[PATCH v4] drm/xe: Allow building as kernel built-in

[Lucas De Marchi]

On Wed, May 21, 2025 at 08:51:24PM +0000, Harry Austen wrote:
>On Wed May 21, 2025 at 4:11 PM BST, Lucas De Marchi wrote:
>> On Wed, May 21, 2025 at 08:35:05AM +0000, Harry Austen wrote:
>> >On Mon May 19, 2025 at 4:14 PM BST, Maarten Lankhorst wrote:
>> >> Hey,
>> >>
>> >> On 2025-05-16 12:42, Harry Austen wrote:
>> >>> Fix Kconfig symbol dependency on KUNIT, which isn't actually required
>> >>> for XE to be built-in. However, if KUNIT is enabled, it must be built-in
>> >>> too.
>> >>>
>> >>> Also, allow DRM_XE_DISPLAY to be built-in. But only as long as DRM_I915
>> >>> isn't, since that results in duplicate symbol errors.
>> >>>
>> >>> Fixes: 08987a8b6820 ("drm/xe: Fix build with KUNIT=m")
>> >>> Cc: Lucas De Marchi <lucas.demarchi at intel.com>
>> >>> Cc: Thomas Hellström <thomas.hellstrom at linux.intel.com>
>> >>> Cc: Jani Nikula <jani.nikula at linux.intel.com>
>> >>> Signed-off-by: Harry Austen <hpausten at protonmail.com>
>> >>> Acked-by: Jani Nikula <jani.nikula at intel.com>
>> >>> ---
>> >>> v4: Add Jani Nikula's Acked-by tag
>> >>> v3: Simplify KUNIT dependency, as suggested by Jani Nikula
>> >>> v2: Ensure DRM_XE_DISPLAY and DRM_I915 can't both be built-in
>> >>>
>> >>>  drivers/gpu/drm/xe/Kconfig | 5 +++--
>> >>>  1 file changed, 3 insertions(+), 2 deletions(-)
>> >>>
>> >>> diff --git a/drivers/gpu/drm/xe/Kconfig b/drivers/gpu/drm/xe/Kconfig
>> >>> index 9bce047901b22..214f40264fa12 100644
>> >>> --- a/drivers/gpu/drm/xe/Kconfig
>> >>> +++ b/drivers/gpu/drm/xe/Kconfig
>> >>> @@ -1,7 +1,8 @@
>> >>>  # SPDX-License-Identifier: GPL-2.0-only
>> >>>  config DRM_XE
>> >>>  	tristate "Intel Xe Graphics"
>> >>> -	depends on DRM && PCI && MMU && (m || (y && KUNIT=y))
>> >>> +	depends on DRM && PCI && MMU
>> >>> +	depends on KUNIT || KUNIT=n

this part looks good and is more related to the title of this commit.
This doesn't need to wait the display thing below.

>> >>>  	select INTERVAL_TREE
>> >>>  	# we need shmfs for the swappable backing store, and in particular
>> >>>  	# the shmem_readpage() which depends upon tmpfs
>> >>> @@ -51,7 +52,7 @@ config DRM_XE
>> >>>
>> >>>  config DRM_XE_DISPLAY
>> >>>  	bool "Enable display support"
>> >>> -	depends on DRM_XE && DRM_XE=m && HAS_IOPORT
>> >>> +	depends on DRM_XE && (DRM_XE=m || DRM_I915!=y) && HAS_IOPORT
>> >>>  	select FB_IOMEM_HELPERS if DRM_FBDEV_EMULATION
>> >>>  	select I2C
>> >>>  	select I2C_ALGOBIT
>> >> Reviewed-by: Maarten Lankhorst <maarten.lankhorst at linux.intel.com>
>> >
>> >Thanks!
>> >
>> >>
>> >> Can we also get rid of the IOSF_MBI select? Not even xe_display depends on it, leftover from initial porting. :)
>> >
>> >As this seems unrelated, I'd be happy to submit a separate patch for this
>> >afterwards. Thanks for the suggestion.
>>
>> same comment as in the previous version of this patch. What I don't like
>> is that it doesn't prevent you to do a bad selection. Instead if you set
>> DRM_XE=y and DRM_I915=y, you lose DRM_XE_DISPLAY. Silently. This is
>> equivalent of chosing the options via make menuconfig:
>>
>> 	# initial state
>> 	./scripts/config --state CONFIG_DRM_XE --state CONFIG_DRM_XE_DISPLAY --state CONFIG_DRM_I915
>> 	m
>> 	y
>> 	m
>>
>> 	# set CONFIG_DRM_XE to built-in
>> 	./scripts/config -e CONFIG_DRM_XE && make -s olddefconfig
>> 	./scripts/config --state CONFIG_DRM_XE --state CONFIG_DRM_XE_DISPLAY --state CONFIG_DRM_I915
>> 	y
>> 	y
>> 	m
>>
>> 	# set CONFIG_DRM_I915 to built-in, we lose display with xe
>> 	$ ~/p/linux-dim/drm-xe-next/scripts/config --state CONFIG_DRM_XE --state CONFIG_DRM_XE_DISPLAY --state CONFIG_DRM_I915
>> 	y
>> 	undef
>> 	y
>>
>> I'm not sure how to prevent setting CONFIG_DRM_I915=y if
>> CONFIG_DRM_XE_DISPLAY is selected.
>
>Ah interesting, thanks for the example. I'll try a few different things and see
>if I can get more sensible behaviour.

I think a reasonable compromise (since it's a temporary thing until we
spinoff the display part) is to move the dependency up, on drm_xe.
This way we don't silently lose display.

If we try to set drm_xe to y (and drm_i915==y), kconfig will complain.
If we try to set drm_i915 to y (and drm_xe==y), we will get drm_xe==m.
Which is better than losing display.


thanks
Lucas De Marchi

>
>>
>> Lucas De Marchi
>>
>> >
>> >Harry
>> >
>
>