On Fri, Nov 28, 2014 at 10:33 AM, Arnd Bergmann [off-list ref] wrote:

On Friday 28 November 2014 14:29:47 Zhou Wang wrote:

quoted

Set ARCH_NR_GPIO for Hisilicon Soc Hip04, which has 4 GPIO
controllers with 32 GPIOs each.

Signed-off-by: Zhou Wang <redacted>
---
 arch/arm/Kconfig                |    1 +
 arch/arm/configs/hisi_defconfig |    3 +++
 2 files changed, 4 insertions(+)

diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index 89c4b5c..26aae1e 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig

@@ -1509,6 +1509,7 @@ config ARCH_NR_GPIO
        default 352 if ARCH_VT8500
        default 288 if ARCH_ROCKCHIP
        default 264 if MACH_H4700
+       default 128 if ARCH_HIP04
        default 0
        help
          Maximum number of GPIOs in the system.

If I remember correctly, you don't actually need to set this if all gpio
clients are using the new gpio descriptor interfaces instead of gpio
numbers.

Unfortunately you still have to. We are working on removing the
dependency on ARCH_NR_GPIO, old habits die hard. But I just
merged this patch:
http://marc.info/?l=linux-gpio&m=141638350328535&w=2

Which makes the situation better.

There is however some other use of this define, so there is
some work required still to get rid of it.

And that patch may blow up in testing too...

Linus, one question for you: Does the ARCH_NR_GPIO actually work as designed?
I see that e.g. OMAP4 defines 192 GPIO lines but does not have an ARCH_NR_GPIO
setting. In an OMAP4-only kernel you would get the default value (512)
from include/asm-generic/gpio.h, and in a multiplatform kernel you get at
least 512 if any of the (ARCH_EXYNOS || ARCH_KEYSTONE || SOC_OMAP5 || \
SOC_DRA7XX || ARCH_S3C24XX || ARCH_S3C64XX || ARCH_S5PV210 || ARCH_SHMOBILE
|| ARCH_TEGRA) are set, so that is still good.

Yeah that is how it kind of works...

However if one builds a kernel that just enables OMAP4 and HIP04, I suspect
it can't work on OMAP4 for any gpio line above 128, which seems to be
a fundamental multiplatform problem.

Yes I guess you are right :(

It's probably just so that so many platforms converge on 512.

Do we neet to increase the default to 512 for all ARCH_MULTIPLATFORM
configurations and just leave ARCH_SHMOBILE, ARCH_TEGRA and MACH_H4700
here as special cases?

That'd be good while we are working to kill off
ARCH_NR_GPIO for good.

I guess there could be arch-specific problems with trying
to get rid of ARCH_NR_GPIO for good, do you have some
input on this? (Arch maintainer hat on...)

Yours,
Linus Walleij