On 10/26/21 11:19 PM, Arnd Bergmann wrote:
On Wed, Oct 27, 2021 at 2:58 AM Randy Dunlap <rdunlap@xxxxxxxxxxxxx> wrote:
Looking at this config, there is:
CONFIG_DRM=y
# CONFIG_DRM_DP_AUX_CHARDEV is not set
# CONFIG_DRM_DEBUG_MM is not set
# CONFIG_DRM_DEBUG_SELFTEST is not set
CONFIG_DRM_KMS_HELPER=m
# CONFIG_DRM_LOAD_EDID_FIRMWARE is not set
# CONFIG_DRM_DP_CEC is not set
CONFIG_DRM_GEM_CMA_HELPER=y
CONFIG_DRM_KMS_CMA_HELPER=y
GEM_CMA_HELPER depends on KMS_HELPER, but the latter is a module. That's probably the cause of the problem. Is it intentionally set this way?
The only drivers that select DRM_KMS_HELPER are both =m, so that's how
DRM_KMS_HELPER is set also.
Symbol: DRM_KMS_HELPER [=m]
Type : tristate
Defined at drivers/gpu/drm/Kconfig:82
Depends on: HAS_IOMEM [=y] && DRM [=y]
Selected by [m]:
- DRM_ATMEL_HLCDC [=m] && HAS_IOMEM [=y] && DRM [=y] && OF [=y] && COMMON_CLK [=y] && MFD_ATMEL_HLCDC [=y] && ARM [=y]
- DRM_ASPEED_GFX [=m] && HAS_IOMEM [=y] && DRM [=y] && OF [=y] && (COMPILE_TEST [=n] || ARCH_ASPEED [=y]) && MMU [=y]
I did the ARM cross-build and also see the linker error.
I didn't understand why -- and still don't, but here is a little
speculation:
In the past (e.g. 10 years ago), we have to move some .o files
in lib/ from lib-y to obj-y so that they would always be included
in the final object file and not cause their user/caller object
files to suffer from undefined references.
These happened because unused functions(?) in lib-y files are
stripped out of the final object files.
The same thing could be happening here (still just guessing).
Does that help any? I dunno.
Adding Arnd to Cc: to see if he has any ideas...
From all I can tell, the problem is that DRM_KMS_CMA_HELPER is a 'bool'
symbol, so if it gets selected by a '=m' driver, it turns into '=y', which
then selects DRM_GEM_CMA_HELPER=y, but that one cannot link
against DRM_KMS_HELPER=m code.
This trivial change makes it all build:
diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig
index c08860db2520..699f434ce813 100644
--- a/drivers/gpu/drm/Kconfig
+++ b/drivers/gpu/drm/Kconfig
@@ -218,7 +218,7 @@ config DRM_GEM_CMA_HELPER
Choose this if you need the GEM CMA helper functions
config DRM_KMS_CMA_HELPER
- bool
+ tristate
depends on DRM
select DRM_GEM_CMA_HELPER
help
but this needs some more testing to make sure it doesn't add
any other regressions.
Interestingly, I never hit the problem in randconfig testing since
there is always some '=y' driver that selects DRM_KMS_HELPER.
Arnd
Hi Arnd,
Thomas had posted a patch that also fixes the problem.
Somehow I missed seeing that the first time around.
Thanks.
--
~Randy
