Skip to contentSkip to conversation

ARM: dts: vf610: lpuart: Add eDMA support

PATCH Superseded Yuan Yao 3 participants 11 messages Jan 15, 2014 – Jan 22, 2014

Subsystems (2)

v3 of 6 Superseded Latest is v6 →

lore
Share & tools
lore

Yuan Yao

author
#0 lore raw

PATCH v3 1/2 ARM: dts: vf610: lpuart: Add eDMA support

Add lpuart dts node properties for eDMA support, them depend on the eDMA driver. See VF610_EDMA_MUXID0_UART*_*X macro on eDMA driver

---
 arch/arm/boot/dts/vf610.dtsi | 12 ++++++++++++
 1 file changed, 12 insertions(+)
arch/arm/boot/dts/vf610.dtsi +12 -0
index 365e0fa..ed6b053 100644
@@ -118,6 +118,9 @@
 				interrupts = <0 61 0x00>;
 				clocks = <&clks VF610_CLK_UART0>;
 				clock-names = "ipg";
+				dma-names = "lpuart-tx","lpuart-rx";
+				dmas = <&edma0 0 VF610_EDMA_MUXID0_UART0_TX>,
+					<&edma0 0 VF610_EDMA_MUXID0_UART0_RX>;
 				status = "disabled";
 			};
 
@@ -127,6 +130,9 @@
 				interrupts = <0 62 0x04>;
 				clocks = <&clks VF610_CLK_UART1>;
 				clock-names = "ipg";
+				dma-names = "lpuart-tx","lpuart-rx";
+				dmas = <&edma0 0 VF610_EDMA_MUXID0_UART1_TX>,
+					<&edma0 0 VF610_EDMA_MUXID0_UART1_RX>;
 				status = "disabled";
 			};
 
@@ -136,6 +142,9 @@
 				interrupts = <0 63 0x04>;
 				clocks = <&clks VF610_CLK_UART2>;
 				clock-names = "ipg";
+				dma-names = "lpuart-tx","lpuart-rx";
+				dmas = <&edma0 0 VF610_EDMA_MUXID0_UART2_TX>,
+					<&edma0 0 VF610_EDMA_MUXID0_UART2_RX>;
 				status = "disabled";
 			};
 
@@ -145,6 +154,9 @@
 				interrupts = <0 64 0x04>;
 				clocks = <&clks VF610_CLK_UART3>;
 				clock-names = "ipg";
+				dma-names = "lpuart-tx","lpuart-rx";
+				dmas = <&edma0 0 VF610_EDMA_MUXID0_UART3_TX>,
+					<&edma0 0 VF610_EDMA_MUXID0_UART3_RX>;
 				status = "disabled";
 			};
 
Signature · 1 line
1.8.4
author
#1 lore raw

PATCH v3 2/2 serial: fsl_lpuart: add DMA support

Add dma support for lpuart. This function depend on DMA driver.
You can turn on it by SERIAL_FSL_LPUART_DMA=y. And It works if dts node has dma properties.

---
 .../devicetree/bindings/serial/fsl-lpuart.txt      |  21 +-
 drivers/tty/serial/Kconfig                         |   7 +
 drivers/tty/serial/fsl_lpuart.c                    | 457 ++++++++++++++++++++-
 3 files changed, 473 insertions(+), 12 deletions(-)
Documentation/devicetree/bindings/serial/fsl-lpuart.txt +17 -4
index 6fd1dd1..7509080 100644
@@ -4,11 +4,24 @@ Required properties:
 - compatible : Should be "fsl,<soc>-lpuart"
 - reg : Address and length of the register set for the device
 - interrupts : Should contain uart interrupt
+- clocks : from common clock binding: handle to uart clock
+- clock-names : from common clock binding: Shall be "ipg"
+
+Optional properties:
+- dma-names: Should contain "lpuart-tx" for transmit and "lpuart-rx" for receive channels
+- dmas: Should contain dma specifiers for transmit and receive channels
+
+Note: Optional properties for DMA support, If need this properties is must.
 
 Example:
 
 uart0: serial at 40027000 {
-	       compatible = "fsl,vf610-lpuart";
-	       reg = <0x40027000 0x1000>;
-	       interrupts = <0 61 0x00>;
-       };
+		compatible = "fsl,vf610-lpuart";
+		reg = <0x40027000 0x1000>;
+		interrupts = <0 61 0x00>;
+		clocks = <&clks VF610_CLK_UART0>;
+		clock-names = "ipg";
+		dma-names = "lpuart-tx","lpuart-rx";
+		dmas = <&edma0 0 VF610_EDMA_MUXID0_UART0_TX>,
+			<&edma0 0 VF610_EDMA_MUXID0_UART0_RX>;
+	};
drivers/tty/serial/Kconfig +7 -0
index a3817ab..ab3910b 100644
@@ -1509,6 +1509,13 @@ config SERIAL_FSL_LPUART_CONSOLE
 	  If you have enabled the lpuart serial port on the Freescale SoCs,
 	  you can make it the console by answering Y to this option.
 
+config SERIAL_FSL_LPUART_DMA
+	bool "Freescale lpuart serial port dma support"
+	depends on FSL_EDMA=y && SERIAL_FSL_LPUART=y
+	help
+	  DMA support for the on-chip lpuart on some Freescale SOCs,
+	  you can turn on the dma support for lpuart by answering Y to this option.
+
 config SERIAL_ST_ASC
 	tristate "ST ASC serial port support"
 	select SERIAL_CORE
drivers/tty/serial/fsl_lpuart.c +449 -8

544 lines not on this page · raw message

Signature · 1 line
1.8.4
#2 lore raw

PATCH v3 2/2 serial: fsl_lpuart: add DMA support

Quoted 4 lines · Yuan Yao

Add dma support for lpuart. This function depend on DMA driver.
You can turn on it by SERIAL_FSL_LPUART_DMA=y. And It works if dts node has dma properties.

Most of the changes you did look good, but I have to say I'm not a fan of the #ifdef you added. I don't think you actually need to have the compile-time selection, and it would be best to just drop that part.

If you have a good reason to keep it, please change the code to more readable "if (IS_ENABLED(CONFIG_FSL_LPUART_DMA))" constructs. If you decide to do this, you will find that you need far fewer such conditionals because a lot of the code will automatically get dropped by the compiler.

Quoted 4 lines

uart0: serial at 40027000 {

Documentation/devicetree/bindings/serial/fsl-lpuart.txt +0 -3
-	       compatible = "fsl,vf610-lpuart";
-	       reg = <0x40027000 0x1000>;
-	       interrupts = <0 61 0x00>;

Documentation/devicetree/bindings/serial/fsl-lpuart.txt Required properties: · line 27

Quoted 10 lines
Documentation/devicetree/bindings/serial/fsl-lpuart.txt +9 -1
-       };
+		compatible = "fsl,vf610-lpuart";
+		reg = <0x40027000 0x1000>;
+		interrupts = <0 61 0x00>;
+		clocks = <&clks VF610_CLK_UART0>;
+		clock-names = "ipg";
+		dma-names = "lpuart-tx","lpuart-rx";
+		dmas = <&edma0 0 VF610_EDMA_MUXID0_UART0_TX>,
+			<&edma0 0 VF610_EDMA_MUXID0_UART0_RX>;
+	};

You still haven't addressed my comment about removing the VF610_EDMA_MUXID0_UART0_TX macros.

drivers/tty/serial/fsl_lpuart.c line 124

Quoted 6 lines
drivers/tty/serial/fsl_lpuart.c +5 -0
+#ifdef CONFIG_SERIAL_FSL_LPUART_DMA
+#define DMA_MAXBURST		16
+#define DMA_MAXBURST_MASK	(DMA_MAXBURST - 1)
+#define FSL_UART_RX_DMA_BUFFER_SIZE	64
+#endif

For this part there was no need for the #ifdef to start with.

Quoted 18 lines

#define DRIVER_NAME "fsl-lpuart"
#define DEV_NAME "ttyLP"
#define UART_NR 6

drivers/tty/serial/fsl_lpuart.c +11 -0
@@ -121,6 +132,26 @@ struct lpuart_port {
 	struct clk		*clk;
 	unsigned int		txfifo_size;
 	unsigned int		rxfifo_size;
+
+#ifdef CONFIG_SERIAL_FSL_LPUART_DMA
+	bool			lpuart_dma_use;
+	struct dma_chan		*dma_tx_chan;
+	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_addr_t		dma_rx_buf_bus;
+	dma_cookie_t		dma_tx_cookie;
+	dma_cookie_t		dma_rx_cookie;

drivers/tty/serial/fsl_lpuart.c struct lpuart_port { · line 155

Quoted 10 lines
drivers/tty/serial/fsl_lpuart.c +9 -0
+	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;
+#endif
 };

This part will result in a slight increase in data size even if dma support is turned off at compile time.

Quoted 7 lines
drivers/tty/serial/fsl_lpuart.c +3 -0
 	temp = readb(port->membase + UARTCR2);
 	writeb(temp | UARTCR2_TIE, port->membase + UARTCR2);
 
+#ifdef CONFIG_SERIAL_FSL_LPUART_DMA
+	if (sport->lpuart_dma_use) {
+		if (!uart_circ_empty(xmit) && !sport->dma_tx_in_progress)

drivers/tty/serial/fsl_lpuart.c static inline void lpuart_transmit_buffer(struct lpuart_port *sport) · line 438

Quoted 10 lines
drivers/tty/serial/fsl_lpuart.c +7 -0
+			lpuart_prepare_tx(sport);
+	} else {
+		if (readb(port->membase + UARTSR1) & UARTSR1_TDRE)
+			lpuart_transmit_buffer(sport);
+	}
+#else
 	if (readb(port->membase + UARTSR1) & UARTSR1_TDRE)
 		lpuart_transmit_buffer(sport);
+#endif
 }

So this can simply become

+ if (IS_ENABLED(CONFIG_SERIAL_FSL_LPUART_DMA) && sport->lpuart_dma_use) {
+ if (!uart_circ_empty(xmit) && !sport->dma_tx_in_progress)
+ lpuart_prepare_tx(sport);
+ } else {
+ if (readb(port->membase + UARTSR1) & UARTSR1_TDRE)
+ lpuart_trans mit_buffer(sport);
+ }

and the compiler will silently drop the lpuart_prepare_tx() function and everything it calls that isn't used elsewhere.

Quoted 4 lines
drivers/tty/serial/fsl_lpuart.c +4 -0
+
+#ifdef CONFIG_SERIAL_FSL_LPUART_DMA
+	struct platform_device *pdev = to_platform_device(port->dev);
+	struct device_node *np = pdev->dev.of_node;

drivers/tty/serial/fsl_lpuart.c static void lpuart_setup_watermark(struct lpuart_port *sport) · line 810

Quoted 10 lines
drivers/tty/serial/fsl_lpuart.c +10 -0
+
+	if (of_get_property(np, "dmas", NULL)) {
+		sport->lpuart_dma_use = true;
+		lpuart_dma_tx_request(port);
+		lpuart_dma_rx_request(port);
+		temp = readb(port->membase + UARTCR5);
+		writeb(temp | UARTCR5_TDMAS, port->membase + UARTCR5);
+	} else
+		sport->lpuart_dma_use = false;	
+#endif

Same here.

drivers/tty/serial/fsl_lpuart.c static int __init lpuart_serial_init(void) · line 1297

Quoted 8 lines
drivers/tty/serial/fsl_lpuart.c +4 -0
@@ -854,6 +1291,10 @@ static int __init lpuart_serial_init(void)
 
 	pr_info("serial: Freescale lpuart driver\n");
 
+#ifdef CONFIG_SERIAL_FSL_LPUART_DMA
+	pr_info("serial: Freescale lpuart dma support\n");
+#endif
+

This message is not helpful in the long run, I'd rather see you drop the other one as well. Note that this function is called on *all* platforms when the driver is enabled, not just on those that actually have the uart.

Arnd

author
#3 lore raw

PATCH v3 2/2 serial: fsl_lpuart: add DMA support

Hi, Arnd
Thanks for your suggestion.

Quoted 21 lines

-----Original Message-----
From: linux-serial-owner at vger.kernel.org [mailto:linux-serial- owner at vger.kernel.org] On Behalf Of Arnd Bergmann
Sent: Wednesday, January 15, 2014 5:26 PM
To: Yuan Yao-B46683
Cc: gregkh at linuxfoundation.org; shawn.guo at linaro.org; linux at arm.linux.org.uk; linux-arm-kernel at lists.infradead.org; linux- serial at vger.kernel.org
Subject: Re: [PATCH v3 2/2] serial: fsl_lpuart: add DMA support

Quoted 2 lines · Yuan Yao

Add dma support for lpuart. This function depend on DMA driver. You can turn on it by SERIAL_FSL_LPUART_DMA=y. And It works if dts node

has dma properties.

Quoted 2 lines

Most of the changes you did look good, but I have to say I'm not a fan of the #ifdef you added. I don't think you actually need to have the compile-time selection, and it would be best to just drop that part.

Quoted 6 lines

If you have a good reason to keep it, please change the code to more readable "if (IS_ENABLED(CONFIG_FSL_LPUART_DMA))" constructs. If you decide to do this, you will find that you need far fewer such conditionals because a lot of the code will automatically get dropped by the compiler.

Thank you! I even not know the IS_ENABLED macro.
At the beginning I just think this can save the size of code. But now I think it's not very necessary.
I tend to cancel it.

Quoted 9 lines
Quoted 9 lines

uart0: serial at 40027000 {

Documentation/devicetree/bindings/serial/fsl-lpuart.txt +4 -4
-	       compatible = "fsl,vf610-lpuart";
-	       reg = <0x40027000 0x1000>;
-	       interrupts = <0 61 0x00>;
-       };
+		compatible = "fsl,vf610-lpuart";
+		reg = <0x40027000 0x1000>;
+		interrupts = <0 61 0x00>;
+		clocks = <&clks VF610_CLK_UART0>;
Quoted 9 lines

Documentation/devicetree/bindings/serial/fsl-lpuart.txt Required properties: · line 27

Quoted 5 lines
Documentation/devicetree/bindings/serial/fsl-lpuart.txt +5 -0
+		clock-names = "ipg";
+		dma-names = "lpuart-tx","lpuart-rx";
+		dmas = <&edma0 0 VF610_EDMA_MUXID0_UART0_TX>,
+			<&edma0 0 VF610_EDMA_MUXID0_UART0_RX>;
+	};

You still haven't addressed my comment about removing the VF610_EDMA_MUXID0_UART0_TX macros.

Very sorry for forget it. At that time the VF610_EDMA_MUXID0_UART0_TX is because of eDMA driver request. But now it also cancel the macro. So this is my fault. I will fix it next.

Quoted 32 lines

drivers/tty/serial/fsl_lpuart.c line 124

Quoted 6 lines
drivers/tty/serial/fsl_lpuart.c +5 -0
+#ifdef CONFIG_SERIAL_FSL_LPUART_DMA
+#define DMA_MAXBURST		16
+#define DMA_MAXBURST_MASK	(DMA_MAXBURST - 1)
+#define FSL_UART_RX_DMA_BUFFER_SIZE	64
+#endif

For this part there was no need for the #ifdef to start with.

Quoted 23 lines

#define DRIVER_NAME "fsl-lpuart"
#define DEV_NAME "ttyLP"
#define UART_NR 6

drivers/tty/serial/fsl_lpuart.c +16 -0
@@ -121,6 +132,26 @@ struct lpuart_port {
 	struct clk		*clk;
 	unsigned int		txfifo_size;
 	unsigned int		rxfifo_size;
+
+#ifdef CONFIG_SERIAL_FSL_LPUART_DMA
+	bool			lpuart_dma_use;
+	struct dma_chan		*dma_tx_chan;
+	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_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;
Quoted 9 lines

drivers/tty/serial/fsl_lpuart.c struct lpuart_port { · line 155

Quoted 5 lines
drivers/tty/serial/fsl_lpuart.c +4 -0
+	int			dma_rx_in_progress;
+	unsigned int		dma_rx_timeout;
+	struct timer_list	lpuart_timer;
+#endif
 };

This part will result in a slight increase in data size even if dma support is turned off at compile time.

Yes, I will fix it.

Quoted 29 lines
Quoted 7 lines
drivers/tty/serial/fsl_lpuart.c +3 -0
 	temp = readb(port->membase + UARTCR2);
 	writeb(temp | UARTCR2_TIE, port->membase + UARTCR2);
 
+#ifdef CONFIG_SERIAL_FSL_LPUART_DMA
+	if (sport->lpuart_dma_use) {
+		if (!uart_circ_empty(xmit) && !sport->dma_tx_in_progress)

drivers/tty/serial/fsl_lpuart.c static inline void lpuart_transmit_buffer(struct lpuart_port *sport) · line 438

Quoted 10 lines
drivers/tty/serial/fsl_lpuart.c +7 -0
+			lpuart_prepare_tx(sport);
+	} else {
+		if (readb(port->membase + UARTSR1) & UARTSR1_TDRE)
+			lpuart_transmit_buffer(sport);
+	}
+#else
 	if (readb(port->membase + UARTSR1) & UARTSR1_TDRE)
 		lpuart_transmit_buffer(sport);
+#endif
 }

So this can simply become

(quoted diff) +1 -0
+	if (IS_ENABLED(CONFIG_SERIAL_FSL_LPUART_DMA) && sport-
Quoted 1 line

lpuart_dma_use) {

drivers/tty/serial/fsl_lpuart.c +6 -0
+		if (!uart_circ_empty(xmit) && !sport->dma_tx_in_progress)
+			lpuart_prepare_tx(sport);
+	} else {
+		if (readb(port->membase + UARTSR1) & UARTSR1_TDRE)
+			lpuart_trans	mit_buffer(sport);
+	}
Quoted 3 lines

and the compiler will silently drop the lpuart_prepare_tx() function and everything it calls that isn't used elsewhere.

It's wonderful.

Quoted 36 lines
Quoted 5 lines
drivers/tty/serial/fsl_lpuart.c +4 -0
+
+#ifdef CONFIG_SERIAL_FSL_LPUART_DMA
+	struct platform_device *pdev = to_platform_device(port->dev);
+	struct device_node *np = pdev->dev.of_node;

drivers/tty/serial/fsl_lpuart.c static void lpuart_setup_watermark(struct lpuart_port *sport) · line 810

Quoted 10 lines
drivers/tty/serial/fsl_lpuart.c +10 -0
+
+	if (of_get_property(np, "dmas", NULL)) {
+		sport->lpuart_dma_use = true;
+		lpuart_dma_tx_request(port);
+		lpuart_dma_rx_request(port);
+		temp = readb(port->membase + UARTCR5);
+		writeb(temp | UARTCR5_TDMAS, port->membase + UARTCR5);
+	} else
+		sport->lpuart_dma_use = false;
+#endif

Same here.

drivers/tty/serial/fsl_lpuart.c static int __init lpuart_serial_init(void) · line 1296

Quoted 7 lines
drivers/tty/serial/fsl_lpuart.c +3 -0
@@ -854,6 +1291,10 @@ static int __init lpuart_serial_init(void)
 
 	pr_info("serial: Freescale lpuart driver\n");
 
+#ifdef CONFIG_SERIAL_FSL_LPUART_DMA
+	pr_info("serial: Freescale lpuart dma support\n"); #endif
+

This message is not helpful in the long run, I'd rather see you drop the other one as well. Note that this function is called on *all* platforms when the driver is enabled, not just on those that actually have the uart.

Arnd
--
To unsubscribe from this list: send the line "unsubscribe linux-serial" in the body of a message to majordomo at vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html

author
#4 lore raw

PATCH v3 2/2 serial: fsl_lpuart: add DMA support

In reply to Arnd Bergmann

Hi, Arnd

Quoted 24 lines

-----Original Message-----

Quoted 23 lines

#define DRIVER_NAME "fsl-lpuart"
#define DEV_NAME "ttyLP"
#define UART_NR 6

drivers/tty/serial/fsl_lpuart.c +16 -0
@@ -121,6 +132,26 @@ struct lpuart_port {
 	struct clk		*clk;
 	unsigned int		txfifo_size;
 	unsigned int		rxfifo_size;
+
+#ifdef CONFIG_SERIAL_FSL_LPUART_DMA
+	bool			lpuart_dma_use;
+	struct dma_chan		*dma_tx_chan;
+	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_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;
Quoted 9 lines

drivers/tty/serial/fsl_lpuart.c struct lpuart_port { · line 155

Quoted 5 lines
drivers/tty/serial/fsl_lpuart.c +4 -0
+	int			dma_rx_in_progress;
+	unsigned int		dma_rx_timeout;
+	struct timer_list	lpuart_timer;
+#endif
 };

This part will result in a slight increase in data size even if dma support is turned off at compile time.

Yes, someone also want me define a independent struct for dma and allocate the memory when dma be use. But I think this way will add a more memory access time, like a->b->c and a->c. And the variable
about dma is just a few bytes but used frequently. So I think I'd better define the variable about dma in lpuart_port struct.

#5 lore raw

PATCH v3 2/2 serial: fsl_lpuart: add DMA support

Quoted 7 lines · Yao Yuan
Quoted 3 lines

This part will result in a slight increase in data size even if dma support is turned off at compile time.

Yes, someone also want me define a independent struct for dma and allocate the memory when dma be use. But I think this way will add a more memory access time, like a->b->c and a->c. And the variable
about dma is just a few bytes but used frequently. So I think I'd better define the variable about dma in lpuart_port struct.

I agree. Just leave it unconditionally in here to make the code simpler.

Arnd

#6 lore raw

PATCH v3 2/2 serial: fsl_lpuart: add DMA support

In reply to Yuan Yao
Quoted 15 lines · Yuan Yao

Add dma support for lpuart. This function depend on DMA driver.
You can turn on it by SERIAL_FSL_LPUART_DMA=y. And It works if dts node has dma properties.

---
 .../devicetree/bindings/serial/fsl-lpuart.txt      |  21 +-
 drivers/tty/serial/Kconfig                         |   7 +
 drivers/tty/serial/fsl_lpuart.c                    | 457 ++++++++++++++++++++-
 3 files changed, 473 insertions(+), 12 deletions(-)
Documentation/devicetree/bindings/serial/fsl-lpuart.txt +0 -0
index 6fd1dd1..7509080 100644

Documentation/devicetree/bindings/serial/fsl-lpuart.txt Required properties: · line 8

Quoted 5 lines
Documentation/devicetree/bindings/serial/fsl-lpuart.txt +2 -0
@@ -4,11 +4,24 @@ Required properties:
 - compatible : Should be "fsl,<soc>-lpuart"
 - reg : Address and length of the register set for the device
 - interrupts : Should contain uart interrupt
+- clocks : from common clock binding: handle to uart clock
+- clock-names : from common clock binding: Shall be "ipg"

Why are these now requried if they weren't previously? That breaks old dts. I can't see any new code touching clocks. Was this an old but undocumented requirement?

Could you please reword this so clocks is defined in terms of clock-names so as to make the relationship between them clear:

- clocks: a list of phandles + clock-specifier pairs, one for each entry in clock-names

- clock-names: should contain:
* "ipg" - the uart clock

Documentation/devicetree/bindings/serial/fsl-lpuart.txt Required properties: · line 12

Quoted 4 lines
Documentation/devicetree/bindings/serial/fsl-lpuart.txt +4 -0
+
+Optional properties:
+- dma-names: Should contain "lpuart-tx" for transmit and "lpuart-rx" for receive channels
+- dmas: Should contain dma specifiers for transmit and receive channels

Likewise could you please define dmas in terms of dma-names please. Why not just "tx" and "rx" as other bindings use?

[...]

Documentation/devicetree/bindings/serial/fsl-lpuart.txt Required properties: · line 22

Quoted 10 lines
Documentation/devicetree/bindings/serial/fsl-lpuart.txt +10 -0
+#ifdef CONFIG_SERIAL_FSL_LPUART_DMA
+       struct platform_device *pdev = to_platform_device(port->dev);
+       struct device_node *np = pdev->dev.of_node;
+
+       if (of_get_property(np, "dmas", NULL)) {
+               sport->lpuart_dma_use = true;
+               lpuart_dma_tx_request(port);
+               lpuart_dma_rx_request(port);
+               temp = readb(port->membase + UARTCR5);
+               writeb(temp | UARTCR5_TDMAS, port->membase + UARTCR5);

Rather than reading the raw dt to find out if you have dmas, can you not just attempt to request the dmas and if either fail give up on using them?

Documentation/devicetree/bindings/serial/fsl-lpuart.txt Required properties: · line 24

Quoted 2 lines
Documentation/devicetree/bindings/serial/fsl-lpuart.txt +2 -0
+       } else
+               sport->lpuart_dma_use = false;

Nit: if you have brackets on one half of an if-else, they should be on the other half too (even if i's a single line).

Cheers,
Mark.

#7 lore raw

PATCH v3 2/2 serial: fsl_lpuart: add DMA support

Quoted 13 lines · On Wednesday 15 January 2014 11:05:37 Mark Rutland wrote:

Documentation/devicetree/bindings/serial/fsl-lpuart.txt Required properties: · line 8

Quoted 8 lines · Yuan Yao
Documentation/devicetree/bindings/serial/fsl-lpuart.txt +2 -0
@@ -4,11 +4,24 @@ Required properties:
 - compatible : Should be "fsl,<soc>-lpuart"
 - reg : Address and length of the register set for the device
 - interrupts : Should contain uart interrupt
+- clocks : from common clock binding: handle to uart clock
+- clock-names : from common clock binding: Shall be "ipg"

Why are these now requried if they weren't previously? That breaks old dts. I can't see any new code touching clocks. Was this an old but undocumented requirement?

It was previously required but not documented. I asked Yuan Yao to add the text to the binding along with the new "dmas" addition.

It would have been cleaner to do this as a separate patch, but at the very least it should be mentioned in the changelog.

Arnd

#8 lore raw

PATCH v3 2/2 serial: fsl_lpuart: add DMA support

Quoted 15 lines · Arnd Bergmann
Quoted 13 lines · On Wednesday 15 January 2014 11:05:37 Mark Rutland wrote:

Documentation/devicetree/bindings/serial/fsl-lpuart.txt Required properties: · line 8

Quoted 8 lines · Yuan Yao
Documentation/devicetree/bindings/serial/fsl-lpuart.txt +2 -0
@@ -4,11 +4,24 @@ Required properties:
 - compatible : Should be "fsl,<soc>-lpuart"
 - reg : Address and length of the register set for the device
 - interrupts : Should contain uart interrupt
+- clocks : from common clock binding: handle to uart clock
+- clock-names : from common clock binding: Shall be "ipg"

Why are these now requried if they weren't previously? That breaks old dts. I can't see any new code touching clocks. Was this an old but undocumented requirement?

Quoted 2 lines

It was previously required but not documented. I asked Yuan Yao to add the text to the binding along with the new "dmas" addition.

Ok.

Quoted 3 lines

It would have been cleaner to do this as a separate patch, but at the very least it should be mentioned in the changelog.

Having it in the changelog would be nice.

Cheers,
Mark.

author
#9 lore raw

PATCH v3 2/2 serial: fsl_lpuart: add DMA support

In reply to Mark Rutland

Hi, mark

Quoted 14 lines

-----Original Message-----

drivers/tty/serial/fsl_lpuart.c static void lpuart_setup_watermark(struct lpuart_port *sport) · line 807

Quoted 10 lines
drivers/tty/serial/fsl_lpuart.c +10 -0
+#ifdef CONFIG_SERIAL_FSL_LPUART_DMA
+       struct platform_device *pdev = to_platform_device(port->dev);
+       struct device_node *np = pdev->dev.of_node;
+
+       if (of_get_property(np, "dmas", NULL)) {
+               sport->lpuart_dma_use = true;
+               lpuart_dma_tx_request(port);
+               lpuart_dma_rx_request(port);
+               temp = readb(port->membase + UARTCR5);
+               writeb(temp | UARTCR5_TDMAS, port->membase + UARTCR5);

Rather than reading the raw dt to find out if you have dmas, can you not just attempt to request the dmas and if either fail give up on using them?

Yes, the dma request function can also confirm it. But maybe it's better that add the judge as a dma entrance first?
I think if the dmas is be written it means dma want be support. At this time, rather than silently change to no dma model, we may throw the error when some errors happened. But dma request failed may have many other reasons. Also the judge will just run only once, it will not waste of performance.
If the dmas is not be written, we don't need to do anything about dma.

author
#10 lore raw

PATCH v3 2/2 serial: fsl_lpuart: add DMA support

Hi, Mark

Think twice.
Yes, you are right. "of_get_property" is not necessary here.

Thanks for your suggestion.

Quoted 29 lines

-----Original Message-----
From: linux-serial-owner at vger.kernel.org [mailto:linux-serial- owner at vger.kernel.org] On Behalf Of Yao Yuan
Sent: Thursday, January 16, 2014 2:54 PM
To: Mark Rutland
Cc: gregkh at linuxfoundation.org; shawn.guo at linaro.org; linux at arm.linux.org.uk; linux-arm-kernel at lists.infradead.org; arnd at arndb.de; linux-serial at vger.kernel.org
Subject: RE: [PATCH v3 2/2] serial: fsl_lpuart: add DMA support

Hi, mark

Quoted 15 lines

-----Original Message-----

drivers/tty/serial/fsl_lpuart.c static void lpuart_setup_watermark(struct lpuart_port *sport)

Quoted 11 lines
drivers/tty/serial/fsl_lpuart.c +11 -0
+#ifdef CONFIG_SERIAL_FSL_LPUART_DMA
+       struct platform_device *pdev = to_platform_device(port->dev);
+       struct device_node *np = pdev->dev.of_node;
+
+       if (of_get_property(np, "dmas", NULL)) {
+               sport->lpuart_dma_use = true;
+               lpuart_dma_tx_request(port);
+               lpuart_dma_rx_request(port);
+               temp = readb(port->membase + UARTCR5);
+               writeb(temp | UARTCR5_TDMAS, port->membase +
+ UARTCR5);

Rather than reading the raw dt to find out if you have dmas, can you not just attempt to request the dmas and if either fail give up on

using them?

Quoted 9 lines

Yes, the dma request function can also confirm it. But maybe it's better that add the judge as a dma entrance first?
I think if the dmas is be written it means dma want be support. At this time, rather than silently change to no dma model, we may throw the error when some errors happened.
But dma request failed may have many other reasons. Also the judge will just run only once, it will not waste of performance.
If the dmas is not be written, we don't need to do anything about dma.  ?

& ~ &  +- ? w ? m b lz ) w*jg

     ?j/   z ?  2 ?
Quoted 1 line

& )? a 

G h  j:+v w ?