Skip to contentSkip to conversation

Preventing OMAP3 serial driver to take control of all UARTs

PATCH New Mika Westerberg 8 participants 22 messages Nov 30, 2009 – Dec 11, 2009

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

Hi Tony,

Current omap serial driver takes control of all 3 (4 on OMAP3640) UARTS. However, we have such a setup where UART2 for example is used by bluetooth driver. It uses the UART as non-standard way (there are some Nokia extensions to H4 protocol) so we cannot use the standard driver for driving the UART but have written special one for that purpose.

Question is: Is there any, upstreamable, way of preventing omap serial driver to do this? Currently this is done with custom #ifdef hackery to mach-omap2/serial.c. Alternative solution that comes into mind is to specify UART configuration in board files and let serial driver to use that instead of hard-coded one. Or do you have some nice alternatives?

Thanks,
MW

#1 lore raw
Quoted 9 lines · Mika Westerberg

Hi Tony,

Current omap serial driver takes control of all 3 (4 on OMAP3640) UARTS. However, we have such a setup where UART2 for example is used by bluetooth driver. It uses the UART as non-standard way (there are some Nokia extensions to H4 protocol) so we cannot use the standard driver for driving the UART but have written special one for that purpose.

Quoted 5 lines

Question is: Is there any, upstreamable, way of preventing omap serial driver to do this? Currently this is done with custom #ifdef hackery to mach-omap2/serial.c. Alternative solution that comes into mind is to specify UART configuration in board files and let serial driver to use that instead of hard-coded one. Or do you have some nice alternatives?

Previously (back around 2.6.28-rc8) in the board file, the omap_uart_config struct controlled which serial ports were enabled on startup. It was used in omap_serial_init, and it looks like that code went away with the following commit:
http://git.kernel.org/?p=linux/kernel/git/tmlind/linux-omap-2.6.git;a=blobdiff;f=arch/arm/mach-omap2/serial.c;h=2e17b57f5b23bb6703a2d621103585af1d8d729b;hp=555e735524381cbf8ef9f20d778ad81f9438e24e;hb=4355c41a635943d30e9396b95185314343dcb551;hpb=7e9ccf7776bb68b5367eb0bb35e519df62bea35c

I'm kinda in the same boat as I want to use some of the unused serial port pins for GPIO, but they are setup as serial ports....

Quoted 6 lines

Thanks,
MW
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#2 lore raw
Quoted 16 lines · Peter Barada
Quoted 9 lines · Mika Westerberg

Hi Tony,

Current omap serial driver takes control of all 3 (4 on OMAP3640) UARTS. However, we have such a setup where UART2 for example is used by bluetooth driver. It uses the UART as non-standard way (there are some Nokia extensions to H4 protocol) so we cannot use the standard driver for driving the UART but have written special one for that purpose.

Quoted 5 lines

Question is: Is there any, upstreamable, way of preventing omap serial driver to do this? Currently this is done with custom #ifdef hackery to mach-omap2/serial.c. Alternative solution that comes into mind is to specify UART configuration in board files and let serial driver to use that instead of hard-coded one. Or do you have some nice alternatives?

Quoted 8 lines

Previously (back around 2.6.28-rc8) in the board file, the omap_uart_config struct controlled which serial ports were enabled on startup.  It was used in omap_serial_init, and it looks like that code went away with the following commit:
http://git.kernel.org/?p=linux/kernel/git/tmlind/linux-omap-2.6.git;a=blobdiff;f=arch/arm/mach-omap2/serial.c;h=2e17b57f5b23bb6703a2d621103585af1d8d729b;hp=555e735524381cbf8ef9f20d778ad81f9438e24e;hb=4355c41a635943d30e9396b95185314343dcb551;hpb=7e9ccf7776bb68b5367eb0bb35e519df62bea35c

I'm kinda in the same boat as I want to use some of the unused serial port pins for GPIO, but they are setup as serial ports....

Not in mainlined yet, but I'm working on porting flattened device tree support to OMAP to solve exactly this sort of problem. Basically, instead of hard coding or #ifdeffing things, a data blob gets handed to the kernel at boot time telling it exactly what hardware is present in a consistent, parsable format. Device drivers then get probed based on data in the device tree. Here's some info on the approach:

http://www.elinux.org/Device_Trees

I expect to have my prototype ready for review mid-January, and most of the common code should be either merged or queued up in linux-next by that time.

Cheers,
g.

Signature · 6 lines
Grant Likely, B.Sc., P.Eng.
Secret Lab Technologies Ltd.
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
maintainer
#3 lore raw
Quoted 17 lines · Grant Likely
Quoted 16 lines · Peter Barada
Quoted 14 lines · Mika Westerberg

Hi Tony,

Current omap serial driver takes control of all 3 (4 on OMAP3640) UARTS. However, we have such a setup where UART2 for example is used by bluetooth driver. It uses the UART as non-standard way (there are some Nokia extensions to H4 protocol) so we cannot use the standard driver for driving the UART but have written special one for that purpose.

Question is: Is there any, upstreamable, way of preventing omap serial driver to do this? Currently this is done with custom #ifdef hackery to mach-omap2/serial.c. Alternative solution that comes into mind is to specify UART configuration in board files and let serial driver to use that instead of hard-coded one. Or do you have some nice alternatives?

Quoted 8 lines
Quoted 8 lines

Previously (back around 2.6.28-rc8) in the board file, the omap_uart_config struct controlled which serial ports were enabled on startup.  It was used in omap_serial_init, and it looks like that code went away with the following commit:
http://git.kernel.org/?p=linux/kernel/git/tmlind/linux-omap-2.6.git;a=blobdiff;f=arch/arm/mach-omap2/serial.c;h=2e17b57f5b23bb6703a2d621103585af1d8d729b;hp=555e735524381cbf8ef9f20d778ad81f9438e24e;hb=4355c41a635943d30e9396b95185314343dcb551;hpb=7e9ccf7776bb68b5367eb0bb35e519df62bea35c

I'm kinda in the same boat as I want to use some of the unused serial port pins for GPIO, but they are setup as serial ports....

Sounds like we need something back to specify the ports to use from board-*.c files. Kevin, got any comments?

Quoted 7 lines

Not in mainlined yet, but I'm working on porting flattened device tree support to OMAP to solve exactly this sort of problem. Basically, instead of hard coding or #ifdeffing things, a data blob gets handed to the kernel at boot time telling it exactly what hardware is present in a consistent, parsable format. Device drivers then get probed based on data in the device tree. Here's some info on the approach:

Quoted 5 lines

http://www.elinux.org/Device_Trees

I expect to have my prototype ready for review mid-January, and most of the common code should be either merged or queued up in linux-next by that time.

While device tree is a nice solution to some of the problems, it still leaves all the issues we already have with buggy and and outdated bootloaders. So we still need to properly initialize the devices in the kernel.

Just for reference, most of the omap bootloader bugs seem to be related to not muxing the pins right or using wrong timings for GPMC.

And then things that mostly change during the board development are the GPIO pins, but those can be easily rewritten in the board-*.c files based on the omap_rev.

But at least the device tree is a standard model, while the earlier omap tag approach was non-standard.

Peter, maybe you've already thought through all this.. But would it be possible to do lightweight device tree that we just use to populate the platform data?

Regards,

Tony
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#4 lore raw
Quoted 55 lines · Tony Lindgren
Quoted 17 lines · Grant Likely
Quoted 16 lines · Peter Barada
Quoted 14 lines · Mika Westerberg

Hi Tony,

Current omap serial driver takes control of all 3 (4 on OMAP3640) UARTS. However, we have such a setup where UART2 for example is used by bluetooth driver. It uses the UART as non-standard way (there are some Nokia extensions to H4 protocol) so we cannot use the standard driver for driving the UART but have written special one for that purpose.

Question is: Is there any, upstreamable, way of preventing omap serial driver to do this? Currently this is done with custom #ifdef hackery to mach-omap2/serial.c. Alternative solution that comes into mind is to specify UART configuration in board files and let serial driver to use that instead of hard-coded one. Or do you have some nice alternatives?

Quoted 8 lines
Quoted 8 lines

Previously (back around 2.6.28-rc8) in the board file, the omap_uart_config struct controlled which serial ports were enabled on startup. It was used in omap_serial_init, and it looks like that code went away with the following commit:
http://git.kernel.org/?p=linux/kernel/git/tmlind/linux-omap-2.6.git;a=blobdiff;f=arch/arm/mach-omap2/serial.c;h=2e17b57f5b23bb6703a2d621103585af1d8d729b;hp=555e735524381cbf8ef9f20d778ad81f9438e24e;hb=4355c41a635943d30e9396b95185314343dcb551;hpb=7e9ccf7776bb68b5367eb0bb35e519df62bea35c

I'm kinda in the same boat as I want to use some of the unused serial port pins for GPIO, but they are setup as serial ports....

Sounds like we need something back to specify the ports to use from board-*.c files. Kevin, got any comments?

Quoted 7 lines

Not in mainlined yet, but I'm working on porting flattened device tree support to OMAP to solve exactly this sort of problem. Basically, instead of hard coding or #ifdeffing things, a data blob gets handed to the kernel at boot time telling it exactly what hardware is present in a consistent, parsable format. Device drivers then get probed based on data in the device tree. Here's some info on the approach:

Quoted 5 lines

http://www.elinux.org/Device_Trees

I expect to have my prototype ready for review mid-January, and most of the common code should be either merged or queued up in linux-next by that time.

While device tree is a nice solution to some of the problems, it still leaves all the issues we already have with buggy and and outdated bootloaders. So we still need to properly initialize the devices in the kernel.

Just for reference, most of the omap bootloader bugs seem to be related to not muxing the pins right or using wrong timings for GPMC.

And then things that mostly change during the board development are the GPIO pins, but those can be easily rewritten in the board-*.c files based on the omap_rev.

Quoted 6 lines

But at least the device tree is a standard model, while the earlier omap tag approach was non-standard.

Peter, maybe you've already thought through all this.. But would it be possible to do lightweight device tree that we just use to populate the platform data?

One possibility is to pass to omap_serial_init() the omap_uart_config struct pointer to specify which parts are enabled. If a NULL is passed in, then enable all the ports available. Since omap_serial_early_init() was already called, the muxing would have to be cleaned up, but since the kernel should mux all the pins it uses, that shouldn't be a problem. omap_serial_init would now look something like(warning, coding on the fly - don't know if it will work as is):

void __init omap_serial_init(struct omap_uart_config *confptr)

{
	int i;

	for (i = 0; i < ARRAY_SIZE(omap_uart); i++) {
		struct omap_uart_state *uart = &omap_uart[i];
		struct platform_device *pdev = &uart->pdev;
		struct device *dev = &pdev->dev;

		/* Only enable
		if (!confptr || confptr->port_enabled & (1<<i)) {
			omap_uart_reset(uart);
			omap_uart_idle_init(uart);

			if (WARN_ON(platform_device_register(pdev)))
				continue;
			if ((cpu_is_omap34xx() && uart->padconf) ||
			    (uart->wk_en && uart->wk_mask)) {
				device_init_wakeup(dev, true);
				DEV_CREATE_FILE(dev, &dev_attr_sleep_timeout);
			}
		}
	}
}
Quoted 3 lines

Regards,

Tony

Signature · 2 lines
Peter Barada <[email protected]>
Logic Product Development, Inc.
maintainer
#5 lore raw
Quoted 87 lines · Peter Barada
Quoted 55 lines · Tony Lindgren
Quoted 17 lines · Grant Likely
Quoted 16 lines · Peter Barada
Quoted 14 lines · Mika Westerberg

Hi Tony,

Current omap serial driver takes control of all 3 (4 on OMAP3640) UARTS. However, we have such a setup where UART2 for example is used by bluetooth driver. It uses the UART as non-standard way (there are some Nokia extensions to H4 protocol) so we cannot use the standard driver for driving the UART but have written special one for that purpose.

Question is: Is there any, upstreamable, way of preventing omap serial driver to do this? Currently this is done with custom #ifdef hackery to mach-omap2/serial.c. Alternative solution that comes into mind is to specify UART configuration in board files and let serial driver to use that instead of hard-coded one. Or do you have some nice alternatives?

Quoted 8 lines
Quoted 8 lines

Previously (back around 2.6.28-rc8) in the board file, the omap_uart_config struct controlled which serial ports were enabled on startup. It was used in omap_serial_init, and it looks like that code went away with the following commit:
http://git.kernel.org/?p=linux/kernel/git/tmlind/linux-omap-2.6.git;a=blobdiff;f=arch/arm/mach-omap2/serial.c;h=2e17b57f5b23bb6703a2d621103585af1d8d729b;hp=555e735524381cbf8ef9f20d778ad81f9438e24e;hb=4355c41a635943d30e9396b95185314343dcb551;hpb=7e9ccf7776bb68b5367eb0bb35e519df62bea35c

I'm kinda in the same boat as I want to use some of the unused serial port pins for GPIO, but they are setup as serial ports....

Sounds like we need something back to specify the ports to use from board-*.c files. Kevin, got any comments?

Quoted 7 lines

Not in mainlined yet, but I'm working on porting flattened device tree support to OMAP to solve exactly this sort of problem. Basically, instead of hard coding or #ifdeffing things, a data blob gets handed to the kernel at boot time telling it exactly what hardware is present in a consistent, parsable format. Device drivers then get probed based on data in the device tree. Here's some info on the approach:

Quoted 5 lines

http://www.elinux.org/Device_Trees

I expect to have my prototype ready for review mid-January, and most of the common code should be either merged or queued up in linux-next by that time.

While device tree is a nice solution to some of the problems, it still leaves all the issues we already have with buggy and and outdated bootloaders. So we still need to properly initialize the devices in the kernel.

Just for reference, most of the omap bootloader bugs seem to be related to not muxing the pins right or using wrong timings for GPMC.

And then things that mostly change during the board development are the GPIO pins, but those can be easily rewritten in the board-*.c files based on the omap_rev.

Quoted 6 lines

But at least the device tree is a standard model, while the earlier omap tag approach was non-standard.

Peter, maybe you've already thought through all this.. But would it be possible to do lightweight device tree that we just use to populate the platform data?

One possibility is to pass to omap_serial_init() the omap_uart_config struct pointer to specify which parts are enabled. If a NULL is passed in, then enable all the ports available. Since omap_serial_early_init() was already called, the muxing would have to be cleaned up, but since the kernel should mux all the pins it uses, that shouldn't be a problem. omap_serial_init would now look something like(warning, coding on the fly - don't know if it will work as is):

void __init omap_serial_init(struct omap_uart_config *confptr)

{
	int i;

	for (i = 0; i < ARRAY_SIZE(omap_uart); i++) {
		struct omap_uart_state *uart = &omap_uart[i];
		struct platform_device *pdev = &uart->pdev;
		struct device *dev = &pdev->dev;

		/* Only enable
		if (!confptr || confptr->port_enabled & (1<<i)) {
			omap_uart_reset(uart);
			omap_uart_idle_init(uart);

			if (WARN_ON(platform_device_register(pdev)))
				continue;
Quoted 8 lines
			if ((cpu_is_omap34xx() && uart->padconf) ||
			    (uart->wk_en && uart->wk_mask)) {
				device_init_wakeup(dev, true);
				DEV_CREATE_FILE(dev, &dev_attr_sleep_timeout);
			}
		}
	}
}

Yeah, I guess something like that would be good. We may want to specify the optional other pins there too, typically it's just RX and TX pins.

It could also be directly struct plat_serial8250_port array we pass? Or does that make things more complicated to use the upcoming omap serial driver?

Regards,

Tony

#6 lore raw
Quoted 97 lines · ext Tony Lindgren
Quoted 87 lines · Peter Barada
Quoted 55 lines · Tony Lindgren
Quoted 17 lines · Grant Likely
Quoted 16 lines · Peter Barada
Quoted 14 lines · Mika Westerberg

Hi Tony,

Current omap serial driver takes control of all 3 (4 on OMAP3640) UARTS. However, we have such a setup where UART2 for example is used by bluetooth driver. It uses the UART as non-standard way (there are some Nokia extensions to H4 protocol) so we cannot use the standard driver for driving the UART but have written special one for that purpose.

Question is: Is there any, upstreamable, way of preventing omap serial driver to do this? Currently this is done with custom #ifdef hackery to mach-omap2/serial.c. Alternative solution that comes into mind is to specify UART configuration in board files and let serial driver to use that instead of hard-coded one. Or do you have some nice alternatives?

Quoted 8 lines
Quoted 8 lines

Previously (back around 2.6.28-rc8) in the board file, the omap_uart_config struct controlled which serial ports were enabled on startup. It was used in omap_serial_init, and it looks like that code went away with the following commit:
http://git.kernel.org/?p=linux/kernel/git/tmlind/linux-omap-2.6.git;a=blobdiff;f=arch/arm/mach-omap2/serial.c;h=2e17b57f5b23bb6703a2d621103585af1d8d729b;hp=555e735524381cbf8ef9f20d778ad81f9438e24e;hb=4355c41a635943d30e9396b95185314343dcb551;hpb=7e9ccf7776bb68b5367eb0bb35e519df62bea35c

I'm kinda in the same boat as I want to use some of the unused serial port pins for GPIO, but they are setup as serial ports....

Sounds like we need something back to specify the ports to use from board-*.c files. Kevin, got any comments?

Quoted 7 lines

Not in mainlined yet, but I'm working on porting flattened device tree support to OMAP to solve exactly this sort of problem. Basically, instead of hard coding or #ifdeffing things, a data blob gets handed to the kernel at boot time telling it exactly what hardware is present in a consistent, parsable format. Device drivers then get probed based on data in the device tree. Here's some info on the approach:

Quoted 5 lines

http://www.elinux.org/Device_Trees

I expect to have my prototype ready for review mid-January, and most of the common code should be either merged or queued up in linux-next by that time.

While device tree is a nice solution to some of the problems, it still leaves all the issues we already have with buggy and and outdated bootloaders. So we still need to properly initialize the devices in the kernel.

Just for reference, most of the omap bootloader bugs seem to be related to not muxing the pins right or using wrong timings for GPMC.

And then things that mostly change during the board development are the GPIO pins, but those can be easily rewritten in the board-*.c files based on the omap_rev.

Quoted 6 lines

But at least the device tree is a standard model, while the earlier omap tag approach was non-standard.

Peter, maybe you've already thought through all this.. But would it be possible to do lightweight device tree that we just use to populate the platform data?

One possibility is to pass to omap_serial_init() the omap_uart_config struct pointer to specify which parts are enabled. If a NULL is passed in, then enable all the ports available. Since omap_serial_early_init() was already called, the muxing would have to be cleaned up, but since the kernel should mux all the pins it uses, that shouldn't be a problem. omap_serial_init would now look something like(warning, coding on the fly - don't know if it will work as is):

void __init omap_serial_init(struct omap_uart_config *confptr)

{
	int i;

	for (i = 0; i < ARRAY_SIZE(omap_uart); i++) {
		struct omap_uart_state *uart = &omap_uart[i];
		struct platform_device *pdev = &uart->pdev;
		struct device *dev = &pdev->dev;

		/* Only enable
		if (!confptr || confptr->port_enabled & (1<<i)) {
			omap_uart_reset(uart);
			omap_uart_idle_init(uart);

			if (WARN_ON(platform_device_register(pdev)))
				continue;
Quoted 8 lines
			if ((cpu_is_omap34xx() && uart->padconf) ||
			    (uart->wk_en && uart->wk_mask)) {
				device_init_wakeup(dev, true);
				DEV_CREATE_FILE(dev, &dev_attr_sleep_timeout);
			}
		}
	}
}
Quoted 7 lines

Yeah, I guess something like that would be good. We may want to specify the optional other pins there too, typically it's just RX and TX pins.

It could also be directly struct plat_serial8250_port array we pass? Or does that make things more complicated to use the upcoming omap serial driver?

How about something like in the patch attached?

Then for example we would do in board-rx51.c:

	...
	omap_serial_init_port(2);

(we only use UART3 as serial port).

I quickly tested this with RX-51 and seems to work.

Thanks,
MW

---

From: Mika Westerberg <[email protected]>
Date: Tue, 1 Dec 2009 12:54:21 +0200
Subject: [PATCH] OMAP3: serial - allow platforms specify which UARTs to initialize

This patch adds new function: omap_serial_init_port(port) that can be used to initialize only selected UARTs as serial ports. Platforms can then in their board files call this function instead of omap_serial_init() if they don't want to use all UARTs as serial ports.

---
 arch/arm/mach-omap2/serial.c             |   59 +++++++++++++++++++++++-------
 arch/arm/plat-omap/include/plat/serial.h |    1 +
 2 files changed, 46 insertions(+), 14 deletions(-)
arch/arm/mach-omap2/serial.c +45 -14
index 2e17b57..fe46560 100644
@@ -631,24 +631,55 @@ void __init omap_serial_early_init(void)
 	}
 }
 
+/**
+ * omap_serial_init_port() - initialize single serial port
+ * @port: serial port number (0-3)
+ *
+ * This function initialies serial driver for given @port only.
+ * Platforms can call this function instead of omap_serial_init()
+ * if they don't plan to use all available UARTs as serial ports.
+ *
+ * Don't mix calls to omap_serial_init_port() and omap_serial_init(),
+ * use only one of the two.
+ */
+void __init omap_serial_init_port(int port)
+{
+	struct omap_uart_state *uart;
+	struct platform_device *pdev;
+	struct device *dev;
+
+	BUG_ON(port < 0);
+	BUG_ON(port >= ARRAY_SIZE(omap_uart));
+
+	uart = &omap_uart[port];
+	pdev = &uart->pdev;
+	dev = &pdev->dev;
+
+	omap_uart_reset(uart);
+	omap_uart_idle_init(uart);
+
+	if (WARN_ON(platform_device_register(pdev)))
+		return;
+
+	if ((cpu_is_omap34xx() && uart->padconf) ||
+	    (uart->wk_en && uart->wk_mask)) {
+		device_init_wakeup(dev, true);
+		DEV_CREATE_FILE(dev, &dev_attr_sleep_timeout);
+	}
+}
+
+/**
+ * omap_serial_init() - intialize all supported serial ports
+ *
+ * Initializes all available UARTs as serial ports. Platforms
+ * can call this function when they want to have default behaviour
+ * for serial ports (e.g initialize them all as serial ports).
+ */
 void __init omap_serial_init(void)
 {
 	int i;
 
 	for (i = 0; i < ARRAY_SIZE(omap_uart); i++) {
-		struct omap_uart_state *uart = &omap_uart[i];
-		struct platform_device *pdev = &uart->pdev;
-		struct device *dev = &pdev->dev;
-
-		omap_uart_reset(uart);
-		omap_uart_idle_init(uart);
-
-		if (WARN_ON(platform_device_register(pdev)))
-			continue;
-		if ((cpu_is_omap34xx() && uart->padconf) ||
-		    (uart->wk_en && uart->wk_mask)) {
-			device_init_wakeup(dev, true);
-			DEV_CREATE_FILE(dev, &dev_attr_sleep_timeout);
-		}
+		omap_serial_init_port(i);
 	}
 }
arch/arm/plat-omap/include/plat/serial.h +1 -0
index 9951345..f5a4a92 100644
@@ -53,6 +53,7 @@
 #ifndef __ASSEMBLER__
 extern void __init omap_serial_early_init(void);
 extern void omap_serial_init(void);
+extern void omap_serial_init_port(int port);
 extern int omap_uart_can_sleep(void);
 extern void omap_uart_check_wakeup(void);
 extern void omap_uart_prepare_suspend(void);
Signature · 1 line
1.5.6.5
#7 lore raw

[sorry for being late to the party, I've been on vacation]

Mika Westerberg <[email protected]> writes:

[...]

Quoted 10 lines

How about something like in the patch attached?

Then for example we would do in board-rx51.c:

	...
	omap_serial_init_port(2);

(we only use UART3 as serial port).

I quickly tested this with RX-51 and seems to work.

I like this approach.

I'm against going back to the previsus platform_data extending approach since it requires doing different stuff when using 8250 or the upcoming omap-serial driver.

Quoted 7 lines

---

From: Mika Westerberg <[email protected]>
Date: Tue, 1 Dec 2009 12:54:21 +0200
Subject: [PATCH] OMAP3: serial - allow platforms specify which UARTs to initialize

Quoted 6 lines

This patch adds new function: omap_serial_init_port(port) that can be used to initialize only selected UARTs as serial ports. Platforms can then in their board files call this function instead of omap_serial_init() if they don't want to use all UARTs as serial ports.

Quoted 98 lines
---
 arch/arm/mach-omap2/serial.c             |   59 +++++++++++++++++++++++-------
 arch/arm/plat-omap/include/plat/serial.h |    1 +
 2 files changed, 46 insertions(+), 14 deletions(-)
arch/arm/mach-omap2/serial.c +45 -14
index 2e17b57..fe46560 100644
@@ -631,24 +631,55 @@ void __init omap_serial_early_init(void)
 	}
 }
 
+/**
+ * omap_serial_init_port() - initialize single serial port
+ * @port: serial port number (0-3)
+ *
+ * This function initialies serial driver for given @port only.
+ * Platforms can call this function instead of omap_serial_init()
+ * if they don't plan to use all available UARTs as serial ports.
+ *
+ * Don't mix calls to omap_serial_init_port() and omap_serial_init(),
+ * use only one of the two.
+ */
+void __init omap_serial_init_port(int port)
+{
+	struct omap_uart_state *uart;
+	struct platform_device *pdev;
+	struct device *dev;
+
+	BUG_ON(port < 0);
+	BUG_ON(port >= ARRAY_SIZE(omap_uart));
+
+	uart = &omap_uart[port];
+	pdev = &uart->pdev;
+	dev = &pdev->dev;
+
+	omap_uart_reset(uart);
+	omap_uart_idle_init(uart);
+
+	if (WARN_ON(platform_device_register(pdev)))
+		return;
+
+	if ((cpu_is_omap34xx() && uart->padconf) ||
+	    (uart->wk_en && uart->wk_mask)) {
+		device_init_wakeup(dev, true);
+		DEV_CREATE_FILE(dev, &dev_attr_sleep_timeout);
+	}
+}
+
+/**
+ * omap_serial_init() - intialize all supported serial ports
+ *
+ * Initializes all available UARTs as serial ports. Platforms
+ * can call this function when they want to have default behaviour
+ * for serial ports (e.g initialize them all as serial ports).
+ */
 void __init omap_serial_init(void)
 {
 	int i;
 
 	for (i = 0; i < ARRAY_SIZE(omap_uart); i++) {
-		struct omap_uart_state *uart = &omap_uart[i];
-		struct platform_device *pdev = &uart->pdev;
-		struct device *dev = &pdev->dev;
-
-		omap_uart_reset(uart);
-		omap_uart_idle_init(uart);
-
-		if (WARN_ON(platform_device_register(pdev)))
-			continue;
-		if ((cpu_is_omap34xx() && uart->padconf) ||
-		    (uart->wk_en && uart->wk_mask)) {
-			device_init_wakeup(dev, true);
-			DEV_CREATE_FILE(dev, &dev_attr_sleep_timeout);
-		}
+		omap_serial_init_port(i);
 	}
 }
arch/arm/plat-omap/include/plat/serial.h +1 -0
index 9951345..f5a4a92 100644
@@ -53,6 +53,7 @@
 #ifndef __ASSEMBLER__
 extern void __init omap_serial_early_init(void);
 extern void omap_serial_init(void);
+extern void omap_serial_init_port(int port);
 extern int omap_uart_can_sleep(void);
 extern void omap_uart_check_wakeup(void);
 extern void omap_uart_prepare_suspend(void);
Signature · 6 lines
1.5.6.5

--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#8 lore raw

PATCH 1/1 OMAP3: serial - allow platforms specify which UARTs to initialize

From: Mika Westerberg <[email protected]>

This patch adds new function: omap_serial_init_port(port) that can be used to initialize only selected UARTs as serial ports. Platforms can then in their board files call this function instead of omap_serial_init() if they don't want to use all UARTs as serial ports.

---
I refreshed this against latest l-o master and added Kevins SoB. Tested with RX-51.

 arch/arm/mach-omap2/serial.c             |   78 ++++++++++++++++++++---------
 arch/arm/plat-omap/include/plat/serial.h |    1 +
 2 files changed, 55 insertions(+), 24 deletions(-)
arch/arm/mach-omap2/serial.c +54 -24
index e853115..eb2ba6e 100644
@@ -649,34 +649,64 @@ void __init omap_serial_early_init(void)
 	}
 }
 
-void __init omap_serial_init(void)
+/**
+ * omap_serial_init_port() - initialize single serial port
+ * @port: serial port number (0-3)
+ *
+ * This function initialies serial driver for given @port only.
+ * Platforms can call this function instead of omap_serial_init()
+ * if they don't plan to use all available UARTs as serial ports.
+ *
+ * Don't mix calls to omap_serial_init_port() and omap_serial_init(),
+ * use only one of the two.
+ */
+void __init omap_serial_init_port(int port)
 {
-	int i;
+	struct omap_uart_state *uart;
+	struct platform_device *pdev;
+	struct device *dev;
 
-	for (i = 0; i < ARRAY_SIZE(omap_uart); i++) {
-		struct omap_uart_state *uart = &omap_uart[i];
-		struct platform_device *pdev = &uart->pdev;
-		struct device *dev = &pdev->dev;
+	BUG_ON(port < 0);
+	BUG_ON(port >= ARRAY_SIZE(omap_uart));
 
-		omap_uart_reset(uart);
-		omap_uart_idle_init(uart);
+	uart = &omap_uart[port];
+	pdev = &uart->pdev;
+	dev = &pdev->dev;
 
-		if (WARN_ON(platform_device_register(pdev)))
-			continue;
-		if ((cpu_is_omap34xx() && uart->padconf) ||
-		    (uart->wk_en && uart->wk_mask)) {
-			device_init_wakeup(dev, true);
-			DEV_CREATE_FILE(dev, &dev_attr_sleep_timeout);
-		}
+	omap_uart_reset(uart);
+	omap_uart_idle_init(uart);
 
-		/* omap44xx: Never read empty UART fifo
-		 * omap3xxx: Never read empty UART fifo on UARTs
-		 * with IP rev >=0x52
-		 */
-		if (cpu_is_omap44xx())
-			uart->p->serial_in = serial_in_override;
-		else if ((serial_read_reg(uart->p, UART_OMAP_MVER) & 0xFF)
-				>= UART_OMAP_NO_EMPTY_FIFO_READ_IP_REV)
-			uart->p->serial_in = serial_in_override;
+	if (WARN_ON(platform_device_register(pdev)))
+		return;
+
+	if ((cpu_is_omap34xx() && uart->padconf) ||
+	    (uart->wk_en && uart->wk_mask)) {
+		device_init_wakeup(dev, true);
+		DEV_CREATE_FILE(dev, &dev_attr_sleep_timeout);
 	}
+
+	/* omap44xx: Never read empty UART fifo
+	 * omap3xxx: Never read empty UART fifo on UARTs
+	 * with IP rev >=0x52
+	 */
+	if (cpu_is_omap44xx())
+		uart->p->serial_in = serial_in_override;
+	else if ((serial_read_reg(uart->p, UART_OMAP_MVER) & 0xFF)
+			>= UART_OMAP_NO_EMPTY_FIFO_READ_IP_REV)
+		uart->p->serial_in = serial_in_override;
+}
+
+/**
+ * omap_serial_init() - intialize all supported serial ports
+ *
+ * Initializes all available UARTs as serial ports. Platforms
+ * can call this function when they want to have default behaviour
+ * for serial ports (e.g initialize them all as serial ports).
+ */
+void __init omap_serial_init(void)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(omap_uart); i++)
+		omap_serial_init_port(i);
 }
arch/arm/plat-omap/include/plat/serial.h +1 -0
index 9951345..f5a4a92 100644
@@ -53,6 +53,7 @@
 #ifndef __ASSEMBLER__
 extern void __init omap_serial_early_init(void);
 extern void omap_serial_init(void);
+extern void omap_serial_init_port(int port);
 extern int omap_uart_can_sleep(void);
 extern void omap_uart_check_wakeup(void);
 extern void omap_uart_prepare_suspend(void);
Signature · 1 line
1.5.6.5
maintainer
#9 lore raw

PATCH 1/1 OMAP3: serial - allow platforms specify which UARTs to

This patch has been applied to the linux-omap
by youw fwiendly patch wobot.

Branch in linux-omap: for-next

Initial commit ID (Likely to change): e663f3a6a9b58cff5b2f67267519ba1f75122b3b

PatchWorks
http://patchwork.kernel.org/patch/66232/

Git (Likely to change, and takes a while to get mirrored)
http://git.kernel.org/?p=linux/kernel/git/tmlind/linux-omap-2.6.git;a=commit;h=e663f3a6a9b58cff5b2f67267519ba1f75122b3b

maintainer
#10 lore raw
In reply to Tony Lindgren
Quoted 27 lines · Tony Lindgren
Quoted 7 lines · Grant Likely

Not in mainlined yet, but I'm working on porting flattened device tree support to OMAP to solve exactly this sort of problem. Basically, instead of hard coding or #ifdeffing things, a data blob gets handed to the kernel at boot time telling it exactly what hardware is present in a consistent, parsable format. Device drivers then get probed based on data in the device tree. Here's some info on the approach:

Quoted 5 lines

http://www.elinux.org/Device_Trees

I expect to have my prototype ready for review mid-January, and most of the common code should be either merged or queued up in linux-next by that time.

While device tree is a nice solution to some of the problems, it still leaves all the issues we already have with buggy and and outdated bootloaders. So we still need to properly initialize the devices in the kernel.

Just for reference, most of the omap bootloader bugs seem to be related to not muxing the pins right or using wrong timings for GPMC.

And then things that mostly change during the board development are the GPIO pins, but those can be easily rewritten in the board-*.c files based on the omap_rev.

Quoted 6 lines

But at least the device tree is a standard model, while the earlier omap tag approach was non-standard.

Peter, maybe you've already thought through all this.. But would it be possible to do lightweight device tree that we just use to populate the platform data?

Sorry, I meant to ask Grant this question as Grant (and not Peter) is working on the device tree.

Tony

#11 lore raw
In reply to Tony Lindgren
Quoted 11 lines · Tony Lindgren
Quoted 5 lines · Grant Likely

http://www.elinux.org/Device_Trees

I expect to have my prototype ready for review mid-January, and most of the common code should be either merged or queued up in linux-next by that time.

While device tree is a nice solution to some of the problems, it still leaves all the issues we already have with buggy and and outdated bootloaders. So we still need to properly initialize the devices in the kernel.

Yes, buggy firmware is still a problem. However, if the required configuration is described in the device tree data, then for some things the driver can handle setting up the device and the amount of board-specific code can be reduced.

Quoted 7 lines

Just for reference, most of the omap bootloader bugs seem to be related to not muxing the pins right or using wrong timings for GPMC.

And then things that mostly change during the board development are the GPIO pins, but those can be easily rewritten in the board-*.c files based on the omap_rev.

Quoted 6 lines

But at least the device tree is a standard model, while the earlier omap tag approach was non-standard.

Peter, maybe you've already thought through all this.. But would it be possible to do lightweight device tree that we just use to populate the platform data?

This is completely possible. Just having the device tree available doesn't force the kernel to use it for everything. I've found it useful to start small and add things as I need them. Most important thing to remember is to follow the documented & established device tree conventions so that common code can understand it.

Oh, and speaking of GPIOs, there is a binding for describing GPIO pin connections in the device tree:
Documentation/powerpc/dts-bindings/gpio/gpio.txt

Cheers,
g.

Signature · 2 lines
Grant Likely, B.Sc., P.Eng.
Secret Lab Technologies Ltd.
#12 lore raw
Quoted 3 lines · Grant Likely

Oh, and speaking of GPIOs, there is a binding for describing GPIO pin connections in the device tree:
Documentation/powerpc/dts-bindings/gpio/gpio.txt

That binding is more about documenting a bank of GPIO pins, while chips like OMAP will need something a bit more flexible.

Given the flexibility and complexity of configuring the pins on mobile-oriented SoCs where each function can come out a variety of different pins, and many pins can 3 or 4 functions associated with them (selected by software config), it will definitely bring a new level of complexity to the device tree descriptions.

I'm definitely not saying that it is impossible, but it might take a little work to hash out a binding that everyone will be happy with.

-Olof

#13 lore raw
Quoted 6 lines · Olof Johansson
Quoted 3 lines · Grant Likely

Oh, and speaking of GPIOs, there is a binding for describing GPIO pin connections in the device tree:
Documentation/powerpc/dts-bindings/gpio/gpio.txt

Quoted 8 lines

That binding is more about documenting a bank of GPIO pins, while chips like OMAP will need something a bit more flexible.

Given the flexibility and complexity of configuring the pins on mobile-oriented SoCs where each function can come out a variety of different pins, and many pins can 3 or 4 functions associated with them (selected by software config), it will definitely bring a new level of complexity to the device tree descriptions.

Ah, you're talking about pin muxing configuration, right? Yes, that GPIO binding deals with controllers, not pin mux. Pin mux is very much an SoC specific thing that isn't mapped easily to a generic binding.

On the 5200, I haven't attempted to describe pin-mux in the device tree at all, and have either expected firmware to set it up correctly, or fixed it up in the platform code.

Quoted 2 lines

I'm definitely not saying that it is impossible, but it might take a little work to hash out a binding that everyone will be happy with.

Right.

g.

Signature · 2 lines
Grant Likely, B.Sc., P.Eng.
Secret Lab Technologies Ltd.
#14 lore raw
Quoted 4 lines · Grant Likely

Ah, you're talking about pin muxing configuration, right? Yes, that GPIO binding deals with controllers, not pin mux. Pin mux is very much an SoC specific thing that isn't mapped easily to a generic binding.

Yep.

Quoted 3 lines

On the 5200, I haven't attempted to describe pin-mux in the device tree at all, and have either expected firmware to set it up correctly, or fixed it up in the platform code.

Yeah. And it's one of the things Tony commented on that firmware tends to get wrong, seems like people aren't doing complete mux configs in u-boot, etc.

So, if it needs to be fixed up in platform code, there will (likely) be need for board-specific code there anyway. A bummer, since the device tree would otherwise make it real easy to bring up new boards without kernel code changes.

-Olof

#15 lore raw
Quoted 13 lines · Olof Johansson
Quoted 4 lines · Grant Likely

Ah, you're talking about pin muxing configuration, right?  Yes, that GPIO binding deals with controllers, not pin mux.  Pin mux is very much an SoC specific thing that isn't mapped easily to a generic binding.

Yep.

Quoted 3 lines

On the 5200, I haven't attempted to describe pin-mux in the device tree at all, and have either expected firmware to set it up correctly, or fixed it up in the platform code.

Quoted 8 lines

Yeah. And it's one of the things Tony commented on that firmware tends to get wrong, seems like people aren't doing complete mux configs in u-boot, etc.

So, if it needs to be fixed up in platform code, there will (likely) be need for board-specific code there anyway. A bummer, since the device tree would otherwise make it real easy to bring up new boards without kernel code changes.

I didn't create a binding on the 5200 because I actually see very little buggy firmware in that regard (partially because I kept telling people to go fix their firmware). :-) If it ends up being the norm that the kernel has to fix it for a given SoC, then I would create an SoC-specific binding for pin mux configuration in the device tree so that some degree of common code can still fix it up.

It should be feasible for board-specific code to be the exception, not the rule.

Cheers,
g.

Signature · 6 lines
Grant Likely, B.Sc., P.Eng.
Secret Lab Technologies Ltd.
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
maintainer
#16 lore raw
Quoted 23 lines · Grant Likely
Quoted 13 lines · Olof Johansson
Quoted 4 lines · Grant Likely

Ah, you're talking about pin muxing configuration, right?  Yes, that GPIO binding deals with controllers, not pin mux.  Pin mux is very much an SoC specific thing that isn't mapped easily to a generic binding.

Yep.

Quoted 3 lines

On the 5200, I haven't attempted to describe pin-mux in the device tree at all, and have either expected firmware to set it up correctly, or fixed it up in the platform code.

Quoted 8 lines

Yeah. And it's one of the things Tony commented on that firmware tends to get wrong, seems like people aren't doing complete mux configs in u-boot, etc.

So, if it needs to be fixed up in platform code, there will (likely) be need for board-specific code there anyway. A bummer, since the device tree would otherwise make it real easy to bring up new boards without kernel code changes.

Quoted 8 lines

I didn't create a binding on the 5200 because I actually see very little buggy firmware in that regard (partially because I kept telling people to go fix their firmware). :-) If it ends up being the norm that the kernel has to fix it for a given SoC, then I would create an SoC-specific binding for pin mux configuration in the device tree so that some degree of common code can still fix it up.

It should be feasible for board-specific code to be the exception, not the rule.

The problem with pin multiplexing is that development boards such as Beagle don't know the desired configuration in the bootloader. There are pins available for connecting external hardware, such as I2C, SPI, or whatever to the 16-bit GPMC bus. Also, the bootloaders tend not to care about pin multiplexing for power management.

We now have kernel cmdline override for pin multiplexing, but that does not help with adding platform_data for custom I2C or SPI devices.

Just something to keep in mind, I still think we can benefit from the open firmware support in various ways :)

Regards,

Tony
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

maintainer
#17 lore raw
In reply to Grant Likely
Quoted 2 lines · Grant Likely
Quoted 1 line · Tony Lindgren

* Grant Likely <[email protected]> [091130 09:01]:

<snip>

Quoted 10 lines
Quoted 4 lines

maybe you've already thought through all this.. But would it be possible to do lightweight device tree that we just use to populate the platform data?

This is completely possible. Just having the device tree available doesn't force the kernel to use it for everything. I've found it useful to start small and add things as I need them. Most important thing to remember is to follow the documented & established device tree conventions so that common code can understand it.

OK, sounds good to me.

Tony

#18 lore raw
Quoted 11 lines · ext Tony Lindgren
Quoted 2 lines · Grant Likely
Quoted 1 line · Tony Lindgren

* Grant Likely <[email protected]> [091130 09:01]:

<snip>

Quoted 5 lines
Quoted 4 lines

maybe you've already thought through all this.. But would it be possible to do lightweight device tree that we just use to populate the platform data?

Quoted 7 lines
Quoted 5 lines

This is completely possible. Just having the device tree available doesn't force the kernel to use it for everything. I've found it useful to start small and add things as I need them. Most important thing to remember is to follow the documented & established device tree conventions so that common code can understand it.

OK, sounds good to me.

Hi,

This device tree stuff sounds like very cool way of doing things. Hope it is ready soon :)

Meanwhile, would it be OK to implement something to get the serial driver taking control of the all the UARTs? Any comments on adding new function to mach-omap2/serial.c: omap_serial_init_port(int port) that could be used from board files instead of omap_serial_init()?

Thanks,
MW

#19 lore raw
Quoted 22 lines · Mika Westerberg
Quoted 11 lines · ext Tony Lindgren
Quoted 2 lines · Grant Likely
Quoted 1 line · Tony Lindgren

* Grant Likely <[email protected]> [091130 09:01]:

<snip>

Quoted 5 lines
Quoted 4 lines

maybe you've already thought through all this.. But would it be possible to do lightweight device tree that we just use to populate the platform data?

Quoted 7 lines
Quoted 5 lines

This is completely possible. Just having the device tree available doesn't force the kernel to use it for everything. I've found it useful to start small and add things as I need them. Most important thing to remember is to follow the documented & established device tree conventions so that common code can understand it.

OK, sounds good to me.

Hi,

Quoted 7 lines

This device tree stuff sounds like very cool way of doing things. Hope it is ready soon :)

Meanwhile, would it be OK to implement something to get the serial driver taking control of the all the UARTs? Any comments on adding new function to mach-omap2/serial.c: omap_serial_init_port(int port) that could be used from board files instead of omap_serial_init()?

Device tree is really promising, but we do need an alternative method anyway, because not everyone will use the device tree. IOW, device tree does not exist yet, and will not be mandatory, so an alternative is required.

Signature · 7 lines
Best Regards,
Artem Bityutskiy (Артём Битюцкий)

--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
maintainer
#20 lore raw
Quoted 31 lines · Artem Bityutskiy
Quoted 22 lines · Mika Westerberg
Quoted 11 lines · ext Tony Lindgren
Quoted 2 lines · Grant Likely
Quoted 1 line · Tony Lindgren

* Grant Likely <[email protected]> [091130 09:01]:

<snip>

Quoted 5 lines
Quoted 4 lines

maybe you've already thought through all this.. But would it be possible to do lightweight device tree that we just use to populate the platform data?

Quoted 7 lines
Quoted 5 lines

This is completely possible. Just having the device tree available doesn't force the kernel to use it for everything. I've found it useful to start small and add things as I need them. Most important thing to remember is to follow the documented & established device tree conventions so that common code can understand it.

OK, sounds good to me.

Hi,

Quoted 7 lines

This device tree stuff sounds like very cool way of doing things. Hope it is ready soon :)

Meanwhile, would it be OK to implement something to get the serial driver taking control of the all the UARTs? Any comments on adding new function to mach-omap2/serial.c: omap_serial_init_port(int port) that could be used from board files instead of omap_serial_init()?

Quoted 4 lines

Device tree is really promising, but we do need an alternative method anyway, because not everyone will use the device tree. IOW, device tree does not exist yet, and will not be mandatory, so an alternative is required.

Right, it sounds like the device tree won't solve all the issues.

But before we change anything with the mach-omap2/serial.c, let's wait until we hear comments from Kevin.

Regards,

Tony

#21 lore raw
Quoted 37 lines · Tony Lindgren
Quoted 31 lines · Artem Bityutskiy
Quoted 22 lines · Mika Westerberg
Quoted 11 lines · ext Tony Lindgren
Quoted 2 lines · Grant Likely
Quoted 1 line · Tony Lindgren

* Grant Likely <[email protected]> [091130 09:01]:

<snip>

Quoted 5 lines
Quoted 4 lines

maybe you've already thought through all this.. But would it be possible to do lightweight device tree that we just use to populate the platform data?

Quoted 7 lines
Quoted 5 lines

This is completely possible. Just having the device tree available doesn't force the kernel to use it for everything. I've found it useful to start small and add things as I need them. Most important thing to remember is to follow the documented & established device tree conventions so that common code can understand it.

OK, sounds good to me.

Hi,

Quoted 7 lines

This device tree stuff sounds like very cool way of doing things. Hope it is ready soon :)

Meanwhile, would it be OK to implement something to get the serial driver taking control of the all the UARTs? Any comments on adding new function to mach-omap2/serial.c: omap_serial_init_port(int port) that could be used from board files instead of omap_serial_init()?

Quoted 4 lines

Device tree is really promising, but we do need an alternative method anyway, because not everyone will use the device tree. IOW, device tree does not exist yet, and will not be mandatory, so an alternative is required.

Quoted 4 lines

Right, it sounds like the device tree won't solve all the issues.

But before we change anything with the mach-omap2/serial.c, let's wait until we hear comments from Kevin.

Tony, Kevin,

Any comments on this?

Thanks,
MW