Re: [RFC v2 2/4] drm/i915/display: add generic to_intel_display() macro

[Rodrigo Vivi <rodrigo.vivi@intel.com>]

On Thu, Mar 07, 2024 at 01:28:57PM +0200, Jani Nikula wrote:
> On Wed, 06 Mar 2024, Rodrigo Vivi <rodrigo.vivi@intel.com> wrote:
> > On Wed, Mar 06, 2024 at 02:24:36PM +0200, Jani Nikula wrote:
> >> Convert various pointers to struct intel_display * using _Generic().
> >> 
> >> Add some macro magic to make adding new conversions easier, and somewhat
> >> abstract the need to cast each generic association. The cast is required
> >> because all associations needs to compile, regardless of the type and
> >> the generic selection.
> >> 
> >> The use of *p in the generic selection assignment expression removes the
> >> need to add separate associations for const pointers.
> >> 
> >> Signed-off-by: Jani Nikula <jani.nikula@intel.com>
> >> ---
> >>  .../drm/i915/display/intel_display_types.h    | 46 +++++++++++++++++++
> >>  1 file changed, 46 insertions(+)
> >> 
> >> diff --git a/drivers/gpu/drm/i915/display/intel_display_types.h b/drivers/gpu/drm/i915/display/intel_display_types.h
> >> index e67cd5b02e84..3f63a5a74d77 100644
> >> --- a/drivers/gpu/drm/i915/display/intel_display_types.h
> >> +++ b/drivers/gpu/drm/i915/display/intel_display_types.h
> >> @@ -2183,4 +2183,50 @@ static inline int to_bpp_x16(int bpp)
> >>  	return bpp << 4;
> >>  }
> >>  
> >> +/*
> >> + * Conversion functions/macros from various pointer types to struct
> >> + * intel_display pointer.
> >> + */
> >> +static inline struct intel_display *__drm_device_to_intel_display(const struct drm_device *drm)
> >> +{
> >> +	/*
> >> +	 * Assume there's a pointer to struct intel_display in memory right
> >> +	 * after struct drm_device.
> >> +	 */
> >> +	struct intel_display **p = (struct intel_display **)(drm + 1);
> >
> > at least this patch and the first one should be together to help the
> > 'magic' to be understood and confirmed safe.
> 
> Yes. I just kept them separate while still juggling the whole thing.
> 
> And it occurs to me we could make *this* the first patch in the series,
> by making the above function:
> 
> static inline struct intel_display *__drm_device_to_intel_display(const struct drm_device *drm)
> {
> 	return &to_i915(drm)->display;
> }
> 
> Then we could only add the struct drm_device *drm backpointer in struct
> intel_display from patch 1, and proceed with patches 3-4, avoiding the
> whole magic thing for starters. It would unblock a whole lot of
> refactoring as the first step.

sounds like a good plan

> 
> >
> >> +
> >> +	return *p;
> >> +}
> >> +
> >> +#define __intel_connector_to_intel_display(p)		\
> >> +	__drm_device_to_intel_display((p)->base.dev)
> >> +#define __intel_crtc_to_intel_display(p)		\
> >> +	__drm_device_to_intel_display((p)->base.dev)
> >> +#define __intel_crtc_state_to_intel_display(p)			\
> >> +	__drm_device_to_intel_display((p)->uapi.crtc->dev)
> >> +#define __intel_digital_port_to_intel_display(p)		\
> >> +	__drm_device_to_intel_display((p)->base.base.dev)
> >> +#define __intel_encoder_to_intel_display(p)		\
> >> +	__drm_device_to_intel_display((p)->base.dev)
> >> +#define __intel_hdmi_to_intel_display(p)	\
> >> +	__drm_device_to_intel_display(hdmi_to_dig_port(p)->base.base.dev)
> >> +#define __intel_dp_to_intel_display(p)	\
> >> +	__drm_device_to_intel_display(dp_to_dig_port(p)->base.base.dev)
> >> +
> >> +/* Helper for generic association. Map types to conversion functions/macros. */
> >> +#define __assoc(type, p) \
> >> +	struct type: __##type##_to_intel_display((struct type *)(p))
> >> +
> >> +/* Convert various pointer types to struct intel_display pointer. */
> >> +#define to_intel_display(p)				\
> >
> > For a moment I almost complained of this because of the generic magic,
> > but mostly because I had guc related code in mind where you can never
> > find the definition, but here it is different. the used 'to_intel_display'
> > can easily be searched by cscope/ctags and then you are able to see
> > below what are the accepted cases, so I ended up liking it.
> 
> Yay!
> 
> I also tried to put effort into making this easily extensible. Add a
> __<FROM-STRUCT>_to_intel_display(p) macro or function, and
> __assoc(<FROM-STRUCT>, p) in the association list below, and it just
> works.

as long as we don't lose the ability to use cscope/ctags to find
these definitions I would be okay.

right now a cscope on intel_crtc would show the _assoc(intel_crtc...
but if we lose that we kind of go to the dark side of the macro
indirections.

> 
> BR,
> Jani.
> 
> >
> >> +	_Generic(*p,					\
> >> +		 __assoc(intel_connector, p),		\
> >> +		 __assoc(intel_crtc, p),		\
> >> +		 __assoc(intel_crtc_state, p),		\
> >> +		 __assoc(intel_digital_port, p),	\
> >> +		 __assoc(intel_encoder, p),		\
> >> +		 __assoc(intel_hdmi, p),		\
> >> +		 __assoc(intel_dp, p),			\
> >> +		 __assoc(drm_device, p))
> >> +
> >>  #endif /*  __INTEL_DISPLAY_TYPES_H__ */
> >> -- 
> >> 2.39.2
> >> 
> 
> -- 
> Jani Nikula, Intel