Skip to contentSkip to conversation

arm64: allwinner: a64: fix video output on Pinebook

PATCH Applied Vasily Khoruzhick 5 participants 15 messages Jan 4, 2025 – Jan 6, 2025

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

v2 cover arm64: allwinner: a64: fix video output on Pinebook

Since commit ca1170b69968 ("clk: sunxi-ng: a64: force select PLL_MIPI in TCON0 mux"), TCON0 clock parent is always set to PLL_MIPI, but unfortunately it breaks video output on Pinebook.

I did an experiment: I manually configured PLL_MIPI and PLL_VIDEO0_2X to the same clock rate and flipped the switch with devmem. Experiment clearly showed that whenever PLL_MIPI is selected as TCON0 clock parent, the video output stops working.

To fix the issue, I partially reverted mentioned commit and added explicit TCON0 clock parent assignment to device tree. By default, it will be PLL_MIPI, and the only users with RGB output - Pinebook and Teres-I will override it in their dts.

v2:
- split device tree binding change into its own patch
- add rationale into commit message and into comment in the code
- gather the tags

Vasily Khoruzhick (4):
dt-bindings: clock: sunxi: Export PLL_VIDEO_2X and PLL_MIPI
clk: sunxi-ng: a64: drop redundant CLK_PLL_VIDEO0_2X and CLK_PLL_MIPI
arm64: dts: allwinner: a64: explicitly assign clock parent for TCON0
clk: sunxi-ng: a64: stop force-selecting PLL-MIPI as TCON0 parent

 .../boot/dts/allwinner/sun50i-a64-pinebook.dts      |  2 ++
 .../arm64/boot/dts/allwinner/sun50i-a64-teres-i.dts |  2 ++
 arch/arm64/boot/dts/allwinner/sun50i-a64.dtsi       |  2 ++
 drivers/clk/sunxi-ng/ccu-sun50i-a64.c               | 13 ++++---------
 drivers/clk/sunxi-ng/ccu-sun50i-a64.h               |  2 --
 include/dt-bindings/clock/sun50i-a64-ccu.h          |  2 ++
 6 files changed, 12 insertions(+), 11 deletions(-)
Signature · 1 line
2.47.1
#1 lore raw

PATCH v2 1/4 dt-bindings: clock: sunxi: Export PLL_VIDEO_2X and PLL_MIPI

Export PLL_VIDEO_2X and PLL_MIPI, these will be used to explicitly select TCON0 clock parent in dts

---
 include/dt-bindings/clock/sun50i-a64-ccu.h | 2 ++
 1 file changed, 2 insertions(+)
include/dt-bindings/clock/sun50i-a64-ccu.h +2 -0
index 175892189e9d..4f220ea7a23c 100644
@@ -44,7 +44,9 @@
 #define _DT_BINDINGS_CLK_SUN50I_A64_H_
 
 #define CLK_PLL_VIDEO0		7
+#define CLK_PLL_VIDEO0_2X	8
 #define CLK_PLL_PERIPH0		11
+#define CLK_PLL_MIPI		17
 
 #define CLK_CPUX		21
 #define CLK_BUS_MIPI_DSI	28
Signature · 1 line
2.47.1
#2 lore raw

Re: [PATCH v2 1/4] dt-bindings: clock: sunxi: Export PLL_VIDEO_2X and PLL_MIPI

Quoted 6 lines · Vasily Khoruzhick

Export PLL_VIDEO_2X and PLL_MIPI, these will be used to explicitly select TCON0 clock parent in dts

Fixes ca1170b69968 ("clk: sunxi-ng: a64: force select PLL_MIPI in TCON0 mux")

Where did this happen?

Quoted 2 lines

And these? I cannot find traces on the list.

Best regards,
Krzysztof

#3 lore raw

Re: [PATCH v2 1/4] dt-bindings: clock: sunxi: Export PLL_VIDEO_2X and PLL_MIPI

Quoted 8 lines · Krzysztof Kozlowski
Quoted 6 lines · Vasily Khoruzhick

Export PLL_VIDEO_2X and PLL_MIPI, these will be used to explicitly select TCON0 clock parent in dts

Fixes ca1170b69968 ("clk: sunxi-ng: a64: force select PLL_MIPI in TCON0 mux")
Quoted 6 lines

Where did this happen?

Quoted 2 lines

And these? I cannot find traces on the list.

I found them, cover letter.

Best regards,
Krzysztof

#4 lore raw

Re: [PATCH v2 1/4] dt-bindings: clock: sunxi: Export PLL_VIDEO_2X and PLL_MIPI

Quoted 10 lines · Krzysztof Kozlowski
Quoted 8 lines · Krzysztof Kozlowski
Quoted 6 lines · Vasily Khoruzhick

Export PLL_VIDEO_2X and PLL_MIPI, these will be used to explicitly select TCON0 clock parent in dts

Fixes ca1170b69968 ("clk: sunxi-ng: a64: force select PLL_MIPI in TCON0 mux")
Quoted 8 lines
Quoted 6 lines

Where did this happen?

Quoted 2 lines

And these? I cannot find traces on the list.

I found them, cover letter.

Yeah, most of the tags are replies to v1 cover letter. Stuart's Tested-by comes from IRC. However all the mentioned people are on CC list, it's not like I'm trying to sneak in counterfeit tags :)

Quoted 4 lines

Best regards,
Krzysztof

#5 lore raw

Re: [PATCH v2 1/4] dt-bindings: clock: sunxi: Export PLL_VIDEO_2X and PLL_MIPI

In reply to Vasily Khoruzhick
Quoted 19 lines · Vasily Khoruzhick

Export PLL_VIDEO_2X and PLL_MIPI, these will be used to explicitly select TCON0 clock parent in dts

---
 include/dt-bindings/clock/sun50i-a64-ccu.h | 2 ++
 1 file changed, 2 insertions(+)
include/dt-bindings/clock/sun50i-a64-ccu.h +0 -0
index 175892189e9d..4f220ea7a23c 100644

include/dt-bindings/clock/sun50i-a64-ccu.h line 49

Quoted 6 lines
include/dt-bindings/clock/sun50i-a64-ccu.h +2 -0
@@ -44,7 +44,9 @@
 #define _DT_BINDINGS_CLK_SUN50I_A64_H_
 
 #define CLK_PLL_VIDEO0         7
+#define CLK_PLL_VIDEO0_2X      8
 #define CLK_PLL_PERIPH0                11
+#define CLK_PLL_MIPI           17

You can't really split code movement into two patches.

With this patch applied the clk driver will fail to build because the macros are now redefined in both header files.

Barring recombining the patches, please add a patch before this adding #ifndef's around the two macros that are moved.

ChenYu

Quoted 5 lines
include/dt-bindings/clock/sun50i-a64-ccu.h +0 -1
 #define CLK_CPUX               21
 #define CLK_BUS_MIPI_DSI       28
--

2.47.1

#6 lore raw

Re: [PATCH v2 1/4] dt-bindings: clock: sunxi: Export PLL_VIDEO_2X and PLL_MIPI

Quoted 29 lines · Chen-Yu Tsai
Quoted 19 lines · Vasily Khoruzhick

Export PLL_VIDEO_2X and PLL_MIPI, these will be used to explicitly select TCON0 clock parent in dts

---
 include/dt-bindings/clock/sun50i-a64-ccu.h | 2 ++
 1 file changed, 2 insertions(+)
include/dt-bindings/clock/sun50i-a64-ccu.h +0 -0
index 175892189e9d..4f220ea7a23c 100644

include/dt-bindings/clock/sun50i-a64-ccu.h line 49

Quoted 6 lines
include/dt-bindings/clock/sun50i-a64-ccu.h +2 -0
@@ -44,7 +44,9 @@
 #define _DT_BINDINGS_CLK_SUN50I_A64_H_
 
 #define CLK_PLL_VIDEO0         7
+#define CLK_PLL_VIDEO0_2X      8
 #define CLK_PLL_PERIPH0                11
+#define CLK_PLL_MIPI           17

You can't really split code movement into two patches.

Quoted 2 lines

With this patch applied the clk driver will fail to build because the macros are now redefined in both header files.

Are you sure? The values seem the same to me... I don't see how this could fail.

Quoted 4 lines

Barring recombining the patches, please add a patch before this adding #ifndef's around the two macros that are moved.

No, not necessary, just churn,

Best regards,
Krzysztof

#7 lore raw

Re: [PATCH v2 1/4] dt-bindings: clock: sunxi: Export PLL_VIDEO_2X and PLL_MIPI

Quoted 30 lines · On Sat, 4 Jan 2025 11:33:23 +0100 Krzysztof Kozlowski <[email protected]> wrote:
Quoted 29 lines · Chen-Yu Tsai
Quoted 19 lines · Vasily Khoruzhick

Export PLL_VIDEO_2X and PLL_MIPI, these will be used to explicitly select TCON0 clock parent in dts

---
 include/dt-bindings/clock/sun50i-a64-ccu.h | 2 ++
 1 file changed, 2 insertions(+)
include/dt-bindings/clock/sun50i-a64-ccu.h +0 -0
index 175892189e9d..4f220ea7a23c 100644

include/dt-bindings/clock/sun50i-a64-ccu.h line 49

Quoted 6 lines
include/dt-bindings/clock/sun50i-a64-ccu.h +2 -0
@@ -44,7 +44,9 @@
 #define _DT_BINDINGS_CLK_SUN50I_A64_H_
 
 #define CLK_PLL_VIDEO0         7
+#define CLK_PLL_VIDEO0_2X      8
 #define CLK_PLL_PERIPH0                11
+#define CLK_PLL_MIPI           17  

You can't really split code movement into two patches.

Quoted 5 lines
Quoted 2 lines

With this patch applied the clk driver will fail to build because the macros are now redefined in both header files.

Are you sure? The values seem the same to me... I don't see how this could fail.

Yes, IIRC the C standard allows repeated definitions with the same value. And I definitely tested this before (and hence recommended this approach to Vasily) and it compiled without any warnings here.

Cheers,
Andre

Quoted 12 lines
Quoted 5 lines

Barring recombining the patches, please add a patch before this adding #ifndef's around the two macros that are moved.

No, not necessary, just churn,

Best regards,
Krzysztof

#8 lore raw

Re: [PATCH v2 1/4] dt-bindings: clock: sunxi: Export PLL_VIDEO_2X and PLL_MIPI

Quoted 37 lines · Andre Przywara
Quoted 33 lines · On Sat, 4 Jan 2025 11:33:23 +0100 Krzysztof Kozlowski <[email protected]> wrote:
Quoted 31 lines · Chen-Yu Tsai
Quoted 19 lines · Vasily Khoruzhick

Export PLL_VIDEO_2X and PLL_MIPI, these will be used to explicitly select TCON0 clock parent in dts

---
 include/dt-bindings/clock/sun50i-a64-ccu.h | 2 ++
 1 file changed, 2 insertions(+)
include/dt-bindings/clock/sun50i-a64-ccu.h +0 -0
index 175892189e9d..4f220ea7a23c 100644

include/dt-bindings/clock/sun50i-a64-ccu.h line 49

Quoted 6 lines
include/dt-bindings/clock/sun50i-a64-ccu.h +2 -0
@@ -44,7 +44,9 @@
 #define _DT_BINDINGS_CLK_SUN50I_A64_H_
 
 #define CLK_PLL_VIDEO0         7
+#define CLK_PLL_VIDEO0_2X      8
 #define CLK_PLL_PERIPH0                11
+#define CLK_PLL_MIPI           17

You can't really split code movement into two patches.

With this patch applied the clk driver will fail to build because the macros are now redefined in both header files.

Quoted 6 lines
Quoted 2 lines

Are you sure? The values seem the same to me... I don't see how this could fail.

Yes, IIRC the C standard allows repeated definitions with the same value. And I definitely tested this before (and hence recommended this approach to Vasily) and it compiled without any warnings here.

Hmm. Didn't know that. Good to know, and I just tried it on my end.

Thanks
ChenYu

Quoted 16 lines

Cheers,
Andre

Quoted 12 lines
Quoted 5 lines

Barring recombining the patches, please add a patch before this adding #ifndef's around the two macros that are moved.

No, not necessary, just churn,

Best regards,
Krzysztof

#9 lore raw

Re: [PATCH v2 1/4] dt-bindings: clock: sunxi: Export PLL_VIDEO_2X and PLL_MIPI

Quoted 5 lines · Chen-Yu Tsai
Quoted 3 lines

Yes, IIRC the C standard allows repeated definitions with the same value. And I definitely tested this before (and hence recommended this approach to Vasily) and it compiled without any warnings here.

Hmm. Didn't know that. Good to know, and I just tried it on my end.

Yeah, I don't like it either. It's an artefact of pretending that the device tree directory(ies) is a separate tree.

Quoted 2 lines

Thanks
ChenYu

#10 lore raw

Re: [PATCH v2 1/4] dt-bindings: clock: sunxi: Export PLL_VIDEO_2X and PLL_MIPI

In reply to Andre Przywara

Hello all,

Quoted 34 lines · Andre Przywara
Quoted 33 lines · On Sat, 4 Jan 2025 11:33:23 +0100 Krzysztof Kozlowski <[email protected]> wrote:
Quoted 31 lines · Chen-Yu Tsai
Quoted 19 lines · Vasily Khoruzhick

Export PLL_VIDEO_2X and PLL_MIPI, these will be used to explicitly select TCON0 clock parent in dts

---
 include/dt-bindings/clock/sun50i-a64-ccu.h | 2 ++
 1 file changed, 2 insertions(+)
include/dt-bindings/clock/sun50i-a64-ccu.h +0 -0
index 175892189e9d..4f220ea7a23c 100644

include/dt-bindings/clock/sun50i-a64-ccu.h line 49

Quoted 6 lines
include/dt-bindings/clock/sun50i-a64-ccu.h +2 -0
@@ -44,7 +44,9 @@
 #define _DT_BINDINGS_CLK_SUN50I_A64_H_
 
 #define CLK_PLL_VIDEO0         7
+#define CLK_PLL_VIDEO0_2X      8
 #define CLK_PLL_PERIPH0                11
+#define CLK_PLL_MIPI           17

You can't really split code movement into two patches.

With this patch applied the clk driver will fail to build because the macros are now redefined in both header files.

Quoted 6 lines
Quoted 2 lines

Are you sure? The values seem the same to me... I don't see how this could fail.

Yes, IIRC the C standard allows repeated definitions with the same value. And I definitely tested this before (and hence recommended this approach to Vasily) and it compiled without any warnings here.

FWIW, I also tested that approach when it was recommended, to make sure that the C compiler(s) are fine with that. And it worked. :)

Quoted 4 lines
Quoted 4 lines
Quoted 2 lines

Barring recombining the patches, please add a patch before this adding #ifndef's around the two macros that are moved.

No, not necessary, just churn,

#11 lore raw

PATCH v2 2/4 clk: sunxi-ng: a64: drop redundant CLK_PLL_VIDEO0_2X and CLK_PLL_MIPI

In reply to Vasily Khoruzhick

Drop redundant CLK_PLL_VIDEO0_2X and CLK_PLL.MIPI. These are now defined in dt-bindings/clock/sun50i-a64-ccu.h

---
 drivers/clk/sunxi-ng/ccu-sun50i-a64.h | 2 --
 1 file changed, 2 deletions(-)
drivers/clk/sunxi-ng/ccu-sun50i-a64.h +0 -2
index a8c11c0b4e06..dfba88a5ad0f 100644
@@ -21,7 +21,6 @@
 
 /* PLL_VIDEO0 exported for HDMI PHY */
 
-#define CLK_PLL_VIDEO0_2X		8
 #define CLK_PLL_VE			9
 #define CLK_PLL_DDR0			10
 
@@ -32,7 +31,6 @@
 #define CLK_PLL_PERIPH1_2X		14
 #define CLK_PLL_VIDEO1			15
 #define CLK_PLL_GPU			16
-#define CLK_PLL_MIPI			17
 #define CLK_PLL_HSIC			18
 #define CLK_PLL_DE			19
 #define CLK_PLL_DDR1			20
Signature · 1 line
2.47.1
#12 lore raw

PATCH v2 3/4 arm64: dts: allwinner: a64: explicitly assign clock parent for TCON0

In reply to Vasily Khoruzhick

TCON0 seems to need a different clock parent depending on output type. For RGB it has to be PLL-VIDEO0-2X, while for DSI it has to be PLL-MIPI, so select it explicitly.

Video output doesn't work if incorrect clock is assigned.

On my Pinebook I manually configured PLL-VIDEO0-2X and PLL-MIPI to the same rate, and while video output works fine with PLL-VIDEO0-2X, it doesn't work at all (as in no picture) with PLL-MIPI.

---
 arch/arm64/boot/dts/allwinner/sun50i-a64-pinebook.dts | 2 ++
 arch/arm64/boot/dts/allwinner/sun50i-a64-teres-i.dts  | 2 ++
 arch/arm64/boot/dts/allwinner/sun50i-a64.dtsi         | 2 ++
 3 files changed, 6 insertions(+)
arch/arm64/boot/dts/allwinner/sun50i-a64-pinebook.dts +2 -0
index 379c2c8466f5..86d44349e095 100644
@@ -390,6 +390,8 @@ &sound {
 &tcon0 {
 	pinctrl-names = "default";
 	pinctrl-0 = <&lcd_rgb666_pins>;
+	assigned-clocks = <&ccu CLK_TCON0>;
+	assigned-clock-parents = <&ccu CLK_PLL_VIDEO0_2X>;
 
 	status = "okay";
 };
arch/arm64/boot/dts/allwinner/sun50i-a64-teres-i.dts +2 -0
index b407e1dd08a7..ec055510af8b 100644
@@ -369,6 +369,8 @@ &sound {
 &tcon0 {
 	pinctrl-names = "default";
 	pinctrl-0 = <&lcd_rgb666_pins>;
+	assigned-clocks = <&ccu CLK_TCON0>;
+	assigned-clock-parents = <&ccu CLK_PLL_VIDEO0_2X>;
 
 	status = "okay";
 };
arch/arm64/boot/dts/allwinner/sun50i-a64.dtsi +2 -0
index a5c3920e0f04..0fecf0abb204 100644
@@ -445,6 +445,8 @@ tcon0: lcd-controller@1c0c000 {
 			clock-names = "ahb", "tcon-ch0";
 			clock-output-names = "tcon-data-clock";
 			#clock-cells = <0>;
+			assigned-clocks = <&ccu CLK_TCON0>;
+			assigned-clock-parents = <&ccu CLK_PLL_MIPI>;
 			resets = <&ccu RST_BUS_TCON0>, <&ccu RST_BUS_LVDS>;
 			reset-names = "lcd", "lvds";
 
Signature · 1 line
2.47.1
#13 lore raw

PATCH v2 4/4 clk: sunxi-ng: a64: stop force-selecting PLL-MIPI as TCON0 parent

In reply to Vasily Khoruzhick

Stop force-selecting PLL-MIPI as TCON0 parent, since it breaks video output on Pinebook that uses RGB to eDP bridge.

Partially revert commit ca1170b69968 ("clk: sunxi-ng: a64: force select PLL_MIPI in TCON0 mux"), while still leaving CLK_SET_RATE_NO_REPARENT flag set, since we do not want the clock to be reparented.

The issue is that apparently different TCON0 outputs require a different clock, or the mux might be selecting the output type.

I did an experiment: I manually configured PLL_MIPI and PLL_VIDEO0_2X to the same clock rate and flipped the switch with devmem. Experiment clearly showed that whenever PLL_MIPI is selected as TCON0 clock parent, the video output stops working.

Therefore, TCON0 clock parent corresponding to the output type must be assigned in the device tree.

---
 drivers/clk/sunxi-ng/ccu-sun50i-a64.c | 13 ++++---------
 1 file changed, 4 insertions(+), 9 deletions(-)
drivers/clk/sunxi-ng/ccu-sun50i-a64.c +4 -9
index 3a7d61c81667..ba1ad267f123 100644
@@ -535,11 +535,11 @@ static SUNXI_CCU_M_WITH_MUX_GATE(de_clk, "de", de_parents,
 				 CLK_SET_RATE_PARENT);
 
 /*
- * DSI output seems to work only when PLL_MIPI selected. Set it and prevent
- * the mux from reparenting.
+ * Experiments showed that RGB output requires pll-video0-2x, while DSI
+ * requires pll-mipi. It will not work with incorrect clock, the screen will
+ * be blank.
+ * sun50i-a64.dtsi assigns pll-mipi as TCON0 parent by default
  */
-#define SUN50I_A64_TCON0_CLK_REG	0x118
-
 static const char * const tcon0_parents[] = { "pll-mipi", "pll-video0-2x" };
 static const u8 tcon0_table[] = { 0, 2, };
 static SUNXI_CCU_MUX_TABLE_WITH_GATE_CLOSEST(tcon0_clk, "tcon0", tcon0_parents,
@@ -959,11 +959,6 @@ static int sun50i_a64_ccu_probe(struct platform_device *pdev)
 
 	writel(0x515, reg + SUN50I_A64_PLL_MIPI_REG);
 
-	/* Set PLL MIPI as parent for TCON0 */
-	val = readl(reg + SUN50I_A64_TCON0_CLK_REG);
-	val &= ~GENMASK(26, 24);
-	writel(val | (0 << 24), reg + SUN50I_A64_TCON0_CLK_REG);
-
 	ret = devm_sunxi_ccu_probe(&pdev->dev, reg, &sun50i_a64_ccu_desc);
 	if (ret)
 		return ret;
Signature · 1 line
2.47.1
Applied
#14 lore raw
In reply to Vasily Khoruzhick
Quoted 10 lines · Vasily Khoruzhick

Since commit ca1170b69968 ("clk: sunxi-ng: a64: force select PLL_MIPI in TCON0 mux"), TCON0 clock parent is always set to PLL_MIPI, but unfortunately it breaks video output on Pinebook.

I did an experiment: I manually configured PLL_MIPI and PLL_VIDEO0_2X to the same clock rate and flipped the switch with devmem. Experiment clearly showed that whenever PLL_MIPI is selected as TCON0 clock parent, the video output stops working.

[...]

Applied to sunxi/for-next in sunxi/linux.git, thanks!

[1/4] dt-bindings: clock: sunxi: Export PLL_VIDEO_2X and PLL_MIPI

      https://git.kernel.org/sunxi/linux/c/9897831de614
[2/4] clk: sunxi-ng: a64: drop redundant CLK_PLL_VIDEO0_2X and CLK_PLL_MIPI
      https://git.kernel.org/sunxi/linux/c/0f368cb7ef10
[3/4] arm64: dts: allwinner: a64: explicitly assign clock parent for TCON0
      https://git.kernel.org/sunxi/linux/c/8715c91a8365
[4/4] clk: sunxi-ng: a64: stop force-selecting PLL-MIPI as TCON0 parent
      https://git.kernel.org/sunxi/linux/c/383ca7bee8a9

Best regards,

Signature · 1 line
Chen-Yu Tsai <[email protected]>