Skip to contentSkip to conversation

drm/panel: add CONFIG_DRM_KMS_HELPER dependencies

PATCH In review Arnd Bergmann 3 participants 9 messages Mar 16, 2022 – Apr 6, 2022

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/panel: add CONFIG_DRM_KMS_HELPER dependencies

From: Arnd Bergmann <[email protected]>

The driver fails to build when the KMS helpers are disabled:

ld.lld: error: undefined symbol: drm_gem_fb_get_obj

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

referenced by drm_mipi_dbi.c
gpu/drm/drm_mipi_dbi.o:(mipi_dbi_buf_copy) in archive drivers/built-in.a
referenced by drm_mipi_dbi.c
gpu/drm/drm_mipi_dbi.o:(mipi_dbi_fb_dirty) in archive drivers/built-in.a

ld.lld: error: undefined symbol: drm_gem_fb_begin_cpu_access

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

referenced by drm_mipi_dbi.c
gpu/drm/drm_mipi_dbi.o:(mipi_dbi_buf_copy) in archive drivers/built-in.a

ld.lld: error: undefined symbol: drm_fb_swab

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

referenced by drm_mipi_dbi.c
gpu/drm/drm_mipi_dbi.o:(mipi_dbi_buf_copy) in archive drivers/built-in.a

ld.lld: error: undefined symbol: drm_fb_xrgb8888_to_rgb565

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

referenced by drm_mipi_dbi.c
gpu/drm/drm_mipi_dbi.o:(mipi_dbi_buf_copy) in archive drivers/built-in.a

ld.lld: error: undefined symbol: drm_fb_memcpy

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

referenced by drm_mipi_dbi.c
gpu/drm/drm_mipi_dbi.o:(mipi_dbi_buf_copy) in archive drivers/built-in.a

This is fairly hard to hit in randconfig drivers, but it eventually did trigger for me in a configuration where all other DRM drivers are loadable modules, but DRM_PANEL_WIDECHIPS_WS2401 was built-in.

Adding a dependency in all drivers that select DRM_MIPI_DBI avoids the problem for now, adding the dependency in DRM_MIPI_DBI as well should help make it easier to figure out why it breaks if someone forgets the dependency the next time.

---
 drivers/gpu/drm/Kconfig       | 2 +-
 drivers/gpu/drm/panel/Kconfig | 4 ++++
 2 files changed, 5 insertions(+), 1 deletion(-)

I see this warning on 5.17-rc8, but did not test it on linux-next, which may already have a fix.

drivers/gpu/drm/Kconfig +1 -1
index b1f22e457fd0..d5ec0b77c010 100644
@@ -30,7 +30,7 @@ menuconfig DRM
 
 config DRM_MIPI_DBI
 	tristate
-	depends on DRM
+	depends on DRM_KMS_HELPER
 
 config DRM_MIPI_DSI
 	bool
drivers/gpu/drm/panel/Kconfig +4 -0
index 0aec5a10b064..96887d0efb9f 100644
@@ -425,6 +425,7 @@ config DRM_PANEL_SAMSUNG_DB7430
 	tristate "Samsung DB7430-based DPI panels"
 	depends on OF && SPI && GPIOLIB
 	depends on BACKLIGHT_CLASS_DEVICE
+	depends on DRM_KMS_HELPER
 	select DRM_MIPI_DBI
 	help
 	  Say Y here if you want to enable support for the Samsung
@@ -440,6 +441,7 @@ config DRM_PANEL_SAMSUNG_S6D16D0
 config DRM_PANEL_SAMSUNG_S6D27A1
 	tristate "Samsung S6D27A1 DPI panel driver"
 	depends on OF && SPI && GPIOLIB
+	depends on DRM_KMS_HELPER
 	select DRM_MIPI_DBI
 	help
 	  Say Y here if you want to enable support for the Samsung
@@ -476,6 +478,7 @@ config DRM_PANEL_SAMSUNG_S6E63M0_SPI
 	depends on SPI
 	depends on DRM_PANEL_SAMSUNG_S6E63M0
 	default DRM_PANEL_SAMSUNG_S6E63M0
+	depends on DRM_KMS_HELPER
 	select DRM_MIPI_DBI
 	help
 	  Say Y here if you want to be able to access the Samsung
@@ -677,6 +680,7 @@ config DRM_PANEL_WIDECHIPS_WS2401
 	tristate "Widechips WS2401 DPI panel driver"
 	depends on SPI && GPIOLIB
 	depends on BACKLIGHT_CLASS_DEVICE
+	depends on DRM_KMS_HELPER
 	select DRM_MIPI_DBI
 	help
 	  Say Y here if you want to enable support for the Widechips WS2401 DPI
Signature · 1 line
2.29.2
#1 lore raw

PATCH 1/1 drm/tegra: vic: fix unused-function warnings

From: Arnd Bergmann <[email protected]>

The use of the old-style SET_RUNTIME_PM_OPS() and SET_SYSTEM_SLEEP_PM_OPS() macros requires function definitions to be hidden to avoid

drivers/gpu/drm/tegra/vic.c:326:12: error: 'vic_runtime_suspend' defined but not used [-Werror=unused-function]
326 | static int vic_runtime_suspend(struct device *dev)

      |            ^~~~~~~~~~~~~~~~~~~
drivers/gpu/drm/tegra/vic.c:292:12: error: 'vic_runtime_resume' defined but not used [-Werror=unused-function]
  292 | static int vic_runtime_resume(struct device *dev)
      |            ^~~~~~~~~~~~~~~~~~

Use the new-style SYSTEM_SLEEP_PM_OPS() and RUNTIME_PM_OPS() instead.

Fixes 1e15f5b911d6 ("drm/tegra: vic: Stop channel on suspend")
---
 drivers/gpu/drm/tegra/vic.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

I see this warning on 5.17-rc8, but did not test it on linux-next, which may already have a fix.

drivers/gpu/drm/tegra/vic.c +2 -3
index 1e342fa3d27b..f56f5921a8c2 100644
@@ -513,9 +513,8 @@ static int vic_remove(struct platform_device *pdev)
 }
 
 static const struct dev_pm_ops vic_pm_ops = {
-	SET_RUNTIME_PM_OPS(vic_runtime_suspend, vic_runtime_resume, NULL)
-	SET_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
-				pm_runtime_force_resume)
+	RUNTIME_PM_OPS(vic_runtime_suspend, vic_runtime_resume, NULL)
+	SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend, pm_runtime_force_resume)
 };
 
 struct platform_driver tegra_vic_driver = {
Signature · 1 line
2.29.2
#2 lore raw

Re: [PATCH] drm/tegra: vic: fix unused-function warnings

Quoted 30 lines · Arnd Bergmann

From: Arnd Bergmann <[email protected]>

The use of the old-style SET_RUNTIME_PM_OPS() and SET_SYSTEM_SLEEP_PM_OPS() macros requires function definitions to be hidden to avoid

drivers/gpu/drm/tegra/vic.c:326:12: error: 'vic_runtime_suspend' defined but not used [-Werror=unused-function]
326 | static int vic_runtime_suspend(struct device *dev)

      |            ^~~~~~~~~~~~~~~~~~~
drivers/gpu/drm/tegra/vic.c:292:12: error: 'vic_runtime_resume' defined but not used [-Werror=unused-function]
  292 | static int vic_runtime_resume(struct device *dev)
      |            ^~~~~~~~~~~~~~~~~~

Use the new-style SYSTEM_SLEEP_PM_OPS() and RUNTIME_PM_OPS() instead.

Fixes 1e15f5b911d6 ("drm/tegra: vic: Stop channel on suspend")
---
 drivers/gpu/drm/tegra/vic.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

I see this warning on 5.17-rc8, but did not test it on linux-next, which may already have a fix.

drivers/gpu/drm/tegra/vic.c +0 -0
index 1e342fa3d27b..f56f5921a8c2 100644
@@ -513,9 +513,8 @@ static int vic_remove(struct platform_device *pdev)
 }

drivers/gpu/drm/tegra/vic.c static int vic_remove(struct platform_device *pdev) · line 520

Quoted 10 lines
drivers/gpu/drm/tegra/vic.c +2 -3
 
 static const struct dev_pm_ops vic_pm_ops = {
-	SET_RUNTIME_PM_OPS(vic_runtime_suspend, vic_runtime_resume, NULL)
-	SET_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
-				pm_runtime_force_resume)
+	RUNTIME_PM_OPS(vic_runtime_suspend, vic_runtime_resume, NULL)
+	SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend, pm_runtime_force_resume)
 };
 
 struct platform_driver tegra_vic_driver = {

Hi Arnd,

is this a replacement for __maybe_unused annotations that we would typically use to address these? Is the ternary operator in PTR_IF enough to eliminate the warning? Does that work the same way for structure definitions as it does for conditionals where we use IS_ENABLED() to use the compiler's DCE for improved coverage?

It looks like it, but just making sure because there's another patch that fixes this warning by adding __maybe_unused.

Thierry

#3 lore raw

Re: [PATCH] drm/tegra: vic: fix unused-function warnings

Quoted 11 lines · Thierry Reding

drivers/gpu/drm/tegra/vic.c static int vic_remove(struct platform_device *pdev) · line 520

Quoted 9 lines · Arnd Bergmann

static const struct dev_pm_ops vic_pm_ops = {

drivers/gpu/drm/tegra/vic.c +2 -3
-     SET_RUNTIME_PM_OPS(vic_runtime_suspend, vic_runtime_resume, NULL)
-     SET_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
-                             pm_runtime_force_resume)
+     RUNTIME_PM_OPS(vic_runtime_suspend, vic_runtime_resume, NULL)
+     SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend, pm_runtime_force_resume)
 };
 
 struct platform_driver tegra_vic_driver = {
Quoted 7 lines

Hi Arnd,

is this a replacement for __maybe_unused annotations that we would typically use to address these? Is the ternary operator in PTR_IF enough to eliminate the warning? Does that work the same way for structure definitions as it does for conditionals where we use IS_ENABLED() to use the compiler's DCE for improved coverage?

Yes to all three.

Quoted 2 lines

It looks like it, but just making sure because there's another patch that fixes this warning by adding __maybe_unused.

I sent a lot of patches in the past to add __maybe_unused, but this was mainly because we could never come up with a good replacement. Paul Cercueil has finally come up with a good solution, so this is how we should do it from now on.

Arnd

#4 lore raw
In reply to Arnd Bergmann

Hi

Am 16.03.22 um 19:36 schrieb Arnd Bergmann:

Quoted 50 lines

From: Arnd Bergmann <[email protected]>

The driver fails to build when the KMS helpers are disabled:

ld.lld: error: undefined symbol: drm_gem_fb_get_obj

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

referenced by drm_mipi_dbi.c
gpu/drm/drm_mipi_dbi.o:(mipi_dbi_buf_copy) in archive drivers/built-in.a
referenced by drm_mipi_dbi.c
gpu/drm/drm_mipi_dbi.o:(mipi_dbi_fb_dirty) in archive drivers/built-in.a

ld.lld: error: undefined symbol: drm_gem_fb_begin_cpu_access

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

referenced by drm_mipi_dbi.c
gpu/drm/drm_mipi_dbi.o:(mipi_dbi_buf_copy) in archive drivers/built-in.a

ld.lld: error: undefined symbol: drm_fb_swab

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

referenced by drm_mipi_dbi.c
gpu/drm/drm_mipi_dbi.o:(mipi_dbi_buf_copy) in archive drivers/built-in.a

ld.lld: error: undefined symbol: drm_fb_xrgb8888_to_rgb565

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

referenced by drm_mipi_dbi.c
gpu/drm/drm_mipi_dbi.o:(mipi_dbi_buf_copy) in archive drivers/built-in.a

ld.lld: error: undefined symbol: drm_fb_memcpy

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

referenced by drm_mipi_dbi.c
gpu/drm/drm_mipi_dbi.o:(mipi_dbi_buf_copy) in archive drivers/built-in.a

This is fairly hard to hit in randconfig drivers, but it eventually did trigger for me in a configuration where all other DRM drivers are loadable modules, but DRM_PANEL_WIDECHIPS_WS2401 was built-in.

Adding a dependency in all drivers that select DRM_MIPI_DBI avoids the problem for now, adding the dependency in DRM_MIPI_DBI as well should help make it easier to figure out why it breaks if someone forgets the dependency the next time.

---
  drivers/gpu/drm/Kconfig       | 2 +-
  drivers/gpu/drm/panel/Kconfig | 4 ++++
  2 files changed, 5 insertions(+), 1 deletion(-)

I see this warning on 5.17-rc8, but did not test it on linux-next, which may already have a fix.

drivers/gpu/drm/Kconfig +0 -0
index b1f22e457fd0..d5ec0b77c010 100644

drivers/gpu/drm/Kconfig menuconfig DRM · line 33

Quoted 5 lines
drivers/gpu/drm/Kconfig +1 -1
@@ -30,7 +30,7 @@ menuconfig DRM
 
  config DRM_MIPI_DBI
  	tristate
-	depends on DRM
+	depends on DRM_KMS_HELPER

This symbol cannot be selected by users, so it's maybe not a good idea to depend on it. In fact, I've had to remove such a statement because it created a cyclic dependency. [1]

Making the drivers depend on KMS helpers is the right thing though. If there's a better solution, please let me know.

Best regards
Thomas

Quoted 39 lines
drivers/gpu/drm/Kconfig +0 -0
  config DRM_MIPI_DSI
  	bool
drivers/gpu/drm/panel/Kconfig +4 -0
index 0aec5a10b064..96887d0efb9f 100644
@@ -425,6 +425,7 @@ config DRM_PANEL_SAMSUNG_DB7430
  	tristate "Samsung DB7430-based DPI panels"
  	depends on OF && SPI && GPIOLIB
  	depends on BACKLIGHT_CLASS_DEVICE
+	depends on DRM_KMS_HELPER
  	select DRM_MIPI_DBI
  	help
  	  Say Y here if you want to enable support for the Samsung
@@ -440,6 +441,7 @@ config DRM_PANEL_SAMSUNG_S6D16D0
  config DRM_PANEL_SAMSUNG_S6D27A1
  	tristate "Samsung S6D27A1 DPI panel driver"
  	depends on OF && SPI && GPIOLIB
+	depends on DRM_KMS_HELPER
  	select DRM_MIPI_DBI
  	help
  	  Say Y here if you want to enable support for the Samsung
@@ -476,6 +478,7 @@ config DRM_PANEL_SAMSUNG_S6E63M0_SPI
  	depends on SPI
  	depends on DRM_PANEL_SAMSUNG_S6E63M0
  	default DRM_PANEL_SAMSUNG_S6E63M0
+	depends on DRM_KMS_HELPER
  	select DRM_MIPI_DBI
  	help
  	  Say Y here if you want to be able to access the Samsung
@@ -677,6 +680,7 @@ config DRM_PANEL_WIDECHIPS_WS2401
  	tristate "Widechips WS2401 DPI panel driver"
  	depends on SPI && GPIOLIB
  	depends on BACKLIGHT_CLASS_DEVICE
+	depends on DRM_KMS_HELPER
  	select DRM_MIPI_DBI
  	help
  	  Say Y here if you want to enable support for the Widechips WS2401 DPI
Signature · 6 lines
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Ivo Totev
#5 lore raw

Am 16.03.22 um 20:12 schrieb Thomas Zimmermann:

Quoted 48 lines

Hi

Am 16.03.22 um 19:36 schrieb Arnd Bergmann:

Quoted 45 lines

From: Arnd Bergmann <[email protected]>

The driver fails to build when the KMS helpers are disabled:

ld.lld: error: undefined symbol: drm_gem_fb_get_obj

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

referenced by drm_mipi_dbi.c
               gpu/drm/drm_mipi_dbi.o:(mipi_dbi_buf_copy) in archive drivers/built-in.a referenced by drm_mipi_dbi.c
               gpu/drm/drm_mipi_dbi.o:(mipi_dbi_fb_dirty) in archive drivers/built-in.a

ld.lld: error: undefined symbol: drm_gem_fb_begin_cpu_access

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

referenced by drm_mipi_dbi.c
               gpu/drm/drm_mipi_dbi.o:(mipi_dbi_buf_copy) in archive drivers/built-in.a

ld.lld: error: undefined symbol: drm_fb_swab

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

referenced by drm_mipi_dbi.c
               gpu/drm/drm_mipi_dbi.o:(mipi_dbi_buf_copy) in archive drivers/built-in.a

ld.lld: error: undefined symbol: drm_fb_xrgb8888_to_rgb565

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

referenced by drm_mipi_dbi.c
               gpu/drm/drm_mipi_dbi.o:(mipi_dbi_buf_copy) in archive drivers/built-in.a

ld.lld: error: undefined symbol: drm_fb_memcpy

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

referenced by drm_mipi_dbi.c
               gpu/drm/drm_mipi_dbi.o:(mipi_dbi_buf_copy) in archive drivers/built-in.a

This is fairly hard to hit in randconfig drivers, but it eventually did trigger for me in a configuration where all other DRM drivers are loadable modules, but DRM_PANEL_WIDECHIPS_WS2401 was built-in.

Adding a dependency in all drivers that select DRM_MIPI_DBI avoids the problem for now, adding the dependency in DRM_MIPI_DBI as well should help make it easier to figure out why it breaks if someone forgets the dependency the next time.

---
  drivers/gpu/drm/Kconfig       | 2 +-
  drivers/gpu/drm/panel/Kconfig | 4 ++++

  2 files changed, 5 insertions(+), 1 deletion(-)

I see this warning on 5.17-rc8, but did not test it on linux-next, which may already have a fix.

Quoted 11 lines
Quoted 9 lines
drivers/gpu/drm/Kconfig +0 -0
index b1f22e457fd0..d5ec0b77c010 100644
@@ -30,7 +30,7 @@ menuconfig DRM

  config DRM_MIPI_DBI
      tristate
-    depends on DRM
+    depends on DRM_KMS_HELPER

This symbol cannot be selected by users, so it's maybe not a good idea to depend on it. In fact, I've had to remove such a statement because it created a cyclic dependency. [1]

[1] https://lore.kernel.org/dri-devel/[email protected]/

Quoted 45 lines

Making the drivers depend on KMS helpers is the right thing though. If there's a better solution, please let me know.

Best regards
Thomas

Quoted 38 lines

  config DRM_MIPI_DSI
      bool

drivers/gpu/drm/panel/Kconfig +0 -0
index 0aec5a10b064..96887d0efb9f 100644
@@ -425,6 +425,7 @@ config DRM_PANEL_SAMSUNG_DB7430

      tristate "Samsung DB7430-based DPI panels" depends on OF && SPI && GPIOLIB
      depends on BACKLIGHT_CLASS_DEVICE
+    depends on DRM_KMS_HELPER
      select DRM_MIPI_DBI
      help
        Say Y here if you want to enable support for the Samsung

drivers/gpu/drm/panel/Kconfig +0 -0
@@ -440,6 +441,7 @@ config DRM_PANEL_SAMSUNG_S6D16D0

  config DRM_PANEL_SAMSUNG_S6D27A1
      tristate "Samsung S6D27A1 DPI panel driver" depends on OF && SPI && GPIOLIB
+    depends on DRM_KMS_HELPER
      select DRM_MIPI_DBI
      help
        Say Y here if you want to enable support for the Samsung

drivers/gpu/drm/panel/Kconfig +0 -0
@@ -476,6 +478,7 @@ config DRM_PANEL_SAMSUNG_S6E63M0_SPI

      depends on SPI
      depends on DRM_PANEL_SAMSUNG_S6E63M0
      default DRM_PANEL_SAMSUNG_S6E63M0
+    depends on DRM_KMS_HELPER
      select DRM_MIPI_DBI
      help
        Say Y here if you want to be able to access the Samsung

drivers/gpu/drm/panel/Kconfig +0 -0
@@ -677,6 +680,7 @@ config DRM_PANEL_WIDECHIPS_WS2401

      tristate "Widechips WS2401 DPI panel driver"
      depends on SPI && GPIOLIB
      depends on BACKLIGHT_CLASS_DEVICE
+    depends on DRM_KMS_HELPER
      select DRM_MIPI_DBI
      help
        Say Y here if you want to enable support for the Widechips WS2401 DPI

Signature · 6 lines
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Ivo Totev
#6 lore raw
Quoted 14 lines · Thomas Zimmermann

Am 16.03.22 um 20:12 schrieb Thomas Zimmermann:

Quoted 13 lines

drivers/gpu/drm/Kconfig menuconfig DRM · line 33

Quoted 9 lines

Adding a dependency in all drivers that select DRM_MIPI_DBI avoids the problem for now, adding the dependency in DRM_MIPI_DBI as well should help make it easier to figure out why it breaks if someone forgets the dependency the next time.

tristate

drivers/gpu/drm/Kconfig +1 -1
-    depends on DRM
+    depends on DRM_KMS_HELPER

This symbol cannot be selected by users, so it's maybe not a good idea to depend on it. In fact, I've had to remove such a statement because it created a cyclic dependency. [1]

I tried to explain above what I was thinking here: the added dependency is both a correct statement (DRM_MIPI_DBI depends on DRM_KMS_HELPER because it cannot be built without DRM_KMS_HELPER) and helpful as an indication what went wrong if we run into the same problem with a new driver, instead of the cryptic link failure you get something like

WARNING: unmet direct dependencies detected for DRM_MIPI_DBI
Depends on [m]: HAS_IOMEM [=y] && DRM_KMS_HELPER [=m] Selected by [y]:
- DRM_PANEL_WIDECHIPS_WS2401 [=y] && HAS_IOMEM [=y] && DRM [=y] &&
DRM_PANEL [=y] && SPI [=y] && GPIOLIB [=y] && BACKLIGHT_CLASS_DEVICE [=y]
Selected by [m]:
- TINYDRM_ILI9225 [=m] && HAS_IOMEM [=y] && DRM [=y] && SPI [=y]

Quoted 2 lines

[1]
https://lore.kernel.org/dri-devel/[email protected]/

I was going for 'depends on' in the panel drivers because I saw the same being done for other panel drivers, and mixing the two methods causes dependency loops. I looked again now, and find that 'select DRM_KMS_HELPER' is more common for other drivers, and makes sense here because it is generally not user-selectable.

The easiest replacement for my patch would then be to just use 'select DRM_KMS_HELPER' from CONFIG_DRM_MIPI_DBI, which makes it
safer and more consistent with your change. If you like, I'll send an updated version.

One thing I'm not sure about is whether there is still use for ever having CONFIG_DRM without CONFIG_DRM_KMS_HELPER if it gets selected by almost every driver anyway. Is this actually a configuration that users rely on, or should we just remove the symbol completely and build the KMS helpers unconditionally?

Arnd

maintainer Acked-by
#7 lore raw

Hi Arnd

Am 16.03.22 um 21:59 schrieb Arnd Bergmann:

Quoted 40 lines · Thomas Zimmermann
Quoted 14 lines

Am 16.03.22 um 20:12 schrieb Thomas Zimmermann:

Quoted 13 lines

drivers/gpu/drm/Kconfig menuconfig DRM · line 33

Quoted 9 lines

Adding a dependency in all drivers that select DRM_MIPI_DBI avoids the problem for now, adding the dependency in DRM_MIPI_DBI as well should help make it easier to figure out why it breaks if someone forgets the dependency the next time.

tristate

drivers/gpu/drm/Kconfig +1 -1
-    depends on DRM
+    depends on DRM_KMS_HELPER

This symbol cannot be selected by users, so it's maybe not a good idea to depend on it. In fact, I've had to remove such a statement because it created a cyclic dependency. [1]

I tried to explain above what I was thinking here: the added dependency is both a correct statement (DRM_MIPI_DBI depends on DRM_KMS_HELPER because it cannot be built without DRM_KMS_HELPER) and helpful as an indication what went wrong if we run into the same problem with a new driver, instead of the cryptic link failure you get something like

WARNING: unmet direct dependencies detected for DRM_MIPI_DBI
Depends on [m]: HAS_IOMEM [=y] && DRM_KMS_HELPER [=m] Selected by [y]:
- DRM_PANEL_WIDECHIPS_WS2401 [=y] && HAS_IOMEM [=y] && DRM [=y] &&
DRM_PANEL [=y] && SPI [=y] && GPIOLIB [=y] && BACKLIGHT_CLASS_DEVICE [=y]
Selected by [m]:
- TINYDRM_ILI9225 [=m] && HAS_IOMEM [=y] && DRM [=y] && SPI [=y]

Quoted 2 lines

[1]
https://lore.kernel.org/dri-devel/[email protected]/

I was going for 'depends on' in the panel drivers because I saw the same being done for other panel drivers, and mixing the two methods causes dependency loops. I looked again now, and find that 'select DRM_KMS_HELPER' is more common for other drivers, and makes sense here because it is generally not user-selectable.

Quoted 4 lines

The easiest replacement for my patch would then be to just use 'select DRM_KMS_HELPER' from CONFIG_DRM_MIPI_DBI, which makes it
safer and more consistent with your change. If you like, I'll send an updated version.

MIPI DBI is another helper and select is not transitive IIRC. So drivers would still have to select KMS helpers as well. (?)

I just added my patch to drm-misc-fixes today and it should show up in upstream soon. Maybe just rebase your patch on top of it and if nothing breaks let's merge it as-is including the 'depends on'. You can add

in this case.

More generally, I think you're right about making DRM helper libraries using 'depends on' to link to other libraries. Drivers would at least know which config symbols to select. A number of config rules would have to be adapted to make that happen, I guess.

One issue is that different submodules of DRM seem to use different logic for expressing such config dependencies. That's been an endless source of problems so far.

Quoted 6 lines

One thing I'm not sure about is whether there is still use for ever having CONFIG_DRM without CONFIG_DRM_KMS_HELPER if it gets selected by almost every driver anyway. Is this actually a configuration that users rely on, or should we just remove the symbol completely and build the KMS helpers unconditionally?

Best leave it as it is. i915 doesn't use it. And since it's a helper, it should not be lumped together with core DRM code simply for reasons of design.

For DRM_KMS_HELPER itself, the mid-term plan is to move some of the code into other modules. KMS helpers used to contain all kind of helpers, but recently there's interest in reducing the minimum size of a built-in DRM with minimal driver support. So the non-essential stuff needs to go into modules for the more-sophisticated DRM drivers.

Best regards
Thomas

Quoted 2 lines

Arnd

Signature · 6 lines
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Ivo Totev
#8 lore raw
Quoted 9 lines · Thomas Zimmermann

Am 16.03.22 um 21:59 schrieb Arnd Bergmann:

Quoted 8 lines · Thomas Zimmermann

I was going for 'depends on' in the panel drivers because I saw the same being done for other panel drivers, and mixing the two methods causes dependency loops. I looked again now, and find that 'select DRM_KMS_HELPER' is more common for other drivers, and makes sense here because it is generally not user-selectable.

Quoted 7 lines
Quoted 4 lines

The easiest replacement for my patch would then be to just use 'select DRM_KMS_HELPER' from CONFIG_DRM_MIPI_DBI, which makes it
safer and more consistent with your change. If you like, I'll send an updated version.

MIPI DBI is another helper and select is not transitive IIRC. So drivers would still have to select KMS helpers as well. (?)

Not sure what you mean here: if a driver selects DRM_MIPI_DBI, and DRM_MIPI_DBI selects DRM_KMS_HELPER, the leaf driver
does not need to select DRM_KMS_HELPER because it is already selected. This is one of the major problems of overusing 'select' because you end up unable to turn things off.

Maybe you are thinking of the case where DRM_MIPI_DBI depends on DRM_KMS_HELPER, and something selects DRM_MIPI_DBI. In this case, the dependency does /not/ get inherited by the leaf driver, it needs a copy of the dependency or it triggers a warning, which is what my patch intended.

Quoted 4 lines

More generally, I think you're right about making DRM helper libraries using 'depends on' to link to other libraries. Drivers would at least know which config symbols to select. A number of config rules would have to be adapted to make that happen, I guess.

Generally speaking, a problem with DRM is that it uses way too much 'select' to enforce other subsystems to be enabled, this is what causes DRM to have more problems with incorrect or circular dependencies, and the only way to avoid that is to be consistent about the dependencies: each symbol should only be referenced with either 'select' or 'depends on' but not both, and 'select' should ideally only be used on hidden symbols.

Quoted 9 lines
Quoted 5 lines

One thing I'm not sure about is whether there is still use for ever having CONFIG_DRM without CONFIG_DRM_KMS_HELPER if it gets selected by almost every driver anyway. Is this actually a configuration that users rely on, or should we just remove the symbol completely and build the KMS helpers unconditionally?

Best leave it as it is. i915 doesn't use it. And since it's a helper, it should not be lumped together with core DRM code simply for reasons of design.

Ok

Quoted 5 lines

For DRM_KMS_HELPER itself, the mid-term plan is to move some of the code into other modules. KMS helpers used to contain all kind of helpers, but recently there's interest in reducing the minimum size of a built-in DRM with minimal driver support. So the non-essential stuff needs to go into modules for the more-sophisticated DRM drivers.

Right, that makes sense.

Arnd