On Thu, 13 Nov 2025, Ville Syrjälä <[email protected]> wrote:
> On Thu, Nov 13, 2025 at 03:37:50PM +0200, Jani Nikula wrote:
>> The sole user of the remaining functions in intel_gmch.[ch] is in
>> display. Move them under display.
>> 
>> This allows us to remove the compat soc/intel_gmch.h from xe.
>> 
>> Signed-off-by: Jani Nikula <[email protected]>
>> ---
>>  drivers/gpu/drm/i915/Makefile                           | 5 +----
>>  drivers/gpu/drm/i915/{soc => display}/intel_gmch.c      | 3 +--
>>  drivers/gpu/drm/i915/{soc => display}/intel_gmch.h      | 0
>>  drivers/gpu/drm/i915/display/intel_vga.c                | 3 +--
>>  drivers/gpu/drm/xe/compat-i915-headers/soc/intel_gmch.h | 6 ------
>>  5 files changed, 3 insertions(+), 14 deletions(-)
>>  rename drivers/gpu/drm/i915/{soc => display}/intel_gmch.c (95%)
>>  rename drivers/gpu/drm/i915/{soc => display}/intel_gmch.h (100%)
>>  delete mode 100644 drivers/gpu/drm/xe/compat-i915-headers/soc/intel_gmch.h
>> 
>> diff --git a/drivers/gpu/drm/i915/Makefile b/drivers/gpu/drm/i915/Makefile
>> index 90588d5bb908..98822c98d960 100644
>> --- a/drivers/gpu/drm/i915/Makefile
>> +++ b/drivers/gpu/drm/i915/Makefile
>> @@ -57,10 +57,6 @@ i915-y += \
>>      vlv_iosf_sb.o \
>>      vlv_suspend.o
>>  
>> -# core peripheral code
>> -i915-y += \
>> -    soc/intel_gmch.o
>> -
>>  # core library code
>>  i915-y += \
>>      i915_memcpy.o \
>> @@ -280,6 +276,7 @@ i915-y += \
>>      display/intel_flipq.o \
>>      display/intel_frontbuffer.o \
>>      display/intel_global_state.o \
>> +    display/intel_gmch.o \
>
> I think I'd just stick the stuff into intel_vga.c since 
> that's where the vgaarb registration is as well.

Moving this directly to intel_vga.c requires a bunch of in-flight
modifications, because currently it's not compiled for xe, and there's a
dummy implementation of intel_gmch_vga_set_decode(). Or I have to wrap
#ifdef I915 around it or something.

See "[PATCH] drm/xe: use the same vga decode code as i915". If that
regresses, it's trivial to revert that, but if I do that as part of
moving the function to intel_vga.c, it's all mixed up.

Having this in intel_vga.c is the pleasing end result, but none of the
paths there are pleasing. Pick your poison, I guess.


BR,
Jani.


-- 
Jani Nikula, Intel

Reply via email to