Enable mvebu pinctrl/gpio for Dove and Kirkwood

We now have a pinctrl driver for arch-mvebu. It supports kirkwood out of the box. Allow ARCH_KIRKWOOD to select it, and make ARCH_KIRKWOOD select it. This then allows the convertion of board-*.c files away from MPP to pinctrl in DT, before they get moved over into mach-mvebu.

Signed-off-by Andrew Lunn <[email protected]>

---
 arch/arm/Kconfig        |    2 ++
 drivers/pinctrl/Kconfig |    2 +-
 2 files changed, 3 insertions(+), 1 deletion(-)

arch/arm/Kconfig +0 -0

index 73067ef..160cd40 100644

Since 3.7 readl/writel require register addresses to be const void* instead of unsigned int. The register addresses are converted using IOMEM() and offsets are added instead of OR'ed.
Also a workaround for the pdma clock is added, that is required as there is still no DT clock provider available on Dove.

Quoted 5 lines · Andrew Lunn

Since 3.7 readl/writel require register addresses to be const void* instead of unsigned int. The register addresses are converted using IOMEM() and offsets are added instead of OR'ed.
Also a workaround for the pdma clock is added, that is required as there is still no DT clock provider available on Dove.

So essentially two unrelated patches squashed into one, and I would apply the first one right off but now the clock change makes me hesitate.

drivers/pinctrl/pinctrl-dove.c static int __devinit dove_pinctrl_probe(struct platform_device *pdev) · line 592

Quoted 4 lines

         clk = devm_clk_get(&pdev->dev, NULL);

drivers/pinctrl/pinctrl-dove.c +3 -0

+
+       /* Currently there is no DT clock provider for pdma clock,
+          this fallback ensures pdma clock is ticking */

/*
* I prefer comment style like so because it's easier to read.
* Maybe it's a bit pedantic but bear with me.
*/

drivers/pinctrl/pinctrl-dove.c static int __devinit dove_pinctrl_probe(struct platform_device *pdev) · line 595

Quoted 3 lines

drivers/pinctrl/pinctrl-dove.c +3 -0

+       if (IS_ERR(clk))
+               clk = clk_get_sys("dove-pdma", NULL);
+

This is a horrible hack. But if the Marvell people agree about it I will live with it.

Quoted 22 lines · Linus Walleij

Quoted 5 lines · Andrew Lunn

Since 3.7 readl/writel require register addresses to be const void* instead of unsigned int. The register addresses are converted using IOMEM() and offsets are added instead of OR'ed.
Also a workaround for the pdma clock is added, that is required as there is still no DT clock provider available on Dove.

So essentially two unrelated patches squashed into one, and I would apply the first one right off but now the clock change makes me hesitate.

drivers/pinctrl/pinctrl-dove.c static int __devinit dove_pinctrl_probe(struct platform_device *pdev) · line 592

Quoted 4 lines

        clk = devm_clk_get(&pdev->dev, NULL);

drivers/pinctrl/pinctrl-dove.c +3 -0

+
+       /* Currently there is no DT clock provider for pdma clock,
+          this fallback ensures pdma clock is ticking */

/*
* I prefer comment style like so because it's easier to read.
* Maybe it's a bit pedantic but bear with me.
*/

Quoted 6 lines

drivers/pinctrl/pinctrl-dove.c static int __devinit dove_pinctrl_probe(struct platform_device *pdev) · line 595

Quoted 3 lines

drivers/pinctrl/pinctrl-dove.c +3 -0

+       if (IS_ERR(clk))
+               clk = clk_get_sys("dove-pdma", NULL);
+

This is a horrible hack. But if the Marvell people agree about it I will live with it.

Unfortunately, it is. This is an chicken-egg-problem here, no DT clk-provider, no clocks properties..

While writing pinctrl-dove I was planing to enable it after the clock provider but with Andrew pushing forward - for a good reason - there comes the trouble ;)

I don't like the hack either but the clk-gate is there and if I don't enable it pinctrl-dove will hang the SoC.

I have a clk-dove DT clk-provider in my pocket somewhere but didn't find the time to prepare it for posting..