Skip to contentSkip to conversation

serial: 8250/ingenic: Add support for the JZ4750

PATCH Superseded Siarhei Volkau 2 participants 9 messages Oct 22, 2022 – Oct 24, 2022

Subsystems (4)

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v3 cover serial: 8250/ingenic: Add support for the JZ4750

JZ4750 and JZ4755 have an extra clock divisor in CGU called CPCCR.ECS. It needs to be handled properly in the early console driver.

v3:
- fix build errors
v2:
- serial moved into separate patchset
- code refactored to avoid peek in CGU register
- Krzysztof's ack picked
v1:
- big patchset for the whole JZ4755 support

Siarhei Volkau (2):
dt-bindings: serial: ingenic: Add support for the JZ4750/55 SoCs
serial: 8250/ingenic: Add support for the JZ4750/JZ4755

 .../bindings/serial/ingenic,uart.yaml         |  4 ++
 drivers/tty/serial/8250/8250_ingenic.c        | 48 ++++++++++++++++---
 2 files changed, 46 insertions(+), 6 deletions(-)
Signature · 1 line
2.36.1
#1 lore raw

PATCH v3 1/2 dt-bindings: serial: ingenic: Add support for the JZ4750/55 SoCs

These SoCs UART block are the same as JZ4725b' one, the difference is outside of the block - it is in the clock generation unit (CGU). The difference requires to make a quirk for early console init.

---
 Documentation/devicetree/bindings/serial/ingenic,uart.yaml | 4 ++++
 1 file changed, 4 insertions(+)
Documentation/devicetree/bindings/serial/ingenic,uart.yaml +4 -0
index 9ca7a18ec..315ceb722 100644
@@ -20,6 +20,7 @@ properties:
     oneOf:
       - enum:
           - ingenic,jz4740-uart
+          - ingenic,jz4750-uart
           - ingenic,jz4760-uart
           - ingenic,jz4780-uart
           - ingenic,x1000-uart
@@ -31,6 +32,9 @@ properties:
       - items:
           - const: ingenic,jz4725b-uart
           - const: ingenic,jz4740-uart
+      - items:
+          - const: ingenic,jz4755-uart
+          - const: ingenic,jz4750-uart
 
   reg:
     maxItems: 1
Signature · 1 line
2.36.1
#2 lore raw

PATCH v3 2/2 serial: 8250/ingenic: Add support for the JZ4750/JZ4755

In reply to Siarhei Volkau

JZ4750/55/60 (but not JZ4760b) have an extra divisor in between extclk and peripheral clock, called CPCCR.ECS, the driver can't figure out the real state of the divisor without dirty hack - peek CGU CPCCR register. However, we can rely on a vendor's bootloader (u-boot 1.1.6) behavior:
if (extclk > 16MHz)
the divisor is enabled, so the UART driving clock is extclk/2.

This behavior relies on hardware differences: most boards (if not all) with those SoCs have 12 or 24 MHz oscillators but many peripherals want 12Mhz to operate properly (AIC and USB-PHY at least).

The patch doesn't affect JZ4760's behavior as it is subject for another patchset with re-classification of all supported ingenic UARTs.

---
 drivers/tty/serial/8250/8250_ingenic.c | 48 ++++++++++++++++++++++----
 1 file changed, 42 insertions(+), 6 deletions(-)
drivers/tty/serial/8250/8250_ingenic.c +42 -6
index 2b2f5d8d2..744705467 100644
@@ -87,24 +87,19 @@ static void __init ingenic_early_console_setup_clock(struct earlycon_device *dev
 	dev->port.uartclk = be32_to_cpup(prop);
 }
 
-static int __init ingenic_early_console_setup(struct earlycon_device *dev,
+static int __init ingenic_earlycon_setup_tail(struct earlycon_device *dev,
 					      const char *opt)
 {
 	struct uart_port *port = &dev->port;
 	unsigned int divisor;
 	int baud = 115200;
 
-	if (!dev->port.membase)
-		return -ENODEV;
-
 	if (opt) {
 		unsigned int parity, bits, flow; /* unused for now */
 
 		uart_parse_options(opt, &baud, &parity, &bits, &flow);
 	}
 
-	ingenic_early_console_setup_clock(dev);
-
 	if (dev->baud)
 		baud = dev->baud;
 	divisor = DIV_ROUND_CLOSEST(port->uartclk, 16 * baud);
@@ -129,9 +124,49 @@ static int __init ingenic_early_console_setup(struct earlycon_device *dev,
 	return 0;
 }
 
+static int __init ingenic_early_console_setup(struct earlycon_device *dev,
+					      const char *opt)
+{
+	if (!dev->port.membase)
+		return -ENODEV;
+
+	ingenic_early_console_setup_clock(dev);
+
+	return ingenic_earlycon_setup_tail(dev, opt);
+}
+
+static int __init jz4750_early_console_setup(struct earlycon_device *dev,
+					     const char *opt)
+{
+	if (!dev->port.membase)
+		return -ENODEV;
+
+	/*
+	 * JZ4750/55/60 (not JZ4760b) have an extra divisor
+	 * between extclk and peripheral clock, the
+	 * driver can't figure out the real state of the
+	 * divisor without dirty hacks (peek CGU register).
+	 * However, we can rely on a vendor's behavior:
+	 * if (extclk > 16MHz)
+	 *   the divisor is enabled.
+	 * This behavior relies on hardware differences:
+	 * most boards with those SoCs have 12 or 24 MHz
+	 * oscillators but many peripherals want 12Mhz
+	 * to operate properly (AIC and USB-phy at least).
+	 */
+	ingenic_early_console_setup_clock(dev);
+	if (dev->port.uartclk > 16000000)
+		dev->port.uartclk /= 2;
+
+	return ingenic_earlycon_setup_tail(dev, opt);
+}
+
 OF_EARLYCON_DECLARE(jz4740_uart, "ingenic,jz4740-uart",
 		    ingenic_early_console_setup);
 
+OF_EARLYCON_DECLARE(jz4750_uart, "ingenic,jz4750-uart",
+		    jz4750_early_console_setup);
+
 OF_EARLYCON_DECLARE(jz4770_uart, "ingenic,jz4770-uart",
 		    ingenic_early_console_setup);
 
@@ -328,6 +363,7 @@ static const struct ingenic_uart_config x1000_uart_config = {
 
 static const struct of_device_id of_match[] = {
 	{ .compatible = "ingenic,jz4740-uart", .data = &jz4740_uart_config },
+	{ .compatible = "ingenic,jz4750-uart", .data = &jz4760_uart_config },
 	{ .compatible = "ingenic,jz4760-uart", .data = &jz4760_uart_config },
 	{ .compatible = "ingenic,jz4770-uart", .data = &jz4760_uart_config },
 	{ .compatible = "ingenic,jz4775-uart", .data = &jz4760_uart_config },
Signature · 1 line
2.36.1
maintainer
#3 lore raw

Re: [PATCH v3 2/2] serial: 8250/ingenic: Add support for the JZ4750/JZ4755

Hi Siarhei,

Quoted 84 lines · Siarhei Volkau

JZ4750/55/60 (but not JZ4760b) have an extra divisor in between extclk and peripheral clock, called CPCCR.ECS, the driver can't figure out the real state of the divisor without dirty hack - peek CGU CPCCR register. However, we can rely on a vendor's bootloader (u-boot 1.1.6) behavior:
if (extclk > 16MHz)
the divisor is enabled, so the UART driving clock is extclk/2.

This behavior relies on hardware differences: most boards (if not all) with those SoCs have 12 or 24 MHz oscillators but many peripherals want 12Mhz to operate properly (AIC and USB-PHY at least).

The patch doesn't affect JZ4760's behavior as it is subject for another patchset with re-classification of all supported ingenic UARTs.

---
 drivers/tty/serial/8250/8250_ingenic.c | 48 
++++++++++++++++++++++----
 1 file changed, 42 insertions(+), 6 deletions(-)
drivers/tty/serial/8250/8250_ingenic.c +0 -0
index 2b2f5d8d2..744705467 100644
@@ -87,24 +87,19 @@ static void __init 

ingenic_early_console_setup_clock(struct earlycon_device *dev

drivers/tty/serial/8250/8250_ingenic.c +0 -1
 	dev->port.uartclk = be32_to_cpup(prop);
 }
 
-static int __init ingenic_early_console_setup(struct earlycon_device 

*dev,
+static int __init ingenic_earlycon_setup_tail(struct earlycon_device *dev,

drivers/tty/serial/8250/8250_ingenic.c +23 -5
 					      const char *opt)
 {
 	struct uart_port *port = &dev->port;
 	unsigned int divisor;
 	int baud = 115200;
 
-	if (!dev->port.membase)
-		return -ENODEV;
-
 	if (opt) {
 		unsigned int parity, bits, flow; /* unused for now */
 
 		uart_parse_options(opt, &baud, &parity, &bits, &flow);
 	}
 
-	ingenic_early_console_setup_clock(dev);
-
 	if (dev->baud)
 		baud = dev->baud;
 	divisor = DIV_ROUND_CLOSEST(port->uartclk, 16 * baud);
@@ -129,9 +124,49 @@ static int __init 
 ingenic_early_console_setup(struct earlycon_device *dev,
 	return 0;
 }
 
+static int __init ingenic_early_console_setup(struct earlycon_device 
 *dev,
+					      const char *opt)
+{
+	if (!dev->port.membase)
+		return -ENODEV;
+
+	ingenic_early_console_setup_clock(dev);
+
+	return ingenic_earlycon_setup_tail(dev, opt);
+}
+
+static int __init jz4750_early_console_setup(struct earlycon_device 
 *dev,
+					     const char *opt)
+{
+	if (!dev->port.membase)
+		return -ENODEV;
+
+	/*
+	 * JZ4750/55/60 (not JZ4760b) have an extra divisor
+	 * between extclk and peripheral clock, the
+	 * driver can't figure out the real state of the
+	 * divisor without dirty hacks (peek CGU register).
+	 * However, we can rely on a vendor's behavior:

drivers/tty/serial/8250/8250_ingenic.c static int __init · line 162

Quoted 10 lines
drivers/tty/serial/8250/8250_ingenic.c +10 -0
+	 * if (extclk > 16MHz)
+	 *   the divisor is enabled.
+	 * This behavior relies on hardware differences:
+	 * most boards with those SoCs have 12 or 24 MHz
+	 * oscillators but many peripherals want 12Mhz
+	 * to operate properly (AIC and USB-phy at least).
+	 */
+	ingenic_early_console_setup_clock(dev);
+	if (dev->port.uartclk > 16000000)
+		dev->port.uartclk /= 2;

I don't understand, didn't we came up to the conclusion in your V1 that it was better to force-enable the EXT/2 divider in the ingenic init code?

-Paul

Quoted 30 lines
drivers/tty/serial/8250/8250_ingenic.c +7 -0
+
+	return ingenic_earlycon_setup_tail(dev, opt);
+}
+
 OF_EARLYCON_DECLARE(jz4740_uart, "ingenic,jz4740-uart",
 		    ingenic_early_console_setup);
 
+OF_EARLYCON_DECLARE(jz4750_uart, "ingenic,jz4750-uart",
+		    jz4750_early_console_setup);
+

OF_EARLYCON_DECLARE(jz4770_uart, "ingenic,jz4770-uart",

drivers/tty/serial/8250/8250_ingenic.c +1 -0
 		    ingenic_early_console_setup);
@@ -328,6 +363,7 @@ static const struct ingenic_uart_config 
 x1000_uart_config = {
 
 static const struct of_device_id of_match[] = {
 	{ .compatible = "ingenic,jz4740-uart", .data = &jz4740_uart_config 
 },
+	{ .compatible = "ingenic,jz4750-uart", .data = &jz4760_uart_config 
 },

{ .compatible = "ingenic,jz4760-uart", .data = &jz4760_uart_config
},
{ .compatible = "ingenic,jz4770-uart", .data = &jz4760_uart_config
},
{ .compatible = "ingenic,jz4775-uart", .data = &jz4760_uart_config
},
--
2.36.1

#4 lore raw

Re: [PATCH v3 2/2] serial: 8250/ingenic: Add support for the JZ4750/JZ4755

сб, 22 окт. 2022 г. в 23:07, Paul Cercueil <[email protected]>:

Quoted 110 lines

Hi Siarhei,

Quoted 90 lines · Siarhei Volkau

JZ4750/55/60 (but not JZ4760b) have an extra divisor in between extclk and peripheral clock, called CPCCR.ECS, the driver can't figure out the
real state of the divisor without dirty hack - peek CGU CPCCR register.
However, we can rely on a vendor's bootloader (u-boot 1.1.6) behavior:
if (extclk > 16MHz)
the divisor is enabled, so the UART driving clock is extclk/2.

This behavior relies on hardware differences: most boards (if not all) with those SoCs have 12 or 24 MHz oscillators but many peripherals want
12Mhz to operate properly (AIC and USB-PHY at least).

The patch doesn't affect JZ4760's behavior as it is subject for another
patchset with re-classification of all supported ingenic UARTs.

Link:
https://github.com/carlos-wong/uboot_jz4755/blob/master/cpu/mips/jz_serial.c#L158

---
 drivers/tty/serial/8250/8250_ingenic.c | 48
++++++++++++++++++++++----
 1 file changed, 42 insertions(+), 6 deletions(-)
drivers/tty/serial/8250/8250_ingenic.c +0 -0
index 2b2f5d8d2..744705467 100644
@@ -87,24 +87,19 @@ static void __init

ingenic_early_console_setup_clock(struct earlycon_device *dev

      dev->port.uartclk = be32_to_cpup(prop);

}

-static int __init ingenic_early_console_setup(struct earlycon_device *dev,
+static int __init ingenic_earlycon_setup_tail(struct earlycon_device *dev,
const char *opt)

 {
      struct uart_port *port = &dev->port;
      unsigned int divisor;
      int baud = 115200;
drivers/tty/serial/8250/8250_ingenic.c +23 -5
-     if (!dev->port.membase)
-             return -ENODEV;
-
      if (opt) {
              unsigned int parity, bits, flow; /* unused for now */
 
              uart_parse_options(opt, &baud, &parity, &bits, &flow);
      }
 
-     ingenic_early_console_setup_clock(dev);
-
      if (dev->baud)
              baud = dev->baud;
      divisor = DIV_ROUND_CLOSEST(port->uartclk, 16 * baud);
@@ -129,9 +124,49 @@ static int __init
 ingenic_early_console_setup(struct earlycon_device *dev,
      return 0;
 }
 
+static int __init ingenic_early_console_setup(struct earlycon_device
 *dev,
+                                           const char *opt)
+{
+     if (!dev->port.membase)
+             return -ENODEV;
+
+     ingenic_early_console_setup_clock(dev);
+
+     return ingenic_earlycon_setup_tail(dev, opt);
+}
+
+static int __init jz4750_early_console_setup(struct earlycon_device
 *dev,
+                                          const char *opt)
+{
+     if (!dev->port.membase)
+             return -ENODEV;
+
+     /*
+      * JZ4750/55/60 (not JZ4760b) have an extra divisor
+      * between extclk and peripheral clock, the
+      * driver can't figure out the real state of the
+      * divisor without dirty hacks (peek CGU register).
+      * However, we can rely on a vendor's behavior:

drivers/tty/serial/8250/8250_ingenic.c static int __init · line 162

Quoted 10 lines
drivers/tty/serial/8250/8250_ingenic.c +10 -0
+      * if (extclk > 16MHz)
+      *   the divisor is enabled.
+      * This behavior relies on hardware differences:
+      * most boards with those SoCs have 12 or 24 MHz
+      * oscillators but many peripherals want 12Mhz
+      * to operate properly (AIC and USB-phy at least).
+      */
+     ingenic_early_console_setup_clock(dev);
+     if (dev->port.uartclk > 16000000)
+             dev->port.uartclk /= 2;

I don't understand, didn't we came up to the conclusion in your V1 that it was better to force-enable the EXT/2 divider in the ingenic init code?

Quoted 2 lines

-Paul

That was better at that moment. I dived deeper in the situation and found a more simple and universal solution, as I think.
Your proposal doesn't cover following situations:
- JZ475x with 12MHz crystal
- JZ4760 with 24MHz crystal
which are legal and might appear in some hardware.

Quoted 32 lines
Quoted 30 lines
drivers/tty/serial/8250/8250_ingenic.c +7 -0
+
+     return ingenic_earlycon_setup_tail(dev, opt);
+}
+
 OF_EARLYCON_DECLARE(jz4740_uart, "ingenic,jz4740-uart",
                  ingenic_early_console_setup);
 
+OF_EARLYCON_DECLARE(jz4750_uart, "ingenic,jz4750-uart",
+                 jz4750_early_console_setup);
+

OF_EARLYCON_DECLARE(jz4770_uart, "ingenic,jz4770-uart",

                  ingenic_early_console_setup);
drivers/tty/serial/8250/8250_ingenic.c +1 -0
@@ -328,6 +363,7 @@ static const struct ingenic_uart_config
 x1000_uart_config = {
 
 static const struct of_device_id of_match[] = {
      { .compatible = "ingenic,jz4740-uart", .data = &jz4740_uart_config
 },
+     { .compatible = "ingenic,jz4750-uart", .data = &jz4760_uart_config
 },
      { .compatible = "ingenic,jz4760-uart", .data = &jz4760_uart_config
},
      { .compatible = "ingenic,jz4770-uart", .data = &jz4760_uart_config
},
      { .compatible = "ingenic,jz4775-uart", .data = &jz4760_uart_config
},

--
2.36.1

maintainer
#5 lore raw

Re: [PATCH v3 2/2] serial: 8250/ingenic: Add support for the JZ4750/JZ4755

Hi,

Quoted 126 lines · Siarhei Volkau

сб, 22 окт. 2022 г. в 23:07, Paul Cercueil <[email protected]>:

Quoted 122 lines

Hi Siarhei,

Quoted 1 line · Siarhei Volkau

JZ4750/55/60 (but not JZ4760b) have an extra divisor in between

extclk

Quoted 1 line

and peripheral clock, called CPCCR.ECS, the driver can't figure

out

Quoted 4 lines

the
real state of the divisor without dirty hack - peek CGU CPCCR register.
However, we can rely on a vendor's bootloader (u-boot 1.1.6)

behavior:

Quoted 4 lines

if (extclk > 16MHz)
the divisor is enabled, so the UART driving clock is extclk/2.

This behavior relies on hardware differences: most boards (if not

all)

Quoted 10 lines

with those SoCs have 12 or 24 MHz oscillators but many peripherals want
12Mhz to operate properly (AIC and USB-PHY at least).

The patch doesn't affect JZ4760's behavior as it is subject for another
patchset with re-classification of all supported ingenic UARTs.

Link:

https://github.com/carlos-wong/uboot_jz4755/blob/master/cpu/mips/jz_serial.c#L158

Quoted 13 lines
---
 drivers/tty/serial/8250/8250_ingenic.c | 48
++++++++++++++++++++++----
 1 file changed, 42 insertions(+), 6 deletions(-)
drivers/tty/serial/8250/8250_ingenic.c +0 -0
index 2b2f5d8d2..744705467 100644
@@ -87,24 +87,19 @@ static void __init

ingenic_early_console_setup_clock(struct earlycon_device *dev

Quoted 4 lines
      dev->port.uartclk = be32_to_cpup(prop);

}

-static int __init ingenic_early_console_setup(struct

earlycon_device

Quoted 2 lines

*dev,
+static int __init ingenic_earlycon_setup_tail(struct

earlycon_device

drivers/tty/serial/8250/8250_ingenic.c static void __init ingenic_early_console_setup_clock(struct earlycon_device *dev

Quoted 12 lines

*dev,
const char *opt)

 {
      struct uart_port *port = &dev->port;
      unsigned int divisor;
      int baud = 115200;
drivers/tty/serial/8250/8250_ingenic.c +0 -3
-     if (!dev->port.membase)
-             return -ENODEV;
-
      if (opt) {
              unsigned int parity, bits, flow; /* unused for now 

*/

Quoted 2 lines

uart_parse_options(opt, &baud, &parity, &bits,

&flow);

drivers/tty/serial/8250/8250_ingenic.c static int __init · line 128

Quoted 13 lines
      }
drivers/tty/serial/8250/8250_ingenic.c +1 -2
-     ingenic_early_console_setup_clock(dev);
-
      if (dev->baud)
              baud = dev->baud;
      divisor = DIV_ROUND_CLOSEST(port->uartclk, 16 * baud);
@@ -129,9 +124,49 @@ static int __init
 ingenic_early_console_setup(struct earlycon_device *dev,
      return 0;
 }
 
+static int __init ingenic_early_console_setup(struct 

earlycon_device

drivers/tty/serial/8250/8250_ingenic.c static int __init ingenic_early_console_setup(struct earlycon_device *dev,

Quoted 12 lines
drivers/tty/serial/8250/8250_ingenic.c +11 -0
 *dev,
+                                           const char *opt)
+{
+     if (!dev->port.membase)
+             return -ENODEV;
+
+     ingenic_early_console_setup_clock(dev);
+
+     return ingenic_earlycon_setup_tail(dev, opt);
+}
+
+static int __init jz4750_early_console_setup(struct 

earlycon_device

Quoted 12 lines
drivers/tty/serial/8250/8250_ingenic.c +11 -0
 *dev,
+                                          const char *opt)
+{
+     if (!dev->port.membase)
+             return -ENODEV;
+
+     /*
+      * JZ4750/55/60 (not JZ4760b) have an extra divisor
+      * between extclk and peripheral clock, the
+      * driver can't figure out the real state of the
+      * divisor without dirty hacks (peek CGU register).
+      * However, we can rely on a vendor's behavior:

drivers/tty/serial/8250/8250_ingenic.c static int __init ingenic_early_console_setup(struct earlycon_device *dev, · line 159

Quoted 10 lines
drivers/tty/serial/8250/8250_ingenic.c +10 -0
+      * if (extclk > 16MHz)
+      *   the divisor is enabled.
+      * This behavior relies on hardware differences:
+      * most boards with those SoCs have 12 or 24 MHz
+      * oscillators but many peripherals want 12Mhz
+      * to operate properly (AIC and USB-phy at least).
+      */
+     ingenic_early_console_setup_clock(dev);
+     if (dev->port.uartclk > 16000000)
+             dev->port.uartclk /= 2;

I don't understand, didn't we came up to the conclusion in your V1
that
it was better to force-enable the EXT/2 divider in the ingenic init code?

Quoted 2 lines

-Paul

Quoted 6 lines

That was better at that moment. I dived deeper in the situation and found a more simple and universal solution, as I think.
Your proposal doesn't cover following situations:
- JZ475x with 12MHz crystal
- JZ4760 with 24MHz crystal
which are legal and might appear in some hardware.

Do you have such hardware?

Don't add support for cases you can't test.

For what we know - all JZ475x use a 24 MHz crystal and all JZ4760(B) use a 12 MHz crystal, until proven otherwise.

-Paul

Quoted 37 lines
Quoted 37 lines
Quoted 14 lines
drivers/tty/serial/8250/8250_ingenic.c +7 -0
+
+     return ingenic_earlycon_setup_tail(dev, opt);
+}
+
 OF_EARLYCON_DECLARE(jz4740_uart, "ingenic,jz4740-uart",
                  ingenic_early_console_setup);
 
+OF_EARLYCON_DECLARE(jz4750_uart, "ingenic,jz4750-uart",
+                 jz4750_early_console_setup);
+

OF_EARLYCON_DECLARE(jz4770_uart, "ingenic,jz4770-uart",

                  ingenic_early_console_setup);

drivers/tty/serial/8250/8250_ingenic.c static const struct ingenic_uart_config · line 366

Quoted 4 lines
drivers/tty/serial/8250/8250_ingenic.c +0 -0
@@ -328,6 +363,7 @@ static const struct ingenic_uart_config
 x1000_uart_config = {
 
 static const struct of_device_id of_match[] = {
      { .compatible = "ingenic,jz4740-uart", .data = 

&jz4740_uart_config

drivers/tty/serial/8250/8250_ingenic.c static const struct ingenic_uart_config · line 368

Quoted 2 lines
drivers/tty/serial/8250/8250_ingenic.c +1 -0
 },
+     { .compatible = "ingenic,jz4750-uart", .data = 

&jz4760_uart_config

Quoted 2 lines
drivers/tty/serial/8250/8250_ingenic.c +0 -0
 },
      { .compatible = "ingenic,jz4760-uart", .data = 

&jz4760_uart_config

Quoted 2 lines

},

      { .compatible = "ingenic,jz4770-uart", .data = 

&jz4760_uart_config

Quoted 2 lines

},

      { .compatible = "ingenic,jz4775-uart", .data = 

&jz4760_uart_config

Quoted 4 lines

},
--
2.36.1

#6 lore raw

Re: [PATCH v3 2/2] serial: 8250/ingenic: Add support for the JZ4750/JZ4755

вс, 23 окт. 2022 г. в 12:16, Paul Cercueil <[email protected]>:

Quoted 1 line

Do you have such hardware?

No

Quoted 1 line

Don't add support for cases you can't test.

It's just a side effect of that approach.

Quoted 2 lines

For what we know - all JZ475x use a 24 MHz crystal and all JZ4760(B) use a 12 MHz crystal, until proven otherwise.

Ouf course it just confirms the rule but I found one exception: JZ4750 & 12MHz

Regarding your proposal:
In my opinion enabling the divisor unconditionally is a bad practice, as it's already enabled (or not) by the bootloader, with respect to the hardware capabilities.I think it's better to keep the driver as it is than adding such things.

BR,
Siarhei

maintainer
#7 lore raw

Re: [PATCH v3 2/2] serial: 8250/ingenic: Add support for the JZ4750/JZ4755

Quoted 7 lines · Siarhei Volkau

вс, 23 окт. 2022 г. в 12:16, Paul Cercueil <[email protected]>:

Quoted 1 line

Do you have such hardware?

No

Quoted 1 line

Don't add support for cases you can't test.

Quoted 7 lines

It's just a side effect of that approach.

Quoted 2 lines

For what we know - all JZ475x use a 24 MHz crystal and all JZ4760(B) use a 12 MHz crystal, until proven otherwise.

Ouf course it just confirms the rule but I found one exception: JZ4750 & 12MHz

Then when this board is upstreamed it will declare a 12 MHz oscillator in its DT, and the ingenic init code won't have to enable the /2 divider for that particular board.

Quoted 5 lines

Regarding your proposal:
In my opinion enabling the divisor unconditionally is a bad practice, as it's already enabled (or not) by the bootloader, with respect to the hardware capabilities.I think it's better to keep the driver as it is than adding such things.

Well, I disagree. Linux should not depend on whatever the bootloader configures.

-Paul

maintainer
#8 lore raw

Re: [PATCH v3 2/2] serial: 8250/ingenic: Add support for the JZ4750/JZ4755

In reply to Siarhei Volkau

Hi Siarhei,

Quoted 34 lines · Siarhei Volkau

JZ4750/55/60 (but not JZ4760b) have an extra divisor in between extclk and peripheral clock, called CPCCR.ECS, the driver can't figure out the real state of the divisor without dirty hack - peek CGU CPCCR register. However, we can rely on a vendor's bootloader (u-boot 1.1.6) behavior:
if (extclk > 16MHz)
the divisor is enabled, so the UART driving clock is extclk/2.

This behavior relies on hardware differences: most boards (if not all) with those SoCs have 12 or 24 MHz oscillators but many peripherals want 12Mhz to operate properly (AIC and USB-PHY at least).

The patch doesn't affect JZ4760's behavior as it is subject for another patchset with re-classification of all supported ingenic UARTs.

---
 drivers/tty/serial/8250/8250_ingenic.c | 48 
++++++++++++++++++++++----
 1 file changed, 42 insertions(+), 6 deletions(-)
drivers/tty/serial/8250/8250_ingenic.c +0 -0
index 2b2f5d8d2..744705467 100644
@@ -87,24 +87,19 @@ static void __init 

ingenic_early_console_setup_clock(struct earlycon_device *dev

drivers/tty/serial/8250/8250_ingenic.c +0 -1
 	dev->port.uartclk = be32_to_cpup(prop);
 }
 
-static int __init ingenic_early_console_setup(struct earlycon_device 

*dev,
+static int __init ingenic_earlycon_setup_tail(struct earlycon_device *dev,

drivers/tty/serial/8250/8250_ingenic.c static void __init ingenic_early_console_setup_clock(struct earlycon_device *dev · line 98

Quoted 8 lines
drivers/tty/serial/8250/8250_ingenic.c +0 -2
 					      const char *opt)
 {
 	struct uart_port *port = &dev->port;
 	unsigned int divisor;
 	int baud = 115200;
 
-	if (!dev->port.membase)
-		return -ENODEV;

Again, as I said on your v2, you can keep this here. Then you won't have to duplicate code.

Quoted 42 lines
drivers/tty/serial/8250/8250_ingenic.c +23 -3
-
 	if (opt) {
 		unsigned int parity, bits, flow; /* unused for now */
 
 		uart_parse_options(opt, &baud, &parity, &bits, &flow);
 	}
 
-	ingenic_early_console_setup_clock(dev);
-
 	if (dev->baud)
 		baud = dev->baud;
 	divisor = DIV_ROUND_CLOSEST(port->uartclk, 16 * baud);
@@ -129,9 +124,49 @@ static int __init 
 ingenic_early_console_setup(struct earlycon_device *dev,
 	return 0;
 }
 
+static int __init ingenic_early_console_setup(struct earlycon_device 
 *dev,
+					      const char *opt)
+{
+	if (!dev->port.membase)
+		return -ENODEV;
+
+	ingenic_early_console_setup_clock(dev);
+
+	return ingenic_earlycon_setup_tail(dev, opt);
+}
+
+static int __init jz4750_early_console_setup(struct earlycon_device 
 *dev,
+					     const char *opt)
+{
+	if (!dev->port.membase)
+		return -ENODEV;
+
+	/*
+	 * JZ4750/55/60 (not JZ4760b) have an extra divisor
+	 * between extclk and peripheral clock, the
+	 * driver can't figure out the real state of the
+	 * divisor without dirty hacks (peek CGU register).
+	 * However, we can rely on a vendor's behavior:

drivers/tty/serial/8250/8250_ingenic.c static int __init · line 162

Quoted 10 lines
drivers/tty/serial/8250/8250_ingenic.c +10 -0
+	 * if (extclk > 16MHz)
+	 *   the divisor is enabled.
+	 * This behavior relies on hardware differences:
+	 * most boards with those SoCs have 12 or 24 MHz
+	 * oscillators but many peripherals want 12Mhz
+	 * to operate properly (AIC and USB-phy at least).
+	 */
+	ingenic_early_console_setup_clock(dev);
+	if (dev->port.uartclk > 16000000)
+		dev->port.uartclk /= 2;

I'm OK with this code, but the comment is not very clear.

What about:

"JZ4750/55/60 have an optional /2 divider between the EXT oscillator and some peripherals including UART, which will be enabled if using a 24 MHz oscillator, and disabled when using a 12 MHz oscillator."

Cheers,
-Paul

Quoted 30 lines
drivers/tty/serial/8250/8250_ingenic.c +7 -0
+
+	return ingenic_earlycon_setup_tail(dev, opt);
+}
+
 OF_EARLYCON_DECLARE(jz4740_uart, "ingenic,jz4740-uart",
 		    ingenic_early_console_setup);
 
+OF_EARLYCON_DECLARE(jz4750_uart, "ingenic,jz4750-uart",
+		    jz4750_early_console_setup);
+

OF_EARLYCON_DECLARE(jz4770_uart, "ingenic,jz4770-uart",

drivers/tty/serial/8250/8250_ingenic.c +1 -0
 		    ingenic_early_console_setup);
@@ -328,6 +363,7 @@ static const struct ingenic_uart_config 
 x1000_uart_config = {
 
 static const struct of_device_id of_match[] = {
 	{ .compatible = "ingenic,jz4740-uart", .data = &jz4740_uart_config 
 },
+	{ .compatible = "ingenic,jz4750-uart", .data = &jz4760_uart_config 
 },

{ .compatible = "ingenic,jz4760-uart", .data = &jz4760_uart_config
},
{ .compatible = "ingenic,jz4770-uart", .data = &jz4760_uart_config
},
{ .compatible = "ingenic,jz4775-uart", .data = &jz4760_uart_config
},
--
2.36.1