diff mbox series

drm/drm_panel: fix export of drm_panel_of_backlight, try #3

Message ID 20200107203231.920256-1-arnd@arndb.de (mailing list archive)
State New, archived
Headers show
Series drm/drm_panel: fix export of drm_panel_of_backlight, try #3 | expand

Commit Message

Arnd Bergmann Jan. 7, 2020, 8:32 p.m. UTC
Making this IS_REACHABLE() was still wrong, as that just determines
whether the lower-level backlight code would be reachable from the panel
driver. However, with CONFIG_DRM=y and CONFIG_BACKLIGHT_CLASS_DEVICE=m,
the drm_panel_of_backlight is left out of drm_panel.o but the condition
tells the driver that it is there, leading to multiple link errors such as

ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-sitronix-st7701.ko] undefined!
ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-sharp-ls043t1le01.ko] undefined!
ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-seiko-43wvf1g.ko] undefined!
ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-ronbo-rb070d30.ko] undefined!
ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-rocktech-jh057n00900.ko] undefined!
ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-panasonic-vvx10f034n00.ko] undefined!
ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-osd-osd101t2587-53ts.ko] undefined!

Change the condition to check for whether the function was actually part
of the drm module. This version of the patch survived a few hundred
randconfig builds, so I have a good feeling this might be the last
one for the export.

Fixes: 4a34a9dcec94 ("drm/drm_panel: Fix EXPORT of drm_panel_of_backlight() one more time")
Fixes: 907aa265fde6 ("drm/drm_panel: fix EXPORT of drm_panel_of_backlight")
Fixes: 152dbdeab1b2 ("drm/panel: add backlight support")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 include/drm/drm_panel.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

Comments

Jani Nikula Jan. 8, 2020, 9:55 a.m. UTC | #1
On Tue, 07 Jan 2020, Arnd Bergmann <arnd@arndb.de> wrote:
> Making this IS_REACHABLE() was still wrong, as that just determines
> whether the lower-level backlight code would be reachable from the panel
> driver. However, with CONFIG_DRM=y and CONFIG_BACKLIGHT_CLASS_DEVICE=m,
> the drm_panel_of_backlight is left out of drm_panel.o but the condition
> tells the driver that it is there, leading to multiple link errors such as
>
> ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-sitronix-st7701.ko] undefined!
> ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-sharp-ls043t1le01.ko] undefined!
> ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-seiko-43wvf1g.ko] undefined!
> ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-ronbo-rb070d30.ko] undefined!
> ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-rocktech-jh057n00900.ko] undefined!
> ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-panasonic-vvx10f034n00.ko] undefined!
> ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-osd-osd101t2587-53ts.ko] undefined!
>
> Change the condition to check for whether the function was actually part
> of the drm module. This version of the patch survived a few hundred
> randconfig builds, so I have a good feeling this might be the last
> one for the export.

Broken record, this will still be wrong, even if it builds and links. No
backlight support for panel despite expectations.

See http://mid.mail-archive.com/87d0cnynst.fsf@intel.com

All of this is just another hack until the backlight config usage is
fixed for good. Do we really want to make this the example to copy paste
wherever we hit the issue next?

I'm not naking, but I'm not acking either.

BR,
Jani.


>
> Fixes: 4a34a9dcec94 ("drm/drm_panel: Fix EXPORT of drm_panel_of_backlight() one more time")
> Fixes: 907aa265fde6 ("drm/drm_panel: fix EXPORT of drm_panel_of_backlight")
> Fixes: 152dbdeab1b2 ("drm/panel: add backlight support")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> ---
>  include/drm/drm_panel.h | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/include/drm/drm_panel.h b/include/drm/drm_panel.h
> index 121f7aabccd1..6193cb555acc 100644
> --- a/include/drm/drm_panel.h
> +++ b/include/drm/drm_panel.h
> @@ -198,7 +198,8 @@ static inline struct drm_panel *of_drm_find_panel(const struct device_node *np)
>  }
>  #endif
>  
> -#if IS_REACHABLE(CONFIG_BACKLIGHT_CLASS_DEVICE)
> +#if IS_ENABLED(CONFIG_DRM_PANEL) && (IS_BUILTIN(CONFIG_BACKLIGHT_CLASS_DEVICE) || \
> +	(IS_MODULE(CONFIG_DRM) && IS_MODULE(CONFIG_BACKLIGHT_CLASS_DEVICE)))
>  int drm_panel_of_backlight(struct drm_panel *panel);
>  #else
>  static inline int drm_panel_of_backlight(struct drm_panel *panel)
Sam Ravnborg Jan. 8, 2020, 10:08 a.m. UTC | #2
Hi Jani.

On Wed, Jan 08, 2020 at 11:55:29AM +0200, Jani Nikula wrote:
> On Tue, 07 Jan 2020, Arnd Bergmann <arnd@arndb.de> wrote:
> > Making this IS_REACHABLE() was still wrong, as that just determines
> > whether the lower-level backlight code would be reachable from the panel
> > driver. However, with CONFIG_DRM=y and CONFIG_BACKLIGHT_CLASS_DEVICE=m,
> > the drm_panel_of_backlight is left out of drm_panel.o but the condition
> > tells the driver that it is there, leading to multiple link errors such as
> >
> > ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-sitronix-st7701.ko] undefined!
> > ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-sharp-ls043t1le01.ko] undefined!
> > ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-seiko-43wvf1g.ko] undefined!
> > ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-ronbo-rb070d30.ko] undefined!
> > ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-rocktech-jh057n00900.ko] undefined!
> > ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-panasonic-vvx10f034n00.ko] undefined!
> > ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-osd-osd101t2587-53ts.ko] undefined!
> >
> > Change the condition to check for whether the function was actually part
> > of the drm module. This version of the patch survived a few hundred
> > randconfig builds, so I have a good feeling this might be the last
> > one for the export.
> 
> Broken record, this will still be wrong, even if it builds and links. No
> backlight support for panel despite expectations.
> 
> See http://mid.mail-archive.com/87d0cnynst.fsf@intel.com
> 
> All of this is just another hack until the backlight config usage is
> fixed for good. Do we really want to make this the example to copy paste
> wherever we hit the issue next?
> 
> I'm not naking, but I'm not acking either.

I will try to take a look at your old BACKLIGHT_CLASS_DEVICE patch this
weekend. I think we need that one fixed - and then we can have this mess
with "drm_panel_of_backlight" fixed in the right way.

Sigh...

	Sam
Arnd Bergmann Jan. 8, 2020, 10:33 a.m. UTC | #3
On Wed, Jan 8, 2020 at 11:08 AM Sam Ravnborg <sam@ravnborg.org> wrote:
> On Wed, Jan 08, 2020 at 11:55:29AM +0200, Jani Nikula wrote:
> > On Tue, 07 Jan 2020, Arnd Bergmann <arnd@arndb.de> wrote:
> > > Making this IS_REACHABLE() was still wrong, as that just determines
> > > whether the lower-level backlight code would be reachable from the panel
> > > driver. However, with CONFIG_DRM=y and CONFIG_BACKLIGHT_CLASS_DEVICE=m,
> > > the drm_panel_of_backlight is left out of drm_panel.o but the condition
> > > tells the driver that it is there, leading to multiple link errors such as
> > >
> > > ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-sitronix-st7701.ko] undefined!
> > > ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-sharp-ls043t1le01.ko] undefined!
> > > ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-seiko-43wvf1g.ko] undefined!
> > > ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-ronbo-rb070d30.ko] undefined!
> > > ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-rocktech-jh057n00900.ko] undefined!
> > > ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-panasonic-vvx10f034n00.ko] undefined!
> > > ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-osd-osd101t2587-53ts.ko] undefined!
> > >
> > > Change the condition to check for whether the function was actually part
> > > of the drm module. This version of the patch survived a few hundred
> > > randconfig builds, so I have a good feeling this might be the last
> > > one for the export.
> >
> > Broken record, this will still be wrong, even if it builds and links. No
> > backlight support for panel despite expectations.
> >
> > See http://mid.mail-archive.com/87d0cnynst.fsf@intel.com
> >
> > All of this is just another hack until the backlight config usage is
> > fixed for good. Do we really want to make this the example to copy paste
> > wherever we hit the issue next?
> >
> > I'm not naking, but I'm not acking either.
>
> I will try to take a look at your old BACKLIGHT_CLASS_DEVICE patch this
> weekend. I think we need that one fixed - and then we can have this mess
> with "drm_panel_of_backlight" fixed in the right way.

I had also attempted to fix the larger mess around 'select' statements in DRM/FB
around BACKLIGHT_CLASS_DEVICE  several times in the past, and even at
some point sent a patch that was acked but never merged and later broke because
of other changes.

If there is a new approach, I'm happy to add patches to my randconfig builder
and see if there are regressions in some corner cases.

      Arnd
Sam Ravnborg Jan. 8, 2020, 4:46 p.m. UTC | #4
Hi Arnd.

> > > All of this is just another hack until the backlight config usage is
> > > fixed for good. Do we really want to make this the example to copy paste
> > > wherever we hit the issue next?
> > >
> > > I'm not naking, but I'm not acking either.
> >
> > I will try to take a look at your old BACKLIGHT_CLASS_DEVICE patch this
> > weekend. I think we need that one fixed - and then we can have this mess
> > with "drm_panel_of_backlight" fixed in the right way.
> 
> I had also attempted to fix the larger mess around 'select' statements in DRM/FB
> around BACKLIGHT_CLASS_DEVICE  several times in the past, and even at
> some point sent a patch that was acked but never merged and later broke because
> of other changes.

Any chance you have the patch around or can dig up a pointer?
My google foo did not turn up anything.

	Sam
Arnd Bergmann Jan. 8, 2020, 8:17 p.m. UTC | #5
On Wed, Jan 8, 2020 at 5:46 PM Sam Ravnborg <sam@ravnborg.org> wrote:
>
> Hi Arnd.
>
> > > > All of this is just another hack until the backlight config usage is
> > > > fixed for good. Do we really want to make this the example to copy paste
> > > > wherever we hit the issue next?
> > > >
> > > > I'm not naking, but I'm not acking either.
> > >
> > > I will try to take a look at your old BACKLIGHT_CLASS_DEVICE patch this
> > > weekend. I think we need that one fixed - and then we can have this mess
> > > with "drm_panel_of_backlight" fixed in the right way.
> >
> > I had also attempted to fix the larger mess around 'select' statements in DRM/FB
> > around BACKLIGHT_CLASS_DEVICE  several times in the past, and even at
> > some point sent a patch that was acked but never merged and later broke because
> > of other changes.
>
> Any chance you have the patch around or can dig up a pointer?
> My google foo did not turn up anything.

I found it now:
https://lore.kernel.org/lkml/20170726135312.2214309-1-arnd@arndb.de/

      Arnd
Sam Ravnborg Jan. 12, 2020, 8:22 p.m. UTC | #6
Hi Arnd.

On Tue, Jan 07, 2020 at 09:32:19PM +0100, Arnd Bergmann wrote:
> Making this IS_REACHABLE() was still wrong, as that just determines
> whether the lower-level backlight code would be reachable from the panel
> driver. However, with CONFIG_DRM=y and CONFIG_BACKLIGHT_CLASS_DEVICE=m,
> the drm_panel_of_backlight is left out of drm_panel.o but the condition
> tells the driver that it is there, leading to multiple link errors such as
> 
> ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-sitronix-st7701.ko] undefined!
> ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-sharp-ls043t1le01.ko] undefined!
> ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-seiko-43wvf1g.ko] undefined!
> ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-ronbo-rb070d30.ko] undefined!
> ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-rocktech-jh057n00900.ko] undefined!
> ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-panasonic-vvx10f034n00.ko] undefined!
> ERROR: "drm_panel_of_backlight" [drivers/gpu/drm/panel/panel-osd-osd101t2587-53ts.ko] undefined!
> 
> Change the condition to check for whether the function was actually part
> of the drm module. This version of the patch survived a few hundred
> randconfig builds, so I have a good feeling this might be the last
> one for the export.
> 
> Fixes: 4a34a9dcec94 ("drm/drm_panel: Fix EXPORT of drm_panel_of_backlight() one more time")
> Fixes: 907aa265fde6 ("drm/drm_panel: fix EXPORT of drm_panel_of_backlight")
> Fixes: 152dbdeab1b2 ("drm/panel: add backlight support")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>

I spent some hours this weekend trying to improve the Kconfig
logic around BACKLIGHT_CLASS_DEVICE and ACPI_VIDEO.

I did manage to change something but I did not manage to make something
that is a real improvement to the current situation.
So everything abandoned.

As we need this fixed I have applied your patch to drm-misc-next.
Thanks,

	Sam

> ---
>  include/drm/drm_panel.h | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/include/drm/drm_panel.h b/include/drm/drm_panel.h
> index 121f7aabccd1..6193cb555acc 100644
> --- a/include/drm/drm_panel.h
> +++ b/include/drm/drm_panel.h
> @@ -198,7 +198,8 @@ static inline struct drm_panel *of_drm_find_panel(const struct device_node *np)
>  }
>  #endif
>  
> -#if IS_REACHABLE(CONFIG_BACKLIGHT_CLASS_DEVICE)
> +#if IS_ENABLED(CONFIG_DRM_PANEL) && (IS_BUILTIN(CONFIG_BACKLIGHT_CLASS_DEVICE) || \
> +	(IS_MODULE(CONFIG_DRM) && IS_MODULE(CONFIG_BACKLIGHT_CLASS_DEVICE)))
>  int drm_panel_of_backlight(struct drm_panel *panel);
>  #else
>  static inline int drm_panel_of_backlight(struct drm_panel *panel)
> -- 
> 2.20.0
diff mbox series

Patch

diff --git a/include/drm/drm_panel.h b/include/drm/drm_panel.h
index 121f7aabccd1..6193cb555acc 100644
--- a/include/drm/drm_panel.h
+++ b/include/drm/drm_panel.h
@@ -198,7 +198,8 @@  static inline struct drm_panel *of_drm_find_panel(const struct device_node *np)
 }
 #endif
 
-#if IS_REACHABLE(CONFIG_BACKLIGHT_CLASS_DEVICE)
+#if IS_ENABLED(CONFIG_DRM_PANEL) && (IS_BUILTIN(CONFIG_BACKLIGHT_CLASS_DEVICE) || \
+	(IS_MODULE(CONFIG_DRM) && IS_MODULE(CONFIG_BACKLIGHT_CLASS_DEVICE)))
 int drm_panel_of_backlight(struct drm_panel *panel);
 #else
 static inline int drm_panel_of_backlight(struct drm_panel *panel)