[PATCH 0/2] tty: serial: Add earlycon support to MT8173 SoC

This series add earlycon support to Mediatek MT8173 SoC. Although 8250 driver
already has earlycon support, but Mediatek UART has highspeed register that
not in 8250_early.c, so I port earlycon in 8250_mtk.c, and add stdout device
tree support.

This patchset is based on 3.18, and my basic support for Mediatek MT8173 Soc [1].

[1] http://lists.infradead.org/pipermail/linux-arm-kernel/2014-December/311238.html

Eddie Huang (2):
  tty: serial: 8250_mtk: Add earlycon
  arm64: dts: Add mediatek MT8173 earlycon support

 arch/arm64/boot/dts/mt8173-evb.dts |  4 +++-
 drivers/tty/serial/8250/8250_mtk.c | 33 +++++++++++++++++++++++++++++++++
 2 files changed, 36 insertions(+), 1 deletion(-)

--
1.8.1.1.dirty

Add earlycon support to mediatek MT8173 evaluation board dts.

Signed-off-by: Eddie Huang <eddie.huang@mediatek.com>
---
 arch/arm64/boot/dts/mt8173-evb.dts | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/arch/arm64/boot/dts/mt8173-evb.dts b/arch/arm64/boot/dts/mt8173-evb.dts
index b8b2621..b8802fd 100644
--- a/arch/arm64/boot/dts/mt8173-evb.dts
+++ b/arch/arm64/boot/dts/mt8173-evb.dts

@@ -30,5 +30,7 @@
 		reg = <0 0x40000000 0 0x80000000>;
 	};
 
-	chosen { };
+	chosen {
+		linux,stdout-path = &uart0;
+	};
 };

-- 
1.8.1.1.dirty

Mediatek UART has highspeed register, but 8250_early.c doesn't
support this, so add earlycon in 8250_mtk.c

Signed-off-by: Eddie Huang <eddie.huang@mediatek.com>
---
 drivers/tty/serial/8250/8250_mtk.c | 33 +++++++++++++++++++++++++++++++++
 1 file changed, 33 insertions(+)

diff --git a/drivers/tty/serial/8250/8250_mtk.c b/drivers/tty/serial/8250/8250_mtk.c
index de7aae5..65dd569 100644
--- a/drivers/tty/serial/8250/8250_mtk.c
+++ b/drivers/tty/serial/8250/8250_mtk.c

@@ -23,6 +23,7 @@
 #include <linux/pm_runtime.h>
 #include <linux/serial_8250.h>
 #include <linux/serial_reg.h>
+#include <linux/console.h>
 
 #include "8250.h"
 

@@ -289,6 +290,38 @@ static struct platform_driver mtk8250_platform_driver = {
 };
 module_platform_driver(mtk8250_platform_driver);
 
+#define BOTH_EMPTY (UART_LSR_TEMT | UART_LSR_THRE)
+
+static void __init mtk8250_serial_putc(struct uart_port *port, int c)
+{
+	while ((readl(port->membase + (UART_LSR << 2)) & BOTH_EMPTY) !=
+		BOTH_EMPTY)
+		;
+	writel(c, port->membase + (UART_TX << 2));
+}
+
+static void __init early_mtk8250_write(struct console *console,
+					const char *s, unsigned int count)
+{
+	struct earlycon_device *dev = console->data;
+
+	uart_console_write(&dev->port, s, count, mtk8250_serial_putc);
+}
+
+static int __init early_mtk8250_setup(struct earlycon_device *device,
+					const char *options)
+{
+	if (!device->port.membase)
+		return -ENODEV;
+
+	device->con->write = early_mtk8250_write;
+
+	return 0;
+}
+
+EARLYCON_DECLARE(mtk8250, early_mtk8250_setup);
+OF_EARLYCON_DECLARE(mtk8250, "mediatek,mt6577-uart", early_mtk8250_setup);
+
 MODULE_AUTHOR("Matthias Brugger");
 MODULE_LICENSE("GPL");
 MODULE_DESCRIPTION("Mediatek 8250 serial port driver");

-- 
1.8.1.1.dirty

2014-12-18 9:33 GMT+01:00 Eddie Huang [off-list ref]:

Please delete newline before the semicolon.

+       writel(c, port->membase + (UART_TX << 2));
+}
+
+static void __init early_mtk8250_write(struct console *console,
+                                       const char *s, unsigned int count)
+{
+       struct earlycon_device *dev = console->data;
+
+       uart_console_write(&dev->port, s, count, mtk8250_serial_putc);
+}
+
+static int __init early_mtk8250_setup(struct earlycon_device *device,
+                                       const char *options)
+{
+       if (!device->port.membase)
+               return -ENODEV;
+
+       device->con->write = early_mtk8250_write;
+
+       return 0;
+}
+
+EARLYCON_DECLARE(mtk8250, early_mtk8250_setup);
+OF_EARLYCON_DECLARE(mtk8250, "mediatek,mt6577-uart", early_mtk8250_setup);
+
 MODULE_AUTHOR("Matthias Brugger");
 MODULE_LICENSE("GPL");
 MODULE_DESCRIPTION("Mediatek 8250 serial port driver");
--
1.8.1.1.dirty

-- 
motzblog.wordpress.com

On Thursday 18 December 2014 09:48:31 Matthias Brugger wrote:

quoted

+static void __init mtk8250_serial_putc(struct uart_port *port, int c)
+{
+       while ((readl(port->membase + (UART_LSR << 2)) & BOTH_EMPTY) !=
+               BOTH_EMPTY)
+               ;

Please delete newline before the semicolon.

No, that newline is best kept in there for clarity. An even better way
to write this would be using cpu_relax():

	while ((readl(port->membase + (UART_LSR << 2)) & BOTH_EMPTY) != BOTH_EMPTY)
		cpu_relax();

which the kernel convention for doing busy-loops.

	Arnd

On Thu, Dec 18, 2014 at 2:33 AM, Eddie Huang [off-list ref] wrote:

Mediatek UART has highspeed register, but 8250_early.c doesn't
support this, so add earlycon in 8250_mtk.c

I don't see any highspeed register setup here. More generically,
aren't you just skipping any UART setup? That may be useful on other
platforms with 8250s, too. With the kernel command line version, you
could perhaps add a "noinit" flag. The DT case is harder, and I'm not
sure how we should handle that. We could perhaps add a
"stdout-path-initialized" flag to chosen.

Is there a reason early_serial8250_write can't work for you other than
it is currently static?

Rob

+{
+       struct earlycon_device *dev = console->data;
+
+       uart_console_write(&dev->port, s, count, mtk8250_serial_putc);
+}
+
+static int __init early_mtk8250_setup(struct earlycon_device *device,
+                                       const char *options)
+{
+       if (!device->port.membase)
+               return -ENODEV;
+
+       device->con->write = early_mtk8250_write;
+
+       return 0;
+}
+
+EARLYCON_DECLARE(mtk8250, early_mtk8250_setup);
+OF_EARLYCON_DECLARE(mtk8250, "mediatek,mt6577-uart", early_mtk8250_setup);
+
 MODULE_AUTHOR("Matthias Brugger");
 MODULE_LICENSE("GPL");
 MODULE_DESCRIPTION("Mediatek 8250 serial port driver");
--
1.8.1.1.dirty

Hi Rob,

On Fri, 2014-12-26 at 15:24 -0600, Rob Herring wrote:

On Thu, Dec 18, 2014 at 2:33 AM, Eddie Huang [off-list ref] wrote:

quoted

Mediatek UART has highspeed register, but 8250_early.c doesn't
support this, so add earlycon in 8250_mtk.c

I don't see any highspeed register setup here. More generically,
aren't you just skipping any UART setup? That may be useful on other
platforms with 8250s, too. With the kernel command line version, you
could perhaps add a "noinit" flag. The DT case is harder, and I'm not
sure how we should handle that. We could perhaps add a
"stdout-path-initialized" flag to chosen.

Yes, I skipped UART setup same as msm_serial.c and amba-pl011.c
(although they are standalone serial driver). Just like earlyprintk, I
think earlycon should reuse UART setting in loader. Since some other
platforms with 8250 already depend on this, it's ok to add flags to
distinguish whether 8250 earlycon driver should init hw or not. As you
said, add "noinit" flag is simple, but "stdout-path-initialized" need
more discussion. 

quoted

Signed-off-by: Eddie Huang <eddie.huang@mediatek.com>
---
 drivers/tty/serial/8250/8250_mtk.c | 33 +++++++++++++++++++++++++++++++++
 1 file changed, 33 insertions(+)

diff --git a/drivers/tty/serial/8250/8250_mtk.c b/drivers/tty/serial/8250/8250_mtk.c
index de7aae5..65dd569 100644
--- a/drivers/tty/serial/8250/8250_mtk.c
+++ b/drivers/tty/serial/8250/8250_mtk.c

@@ -23,6 +23,7 @@
 #include <linux/pm_runtime.h>
 #include <linux/serial_8250.h>
 #include <linux/serial_reg.h>
+#include <linux/console.h>

 #include "8250.h"

@@ -289,6 +290,38 @@ static struct platform_driver mtk8250_platform_driver = {
 };
 module_platform_driver(mtk8250_platform_driver);

+#define BOTH_EMPTY (UART_LSR_TEMT | UART_LSR_THRE)
+
+static void __init mtk8250_serial_putc(struct uart_port *port, int c)
+{
+       while ((readl(port->membase + (UART_LSR << 2)) & BOTH_EMPTY) !=
+               BOTH_EMPTY)
+               ;
+       writel(c, port->membase + (UART_TX << 2));
+}
+
+static void __init early_mtk8250_write(struct console *console,
+                                       const char *s, unsigned int count)

Is there a reason early_serial8250_write can't work for you other than
it is currently static?

Rob

The reason is only static. I want to keep 8250_early.c untouchable, all
modifications in 8250_mtk.c

On Sun, Dec 28, 2014 at 10:24 PM, Eddie Huang [off-list ref] wrote:

Hi Rob,

On Fri, 2014-12-26 at 15:24 -0600, Rob Herring wrote:

quoted

On Thu, Dec 18, 2014 at 2:33 AM, Eddie Huang [off-list ref] wrote:

quoted

Mediatek UART has highspeed register, but 8250_early.c doesn't
support this, so add earlycon in 8250_mtk.c

I don't see any highspeed register setup here. More generically,
aren't you just skipping any UART setup? That may be useful on other
platforms with 8250s, too. With the kernel command line version, you
could perhaps add a "noinit" flag. The DT case is harder, and I'm not
sure how we should handle that. We could perhaps add a
"stdout-path-initialized" flag to chosen.

Yes, I skipped UART setup same as msm_serial.c and amba-pl011.c
(although they are standalone serial driver). Just like earlyprintk, I
think earlycon should reuse UART setting in loader. Since some other
platforms with 8250 already depend on this, it's ok to add flags to
distinguish whether 8250 earlycon driver should init hw or not. As you
said, add "noinit" flag is simple, but "stdout-path-initialized" need
more discussion.

quoted

quoted

Signed-off-by: Eddie Huang <eddie.huang@mediatek.com>
---
 drivers/tty/serial/8250/8250_mtk.c | 33 +++++++++++++++++++++++++++++++++
 1 file changed, 33 insertions(+)

diff --git a/drivers/tty/serial/8250/8250_mtk.c b/drivers/tty/serial/8250/8250_mtk.c
index de7aae5..65dd569 100644
--- a/drivers/tty/serial/8250/8250_mtk.c
+++ b/drivers/tty/serial/8250/8250_mtk.c

@@ -23,6 +23,7 @@
 #include <linux/pm_runtime.h>
 #include <linux/serial_8250.h>
 #include <linux/serial_reg.h>
+#include <linux/console.h>

 #include "8250.h"

@@ -289,6 +290,38 @@ static struct platform_driver mtk8250_platform_driver = {
 };
 module_platform_driver(mtk8250_platform_driver);

+#define BOTH_EMPTY (UART_LSR_TEMT | UART_LSR_THRE)
+
+static void __init mtk8250_serial_putc(struct uart_port *port, int c)
+{
+       while ((readl(port->membase + (UART_LSR << 2)) & BOTH_EMPTY) !=
+               BOTH_EMPTY)
+               ;
+       writel(c, port->membase + (UART_TX << 2));
+}
+
+static void __init early_mtk8250_write(struct console *console,
+                                       const char *s, unsigned int count)

Is there a reason early_serial8250_write can't work for you other than
it is currently static?

Rob

The reason is only static. I want to keep 8250_early.c untouchable, all
modifications in 8250_mtk.c

That is not how kernel development works. You should fix 8250_early.c
to do what you need rather than duplicating code.

Rob

On Monday 29 December 2014 12:24:56 Eddie Huang wrote:

Hi Rob,

On Fri, 2014-12-26 at 15:24 -0600, Rob Herring wrote:

quoted

On Thu, Dec 18, 2014 at 2:33 AM, Eddie Huang [off-list ref] wrote:

quoted

Mediatek UART has highspeed register, but 8250_early.c doesn't
support this, so add earlycon in 8250_mtk.c

I don't see any highspeed register setup here. More generically,
aren't you just skipping any UART setup? That may be useful on other
platforms with 8250s, too. With the kernel command line version, you
could perhaps add a "noinit" flag. The DT case is harder, and I'm not
sure how we should handle that. We could perhaps add a
"stdout-path-initialized" flag to chosen.

Yes, I skipped UART setup same as msm_serial.c and amba-pl011.c
(although they are standalone serial driver). Just like earlyprintk, I
think earlycon should reuse UART setting in loader. Since some other
platforms with 8250 already depend on this, it's ok to add flags to
distinguish whether 8250 earlycon driver should init hw or not. As you
said, add "noinit" flag is simple, but "stdout-path-initialized" need
more discussion. 

It would definitely be nice to have consistent behavior, and I also
think it would be good to default to noinit for the console, because that
is the more likely case. Basically anything that can do debug_ll on ARM
already depends on this behavior because the debug_ll infrastructure
does not initialize the console before using it. I believe the same
is true for powerpc udbg.

	Arnd