Skip to contentSkip to conversation

arm64 defconfig breakage (gpio) in next-20130218

PATCH New Mark Rutland 6 participants 10 messages Feb 18, 2013 – Feb 22, 2013 1 message not in the archive arm64

lore
Share & tools
lore

Mark Rutland

author reviewer
#0 lore raw

Hello,

When I attempt to build a defconfig arm64 kernel from next-20130218, it fails with the following complaints from gcc:

In file included from include/linux/gpio.h:48:0,
from drivers/gpio/devres.c:18:
include/asm-generic/gpio.h: In function ?gpio_get_value_cansleep?:
include/asm-generic/gpio.h:270:2: error: implicit declaration of function ?__gpio_get_value? [-Werror=implicit-function-declaration] include/asm-generic/gpio.h: In function ?gpio_set_value_cansleep?:
include/asm-generic/gpio.h:276:2: error: implicit declaration of function ?__gpio_set_value? [-Werror=implicit-function-declaration] In file included from drivers/gpio/devres.c:18:0:
include/linux/gpio.h: At top level:
include/linux/gpio.h:60:19: error: redefinition of ?gpio_cansleep?
In file included from include/linux/gpio.h:48:0,
from drivers/gpio/devres.c:18:
include/asm-generic/gpio.h:262:19: note: previous definition of ?gpio_cansleep? was here
In file included from drivers/gpio/devres.c:18:0:
include/linux/gpio.h: In function ?gpio_cansleep?:
include/linux/gpio.h:62:2: error: implicit declaration of function ?__gpio_cansleep? [-Werror=implicit-function-declaration] include/linux/gpio.h: In function ?gpio_to_irq?:
include/linux/gpio.h:67:2: error: implicit declaration of function ?__gpio_to_irq? [-Werror=implicit-function-declaration] drivers/gpio/devres.c: In function ?devm_gpio_release?:
drivers/gpio/devres.c:26:2: error: implicit declaration of function ?gpio_free? [-Werror=implicit-function-declaration] drivers/gpio/devres.c: In function ?devm_gpio_request?:
drivers/gpio/devres.c:60:2: error: implicit declaration of function ?gpio_request? [-Werror=implicit-function-declaration] drivers/gpio/devres.c: In function ?devm_gpio_request_one?:
drivers/gpio/devres.c:90:2: error: implicit declaration of function ?gpio_request_one? [-Werror=implicit-function-declaration]
cc1: some warnings being treated as errors
make[2]: *** [drivers/gpio/devres.o] Error 1
make[1]: *** [drivers/gpio] Error 2

I've bisected this down to 0fa2fd9a0d: "Merge branch 'linusw/devel' of
git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-gpio.git into gpio/next". Both parent commits build fine.

I've played around a bit, and selecting ARCH_REQUIRE_GPIOLIB in the arm64 Kconfig fixes the issue, but I'm unfamilar with gpio{,lib} and I'm not sure that's the best/correct way of solving this.

Thoughts?

Thanks,
Mark.

#1 lore raw
Quoted 32 lines · Mark Rutland

Hello,

When I attempt to build a defconfig arm64 kernel from next-20130218, it fails with the following complaints from gcc:

In file included from include/linux/gpio.h:48:0,
from drivers/gpio/devres.c:18:
include/asm-generic/gpio.h: In function ?gpio_get_value_cansleep?:
include/asm-generic/gpio.h:270:2: error: implicit declaration of function ?__gpio_get_value? [-Werror=implicit-function-declaration] include/asm-generic/gpio.h: In function ?gpio_set_value_cansleep?:
include/asm-generic/gpio.h:276:2: error: implicit declaration of function ?__gpio_set_value? [-Werror=implicit-function-declaration] In file included from drivers/gpio/devres.c:18:0:
include/linux/gpio.h: At top level:
include/linux/gpio.h:60:19: error: redefinition of ?gpio_cansleep?
In file included from include/linux/gpio.h:48:0,
from drivers/gpio/devres.c:18:
include/asm-generic/gpio.h:262:19: note: previous definition of ?gpio_cansleep? was here
In file included from drivers/gpio/devres.c:18:0:
include/linux/gpio.h: In function ?gpio_cansleep?:
include/linux/gpio.h:62:2: error: implicit declaration of function ?__gpio_cansleep? [-Werror=implicit-function-declaration] include/linux/gpio.h: In function ?gpio_to_irq?:
include/linux/gpio.h:67:2: error: implicit declaration of function ?__gpio_to_irq? [-Werror=implicit-function-declaration] drivers/gpio/devres.c: In function ?devm_gpio_release?:
drivers/gpio/devres.c:26:2: error: implicit declaration of function ?gpio_free? [-Werror=implicit-function-declaration] drivers/gpio/devres.c: In function ?devm_gpio_request?:
drivers/gpio/devres.c:60:2: error: implicit declaration of function ?gpio_request? [-Werror=implicit-function-declaration] drivers/gpio/devres.c: In function ?devm_gpio_request_one?:
drivers/gpio/devres.c:90:2: error: implicit declaration of function ?gpio_request_one? [-Werror=implicit-function-declaration]
cc1: some warnings being treated as errors
make[2]: *** [drivers/gpio/devres.o] Error 1
make[1]: *** [drivers/gpio] Error 2

Quoted 7 lines

I've bisected this down to 0fa2fd9a0d: "Merge branch 'linusw/devel' of
git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-gpio.git into gpio/next". Both parent commits build fine.

I've played around a bit, and selecting ARCH_REQUIRE_GPIOLIB in the arm64 Kconfig fixes the issue, but I'm unfamilar with gpio{,lib} and I'm not sure that's the best/correct way of solving this.

Kconfig GENERIC_GPIO is designed as an option defined at architecture level and select by platform who knows how gpio driver/API is implemented. The following change should fix the error.

Shawn

arch/arm64/Kconfig +1 -1
index f532ce5..aca4a25 100644
@@ -93,7 +93,7 @@ config IOMMU_HELPER
        def_bool SWIOTLB
 
 config GENERIC_GPIO
-       def_bool y
+       bool
 
 source "init/Kconfig"
#2 lore raw
Quoted 18 lines · On Thursday 21 February 2013 17:31:07 Shawn Guo wrote:
Quoted 8 lines

I've bisected this down to 0fa2fd9a0d: "Merge branch 'linusw/devel' of
git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-gpio.git into gpio/next". Both parent commits build fine.

I've played around a bit, and selecting ARCH_REQUIRE_GPIOLIB in the arm64 Kconfig fixes the issue, but I'm unfamilar with gpio{,lib} and I'm not sure that's the best/correct way of solving this.

Kconfig GENERIC_GPIO is designed as an option defined at architecture level and select by platform who knows how gpio driver/API is implemented. The following change should fix the error.

arch/arm64/Kconfig +0 -0
index f532ce5..aca4a25 100644

arch/arm64/Kconfig config IOMMU_HELPER · line 98

Quoted 7 lines
arch/arm64/Kconfig +1 -1
@@ -93,7 +93,7 @@ config IOMMU_HELPER
        def_bool SWIOTLB
 
 config GENERIC_GPIO
-       def_bool y
+       bool
 
 source "init/Kconfig"

But we have no arm64 platforms that would pick GENERIC_GPIO, it's expected to come from GPIOLIB anyway. I think the right fix is this:

arch/arm64/Kconfig +1 -3
index ab4aa54..3fab0db 100644
@@ -3,6 +3,7 @@ config ARM64
 	select ARCH_HAS_ATOMIC64_DEC_IF_POSITIVE
 	select ARCH_WANT_COMPAT_IPC_PARSE_VERSION
 	select ARCH_WANT_FRAME_POINTERS
+	select ARCH_WANT_OPTIONAL_GPIOLIB
 	select ARM_AMBA
 	select ARM_ARCH_TIMER
 	select CLONE_BACKWARDS
@@ -92,9 +93,6 @@ config SWIOTLB
 config IOMMU_HELPER
 	def_bool SWIOTLB
 
-config GENERIC_GPIO
-	def_bool y
-
 source "init/Kconfig"
 
 source "kernel/Kconfig.freezer"

Arnd

#3 lore raw
Quoted 43 lines · Arnd Bergmann
Quoted 18 lines · On Thursday 21 February 2013 17:31:07 Shawn Guo wrote:
Quoted 8 lines

I've bisected this down to 0fa2fd9a0d: "Merge branch 'linusw/devel' of
git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-gpio.git into gpio/next". Both parent commits build fine.

I've played around a bit, and selecting ARCH_REQUIRE_GPIOLIB in the arm64 Kconfig fixes the issue, but I'm unfamilar with gpio{,lib} and I'm not sure that's the best/correct way of solving this.

Kconfig GENERIC_GPIO is designed as an option defined at architecture level and select by platform who knows how gpio driver/API is implemented. The following change should fix the error.

arch/arm64/Kconfig +0 -0
index f532ce5..aca4a25 100644

arch/arm64/Kconfig config IOMMU_HELPER · line 98

Quoted 7 lines
arch/arm64/Kconfig +1 -1
@@ -93,7 +93,7 @@ config IOMMU_HELPER
        def_bool SWIOTLB
 
 config GENERIC_GPIO
-       def_bool y
+       bool
 
 source "init/Kconfig"

But we have no arm64 platforms that would pick GENERIC_GPIO, it's expected to come from GPIOLIB anyway. I think the right fix is this:

arch/arm64/Kconfig +1 -0
index ab4aa54..3fab0db 100644
@@ -3,6 +3,7 @@ config ARM64
        select ARCH_HAS_ATOMIC64_DEC_IF_POSITIVE
        select ARCH_WANT_COMPAT_IPC_PARSE_VERSION
        select ARCH_WANT_FRAME_POINTERS
+       select ARCH_WANT_OPTIONAL_GPIOLIB
        select ARM_AMBA
        select ARM_ARCH_TIMER
        select CLONE_BACKWARDS

arch/arm64/Kconfig config SWIOTLB · line 98

Quoted 9 lines
arch/arm64/Kconfig +0 -3
@@ -92,9 +93,6 @@ config SWIOTLB
 config IOMMU_HELPER
        def_bool SWIOTLB
 
-config GENERIC_GPIO
-       def_bool y
-
 source "init/Kconfig"
 
 source "kernel/Kconfig.freezer"

I've not asked Linus to pull the GPIO tree yet because of this issue. Can someone please test this on aarch64 and send me a properly formatted patch?

g.

Signature · 2 lines
Grant Likely, B.Sc., P.Eng.
Secret Lab Technologies Ltd.
#4 lore raw

PATCH 1/1 arm64: select ARCH_WANT_OPTIONAL_GPIOLIB

An architecture should not unconditionally enable 'GENERIC_GPIO' without providing an implementation. In case of arm64, selecting ARCH_WANT_OPTIONAL_GPIOLIB is the right solution, because it lets us enable GPIOLIB when configuring the kernel, and that implicitly turns on GENERIC_GPIO.

----

Quoted 21 lines · On Thursday 21 February 2013 10:36:41 Grant Likely wrote:
Quoted 21 lines
arch/arm64/Kconfig +1 -3
index ab4aa54..3fab0db 100644
@@ -3,6 +3,7 @@ config ARM64
        select ARCH_HAS_ATOMIC64_DEC_IF_POSITIVE
        select ARCH_WANT_COMPAT_IPC_PARSE_VERSION
        select ARCH_WANT_FRAME_POINTERS
+       select ARCH_WANT_OPTIONAL_GPIOLIB
        select ARM_AMBA
        select ARM_ARCH_TIMER
        select CLONE_BACKWARDS
@@ -92,9 +93,6 @@ config SWIOTLB
 config IOMMU_HELPER
        def_bool SWIOTLB
 
-config GENERIC_GPIO
-       def_bool y
-
 source "init/Kconfig"
 
Quoted 5 lines

arch/arm64/Kconfig config SWIOTLB · line 98

Quoted 1 line
arch/arm64/Kconfig +0 -0
 source "kernel/Kconfig.freezer"

I've not asked Linus to pull the GPIO tree yet because of this issue. Can someone please test this on aarch64 and send me a properly formatted patch?

It was wrong, the new version below is good though. Tested with defconfig and allyesconfig on arm64. I found a few more trivial issues with arm64 allyesconfig that I fixed up as well. I'll send separate patches for those.

arch/arm64/Kconfig +2 -1
index ab4aa54..38d3f34 100644
@@ -3,6 +3,7 @@ config ARM64
 	select ARCH_HAS_ATOMIC64_DEC_IF_POSITIVE
 	select ARCH_WANT_COMPAT_IPC_PARSE_VERSION
 	select ARCH_WANT_FRAME_POINTERS
+	select ARCH_WANT_OPTIONAL_GPIOLIB
 	select ARM_AMBA
 	select ARM_ARCH_TIMER
 	select CLONE_BACKWARDS
@@ -93,7 +94,7 @@ config IOMMU_HELPER
 	def_bool SWIOTLB
 
 config GENERIC_GPIO
-	def_bool y
+	bool
 
 source "init/Kconfig"
 
#5 lore raw

PATCH 1/1 arm64: select ARCH_WANT_OPTIONAL_GPIOLIB

Quoted 7 lines · Arnd Bergmann

An architecture should not unconditionally enable 'GENERIC_GPIO' without providing an implementation. In case of arm64, selecting ARCH_WANT_OPTIONAL_GPIOLIB is the right solution, because it lets us enable GPIOLIB when configuring the kernel, and that implicitly turns on GENERIC_GPIO.

GENERIC_GPIO must die, it's just causing trouble like this all the time :-(

Yours,
Linus Walleij

#6 lore raw

PATCH 1/1 arm64: select ARCH_WANT_OPTIONAL_GPIOLIB

Quoted 11 lines · Linus Walleij
Quoted 7 lines · Arnd Bergmann

An architecture should not unconditionally enable 'GENERIC_GPIO' without providing an implementation. In case of arm64, selecting ARCH_WANT_OPTIONAL_GPIOLIB is the right solution, because it lets us enable GPIOLIB when configuring the kernel, and that implicitly turns on GENERIC_GPIO.

Okay, I've put this into my gpio/next branch and I'll ask Linus to pull shortly. Thanks everyone.

g.

A message that is not in the archive (sent off-list or not yet received): 1636745.vSza8MrtIn@wuerfel

maintainer Acked-by
#8 lore raw

arm64: select ARCH_WANT_OPTIONAL_GPIOLIB

Quoted 36 lines · Arnd Bergmann

An architecture should not unconditionally enable 'GENERIC_GPIO' without providing an implementation. In case of arm64, selecting ARCH_WANT_OPTIONAL_GPIOLIB is the right solution, because it lets us enable GPIOLIB when configuring the kernel, and that implicitly turns on GENERIC_GPIO.

----

Quoted 21 lines · On Thursday 21 February 2013 10:36:41 Grant Likely wrote:
Quoted 21 lines
arch/arm64/Kconfig +1 -3
index ab4aa54..3fab0db 100644
@@ -3,6 +3,7 @@ config ARM64
        select ARCH_HAS_ATOMIC64_DEC_IF_POSITIVE
        select ARCH_WANT_COMPAT_IPC_PARSE_VERSION
        select ARCH_WANT_FRAME_POINTERS
+       select ARCH_WANT_OPTIONAL_GPIOLIB
        select ARM_AMBA
        select ARM_ARCH_TIMER
        select CLONE_BACKWARDS
@@ -92,9 +93,6 @@ config SWIOTLB
 config IOMMU_HELPER
        def_bool SWIOTLB
 
-config GENERIC_GPIO
-       def_bool y
-
 source "init/Kconfig"
 
Quoted 5 lines

arch/arm64/Kconfig config SWIOTLB · line 98

Quoted 1 line
arch/arm64/Kconfig +0 -0
 source "kernel/Kconfig.freezer"

I've not asked Linus to pull the GPIO tree yet because of this issue. Can someone please test this on aarch64 and send me a properly formatted patch?

Quoted 4 lines

It was wrong, the new version below is good though. Tested with defconfig and allyesconfig on arm64. I found a few more trivial issues with arm64 allyesconfig that I fixed up as well. I'll send separate patches for those.

That's what I just tried, it looks ok to me:

Maybe Mark can try it with -next (I only tested with 3.8).

Thanks.

Signature · 1 line
Catalin
author reviewer
#9 lore raw

arm64: select ARCH_WANT_OPTIONAL_GPIOLIB

Quoted 42 lines · Catalin Marinas
Quoted 36 lines · Arnd Bergmann

An architecture should not unconditionally enable 'GENERIC_GPIO' without providing an implementation. In case of arm64, selecting ARCH_WANT_OPTIONAL_GPIOLIB is the right solution, because it lets us enable GPIOLIB when configuring the kernel, and that implicitly turns on GENERIC_GPIO.

----

Quoted 21 lines · On Thursday 21 February 2013 10:36:41 Grant Likely wrote:
Quoted 21 lines
arch/arm64/Kconfig +1 -3
index ab4aa54..3fab0db 100644
@@ -3,6 +3,7 @@ config ARM64
        select ARCH_HAS_ATOMIC64_DEC_IF_POSITIVE
        select ARCH_WANT_COMPAT_IPC_PARSE_VERSION
        select ARCH_WANT_FRAME_POINTERS
+       select ARCH_WANT_OPTIONAL_GPIOLIB
        select ARM_AMBA
        select ARM_ARCH_TIMER
        select CLONE_BACKWARDS
@@ -92,9 +93,6 @@ config SWIOTLB
 config IOMMU_HELPER
        def_bool SWIOTLB
 
-config GENERIC_GPIO
-       def_bool y
-
 source "init/Kconfig"
 
Quoted 5 lines

arch/arm64/Kconfig config SWIOTLB · line 98

Quoted 1 line
arch/arm64/Kconfig +0 -0
 source "kernel/Kconfig.freezer"

I've not asked Linus to pull the GPIO tree yet because of this issue. Can someone please test this on aarch64 and send me a properly formatted patch?

Quoted 4 lines

It was wrong, the new version below is good though. Tested with defconfig and allyesconfig on arm64. I found a few more trivial issues with arm64 allyesconfig that I fixed up as well. I'll send separate patches for those.

Quoted 5 lines

That's what I just tried, it looks ok to me:

Maybe Mark can try it with -next (I only tested with 3.8).

I've just tried the updated patch on next-20130222; arm64 defconfig builds successfully.

Thanks,
Mark.

maintainer
#10 lore raw
In reply to Arnd Bergmann
Quoted 43 lines · Arnd Bergmann
Quoted 18 lines · On Thursday 21 February 2013 17:31:07 Shawn Guo wrote:
Quoted 8 lines

I've bisected this down to 0fa2fd9a0d: "Merge branch 'linusw/devel' of
git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-gpio.git into gpio/next". Both parent commits build fine.

I've played around a bit, and selecting ARCH_REQUIRE_GPIOLIB in the arm64 Kconfig fixes the issue, but I'm unfamilar with gpio{,lib} and I'm not sure that's the best/correct way of solving this.

Kconfig GENERIC_GPIO is designed as an option defined at architecture level and select by platform who knows how gpio driver/API is implemented. The following change should fix the error.

arch/arm64/Kconfig +0 -0
index f532ce5..aca4a25 100644

arch/arm64/Kconfig config IOMMU_HELPER · line 98

Quoted 7 lines
arch/arm64/Kconfig +1 -1
@@ -93,7 +93,7 @@ config IOMMU_HELPER
        def_bool SWIOTLB
 
 config GENERIC_GPIO
-       def_bool y
+       bool
 
 source "init/Kconfig"

But we have no arm64 platforms that would pick GENERIC_GPIO, it's expected to come from GPIOLIB anyway. I think the right fix is this:

arch/arm64/Kconfig +1 -0
index ab4aa54..3fab0db 100644
@@ -3,6 +3,7 @@ config ARM64
 	select ARCH_HAS_ATOMIC64_DEC_IF_POSITIVE
 	select ARCH_WANT_COMPAT_IPC_PARSE_VERSION
 	select ARCH_WANT_FRAME_POINTERS
+	select ARCH_WANT_OPTIONAL_GPIOLIB
 	select ARM_AMBA
 	select ARM_ARCH_TIMER
 	select CLONE_BACKWARDS

arch/arm64/Kconfig config SWIOTLB · line 97

Quoted 6 lines
arch/arm64/Kconfig +0 -3
@@ -92,9 +93,6 @@ config SWIOTLB
 config IOMMU_HELPER
 	def_bool SWIOTLB
 
-config GENERIC_GPIO
-	def_bool y
-

In my soc-armv8-model branch, the ARCH_VEXPRESS selects ARCH_REQUIRE_GPIOLIB which in turn selects GENERIC_GPIO but this seems to be defined in the arch Kconfig.

Signature · 1 line
Catalin