Skip to contentSkip to conversation

PCI: armada8k: Fix clock resource for Armada 7K/8K

PATCH Superseded Gregory CLEMENT 3 participants 7 messages Feb 28, 2018

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover PCI: armada8k: Fix clock resource for Armada 7K/8K

Hi,

This short series fixes the way the clocks are used for the PCIe host controller embedded in the Marvell Armada 7K/8K SoCs. On these SoCs a second one is needed in order to clock the registers. It was not noticed until now because we relied on the bootloader and also because the clock driver was wrong.

Thanks to this fix, it would be possible to fix the clock driver without introducing a regression.

The first patch is just a small cleanup found when I wrote the main patch.

Gregory CLEMENT (2):
PCI: armada8k: Remove useless test before clk_disable_unprepare
PCI: armada8k: Fix clock resource by adding a register clock

 Documentation/devicetree/bindings/pci/pci-armada8k.txt |  6 +++++-
 drivers/pci/dwc/pcie-armada8k.c                        | 14 ++++++++++++--
 2 files changed, 17 insertions(+), 3 deletions(-)
Signature · 1 line
2.16.1
#1 lore raw

PATCH 1/2 PCI: armada8k: Remove useless test before clk_disable_unprepare

clk_disable_unprepare() already checks that the clock pointer is valid. No need to test it before calling it.

---
 drivers/pci/dwc/pcie-armada8k.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/pci/dwc/pcie-armada8k.c +1 -2
index b587352f8b9f..f9b1aec25c5c 100644
@@ -247,8 +247,7 @@ static int armada8k_pcie_probe(struct platform_device *pdev)
 	return 0;
 
 fail:
-	if (!IS_ERR(pcie->clk))
-		clk_disable_unprepare(pcie->clk);
+	clk_disable_unprepare(pcie->clk);
 
 	return ret;
 }
Signature · 1 line
2.16.1
#2 lore raw

PATCH 2/2 PCI: armada8k: Fix clock resource by adding a register clock

In reply to Gregory CLEMENT

On Armada 7K/8K we need to explicitly enable the register clock. This clock is optional because not all the SoCs using this IP need it but at least for Armada 7K/8K it is actually mandatory.

The binding documentation is updated accordingly.

---
 Documentation/devicetree/bindings/pci/pci-armada8k.txt |  6 +++++-
 drivers/pci/dwc/pcie-armada8k.c                        | 11 +++++++++++
 2 files changed, 16 insertions(+), 1 deletion(-)
Documentation/devicetree/bindings/pci/pci-armada8k.txt +5 -1
index c1e4c3d10a74..9948b1e9a8e5 100644
@@ -12,7 +12,11 @@ Required properties:
    - "ctrl" for the control register region
    - "config" for the config space region
 - interrupts: Interrupt specifier for the PCIe controler
-- clocks: reference to the PCIe controller clock
+- clocks: reference to the PCIe controller clocks
+- clock-names: mandatory if there is a second clock, in this case the
+   name must be "core" for the first clock and "reg" for the second
+   one
+
 
 Example:
 
drivers/pci/dwc/pcie-armada8k.c +11 -0
index f9b1aec25c5c..aa4e5cc4ab7b 100644
@@ -28,6 +28,7 @@
 struct armada8k_pcie {
 	struct dw_pcie *pci;
 	struct clk *clk;
+	struct clk *clk_reg;
 };
 
 #define PCIE_VENDOR_REGS_OFFSET		0x8000
@@ -229,6 +230,15 @@ static int armada8k_pcie_probe(struct platform_device *pdev)
 	if (ret)
 		return ret;
 
+	if (IS_ERR(pcie->clk_reg) && PTR_ERR(pcie->clk_reg) == -EPROBE_DEFER) {
+		clk_disable_unprepare(pcie->clk);
+		return -EPROBE_DEFER;
+	}
+	if (!IS_ERR(pcie->clk_reg)) {
+		ret = clk_prepare_enable(pcie->clk_reg);
+		if (ret)
+			goto fail;
+	}
 	/* Get the dw-pcie unit configuration/control registers base. */
 	base = platform_get_resource_byname(pdev, IORESOURCE_MEM, "ctrl");
 	pci->dbi_base = devm_pci_remap_cfg_resource(dev, base);
@@ -247,6 +257,7 @@ static int armada8k_pcie_probe(struct platform_device *pdev)
 	return 0;
 
 fail:
+	clk_disable_unprepare(pcie->clk_reg);
 	clk_disable_unprepare(pcie->clk);
 
 	return ret;
Signature · 1 line
2.16.1
#3 lore raw

PATCH 2/2 PCI: armada8k: Fix clock resource by adding a register clock

Hello,

Quoted 17 lines · Gregory CLEMENT

On Armada 7K/8K we need to explicitly enable the register clock. This clock is optional because not all the SoCs using this IP need it but at least for Armada 7K/8K it is actually mandatory.

The binding documentation is updated accordingly.

---
 Documentation/devicetree/bindings/pci/pci-armada8k.txt |  6 +++++-
 drivers/pci/dwc/pcie-armada8k.c                        | 11 +++++++++++
 2 files changed, 16 insertions(+), 1 deletion(-)
Documentation/devicetree/bindings/pci/pci-armada8k.txt +0 -0
index c1e4c3d10a74..9948b1e9a8e5 100644

Documentation/devicetree/bindings/pci/pci-armada8k.txt Required properties: · line 19

Quoted 9 lines
Documentation/devicetree/bindings/pci/pci-armada8k.txt +5 -1
@@ -12,7 +12,11 @@ Required properties:
    - "ctrl" for the control register region
    - "config" for the config space region
 - interrupts: Interrupt specifier for the PCIe controler
-- clocks: reference to the PCIe controller clock
+- clocks: reference to the PCIe controller clocks
+- clock-names: mandatory if there is a second clock, in this case the
+   name must be "core" for the first clock and "reg" for the second
+   one
+

Unneeded new line added here.

Quoted 19 lines
Documentation/devicetree/bindings/pci/pci-armada8k.txt +0 -0
 Example:
 
drivers/pci/dwc/pcie-armada8k.c +1 -0
index f9b1aec25c5c..aa4e5cc4ab7b 100644
@@ -28,6 +28,7 @@
 struct armada8k_pcie {
 	struct dw_pcie *pci;
 	struct clk *clk;
+	struct clk *clk_reg;
 };
 
 #define PCIE_VENDOR_REGS_OFFSET		0x8000
@@ -229,6 +230,15 @@ static int armada8k_pcie_probe(struct platform_device *pdev)
 	if (ret)
 		return ret;
 

drivers/pci/dwc/pcie-armada8k.c static int armada8k_pcie_probe(struct platform_device *pdev) · line 242

Quoted 10 lines
drivers/pci/dwc/pcie-armada8k.c +9 -0
+	if (IS_ERR(pcie->clk_reg) && PTR_ERR(pcie->clk_reg) == -EPROBE_DEFER) {
+		clk_disable_unprepare(pcie->clk);
+		return -EPROBE_DEFER;
+	}
+	if (!IS_ERR(pcie->clk_reg)) {
+		ret = clk_prepare_enable(pcie->clk_reg);
+		if (ret)
+			goto fail;
+	}
 	/* Get the dw-pcie unit configuration/control registers base. */

Missing new line between the end of the block and the next comment.

Regarding the error handling, doesn't it make more sense to also use a goto label to disable pcie->clk when getting the second clock gets a -EPROBE_DEFER ?

drivers/pci/dwc/pcie-armada8k.c static int armada8k_pcie_probe(struct platform_device *pdev) · line 260

Quoted 7 lines
drivers/pci/dwc/pcie-armada8k.c +1 -0
 	base = platform_get_resource_byname(pdev, IORESOURCE_MEM, "ctrl");
 	pci->dbi_base = devm_pci_remap_cfg_resource(dev, base);
@@ -247,6 +257,7 @@ static int armada8k_pcie_probe(struct platform_device *pdev)
 	return 0;
 
 fail:
+	clk_disable_unprepare(pcie->clk_reg);

So you are disabling/unpreparing the clock, which failed to prepare/enable ?

drivers/pci/dwc/pcie-armada8k.c static int armada8k_pcie_probe(struct platform_device *pdev) · line 263

Quoted 3 lines
drivers/pci/dwc/pcie-armada8k.c +0 -0
 	clk_disable_unprepare(pcie->clk);
 
 	return ret;

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin (formerly Free Electrons)
Embedded Linux and Kernel engineering
http://bootlin.com
#4 lore raw

PATCH 2/2 PCI: armada8k: Fix clock resource by adding a register clock

Hi Thomas,

Quoted 15 lines · Thomas Petazzoni

Hello,

Quoted 12 lines · Gregory CLEMENT

On Armada 7K/8K we need to explicitly enable the register clock. This clock is optional because not all the SoCs using this IP need it but at least for Armada 7K/8K it is actually mandatory.

The binding documentation is updated accordingly.

---
 Documentation/devicetree/bindings/pci/pci-armada8k.txt |  6 +++++-
 drivers/pci/dwc/pcie-armada8k.c                        | 11 +++++++++++
 2 files changed, 16 insertions(+), 1 deletion(-)
Quoted 16 lines
Quoted 5 lines
Documentation/devicetree/bindings/pci/pci-armada8k.txt +0 -0
index c1e4c3d10a74..9948b1e9a8e5 100644

Documentation/devicetree/bindings/pci/pci-armada8k.txt Required properties: · line 19

Quoted 9 lines
Documentation/devicetree/bindings/pci/pci-armada8k.txt +5 -1
@@ -12,7 +12,11 @@ Required properties:
    - "ctrl" for the control register region
    - "config" for the config space region
 - interrupts: Interrupt specifier for the PCIe controler
-- clocks: reference to the PCIe controller clock
+- clocks: reference to the PCIe controller clocks
+- clock-names: mandatory if there is a second clock, in this case the
+   name must be "core" for the first clock and "reg" for the second
+   one
+

Unneeded new line added here.

will removed it

Quoted 20 lines
Quoted 20 lines
Documentation/devicetree/bindings/pci/pci-armada8k.txt +0 -0
 Example:
 
drivers/pci/dwc/pcie-armada8k.c +1 -0
index f9b1aec25c5c..aa4e5cc4ab7b 100644
@@ -28,6 +28,7 @@
 struct armada8k_pcie {
 	struct dw_pcie *pci;
 	struct clk *clk;
+	struct clk *clk_reg;
 };
 
 #define PCIE_VENDOR_REGS_OFFSET		0x8000
@@ -229,6 +230,15 @@ static int armada8k_pcie_probe(struct platform_device *pdev)
 	if (ret)
 		return ret;
 
Quoted 13 lines

drivers/pci/dwc/pcie-armada8k.c static int armada8k_pcie_probe(struct platform_device *pdev) · line 242

Quoted 10 lines
drivers/pci/dwc/pcie-armada8k.c +9 -0
+	if (IS_ERR(pcie->clk_reg) && PTR_ERR(pcie->clk_reg) == -EPROBE_DEFER) {
+		clk_disable_unprepare(pcie->clk);
+		return -EPROBE_DEFER;
+	}
+	if (!IS_ERR(pcie->clk_reg)) {
+		ret = clk_prepare_enable(pcie->clk_reg);
+		if (ret)
+			goto fail;
+	}
 	/* Get the dw-pcie unit configuration/control registers base. */

Missing new line between the end of the block and the next comment.

OK

Quoted 9 lines

Regarding the error handling, doesn't it make more sense to also use a goto label to disable pcie->clk when getting the second clock gets a -EPROBE_DEFER ?

Quoted 5 lines
drivers/pci/dwc/pcie-armada8k.c +0 -0
 	base = platform_get_resource_byname(pdev, IORESOURCE_MEM, "ctrl");
 	pci->dbi_base = devm_pci_remap_cfg_resource(dev, base);
@@ -247,6 +257,7 @@ static int armada8k_pcie_probe(struct platform_device *pdev)
 	return 0;
 
Quoted 5 lines

drivers/pci/dwc/pcie-armada8k.c static int armada8k_pcie_probe(struct platform_device *pdev) · line 260

Quoted 2 lines
drivers/pci/dwc/pcie-armada8k.c +1 -0
 fail:
+	clk_disable_unprepare(pcie->clk_reg);

So you are disabling/unpreparing the clock, which failed to prepare/enable ?

I was thinking to a single user, in this case if the prepare/enable failed then the disabling/unpreparing do nothing because the counter is already to 0. But indeed in case of multiple users of the clock, then the counter could be wrongly decrease. I will modify it by adding a other label.

Thanks,

Gregory

Quoted 10 lines

drivers/pci/dwc/pcie-armada8k.c static int armada8k_pcie_probe(struct platform_device *pdev) · line 263

Quoted 4 lines
drivers/pci/dwc/pcie-armada8k.c +0 -0
 	clk_disable_unprepare(pcie->clk);
 
 	return ret;

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin (formerly Free Electrons)
Embedded Linux and Kernel engineering
http://bootlin.com
Signature · 3 lines
Gregory Clement, Bootlin (formerly Free Electrons)
Embedded Linux and Kernel engineering
http://bootlin.com
#5 lore raw

PATCH 2/2 PCI: armada8k: Fix clock resource by adding a register clock

In reply to Gregory CLEMENT

drivers/pci/dwc/pcie-armada8k.c static int armada8k_pcie_probe(struct platform_device *pdev) · line 233

Quoted 5 lines · Gregory CLEMENT
drivers/pci/dwc/pcie-armada8k.c +1 -0
@@ -229,6 +230,15 @@ static int armada8k_pcie_probe(struct platform_device *pdev)
 	if (ret)
 		return ret;
 
+	if (IS_ERR(pcie->clk_reg) && PTR_ERR(pcie->clk_reg) == -EPROBE_DEFER) {

You do realise this is needlessly complex.

Pointer errors are unique, so:

	if (pcie->clk_reg == ERR_PTR(-EPROBE_DEFER)) {

will do the same thing but without the complexity. Transforming the constant rather than the variable is also a good habbit to get into - the compiler can optimise transforms to constants, but can't with variables, so comparisons involving things like endian conversion should always be done by transforming the constant not the variable.

Signature · 3 lines
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 8.8Mbps down 630kbps up
According to speedtest.net: 8.21Mbps down 510kbps up
#6 lore raw

PATCH 2/2 PCI: armada8k: Fix clock resource by adding a register clock

Hi Russell King,

Quoted 11 lines · Russell King - ARM Linux

drivers/pci/dwc/pcie-armada8k.c static int armada8k_pcie_probe(struct platform_device *pdev) · line 233

Quoted 5 lines · Gregory CLEMENT
drivers/pci/dwc/pcie-armada8k.c +1 -0
@@ -229,6 +230,15 @@ static int armada8k_pcie_probe(struct platform_device *pdev)
 	if (ret)
 		return ret;
 
+	if (IS_ERR(pcie->clk_reg) && PTR_ERR(pcie->clk_reg) == -EPROBE_DEFER) {

You do realise this is needlessly complex.

Pointer errors are unique, so:

Quoted 7 lines
	if (pcie->clk_reg == ERR_PTR(-EPROBE_DEFER)) {

will do the same thing but without the complexity. Transforming the constant rather than the variable is also a good habbit to get into - the compiler can optimise transforms to constants, but can't with variables, so comparisons involving things like endian conversion should always be done by transforming the constant not the variable.

Thanks for the tip, I will use it in the next version.

Gregory

Quoted 5 lines
Signature · 4 lines
-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 8.8Mbps down 630kbps up
According to speedtest.net: 8.21Mbps down 510kbps up
Signature · 3 lines
Gregory Clement, Bootlin (formerly Free Electrons)
Embedded Linux and Kernel engineering
http://bootlin.com