Skip to contentSkip to conversation

Alternative clk gate ctrl for Kirkwood variants

RFC PATCH New Valentin Longchamp 5 participants 11 messages May 7, 2013 – May 8, 2013

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 Alternative clk gate ctrl for Kirkwood variants

This series introduces a clk gate ctrl for the kirkwood variants embedded in the Prestera SoCs.

This would allow to get rid of the workaround for the km_kirkwood board support file that was added previously because of the lack of DT clk gate support.

Valentin Longchamp (3):
clk: mvebu: add gate ctrl for Prestera kirkwood variants
ARM: kirkwood: look for alternative gate ctrl of node
ARM: kirkwood: remove clock gating disabling for km_kirkwood

 .../devicetree/bindings/clock/mvebu-gated-clock.txt   | 17 +++++++++++++++++
 arch/arm/boot/dts/kirkwood-98dx4122.dtsi              |  7 +++++++
 arch/arm/mach-kirkwood/board-dt.c                     |  4 ++++
 arch/arm/mach-kirkwood/board-km_kirkwood.c            | 15 ---------------
 drivers/clk/mvebu/clk-gating-ctrl.c                   | 19 +++++++++++++++++++
 5 files changed, 47 insertions(+), 15 deletions(-)
Signature · 1 line
1.8.0.1
#1 lore raw

PATCH 1/3 clk: mvebu: add gate ctrl for Prestera kirkwood variants

The kirkwood device found in the Prestera SoCs does not have all the peripherals of its the usual kirkwood SoCs. There are hence missing clocks in the SoCs.

This patch registers another gate controller for the kirkwood that registers only the available clocks of this kirkwood variant.

---
 .../devicetree/bindings/clock/mvebu-gated-clock.txt   | 17 +++++++++++++++++
 drivers/clk/mvebu/clk-gating-ctrl.c                   | 19 +++++++++++++++++++
 2 files changed, 36 insertions(+)
Documentation/devicetree/bindings/clock/mvebu-gated-clock.txt +17 -0
index cffc93d..7f494e90 100644
@@ -92,10 +92,27 @@ ID	Clock	Peripheral
 19	ge1	Gigabit Ethernet 1
 20	tdm	Time Division Mplx
 
+The following is a list of provided IDs for Kirkwood Prestera variant:
+ID	Clock	Peripheral
+-----------------------------------
+0	ge0	Gigabit Ethernet 0
+2	pex0	PCIe Cntrl 0
+3	usb0	USB Host 0
+4	sdio	SDIO Cntrl
+5	tsu	Transp. Stream Unit
+6	dunit	SDRAM Cntrl
+7	runit	Runit
+8	xor0	XOR DMA 0
+16	xor1	XOR DMA 1
+17	crypto	CESA engine
+19	ge1	Gigabit Ethernet 1
+
+
 Required properties:
 - compatible : shall be one of the following:
 	"marvell,dove-gating-clock" - for Dove SoC clock gating
 	"marvell,kirkwood-gating-clock" - for Kirkwood SoC clock gating
+	"marvell,prestera-kw-gating-clock" - for Preseta SoC clock gating
 - reg : shall be the register address of the Clock Gating Control register
 - #clock-cells : from common clock binding; shall be set to 1
 
drivers/clk/mvebu/clk-gating-ctrl.c +19 -0
index ebf141d..ba37802 100644
@@ -203,6 +203,21 @@ static const struct mvebu_soc_descr __initconst kirkwood_gating_descr[] = {
 	{ "tdm", NULL, 20 },
 	{ }
 };
+
+static const struct mvebu_soc_descr __initconst prestera_kw_gating_descr[] = {
+	{ "ge0", NULL, 0 },
+	{ "pex0", NULL, 2 },
+	{ "usb0", NULL, 3 },
+	{ "sdio", NULL, 4 },
+	{ "tsu", NULL, 5 },
+	{ "runit", NULL, 7 },
+	{ "xor0", NULL, 8 },
+	{ "powersave", "cpuclk", 11 },
+	{ "xor1", NULL, 16 },
+	{ "crypto", NULL, 17 },
+	{ "ge1", NULL, 19 },
+	{ }
+};
 #endif
 
 static const __initdata struct of_device_id clk_gating_match[] = {
@@ -232,6 +247,10 @@ static const __initdata struct of_device_id clk_gating_match[] = {
 		.compatible = "marvell,kirkwood-gating-clock",
 		.data = kirkwood_gating_descr,
 	},
+	{
+		.compatible = "marvell,prestera-kw-gating-clock",
+		.data = prestera_kw_gating_descr,
+	},
 #endif
 
 	{ }
Signature · 1 line
1.8.0.1
#2 lore raw

PATCH 1/3 clk: mvebu: add gate ctrl for Prestera kirkwood variants

Quoted 6 lines · Valentin Longchamp

The kirkwood device found in the Prestera SoCs does not have all the peripherals of its the usual kirkwood SoCs. There are hence missing clocks in the SoCs.

This patch registers another gate controller for the kirkwood that registers only the available clocks of this kirkwood variant.

Hi Valentin,

Overall the patch set looks OK, however I don't understand why you need to declare a new gate controller. The list you delcared seemed to be just a subset of the Kirkwood one, why can't you use this one?

Note that I am not an expert for the kirkwood hardware, so maybe I missed something.

Regards,

Quoted 77 lines
---
 .../devicetree/bindings/clock/mvebu-gated-clock.txt   | 17 +++++++++++++++++
 drivers/clk/mvebu/clk-gating-ctrl.c                   | 19 +++++++++++++++++++
 2 files changed, 36 insertions(+)
Documentation/devicetree/bindings/clock/mvebu-gated-clock.txt +17 -0
index cffc93d..7f494e90 100644
@@ -92,10 +92,27 @@ ID	Clock	Peripheral
 19	ge1	Gigabit Ethernet 1
 20	tdm	Time Division Mplx
 
+The following is a list of provided IDs for Kirkwood Prestera variant:
+ID	Clock	Peripheral
+-----------------------------------
+0	ge0	Gigabit Ethernet 0
+2	pex0	PCIe Cntrl 0
+3	usb0	USB Host 0
+4	sdio	SDIO Cntrl
+5	tsu	Transp. Stream Unit
+6	dunit	SDRAM Cntrl
+7	runit	Runit
+8	xor0	XOR DMA 0
+16	xor1	XOR DMA 1
+17	crypto	CESA engine
+19	ge1	Gigabit Ethernet 1
+
+
 Required properties:
 - compatible : shall be one of the following:
 	"marvell,dove-gating-clock" - for Dove SoC clock gating
 	"marvell,kirkwood-gating-clock" - for Kirkwood SoC clock gating
+	"marvell,prestera-kw-gating-clock" - for Preseta SoC clock gating
 - reg : shall be the register address of the Clock Gating Control register
 - #clock-cells : from common clock binding; shall be set to 1
 
drivers/clk/mvebu/clk-gating-ctrl.c +19 -0
index ebf141d..ba37802 100644
@@ -203,6 +203,21 @@ static const struct mvebu_soc_descr __initconst kirkwood_gating_descr[] = {
 	{ "tdm", NULL, 20 },
 	{ }
 };
+
+static const struct mvebu_soc_descr __initconst prestera_kw_gating_descr[] = {
+	{ "ge0", NULL, 0 },
+	{ "pex0", NULL, 2 },
+	{ "usb0", NULL, 3 },
+	{ "sdio", NULL, 4 },
+	{ "tsu", NULL, 5 },
+	{ "runit", NULL, 7 },
+	{ "xor0", NULL, 8 },
+	{ "powersave", "cpuclk", 11 },
+	{ "xor1", NULL, 16 },
+	{ "crypto", NULL, 17 },
+	{ "ge1", NULL, 19 },
+	{ }
+};
 #endif
 
 static const __initdata struct of_device_id clk_gating_match[] = {
@@ -232,6 +247,10 @@ static const __initdata struct of_device_id clk_gating_match[] = {
 		.compatible = "marvell,kirkwood-gating-clock",
 		.data = kirkwood_gating_descr,
 	},
+	{
+		.compatible = "marvell,prestera-kw-gating-clock",
+		.data = prestera_kw_gating_descr,
+	},
 #endif
 
 	{ }
Signature · 4 lines
Gregory Clement, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com
maintainer
#3 lore raw

PATCH 1/3 clk: mvebu: add gate ctrl for Prestera kirkwood variants

Quoted 5 lines · Gregory CLEMENT
Quoted 4 lines · Valentin Longchamp

The kirkwood device found in the Prestera SoCs does not have all the peripherals of its the usual kirkwood SoCs. There are hence missing clocks in the SoCs.

Quoted 8 lines
Quoted 2 lines

This patch registers another gate controller for the kirkwood that registers only the available clocks of this kirkwood variant.

Hi Valentin,

Overall the patch set looks OK, however I don't understand why you need to declare a new gate controller. The list you delcared seemed to be just a subset of the Kirkwood one, why can't you use this one?

Hi Gregory

Take a look at the code which gets deleted in patch 3/3.

/*
* Our variant of kirkwood (integrated in the Bobcat) hangs on accessing
* SATA bits (14-15) of the Clock Gating Control Register. Since these
* devices are also not present in this variant, their clocks get
* disabled because unused when clk_disable_unused() gets called.
* That's why we change the flags to these clocks to CLK_IGNORE_UNUSED */

By not defining these clocks in the first place, they never get turned off...

Andrew

#4 lore raw

PATCH 1/3 clk: mvebu: add gate ctrl for Prestera kirkwood variants

In reply to Gregory CLEMENT
Quoted 8 lines · Gregory CLEMENT
Quoted 6 lines · Valentin Longchamp

The kirkwood device found in the Prestera SoCs does not have all the peripherals of its the usual kirkwood SoCs. There are hence missing clocks in the SoCs.

This patch registers another gate controller for the kirkwood that registers only the available clocks of this kirkwood variant.

Quoted 8 lines

Hi Valentin,

Overall the patch set looks OK, however I don't understand why you need to declare a new gate controller. The list you delcared seemed to be just a subset of the Kirkwood one, why can't you use this one?

Note that I am not an expert for the kirkwood hardware, so maybe I missed something.

I just checked my mails from late 2012 and there Valentin and I agreed, that not the missing bits in clock gating control registers was the root cause of km_kirkwood to hang. It was rather the phy gates (that do not/ not yet exist on DT) that caused it to hang, i.e. when accessing SATA PHY registers.

Valentin, I am not against an extra clock-gating-ctrl for Prestera but maybe having an kirkwood.dtsi fork without SATA and other peripherals will also help?

And: Does DT-enabled km_kirkwood ever hang without the hack in it's board setup? There should be no PHY gates on DT boards..

Sebastian

#5 lore raw

PATCH 1/3 clk: mvebu: add gate ctrl for Prestera kirkwood variants

Quoted 15 lines · Sebastian Hesselbarth
Quoted 14 lines · Gregory CLEMENT
Quoted 6 lines · Valentin Longchamp

The kirkwood device found in the Prestera SoCs does not have all the peripherals of its the usual kirkwood SoCs. There are hence missing clocks in the SoCs.

This patch registers another gate controller for the kirkwood that registers only the available clocks of this kirkwood variant.

Hi Valentin,

Overall the patch set looks OK, however I don't understand why you need to declare a new gate controller. The list you delcared seemed to be just a subset of the Kirkwood one, why can't you use this one?

Quoted 8 lines
Quoted 2 lines

Note that I am not an expert for the kirkwood hardware, so maybe I missed something.

I just checked my mails from late 2012 and there Valentin and I agreed, that not the missing bits in clock gating control registers was the root cause of km_kirkwood to hang. It was rather the phy gates (that do not/ not yet exist on DT) that caused it to hang, i.e. when accessing SATA PHY registers.

Yeah you are right, I had forgotten about this test. I just had a look at the board-km_kirkwood.c and I tried to figure out a way to make sure the clock gates bits never get read/written, so never get initialized. But as you say, that's not the real cause of the hang but

Quoted 4 lines

Valentin, I am not against an extra clock-gating-ctrl for Prestera but maybe having an kirkwood.dtsi fork without SATA and other peripherals will also help?

Since it's not the real problem having the extra clock-gating-ctrl would me more consistent with the real Prestera HW, but maybe it's better to keep only one that works for both as it's already the case, it's less code to maintain.

A better approach would be to work on the kirkwood.dtsi as you mention it. The SATA node in kirkwood.dtsi is by default disabled, is this enough or should it be moved to kirkwood-6281.dtsi for instance ?

Quoted 4 lines

And: Does DT-enabled km_kirkwood ever hang without the hack in it's board setup? There should be no PHY gates on DT boards..

I have just tested it and it does not hang, so I will submit and patch that simply removes this workaround since the PHY gates are not accessed anymore (thanks to DT clock gates and that km_kirkwood does not enable the sata of node).

Valentin

#6 lore raw

PATCH 1/3 clk: mvebu: add gate ctrl for Prestera kirkwood variants

Quoted 16 lines · Valentin Longchamp
Quoted 5 lines · Sebastian Hesselbarth

I just checked my mails from late 2012 and there Valentin and I agreed, that not the missing bits in clock gating control registers was the root cause of km_kirkwood to hang. It was rather the phy gates (that do not/ not yet exist on DT) that caused it to hang, i.e. when accessing SATA PHY registers.

Yeah you are right, I had forgotten about this test. I just had a look at the board-km_kirkwood.c and I tried to figure out a way to make sure the clock gates bits never get read/written, so never get initialized. But as you say, that's not the real cause of the hang but

Quoted 3 lines

Valentin, I am not against an extra clock-gating-ctrl for Prestera but maybe having an kirkwood.dtsi fork without SATA and other peripherals will also help?

Quoted 7 lines

Since it's not the real problem having the extra clock-gating-ctrl would me more consistent with the real Prestera HW, but maybe it's better to keep only one that works for both as it's already the case, it's less code to maintain.

A better approach would be to work on the kirkwood.dtsi as you mention it. The SATA node in kirkwood.dtsi is by default disabled, is this enough or should it be moved to kirkwood-6281.dtsi for instance ?

Valentin,

I guess for the long run, we will re-introduce phy gates either by abusing clk gates or some other way. But they will depend on existing OF nodes.

Considering this, it would be best to fork kirkwood.dtsi for Prestera and leave out all OF nodes that are not implemented there. We could have

kirkwood.dtsi -+-> kirkwood-kirkwood.dtsi -+-> kirkwood-6281.dtsi

                +-> kirkwood-prestera.dtsi  +-> kirkwood-6282.dtsi

or any other naming scheme. Maybe Andrew or Jason can comment on this.

Quoted 6 lines
Quoted 2 lines

And: Does DT-enabled km_kirkwood ever hang without the hack in it's board setup? There should be no PHY gates on DT boards..

I have just tested it and it does not hang, so I will submit and patch that simply removes this workaround since the PHY gates are not accessed anymore (thanks to DT clock gates and that km_kirkwood does not enable the sata of node).

Great! I suggest to also submit the dtsi changes above within the same patch set.

Sebastian

#7 lore raw

PATCH 1/3 clk: mvebu: add gate ctrl for Prestera kirkwood variants

Quoted 31 lines · Sebastian Hesselbarth
Quoted 16 lines · Valentin Longchamp
Quoted 5 lines · Sebastian Hesselbarth

I just checked my mails from late 2012 and there Valentin and I agreed, that not the missing bits in clock gating control registers was the root cause of km_kirkwood to hang. It was rather the phy gates (that do not/ not yet exist on DT) that caused it to hang, i.e. when accessing SATA PHY registers.

Yeah you are right, I had forgotten about this test. I just had a look at the board-km_kirkwood.c and I tried to figure out a way to make sure the clock gates bits never get read/written, so never get initialized. But as you say, that's not the real cause of the hang but

Quoted 3 lines

Valentin, I am not against an extra clock-gating-ctrl for Prestera but maybe having an kirkwood.dtsi fork without SATA and other peripherals will also help?

Quoted 7 lines

Since it's not the real problem having the extra clock-gating-ctrl would me more consistent with the real Prestera HW, but maybe it's better to keep only one that works for both as it's already the case, it's less code to maintain.

A better approach would be to work on the kirkwood.dtsi as you mention it. The SATA node in kirkwood.dtsi is by default disabled, is this enough or should it be moved to kirkwood-6281.dtsi for instance ?

Valentin,

I guess for the long run, we will re-introduce phy gates either by abusing clk gates or some other way. But they will depend on existing OF nodes.

Quoted 6 lines

Considering this, it would be best to fork kirkwood.dtsi for Prestera and leave out all OF nodes that are not implemented there. We could have

kirkwood.dtsi -+-> kirkwood-kirkwood.dtsi -+-> kirkwood-6281.dtsi

               +-> kirkwood-prestera.dtsi  +-> kirkwood-6282.dtsi

or,

kirkwood.dtsi
prestera.dtsi

-pinctrl-
kirkwood-98dx4122.dtsi (should rename to prestera-, but seems churn-ish) kirkwood-6281.dtsi
kirkwood-6282.dtsi

-dts-
kirkwood-km_kirkwood.dts (rename/churn prestera-?)
|-prestera.dtsi
\-kirkwood-98dx4122.dtsi

or, honestly, how many boards are going to be based on prestera? If it's just the one, why not combine all the nodes into kirkwood-km_kirkwood.dts?

thx,

Jason.

#8 lore raw

PATCH 2/3 ARM: kirkwood: look for alternative gate ctrl of node

In reply to Valentin Longchamp

The search for the gate ctrl of node must now also look for the introduced compatible node.

---
 arch/arm/mach-kirkwood/board-dt.c | 4 ++++
 1 file changed, 4 insertions(+)
arch/arm/mach-kirkwood/board-dt.c +4 -0
index 7904758..3e607c9 100644
@@ -44,6 +44,10 @@ static void __init kirkwood_legacy_clk_init(void)
 	struct of_phandle_args clkspec;
 	struct clk *clk;
 
+	if (np == NULL)
+		np = of_find_compatible_node(NULL, NULL,
+			"marvell,prestera-kw-gating-clock");
+
 	clkspec.np = np;
 	clkspec.args_count = 1;
 
Signature · 1 line
1.8.0.1
#9 lore raw

PATCH 3/3 ARM: kirkwood: remove clock gating disabling for km_kirkwood

In reply to Valentin Longchamp

A new clock gating controller is available for the prestera-kw kirkwood variant. This controller does not enable the unavailable clk gates.

This patch removes the fixes from the board-km_kirkwood file and enables the correct clk gating controller for kirkwood-98dx4122.dtsi.

---
 arch/arm/boot/dts/kirkwood-98dx4122.dtsi   |  7 +++++++
 arch/arm/mach-kirkwood/board-km_kirkwood.c | 15 ---------------
 2 files changed, 7 insertions(+), 15 deletions(-)
arch/arm/boot/dts/kirkwood-98dx4122.dtsi +7 -0
index 3271e4c..8c73673 100644
@@ -27,5 +27,12 @@
 				marvell,function = "uart1";
 			};
 		};
+
+		gate_clk: clock-gating-control at 2011c {
+			compatible = "marvell,prestera-kw-gating-clock";
+			reg = <0x2011c 0x4>;
+			clocks = <&core_clk 0>;
+			#clock-cells = <1>;
+		};
 	};
 };
arch/arm/mach-kirkwood/board-km_kirkwood.c +0 -15
index 44e4605..f1a6ab7 100644
@@ -25,20 +25,5 @@ static struct mv643xx_eth_platform_data km_kirkwood_ge00_data = {
 
 void __init km_kirkwood_init(void)
 {
-	struct clk *sata_clk;
-	/*
-	 * Our variant of kirkwood (integrated in the Bobcat) hangs on accessing
-	 * SATA bits (14-15) of the Clock Gating Control Register. Since these
-	 * devices are also not present in this variant, their clocks get
-	 * disabled because unused when clk_disable_unused() gets called.
-	 * That's why we change the flags to these clocks to CLK_IGNORE_UNUSED
-	 */
-	sata_clk = clk_get_sys("sata_mv.0", "0");
-	if (!IS_ERR(sata_clk))
-		sata_clk->flags |= CLK_IGNORE_UNUSED;
-	sata_clk = clk_get_sys("sata_mv.0", "1");
-	if (!IS_ERR(sata_clk))
-		sata_clk->flags |= CLK_IGNORE_UNUSED;
-
 	kirkwood_ge00_init(&km_kirkwood_ge00_data);
 }
Signature · 1 line
1.8.0.1
#10 lore raw

PATCH 3/3 ARM: kirkwood: remove clock gating disabling for km_kirkwood

Valentin,

Nice series! I thought once the mv643xx_eth DT bindings take hold that this was going to be one of the holdovers with a custom board init file. Looks like that won't be the case ;-)

Quoted 19 lines · Valentin Longchamp

A new clock gating controller is available for the prestera-kw kirkwood variant. This controller does not enable the unavailable clk gates.

This patch removes the fixes from the board-km_kirkwood file and enables the correct clk gating controller for kirkwood-98dx4122.dtsi.

---
 arch/arm/boot/dts/kirkwood-98dx4122.dtsi   |  7 +++++++
 arch/arm/mach-kirkwood/board-km_kirkwood.c | 15 ---------------
 2 files changed, 7 insertions(+), 15 deletions(-)
arch/arm/boot/dts/kirkwood-98dx4122.dtsi +0 -0
index 3271e4c..8c73673 100644
@@ -27,5 +27,12 @@
 				marvell,function = "uart1";
 			};

arch/arm/boot/dts/kirkwood-98dx4122.dtsi line 38

Quoted 10 lines
arch/arm/boot/dts/kirkwood-98dx4122.dtsi +7 -0
 		};
+
+		gate_clk: clock-gating-control at 2011c {
+			compatible = "marvell,prestera-kw-gating-clock";
+			reg = <0x2011c 0x4>;
+			clocks = <&core_clk 0>;
+			#clock-cells = <1>;
+		};
 	};
 };

--->8----

Please split this into two patches, one for the dtsi, and one for code removal. It'll prevent merge conflicts and branch dependencies for us.

thx,

Jason.

Quoted 28 lines
arch/arm/mach-kirkwood/board-km_kirkwood.c +0 -15
index 44e4605..f1a6ab7 100644
@@ -25,20 +25,5 @@ static struct mv643xx_eth_platform_data km_kirkwood_ge00_data = {
 
 void __init km_kirkwood_init(void)
 {
-	struct clk *sata_clk;
-	/*
-	 * Our variant of kirkwood (integrated in the Bobcat) hangs on accessing
-	 * SATA bits (14-15) of the Clock Gating Control Register. Since these
-	 * devices are also not present in this variant, their clocks get
-	 * disabled because unused when clk_disable_unused() gets called.
-	 * That's why we change the flags to these clocks to CLK_IGNORE_UNUSED
-	 */
-	sata_clk = clk_get_sys("sata_mv.0", "0");
-	if (!IS_ERR(sata_clk))
-		sata_clk->flags |= CLK_IGNORE_UNUSED;
-	sata_clk = clk_get_sys("sata_mv.0", "1");
-	if (!IS_ERR(sata_clk))
-		sata_clk->flags |= CLK_IGNORE_UNUSED;
-
 	kirkwood_ge00_init(&km_kirkwood_ge00_data);
 }
Signature · 1 line
1.8.0.1