[PATCH] ARM64: tegra: Add support for Google Pixel C

On 07/01/16 14:11, Arnd Bergmann wrote:

On Thursday 07 January 2016 13:19:44 Jon Hunter wrote:

quoted

Adding Arnd.

Hmmm ... well apparently stdout-path does not work for tegra and in
order to make this work I had to do the following ...


[PATCH] serial: 8250: of: Add earlycon support for Tegra

Currently, early console support only works for Tegra when the serial
port information is passed via the earlycon boot parameter. If the
serial port information is specified via device-tree using the
"stdout-path" then the early console does not work because:

1. The tegra serial ports compatibility parameter does not match any
   of the supported serial drivers for early console.
2. The of_setup_earlycon() function assumes that serial port registers
   are byte aligned and for tegra they are 32-bit aligned.

Add an early console setup function for tegra so that the early console
can be specified via the device-tree "stdout-path" variable.

Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
---
 drivers/tty/serial/8250/8250_of.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/drivers/tty/serial/8250/8250_of.c b/drivers/tty/serial/8250/8250_of.c
index 33021c1f7d55..98adf83e83c2 100644
--- a/drivers/tty/serial/8250/8250_of.c
+++ b/drivers/tty/serial/8250/8250_of.c

@@ -44,6 +44,16 @@ void tegra_serial_handle_break(struct uart_port *p)
                udelay(1);
        } while (1);
 }
+
+int __init tegra_earlycon_setup(struct earlycon_device *device,
+                               const char *options)
+{
+       device->port.iotype = UPIO_MEM32;
+       device->port.regshift = 2;
+
+       return early_serial8250_setup(device, options);
+}
+OF_EARLYCON_DECLARE(tegra20_uart, "nvidia,tegra20-uart", tegra_earlycon_setup);
 #else
 static inline void tegra_serial_handle_break(struct uart_port *port)
 {

Arnd, does the above look ok, or should there be a generic
early_serial8250x32_setup() somewhere?

I think it would be better to put it into 8250_early.c rather than 
8250_of.c, as there are already some other definitions in there,
and the #ifdef CONFIG_TEGRA in 8250_of.c is for some other workaround.

Ok so something like this ...

diff --git a/drivers/tty/serial/8250/8250_early.c

b/drivers/tty/serial/8250/8250_early.c
index af62131af21e..4ce6e2b57534 100644

--- a/drivers/tty/serial/8250/8250_early.c
+++ b/drivers/tty/serial/8250/8250_early.c

@@ -141,7 +141,18 @@ int __init early_serial8250_setup(struct

earlycon_device *device,
        device->con->write = early_serial8250_write;
        return 0;
 }
+
+int __init early_serial8250_mem32_setup(struct earlycon_device *device,
+                                       const char *options)
+{
+       device->port.iotype = UPIO_MEM32;
+       device->port.regshift = 2;
+
+       return early_serial8250_setup(device, options);
+}
 EARLYCON_DECLARE(uart8250, early_serial8250_setup);
 EARLYCON_DECLARE(uart, early_serial8250_setup);
 OF_EARLYCON_DECLARE(ns16550, "ns16550", early_serial8250_setup);
 OF_EARLYCON_DECLARE(ns16550a, "ns16550a", early_serial8250_setup);
+OF_EARLYCON_DECLARE(tegra20_uart, "nvidia,tegra20-uart",
+                   early_serial8250_mem32_setup);

Would it be possible to handle the "reg-io-width" parsing in
early_serial8250_setup instead of keying it off the string?

Good point. I spent a bit of time looking at this, but I cannot see an
easy way to do this unless we check for "reg-io-width" and "reg-shift"
in early_init_dt_scan_chosen_serial() and pass to of_setup_earlycon(),
however, I am guessing that this would not be ideal as this would happen
for non-8250 devices. May be that would be ok, but I am not sure.

Cheers
Jon