Skip to contentSkip to conversation

ARM: dts: marvell: add missing SDHCI clock-names

PATCH In review Rosen Penev 3 participants 3 messages Aug 23 – Aug 28

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ARM: dts: marvell: add missing SDHCI clock-names

The marvell,armada-380-sdhci binding requires "clock-names", and the sdhci-pxav3 driver requests its io clock by name before falling back to an unnamed lookup. Both Armada 38x and 39x nodes only carry an unnamed clocks reference, so dtbs_check reports "'clock-names' is a required property" on every board.

Add clock-names = "io" to both nodes. No functional change: the driver already obtains the clock through its fallback.

Assisted-by opencode:big-pickle
---
 arch/arm/boot/dts/marvell/armada-38x.dtsi | 1 +
 arch/arm/boot/dts/marvell/armada-39x.dtsi | 1 +
 2 files changed, 2 insertions(+)
arch/arm/boot/dts/marvell/armada-38x.dtsi +1 -0
index 1155bee20523..598022645dd2 100644
@@ -627,6 +627,7 @@ sdhci: mmc@d8000 {
 					<0x18454 0x4>;
 				interrupts = <GIC_SPI 25 IRQ_TYPE_LEVEL_HIGH>;
 				clocks = <&gateclk 17>;
+				clock-names = "io";
 				mrvl,clk-delay-cycles = <0x1F>;
 				status = "disabled";
 			};
arch/arm/boot/dts/marvell/armada-39x.dtsi +1 -0
index 218fcf99ae07..81dd4ca5fbc9 100644
@@ -381,6 +381,7 @@ mmc@d8000 {
 					<0x18454 0x4>;
 				interrupts = <GIC_SPI 25 IRQ_TYPE_LEVEL_HIGH>;
 				clocks = <&gateclk 17>;
+				clock-names = "io";
 				mrvl,clk-delay-cycles = <0x1F>;
 				status = "disabled";
 			};
Signature · 1 line
2.55.0
#1 lore raw
Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Pre-existing issues:
- [High] Unchecked return value of clk_prepare_enable() leading to potential unclocked hardware accesses.
- [High] Suppression of legitimate errors (e.g., -EPROBE_DEFER) when fetching the optional 'core' clock. --

--- Patch [1]: [PATCH] ARM: dts: marvell: add missing SDHCI clock-names --- commit 4be703994612e3151cc7dd6a9b229ae270b6b4da
Author: Rosen Penev <[email protected]>

ARM: dts: marvell: add missing SDHCI clock-names

The marvell,armada-380-sdhci binding requires "clock-names", and the sdhci-pxav3 driver requests its io clock by name before falling back to an unnamed lookup. Both Armada 38x and 39x nodes only carry an unnamed clocks reference, so dtbs_check reports "'clock-names' is a required property" on every board.

Add clock-names = "io" to both nodes. No functional change: the driver already obtains the clock through its fallback.

Quoted 4 lines
arch/arm/boot/dts/marvell/armada-38x.dtsi +0 -0
index 1d616edda322d..0618779090d04 100644

[ ... ]

arch/arm/boot/dts/marvell/armada-38x.dtsi sdhci: sdhci@d8000 { · line 630

Quoted 5 lines
arch/arm/boot/dts/marvell/armada-38x.dtsi +1 -0
@@ -627,6 +627,7 @@ sdhci: sdhci@d8000 {
 					<0x18454 0x4>;
 				interrupts = <GIC_SPI 25 IRQ_TYPE_LEVEL_HIGH>;
 				clocks = <&gateclk 17>;
+				clock-names = "io";

[Severity: High]
This isn't a bug introduced by this patch, but while reviewing the sdhci-pxav3 driver mentioned in the commit message, I noticed a pre-existing issue in how these clocks are handled.

Looking at sdhci_pxav3_probe() in drivers/mmc/host/sdhci-pxav3.c:

drivers/mmc/host/sdhci-pxav3.c:sdhci_pxav3_probe() {
    ...
	pltfm_host->clk = pxa->clk_io;
	clk_prepare_enable(pxa->clk_io);
    ...
	pxa->clk_core = devm_clk_get(dev, "core");
	if (!IS_ERR(pxa->clk_core))
		clk_prepare_enable(pxa->clk_core);
    ...
}

Are we ignoring the return values of clk_prepare_enable() here? If the clock subsystem fails to enable the IO or core clocks during probe, the driver might access hardware registers while the clock is disabled or failed to enable, which could lead to a synchronous external abort.

[Severity: High]
This is also a pre-existing issue, but does this code suppress legitimate errors like -EPROBE_DEFER when fetching the optional 'core' clock?

drivers/mmc/host/sdhci-pxav3.c:sdhci_pxav3_probe() {
    ...
	pxa->clk_core = devm_clk_get(dev, "core");
	if (!IS_ERR(pxa->clk_core))
		clk_prepare_enable(pxa->clk_core);
    ...
}

If devm_clk_get() returns -EPROBE_DEFER because the clock provider is not yet ready, the if (!IS_ERR(pxa->clk_core)) condition treats this as if the clock is absent and proceeds silently without it. Could this break device initialization if the clock is actually required but deferred?

The unchecked clk_prepare_enable() issue also appears in the runtime resume path:

drivers/mmc/host/sdhci-pxav3.c:sdhci_pxav3_runtime_resume() {
    ...
	clk_prepare_enable(pxa->clk_io);
	if (!IS_ERR(pxa->clk_core))
		clk_prepare_enable(pxa->clk_core);

	sdhci_runtime_resume_host(host, 0);
	return 0;
}

Can the clocks fail to enable during runtime resume, leading to the same risk of unclocked hardware access when sdhci_runtime_resume_host() is called?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
maintainer
#2 lore raw
In reply to Rosen Penev
Quoted 11 lines · Rosen Penev

The marvell,armada-380-sdhci binding requires "clock-names", and the sdhci-pxav3 driver requests its io clock by name before falling back to an unnamed lookup. Both Armada 38x and 39x nodes only carry an unnamed clocks reference, so dtbs_check reports "'clock-names' is a required property" on every board.

Add clock-names = "io" to both nodes. No functional change: the driver already obtains the clock through its fallback.

Assisted-by opencode:big-pickle

Applied on mvebu/dt

Thanks,

Gregory

Quoted 32 lines
---
 arch/arm/boot/dts/marvell/armada-38x.dtsi | 1 +
 arch/arm/boot/dts/marvell/armada-39x.dtsi | 1 +
 2 files changed, 2 insertions(+)
arch/arm/boot/dts/marvell/armada-38x.dtsi +1 -0
index 1155bee20523..598022645dd2 100644
@@ -627,6 +627,7 @@ sdhci: mmc@d8000 {
 					<0x18454 0x4>;
 				interrupts = <GIC_SPI 25 IRQ_TYPE_LEVEL_HIGH>;
 				clocks = <&gateclk 17>;
+				clock-names = "io";
 				mrvl,clk-delay-cycles = <0x1F>;
 				status = "disabled";
 			};
arch/arm/boot/dts/marvell/armada-39x.dtsi +1 -0
index 218fcf99ae07..81dd4ca5fbc9 100644
@@ -381,6 +381,7 @@ mmc@d8000 {
 					<0x18454 0x4>;
 				interrupts = <GIC_SPI 25 IRQ_TYPE_LEVEL_HIGH>;
 				clocks = <&gateclk 17>;
+				clock-names = "io";
 				mrvl,clk-delay-cycles = <0x1F>;
 				status = "disabled";
 			};
Signature · 1 line
2.55.0
Signature · 3 lines
Grégory CLEMENT, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com