Skip to contentSkip to conversation

clk: fixed-rate: use full DT node name

PATCH New Stephen Warren 3 participants 9 messages Feb 14, 2014 – Feb 18, 2014

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/2 clk: fixed-rate: use full DT node name

clk-fixed-rate currently names clocks according to a node's name without the unit address. When faced with the legal and technically correct DT structure below, this causes rgistration attempts for 3 clocks with the same name, 2 of which fail.

	clocks {
		compatible = "simple-bus";
		#address-cells = <1>;
		#size-cells = <0>;

		clk_mmc: clock at 0 {
			compatible = "fixed-clock";
			reg = <0>;
...
		clk_i2c: clock at 1 {
			compatible = "fixed-clock";
			reg = <1>;
...
		clk_spi: clock at 2 {
			compatible = "fixed-clock";
			reg = <2>;

...

Solve this by naming the clocks after the full node name rather than the short version (e.g. /clocks/clock at 0).

---
Note that if this is accepted, I intend to submit a patch for the RPi DTS which uses the naming structure above, so it might be useful to place this patch in its own branch. Or, I could submit the cleanup after 3.15-rc1.

---
 drivers/clk/clk-fixed-rate.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/clk/clk-fixed-rate.c +1 -1
index 0fc56ab..3335b3c 100644
@@ -115,7 +115,7 @@ EXPORT_SYMBOL_GPL(clk_register_fixed_rate);
 void of_fixed_clk_setup(struct device_node *node)
 {
 	struct clk *clk;
-	const char *clk_name = node->name;
+	const char *clk_name = node->full_name;
 	u32 rate;
 	u32 accuracy = 0;
 
Signature · 1 line
1.8.3.2
#1 lore raw

PATCH 2/2 ARM: bcm2835: fix clock DT node names

DT nodes should be named according to the type of object that they represent rather than the identity. DT nodes that contain a reg property should include a unit address in their name. Fix these issues.

---
This is the bcm2835 patch which depends on patch 1. I guess this could also go through the clk tree if that makes it easier, although there's always the small risk of conflicts if you do that.

 arch/arm/boot/dts/bcm2835.dtsi | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
arch/arm/boot/dts/bcm2835.dtsi +3 -3
index a2d4889..1cf1ae7 100644
@@ -155,21 +155,21 @@
 		#address-cells = <1>;
 		#size-cells = <0>;
 
-		clk_mmc: mmc {
+		clk_mmc: clock at 0 {
 			compatible = "fixed-clock";
 			reg = <0>;
 			#clock-cells = <0>;
 			clock-frequency = <100000000>;
 		};
 
-		clk_i2c: i2c {
+		clk_i2c: clock at 1 {
 			compatible = "fixed-clock";
 			reg = <1>;
 			#clock-cells = <0>;
 			clock-frequency = <250000000>;
 		};
 
-		clk_spi: spi {
+		clk_spi: clock at 2 {
 			compatible = "fixed-clock";
 			reg = <2>;
 			#clock-cells = <0>;
Signature · 1 line
1.8.3.2
#2 lore raw

PATCH 2/2 ARM: bcm2835: fix clock DT node names

Quoted 3 lines · Stephen Warren

DT nodes should be named according to the type of object that they represent rather than the identity. DT nodes that contain a reg property should include a unit address in their name. Fix these issues.

As mentioned on patch one, I don't think this makes sense. clock at N is simply not correct, and if simple-bus weren't being abused it would be far clearer that that is the case.

Please use clock_N rather than clock at N here, and get rid of the meaningless reg values.

The simple-bus should either be removed or fixed up to meet the requirements of the simple-bus binding (i.e. add a ranges property). I would prefer the former.

Thanks,
Mark.

Quoted 47 lines

---
This is the bcm2835 patch which depends on patch 1. I guess this could also go through the clk tree if that makes it easier, although there's always the small risk of conflicts if you do that.

 arch/arm/boot/dts/bcm2835.dtsi | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
arch/arm/boot/dts/bcm2835.dtsi +3 -3
index a2d4889..1cf1ae7 100644
@@ -155,21 +155,21 @@
 		#address-cells = <1>;
 		#size-cells = <0>;
 
-		clk_mmc: mmc {
+		clk_mmc: clock at 0 {
 			compatible = "fixed-clock";
 			reg = <0>;
 			#clock-cells = <0>;
 			clock-frequency = <100000000>;
 		};
 
-		clk_i2c: i2c {
+		clk_i2c: clock at 1 {
 			compatible = "fixed-clock";
 			reg = <1>;
 			#clock-cells = <0>;
 			clock-frequency = <250000000>;
 		};
 
-		clk_spi: spi {
+		clk_spi: clock at 2 {
 			compatible = "fixed-clock";
 			reg = <2>;
 			#clock-cells = <0>;
Signature · 6 lines
1.8.3.2

--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo at vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#3 lore raw

PATCH 1/2 clk: fixed-rate: use full DT node name

In reply to Stephen Warren
Quoted 21 lines · Stephen Warren

clk-fixed-rate currently names clocks according to a node's name without the unit address. When faced with the legal and technically correct DT structure below, this causes rgistration attempts for 3 clocks with the same name, 2 of which fail.

	clocks {
		compatible = "simple-bus";
		#address-cells = <1>;
		#size-cells = <0>;

		clk_mmc: clock at 0 {
			compatible = "fixed-clock";
			reg = <0>;
...
		clk_i2c: clock at 1 {
			compatible = "fixed-clock";
			reg = <1>;
...
		clk_spi: clock at 2 {
			compatible = "fixed-clock";
			reg = <2>;
Quoted 1 line

...

I'd argue that this case isn't valid.

The fixed-clock binding doesn't define a reg, yet simple bus binding implies that the reg property of child nodes should be interpretted as the same address space as their parent (MMIO in this case?). The fixed-clock nodes reg proeprties clearly aren't MMIO addresses.

Additionally, the _requred_ ranges property is missing.

It's just nonsensical; rename them to clock_{0,1,..} instead and get rid of the reg properties. Then they're named uniquely.

However, for cases where the reg value is meaningful the below patch makes sense.

Thanks,
Mark.

Quoted 34 lines

Solve this by naming the clocks after the full node name rather than the short version (e.g. /clocks/clock at 0).

---
Note that if this is accepted, I intend to submit a patch for the RPi DTS which uses the naming structure above, so it might be useful to place this patch in its own branch. Or, I could submit the cleanup after 3.15-rc1.

---
 drivers/clk/clk-fixed-rate.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/clk/clk-fixed-rate.c +1 -1
index 0fc56ab..3335b3c 100644
@@ -115,7 +115,7 @@ EXPORT_SYMBOL_GPL(clk_register_fixed_rate);
 void of_fixed_clk_setup(struct device_node *node)
 {
 	struct clk *clk;
-	const char *clk_name = node->name;
+	const char *clk_name = node->full_name;
 	u32 rate;
 	u32 accuracy = 0;
 
Signature · 6 lines
1.8.3.2

--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo at vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#4 lore raw

PATCH 1/2 clk: fixed-rate: use full DT node name

Quoted 24 lines · Mark Rutland
Quoted 21 lines · Stephen Warren

clk-fixed-rate currently names clocks according to a node's name without the unit address. When faced with the legal and technically correct DT structure below, this causes rgistration attempts for 3 clocks with the same name, 2 of which fail.

	clocks {
		compatible = "simple-bus";
		#address-cells = <1>;
		#size-cells = <0>;

		clk_mmc: clock at 0 {
			compatible = "fixed-clock";
			reg = <0>;
...
		clk_i2c: clock at 1 {
			compatible = "fixed-clock";
			reg = <1>;
...
		clk_spi: clock at 2 {
			compatible = "fixed-clock";
			reg = <2>;
Quoted 1 line

...

Quoted 1 line

I'd argue that this case isn't valid.

Well, it's very widely used, and was the result of numerous discussions of how this kind of thing should be represented:-/

Quoted 6 lines

The fixed-clock binding doesn't define a reg, yet simple bus binding implies that the reg property of child nodes should be interpretted as the same address space as their parent (MMIO in this case?). The fixed-clock nodes reg proeprties clearly aren't MMIO addresses.

Additionally, the _requred_ ranges property is missing.

Perhaps we need to invent a simple-container instead then?

Quoted 2 lines

It's just nonsensical; rename them to clock_{0,1,..} instead and get rid of the reg properties. Then they're named uniquely.

That's not legal either. DT node names are supposed to represent the type of device/object (i.e. just "clock"), not the identity of the device/object (i.e. not include IDs etc.). Hence, the node name needs to be "clock" for all of them, and the unit address must be used to differentiate them.

#5 lore raw

PATCH 1/2 clk: fixed-rate: use full DT node name

Quoted 25 lines · Stephen Warren
Quoted 24 lines · Mark Rutland
Quoted 22 lines · Stephen Warren

clk-fixed-rate currently names clocks according to a node's name without the unit address. When faced with the legal and technically correct DT structure below, this causes rgistration attempts for 3 clocks with the same name, 2 of which fail.

	clocks {
		compatible = "simple-bus";
		#address-cells = <1>;
		#size-cells = <0>;

		clk_mmc: clock at 0 {
			compatible = "fixed-clock";
			reg = <0>;
...
		clk_i2c: clock at 1 {
			compatible = "fixed-clock";
			reg = <1>;
...
		clk_spi: clock at 2 {
			compatible = "fixed-clock";
			reg = <2>;

...

Quoted 4 lines
Quoted 1 line

I'd argue that this case isn't valid.

Well, it's very widely used, and was the result of numerous discussions of how this kind of thing should be represented:-/

Maybe we have to live with it then. :/

Quoted 9 lines
Quoted 7 lines

The fixed-clock binding doesn't define a reg, yet simple bus binding implies that the reg property of child nodes should be interpretted as the same address space as their parent (MMIO in this case?). The fixed-clock nodes reg proeprties clearly aren't MMIO addresses.

Additionally, the _requred_ ranges property is missing.

Perhaps we need to invent a simple-container instead then?

As you mention in your other reply, it makes some sense to omit ranges if the address space is disjoint. My concern is that the address space is meaningless and arbitrary. If we had a real disjoint address space we'd have another kind of bus node as the container.

Quoted 9 lines
Quoted 3 lines

It's just nonsensical; rename them to clock_{0,1,..} instead and get rid of the reg properties. Then they're named uniquely.

That's not legal either. DT node names are supposed to represent the type of device/object (i.e. just "clock"), not the identity of the device/object (i.e. not include IDs etc.). Hence, the node name needs to be "clock" for all of them, and the unit address must be used to differentiate them.

As far as I can see from ePAPR, the only requriement is:

The node-name shall start with a lower or uppercase character and should describe the general class of device.

IMO clock_1 describes the general class of device as well as clock at 1, while also not filling a unexpected property with a meaningless value.

Thanks,
Mark.

#6 lore raw

PATCH 1/2 clk: fixed-rate: use full DT node name

Quoted 26 lines · Mark Rutland
Quoted 25 lines · Stephen Warren
Quoted 24 lines · Mark Rutland
Quoted 22 lines · Stephen Warren

clk-fixed-rate currently names clocks according to a node's name without the unit address. When faced with the legal and technically correct DT structure below, this causes rgistration attempts for 3 clocks with the same name, 2 of which fail.

	clocks {
		compatible = "simple-bus";
		#address-cells = <1>;
		#size-cells = <0>;

		clk_mmc: clock at 0 {
			compatible = "fixed-clock";
			reg = <0>;
...
		clk_i2c: clock at 1 {
			compatible = "fixed-clock";
			reg = <1>;
...
		clk_spi: clock at 2 {
			compatible = "fixed-clock";
			reg = <2>;

...

Quoted 6 lines
Quoted 4 lines
Quoted 1 line

I'd argue that this case isn't valid.

Well, it's very widely used, and was the result of numerous discussions of how this kind of thing should be represented:-/

Maybe we have to live with it then. :/

Great:-)

...

Quoted 9 lines
Quoted 8 lines
Quoted 2 lines

It's just nonsensical; rename them to clock_{0,1,..} instead and get rid of the reg properties. Then they're named uniquely.

That's not legal either. DT node names are supposed to represent the type of device/object (i.e. just "clock"), not the identity of the device/object (i.e. not include IDs etc.). Hence, the node name needs to be "clock" for all of them, and the unit address must be used to differentiate them.

Quoted 7 lines

As far as I can see from ePAPR, the only requriement is:

The node-name shall start with a lower or uppercase character and should describe the general class of device.

IMO clock_1 describes the general class of device as well as clock at 1, while also not filling a unexpected property with a meaningless value.

I believe section 2.2.2 "Generic Names Recommendation" is the source of the rule that nodes should be named after the type of object rather than the identity.

"""
The name of a node should be somewhat generic, reflecting the function of the device and not its precise programming model. If appropriate, the name should be one of the following choices:

* atm
* cache-controller
* compact-flash
* can
* cpu
...
"""

(I note e.g. "cpu" not "cpu-1", "cpu_2", etc.)

#7 lore raw

PATCH 1/2 clk: fixed-rate: use full DT node name

In reply to Mark Rutland
Quoted 24 lines · Mark Rutland
Quoted 21 lines · Stephen Warren

clk-fixed-rate currently names clocks according to a node's name without the unit address. When faced with the legal and technically correct DT structure below, this causes rgistration attempts for 3 clocks with the same name, 2 of which fail.

	clocks {
		compatible = "simple-bus";
		#address-cells = <1>;
		#size-cells = <0>;

		clk_mmc: clock at 0 {
			compatible = "fixed-clock";
			reg = <0>;
...
		clk_i2c: clock at 1 {
			compatible = "fixed-clock";
			reg = <1>;
...
		clk_spi: clock at 2 {
			compatible = "fixed-clock";
			reg = <2>;
Quoted 1 line

...

Quoted 8 lines

I'd argue that this case isn't valid.

The fixed-clock binding doesn't define a reg, yet simple bus binding implies that the reg property of child nodes should be interpretted as the same address space as their parent (MMIO in this case?). The fixed-clock nodes reg proeprties clearly aren't MMIO addresses.

Additionally, the _requred_ ranges property is missing.

Oh, IIRC that was deliberate to indicate that the child address space was disjoint from the parent address space.

#8 lore raw

PATCH 1/2 clk: fixed-rate: use full DT node name

In reply to Stephen Warren

Hello Stephen,

El 14/02/14 03:16, Stephen Warren escribi?:

Quoted 21 lines

clk-fixed-rate currently names clocks according to a node's name without the unit address. When faced with the legal and technically correct DT structure below, this causes rgistration attempts for 3 clocks with the same name, 2 of which fail.

	clocks {
		compatible = "simple-bus";
		#address-cells = <1>;
		#size-cells = <0>;

		clk_mmc: clock at 0 {
			compatible = "fixed-clock";
			reg = <0>;
...
		clk_i2c: clock at 1 {
			compatible = "fixed-clock";
			reg = <1>;
...
		clk_spi: clock at 2 {
			compatible = "fixed-clock";
			reg = <2>;
Quoted 4 lines

...

Solve this by naming the clocks after the full node name rather than the short version (e.g. /clocks/clock at 0).

An alternative that doesn't require any change in the driver would be to use the optional but recommended clock-output-names property from the common clock binding.

Your DT would then look something like the following

Quoted 12 lines

...

		clk_mmc: clk at 0 {
			compatible = "fixed-clock";
			clock-output-names = "mmc";
...
		clk_i2c: clk at 1 {
			compatible = "fixed-clock";
			clock-output-names = "i2c";
...
		clk_spi: clk at 2 {
			compatible = "fixed-clock";
			clock-output-names = "spi";
Quoted 1 line

...

Cheers,

Emilio