clk_gate2("uart4", "ipg_bus", CCM_CCGR6, CCM_CCGRx_CGn(9));
-	clk[VF610_CLK_UART5] = imx_clk_gate2("uart5", "ipg_bus", CCM_CCGR6, CCM_CCGRx_CGn(10));
+	clk[VF610_CLK_UART0] = imx_clk_gate2_cgr("uart0", "ipg_bus", CCM_CCGR0, CCM_CCGRx_CGn(7), 0x2);
+	clk[VF610_CLK_UART1] = imx_clk_gate2_cgr("uart1", "ipg_bus", CCM_CCGR0, CCM_CCGRx_CGn(8), 0x2);
+	clk[VF610_CLK_UART2] = imx_clk_gate2_cgr("uart2", "ipg_bus", CCM_CCGR0, CCM_CCGRx_CGn(9), 0x2);
+	clk[VF610_CLK_UART3] = imx_clk_gate2_cgr("uart3", "ipg_bus", CCM_CCGR0, CCM_CCGRx_CGn(10), 0x2);
+	clk[VF610_CLK_UART4] = imx_clk_gate2_cgr("uart4", "ipg_bus", CCM_CCGR6, CCM_CCGRx_CGn(9), 0x2);
+	clk[VF610_CLK_UART5] = imx_clk_gate2_cgr("uart5", "ipg_bus", CCM_CCGR6, CCM_CCGRx_CGn(10), 0x2);
 
 	clk[VF610_CLK_I2C0] = imx_clk_gate2("i2c0", "ipg_bus", CCM_CCGR4, CCM_CCGRx_CGn(6));
 	clk[VF610_CLK_I2C1] = imx_clk_gate2("i2c1", "ipg_bus", CCM_CCGR4, CCM_CCGRx_CGn(7));
-- 
2.9.0

[PATCH v2 3/9] tty: serial: fsl_lpuart: support suspend/resume

[Bhuvanchandra DV]

From: Stefan Agner <stefan@agner.ch>

Add suspend/resume support.

Signed-off-by: Stefan Agner <stefan@agner.ch>
Signed-off-by: Bhuvanchandra DV <bhuvanchandra.dv@toradex.com>
---
 drivers/tty/serial/fsl_lpuart.c | 16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)

diff --git a/drivers/tty/serial/fsl_lpuart.c b/drivers/tty/serial/fsl_lpuart.c
index fabfa7e..75a2098 100644
--- a/drivers/tty/serial/fsl_lpuart.c
+++ b/drivers/tty/serial/fsl_lpuart.c
@@ -483,9 +483,8 @@ static void lpuart_dma_rx_complete(void *arg)
 	spin_unlock_irqrestore(&sport->port.lock, flags);
 }
 
-static void lpuart_timer_func(unsigned long data)
+static void lpuart_dma_rx_terminate(struct lpuart_port *sport)
 {
-	struct lpuart_port *sport = (struct lpuart_port *)data;
 	struct tty_port *port = &sport->port.state->port;
 	struct dma_tx_state state;
 	unsigned long flags;
@@ -510,6 +509,11 @@ static void lpuart_timer_func(unsigned long data)
 	spin_unlock_irqrestore(&sport->port.lock, flags);
 }
 
+static void lpuart_timer_func(unsigned long data)
+{
+	lpuart_dma_rx_terminate((struct lpuart_port *)data);
+}
+
 static inline void lpuart_prepare_rx(struct lpuart_port *sport)
 {
 	unsigned long flags;
@@ -1925,7 +1929,12 @@ static int lpuart_suspend(struct device *dev)
 		writeb(temp, sport->port.membase + UARTCR2);
 	}
 
+	if (sport->dma_rx_in_progress)
+		lpuart_dma_rx_terminate(sport);
+
 	uart_suspend_port(&lpuart_reg, &sport->port);
+	if (sport->port.suspended && !sport->port.irq_wake)
+		clk_disable_unprepare(sport->clk);
 
 	return 0;
 }
@@ -1935,6 +1944,9 @@ static int lpuart_resume(struct device *dev)
 	struct lpuart_port *sport = dev_get_drvdata(dev);
 	unsigned long temp;
 
+	if (sport->port.suspended && !sport->port.irq_wake)
+		clk_prepare_enable(sport->clk);
+
 	if (sport->lpuart32) {
 		lpuart32_setup_watermark(sport);
 		temp = lpuart32_read(sport->port.membase + UARTCTRL);
-- 
2.9.0

[PATCH v2 4/9] tty: serial: fsl_lpuart: fix clearing of receive flag

[Bhuvanchandra DV]

From: Stefan Agner <stefan@agner.ch>

Commit 8e4934c6d6c6 ("tty: serial: fsl_lpuart: clear receive flag on FIFO
flush") implemented clearing of the receive flag by reading the status register
only. It turned out that even though we flush the FIFO afterwards, a explicit
read of the data register is still required.

This leads to a FIFO underrun. To avoid this, follow the advice in the overrun
"Operation section": Unconditionally clear RXUF after using RXFLUSH.

Signed-off-by: Stefan Agner <stefan@agner.ch>
Signed-off-by: Bhuvanchandra DV <bhuvanchandra.dv@toradex.com>
---
 drivers/tty/serial/fsl_lpuart.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/tty/serial/fsl_lpuart.c b/drivers/tty/serial/fsl_lpuart.c
index 75a2098..97c1fda 100644
--- a/drivers/tty/serial/fsl_lpuart.c
+++ b/drivers/tty/serial/fsl_lpuart.c
@@ -935,13 +935,16 @@ static void lpuart_setup_watermark(struct lpuart_port *sport)
 	writeb(val | UARTPFIFO_TXFE | UARTPFIFO_RXFE,
 			sport->port.membase + UARTPFIFO);
 
-	/* explicitly clear RDRF */
-	readb(sport->port.membase + UARTSR1);
-
 	/* flush Tx and Rx FIFO */
 	writeb(UARTCFIFO_TXFLUSH | UARTCFIFO_RXFLUSH,
 			sport->port.membase + UARTCFIFO);
 
+	/* explicitly clear RDRF */
+	if (readb(sport->port.membase + UARTSR1) & UARTSR1_RDRF) {
+		readb(sport->port.membase + UARTDR);
+		writeb(UARTSFIFO_RXUF, sport->port.membase + UARTSFIFO);
+	}
+
 	writeb(0, sport->port.membase + UARTTWFIFO);
 	writeb(1, sport->port.membase + UARTRWFIFO);
 
-- 
2.9.0

[PATCH v2 5/9] tty: serial: fsl_lpuart: Fix broken 8m/s1 support

[Bhuvanchandra DV]

By default the driver always configure the mode as 8s1 even when 8m1 mode is
selected. Fix this by adding support to control the space/mark bit.

Signed-off-by: Bhuvanchandra DV <bhuvanchandra.dv@toradex.com>
---
 drivers/tty/serial/fsl_lpuart.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/drivers/tty/serial/fsl_lpuart.c b/drivers/tty/serial/fsl_lpuart.c
index 97c1fda..615f191 100644
--- a/drivers/tty/serial/fsl_lpuart.c
+++ b/drivers/tty/serial/fsl_lpuart.c
@@ -1220,13 +1220,14 @@ lpuart_set_termios(struct uart_port *port, struct ktermios *termios,
 {
 	struct lpuart_port *sport = container_of(port, struct lpuart_port, port);
 	unsigned long flags;
-	unsigned char cr1, old_cr1, old_cr2, cr4, bdh, modem;
+	unsigned char cr1, old_cr1, old_cr2, cr3, cr4, bdh, modem;
 	unsigned int  baud;
 	unsigned int old_csize = old ? old->c_cflag & CSIZE : CS8;
 	unsigned int sbr, brfa;
 
 	cr1 = old_cr1 = readb(sport->port.membase + UARTCR1);
 	old_cr2 = readb(sport->port.membase + UARTCR2);
+	cr3 = readb(sport->port.membase + UARTCR3);
 	cr4 = readb(sport->port.membase + UARTCR4);
 	bdh = readb(sport->port.membase + UARTBDH);
 	modem = readb(sport->port.membase + UARTMODEM);
@@ -1274,7 +1275,10 @@ lpuart_set_termios(struct uart_port *port, struct ktermios *termios,
 	if ((termios->c_cflag & PARENB)) {
 		if (termios->c_cflag & CMSPAR) {
 			cr1 &= ~UARTCR1_PE;
-			cr1 |= UARTCR1_M;
+			if (termios->c_cflag & PARODD)
+				cr3 |= UARTCR3_T8;
+			else
+				cr3 &= ~UARTCR3_T8;
 		} else {
 			cr1 |= UARTCR1_PE;
 			if ((termios->c_cflag & CSIZE) == CS8)
@@ -1342,6 +1346,7 @@ lpuart_set_termios(struct uart_port *port, struct ktermios *termios,
 	writeb(cr4 | brfa, sport->port.membase + UARTCR4);
 	writeb(bdh, sport->port.membase + UARTBDH);
 	writeb(sbr & 0xFF, sport->port.membase + UARTBDL);
+	writeb(cr3, sport->port.membase + UARTCR3);
 	writeb(cr1, sport->port.membase + UARTCR1);
 	writeb(modem, sport->port.membase + UARTMODEM);
 
-- 
2.9.0

[PATCH v2 6/9] tty: serial: fsl-lpuart: Use cyclic DMA for Rx

[Bhuvanchandra DV]

The initial approach of DMA implementatin for RX is inefficient due to switching
from PIO to DMA, this leads to overruns especially on instances with the smaller
FIFO. To address these issues this patch uses a cyclic DMA for receiver path.

Some part of the code is borrowed from atmel serial driver.

Signed-off-by: Bhuvanchandra DV <bhuvanchandra.dv@toradex.com>
---
 drivers/tty/serial/fsl_lpuart.c | 483 +++++++++++++++++++++-------------------
 1 file changed, 258 insertions(+), 225 deletions(-)

diff --git a/drivers/tty/serial/fsl_lpuart.c b/drivers/tty/serial/fsl_lpuart.c
index 615f191..51d2b5a 100644
--- a/drivers/tty/serial/fsl_lpuart.c
+++ b/drivers/tty/serial/fsl_lpuart.c
@@ -224,7 +224,8 @@
 #define UARTWATER_TXWATER_OFF	0
 #define UARTWATER_RXWATER_OFF	16
 
-#define FSL_UART_RX_DMA_BUFFER_SIZE	64
+/* Rx DMA timeout in ms, which is used to calculate Rx ring buffer size */
+#define DMA_RX_TIMEOUT		(10)
 
 #define DRIVER_NAME	"fsl-lpuart"
 #define DEV_NAME	"ttyLP"
@@ -244,17 +245,17 @@ struct lpuart_port {
 	struct dma_async_tx_descriptor  *dma_tx_desc;
 	struct dma_async_tx_descriptor  *dma_rx_desc;
 	dma_addr_t		dma_tx_buf_bus;
-	dma_addr_t		dma_rx_buf_bus;
 	dma_cookie_t		dma_tx_cookie;
 	dma_cookie_t		dma_rx_cookie;
 	unsigned char		*dma_tx_buf_virt;
-	unsigned char		*dma_rx_buf_virt;
 	unsigned int		dma_tx_bytes;
 	unsigned int		dma_rx_bytes;
 	int			dma_tx_in_progress;
-	int			dma_rx_in_progress;
 	unsigned int		dma_rx_timeout;
 	struct timer_list	lpuart_timer;
+	struct scatterlist	rx_sgl;
+	struct circ_buf		rx_ring;
+	int			rx_dma_rng_buf_len;
 };
 
 static const struct of_device_id lpuart_dt_ids[] = {
@@ -270,7 +271,6 @@ MODULE_DEVICE_TABLE(of, lpuart_dt_ids);
 
 /* Forward declare this for the dma callbacks*/
 static void lpuart_dma_tx_complete(void *arg);
-static void lpuart_dma_rx_complete(void *arg);
 
 static u32 lpuart32_read(void __iomem *addr)
 {
@@ -316,32 +316,6 @@ static void lpuart32_stop_rx(struct uart_port *port)
 	lpuart32_write(temp & ~UARTCTRL_RE, port->membase + UARTCTRL);
 }
 
-static void lpuart_copy_rx_to_tty(struct lpuart_port *sport,
-		struct tty_port *tty, int count)
-{
-	int copied;
-
-	sport->port.icount.rx += count;
-
-	if (!tty) {
-		dev_err(sport->port.dev, "No tty port\n");
-		return;
-	}
-
-	dma_sync_single_for_cpu(sport->port.dev, sport->dma_rx_buf_bus,
-			FSL_UART_RX_DMA_BUFFER_SIZE, DMA_FROM_DEVICE);
-	copied = tty_insert_flip_string(tty,
-			((unsigned char *)(sport->dma_rx_buf_virt)), count);
-
-	if (copied != count) {
-		WARN_ON(1);
-		dev_err(sport->port.dev, "RxData copy to tty layer failed\n");
-	}
-
-	dma_sync_single_for_device(sport->port.dev, sport->dma_rx_buf_bus,
-			FSL_UART_RX_DMA_BUFFER_SIZE, DMA_TO_DEVICE);
-}
-
 static void lpuart_pio_tx(struct lpuart_port *sport)
 {
 	struct circ_buf *xmit = &sport->port.state->xmit;
@@ -433,28 +407,6 @@ static void lpuart_dma_tx_complete(void *arg)
 	spin_unlock_irqrestore(&sport->port.lock, flags);
 }
 
-static int lpuart_dma_rx(struct lpuart_port *sport)
-{
-	dma_sync_single_for_device(sport->port.dev, sport->dma_rx_buf_bus,
-			FSL_UART_RX_DMA_BUFFER_SIZE, DMA_TO_DEVICE);
-	sport->dma_rx_desc = dmaengine_prep_slave_single(sport->dma_rx_chan,
-			sport->dma_rx_buf_bus, FSL_UART_RX_DMA_BUFFER_SIZE,
-			DMA_DEV_TO_MEM, DMA_PREP_INTERRUPT);
-
-	if (!sport->dma_rx_desc) {
-		dev_err(sport->port.dev, "Not able to get desc for rx\n");
-		return -EIO;
-	}
-
-	sport->dma_rx_desc->callback = lpuart_dma_rx_complete;
-	sport->dma_rx_desc->callback_param = sport;
-	sport->dma_rx_in_progress = 1;
-	sport->dma_rx_cookie = dmaengine_submit(sport->dma_rx_desc);
-	dma_async_issue_pending(sport->dma_rx_chan);
-
-	return 0;
-}
-
 static void lpuart_flush_buffer(struct uart_port *port)
 {
 	struct lpuart_port *sport = container_of(port, struct lpuart_port, port);
@@ -464,73 +416,6 @@ static void lpuart_flush_buffer(struct uart_port *port)
 	}
 }
 
-static void lpuart_dma_rx_complete(void *arg)
-{
-	struct lpuart_port *sport = arg;
-	struct tty_port *port = &sport->port.state->port;
-	unsigned long flags;
-
-	async_tx_ack(sport->dma_rx_desc);
-	mod_timer(&sport->lpuart_timer, jiffies + sport->dma_rx_timeout);
-
-	spin_lock_irqsave(&sport->port.lock, flags);
-
-	sport->dma_rx_in_progress = 0;
-	lpuart_copy_rx_to_tty(sport, port, FSL_UART_RX_DMA_BUFFER_SIZE);
-	tty_flip_buffer_push(port);
-	lpuart_dma_rx(sport);
-
-	spin_unlock_irqrestore(&sport->port.lock, flags);
-}
-
-static void lpuart_dma_rx_terminate(struct lpuart_port *sport)
-{
-	struct tty_port *port = &sport->port.state->port;
-	struct dma_tx_state state;
-	unsigned long flags;
-	unsigned char temp;
-	int count;
-
-	del_timer(&sport->lpuart_timer);
-	dmaengine_pause(sport->dma_rx_chan);
-	dmaengine_tx_status(sport->dma_rx_chan, sport->dma_rx_cookie, &state);
-	dmaengine_terminate_all(sport->dma_rx_chan);
-	count = FSL_UART_RX_DMA_BUFFER_SIZE - state.residue;
-	async_tx_ack(sport->dma_rx_desc);
-
-	spin_lock_irqsave(&sport->port.lock, flags);
-
-	sport->dma_rx_in_progress = 0;
-	lpuart_copy_rx_to_tty(sport, port, count);
-	tty_flip_buffer_push(port);
-	temp = readb(sport->port.membase + UARTCR5);
-	writeb(temp & ~UARTCR5_RDMAS, sport->port.membase + UARTCR5);
-
-	spin_unlock_irqrestore(&sport->port.lock, flags);
-}
-
-static void lpuart_timer_func(unsigned long data)
-{
-	lpuart_dma_rx_terminate((struct lpuart_port *)data);
-}
-
-static inline void lpuart_prepare_rx(struct lpuart_port *sport)
-{
-	unsigned long flags;
-	unsigned char temp;
-
-	spin_lock_irqsave(&sport->port.lock, flags);
-
-	sport->lpuart_timer.expires = jiffies + sport->dma_rx_timeout;
-	add_timer(&sport->lpuart_timer);
-
-	lpuart_dma_rx(sport);
-	temp = readb(sport->port.membase + UARTCR5);
-	writeb(temp | UARTCR5_RDMAS, sport->port.membase + UARTCR5);
-
-	spin_unlock_irqrestore(&sport->port.lock, flags);
-}
-
 static inline void lpuart_transmit_buffer(struct lpuart_port *sport)
 {
 	struct circ_buf *xmit = &sport->port.state->xmit;
@@ -770,18 +655,14 @@ out:
 static irqreturn_t lpuart_int(int irq, void *dev_id)
 {
 	struct lpuart_port *sport = dev_id;
-	unsigned char sts, crdma;
+	unsigned char sts;
 
 	sts = readb(sport->port.membase + UARTSR1);
-	crdma = readb(sport->port.membase + UARTCR5);
 
-	if (sts & UARTSR1_RDRF && !(crdma & UARTCR5_RDMAS)) {
-		if (sport->lpuart_dma_rx_use)
-			lpuart_prepare_rx(sport);
-		else
-			lpuart_rxint(irq, dev_id);
-	}
-	if (sts & UARTSR1_TDRE && !(crdma & UARTCR5_TDMAS)) {
+	if (sts & UARTSR1_RDRF)
+		lpuart_rxint(irq, dev_id);
+
+	if (sts & UARTSR1_TDRE) {
 		if (sport->lpuart_dma_tx_use)
 			lpuart_pio_tx(sport);
 		else
@@ -834,6 +715,209 @@ static unsigned int lpuart32_tx_empty(struct uart_port *port)
 		TIOCSER_TEMT : 0;
 }
 
+static void lpuart_copy_rx_to_tty(struct lpuart_port *sport)
+{
+	struct tty_port *port = &sport->port.state->port;
+	struct dma_tx_state state;
+	enum dma_status dmastat;
+	struct circ_buf *ring = &sport->rx_ring;
+	unsigned long flags;
+	int count = 0;
+	unsigned char sr;
+
+	sr = readb(sport->port.membase + UARTSR1);
+
+	if (sr & (UARTSR1_PE | UARTSR1_FE)) {
+		/* Read DR to clear the error flags */
+		readb(sport->port.membase + UARTDR);
+
+		if (sr & UARTSR1_PE)
+		    sport->port.icount.parity++;
+		else if (sr & UARTSR1_FE)
+		    sport->port.icount.frame++;
+	}
+
+	async_tx_ack(sport->dma_rx_desc);
+
+	spin_lock_irqsave(&sport->port.lock, flags);
+
+	dmastat = dmaengine_tx_status(sport->dma_rx_chan,
+				sport->dma_rx_cookie,
+				&state);
+
+	if (dmastat == DMA_ERROR) {
+		dev_err(sport->port.dev, "Rx DMA transfer failed!\n");
+		spin_unlock_irqrestore(&sport->port.lock, flags);
+		return;
+	}
+
+	/* CPU claims ownership of RX DMA buffer */
+	dma_sync_sg_for_cpu(sport->port.dev, &sport->rx_sgl, 1, DMA_FROM_DEVICE);
+
+	/*
+	 * ring->head points to the end of data already written by the DMA.
+	 * ring->tail points to the beginning of data to be read by the
+	 * framework.
+	 * The current transfer size should not be larger than the dma buffer
+	 * length.
+	 */
+	ring->head = sport->rx_sgl.length - state.residue;
+	BUG_ON(ring->head > sport->rx_sgl.length);
+	/*
+	 * At this point ring->head may point to the first byte right after the
+	 * last byte of the dma buffer:
+	 * 0 <= ring->head <= sport->rx_sgl.length
+	 *
+	 * However ring->tail must always points inside the dma buffer:
+	 * 0 <= ring->tail <= sport->rx_sgl.length - 1
+	 *
+	 * Since we use a ring buffer, we have to handle the case
+	 * where head is lower than tail. In such a case, we first read from
+	 * tail to the end of the buffer then reset tail.
+	 */
+	if (ring->head < ring->tail) {
+		count = sport->rx_sgl.length - ring->tail;
+
+		tty_insert_flip_string(port, ring->buf + ring->tail, count);
+		ring->tail = 0;
+		sport->port.icount.rx += count;
+	}
+
+	/* Finally we read data from tail to head */
+	if (ring->tail < ring->head) {
+		count = ring->head - ring->tail;
+		tty_insert_flip_string(port, ring->buf + ring->tail, count);
+		/* Wrap ring->head if needed */
+		if (ring->head >= sport->rx_sgl.length)
+			ring->head = 0;
+		ring->tail = ring->head;
+		sport->port.icount.rx += count;
+	}
+
+	dma_sync_sg_for_device(sport->port.dev, &sport->rx_sgl, 1,
+			       DMA_FROM_DEVICE);
+
+	spin_unlock_irqrestore(&sport->port.lock, flags);
+
+	tty_flip_buffer_push(port);
+	mod_timer(&sport->lpuart_timer, jiffies + sport->dma_rx_timeout);
+}
+
+static void lpuart_dma_rx_complete(void *arg)
+{
+	struct lpuart_port *sport = arg;
+
+	lpuart_copy_rx_to_tty(sport);
+}
+
+static void lpuart_timer_func(unsigned long data)
+{
+	struct lpuart_port *sport = (struct lpuart_port *)data;
+
+	lpuart_copy_rx_to_tty(sport);
+}
+
+static inline int lpuart_start_rx_dma(struct lpuart_port *sport)
+{
+	struct dma_slave_config dma_rx_sconfig = {};
+	struct circ_buf *ring = &sport->rx_ring;
+	int ret, nent;
+	int bits, baud;
+	struct tty_struct *tty = tty_port_tty_get(&sport->port.state->port);
+	struct ktermios *termios = &tty->termios;
+
+	baud = tty_get_baud_rate(tty);
+
+	bits = (termios->c_cflag & CSIZE) == CS7 ? 9 : 10;
+	if (termios->c_cflag & PARENB)
+		bits++;
+
+	/*
+	 * Calculate length of one DMA buffer size to keep latency below
+	 * 10ms at any baud rate.
+	 */
+	sport->rx_dma_rng_buf_len = (DMA_RX_TIMEOUT * baud /  bits / 1000) * 2;
+	sport->rx_dma_rng_buf_len = (1 << (fls(sport->rx_dma_rng_buf_len) - 1));
+	if (sport->rx_dma_rng_buf_len < 16)
+		sport->rx_dma_rng_buf_len = 16;
+
+	ring->buf = kmalloc(sport->rx_dma_rng_buf_len, GFP_KERNEL);
+	if (!ring->buf) {
+		dev_err(sport->port.dev, "Ring buf alloc failed\n");
+		return -ENOMEM;
+	}
+
+	sg_init_one(&sport->rx_sgl, ring->buf, sport->rx_dma_rng_buf_len);
+	sg_set_buf(&sport->rx_sgl, ring->buf, sport->rx_dma_rng_buf_len);
+	nent = dma_map_sg(sport->port.dev, &sport->rx_sgl, 1, DMA_FROM_DEVICE);
+
+	if (!nent) {
+		dev_err(sport->port.dev, "DMA Rx mapping error\n");
+		return -EINVAL;
+	}
+
+	dma_rx_sconfig.src_addr = sport->port.mapbase + UARTDR;
+	dma_rx_sconfig.src_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
+	dma_rx_sconfig.src_maxburst = 1;
+	dma_rx_sconfig.direction = DMA_DEV_TO_MEM;
+	ret = dmaengine_slave_config(sport->dma_rx_chan, &dma_rx_sconfig);
+
+	if (ret < 0) {
+		dev_err(sport->port.dev,
+				"DMA Rx slave config failed, err = %d\n", ret);
+		return ret;
+	}
+
+	sport->dma_rx_desc = dmaengine_prep_dma_cyclic(sport->dma_rx_chan,
+				 sg_dma_address(&sport->rx_sgl),
+				 sport->rx_sgl.length,
+				 sport->rx_sgl.length / 2,
+				 DMA_DEV_TO_MEM,
+				 DMA_PREP_INTERRUPT);
+	if (!sport->dma_rx_desc) {
+		dev_err(sport->port.dev, "Cannot prepare cyclic DMA\n");
+		return -EFAULT;
+	}
+
+	sport->dma_rx_desc->callback = lpuart_dma_rx_complete;
+	sport->dma_rx_desc->callback_param = sport;
+	sport->dma_rx_cookie = dmaengine_submit(sport->dma_rx_desc);
+	dma_async_issue_pending(sport->dma_rx_chan);
+
+	writeb(readb(sport->port.membase + UARTCR5) | UARTCR5_RDMAS,
+				sport->port.membase + UARTCR5);
+
+	return 0;
+}
+
+static void lpuart_dma_tx_free(struct uart_port *port)
+{
+	struct lpuart_port *sport = container_of(port,
+					struct lpuart_port, port);
+
+	dma_unmap_single(sport->port.dev, sport->dma_tx_buf_bus,
+			UART_XMIT_SIZE, DMA_TO_DEVICE);
+
+	sport->dma_tx_buf_bus = 0;
+	sport->dma_tx_buf_virt = NULL;
+}
+
+static void lpuart_dma_rx_free(struct uart_port *port)
+{
+	struct lpuart_port *sport = container_of(port,
+					struct lpuart_port, port);
+
+	if (sport->dma_rx_chan)
+		dmaengine_terminate_all(sport->dma_rx_chan);
+
+	dma_unmap_sg(sport->port.dev, &sport->rx_sgl, 1, DMA_FROM_DEVICE);
+	kfree(sport->rx_ring.buf);
+	sport->rx_ring.tail = 0;
+	sport->rx_ring.head = 0;
+	sport->dma_rx_desc = NULL;
+	sport->dma_rx_cookie = -EINVAL;
+}
+
 static unsigned int lpuart_get_mctrl(struct uart_port *port)
 {
 	unsigned int temp = 0;
@@ -1015,72 +1099,12 @@ static int lpuart_dma_tx_request(struct uart_port *port)
 	return 0;
 }
 
-static int lpuart_dma_rx_request(struct uart_port *port)
+static void rx_dma_timer_init(struct lpuart_port *sport)
 {
-	struct lpuart_port *sport = container_of(port,
-					struct lpuart_port, port);
-	struct dma_slave_config dma_rx_sconfig;
-	dma_addr_t dma_bus;
-	unsigned char *dma_buf;
-	int ret;
-
-	dma_buf = devm_kzalloc(sport->port.dev,
-				FSL_UART_RX_DMA_BUFFER_SIZE, GFP_KERNEL);
-
-	if (!dma_buf) {
-		dev_err(sport->port.dev, "Dma rx alloc failed\n");
-		return -ENOMEM;
-	}
-
-	dma_bus = dma_map_single(sport->dma_rx_chan->device->dev, dma_buf,
-				FSL_UART_RX_DMA_BUFFER_SIZE, DMA_FROM_DEVICE);
-
-	if (dma_mapping_error(sport->dma_rx_chan->device->dev, dma_bus)) {
-		dev_err(sport->port.dev, "dma_map_single rx failed\n");
-		return -ENOMEM;
-	}
-
-	dma_rx_sconfig.src_addr = sport->port.mapbase + UARTDR;
-	dma_rx_sconfig.src_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
-	dma_rx_sconfig.src_maxburst = 1;
-	dma_rx_sconfig.direction = DMA_DEV_TO_MEM;
-	ret = dmaengine_slave_config(sport->dma_rx_chan, &dma_rx_sconfig);
-
-	if (ret < 0) {
-		dev_err(sport->port.dev,
-				"Dma slave config failed, err = %d\n", ret);
-		return ret;
-	}
-
-	sport->dma_rx_buf_virt = dma_buf;
-	sport->dma_rx_buf_bus = dma_bus;
-	sport->dma_rx_in_progress = 0;
-
-	return 0;
-}
-
-static void lpuart_dma_tx_free(struct uart_port *port)
-{
-	struct lpuart_port *sport = container_of(port,
-					struct lpuart_port, port);
-
-	dma_unmap_single(sport->port.dev, sport->dma_tx_buf_bus,
-			UART_XMIT_SIZE, DMA_TO_DEVICE);
-
-	sport->dma_tx_buf_bus = 0;
-	sport->dma_tx_buf_virt = NULL;
-}
-
-static void lpuart_dma_rx_free(struct uart_port *port)
-{
-	struct lpuart_port *sport = container_of(port,
-					struct lpuart_port, port);
-
-	dma_unmap_single(sport->port.dev, sport->dma_rx_buf_bus,
-			FSL_UART_RX_DMA_BUFFER_SIZE, DMA_FROM_DEVICE);
-
-	sport->dma_rx_buf_bus = 0;
-	sport->dma_rx_buf_virt = NULL;
+		setup_timer(&sport->lpuart_timer, lpuart_timer_func,
+				(unsigned long)sport);
+		sport->lpuart_timer.expires = jiffies + sport->dma_rx_timeout;
+		add_timer(&sport->lpuart_timer);
 }
 
 static int lpuart_startup(struct uart_port *port)
@@ -1101,22 +1125,6 @@ static int lpuart_startup(struct uart_port *port)
 	sport->rxfifo_size = 0x1 << (((temp >> UARTPFIFO_RXSIZE_OFF) &
 		UARTPFIFO_FIFOSIZE_MASK) + 1);
 
-	if (sport->dma_rx_chan && !lpuart_dma_rx_request(port)) {
-		sport->lpuart_dma_rx_use = true;
-		setup_timer(&sport->lpuart_timer, lpuart_timer_func,
-			    (unsigned long)sport);
-	} else
-		sport->lpuart_dma_rx_use = false;
-
-
-	if (sport->dma_tx_chan && !lpuart_dma_tx_request(port)) {
-		sport->lpuart_dma_tx_use = true;
-		temp = readb(port->membase + UARTCR5);
-		temp &= ~UARTCR5_RDMAS;
-		writeb(temp | UARTCR5_TDMAS, port->membase + UARTCR5);
-	} else
-		sport->lpuart_dma_tx_use = false;
-
 	ret = devm_request_irq(port->dev, port->irq, lpuart_int, 0,
 				DRIVER_NAME, sport);
 	if (ret)
@@ -1130,7 +1138,28 @@ static int lpuart_startup(struct uart_port *port)
 	temp |= (UARTCR2_RIE | UARTCR2_TIE | UARTCR2_RE | UARTCR2_TE);
 	writeb(temp, sport->port.membase + UARTCR2);
 
+	if (sport->dma_rx_chan && !lpuart_start_rx_dma(sport)) {
+		/* set Rx DMA timeout */
+		sport->dma_rx_timeout = msecs_to_jiffies(DMA_RX_TIMEOUT);
+		if (!sport->dma_rx_timeout)
+		     sport->dma_rx_timeout = 1;
+
+		sport->lpuart_dma_rx_use = true;
+		rx_dma_timer_init(sport);
+	} else {
+		sport->lpuart_dma_rx_use = false;
+	}
+
+	if (sport->dma_tx_chan && !lpuart_dma_tx_request(port)) {
+		sport->lpuart_dma_tx_use = true;
+		temp = readb(port->membase + UARTCR5);
+		writeb(temp | UARTCR5_TDMAS, port->membase + UARTCR5);
+	} else {
+		sport->lpuart_dma_tx_use = false;
+	}
+
 	spin_unlock_irqrestore(&sport->port.lock, flags);
+
 	return 0;
 }
 
@@ -1187,8 +1216,8 @@ static void lpuart_shutdown(struct uart_port *port)
 	devm_free_irq(port->dev, port->irq, sport);
 
 	if (sport->lpuart_dma_rx_use) {
-		lpuart_dma_rx_free(&sport->port);
 		del_timer_sync(&sport->lpuart_timer);
+		lpuart_dma_rx_free(&sport->port);
 	}
 
 	if (sport->lpuart_dma_tx_use)
@@ -1318,17 +1347,6 @@ lpuart_set_termios(struct uart_port *port, struct ktermios *termios,
 	/* update the per-port timeout */
 	uart_update_timeout(port, termios->c_cflag, baud);
 
-	if (sport->lpuart_dma_rx_use) {
-		/* Calculate delay for 1.5 DMA buffers */
-		sport->dma_rx_timeout = (sport->port.timeout - HZ / 50) *
-					FSL_UART_RX_DMA_BUFFER_SIZE * 3 /
-					sport->rxfifo_size / 2;
-		dev_dbg(port->dev, "DMA Rx t-out %ums, tty t-out %u jiffies\n",
-			sport->dma_rx_timeout * 1000 / HZ, sport->port.timeout);
-		if (sport->dma_rx_timeout < msecs_to_jiffies(20))
-			sport->dma_rx_timeout = msecs_to_jiffies(20);
-	}
-
 	/* wait transmit engin complete */
 	while (!(readb(sport->port.membase + UARTSR1) & UARTSR1_TC))
 		barrier();
@@ -1353,6 +1371,24 @@ lpuart_set_termios(struct uart_port *port, struct ktermios *termios,
 	/* restore control register */
 	writeb(old_cr2, sport->port.membase + UARTCR2);
 
+	/*
+	 * If new baud rate is set, we will also need to update the Ring buffer
+	 * length according to the selected baud rate and restart Rx DMA path.
+	 */
+	if (old) {
+		if (sport->lpuart_dma_rx_use) {
+			del_timer_sync(&sport->lpuart_timer);
+			lpuart_dma_rx_free(&sport->port);
+		}
+
+		if (sport->dma_rx_chan && !lpuart_start_rx_dma(sport)) {
+			sport->lpuart_dma_rx_use = true;
+			rx_dma_timer_init(sport);
+		} else {
+			sport->lpuart_dma_rx_use = false;
+		}
+	}
+
 	spin_unlock_irqrestore(&sport->port.lock, flags);
 }
 
@@ -1937,9 +1973,6 @@ static int lpuart_suspend(struct device *dev)
 		writeb(temp, sport->port.membase + UARTCR2);
 	}
 
-	if (sport->dma_rx_in_progress)
-		lpuart_dma_rx_terminate(sport);
-
 	uart_suspend_port(&lpuart_reg, &sport->port);
 	if (sport->port.suspended && !sport->port.irq_wake)
 		clk_disable_unprepare(sport->clk);
-- 
2.9.0

[PATCH v2 7/9] tty: serial: fsl-lpuart: Use scatter/gather DMA for Tx

[Bhuvanchandra DV]

Drop PIO to DMA switching and use scatter/gather DMA for Tx path to improve
performance.

Some part of the code is borrowed from imx serial driver.

Signed-off-by: Bhuvanchandra DV <bhuvanchandra.dv@toradex.com>
---
 drivers/tty/serial/fsl_lpuart.c | 257 ++++++++++++++++++----------------------
 1 file changed, 113 insertions(+), 144 deletions(-)

diff --git a/drivers/tty/serial/fsl_lpuart.c b/drivers/tty/serial/fsl_lpuart.c
index 51d2b5a..27687d5 100644
--- a/drivers/tty/serial/fsl_lpuart.c
+++ b/drivers/tty/serial/fsl_lpuart.c
@@ -244,18 +244,18 @@ struct lpuart_port {
 	struct dma_chan		*dma_rx_chan;
 	struct dma_async_tx_descriptor  *dma_tx_desc;
 	struct dma_async_tx_descriptor  *dma_rx_desc;
-	dma_addr_t		dma_tx_buf_bus;
 	dma_cookie_t		dma_tx_cookie;
 	dma_cookie_t		dma_rx_cookie;
-	unsigned char		*dma_tx_buf_virt;
 	unsigned int		dma_tx_bytes;
 	unsigned int		dma_rx_bytes;
-	int			dma_tx_in_progress;
+	bool			dma_tx_in_progress;
 	unsigned int		dma_rx_timeout;
 	struct timer_list	lpuart_timer;
-	struct scatterlist	rx_sgl;
+	struct scatterlist	rx_sgl, tx_sgl[2];
 	struct circ_buf		rx_ring;
 	int			rx_dma_rng_buf_len;
+	unsigned int		dma_tx_nents;
+	wait_queue_head_t	dma_wait;
 };
 
 static const struct of_device_id lpuart_dt_ids[] = {
@@ -316,103 +316,118 @@ static void lpuart32_stop_rx(struct uart_port *port)
 	lpuart32_write(temp & ~UARTCTRL_RE, port->membase + UARTCTRL);
 }
 
-static void lpuart_pio_tx(struct lpuart_port *sport)
+static void lpuart_dma_tx(struct lpuart_port *sport)
 {
 	struct circ_buf *xmit = &sport->port.state->xmit;
-	unsigned long flags;
-
-	spin_lock_irqsave(&sport->port.lock, flags);
+	struct scatterlist *sgl = sport->tx_sgl;
+	struct device *dev = sport->port.dev;
+	int ret;
 
-	while (!uart_circ_empty(xmit) &&
-		readb(sport->port.membase + UARTTCFIFO) < sport->txfifo_size) {
-		writeb(xmit->buf[xmit->tail], sport->port.membase + UARTDR);
-		xmit->tail = (xmit->tail + 1) & (UART_XMIT_SIZE - 1);
-		sport->port.icount.tx++;
-	}
+	if (sport->dma_tx_in_progress)
+		return;
 
-	if (uart_circ_chars_pending(xmit) < WAKEUP_CHARS)
-		uart_write_wakeup(&sport->port);
+	sport->dma_tx_bytes = uart_circ_chars_pending(xmit);
 
-	if (uart_circ_empty(xmit))
-		writeb(readb(sport->port.membase + UARTCR5) | UARTCR5_TDMAS,
-			sport->port.membase + UARTCR5);
+	if (xmit->tail < xmit->head) {
+		sport->dma_tx_nents = 1;
+		sg_init_one(sgl, xmit->buf + xmit->tail, sport->dma_tx_bytes);
+	} else {
+		sport->dma_tx_nents = 2;
+		sg_init_table(sgl, 2);
+		sg_set_buf(sgl, xmit->buf + xmit->tail,
+				UART_XMIT_SIZE - xmit->tail);
+		sg_set_buf(sgl + 1, xmit->buf, xmit->head);
+	}
 
-	spin_unlock_irqrestore(&sport->port.lock, flags);
-}
+	ret = dma_map_sg(dev, sgl, sport->dma_tx_nents, DMA_TO_DEVICE);
+	if (!ret) {
+		dev_err(dev, "DMA mapping error for TX.\n");
+		return;
+	}
 
-static int lpuart_dma_tx(struct lpuart_port *sport, unsigned long count)
-{
-	struct circ_buf *xmit = &sport->port.state->xmit;
-	dma_addr_t tx_bus_addr;
-
-	dma_sync_single_for_device(sport->port.dev, sport->dma_tx_buf_bus,
-				UART_XMIT_SIZE, DMA_TO_DEVICE);
-	sport->dma_tx_bytes = count & ~(sport->txfifo_size - 1);
-	tx_bus_addr = sport->dma_tx_buf_bus + xmit->tail;
-	sport->dma_tx_desc = dmaengine_prep_slave_single(sport->dma_tx_chan,
-					tx_bus_addr, sport->dma_tx_bytes,
+	sport->dma_tx_desc = dmaengine_prep_slave_sg(sport->dma_tx_chan, sgl,
+					sport->dma_tx_nents,
 					DMA_MEM_TO_DEV, DMA_PREP_INTERRUPT);
-
 	if (!sport->dma_tx_desc) {
-		dev_err(sport->port.dev, "Not able to get desc for tx\n");
-		return -EIO;
+		dma_unmap_sg(dev, sgl, sport->dma_tx_nents, DMA_TO_DEVICE);
+		dev_err(dev, "Cannot prepare TX slave DMA!\n");
+		return;
 	}
 
 	sport->dma_tx_desc->callback = lpuart_dma_tx_complete;
 	sport->dma_tx_desc->callback_param = sport;
-	sport->dma_tx_in_progress = 1;
+	sport->dma_tx_in_progress = true;
 	sport->dma_tx_cookie = dmaengine_submit(sport->dma_tx_desc);
 	dma_async_issue_pending(sport->dma_tx_chan);
 
-	return 0;
-}
-
-static void lpuart_prepare_tx(struct lpuart_port *sport)
-{
-	struct circ_buf *xmit = &sport->port.state->xmit;
-	unsigned long count =  CIRC_CNT_TO_END(xmit->head,
-					xmit->tail, UART_XMIT_SIZE);
-
-	if (!count)
-		return;
-
-	if (count < sport->txfifo_size)
-		writeb(readb(sport->port.membase + UARTCR5) & ~UARTCR5_TDMAS,
-				sport->port.membase + UARTCR5);
-	else {
-		writeb(readb(sport->port.membase + UARTCR5) | UARTCR5_TDMAS,
-				sport->port.membase + UARTCR5);
-		lpuart_dma_tx(sport, count);
-	}
 }
 
 static void lpuart_dma_tx_complete(void *arg)
 {
 	struct lpuart_port *sport = arg;
+	struct scatterlist *sgl = &sport->tx_sgl[0];
 	struct circ_buf *xmit = &sport->port.state->xmit;
 	unsigned long flags;
 
-	async_tx_ack(sport->dma_tx_desc);
-
 	spin_lock_irqsave(&sport->port.lock, flags);
 
+	dma_unmap_sg(sport->port.dev, sgl, sport->dma_tx_nents, DMA_TO_DEVICE);
+
 	xmit->tail = (xmit->tail + sport->dma_tx_bytes) & (UART_XMIT_SIZE - 1);
-	sport->dma_tx_in_progress = 0;
+
+	sport->port.icount.tx += sport->dma_tx_bytes;
+	sport->dma_tx_in_progress = false;
+	spin_unlock_irqrestore(&sport->port.lock, flags);
 
 	if (uart_circ_chars_pending(xmit) < WAKEUP_CHARS)
 		uart_write_wakeup(&sport->port);
 
-	lpuart_prepare_tx(sport);
+	if (waitqueue_active(&sport->dma_wait)) {
+		wake_up(&sport->dma_wait);
+		return;
+	}
+
+	spin_lock_irqsave(&sport->port.lock, flags);
+
+	if (!uart_circ_empty(xmit) && !uart_tx_stopped(&sport->port))
+		lpuart_dma_tx(sport);
 
 	spin_unlock_irqrestore(&sport->port.lock, flags);
 }
 
+static int lpuart_dma_tx_request(struct uart_port *port)
+{
+	struct lpuart_port *sport = container_of(port,
+					struct lpuart_port, port);
+	struct dma_slave_config dma_tx_sconfig = {};
+	int ret;
+
+	dma_tx_sconfig.dst_addr = sport->port.mapbase + UARTDR;
+	dma_tx_sconfig.dst_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
+	dma_tx_sconfig.dst_maxburst = 1;
+	dma_tx_sconfig.direction = DMA_MEM_TO_DEV;
+	ret = dmaengine_slave_config(sport->dma_tx_chan, &dma_tx_sconfig);
+
+	if (ret) {
+		dev_err(sport->port.dev,
+				"DMA slave config failed, err = %d\n", ret);
+		return ret;
+	}
+
+	return 0;
+}
+
 static void lpuart_flush_buffer(struct uart_port *port)
 {
 	struct lpuart_port *sport = container_of(port, struct lpuart_port, port);
+
 	if (sport->lpuart_dma_tx_use) {
+		if (sport->dma_tx_in_progress) {
+			dma_unmap_sg(sport->port.dev, &sport->tx_sgl[0],
+				sport->dma_tx_nents, DMA_TO_DEVICE);
+			sport->dma_tx_in_progress = false;
+		}
 		dmaengine_terminate_all(sport->dma_tx_chan);
-		sport->dma_tx_in_progress = 0;
 	}
 }
 
@@ -469,8 +484,8 @@ static void lpuart_start_tx(struct uart_port *port)
 	writeb(temp | UARTCR2_TIE, port->membase + UARTCR2);
 
 	if (sport->lpuart_dma_tx_use) {
-		if (!uart_circ_empty(xmit) && !sport->dma_tx_in_progress)
-			lpuart_prepare_tx(sport);
+		if (!uart_circ_empty(xmit) && !uart_tx_stopped(port))
+			lpuart_dma_tx(sport);
 	} else {
 		if (readb(port->membase + UARTSR1) & UARTSR1_TDRE)
 			lpuart_transmit_buffer(sport);
@@ -489,6 +504,29 @@ static void lpuart32_start_tx(struct uart_port *port)
 		lpuart32_transmit_buffer(sport);
 }
 
+/* return TIOCSER_TEMT when transmitter is not busy */
+static unsigned int lpuart_tx_empty(struct uart_port *port)
+{
+	struct lpuart_port *sport = container_of(port,
+			struct lpuart_port, port);
+	unsigned char sr1 = readb(port->membase + UARTSR1);
+	unsigned char sfifo = readb(port->membase + UARTSFIFO);
+
+	if (sport->dma_tx_in_progress)
+		return 0;
+
+	if (sr1 & UARTSR1_TC && sfifo & UARTSFIFO_TXEMPT)
+		return TIOCSER_TEMT;
+
+	return 0;
+}
+
+static unsigned int lpuart32_tx_empty(struct uart_port *port)
+{
+	return (lpuart32_read(port->membase + UARTSTAT) & UARTSTAT_TC) ?
+		TIOCSER_TEMT : 0;
+}
+
 static irqreturn_t lpuart_txint(int irq, void *dev_id)
 {
 	struct lpuart_port *sport = dev_id;
@@ -662,12 +700,8 @@ static irqreturn_t lpuart_int(int irq, void *dev_id)
 	if (sts & UARTSR1_RDRF)
 		lpuart_rxint(irq, dev_id);
 
-	if (sts & UARTSR1_TDRE) {
-		if (sport->lpuart_dma_tx_use)
-			lpuart_pio_tx(sport);
-		else
-			lpuart_txint(irq, dev_id);
-	}
+	if (sts & UARTSR1_TDRE)
+		lpuart_txint(irq, dev_id);
 
 	return IRQ_HANDLED;
 }
@@ -692,29 +726,6 @@ static irqreturn_t lpuart32_int(int irq, void *dev_id)
 	return IRQ_HANDLED;
 }
 
-/* return TIOCSER_TEMT when transmitter is not busy */
-static unsigned int lpuart_tx_empty(struct uart_port *port)
-{
-	struct lpuart_port *sport = container_of(port,
-			struct lpuart_port, port);
-	unsigned char sr1 = readb(port->membase + UARTSR1);
-	unsigned char sfifo = readb(port->membase + UARTSFIFO);
-
-	if (sport->dma_tx_in_progress)
-		return 0;
-
-	if (sr1 & UARTSR1_TC && sfifo & UARTSFIFO_TXEMPT)
-		return TIOCSER_TEMT;
-
-	return 0;
-}
-
-static unsigned int lpuart32_tx_empty(struct uart_port *port)
-{
-	return (lpuart32_read(port->membase + UARTSTAT) & UARTSTAT_TC) ?
-		TIOCSER_TEMT : 0;
-}
-
 static void lpuart_copy_rx_to_tty(struct lpuart_port *sport)
 {
 	struct tty_port *port = &sport->port.state->port;
@@ -890,18 +901,6 @@ static inline int lpuart_start_rx_dma(struct lpuart_port *sport)
 	return 0;
 }
 
-static void lpuart_dma_tx_free(struct uart_port *port)
-{
-	struct lpuart_port *sport = container_of(port,
-					struct lpuart_port, port);
-
-	dma_unmap_single(sport->port.dev, sport->dma_tx_buf_bus,
-			UART_XMIT_SIZE, DMA_TO_DEVICE);
-
-	sport->dma_tx_buf_bus = 0;
-	sport->dma_tx_buf_virt = NULL;
-}
-
 static void lpuart_dma_rx_free(struct uart_port *port)
 {
 	struct lpuart_port *sport = container_of(port,
@@ -1061,44 +1060,6 @@ static void lpuart32_setup_watermark(struct lpuart_port *sport)
 	lpuart32_write(ctrl_saved, sport->port.membase + UARTCTRL);
 }
 
-static int lpuart_dma_tx_request(struct uart_port *port)
-{
-	struct lpuart_port *sport = container_of(port,
-					struct lpuart_port, port);
-	struct dma_slave_config dma_tx_sconfig;
-	dma_addr_t dma_bus;
-	unsigned char *dma_buf;
-	int ret;
-
-	dma_bus = dma_map_single(sport->dma_tx_chan->device->dev,
-				sport->port.state->xmit.buf,
-				UART_XMIT_SIZE, DMA_TO_DEVICE);
-
-	if (dma_mapping_error(sport->dma_tx_chan->device->dev, dma_bus)) {
-		dev_err(sport->port.dev, "dma_map_single tx failed\n");
-		return -ENOMEM;
-	}
-
-	dma_buf = sport->port.state->xmit.buf;
-	dma_tx_sconfig.dst_addr = sport->port.mapbase + UARTDR;
-	dma_tx_sconfig.dst_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
-	dma_tx_sconfig.dst_maxburst = sport->txfifo_size;
-	dma_tx_sconfig.direction = DMA_MEM_TO_DEV;
-	ret = dmaengine_slave_config(sport->dma_tx_chan, &dma_tx_sconfig);
-
-	if (ret < 0) {
-		dev_err(sport->port.dev,
-				"Dma slave config failed, err = %d\n", ret);
-		return ret;
-	}
-
-	sport->dma_tx_buf_virt = dma_buf;
-	sport->dma_tx_buf_bus = dma_bus;
-	sport->dma_tx_in_progress = 0;
-
-	return 0;
-}
-
 static void rx_dma_timer_init(struct lpuart_port *sport)
 {
 		setup_timer(&sport->lpuart_timer, lpuart_timer_func,
@@ -1151,6 +1112,7 @@ static int lpuart_startup(struct uart_port *port)
 	}
 
 	if (sport->dma_tx_chan && !lpuart_dma_tx_request(port)) {
+		init_waitqueue_head(&sport->dma_wait);
 		sport->lpuart_dma_tx_use = true;
 		temp = readb(port->membase + UARTCR5);
 		writeb(temp | UARTCR5_TDMAS, port->membase + UARTCR5);
@@ -1220,8 +1182,15 @@ static void lpuart_shutdown(struct uart_port *port)
 		lpuart_dma_rx_free(&sport->port);
 	}
 
-	if (sport->lpuart_dma_tx_use)
-		lpuart_dma_tx_free(&sport->port);
+	if (sport->lpuart_dma_tx_use) {
+		if (wait_event_interruptible(sport->dma_wait,
+			!sport->dma_tx_in_progress) != false) {
+			sport->dma_tx_in_progress = false;
+			dmaengine_terminate_all(sport->dma_tx_chan);
+		}
+
+		lpuart_stop_tx(port);
+	}
 }
 
 static void lpuart32_shutdown(struct uart_port *port)
-- 
2.9.0

[PATCH v2 8/9] tty: serial: fsl-lpuart: Update suspend/resume for DMA mode

[Bhuvanchandra DV]

When DMA mode is enabled one need to make sure the DMA channels are idle before
entering suspend mode especially when UART ports which are set as wakeup source
and console port with no_console_suspend is set. This patch takes care of
gracefully releasing DMA channels for the above two cases and start the DMA at
resume.

Signed-off-by: Bhuvanchandra DV <bhuvanchandra.dv@toradex.com>
---
 drivers/tty/serial/fsl_lpuart.c | 44 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 44 insertions(+)

diff --git a/drivers/tty/serial/fsl_lpuart.c b/drivers/tty/serial/fsl_lpuart.c
index 27687d5..134090a 100644
--- a/drivers/tty/serial/fsl_lpuart.c
+++ b/drivers/tty/serial/fsl_lpuart.c
@@ -1943,6 +1943,30 @@ static int lpuart_suspend(struct device *dev)
 	}
 
 	uart_suspend_port(&lpuart_reg, &sport->port);
+
+	if (sport->lpuart_dma_rx_use) {
+		/*
+		 * EDMA driver during suspend will forcefully release any
+		 * non-idle DMA channels. If port wakeup is enabled or if port
+		 * is console port or 'no_console_suspend' is set the Rx DMA
+		 * cannot resume as as expected, hence gracefully release the
+		 * Rx DMA path before suspend and start Rx DMA path on resume.
+		 */
+		if (sport->port.irq_wake) {
+			del_timer_sync(&sport->lpuart_timer);
+			lpuart_dma_rx_free(&sport->port);
+		}
+
+		/* Disable Rx DMA to use UART port as wakeup source */
+		writeb(readb(sport->port.membase + UARTCR5) & ~UARTCR5_RDMAS,
+					sport->port.membase + UARTCR5);
+	}
+
+	if (sport->lpuart_dma_tx_use) {
+		sport->dma_tx_in_progress = false;
+		dmaengine_terminate_all(sport->dma_tx_chan);
+	}
+
 	if (sport->port.suspended && !sport->port.irq_wake)
 		clk_disable_unprepare(sport->clk);
 
@@ -1970,6 +1994,26 @@ static int lpuart_resume(struct device *dev)
 		writeb(temp, sport->port.membase + UARTCR2);
 	}
 
+	if (sport->lpuart_dma_rx_use) {
+		if (sport->port.irq_wake) {
+			if (!lpuart_start_rx_dma(sport)) {
+				sport->lpuart_dma_rx_use = true;
+				rx_dma_timer_init(sport);
+			} else {
+				sport->lpuart_dma_rx_use = false;
+			}
+		}
+	}
+
+	if (sport->dma_tx_chan && !lpuart_dma_tx_request(&sport->port)) {
+			init_waitqueue_head(&sport->dma_wait);
+			sport->lpuart_dma_tx_use = true;
+			writeb(readb(sport->port.membase + UARTCR5) |
+				UARTCR5_TDMAS, sport->port.membase + UARTCR5);
+	} else {
+		sport->lpuart_dma_tx_use = false;
+	}
+
 	uart_resume_port(&lpuart_reg, &sport->port);
 
 	return 0;
-- 
2.9.0

[PATCH v2 9/9] tty: serial: fsl_lpuart: Add support for RS-485

[Bhuvanchandra DV]

Enable Vybrid's build-in support for RS-485 auto RTS for controlling line
direction of RS-485 transceiver driver.

Enable RS485 feature by either using ioctrl 'TIOCSRS485' or enable it in the
device tree by setting 'linux,rs485-enabled-at-boot-time' property.

Signed-off-by: Bhuvanchandra DV <bhuvanchandra.dv@toradex.com>
---
 drivers/tty/serial/fsl_lpuart.c | 78 +++++++++++++++++++++++++++++++++++++----
 1 file changed, 72 insertions(+), 6 deletions(-)

diff --git a/drivers/tty/serial/fsl_lpuart.c b/drivers/tty/serial/fsl_lpuart.c
index 134090a..4d1fca4 100644
--- a/drivers/tty/serial/fsl_lpuart.c
+++ b/drivers/tty/serial/fsl_lpuart.c
@@ -917,6 +917,52 @@ static void lpuart_dma_rx_free(struct uart_port *port)
 	sport->dma_rx_cookie = -EINVAL;
 }
 
+static int lpuart_config_rs485(struct uart_port *port,
+			struct serial_rs485 *rs485)
+{
+	struct lpuart_port *sport = container_of(port,
+			struct lpuart_port, port);
+
+	u8 modem = readb(sport->port.membase + UARTMODEM) &
+		~(UARTMODEM_TXRTSPOL | UARTMODEM_TXRTSE);
+	writeb(modem, sport->port.membase + UARTMODEM);
+
+	if (rs485->flags & SER_RS485_ENABLED) {
+		/* Enable auto RS-485 RTS mode */
+		modem |= UARTMODEM_TXRTSE;
+
+		/*
+		 * RTS needs to be logic HIGH either during transer _or_ after
+		 * transfer, other variants are not supported by the hardware.
+		 */
+
+		if (!(rs485->flags & (SER_RS485_RTS_ON_SEND |
+				SER_RS485_RTS_AFTER_SEND)))
+			rs485->flags |= SER_RS485_RTS_ON_SEND;
+
+		if (rs485->flags & SER_RS485_RTS_ON_SEND &&
+				rs485->flags & SER_RS485_RTS_AFTER_SEND)
+			rs485->flags &= ~SER_RS485_RTS_AFTER_SEND;
+
+		/*
+		 * The hardware defaults to RTS logic HIGH while transfer.
+		 * Switch polarity in case RTS shall be logic HIGH
+		 * after transfer.
+		 * Note: UART is assumed to be active high.
+		 */
+		if (rs485->flags & SER_RS485_RTS_ON_SEND)
+			modem &= ~UARTMODEM_TXRTSPOL;
+		else if (rs485->flags & SER_RS485_RTS_AFTER_SEND)
+			modem |= UARTMODEM_TXRTSPOL;
+	}
+
+	/* Store the new configuration */
+	sport->port.rs485 = *rs485;
+
+	writeb(modem, sport->port.membase + UARTMODEM);
+	return 0;
+}
+
 static unsigned int lpuart_get_mctrl(struct uart_port *port)
 {
 	unsigned int temp = 0;
@@ -950,17 +996,22 @@ static unsigned int lpuart32_get_mctrl(struct uart_port *port)
 static void lpuart_set_mctrl(struct uart_port *port, unsigned int mctrl)
 {
 	unsigned char temp;
+	struct lpuart_port *sport = container_of(port,
+				struct lpuart_port, port);
 
-	temp = readb(port->membase + UARTMODEM) &
+	/* Make sure RXRTSE bit is not set when RS485 is enabled */
+	if (!(sport->port.rs485.flags & SER_RS485_ENABLED)) {
+		temp = readb(sport->port.membase + UARTMODEM) &
 			~(UARTMODEM_RXRTSE | UARTMODEM_TXCTSE);
 
-	if (mctrl & TIOCM_RTS)
-		temp |= UARTMODEM_RXRTSE;
+		if (mctrl & TIOCM_RTS)
+			temp |= UARTMODEM_RXRTSE;
 
-	if (mctrl & TIOCM_CTS)
-		temp |= UARTMODEM_TXCTSE;
+		if (mctrl & TIOCM_CTS)
+			temp |= UARTMODEM_TXCTSE;
 
-	writeb(temp, port->membase + UARTMODEM);
+		writeb(temp, port->membase + UARTMODEM);
+	}
 }
 
 static void lpuart32_set_mctrl(struct uart_port *port, unsigned int mctrl)
@@ -1256,6 +1307,13 @@ lpuart_set_termios(struct uart_port *port, struct ktermios *termios,
 		cr1 |= UARTCR1_M;
 	}
 
+	/*
+	 * When auto RS-485 RTS mode is enabled,
+	 * hardware flow control need to be disabled.
+	 */
+	if (sport->port.rs485.flags & SER_RS485_ENABLED)
+		termios->c_cflag &= ~CRTSCTS;
+
 	if (termios->c_cflag & CRTSCTS) {
 		modem |= (UARTMODEM_RXRTSE | UARTMODEM_TXCTSE);
 	} else {
@@ -1864,6 +1922,8 @@ static int lpuart_probe(struct platform_device *pdev)
 		sport->port.ops = &lpuart_pops;
 	sport->port.flags = UPF_BOOT_AUTOCONF;
 
+	sport->port.rs485_config = lpuart_config_rs485;
+
 	sport->clk = devm_clk_get(&pdev->dev, "ipg");
 	if (IS_ERR(sport->clk)) {
 		ret = PTR_ERR(sport->clk);
@@ -1904,6 +1964,12 @@ static int lpuart_probe(struct platform_device *pdev)
 		dev_info(sport->port.dev, "DMA rx channel request failed, "
 				"operating without rx DMA\n");
 
+	if (of_property_read_bool(np, "linux,rs485-enabled-at-boot-time")) {
+		sport->port.rs485.flags |= SER_RS485_ENABLED;
+		sport->port.rs485.flags |= SER_RS485_RTS_ON_SEND;
+		writeb(UARTMODEM_TXRTSE, sport->port.membase + UARTMODEM);
+	}
+
 	return 0;
 }
 
-- 
2.9.0

[PATCH v2 0/9] Improve Rx/Tx DMA implementation

[Bhuvanchandra DV]

Ping!

On 06/28/16 11:02, Bhuvanchandra DV wrote:

> Changes since v1:
> Split suspend/resume patch.
>
> Bhuvanchandra DV (5):
>    tty: serial: fsl_lpuart: Fix broken 8m/s1 support
>    tty: serial: fsl-lpuart: Use cyclic DMA for Rx
>    tty: serial: fsl-lpuart: Use scatter/gather DMA for Tx
>    tty: serial: fsl-lpuart: Update suspend/resume for DMA mode
>    tty: serial: fsl_lpuart: Add support for RS-485
>
> Stefan Agner (4):
>    tty: serial: fsl_lpuart: consider TX FIFO too in tx_empty
>    clk: imx: vf610: Disable automatic clock gating for lpuart in LPSTOP mode
>    tty: serial: fsl_lpuart: support suspend/resume
>    tty: serial: fsl_lpuart: fix clearing of receive flag
>
>   drivers/clk/imx/clk-vf610.c     |  12 +-
>   drivers/tty/serial/fsl_lpuart.c | 810 +++++++++++++++++++++++-----------------
>   2 files changed, 482 insertions(+), 340 deletions(-)
>

[PATCH v2 2/9] clk: imx: vf610: Disable automatic clock gating for lpuart in LPSTOP mode

[Michael Turquette]

Quoting Bhuvanchandra DV (2016-06-27 22:32:28)
> From: Stefan Agner <stefan@agner.ch>
> 
> In order to allow wake support in STOP sleep mode, clocks are needed. Use
> imx_clk_gate2_cgr to disable automatic clock gating in low power mode STOP.
> This allows to enable wake by UART using:
> echo enabled > /sys/class/tty/ttyLP0/power/wakeup
> 
> However, if wake is not enabled, the driver should disable the clocks explicitly
> to save power.
> 
> Signed-off-by: Stefan Agner <stefan@agner.ch>
> Signed-off-by: Bhuvanchandra DV <bhuvanchandra.dv@toradex.com>

Applied.

Regards,
Mike

> ---
>  drivers/clk/imx/clk-vf610.c | 12 ++++++------
>  1 file changed, 6 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/clk/imx/clk-vf610.c b/drivers/clk/imx/clk-vf610.c
> index 3a1f244..0476353 100644
> --- a/drivers/clk/imx/clk-vf610.c
> +++ b/drivers/clk/imx/clk-vf610.c
> @@ -315,12 +315,12 @@ static void __init vf610_clocks_init(struct device_node *ccm_node)
>  
>         clk[VF610_CLK_PIT] = imx_clk_gate2("pit", "ipg_bus", CCM_CCGR1, CCM_CCGRx_CGn(7));
>  
> -       clk[VF610_CLK_UART0] = imx_clk_gate2("uart0", "ipg_bus", CCM_CCGR0, CCM_CCGRx_CGn(7));
> -       clk[VF610_CLK_UART1] = imx_clk_gate2("uart1", "ipg_bus", CCM_CCGR0, CCM_CCGRx_CGn(8));
> -       clk[VF610_CLK_UART2] = imx_clk_gate2("uart2", "ipg_bus", CCM_CCGR0, CCM_CCGRx_CGn(9));
> -       clk[VF610_CLK_UART3] = imx_clk_gate2("uart3", "ipg_bus", CCM_CCGR0, CCM_CCGRx_CGn(10));
> -       clk[VF610_CLK_UART4] = imx_clk_gate2("uart4", "ipg_bus", CCM_CCGR6, CCM_CCGRx_CGn(9));
> -       clk[VF610_CLK_UART5] = imx_clk_gate2("uart5", "ipg_bus", CCM_CCGR6, CCM_CCGRx_CGn(10));
> +       clk[VF610_CLK_UART0] = imx_clk_gate2_cgr("uart0", "ipg_bus", CCM_CCGR0, CCM_CCGRx_CGn(7), 0x2);
> +       clk[VF610_CLK_UART1] = imx_clk_gate2_cgr("uart1", "ipg_bus", CCM_CCGR0, CCM_CCGRx_CGn(8), 0x2);
> +       clk[VF610_CLK_UART2] = imx_clk_gate2_cgr("uart2", "ipg_bus", CCM_CCGR0, CCM_CCGRx_CGn(9), 0x2);
> +       clk[VF610_CLK_UART3] = imx_clk_gate2_cgr("uart3", "ipg_bus", CCM_CCGR0, CCM_CCGRx_CGn(10), 0x2);
> +       clk[VF610_CLK_UART4] = imx_clk_gate2_cgr("uart4", "ipg_bus", CCM_CCGR6, CCM_CCGRx_CGn(9), 0x2);
> +       clk[VF610_CLK_UART5] = imx_clk_gate2_cgr("uart5", "ipg_bus", CCM_CCGR6, CCM_CCGRx_CGn(10), 0x2);
>  
>         clk[VF610_CLK_I2C0] = imx_clk_gate2("i2c0", "ipg_bus", CCM_CCGR4, CCM_CCGRx_CGn(6));
>         clk[VF610_CLK_I2C1] = imx_clk_gate2("i2c1", "ipg_bus", CCM_CCGR4, CCM_CCGRx_CGn(7));
> -- 
> 2.9.0
>