Skip to contentSkip to conversation

drm/bridge: fix RC_CORE dependency

PATCH In review Arnd Bergmann 3 participants 11 messages Jul 18, 2019 – Jul 19, 2019

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/bridge: fix RC_CORE dependency

Using 'imply' causes a new problem, as it allows the case of CONFIG_INPUT=m with RC_CORE=y, which fails to link:

drivers/media/rc/rc-main.o: In function `ir_do_keyup':
rc-main.c:(.text+0x2b4): undefined reference to input_event' drivers/media/rc/rc-main.o: In function rc_repeat':
rc-main.c:(.text+0x350): undefined reference to input_event' drivers/media/rc/rc-main.o: In function rc_allocate_device':
rc-main.c:(.text+0x90c): undefined reference to `input_allocate_device'

Add a 'depends on' that allows building both with and without CONFIG_RC_CORE, but disallows combinations that don't link.

Fixes 5023cf32210d ("drm/bridge: make remote control optional")
---
 drivers/gpu/drm/bridge/Kconfig | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/bridge/Kconfig +1 -1
index f64c91defdc3..70a8ed2505aa 100644
@@ -85,8 +85,8 @@ config DRM_SIL_SII8620
 	tristate "Silicon Image SII8620 HDMI/MHL bridge"
 	depends on OF
 	select DRM_KMS_HELPER
+	depends on RC_CORE || !RC_CORE
 	imply EXTCON
-	imply RC_CORE
 	help
 	  Silicon Image SII8620 HDMI/MHL bridge chip driver.
 
Signature · 1 line
2.20.0
#1 lore raw

Hi Arnd,

Quoted 10 lines · Arnd Bergmann

Using 'imply' causes a new problem, as it allows the case of CONFIG_INPUT=m with RC_CORE=y, which fails to link:

drivers/media/rc/rc-main.o: In function `ir_do_keyup':
rc-main.c:(.text+0x2b4): undefined reference to input_event' drivers/media/rc/rc-main.o: In function rc_repeat':
rc-main.c:(.text+0x350): undefined reference to input_event' drivers/media/rc/rc-main.o: In function rc_allocate_device':
rc-main.c:(.text+0x90c): undefined reference to `input_allocate_device'

Quoted 5 lines

Add a 'depends on' that allows building both with and without CONFIG_RC_CORE, but disallows combinations that don't link.

Fixes 5023cf32210d ("drm/bridge: make remote control optional")

Proper solution has been already merged via input tree[1].

[1]:
https://lore.kernel.org/lkml/CAKdAkRTGXNbUsuKASNGLfwUwC7Asod9K5baYLPWPU7EX-42-yA@mail.gmail.com/

Regards

Andrzej

Quoted 18 lines
---
 drivers/gpu/drm/bridge/Kconfig | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/bridge/Kconfig +1 -1
index f64c91defdc3..70a8ed2505aa 100644
@@ -85,8 +85,8 @@ config DRM_SIL_SII8620
 	tristate "Silicon Image SII8620 HDMI/MHL bridge"
 	depends on OF
 	select DRM_KMS_HELPER
+	depends on RC_CORE || !RC_CORE
 	imply EXTCON
-	imply RC_CORE
 	help
 	  Silicon Image SII8620 HDMI/MHL bridge chip driver.
 
#2 lore raw
Quoted 21 lines · Andrzej Hajda

Hi Arnd,

Quoted 10 lines · Arnd Bergmann

Using 'imply' causes a new problem, as it allows the case of CONFIG_INPUT=m with RC_CORE=y, which fails to link:

drivers/media/rc/rc-main.o: In function `ir_do_keyup':
rc-main.c:(.text+0x2b4): undefined reference to input_event' drivers/media/rc/rc-main.o: In function rc_repeat':
rc-main.c:(.text+0x350): undefined reference to input_event' drivers/media/rc/rc-main.o: In function rc_allocate_device':
rc-main.c:(.text+0x90c): undefined reference to `input_allocate_device'

Quoted 5 lines

Add a 'depends on' that allows building both with and without CONFIG_RC_CORE, but disallows combinations that don't link.

Fixes 5023cf32210d ("drm/bridge: make remote control optional")
Quoted 5 lines

Proper solution has been already merged via input tree[1].

[1]:
https://lore.kernel.org/lkml/CAKdAkRTGXNbUsuKASNGLfwUwC7Asod9K5baYLPWPU7EX-42-yA@mail.gmail.com/

At that link, I only see the patch that caused the regression, not the solution. Are you sure it's fixed?

Arnd

#3 lore raw
Quoted 20 lines · Arnd Bergmann
Quoted 19 lines · Andrzej Hajda

Hi Arnd,

Quoted 15 lines · Arnd Bergmann

Using 'imply' causes a new problem, as it allows the case of CONFIG_INPUT=m with RC_CORE=y, which fails to link:

drivers/media/rc/rc-main.o: In function `ir_do_keyup':
rc-main.c:(.text+0x2b4): undefined reference to input_event' drivers/media/rc/rc-main.o: In function rc_repeat':
rc-main.c:(.text+0x350): undefined reference to input_event' drivers/media/rc/rc-main.o: In function rc_allocate_device':
rc-main.c:(.text+0x90c): undefined reference to `input_allocate_device'

Add a 'depends on' that allows building both with and without CONFIG_RC_CORE, but disallows combinations that don't link.

Fixes 5023cf32210d ("drm/bridge: make remote control optional")
Quoted 7 lines
Quoted 5 lines

Proper solution has been already merged via input tree[1].

[1]:
https://lore.kernel.org/lkml/CAKdAkRTGXNbUsuKASNGLfwUwC7Asod9K5baYLPWPU7EX-42-yA@mail.gmail.com/

At that link, I only see the patch that caused the regression, not the solution. Are you sure it's fixed?

Ups, you are right, I though you are fixing what this patch attempted to fix :)

Anyway, we want to avoid dependency on RC_CORE - this driver does not require it, but with RC_CORE it has additional features.

Maybe "imply INPUT" would help?

Regards

Andrzej

Quoted 3 lines

Arnd

#4 lore raw
Quoted 30 lines · Andrzej Hajda
Quoted 20 lines · Arnd Bergmann
Quoted 19 lines · Andrzej Hajda

Hi Arnd,

Quoted 15 lines · Arnd Bergmann

Using 'imply' causes a new problem, as it allows the case of CONFIG_INPUT=m with RC_CORE=y, which fails to link:

drivers/media/rc/rc-main.o: In function `ir_do_keyup':
rc-main.c:(.text+0x2b4): undefined reference to input_event' drivers/media/rc/rc-main.o: In function rc_repeat':
rc-main.c:(.text+0x350): undefined reference to input_event' drivers/media/rc/rc-main.o: In function rc_allocate_device':
rc-main.c:(.text+0x90c): undefined reference to `input_allocate_device'

Add a 'depends on' that allows building both with and without CONFIG_RC_CORE, but disallows combinations that don't link.

Fixes 5023cf32210d ("drm/bridge: make remote control optional")
Quoted 7 lines
Quoted 5 lines

Proper solution has been already merged via input tree[1].

[1]:
https://lore.kernel.org/lkml/CAKdAkRTGXNbUsuKASNGLfwUwC7Asod9K5baYLPWPU7EX-42-yA@mail.gmail.com/

At that link, I only see the patch that caused the regression, not the solution. Are you sure it's fixed?

Quoted 5 lines

Ups, you are right, I though you are fixing what this patch attempted to fix :)

Anyway, we want to avoid dependency on RC_CORE - this driver does not require it, but with RC_CORE it has additional features.

Right, that's what my patch does: if RC_CORE is disabled, you can still set DRM_SIL_SII8620=y, but if RC_CORE=m, DRM_SIL_SII8620 can only be =m or =n.

Quoted 1 line

Maybe "imply INPUT" would help?

No, that would make it worse. Device drivers really have no business turning on other subsystems.

Arnd

#5 lore raw
Quoted 38 lines · Arnd Bergmann
Quoted 30 lines · Andrzej Hajda
Quoted 20 lines · Arnd Bergmann
Quoted 19 lines · Andrzej Hajda

Hi Arnd,

Quoted 15 lines · Arnd Bergmann

Using 'imply' causes a new problem, as it allows the case of CONFIG_INPUT=m with RC_CORE=y, which fails to link:

drivers/media/rc/rc-main.o: In function `ir_do_keyup':
rc-main.c:(.text+0x2b4): undefined reference to input_event' drivers/media/rc/rc-main.o: In function rc_repeat':
rc-main.c:(.text+0x350): undefined reference to input_event' drivers/media/rc/rc-main.o: In function rc_allocate_device':
rc-main.c:(.text+0x90c): undefined reference to `input_allocate_device'

Add a 'depends on' that allows building both with and without CONFIG_RC_CORE, but disallows combinations that don't link.

Fixes 5023cf32210d ("drm/bridge: make remote control optional")
Quoted 7 lines
Quoted 5 lines

Proper solution has been already merged via input tree[1].

[1]:
https://lore.kernel.org/lkml/CAKdAkRTGXNbUsuKASNGLfwUwC7Asod9K5baYLPWPU7EX-42-yA@mail.gmail.com/

At that link, I only see the patch that caused the regression, not the solution. Are you sure it's fixed?

Quoted 5 lines

Ups, you are right, I though you are fixing what this patch attempted to fix :)

Anyway, we want to avoid dependency on RC_CORE - this driver does not require it, but with RC_CORE it has additional features.

Quoted 9 lines

Right, that's what my patch does: if RC_CORE is disabled, you can still set DRM_SIL_SII8620=y, but if RC_CORE=m, DRM_SIL_SII8620 can only be =m or =n.

Quoted 1 line

Maybe "imply INPUT" would help?

No, that would make it worse. Device drivers really have no business turning on other subsystems.

OK, in the meantime I will redo the branch by dropping the sil-sii8620.c Kconfig changes and also drop all "imply" business from applespi driver as they give us more trouble than they are worth. We do not have "imply" for i801_smbus for Symaptics SMBUS mode and it works fine. It it distro's task to configure the kernel properly.

Thanks.

Signature · 1 line
Dmitry
#6 lore raw
Quoted 29 lines · Dmitry Torokhov
Quoted 18 lines · Arnd Bergmann
Quoted 11 lines · Andrzej Hajda
Quoted 8 lines · Arnd Bergmann
Quoted 5 lines · Andrzej Hajda

Proper solution has been already merged via input tree[1].

[1]:
https://lore.kernel.org/lkml/CAKdAkRTGXNbUsuKASNGLfwUwC7Asod9K5baYLPWPU7EX-42-yA@mail.gmail.com/

At that link, I only see the patch that caused the regression, not the solution. Are you sure it's fixed?

Quoted 5 lines

Ups, you are right, I though you are fixing what this patch attempted to fix :)

Anyway, we want to avoid dependency on RC_CORE - this driver does not require it, but with RC_CORE it has additional features.

Quoted 9 lines

Right, that's what my patch does: if RC_CORE is disabled, you can still set DRM_SIL_SII8620=y, but if RC_CORE=m, DRM_SIL_SII8620 can only be =m or =n.

Quoted 1 line

Maybe "imply INPUT" would help?

No, that would make it worse. Device drivers really have no business turning on other subsystems.

Quoted 5 lines

OK, in the meantime I will redo the branch by dropping the sil-sii8620.c Kconfig changes and also drop all "imply" business from applespi driver as they give us more trouble than they are worth. We do not have "imply" for i801_smbus for Symaptics SMBUS mode and it works fine. It it distro's task to configure the kernel properly.

Thanks!

I think the "drm/bridge: make remote control optional" patch is fine with my fixup, the IS_ENABLED() checks take care of the case where RC_CORE is unavailable, and the 'depends on RC_CORE || !RC_CORE' line takes care of the RC_CORE=m case.

I suppose Ronald could send a replacement patch with my fixup after the merge window.

Arnd

#7 lore raw
In reply to Andrzej Hajda
Quoted 31 lines · Andrzej Hajda
Quoted 20 lines · Arnd Bergmann
Quoted 19 lines · Andrzej Hajda

Hi Arnd,

Quoted 15 lines · Arnd Bergmann

Using 'imply' causes a new problem, as it allows the case of CONFIG_INPUT=m with RC_CORE=y, which fails to link:

drivers/media/rc/rc-main.o: In function `ir_do_keyup':
rc-main.c:(.text+0x2b4): undefined reference to input_event' drivers/media/rc/rc-main.o: In function rc_repeat':
rc-main.c:(.text+0x350): undefined reference to input_event' drivers/media/rc/rc-main.o: In function rc_allocate_device':
rc-main.c:(.text+0x90c): undefined reference to `input_allocate_device'

Add a 'depends on' that allows building both with and without CONFIG_RC_CORE, but disallows combinations that don't link.

Fixes 5023cf32210d ("drm/bridge: make remote control optional")
Quoted 7 lines
Quoted 5 lines

Proper solution has been already merged via input tree[1].

[1]:
https://lore.kernel.org/lkml/CAKdAkRTGXNbUsuKASNGLfwUwC7Asod9K5baYLPWPU7EX-42-yA@mail.gmail.com/

At that link, I only see the patch that caused the regression, not the solution. Are you sure it's fixed?

Quoted 7 lines

Ups, you are right, I though you are fixing what this patch attempted to fix :)

Anyway, we want to avoid dependency on RC_CORE - this driver does not require it, but with RC_CORE it has additional features.

Maybe "imply INPUT" would help?

No, it won't. I am sorry, I should have looked closer, but as written, drivers/gpu/drm/bridge/sil-sii8620.c has a hard dependency on the RC core and "imply" was the wrong solution for this, we need "depends on RC_CORE". If we want to make RC support optional than we should stub out paths that use RC_CORE (such as sii8620_init_rcp_input_dev()) and guard them by "#ifdef CONFIG_RC_CORE". Then we could keep "imply" on RC_CORE.

I am surprised though that imply allows violating the constraint on implied symbols, as RC_CORE has straight "depends on" for INPUT.

Thanks.

Signature · 1 line
Dmitry
#8 lore raw
In reply to Arnd Bergmann
Quoted 2 lines · Arnd Bergmann

Using 'imply' causes a new problem, as it allows the case of CONFIG_INPUT=m with RC_CORE=y, which fails to link:

I have reviewed dependencies and I wonder how such configuration is possible at all.

RC_CORE depends on INPUT (at least on today's next branch) so if INPUT=m then RC_CORE should be either n either m, am I right?

Arnd, are there unknown to me changes in RC/INPUT dependencies?

Regards

Andrzej

Quoted 31 lines

drivers/media/rc/rc-main.o: In function `ir_do_keyup':
rc-main.c:(.text+0x2b4): undefined reference to input_event' drivers/media/rc/rc-main.o: In function rc_repeat':
rc-main.c:(.text+0x350): undefined reference to input_event' drivers/media/rc/rc-main.o: In function rc_allocate_device':
rc-main.c:(.text+0x90c): undefined reference to `input_allocate_device'

Add a 'depends on' that allows building both with and without CONFIG_RC_CORE, but disallows combinations that don't link.

Fixes 5023cf32210d ("drm/bridge: make remote control optional")
---
 drivers/gpu/drm/bridge/Kconfig | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/bridge/Kconfig +1 -1
index f64c91defdc3..70a8ed2505aa 100644
@@ -85,8 +85,8 @@ config DRM_SIL_SII8620
 	tristate "Silicon Image SII8620 HDMI/MHL bridge"
 	depends on OF
 	select DRM_KMS_HELPER
+	depends on RC_CORE || !RC_CORE
 	imply EXTCON
-	imply RC_CORE
 	help
 	  Silicon Image SII8620 HDMI/MHL bridge chip driver.
 
#9 lore raw
Quoted 11 lines · Andrzej Hajda
Quoted 2 lines · Arnd Bergmann

Using 'imply' causes a new problem, as it allows the case of CONFIG_INPUT=m with RC_CORE=y, which fails to link:

I have reviewed dependencies and I wonder how such configuration is possible at all.

RC_CORE depends on INPUT (at least on today's next branch) so if INPUT=m then RC_CORE should be either n either m, am I right?

Right.

Quoted 1 line

Arnd, are there unknown to me changes in RC/INPUT dependencies?

I think this is 'imply' behaving oddly when we have conflicting requirements:

- INPUT=m forces RC_CORE to be =m or =n
- DRM_SIL_SII8620=y asks RC_CORE to be =y unless it cannot be enabled

Kconfig decided to make this RC_CORE=y, which caused the link failure. Making it RC_CORE=m however would not work either because then we'd get a link failure from the sii8620 driver to rc_core.

so a pure 'imply' cannot work here, and we need a dependency, one of:

a)
depends on INPUT || !INPUT
select RC_CORE if INPUT

b) depends on RC_CORE || !RC_CORE

b) is what othe drivers use, e.g. SMS_SDIO_DRV

Arnd

#10 lore raw
Quoted 24 lines · Arnd Bergmann
Quoted 9 lines · Andrzej Hajda
Quoted 2 lines · Arnd Bergmann

Using 'imply' causes a new problem, as it allows the case of CONFIG_INPUT=m with RC_CORE=y, which fails to link:

I have reviewed dependencies and I wonder how such configuration is possible at all.

RC_CORE depends on INPUT (at least on today's next branch) so if INPUT=m then RC_CORE should be either n either m, am I right?

Right.

Quoted 1 line

Arnd, are there unknown to me changes in RC/INPUT dependencies?

I think this is 'imply' behaving oddly when we have conflicting requirements:

- INPUT=m forces RC_CORE to be =m or =n
- DRM_SIL_SII8620=y asks RC_CORE to be =y unless it cannot be enabled

Kconfig decided to make this RC_CORE=y, which caused the link failure. Making it RC_CORE=m however would not work either because then we'd get a link failure from the sii8620 driver to rc_core.

so a pure 'imply' cannot work here, and we need a dependency, one of:

Quoted 7 lines

a)
depends on INPUT || !INPUT
select RC_CORE if INPUT

b) depends on RC_CORE || !RC_CORE

b) is what othe drivers use, e.g. SMS_SDIO_DRV

OK, thanks for explanation, really weird.

I though about imply as "weak dependency", but it is "weak select" with drawbacks of select.

Anyway I am surprised that Kconfig did not complain about contradictory requirements on RC_CORE symbol.

Regards

Andrzej

Quoted 3 lines

Arnd