[PATCH 40/41] drm/amd/display: Migrate color manager HW and fix MCM blend LUT issues

<[email protected]> Fri, 31 Jul 2026 17:13:01 -0400
Newsgroups org.freedesktop.lists.amd-gfx
Message-ID <[email protected]>
From: Rafal Ostrowski <[email protected]>

Combine two related color manager improvements:

1. Migrate MCM/RMCM, shaper and 3DLUT HW programming to dc_plane_cm
   - Continue convergence with upstream Color Manager refactor
   - Migrate MCM/RMCM, shaper and 3DLUT programming for dcn401+ to dc_plane_cm
   - Unify MCM programming across both legacy and 3DLUT DMA paths
   - Add 3DLUT DMA broadcast support for multi-pipe planes

2. Reprogram MCM blend LUT on blend_enable falling edge
   - Fix issue where SDR desktop remained desaturated after video playback
   - Detect blend_enable falling edge (1->0) in surface update detection
   - Force MCM LUT reprogramming on blend state change

Reviewed-by: Dillon Varone <[email protected]>
Signed-off-by: Rafal Ostrowski <[email protected]>
Signed-off-by: Roman Li <[email protected]>
---
 drivers/gpu/drm/amd/display/dc/core/dc.c      |  66 +--
 .../drm/amd/display/dc/core/dc_hw_sequencer.c |  18 +
 .../gpu/drm/amd/display/dc/core/dc_stream.c   |   3 +-
 drivers/gpu/drm/amd/display/dc/dc.h           |  12 +-
 .../amd/display/dc/hubp/dcn401/dcn401_hubp.c  | 201 +++++---
 .../amd/display/dc/hubp/dcn401/dcn401_hubp.h  |  23 +-
 .../amd/display/dc/hubp/dcn42/dcn42_hubp.c    |  78 ++-
 .../amd/display/dc/hubp/dcn42/dcn42_hubp.h    |  10 +-
 .../amd/display/dc/hubp/dcn50/dcn50_hubp.c    |  10 +-
 .../amd/display/dc/hubp/dcn60/dcn60_hubp.c    |   7 +-
 .../amd/display/dc/hwss/dcn401/dcn401_hwseq.c | 428 ++++++---------
 .../amd/display/dc/hwss/dcn401/dcn401_hwseq.h |   7 +-
 .../amd/display/dc/hwss/dcn401/dcn401_init.c  |   1 -
 .../amd/display/dc/hwss/dcn42/dcn42_hwseq.c   | 488 ++----------------
 .../amd/display/dc/hwss/dcn42/dcn42_hwseq.h   |   6 -
 .../amd/display/dc/hwss/dcn42/dcn42_init.c    |   2 -
 .../amd/display/dc/hwss/dcn60/dcn60_init.c    |   2 -
 .../drm/amd/display/dc/hwss/hw_sequencer.h    |   2 +-
 .../display/dc/hwss/hw_sequencer_private.h    |   5 -
 drivers/gpu/drm/amd/display/dc/inc/hw/hubp.h  |  35 +-
 drivers/gpu/drm/amd/display/dc/inc/hw/mpc.h   |  81 +--
 .../amd/display/dc/mpc/dcn401/dcn401_mpc.c    | 177 ++++---
 .../amd/display/dc/mpc/dcn401/dcn401_mpc.h    |  25 +-
 .../drm/amd/display/dc/mpc/dcn42/dcn42_mpc.c  | 405 ++-------------
 .../drm/amd/display/dc/mpc/dcn42/dcn42_mpc.h  |  55 +-
 .../drm/amd/display/dc/mpc/dcn60/dcn60_mpc.c  | 235 ++++-----
 .../drm/amd/display/dc/mpc/dcn60/dcn60_mpc.h  |  15 -
 .../dc/resource/dcn42/dcn42_resource.c        |   1 +
 .../dc/resource/dcn60/dcn60_resource.c        |   1 +
 29 files changed, 778 insertions(+), 1621 deletions(-)

diff --git a/drivers/gpu/drm/amd/display/dc/core/dc.c b/drivers/gpu/drm/amd/display/dc/core/dc.c
index 7f378b5e896c..68f228014305 100644
--- a/drivers/gpu/drm/amd/display/dc/core/dc.c
+++ b/drivers/gpu/drm/amd/display/dc/core/dc.c
@@ -3027,13 +3027,14 @@ static struct dc_update_descriptor det_surface_update(
 		elevate_update_type(&overall_type, UPDATE_TYPE_FAST, LOCK_DESCRIPTOR_STREAM);
 	}
 
-	if ((u->cm && u->cm->flags.bits.blend_enable) ||
+	if ((u->cm && (u->cm->flags.bits.blend_enable ||
+			u->cm->flags.bits.blend_enable != u->surface->cm.flags.bits.blend_enable)) ||
 			(u->gamma && dce_use_lut(u->plane_info ? u->plane_info->format : u->surface->format))) {
 		update_bits->gamma_change = 1;
 		elevate_update_type(&overall_type, UPDATE_TYPE_FAST, LOCK_DESCRIPTOR_STREAM);
 	}
 
-	if (u->cm && (u->cm->flags.bits.lut3d_enable || u->cm->flags.bits.shaper_enable)) {
+	if (u->cm && (u->cm->flags.bits.lut3d_enable || u->surface->cm.flags.bits.lut3d_enable)) {
 		update_bits->lut_3d = 1;
 		elevate_update_type(&overall_type, UPDATE_TYPE_FAST, LOCK_DESCRIPTOR_STREAM);
 	}
@@ -3075,25 +3076,12 @@ static struct dc_update_descriptor det_surface_update(
 			}
 		};
 
-		if (u->cm->flags.bits.shaper_enable != u->surface->cm.flags.bits.shaper_enable
-				|| u->cm->flags.bits.blend_enable != u->surface->cm.flags.bits.blend_enable
-				|| u->cm->flags.bits.lut3d_enable != u->surface->cm.flags.bits.lut3d_enable
-				|| u->cm->flags.bits.lut3d_dma_enable != u->surface->cm.flags.bits.lut3d_dma_enable) {
-			update_bits->mcm_transfer_function_enable_change = 1;
-			elevate_update_type(&overall_type, UPDATE_TYPE_FULL, LOCK_DESCRIPTOR_GLOBAL);
-		}
-
 		if ((u->cm->flags.all != blend_only_flags.all && u->cm->flags.all != 0) ||
 				(u->surface->cm.flags.all != blend_only_flags.all && u->surface->cm.flags.all != 0)) {
 			elevate_update_type(&overall_type, UPDATE_TYPE_FULL, LOCK_DESCRIPTOR_GLOBAL);
 		}
 	}
 
-	if (update_bits->lut_3d &&
-			!u->surface->cm.flags.bits.lut3d_dma_enable) {
-		elevate_update_type(&overall_type, UPDATE_TYPE_FULL, LOCK_DESCRIPTOR_GLOBAL);
-	}
-
 	if (check_config->enable_legacy_fast_update &&
 			(update_bits->gamma_change ||
 			update_bits->gamut_remap_change ||
@@ -3446,46 +3434,6 @@ static void copy_surface_update_to_plane(
 		memcpy(&surface->cm, srf_update->cm, sizeof(surface->cm));
 		surface->cm.refcount = refcount;
 
-#ifndef TRIM_CM2
-		/* Populate mcm_luts from cm for legacy consumers (dml2, hwseq) */
-		surface->mcm_luts.lut1d_func = &surface->cm.blend_func;
-		surface->mcm_luts.shaper = &surface->cm.shaper_func;
-		if (srf_update->cm->flags.bits.lut3d_dma_enable) {
-			surface->mcm_luts.lut3d_data.lut3d_src = DC_CM2_TRANSFER_FUNC_SOURCE_VIDMEM;
-			surface->mcm_luts.lut3d_data.gpu_mem_params.addr = surface->cm.lut3d_dma.addr;
-			surface->mcm_luts.lut3d_data.gpu_mem_params.layout =
-				(surface->cm.lut3d_dma.swizzle == CM_LUT_3D_SWIZZLE_LINEAR_RGB) ?
-					DC_CM2_GPU_MEM_LAYOUT_3D_SWIZZLE_LINEAR_RGB :
-				(surface->cm.lut3d_dma.swizzle == CM_LUT_3D_SWIZZLE_LINEAR_BGR) ?
-					DC_CM2_GPU_MEM_LAYOUT_3D_SWIZZLE_LINEAR_BGR :
-					DC_CM2_GPU_MEM_LAYOUT_1D_PACKED_LINEAR;
-			surface->mcm_luts.lut3d_data.gpu_mem_params.format_params.format =
-				(surface->cm.lut3d_dma.format == CM_LUT_PIXEL_FORMAT_RGBA16161616_UNORM_12MSB) ?
-					DC_CM2_GPU_MEM_FORMAT_16161616_UNORM_12MSB :
-				(surface->cm.lut3d_dma.format == CM_LUT_PIXEL_FORMAT_RGBA16161616_UNORM_12LSB) ?
-					DC_CM2_GPU_MEM_FORMAT_16161616_UNORM_12LSB :
-					DC_CM2_GPU_MEM_FORMAT_16161616_FLOAT_FP1_5_10;
-			surface->mcm_luts.lut3d_data.gpu_mem_params.format_params.float_params.bias =
-				surface->cm.lut3d_dma.bias;
-			surface->mcm_luts.lut3d_data.gpu_mem_params.format_params.float_params.scale =
-				surface->cm.lut3d_dma.scale;
-			surface->mcm_luts.lut3d_data.gpu_mem_params.component_order =
-				DC_CM2_GPU_MEM_PIXEL_COMPONENT_ORDER_RGBA;
-			surface->mcm_luts.lut3d_data.gpu_mem_params.size = DC_CM2_GPU_MEM_SIZE_TRANSFORMED;
-			surface->mcm_luts.lut3d_data.mpc_3dlut_enable = (srf_update->cm->flags.bits.lut3d_enable != 0);
-		} else {
-			surface->mcm_luts.lut3d_data.lut3d_src = DC_CM2_TRANSFER_FUNC_SOURCE_SYSMEM;
-			surface->mcm_luts.lut3d_data.lut3d_func = &surface->cm.lut3d_func;
-		}
-
-		if (srf_update->cm->flags.bits.shaper_enable &&
-				srf_update->cm->flags.bits.lut3d_enable)
-			surface->mcm_shaper_3dlut_setting = DC_CM2_SHAPER_3DLUT_SETTING_ENABLE_SHAPER_3DLUT;
-		else if (srf_update->cm->flags.bits.shaper_enable)
-			surface->mcm_shaper_3dlut_setting = DC_CM2_SHAPER_3DLUT_SETTING_ENABLE_SHAPER;
-		else
-			surface->mcm_shaper_3dlut_setting = DC_CM2_SHAPER_3DLUT_SETTING_BYPASS_ALL;
-#endif /* TRIM_CM2 */
 	}
 
 	if (srf_update->hdr_mult.value)
@@ -3496,12 +3444,6 @@ static void copy_surface_update_to_plane(
 		surface->sdr_white_level_nits =
 				srf_update->sdr_white_level_nits;
 
-	if (srf_update->cm &&
-			(srf_update->cm->flags.bits.blend_enable ||
-			srf_update->cm->flags.bits.shaper_enable ||
-			srf_update->cm->flags.bits.lut3d_enable))
-		surface->lut_bank_a = !surface->lut_bank_a;
-
 	if (srf_update->input_csc_color_matrix)
 		surface->input_csc_color_matrix =
 			*srf_update->input_csc_color_matrix;
@@ -5243,7 +5185,7 @@ static void commit_planes_for_stream(struct dc *dc,
 						srf_updates[i].cm->flags.bits.lut3d_enable &&
 						srf_updates[i].cm->flags.bits.lut3d_dma_enable &&
 						dc->hwss.trigger_3dlut_dma_load)
-					dc->hwss.trigger_3dlut_dma_load(dc, pipe_ctx);
+					dc->hwss.trigger_3dlut_dma_load(pipe_ctx);
 
 				/*program triple buffer after lock based on flip type*/
 				if (dc->hwss.program_triplebuffer != NULL && dc->debug.enable_tri_buf) {
diff --git a/drivers/gpu/drm/amd/display/dc/core/dc_hw_sequencer.c b/drivers/gpu/drm/amd/display/dc/core/dc_hw_sequencer.c
index 8878652434d2..76e3509f73c5 100644
--- a/drivers/gpu/drm/amd/display/dc/core/dc_hw_sequencer.c
+++ b/drivers/gpu/drm/amd/display/dc/core/dc_hw_sequencer.c
@@ -1457,6 +1457,24 @@ void hwss_build_fast_sequence(struct dc *dc,
 					(*num_steps)++;
 				}
 
+				if (current_mpc_pipe->plane_state->update_bits.hdr_mult) {
+					struct fixed31_32 multiplier = current_mpc_pipe->plane_state->hdr_mult;
+					uint32_t hw_mult = 0x1f000; // 1.0 default multiplier
+					struct custom_float_format fmt;
+
+					fmt.exponenta_bits = 6;
+					fmt.mantissa_bits = 12;
+					fmt.sign = true;
+
+					if (!dc_fixpt_eq(multiplier, dc_fixpt_from_int(0)) && // check != 0
+						convert_to_custom_float_format(multiplier, &fmt, &hw_mult)) {
+						block_sequence[*num_steps].params.dpp_set_hdr_multiplier_params.dpp = current_mpc_pipe->plane_res.dpp;
+						block_sequence[*num_steps].params.dpp_set_hdr_multiplier_params.hw_mult = hw_mult;
+						block_sequence[*num_steps].func = DPP_SET_HDR_MULTIPLIER;
+						(*num_steps)++;
+					}
+				}
+
 				if (dc->hwss.program_gamut_remap &&
 						(current_mpc_pipe->plane_state->update_bits.gamut_remap_change ||
 						 current_mpc_pipe->stream->update_flags.bits.gamut_remap)) {
diff --git a/drivers/gpu/drm/amd/display/dc/core/dc_stream.c b/drivers/gpu/drm/amd/display/dc/core/dc_stream.c
index ce5b9e82db8d..9f3068e899e1 100644
--- a/drivers/gpu/drm/amd/display/dc/core/dc_stream.c
+++ b/drivers/gpu/drm/amd/display/dc/core/dc_stream.c
@@ -991,6 +991,7 @@ void dc_stream_log(const struct dc *dc, const struct dc_stream_state *stream)
 	}
 }
 
+/* TODO - move to per plane ownership? */
 /*
 *	dc_stream_get_3dlut()
 *	Requirements:
@@ -1042,7 +1043,6 @@ void dc_stream_release_3dlut_for_stream(
 	if (rmcm_3dlut) {
 		rmcm_3dlut->isInUse = false;
 		rmcm_3dlut->stream  = NULL;
-		rmcm_3dlut->protection_bits = 0;
 	}
 }
 
@@ -1054,7 +1054,6 @@ void dc_stream_init_rmcm_3dlut(struct dc *dc)
 	for (unsigned int i = 0; i < num_rmcm; i++) {
 		dc->res_pool->rmcm_3dlut[i].isInUse = false;
 		dc->res_pool->rmcm_3dlut[i].stream = NULL;
-		dc->res_pool->rmcm_3dlut[i].protection_bits = 0;
 	}
 }
 
diff --git a/drivers/gpu/drm/amd/display/dc/dc.h b/drivers/gpu/drm/amd/display/dc/dc.h
index 683b5fb1945d..b2fd3a94f067 100644
--- a/drivers/gpu/drm/amd/display/dc/dc.h
+++ b/drivers/gpu/drm/amd/display/dc/dc.h
@@ -1459,6 +1459,7 @@ struct dc_transfer_func {
 	enum dc_transfer_func_predefined tf;
 	/* FP16 1.0 reference level in nits, default is 80 nits, only for PQ*/
 	uint32_t sdr_ref_white_level;
+	struct fixed31_32 hdr_multiplier;
 	union {
 		struct pwl_params pwl;
 		struct dc_transfer_func_distributed_points tf_pts;
@@ -1509,13 +1510,11 @@ struct lut_mem_mapping {
 struct dc_rmcm_3dlut {
 	bool isInUse;
 	const struct dc_stream_state *stream;
-	uint8_t protection_bits;
 };
 
 struct dc_3dlut {
 	struct kref refcount;
 	struct tetrahedral_params lut_3d;
-	struct fixed31_32 hdr_multiplier;
 	union dc_3dlut_state state;
 };
 
@@ -1595,7 +1594,6 @@ struct pipe_update_bits {
 	uint32_t stereo_format_change:1;
 	uint32_t lut_3d:1;
 	uint32_t tmz_changed:1;
-	uint32_t mcm_transfer_function_enable_change:1; /* disable or enable MCM transfer func */
 	uint32_t full_update:1;
 	uint32_t sdr_white_level_nits:1;
 	uint32_t cm_hist_change:1;
@@ -1640,7 +1638,6 @@ static inline void dc_pipe_update_bits_set_full(struct pipe_update_bits *flags)
 	flags->stereo_format_change = 1;
 	flags->lut_3d = 1;
 	flags->tmz_changed = 1;
-	flags->mcm_transfer_function_enable_change = 1;
 	flags->full_update = 1;
 	flags->sdr_white_level_nits = 1;
 	flags->cm_hist_change = 1;
@@ -1674,7 +1671,6 @@ static inline bool dc_pipe_update_bits_is_any_set(const struct pipe_update_bits
 		flags->stereo_format_change ||
 		flags->lut_3d ||
 		flags->tmz_changed ||
-		flags->mcm_transfer_function_enable_change ||
 		flags->full_update ||
 		flags->sdr_white_level_nits ||
 		flags->cm_hist_change;
@@ -1709,18 +1705,16 @@ struct dc_plane_state {
 	enum dc_color_space color_space;
 
 #ifndef TRIM_CM2
-	// TODO: No longer used, remove
+	bool lut_bank_a;
 	struct dc_hdr_static_metadata hdr_static_ctx;
-
 	struct dc_3dlut lut3d_func;
 	struct dc_transfer_func in_shaper_func;
 	struct dc_transfer_func blend_tf;
 	enum dc_cm2_shaper_3dlut_setting mcm_shaper_3dlut_setting;
 	bool mcm_lut1d_enable;
 	struct dc_cm2_func_luts mcm_luts;
-#endif /* TRIM_CM2 */
-	bool lut_bank_a;
 	enum mpcc_movable_cm_location mcm_location;
+#endif /* TRIM_CM2 */
 	struct dc_plane_cm cm;
 
 	struct dc_transfer_func *gamcor_tf;
diff --git a/drivers/gpu/drm/amd/display/dc/hubp/dcn401/dcn401_hubp.c b/drivers/gpu/drm/amd/display/dc/hubp/dcn401/dcn401_hubp.c
index 9965cf572354..a2e0bafb711a 100644
--- a/drivers/gpu/drm/amd/display/dc/hubp/dcn401/dcn401_hubp.c
+++ b/drivers/gpu/drm/amd/display/dc/hubp/dcn401/dcn401_hubp.c
@@ -41,12 +41,12 @@
 	hubp2->hubp_shift->field_name, hubp2->hubp_mask->field_name
 
 void hubp401_program_3dlut_fl_addr(struct hubp *hubp,
-	const struct dc_plane_address address)
+	const struct dc_plane_address *address)
 {
 	struct dcn20_hubp *hubp2 = TO_DCN20_HUBP(hubp);
 
-	REG_UPDATE(HUBP_3DLUT_ADDRESS_HIGH, HUBP_3DLUT_ADDRESS_HIGH, address.lut3d.addr.high_part);
-	REG_WRITE(HUBP_3DLUT_ADDRESS_LOW, address.lut3d.addr.low_part);
+	REG_UPDATE(HUBP_3DLUT_ADDRESS_HIGH, HUBP_3DLUT_ADDRESS_HIGH, address->lut3d.addr.high_part);
+	REG_WRITE(HUBP_3DLUT_ADDRESS_LOW, address->lut3d.addr.low_part);
 }
 
 void hubp401_program_3dlut_fl_dlg_param(struct hubp *hubp, int refcyc_per_3dlut_group)
@@ -72,33 +72,46 @@ uint32_t hubp401_get_3dlut_fl_done(struct hubp *hubp)
 	return ret;
 }
 
-void hubp401_program_3dlut_fl_addressing_mode(struct hubp *hubp, enum hubp_3dlut_fl_addressing_mode addr_mode)
+static void hubp401_get_3dlut_fl_xbar_map(
+		const enum dc_cm_lut_pixel_format format,
+		enum hubp_3dlut_fl_crossbar_bit_slice *bit_slice_y_g,
+		enum hubp_3dlut_fl_crossbar_bit_slice *bit_slice_cb_b,
+		enum hubp_3dlut_fl_crossbar_bit_slice *bit_slice_cr_r)
 {
-	struct dcn20_hubp *hubp2 = TO_DCN20_HUBP(hubp);
-
-	REG_UPDATE(HUBP_3DLUT_CONTROL, HUBP_3DLUT_ADDRESSING_MODE, addr_mode);
-}
-
-void hubp401_program_3dlut_fl_width(struct hubp *hubp, enum hubp_3dlut_fl_width width)
-{
-	struct dcn20_hubp *hubp2 = TO_DCN20_HUBP(hubp);
-
-	REG_UPDATE(HUBP_3DLUT_CONTROL, HUBP_3DLUT_WIDTH, width);
+	switch (format) {
+	case CM_LUT_PIXEL_FORMAT_BGRA16161616_UNORM_12MSB:
+	case CM_LUT_PIXEL_FORMAT_BGRA16161616_UNORM_12LSB:
+	case CM_LUT_PIXEL_FORMAT_BGRA16161616_FLOAT_FP1_5_10:
+		/* BGRA */
+		*bit_slice_cr_r = hubp_3dlut_fl_crossbar_bit_slice_32_47;
+		*bit_slice_y_g = hubp_3dlut_fl_crossbar_bit_slice_16_31;
+		*bit_slice_cb_b =  hubp_3dlut_fl_crossbar_bit_slice_0_15;
+		break;
+	case CM_LUT_PIXEL_FORMAT_RGBA16161616_UNORM_12MSB:
+	case CM_LUT_PIXEL_FORMAT_RGBA16161616_UNORM_12LSB:
+	case CM_LUT_PIXEL_FORMAT_RGBA16161616_FLOAT_FP1_5_10:
+	default:
+		/* RGBA */
+		*bit_slice_cr_r = hubp_3dlut_fl_crossbar_bit_slice_0_15;
+		*bit_slice_y_g = hubp_3dlut_fl_crossbar_bit_slice_16_31;
+		*bit_slice_cb_b = hubp_3dlut_fl_crossbar_bit_slice_32_47;
+		break;
+	}
 }
 
-void hubp401_program_3dlut_fl_tmz_protected(struct hubp *hubp, uint8_t protection_bits)
+void hubp401_program_3dlut_fl_crossbar(struct hubp *hubp,
+		const enum dc_cm_lut_pixel_format format)
 {
 	struct dcn20_hubp *hubp2 = TO_DCN20_HUBP(hubp);
 
-	REG_UPDATE(HUBP_3DLUT_CONTROL, HUBP_3DLUT_TMZ, protection_bits);
-}
+	enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_y_g = 0;
+	enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_cb_b = 0;
+	enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_cr_r = 0;
 
-void hubp401_program_3dlut_fl_crossbar(struct hubp *hubp,
-			enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_y_g,
-			enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_cb_b,
-			enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_cr_r)
-{
-	struct dcn20_hubp *hubp2 = TO_DCN20_HUBP(hubp);
+	hubp401_get_3dlut_fl_xbar_map(format,
+			&bit_slice_y_g,
+			&bit_slice_cb_b,
+			&bit_slice_cr_r);
 
 	REG_UPDATE_3(HUBP_3DLUT_CONTROL,
 			HUBP_3DLUT_CROSSBAR_SELECT_Y_G, bit_slice_y_g,
@@ -106,62 +119,122 @@ void hubp401_program_3dlut_fl_crossbar(struct hubp *hubp,
 			HUBP_3DLUT_CROSSBAR_SELECT_CR_R, bit_slice_cr_r);
 }
 
-void hubp401_update_3dlut_fl_bias_scale(struct hubp *hubp, uint16_t bias, uint16_t scale)
+static enum hubp_3dlut_fl_width hubp401_get_3dlut_fl_width(
+		const enum dc_cm_lut_size size,
+		const enum dc_cm_lut_swizzle swizzle)
 {
-	struct dcn20_hubp *hubp2 = TO_DCN20_HUBP(hubp);
+	enum hubp_3dlut_fl_width width = 0;
+
+	switch (size) {
+	case CM_LUT_SIZE_333333:
+		ASSERT(swizzle != CM_LUT_1D_PACKED_LINEAR);
+		width = hubp_3dlut_fl_width_33;
+		break;
+	case CM_LUT_SIZE_171717:
+		if (swizzle != CM_LUT_1D_PACKED_LINEAR) {
+			width = hubp_3dlut_fl_width_17;
+		} else {
+			width = hubp_3dlut_fl_width_17_transformed;
+		}
+		break;
+	default:
+		width = 0;
+		break;
+	}
 
-	REG_UPDATE_2(_3DLUT_FL_BIAS_SCALE, HUBP0_3DLUT_FL_BIAS, bias, HUBP0_3DLUT_FL_SCALE, scale);
+	return width;
 }
 
-void hubp401_program_3dlut_fl_mode(struct hubp *hubp, enum hubp_3dlut_fl_mode mode)
+static enum hubp_3dlut_fl_format hubp401_get_3dlut_fl_format(
+		const enum dc_cm_lut_pixel_format dc_format)
 {
-	struct dcn20_hubp *hubp2 = TO_DCN20_HUBP(hubp);
+	enum hubp_3dlut_fl_format hubp_format = hubp_3dlut_fl_format_unorm_12msb_bitslice;
 
-	REG_UPDATE(_3DLUT_FL_CONFIG, HUBP0_3DLUT_FL_MODE, mode);
+	switch (dc_format) {
+	case CM_LUT_PIXEL_FORMAT_RGBA16161616_UNORM_12MSB:
+	case CM_LUT_PIXEL_FORMAT_BGRA16161616_UNORM_12MSB:
+		hubp_format = hubp_3dlut_fl_format_unorm_12msb_bitslice;
+		break;
+	case CM_LUT_PIXEL_FORMAT_RGBA16161616_UNORM_12LSB:
+	case CM_LUT_PIXEL_FORMAT_BGRA16161616_UNORM_12LSB:
+		hubp_format = hubp_3dlut_fl_format_unorm_12lsb_bitslice;
+		break;
+	case CM_LUT_PIXEL_FORMAT_RGBA16161616_FLOAT_FP1_5_10:
+	case CM_LUT_PIXEL_FORMAT_BGRA16161616_FLOAT_FP1_5_10:
+		hubp_format = hubp_3dlut_fl_format_float_fp1_5_10;
+		break;
+	default:
+		BREAK_TO_DEBUGGER();
+		break;
+	}
+
+	return hubp_format;
 }
 
-void hubp401_program_3dlut_fl_format(struct hubp *hubp, enum hubp_3dlut_fl_format format)
+static enum hubp_3dlut_fl_addressing_mode hubp401_get_3dlut_fl_addr_mode(
+		const enum dc_cm_lut_swizzle swizzle)
 {
-	struct dcn20_hubp *hubp2 = TO_DCN20_HUBP(hubp);
+	enum hubp_3dlut_fl_addressing_mode addr_mode;
+
+	switch (swizzle) {
+	case CM_LUT_1D_PACKED_LINEAR:
+		addr_mode = hubp_3dlut_fl_addressing_mode_simple_linear;
+		break;
+	case CM_LUT_3D_SWIZZLE_LINEAR_RGB:
+	case CM_LUT_3D_SWIZZLE_LINEAR_BGR:
+	default:
+		addr_mode = hubp_3dlut_fl_addressing_mode_sw_linear;
+		break;
+	}
 
-	REG_UPDATE(_3DLUT_FL_CONFIG, HUBP0_3DLUT_FL_FORMAT, format);
+	return addr_mode;
 }
 
-void hubp401_program_3dlut_fl_config(
-	struct hubp *hubp,
-	struct hubp_fl_3dlut_config *cfg)
+static enum hubp_3dlut_fl_mode hubp401_get_3dlut_fl_mode(
+		const enum dc_cm_lut_swizzle swizzle)
 {
-	struct dcn20_hubp *hubp2 = TO_DCN20_HUBP(hubp);
+	enum hubp_3dlut_fl_mode mode;
 
-	uint32_t mpc_width = {(cfg->width == 17) ? 0 : 1};
-	uint32_t width = {cfg->width};
+	switch (swizzle) {
+	case CM_LUT_3D_SWIZZLE_LINEAR_RGB:
+		mode = hubp_3dlut_fl_mode_native_1;
+		break;
+	case CM_LUT_3D_SWIZZLE_LINEAR_BGR:
+		mode = hubp_3dlut_fl_mode_native_2;
+		break;
+	case CM_LUT_1D_PACKED_LINEAR:
+		mode = hubp_3dlut_fl_mode_transform;
+		break;
+	default:
+		mode = hubp_3dlut_fl_mode_disable;
+		break;
+	}
 
-	if (cfg->layout == CM_LUT_1D_PACKED_LINEAR)
-		width = (cfg->width == 17) ? 4916 : 35940;
+	return mode;
+}
+
+void hubp401_program_3dlut_fl_config(struct hubp *hubp,
+		const struct dc_3dlut_dma *config)
+{
+	struct dcn20_hubp *hubp2 = TO_DCN20_HUBP(hubp);
+
+	enum hubp_3dlut_fl_width width = hubp401_get_3dlut_fl_width(config->size, config->swizzle);
+	enum hubp_3dlut_fl_format format = hubp401_get_3dlut_fl_format(config->format);
+	enum hubp_3dlut_fl_addressing_mode addr_mode = hubp401_get_3dlut_fl_addr_mode(config->swizzle);
+	enum hubp_3dlut_fl_mode mode = hubp401_get_3dlut_fl_mode(config->swizzle);
 
 	REG_UPDATE_2(_3DLUT_FL_CONFIG,
-		HUBP0_3DLUT_FL_MODE, cfg->mode,
-		HUBP0_3DLUT_FL_FORMAT, cfg->format);
+			HUBP0_3DLUT_FL_MODE, mode,
+			HUBP0_3DLUT_FL_FORMAT, format);
 
 	REG_UPDATE_2(_3DLUT_FL_BIAS_SCALE,
-		HUBP0_3DLUT_FL_BIAS, cfg->bias,
-		HUBP0_3DLUT_FL_SCALE, cfg->scale);
-
-	REG_UPDATE(HUBP_3DLUT_ADDRESS_HIGH,
-		HUBP_3DLUT_ADDRESS_HIGH, cfg->address.lut3d.addr.high_part);
-	REG_UPDATE(HUBP_3DLUT_ADDRESS_LOW,
-		HUBP_3DLUT_ADDRESS_LOW, cfg->address.lut3d.addr.low_part);
-
-	//cross bar
-	REG_UPDATE_8(HUBP_3DLUT_CONTROL,
-		HUBP_3DLUT_MPC_WIDTH, mpc_width,
-		HUBP_3DLUT_WIDTH, width,
-		HUBP_3DLUT_CROSSBAR_SELECT_CR_R, cfg->crossbar_bit_slice_cr_r,
-		HUBP_3DLUT_CROSSBAR_SELECT_Y_G, cfg->crossbar_bit_slice_y_g,
-		HUBP_3DLUT_CROSSBAR_SELECT_CB_B, cfg->crossbar_bit_slice_cb_b,
-		HUBP_3DLUT_ADDRESSING_MODE, cfg->addr_mode,
-		HUBP_3DLUT_TMZ, cfg->protection_bits,
-		HUBP_3DLUT_ENABLE, cfg->enabled ? 1 : 0);
+			HUBP0_3DLUT_FL_BIAS, config->bias,
+			HUBP0_3DLUT_FL_SCALE, config->scale);
+
+	REG_UPDATE_3(HUBP_3DLUT_CONTROL,
+			HUBP_3DLUT_WIDTH, width,
+			HUBP_3DLUT_ADDRESSING_MODE, addr_mode,
+			HUBP_3DLUT_TMZ, config->addr.tmz_surface);
 }
 
 void hubp401_update_mall_sel(struct hubp *hubp, uint32_t mall_sel, bool c_cursor)
@@ -1062,19 +1135,13 @@ static struct hubp_funcs dcn401_hubp_funcs = {
 	.hubp_update_mall_sel = hubp401_update_mall_sel,
 	.hubp_prepare_subvp_buffering = hubp32_prepare_subvp_buffering,
 	.hubp_program_mcache_id_and_split_coordinate = hubp401_program_mcache_id_and_split_coordinate,
-	.hubp_update_3dlut_fl_bias_scale = hubp401_update_3dlut_fl_bias_scale,
-	.hubp_program_3dlut_fl_mode = hubp401_program_3dlut_fl_mode,
-	.hubp_program_3dlut_fl_format = hubp401_program_3dlut_fl_format,
 	.hubp_program_3dlut_fl_addr = hubp401_program_3dlut_fl_addr,
+	.hubp_program_3dlut_fl_config = hubp401_program_3dlut_fl_config,
 	.hubp_program_3dlut_fl_dlg_param = hubp401_program_3dlut_fl_dlg_param,
 	.hubp_enable_3dlut_fl = hubp401_enable_3dlut_fl,
-	.hubp_program_3dlut_fl_addressing_mode = hubp401_program_3dlut_fl_addressing_mode,
-	.hubp_program_3dlut_fl_width = hubp401_program_3dlut_fl_width,
-	.hubp_program_3dlut_fl_tmz_protected = hubp401_program_3dlut_fl_tmz_protected,
 	.hubp_program_3dlut_fl_crossbar = hubp401_program_3dlut_fl_crossbar,
 	.hubp_get_3dlut_fl_done = hubp401_get_3dlut_fl_done,
 	.hubp_clear_tiling = hubp401_clear_tiling,
-	.hubp_program_3dlut_fl_config = hubp401_program_3dlut_fl_config,
 	.hubp_read_reg_state = hubp3_read_reg_state
 };
 
diff --git a/drivers/gpu/drm/amd/display/dc/hubp/dcn401/dcn401_hubp.h b/drivers/gpu/drm/amd/display/dc/hubp/dcn401/dcn401_hubp.h
index 4116834c552d..adf343837652 100644
--- a/drivers/gpu/drm/amd/display/dc/hubp/dcn401/dcn401_hubp.h
+++ b/drivers/gpu/drm/amd/display/dc/hubp/dcn401/dcn401_hubp.h
@@ -328,32 +328,17 @@ uint32_t hubp401_get_3dlut_fl_done(struct hubp *hubp);
 
 void hubp401_set_unbounded_requesting(struct hubp *hubp, bool enable);
 
-void hubp401_update_3dlut_fl_bias_scale(struct hubp *hubp, uint16_t bias, uint16_t scale);
-
 void hubp401_program_3dlut_fl_crossbar(struct hubp *hubp,
-	enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_y_g,
-	enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_cb_b,
-	enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_cr_r);
-
-void hubp401_program_3dlut_fl_tmz_protected(struct hubp *hubp, uint8_t protection_bits);
-
-void hubp401_program_3dlut_fl_width(struct hubp *hubp, enum hubp_3dlut_fl_width width);
-
-void hubp401_program_3dlut_fl_addressing_mode(struct hubp *hubp, enum hubp_3dlut_fl_addressing_mode addr_mode);
+		const enum dc_cm_lut_pixel_format format);
 
 void hubp401_enable_3dlut_fl(struct hubp *hubp, bool enable);
 
 void hubp401_program_3dlut_fl_dlg_param(struct hubp *hubp, int refcyc_per_3dlut_group);
 
-void hubp401_program_3dlut_fl_addr(struct hubp *hubp, const struct dc_plane_address address);
+void hubp401_program_3dlut_fl_addr(struct hubp *hubp, const struct dc_plane_address *address);
 
-void hubp401_program_3dlut_fl_format(struct hubp *hubp, enum hubp_3dlut_fl_format format);
-
-void hubp401_program_3dlut_fl_mode(struct hubp *hubp, enum hubp_3dlut_fl_mode mode);
-
-void hubp401_program_3dlut_fl_config(
-	struct hubp *hubp,
-	struct hubp_fl_3dlut_config *cfg);
+void hubp401_program_3dlut_fl_config(struct hubp *hubp,
+		const struct dc_3dlut_dma *config);
 
 void hubp401_clear_tiling(struct hubp *hubp);
 
diff --git a/drivers/gpu/drm/amd/display/dc/hubp/dcn42/dcn42_hubp.c b/drivers/gpu/drm/amd/display/dc/hubp/dcn42/dcn42_hubp.c
index 94730f3cff42..86e49eb84a54 100644
--- a/drivers/gpu/drm/amd/display/dc/hubp/dcn42/dcn42_hubp.c
+++ b/drivers/gpu/drm/amd/display/dc/hubp/dcn42/dcn42_hubp.c
@@ -317,19 +317,84 @@ static void hubp42_program_surface_config(
 	hubp42_program_pixel_format(hubp, format);
 }
 
+static void hubp42_get_3dlut_fl_xbar_map(
+		const enum dc_cm_lut_pixel_format format,
+		enum hubp_3dlut_fl_crossbar_bit_slice *bit_slice_y_g,
+		enum hubp_3dlut_fl_crossbar_bit_slice *bit_slice_cb_b,
+		enum hubp_3dlut_fl_crossbar_bit_slice *bit_slice_cr_r)
+{
+	switch (format) {
+	case CM_LUT_PIXEL_FORMAT_BGRA16161616_UNORM_12MSB:
+	case CM_LUT_PIXEL_FORMAT_BGRA16161616_UNORM_12LSB:
+	case CM_LUT_PIXEL_FORMAT_BGRA16161616_FLOAT_FP1_5_10:
+		/* BGRA */
+		*bit_slice_cr_r = hubp_3dlut_fl_crossbar_bit_slice_32_47;
+		*bit_slice_y_g = hubp_3dlut_fl_crossbar_bit_slice_16_31;
+		*bit_slice_cb_b =  hubp_3dlut_fl_crossbar_bit_slice_0_15;
+		break;
+	case CM_LUT_PIXEL_FORMAT_RGBA16161616_UNORM_12MSB:
+	case CM_LUT_PIXEL_FORMAT_RGBA16161616_UNORM_12LSB:
+	case CM_LUT_PIXEL_FORMAT_RGBA16161616_FLOAT_FP1_5_10:
+	default:
+		/* RGBA */
+		*bit_slice_cr_r = hubp_3dlut_fl_crossbar_bit_slice_0_15;
+		*bit_slice_y_g = hubp_3dlut_fl_crossbar_bit_slice_16_31;
+		*bit_slice_cb_b = hubp_3dlut_fl_crossbar_bit_slice_32_47;
+		break;
+	}
+}
+
 void hubp42_program_3dlut_fl_crossbar(struct hubp *hubp,
-	enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_r,
-	enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_g,
-	enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_b)
+		const enum dc_cm_lut_pixel_format format)
 {
 	struct dcn20_hubp *hubp2 = TO_DCN20_HUBP(hubp);
 
+	enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_g = 0;
+	enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_b = 0;
+	enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_r = 0;
+
+	hubp42_get_3dlut_fl_xbar_map(format,
+			&bit_slice_g,
+			&bit_slice_b,
+			&bit_slice_r);
+
 	REG_UPDATE_3(HUBP_3DLUT_CONTROL,
 		HUBP_3DLUT_CROSSBAR_SEL_R, bit_slice_r,
 		HUBP_3DLUT_CROSSBAR_SEL_G, bit_slice_g,
 		HUBP_3DLUT_CROSSBAR_SEL_B, bit_slice_b);
 }
 
+static uint32_t hubp42_get_3dlut_fl_mpc_width(
+		const enum dc_cm_lut_size size)
+{
+	uint32_t width = 0;
+
+	switch (size) {
+	case CM_LUT_SIZE_333333:
+		width = 1;
+		break;
+	case CM_LUT_SIZE_171717:
+	default:
+		width = 0;
+		break;
+	}
+
+	return width;
+}
+
+void hubp42_program_3dlut_fl_config(struct hubp *hubp,
+		const struct dc_3dlut_dma *config)
+{
+	struct dcn20_hubp *hubp2 = TO_DCN20_HUBP(hubp);
+
+	uint32_t mpc_width = hubp42_get_3dlut_fl_mpc_width(config->size);
+
+	REG_UPDATE(HUBP_3DLUT_CONTROL,
+			HUBP_3DLUT_MPC_WIDTH, mpc_width);
+
+	hubp401_program_3dlut_fl_config(hubp, config);
+}
+
 static bool hubp42_program_surface_flip_and_addr(
 	struct hubp *hubp,
 	const struct dc_plane_address *address,
@@ -660,15 +725,10 @@ struct hubp_funcs dcn42_hubp_funcs = {
 	.hubp_set_flip_int = hubp1_set_flip_int,
 	.hubp_in_blank = hubp1_in_blank,
 	.program_extended_blank = hubp31_program_extended_blank_value,
-	.hubp_update_3dlut_fl_bias_scale = hubp401_update_3dlut_fl_bias_scale,
-	.hubp_program_3dlut_fl_mode = hubp401_program_3dlut_fl_mode,
-	.hubp_program_3dlut_fl_format = hubp401_program_3dlut_fl_format,
 	.hubp_program_3dlut_fl_addr = hubp401_program_3dlut_fl_addr,
+	.hubp_program_3dlut_fl_config = hubp42_program_3dlut_fl_config,
 	.hubp_program_3dlut_fl_dlg_param = hubp401_program_3dlut_fl_dlg_param,
 	.hubp_enable_3dlut_fl = hubp401_enable_3dlut_fl,
-	.hubp_program_3dlut_fl_addressing_mode = hubp401_program_3dlut_fl_addressing_mode,
-	.hubp_program_3dlut_fl_width = hubp401_program_3dlut_fl_width,
-	.hubp_program_3dlut_fl_tmz_protected = hubp401_program_3dlut_fl_tmz_protected,
 	.hubp_program_3dlut_fl_crossbar = hubp42_program_3dlut_fl_crossbar,
 	.hubp_get_3dlut_fl_done = hubp401_get_3dlut_fl_done,
 	.hubp_clear_tiling = hubp3_clear_tiling,
diff --git a/drivers/gpu/drm/amd/display/dc/hubp/dcn42/dcn42_hubp.h b/drivers/gpu/drm/amd/display/dc/hubp/dcn42/dcn42_hubp.h
index ba4a58c353dd..3a030e5e517e 100644
--- a/drivers/gpu/drm/amd/display/dc/hubp/dcn42/dcn42_hubp.h
+++ b/drivers/gpu/drm/amd/display/dc/hubp/dcn42/dcn42_hubp.h
@@ -79,11 +79,11 @@ bool hubp42_construct(
 	const struct dcn_hubp2_shift *hubp_shift,
 	const struct dcn_hubp2_mask *hubp_mask);
 
-void hubp42_program_3dlut_fl_crossbar(
-	struct hubp *hubp,
-	enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_r,
-	enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_g,
-	enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_b);
+void hubp42_program_3dlut_fl_crossbar(struct hubp *hubp,
+		const enum dc_cm_lut_pixel_format format);
+
+void hubp42_program_3dlut_fl_config(struct hubp *hubp,
+		const struct dc_3dlut_dma *config);
 
 void hubp42_read_state(struct hubp *hubp);
 
diff --git a/drivers/gpu/drm/amd/display/dc/hubp/dcn50/dcn50_hubp.c b/drivers/gpu/drm/amd/display/dc/hubp/dcn50/dcn50_hubp.c
index e704ce5cd0a0..0230f03590cf 100644
--- a/drivers/gpu/drm/amd/display/dc/hubp/dcn50/dcn50_hubp.c
+++ b/drivers/gpu/drm/amd/display/dc/hubp/dcn50/dcn50_hubp.c
@@ -7,6 +7,7 @@
 #include "reg_helper.h"
 #include "basics/conversion.h"
 #include "dcn50_hubp.h"
+#include "dcn42/dcn42_hubp.h"
 
 #define REG(reg)\
 	hubp2->hubp_regs->reg
@@ -500,16 +501,11 @@ static struct hubp_funcs dcn50_hubp_funcs = {
 	.hubp_update_mall_sel = hubp401_update_mall_sel,
 	.hubp_prepare_subvp_buffering = hubp32_prepare_subvp_buffering,
 	.hubp_program_mcache_id_and_split_coordinate = hubp401_program_mcache_id_and_split_coordinate,
-	.hubp_update_3dlut_fl_bias_scale = hubp401_update_3dlut_fl_bias_scale,
-	.hubp_program_3dlut_fl_mode = hubp401_program_3dlut_fl_mode,
-	.hubp_program_3dlut_fl_format = hubp401_program_3dlut_fl_format,
 	.hubp_program_3dlut_fl_addr = hubp401_program_3dlut_fl_addr,
+	.hubp_program_3dlut_fl_config = hubp42_program_3dlut_fl_config,
 	.hubp_program_3dlut_fl_dlg_param = hubp401_program_3dlut_fl_dlg_param,
 	.hubp_enable_3dlut_fl = hubp401_enable_3dlut_fl,
-	.hubp_program_3dlut_fl_addressing_mode = hubp401_program_3dlut_fl_addressing_mode,
-	.hubp_program_3dlut_fl_width = hubp401_program_3dlut_fl_width,
-	.hubp_program_3dlut_fl_tmz_protected = hubp401_program_3dlut_fl_tmz_protected,
-	.hubp_program_3dlut_fl_crossbar = hubp401_program_3dlut_fl_crossbar,
+	.hubp_program_3dlut_fl_crossbar = hubp42_program_3dlut_fl_crossbar,
 	.hubp_get_3dlut_fl_done = hubp401_get_3dlut_fl_done,
 	.hubp_clear_tiling = hubp401_clear_tiling,
 	.hubp_read_reg_state = hubp3_read_reg_state
diff --git a/drivers/gpu/drm/amd/display/dc/hubp/dcn60/dcn60_hubp.c b/drivers/gpu/drm/amd/display/dc/hubp/dcn60/dcn60_hubp.c
index 1065303f1d72..d22fe26aae62 100644
--- a/drivers/gpu/drm/amd/display/dc/hubp/dcn60/dcn60_hubp.c
+++ b/drivers/gpu/drm/amd/display/dc/hubp/dcn60/dcn60_hubp.c
@@ -441,15 +441,10 @@ static struct hubp_funcs dcn60_hubp_funcs = {
 	.hubp_update_mall_sel = NULL,
 	.hubp_prepare_subvp_buffering = hubp32_prepare_subvp_buffering,
 	.hubp_program_mcache_id_and_split_coordinate = hubp401_program_mcache_id_and_split_coordinate,
-	.hubp_update_3dlut_fl_bias_scale = hubp401_update_3dlut_fl_bias_scale,
-	.hubp_program_3dlut_fl_mode = hubp401_program_3dlut_fl_mode,
-	.hubp_program_3dlut_fl_format = hubp401_program_3dlut_fl_format,
 	.hubp_program_3dlut_fl_addr = hubp401_program_3dlut_fl_addr,
+	.hubp_program_3dlut_fl_config = hubp42_program_3dlut_fl_config,
 	.hubp_program_3dlut_fl_dlg_param = hubp401_program_3dlut_fl_dlg_param,
 	.hubp_enable_3dlut_fl = hubp401_enable_3dlut_fl,
-	.hubp_program_3dlut_fl_addressing_mode = hubp401_program_3dlut_fl_addressing_mode,
-	.hubp_program_3dlut_fl_width = hubp401_program_3dlut_fl_width,
-	.hubp_program_3dlut_fl_tmz_protected = hubp401_program_3dlut_fl_tmz_protected,
 	.hubp_program_3dlut_fl_crossbar = hubp42_program_3dlut_fl_crossbar,
 	.hubp_get_3dlut_fl_done = hubp401_get_3dlut_fl_done,
 	.hubp_clear_tiling = hubp401_clear_tiling,
diff --git a/drivers/gpu/drm/amd/display/dc/hwss/dcn401/dcn401_hwseq.c b/drivers/gpu/drm/amd/display/dc/hwss/dcn401/dcn401_hwseq.c
index 904780e02f70..51596f445a09 100644
--- a/drivers/gpu/drm/amd/display/dc/hwss/dcn401/dcn401_hwseq.c
+++ b/drivers/gpu/drm/amd/display/dc/hwss/dcn401/dcn401_hwseq.c
@@ -99,10 +99,6 @@ void dcn401_program_gamut_remap(struct program_gamut_remap_params *params)
 	unsigned int i = 0;
 	struct mpc_grph_gamut_adjustment mpc_adjust;
 
-	//For now assert if location is not pre-blend
-	if (plane)
-		ASSERT(plane->mcm_location == MPCC_MOVABLE_CM_LOCATION_BEFORE);
-
 	// program MPCC_MCM_FIRST_GAMUT_REMAP
 	memset(&mpc_adjust, 0, sizeof(mpc_adjust));
 	mpc_adjust.gamut_adjust_type = GRAPHICS_GAMUT_ADJUST_TYPE_BYPASS;
@@ -413,289 +409,200 @@ void dcn401_init_hw(struct dc *dc)
 	}
 }
 
-static void dcn401_get_mcm_lut_xable_from_pipe_ctx(struct dc *dc, struct pipe_ctx *pipe_ctx,
-		enum MCM_LUT_XABLE *shaper_xable,
-		enum MCM_LUT_XABLE *lut3d_xable,
-		enum MCM_LUT_XABLE *lut1d_xable)
+void dcn401_trigger_3dlut_dma_load(struct pipe_ctx *pipe_ctx)
 {
-	struct mpc *mpc = dc->res_pool->mpc;
-	int mpcc_id = pipe_ctx->plane_res.hubp->inst;
-
-	if (!pipe_ctx->plane_state)
-		return;
+	const struct pipe_ctx *primary_dpp_pipe_ctx = resource_get_primary_dpp_pipe(pipe_ctx);
+	struct hubp *primary_hubp = primary_dpp_pipe_ctx ?
+			primary_dpp_pipe_ctx->plane_res.hubp : NULL;
 
-	mpc->funcs->set_movable_cm_location(mpc, MPCC_MOVABLE_CM_LOCATION_BEFORE, mpcc_id);
-	pipe_ctx->plane_state->mcm_location = MPCC_MOVABLE_CM_LOCATION_BEFORE;
-
-	*lut1d_xable = pipe_ctx->plane_state->cm.flags.bits.blend_enable ?
-		MCM_LUT_ENABLE : MCM_LUT_DISABLE;
-	*shaper_xable = pipe_ctx->plane_state->cm.flags.bits.shaper_enable ?
-		MCM_LUT_ENABLE : MCM_LUT_DISABLE;
-	*lut3d_xable = (pipe_ctx->plane_state->cm.flags.bits.shaper_enable &&
-			pipe_ctx->plane_state->cm.flags.bits.lut3d_enable) ?
-		MCM_LUT_ENABLE : MCM_LUT_DISABLE;
+	if (primary_hubp && primary_hubp->funcs->hubp_enable_3dlut_fl) {
+		primary_hubp->funcs->hubp_enable_3dlut_fl(primary_hubp, true);
+	}
 }
 
-void dcn401_populate_mcm_luts(struct dc *dc,
-		struct pipe_ctx *pipe_ctx,
-		const struct dc_plane_cm *cm,
-		bool lut_bank_a)
+bool dcn401_set_mcm_luts(struct pipe_ctx *pipe_ctx,
+				const struct dc_plane_state *plane_state)
 {
+	struct dc *dc = pipe_ctx->plane_res.hubp->ctx->dc;
+	const struct pipe_ctx *primary_dpp_pipe_ctx = resource_get_primary_dpp_pipe(pipe_ctx);
 	struct dpp *dpp_base = pipe_ctx->plane_res.dpp;
 	struct hubp *hubp = pipe_ctx->plane_res.hubp;
+	struct hubp *primary_hubp = primary_dpp_pipe_ctx ?
+			primary_dpp_pipe_ctx->plane_res.hubp : hubp;   /* fall back to current pipe */
+	const struct dc_plane_cm *cm = &plane_state->cm;
 	int mpcc_id = hubp->inst;
 	struct mpc *mpc = dc->res_pool->mpc;
 	union mcm_lut_params m_lut_params;
-	const bool lut3d_dma = !!cm->flags.bits.lut3d_dma_enable;
-	enum hubp_3dlut_fl_format format = 0;
-	enum hubp_3dlut_fl_mode mode;
-	/* Width was previously hard-coded to TRANSFORMED via local_mcm build,
-	 * preserve identical behavior.
-	 */
-	enum hubp_3dlut_fl_width width = hubp_3dlut_fl_width_transformed;
-	enum hubp_3dlut_fl_addressing_mode addr_mode;
-	enum hubp_3dlut_fl_crossbar_bit_slice crossbar_bit_slice_y_g;
-	enum hubp_3dlut_fl_crossbar_bit_slice crossbar_bit_slice_cb_b;
-	enum hubp_3dlut_fl_crossbar_bit_slice crossbar_bit_slice_cr_r;
-	enum MCM_LUT_XABLE shaper_xable = MCM_LUT_DISABLE;
-	enum MCM_LUT_XABLE lut3d_xable = MCM_LUT_DISABLE;
-	enum MCM_LUT_XABLE lut1d_xable = MCM_LUT_DISABLE;
+	struct dc_3dlut_dma lut3d_dma;
+	bool lut_enable;
+	bool lut_bank_a;
 	bool rval;
+	bool result = true;
+
+	/* decide LUT bank based on current in use */
+	mpc->funcs->get_lut_mode(mpc, MCM_LUT_1DLUT, mpcc_id, &lut_enable, &lut_bank_a);
+	if (!lut_enable) {
+		mpc->funcs->get_lut_mode(mpc, MCM_LUT_SHAPER, mpcc_id, &lut_enable, &lut_bank_a);
+	}
+	if (!lut_enable) {
+		mpc->funcs->get_lut_mode(mpc, MCM_LUT_3DLUT, mpcc_id, &lut_enable, &lut_bank_a);
+	}
 
-	dcn401_get_mcm_lut_xable_from_pipe_ctx(dc, pipe_ctx, &shaper_xable, &lut3d_xable, &lut1d_xable);
+	/* switch to the next bank */
+	if (lut_enable) {
+		lut_bank_a = !lut_bank_a;
+	}
+
+	/* MCM location fixed to pre-blend */
+	mpc->funcs->set_movable_cm_location(mpc, MPCC_MOVABLE_CM_LOCATION_BEFORE, mpcc_id);
 
 	/* 1D LUT */
-	{
-		memset(&m_lut_params, 0, sizeof(m_lut_params));
+	lut_enable = cm->flags.bits.blend_enable != 0u;
+	memset(&m_lut_params, 0, sizeof(m_lut_params));
+	if (lut_enable) {
 		if (cm->blend_func.type == TF_TYPE_HWPWL)
 			m_lut_params.pwl = &cm->blend_func.pwl;
 		else if (cm->blend_func.type == TF_TYPE_DISTRIBUTED_POINTS) {
-			rval = cm3_helper_translate_curve_to_hw_format(mpc->ctx,
+			rval = cm3_helper_translate_curve_to_hw_format(plane_state->ctx,
 					&cm->blend_func,
-					&dpp_base->regamma_params, false);
+					&dpp_base->regamma_params,
+					false);
 			m_lut_params.pwl = rval ? &dpp_base->regamma_params : NULL;
 		}
-		if (m_lut_params.pwl) {
-			if (mpc->funcs->populate_lut)
-				mpc->funcs->populate_lut(mpc, MCM_LUT_1DLUT, m_lut_params, lut_bank_a, mpcc_id);
+
+		if (!m_lut_params.pwl) {
+			lut_enable = false;
 		}
-		if (mpc->funcs->program_lut_mode)
-			mpc->funcs->program_lut_mode(mpc, MCM_LUT_1DLUT, lut1d_xable && m_lut_params.pwl, lut_bank_a, mpcc_id);
+	} else {
+		lut_enable = false;
 	}
 
+	if (mpc->funcs->program_lut_mode)
+		mpc->funcs->program_lut_mode(mpc, MCM_LUT_1DLUT, lut_enable, lut_bank_a, CM_LUT_SIZE_NONE, mpcc_id);
+	if (lut_enable && mpc->funcs->populate_lut)
+		mpc->funcs->populate_lut(mpc, MCM_LUT_1DLUT, &m_lut_params, lut_bank_a, mpcc_id);
+
 	/* Shaper */
-	if (cm->flags.bits.lut3d_enable) {
+	lut_enable = cm->flags.bits.shaper_enable != 0u;
+	if (lut_enable) {
 		memset(&m_lut_params, 0, sizeof(m_lut_params));
 		if (cm->shaper_func.type == TF_TYPE_HWPWL)
 			m_lut_params.pwl = &cm->shaper_func.pwl;
 		else if (cm->shaper_func.type == TF_TYPE_DISTRIBUTED_POINTS) {
 			ASSERT(false);
-			rval = cm3_helper_translate_curve_to_hw_format(mpc->ctx,
+			rval = cm3_helper_translate_curve_to_hw_format(plane_state->ctx,
 					&cm->shaper_func,
-					&dpp_base->regamma_params, true);
-			m_lut_params.pwl = rval ? &dpp_base->regamma_params : NULL;
-		}
-		if (m_lut_params.pwl) {
-			if (mpc->funcs->mcm.populate_lut)
-				mpc->funcs->mcm.populate_lut(mpc, m_lut_params, lut_bank_a, mpcc_id);
-			if (mpc->funcs->program_lut_mode)
-				mpc->funcs->program_lut_mode(mpc, MCM_LUT_SHAPER, MCM_LUT_ENABLE, lut_bank_a, mpcc_id);
+					&dpp_base->shaper_params,
+					true);
+			m_lut_params.pwl = rval ? &dpp_base->shaper_params : NULL;
 		}
-	}
-
-	/* 3DLUT */
-	if (!lut3d_dma) {
-		/* SYSMEM (legacy lut3d_func) */
-		memset(&m_lut_params, 0, sizeof(m_lut_params));
-		if (hubp->funcs->hubp_enable_3dlut_fl)
-			hubp->funcs->hubp_enable_3dlut_fl(hubp, false);
-
-		if (cm->lut3d_func.state.bits.initialized) {
-			m_lut_params.lut3d = &cm->lut3d_func.lut_3d;
-			if (mpc->funcs->populate_lut)
-				mpc->funcs->populate_lut(mpc, MCM_LUT_3DLUT, m_lut_params, lut_bank_a, mpcc_id);
-			if (mpc->funcs->program_lut_mode)
-				mpc->funcs->program_lut_mode(mpc, MCM_LUT_3DLUT, lut3d_xable, lut_bank_a,
-						mpcc_id);
+		if (!m_lut_params.pwl) {
+			lut_enable = false;
 		}
 	} else {
-		/* VIDMEM (3DLUT DMA Fast Load) */
+		lut_enable = false;
+	}
 
-		/* Select width based on the requested LUT size */
-		switch (cm->lut3d_dma.size) {
-		case CM_LUT_SIZE_333333:
-			if (dc->caps.color.mpc.rmcm_3d_lut_caps.lut_dim_caps.dim_33)
-				width = hubp_3dlut_fl_width_33;
-			break;
-		case CM_LUT_SIZE_171717:
-			width = hubp_3dlut_fl_width_17;
-			break;
-		default:
-			/* keep default hubp_3dlut_fl_width_transformed */
-			break;
-		}
+	if (mpc->funcs->program_lut_mode)
+		mpc->funcs->program_lut_mode(mpc, MCM_LUT_SHAPER, lut_enable, lut_bank_a, CM_LUT_SIZE_NONE, mpcc_id);
+	if (lut_enable && mpc->funcs->populate_lut)
+		mpc->funcs->populate_lut(mpc, MCM_LUT_SHAPER, &m_lut_params, lut_bank_a, mpcc_id);
 
-		//check for support
-		if (mpc->funcs->mcm.is_config_supported &&
-			!mpc->funcs->mcm.is_config_supported(width))
-			return;
+	/* NOTE: Toggling from DMA->Host is not supported atomically as hardware
+	 * blocks writes until 3DLUT FL mode is cleared from HUBP on VUpdate.
+	 * Expectation is either option is used consistently.
+	*/
 
-		if (mpc->funcs->program_lut_read_write_control)
-			mpc->funcs->program_lut_read_write_control(mpc, MCM_LUT_3DLUT, lut_bank_a, mpcc_id);
+	/* 3DLUT */
+	lut_enable = cm->flags.bits.lut3d_enable != 0u;
+	if (lut_enable && cm->flags.bits.lut3d_dma_enable) {
+		/* Fast (DMA) Load Mode */
+		/* MPC */
 		if (mpc->funcs->program_lut_mode)
-			mpc->funcs->program_lut_mode(mpc, MCM_LUT_3DLUT, lut3d_xable, lut_bank_a, mpcc_id);
+			mpc->funcs->program_lut_mode(mpc, MCM_LUT_3DLUT, lut_enable, lut_bank_a, cm->lut3d_dma.size, mpcc_id);
 
-		if (hubp->funcs->hubp_program_3dlut_fl_addr)
-			hubp->funcs->hubp_program_3dlut_fl_addr(hubp, cm->lut3d_dma.addr);
-
-		/* bit_depth was previously zero-initialized in local_mcm,
-		 * preserve identical behavior.
-		 */
-		if (mpc->funcs->mcm.program_bit_depth)
-			mpc->funcs->mcm.program_bit_depth(mpc, 0, mpcc_id);
-
-		switch (cm->lut3d_dma.swizzle) {
-		case CM_LUT_3D_SWIZZLE_LINEAR_RGB:
-			mode = hubp_3dlut_fl_mode_native_1;
-			addr_mode = hubp_3dlut_fl_addressing_mode_sw_linear;
-			break;
-		case CM_LUT_3D_SWIZZLE_LINEAR_BGR:
-			mode = hubp_3dlut_fl_mode_native_2;
-			addr_mode = hubp_3dlut_fl_addressing_mode_sw_linear;
-			break;
-		case CM_LUT_1D_PACKED_LINEAR:
-			mode = hubp_3dlut_fl_mode_transform;
-			addr_mode = hubp_3dlut_fl_addressing_mode_simple_linear;
-			break;
-		default:
-			mode = hubp_3dlut_fl_mode_disable;
-			addr_mode = hubp_3dlut_fl_addressing_mode_sw_linear;
-			break;
-		}
-		if (hubp->funcs->hubp_program_3dlut_fl_mode)
-			hubp->funcs->hubp_program_3dlut_fl_mode(hubp, mode);
+		/* only supports 12 bit */
+		if (mpc->funcs->program_lut_read_write_control)
+			mpc->funcs->program_lut_read_write_control(mpc, MCM_LUT_3DLUT, lut_bank_a, 12, mpcc_id);
 
-		if (hubp->funcs->hubp_program_3dlut_fl_addressing_mode)
-			hubp->funcs->hubp_program_3dlut_fl_addressing_mode(hubp, addr_mode);
+		if (mpc->funcs->update_3dlut_fast_load_select)
+			mpc->funcs->update_3dlut_fast_load_select(mpc, mpcc_id, primary_hubp->inst);
 
-		switch (cm->lut3d_dma.format) {
-		case CM_LUT_PIXEL_FORMAT_RGBA16161616_UNORM_12MSB:
-			format = hubp_3dlut_fl_format_unorm_12msb_bitslice;
-			break;
-		case CM_LUT_PIXEL_FORMAT_RGBA16161616_UNORM_12LSB:
-			format = hubp_3dlut_fl_format_unorm_12lsb_bitslice;
-			break;
-		case CM_LUT_PIXEL_FORMAT_RGBA16161616_FLOAT_FP1_5_10:
-			format = hubp_3dlut_fl_format_float_fp1_5_10;
-			break;
-		default:
-			break;
-		}
-		if (hubp->funcs->hubp_program_3dlut_fl_format)
-			hubp->funcs->hubp_program_3dlut_fl_format(hubp, format);
-		if (hubp->funcs->hubp_update_3dlut_fl_bias_scale &&
-				mpc->funcs->mcm.program_bias_scale) {
-			mpc->funcs->mcm.program_bias_scale(mpc,
-				cm->lut3d_dma.bias,
-				cm->lut3d_dma.scale,
-				mpcc_id);
-			hubp->funcs->hubp_update_3dlut_fl_bias_scale(hubp,
-				cm->lut3d_dma.bias,
-				cm->lut3d_dma.scale);
-		}
+		/* HUBP */
+		if (primary_hubp->inst == hubp->inst) {
+			/* only program if this is the primary dpp pipe for the given plane */
+			if (hubp->funcs->hubp_program_3dlut_fl_config)
+				hubp->funcs->hubp_program_3dlut_fl_config(hubp, &cm->lut3d_dma);
 
-		/* component_order was previously hard-coded to RGBA in local_mcm,
-		 * preserve identical behavior.
-		 */
-		crossbar_bit_slice_cr_r = hubp_3dlut_fl_crossbar_bit_slice_0_15;
-		crossbar_bit_slice_y_g = hubp_3dlut_fl_crossbar_bit_slice_16_31;
-		crossbar_bit_slice_cb_b = hubp_3dlut_fl_crossbar_bit_slice_32_47;
+			if (hubp->funcs->hubp_program_3dlut_fl_crossbar)
+				hubp->funcs->hubp_program_3dlut_fl_crossbar(hubp, cm->lut3d_dma.format);
 
-		if (hubp->funcs->hubp_program_3dlut_fl_crossbar)
-			hubp->funcs->hubp_program_3dlut_fl_crossbar(hubp,
-					crossbar_bit_slice_cr_r,
-					crossbar_bit_slice_y_g,
-					crossbar_bit_slice_cb_b);
+			if (hubp->funcs->hubp_program_3dlut_fl_addr)
+				hubp->funcs->hubp_program_3dlut_fl_addr(hubp, &cm->lut3d_dma.addr);
 
-		if (mpc->funcs->mcm.program_lut_read_write_control)
-			mpc->funcs->mcm.program_lut_read_write_control(mpc, MCM_LUT_3DLUT, lut_bank_a, true, mpcc_id);
+			if (hubp->funcs->hubp_enable_3dlut_fl) {
+				hubp->funcs->hubp_enable_3dlut_fl(hubp, true);
+			} else {
+				/* GPU memory only supports fast load path */
+				BREAK_TO_DEBUGGER();
+				lut_enable = false;
+				result = false;
+			}
+		} else {
+			/* re-trigger primary HUBP to load 3DLUT */
+			if (primary_hubp->funcs->hubp_enable_3dlut_fl) {
+				primary_hubp->funcs->hubp_enable_3dlut_fl(primary_hubp, true);
+			}
 
-		if (mpc->funcs->mcm.program_3dlut_size)
-			mpc->funcs->mcm.program_3dlut_size(mpc, width, mpcc_id);
+			/* clear FL setup on this pipe's HUBP */
+			memset(&lut3d_dma, 0, sizeof(lut3d_dma));
+			if (hubp->funcs->hubp_program_3dlut_fl_config)
+				hubp->funcs->hubp_program_3dlut_fl_config(hubp, &lut3d_dma);
 
-		if (mpc->funcs->update_3dlut_fast_load_select)
-			mpc->funcs->update_3dlut_fast_load_select(mpc, mpcc_id, hubp->inst);
+			if (hubp->funcs->hubp_enable_3dlut_fl)
+				hubp->funcs->hubp_enable_3dlut_fl(hubp, false);
+		}
+	} else {
+		/* Legacy (Host) Load Mode */
+		memset(&m_lut_params, 0, sizeof(m_lut_params));
 
-		if (hubp->funcs->hubp_enable_3dlut_fl)
-			hubp->funcs->hubp_enable_3dlut_fl(hubp, true);
-		else {
-			if (mpc->funcs->program_lut_mode) {
-				mpc->funcs->program_lut_mode(mpc, MCM_LUT_SHAPER, MCM_LUT_DISABLE, lut_bank_a, mpcc_id);
-				mpc->funcs->program_lut_mode(mpc, MCM_LUT_3DLUT, MCM_LUT_DISABLE, lut_bank_a, mpcc_id);
-				mpc->funcs->program_lut_mode(mpc, MCM_LUT_1DLUT, MCM_LUT_DISABLE, lut_bank_a, mpcc_id);
-			}
+		if (cm->flags.bits.lut3d_enable && cm->lut3d_func.state.bits.initialized) {
+			m_lut_params.lut3d = &cm->lut3d_func.lut_3d;
+		} else {
+			lut_enable = false;
 		}
-	}
-}
 
-void dcn401_trigger_3dlut_dma_load(struct dc *dc, struct pipe_ctx *pipe_ctx)
-{
-	(void)dc;
-	struct hubp *hubp = pipe_ctx->plane_res.hubp;
+		/* MPC */
+		if (mpc->funcs->program_lut_mode)
+			mpc->funcs->program_lut_mode(mpc,
+					MCM_LUT_3DLUT,
+					lut_enable,
+					lut_bank_a,
+					cm->lut3d_func.lut_3d.use_tetrahedral_9 ? CM_LUT_SIZE_999 : CM_LUT_SIZE_171717,
+					mpcc_id);
+
+		if (lut_enable) {
+			if (mpc->funcs->program_lut_read_write_control)
+				mpc->funcs->program_lut_read_write_control(mpc,
+						MCM_LUT_3DLUT,
+						lut_bank_a,
+						cm->lut3d_func.lut_3d.use_12bits ? 12 : 10,
+						mpcc_id);
 
-	if (hubp->funcs->hubp_enable_3dlut_fl) {
-		hubp->funcs->hubp_enable_3dlut_fl(hubp, true);
-	}
-}
+			if (mpc->funcs->update_3dlut_fast_load_select)
+				mpc->funcs->update_3dlut_fast_load_select(mpc, mpcc_id, 0xf);
 
-bool dcn401_set_mcm_luts(struct pipe_ctx *pipe_ctx,
-				const struct dc_plane_state *plane_state)
-{
-	struct dpp *dpp_base = pipe_ctx->plane_res.dpp;
-	int mpcc_id = pipe_ctx->plane_res.hubp->inst;
-	struct dc *dc = pipe_ctx->stream_res.opp->ctx->dc;
-	struct mpc *mpc = dc->res_pool->mpc;
-	bool result;
-	const struct pwl_params *lut_params = NULL;
-	bool rval;
+			if (mpc->funcs->populate_lut)
+				mpc->funcs->populate_lut(mpc, MCM_LUT_3DLUT, &m_lut_params, lut_bank_a, mpcc_id);
+		}
 
-	if (plane_state->cm.flags.bits.lut3d_dma_enable) {
-		dcn401_populate_mcm_luts(dc, pipe_ctx, &plane_state->cm, plane_state->lut_bank_a);
-		return true;
-	}
+		/* HUBP */
+		memset(&lut3d_dma, 0, sizeof(lut3d_dma));
+		if (hubp->funcs->hubp_program_3dlut_fl_config)
+			hubp->funcs->hubp_program_3dlut_fl_config(hubp, &lut3d_dma);
 
-	mpc->funcs->set_movable_cm_location(mpc, MPCC_MOVABLE_CM_LOCATION_BEFORE, mpcc_id);
-	pipe_ctx->plane_state->mcm_location = MPCC_MOVABLE_CM_LOCATION_BEFORE;
-	// 1D LUT
-	if (plane_state->cm.blend_func.type == TF_TYPE_HWPWL)
-		lut_params = &plane_state->cm.blend_func.pwl;
-	else if (plane_state->cm.blend_func.type == TF_TYPE_DISTRIBUTED_POINTS) {
-		rval = cm3_helper_translate_curve_to_hw_format(plane_state->ctx,
-							       &plane_state->cm.blend_func,
-							       &dpp_base->regamma_params, false);
-		lut_params = rval ? &dpp_base->regamma_params : NULL;
-	}
-	result = mpc->funcs->program_1dlut(mpc, lut_params, mpcc_id);
-	lut_params = NULL;
-
-	// Shaper
-	if (plane_state->cm.shaper_func.type == TF_TYPE_HWPWL)
-		lut_params = &plane_state->cm.shaper_func.pwl;
-	else if (plane_state->cm.shaper_func.type == TF_TYPE_DISTRIBUTED_POINTS) {
-		// TODO: dpp_base replace
-		rval = cm3_helper_translate_curve_to_hw_format(plane_state->ctx,
-							       &plane_state->cm.shaper_func,
-							       &dpp_base->shaper_params, true);
-		lut_params = rval ? &dpp_base->shaper_params : NULL;
-	}
-	result &= mpc->funcs->program_shaper(mpc, lut_params, mpcc_id);
-
-	// 3D
-	if (mpc->funcs->program_3dlut) {
-		if (plane_state->cm.lut3d_func.state.bits.initialized == 1)
-			result &= mpc->funcs->program_3dlut(mpc, &plane_state->cm.lut3d_func.lut_3d, mpcc_id);
-		else
-			result &= mpc->funcs->program_3dlut(mpc, NULL, mpcc_id);
+		if (hubp->funcs->hubp_enable_3dlut_fl)
+			hubp->funcs->hubp_enable_3dlut_fl(hubp, false);
 	}
 
 	return result;
@@ -2096,42 +2003,41 @@ void dcn401_perform_3dlut_wa_unlock(struct pipe_ctx *pipe_ctx)
 	if (!pipe_ctx)
 		return;
 
-	struct pipe_ctx *wa_pipes[MAX_PIPES] = { NULL };
-	struct pipe_ctx *odm_pipe, *mpc_pipe;
-	int i, wa_pipe_ct = 0;
+	const struct pipe_ctx *otg_master_pipe_ctx = resource_get_otg_master(pipe_ctx);
+	struct timing_generator *tg = otg_master_pipe_ctx ?
+			otg_master_pipe_ctx->stream_res.tg : NULL;
+	const struct pipe_ctx *primary_dpp_pipe_ctx = resource_is_pipe_type(pipe_ctx, DPP_PIPE) ?
+			resource_get_primary_dpp_pipe(pipe_ctx) : pipe_ctx;
+	struct hubp *primary_hubp = primary_dpp_pipe_ctx ?
+			primary_dpp_pipe_ctx->plane_res.hubp : NULL;
 
-	for (odm_pipe = pipe_ctx; odm_pipe != NULL; odm_pipe = odm_pipe->next_odm_pipe) {
-		for (mpc_pipe = odm_pipe; mpc_pipe != NULL; mpc_pipe = mpc_pipe->bottom_pipe) {
-			if (mpc_pipe->plane_state &&
-					mpc_pipe->plane_state->cm.flags.bits.lut3d_enable &&
-					mpc_pipe->plane_state->cm.flags.bits.lut3d_dma_enable) {
-				wa_pipes[wa_pipe_ct++] = mpc_pipe;
-			}
-		}
+	if (!otg_master_pipe_ctx || !tg) {
+		return;
 	}
 
-	if (wa_pipe_ct > 0) {
-		if (pipe_ctx->stream_res.tg->funcs->set_vupdate_keepout)
-			pipe_ctx->stream_res.tg->funcs->set_vupdate_keepout(pipe_ctx->stream_res.tg, true);
+	if (primary_dpp_pipe_ctx &&
+			primary_dpp_pipe_ctx->plane_state &&
+			primary_dpp_pipe_ctx->plane_state->cm.flags.bits.lut3d_enable &&
+			primary_dpp_pipe_ctx->plane_state->cm.flags.bits.lut3d_dma_enable) {
+		if (tg->funcs->set_vupdate_keepout)
+			tg->funcs->set_vupdate_keepout(tg, true);
 
-		for (i = 0; i < wa_pipe_ct; ++i) {
-			if (wa_pipes[i]->plane_res.hubp->funcs->hubp_enable_3dlut_fl)
-				wa_pipes[i]->plane_res.hubp->funcs->hubp_enable_3dlut_fl(wa_pipes[i]->plane_res.hubp, true);
+		if (primary_hubp && primary_hubp->funcs->hubp_enable_3dlut_fl) {
+			primary_hubp->funcs->hubp_enable_3dlut_fl(primary_hubp, true);
 		}
 
-		pipe_ctx->stream_res.tg->funcs->unlock(pipe_ctx->stream_res.tg);
-		if (pipe_ctx->stream_res.tg->funcs->wait_update_lock_status)
-			pipe_ctx->stream_res.tg->funcs->wait_update_lock_status(pipe_ctx->stream_res.tg, false);
+		tg->funcs->unlock(tg);
+		if (tg->funcs->wait_update_lock_status)
+			tg->funcs->wait_update_lock_status(tg, false);
 
-		for (i = 0; i < wa_pipe_ct; ++i) {
-			if (wa_pipes[i]->plane_res.hubp->funcs->hubp_enable_3dlut_fl)
-				wa_pipes[i]->plane_res.hubp->funcs->hubp_enable_3dlut_fl(wa_pipes[i]->plane_res.hubp, true);
+		if (primary_hubp && primary_hubp->funcs->hubp_enable_3dlut_fl) {
+			primary_hubp->funcs->hubp_enable_3dlut_fl(primary_hubp, true);
 		}
 
-		if (pipe_ctx->stream_res.tg->funcs->set_vupdate_keepout)
-			pipe_ctx->stream_res.tg->funcs->set_vupdate_keepout(pipe_ctx->stream_res.tg, false);
+		if (tg->funcs->set_vupdate_keepout)
+			tg->funcs->set_vupdate_keepout(tg, false);
 	} else {
-		pipe_ctx->stream_res.tg->funcs->unlock(pipe_ctx->stream_res.tg);
+		tg->funcs->unlock(tg);
 	}
 }
 
diff --git a/drivers/gpu/drm/amd/display/dc/hwss/dcn401/dcn401_hwseq.h b/drivers/gpu/drm/amd/display/dc/hwss/dcn401/dcn401_hwseq.h
index f90e25243ead..69c7dc73a29a 100644
--- a/drivers/gpu/drm/amd/display/dc/hwss/dcn401/dcn401_hwseq.h
+++ b/drivers/gpu/drm/amd/display/dc/hwss/dcn401/dcn401_hwseq.h
@@ -39,8 +39,7 @@ void dcn401_init_hw(struct dc *dc);
 bool dcn401_set_mcm_luts(struct pipe_ctx *pipe_ctx,
 				const struct dc_plane_state *plane_state);
 bool dcn401_set_output_transfer_func(struct set_output_transfer_func_params *params);
-void dcn401_trigger_3dlut_dma_load(struct dc *dc,
-				struct pipe_ctx *pipe_ctx);
+void dcn401_trigger_3dlut_dma_load(struct pipe_ctx *pipe_ctx);
 void dcn401_calculate_dccg_tmds_div_value(struct pipe_ctx *pipe_ctx,
 				unsigned int *tmds_div);
 enum dc_status dcn401_enable_stream_timing(
@@ -48,10 +47,6 @@ enum dc_status dcn401_enable_stream_timing(
 				struct dc_state *context,
 				struct dc *dc);
 void dcn401_enable_stream(struct pipe_ctx *pipe_ctx);
-void dcn401_populate_mcm_luts(struct dc *dc,
-		struct pipe_ctx *pipe_ctx,
-		const struct dc_plane_cm *cm,
-		bool lut_bank_a);
 void dcn401_setup_hpo_hw_control(const struct dce_hwseq *hws, bool enable);
 
 void dcn401_disable_link_output(struct dc_link *link,
diff --git a/drivers/gpu/drm/amd/display/dc/hwss/dcn401/dcn401_init.c b/drivers/gpu/drm/amd/display/dc/hwss/dcn401/dcn401_init.c
index f206e221f926..7039f40ea880 100644
--- a/drivers/gpu/drm/amd/display/dc/hwss/dcn401/dcn401_init.c
+++ b/drivers/gpu/drm/amd/display/dc/hwss/dcn401/dcn401_init.c
@@ -166,7 +166,6 @@ static const struct hwseq_private_funcs dcn401_private_funcs = {
 	.calculate_dccg_k1_k2_values = NULL,
 	.apply_single_controller_ctx_to_hw = dce110_apply_single_controller_ctx_to_hw,
 	.reset_back_end_for_pipe = dcn401_reset_back_end_for_pipe,
-	.populate_mcm_luts = NULL,
 	.perform_3dlut_wa_unlock = dcn401_perform_3dlut_wa_unlock,
 	.program_pipe_sequence = dcn401_program_pipe_sequence,
 	.dc_ip_request_cntl = dcn401_dc_ip_request_cntl,
diff --git a/drivers/gpu/drm/amd/display/dc/hwss/dcn42/dcn42_hwseq.c b/drivers/gpu/drm/amd/display/dc/hwss/dcn42/dcn42_hwseq.c
index 151a29bf0e9d..13580d9f9303 100644
--- a/drivers/gpu/drm/amd/display/dc/hwss/dcn42/dcn42_hwseq.c
+++ b/drivers/gpu/drm/amd/display/dc/hwss/dcn42/dcn42_hwseq.c
@@ -402,64 +402,6 @@ void dcn42_program_cm_hist(
 			plane_state->cm_hist_control, plane_state->color_space);
 }
 
-static void dc_get_lut_xbar(
-	enum hubp_3dlut_fl_crossbar_bit_slice *cr_r,
-	enum hubp_3dlut_fl_crossbar_bit_slice *y_g,
-	enum hubp_3dlut_fl_crossbar_bit_slice *cb_b)
-{
-	/* component_order was previously hard-coded to RGBA in local_mcm,
-	 * preserve identical behavior.
-	 */
-	*cr_r = hubp_3dlut_fl_crossbar_bit_slice_32_47;
-	*y_g = hubp_3dlut_fl_crossbar_bit_slice_16_31;
-	*cb_b = hubp_3dlut_fl_crossbar_bit_slice_0_15;
-}
-
-static void dc_get_lut_mode(
-	enum dc_cm_lut_swizzle swizzle,
-	enum hubp_3dlut_fl_mode *mode,
-	enum hubp_3dlut_fl_addressing_mode *addr_mode)
-{
-	switch (swizzle) {
-	case CM_LUT_3D_SWIZZLE_LINEAR_RGB:
-		*mode = hubp_3dlut_fl_mode_native_1;
-		*addr_mode = hubp_3dlut_fl_addressing_mode_sw_linear;
-		break;
-	case CM_LUT_3D_SWIZZLE_LINEAR_BGR:
-		*mode = hubp_3dlut_fl_mode_native_2;
-		*addr_mode = hubp_3dlut_fl_addressing_mode_sw_linear;
-		break;
-	case CM_LUT_1D_PACKED_LINEAR:
-		*mode = hubp_3dlut_fl_mode_transform;
-		*addr_mode = hubp_3dlut_fl_addressing_mode_simple_linear;
-		break;
-	default:
-		*mode = hubp_3dlut_fl_mode_disable;
-		*addr_mode = hubp_3dlut_fl_addressing_mode_sw_linear;
-		break;
-	}
-}
-
-static void dc_get_lut_format(
-	enum dc_cm_lut_pixel_format dc_format,
-	enum hubp_3dlut_fl_format *format)
-{
-	switch (dc_format) {
-	case CM_LUT_PIXEL_FORMAT_RGBA16161616_UNORM_12MSB:
-		*format = hubp_3dlut_fl_format_unorm_12msb_bitslice;
-		break;
-	case CM_LUT_PIXEL_FORMAT_RGBA16161616_UNORM_12LSB:
-		*format = hubp_3dlut_fl_format_unorm_12lsb_bitslice;
-		break;
-	case CM_LUT_PIXEL_FORMAT_RGBA16161616_FLOAT_FP1_5_10:
-		*format = hubp_3dlut_fl_format_float_fp1_5_10;
-		break;
-	default:
-		*format = hubp_3dlut_fl_format_unorm_12msb_bitslice;
-		break;
-	}
-}
-
 static bool dc_is_rmcm_3dlut_supported(struct hubp *hubp, struct mpc *mpc)
 {
 	if (mpc->funcs->rmcm.power_on_shaper_3dlut &&
@@ -470,106 +412,17 @@ static bool dc_is_rmcm_3dlut_supported(struct hubp *hubp, struct mpc *mpc)
 	return false;
 }
 
-static bool is_rmcm_3dlut_fl_supported(struct dc *dc)
-{
-	/* size was previously hard-coded to TRANSFORMED in local_mcm,
-	 * which mapped to dim_17. Preserve identical behavior.
-	 */
-	if (!dc->caps.color.mpc.rmcm_3d_lut_caps.dma_3d_lut)
-		return false;
-	return dc->caps.color.mpc.rmcm_3d_lut_caps.lut_dim_caps.dim_17 != 0u;
-}
-
-static void dcn42_set_mcm_location_post_blend(struct dc *dc, struct pipe_ctx *pipe_ctx, bool bPostBlend)
-{
-	struct mpc *mpc = dc->res_pool->mpc;
-	int mpcc_id = pipe_ctx->plane_res.hubp->inst;
-
-	if (!pipe_ctx->plane_state)
-		return;
-
-	mpc->funcs->set_movable_cm_location(mpc, MPCC_MOVABLE_CM_LOCATION_BEFORE, mpcc_id);
-	pipe_ctx->plane_state->mcm_location = (bPostBlend) ?
-											MPCC_MOVABLE_CM_LOCATION_AFTER :
-											MPCC_MOVABLE_CM_LOCATION_BEFORE;
-}
-
-static void dcn42_get_mcm_lut_xable_from_pipe_ctx(struct dc *dc, struct pipe_ctx *pipe_ctx,
-		enum MCM_LUT_XABLE *shaper_xable,
-		enum MCM_LUT_XABLE *lut3d_xable,
-		enum MCM_LUT_XABLE *lut1d_xable)
-{
-	struct mpc *mpc = dc->res_pool->mpc;
-	int mpcc_id = pipe_ctx->plane_res.hubp->inst;
-
-	if (!pipe_ctx->plane_state)
-		return;
-
-	mpc->funcs->set_movable_cm_location(mpc, MPCC_MOVABLE_CM_LOCATION_BEFORE, mpcc_id);
-	pipe_ctx->plane_state->mcm_location = MPCC_MOVABLE_CM_LOCATION_BEFORE;
-
-	*lut1d_xable = pipe_ctx->plane_state->cm.flags.bits.blend_enable ?
-		MCM_LUT_ENABLE : MCM_LUT_DISABLE;
-	*shaper_xable = pipe_ctx->plane_state->cm.flags.bits.shaper_enable ?
-		MCM_LUT_ENABLE : MCM_LUT_DISABLE;
-	*lut3d_xable = (pipe_ctx->plane_state->cm.flags.bits.shaper_enable &&
-			pipe_ctx->plane_state->cm.flags.bits.lut3d_enable) ?
-		MCM_LUT_ENABLE : MCM_LUT_DISABLE;
-}
-
-static void fl_get_lut_mode(
-	enum dc_cm_lut_swizzle swizzle,
-	enum hubp_3dlut_fl_mode *mode,
-	enum hubp_3dlut_fl_addressing_mode *addr_mode,
-	enum hubp_3dlut_fl_width *width)
-{
-	/* size was previously hard-coded to TRANSFORMED in local_mcm,
-	 * preserve identical behavior (transformed width).
-	 */
-	*width = hubp_3dlut_fl_width_17;
-
-	switch (swizzle) {
-	case CM_LUT_3D_SWIZZLE_LINEAR_RGB:
-		*mode = hubp_3dlut_fl_mode_native_1;
-		*addr_mode = hubp_3dlut_fl_addressing_mode_sw_linear;
-		break;
-	case CM_LUT_3D_SWIZZLE_LINEAR_BGR:
-		*mode = hubp_3dlut_fl_mode_native_2;
-		*addr_mode = hubp_3dlut_fl_addressing_mode_sw_linear;
-		break;
-	case CM_LUT_1D_PACKED_LINEAR:
-		*mode = hubp_3dlut_fl_mode_transform;
-		*addr_mode = hubp_3dlut_fl_addressing_mode_simple_linear;
-		break;
-	default:
-		*mode = hubp_3dlut_fl_mode_disable;
-		*addr_mode = hubp_3dlut_fl_addressing_mode_sw_linear;
-		break;
-	}
-}
-
 bool dcn42_program_rmcm_luts(
 	struct hubp *hubp,
 	struct pipe_ctx *pipe_ctx,
 	const struct dc_plane_cm *cm,
 	struct mpc *mpc,
-	bool lut_bank_a,
 	int mpcc_id)
 {
 	struct dpp *dpp_base = pipe_ctx->plane_res.dpp;
 	union mcm_lut_params m_lut_params = {0};
-	enum MCM_LUT_XABLE shaper_xable, lut3d_xable = MCM_LUT_DISABLE, lut1d_xable;
-		enum hubp_3dlut_fl_mode mode;
-	enum hubp_3dlut_fl_addressing_mode addr_mode;
-	enum hubp_3dlut_fl_format format = hubp_3dlut_fl_format_unorm_12msb_bitslice;
-	enum hubp_3dlut_fl_crossbar_bit_slice crossbar_bit_slice_y_g = hubp_3dlut_fl_crossbar_bit_slice_16_31;
-	enum hubp_3dlut_fl_crossbar_bit_slice crossbar_bit_slice_cb_b = hubp_3dlut_fl_crossbar_bit_slice_0_15;
-	enum hubp_3dlut_fl_crossbar_bit_slice crossbar_bit_slice_cr_r = hubp_3dlut_fl_crossbar_bit_slice_32_47;
-	enum hubp_3dlut_fl_width width = hubp_3dlut_fl_width_17;
-
 
 	struct dc *dc = hubp->ctx->dc;
-	struct hubp_fl_3dlut_config fl_config;
 	struct mpc_fl_3dlut_config mpc_fl_config;
 
 	struct dc_stream_state *stream = pipe_ctx->stream;
@@ -577,19 +430,24 @@ bool dcn42_program_rmcm_luts(
 	// true->false when it can be allocated at DI time
 	struct dc_rmcm_3dlut *rmcm_3dlut = dc_stream_get_3dlut_for_stream(dc, stream, false);
 
+	bool lut_enable = false;
+	bool lut_bank_a = true;
+
 	//check to see current pipe is part of a stream with allocated rmcm 3dlut
 	if (!rmcm_3dlut)
 		return false;
 
-	/* rmcm_tmz was previously zero-initialized in local_mcm,
-	 * preserve identical behavior.
+	/* Determine the LUT bank currently active in HW and switch to the other
+	 * bank to preserve double-buffering (mirrors dcn401_set_mcm_luts).
 	 */
-	rmcm_3dlut->protection_bits = 0;
+	if (mpc->funcs->rmcm.get_3dlut_mode)
+		mpc->funcs->rmcm.get_3dlut_mode(mpc, mpcc_id, &lut_enable, &lut_bank_a);
 
-	dcn42_get_mcm_lut_xable_from_pipe_ctx(dc, pipe_ctx, &shaper_xable, &lut3d_xable, &lut1d_xable);
+	if (lut_enable)
+		lut_bank_a = !lut_bank_a;
 
 	/* Shaper */
-	{
+	if (cm->flags.bits.shaper_enable) {
 		memset(&m_lut_params, 0, sizeof(m_lut_params));
 
 		if (cm->shaper_func.type == TF_TYPE_HWPWL) {
@@ -615,56 +473,19 @@ bool dcn42_program_rmcm_luts(
 
 	/* 3DLUT */
 	if (!cm->flags.bits.lut3d_dma_enable) {
-		/* SYSMEM path — no DMA 3DLUT available.
-		 * Previously this was treated as a no-op for the DMA/VIDMEM
-		 * programming, preserve identical behavior.
+		/* RMCM host (non-DMA) 3DLUT load is not implemented; fail fast
+		 * instead of silently reporting success
 		 */
-		memset(&m_lut_params, 0, sizeof(m_lut_params));
+		BREAK_TO_DEBUGGER();
+		return false;
 	} else {
-		/* VIDMEM (3DLUT DMA Fast Load) */
-
-		fl_get_lut_mode(cm->lut3d_dma.swizzle,
-				&mode,
-				&addr_mode,
-				&width);
-
-		if (!dc_is_rmcm_3dlut_supported(hubp, mpc) ||
-			!mpc->funcs->rmcm.is_config_supported(
-				(width == hubp_3dlut_fl_width_17 ||
-				 width == hubp_3dlut_fl_width_transformed) ? 17 : 33))
+		if (!dc_is_rmcm_3dlut_supported(hubp, mpc))
 			return false;
 
-		// setting native or transformed mode,
-		dc_get_lut_mode(cm->lut3d_dma.swizzle, &mode, &addr_mode);
-
 		//seems to be only for the MCM
-		dc_get_lut_format(cm->lut3d_dma.format, &format);
-
-		dc_get_lut_xbar(
-			&crossbar_bit_slice_cr_r,
-			&crossbar_bit_slice_y_g,
-			&crossbar_bit_slice_cb_b);
-
-		fl_config.mode					= mode;
-		fl_config.enabled				= lut3d_xable != MCM_LUT_DISABLE;
-		fl_config.address				= cm->lut3d_dma.addr;
-		fl_config.format				= format;
-		fl_config.crossbar_bit_slice_y_g  = crossbar_bit_slice_y_g;
-		fl_config.crossbar_bit_slice_cb_b = crossbar_bit_slice_cb_b;
-		fl_config.crossbar_bit_slice_cr_r = crossbar_bit_slice_cr_r;
-		fl_config.width				    = width;
-		fl_config.protection_bits		= rmcm_3dlut->protection_bits;
-		fl_config.addr_mode			    = addr_mode;
-		fl_config.layout                = cm->lut3d_dma.swizzle;
-		fl_config.bias	= cm->lut3d_dma.bias;
-		fl_config.scale	= cm->lut3d_dma.scale;
-
-		mpc_fl_config.enabled			= fl_config.enabled;
-		mpc_fl_config.width	            = width;
+		mpc_fl_config.enabled			= cm->flags.bits.lut3d_enable != 0u;
+		mpc_fl_config.size	            = cm->lut3d_dma.size;
 		mpc_fl_config.select_lut_bank_a = lut_bank_a;
-		/* bit_depth was previously zero-initialized in local_mcm,
-		 * preserve identical behavior.
-		 */
 		mpc_fl_config.bit_depth		    = 0;
 		mpc_fl_config.hubp_index		= hubp->inst;
 		mpc_fl_config.bias	= cm->lut3d_dma.bias;
@@ -676,7 +497,12 @@ bool dcn42_program_rmcm_luts(
 		//2. program RMCM - 3dlut reg programming
 		mpc->funcs->rmcm.fl_3dlut_configure(mpc, &mpc_fl_config, mpcc_id);
 
-		hubp->funcs->hubp_program_3dlut_fl_config(hubp, &fl_config);
+		/* HUBP */
+		if (hubp->funcs->hubp_program_3dlut_fl_config)
+			hubp->funcs->hubp_program_3dlut_fl_config(hubp, &cm->lut3d_dma);
+
+		if (hubp->funcs->hubp_program_3dlut_fl_addr)
+			hubp->funcs->hubp_program_3dlut_fl_addr(hubp, &cm->lut3d_dma.addr);
 
 		//3. power on the block
 		mpc->funcs->rmcm.power_on_shaper_3dlut(mpc, mpcc_id, true);
@@ -685,261 +511,31 @@ bool dcn42_program_rmcm_luts(
 	return true;
 }
 
-void dcn42_populate_mcm_luts(struct dc *dc,
-		struct pipe_ctx *pipe_ctx,
-		const struct dc_plane_cm *cm,
-		bool lut_bank_a)
-{
-	struct dpp *dpp_base = pipe_ctx->plane_res.dpp;
-	struct hubp *hubp = pipe_ctx->plane_res.hubp;
-	int mpcc_id = hubp->inst;
-	struct mpc *mpc = dc->res_pool->mpc;
-	union mcm_lut_params m_lut_params;
-	const bool lut3d_dma = !!cm->flags.bits.lut3d_dma_enable;
-	enum hubp_3dlut_fl_format format = 0;
-	enum hubp_3dlut_fl_mode mode;
-	/* Width was previously hard-coded to TRANSFORMED via local_mcm build,
-	 * preserve identical behavior.
-	 */
-	enum hubp_3dlut_fl_width width = hubp_3dlut_fl_width_transformed;
-	enum hubp_3dlut_fl_addressing_mode addr_mode;
-	enum hubp_3dlut_fl_crossbar_bit_slice crossbar_bit_slice_y_g;
-	enum hubp_3dlut_fl_crossbar_bit_slice crossbar_bit_slice_cb_b;
-	enum hubp_3dlut_fl_crossbar_bit_slice crossbar_bit_slice_cr_r;
-	enum MCM_LUT_XABLE shaper_xable = MCM_LUT_DISABLE;
-	enum MCM_LUT_XABLE lut3d_xable = MCM_LUT_DISABLE;
-	enum MCM_LUT_XABLE lut1d_xable = MCM_LUT_DISABLE;
-	bool rval;
-
-	dcn42_get_mcm_lut_xable_from_pipe_ctx(dc, pipe_ctx, &shaper_xable, &lut3d_xable, &lut1d_xable);
-
-	//MCM - setting its location (Before/After) blender
-	//mpc_mcm_post_blend was previously zero-initialized in local_mcm,
-	//preserve identical behavior.
-	dcn42_set_mcm_location_post_blend(
-		dc,
-		pipe_ctx,
-		false);
-
-	//RMCM - 3dLUT+Shaper
-	if (cm->flags.bits.rmcm_enable &&
-		is_rmcm_3dlut_fl_supported(dc))
-		dcn42_program_rmcm_luts(
-			hubp,
-			pipe_ctx,
-			cm,
-			mpc,
-			lut_bank_a,
-			mpcc_id);
-
-	/* 1D LUT */
-	{
-		memset(&m_lut_params, 0, sizeof(m_lut_params));
-		if (cm->blend_func.type == TF_TYPE_HWPWL)
-			m_lut_params.pwl = &cm->blend_func.pwl;
-		else if (cm->blend_func.type == TF_TYPE_DISTRIBUTED_POINTS) {
-			rval = cm3_helper_translate_curve_to_hw_format(mpc->ctx,
-					&cm->blend_func,
-					&dpp_base->regamma_params, false);
-			m_lut_params.pwl = rval ? &dpp_base->regamma_params : NULL;
-		}
-		if (m_lut_params.pwl) {
-			if (mpc->funcs->populate_lut)
-				mpc->funcs->populate_lut(mpc, MCM_LUT_1DLUT, m_lut_params, lut_bank_a, mpcc_id);
-		}
-		if (mpc->funcs->program_lut_mode)
-			mpc->funcs->program_lut_mode(mpc, MCM_LUT_1DLUT, lut1d_xable && m_lut_params.pwl, lut_bank_a, mpcc_id);
-	}
-
-	/* Shaper */
-	if (cm->flags.bits.lut3d_enable) {
-		memset(&m_lut_params, 0, sizeof(m_lut_params));
-		if (cm->shaper_func.type == TF_TYPE_HWPWL)
-			m_lut_params.pwl = &cm->shaper_func.pwl;
-		else if (cm->shaper_func.type == TF_TYPE_DISTRIBUTED_POINTS) {
-			ASSERT(false);
-			rval = cm3_helper_translate_curve_to_hw_format(mpc->ctx,
-					&cm->shaper_func,
-					&dpp_base->regamma_params, true);
-			m_lut_params.pwl = rval ? &dpp_base->regamma_params : NULL;
-		}
-		if (m_lut_params.pwl) {
-			if (mpc->funcs->mcm.populate_lut)
-				mpc->funcs->mcm.populate_lut(mpc, m_lut_params, lut_bank_a, mpcc_id);
-			if (mpc->funcs->program_lut_mode)
-				mpc->funcs->program_lut_mode(mpc, MCM_LUT_SHAPER, MCM_LUT_ENABLE, lut_bank_a, mpcc_id);
-		}
-	}
-
-	/* 3DLUT */
-	if (!lut3d_dma) {
-		/* SYSMEM (legacy lut3d_func) */
-		memset(&m_lut_params, 0, sizeof(m_lut_params));
-		if (hubp->funcs->hubp_enable_3dlut_fl)
-			hubp->funcs->hubp_enable_3dlut_fl(hubp, false);
-
-		if (cm->lut3d_func.state.bits.initialized) {
-			m_lut_params.lut3d = &cm->lut3d_func.lut_3d;
-			if (mpc->funcs->populate_lut)
-				mpc->funcs->populate_lut(mpc, MCM_LUT_3DLUT, m_lut_params, lut_bank_a, mpcc_id);
-			if (mpc->funcs->program_lut_mode)
-				mpc->funcs->program_lut_mode(mpc, MCM_LUT_3DLUT, lut3d_xable, lut_bank_a,
-						mpcc_id);
-		}
-	} else {
-		/* VIDMEM (3DLUT DMA Fast Load) */
-
-		//check for support
-		if (mpc->funcs->mcm.is_config_supported &&
-			!mpc->funcs->mcm.is_config_supported(width))
-			return;
-
-		if (mpc->funcs->program_lut_read_write_control)
-			mpc->funcs->program_lut_read_write_control(mpc, MCM_LUT_3DLUT, lut_bank_a, mpcc_id);
-		if (mpc->funcs->program_lut_mode)
-			mpc->funcs->program_lut_mode(mpc, MCM_LUT_3DLUT, lut3d_xable, lut_bank_a, mpcc_id);
-
-		if (hubp->funcs->hubp_program_3dlut_fl_addr)
-			hubp->funcs->hubp_program_3dlut_fl_addr(hubp, cm->lut3d_dma.addr);
-
-		/* bit_depth was previously zero-initialized in local_mcm,
-		 * preserve identical behavior.
-		 */
-		if (mpc->funcs->mcm.program_bit_depth)
-			mpc->funcs->mcm.program_bit_depth(mpc, 0, mpcc_id);
-
-		switch (cm->lut3d_dma.swizzle) {
-		case CM_LUT_3D_SWIZZLE_LINEAR_RGB:
-			mode = hubp_3dlut_fl_mode_native_1;
-			addr_mode = hubp_3dlut_fl_addressing_mode_sw_linear;
-			break;
-		case CM_LUT_3D_SWIZZLE_LINEAR_BGR:
-			mode = hubp_3dlut_fl_mode_native_2;
-			addr_mode = hubp_3dlut_fl_addressing_mode_sw_linear;
-			break;
-		case CM_LUT_1D_PACKED_LINEAR:
-			mode = hubp_3dlut_fl_mode_transform;
-			addr_mode = hubp_3dlut_fl_addressing_mode_simple_linear;
-			break;
-		default:
-			mode = hubp_3dlut_fl_mode_disable;
-			addr_mode = hubp_3dlut_fl_addressing_mode_sw_linear;
-			break;
-		}
-		if (hubp->funcs->hubp_program_3dlut_fl_mode)
-			hubp->funcs->hubp_program_3dlut_fl_mode(hubp, mode);
-
-		if (hubp->funcs->hubp_program_3dlut_fl_addressing_mode)
-			hubp->funcs->hubp_program_3dlut_fl_addressing_mode(hubp, addr_mode);
-
-		switch (cm->lut3d_dma.format) {
-		case CM_LUT_PIXEL_FORMAT_RGBA16161616_UNORM_12MSB:
-			format = hubp_3dlut_fl_format_unorm_12msb_bitslice;
-			break;
-		case CM_LUT_PIXEL_FORMAT_RGBA16161616_UNORM_12LSB:
-			format = hubp_3dlut_fl_format_unorm_12lsb_bitslice;
-			break;
-		case CM_LUT_PIXEL_FORMAT_RGBA16161616_FLOAT_FP1_5_10:
-			format = hubp_3dlut_fl_format_float_fp1_5_10;
-			break;
-		default:
-			break;
-		}
-		if (hubp->funcs->hubp_program_3dlut_fl_format)
-			hubp->funcs->hubp_program_3dlut_fl_format(hubp, format);
-		if (hubp->funcs->hubp_update_3dlut_fl_bias_scale &&
-				mpc->funcs->mcm.program_bias_scale) {
-			mpc->funcs->mcm.program_bias_scale(mpc,
-				cm->lut3d_dma.bias,
-				cm->lut3d_dma.scale,
-				mpcc_id);
-			hubp->funcs->hubp_update_3dlut_fl_bias_scale(hubp,
-				cm->lut3d_dma.bias,
-				cm->lut3d_dma.scale);
-		}
-
-		/* component_order was previously hard-coded to RGBA in local_mcm,
-		 * preserve identical behavior.
-		 */
-		crossbar_bit_slice_cr_r = hubp_3dlut_fl_crossbar_bit_slice_0_15;
-		crossbar_bit_slice_y_g = hubp_3dlut_fl_crossbar_bit_slice_16_31;
-		crossbar_bit_slice_cb_b = hubp_3dlut_fl_crossbar_bit_slice_32_47;
-
-		if (hubp->funcs->hubp_program_3dlut_fl_crossbar)
-			hubp->funcs->hubp_program_3dlut_fl_crossbar(hubp,
-					crossbar_bit_slice_cr_r,
-					crossbar_bit_slice_y_g,
-					crossbar_bit_slice_cb_b);
-
-		if (mpc->funcs->mcm.program_lut_read_write_control)
-			mpc->funcs->mcm.program_lut_read_write_control(mpc, MCM_LUT_3DLUT, lut_bank_a, true, mpcc_id);
-
-		if (mpc->funcs->mcm.program_3dlut_size)
-			mpc->funcs->mcm.program_3dlut_size(mpc, width, mpcc_id);
-
-		if (mpc->funcs->update_3dlut_fast_load_select)
-			mpc->funcs->update_3dlut_fast_load_select(mpc, mpcc_id, hubp->inst);
-
-		if (hubp->funcs->hubp_enable_3dlut_fl)
-			hubp->funcs->hubp_enable_3dlut_fl(hubp, true);
-		else {
-			if (mpc->funcs->program_lut_mode) {
-				mpc->funcs->program_lut_mode(mpc, MCM_LUT_SHAPER, MCM_LUT_DISABLE, lut_bank_a, mpcc_id);
-				mpc->funcs->program_lut_mode(mpc, MCM_LUT_3DLUT, MCM_LUT_DISABLE, lut_bank_a, mpcc_id);
-				mpc->funcs->program_lut_mode(mpc, MCM_LUT_1DLUT, MCM_LUT_DISABLE, lut_bank_a, mpcc_id);
-			}
-		}
-	}
-}
-
 bool dcn42_set_mcm_luts(struct pipe_ctx *pipe_ctx,
 				const struct dc_plane_state *plane_state)
 {
-	struct dpp *dpp_base = pipe_ctx->plane_res.dpp;
-	int mpcc_id = pipe_ctx->plane_res.hubp->inst;
-	struct dc *dc = pipe_ctx->stream_res.opp->ctx->dc;
-	struct mpc *mpc = dc->res_pool->mpc;
 	bool result;
-	const struct pwl_params *lut_params = NULL;
-	bool rval;
-
-	if (plane_state->cm.flags.bits.lut3d_dma_enable) {
-		dcn42_populate_mcm_luts(dc, pipe_ctx, &plane_state->cm, plane_state->lut_bank_a);
-		return true;
-	}
 
-	mpc->funcs->set_movable_cm_location(mpc, MPCC_MOVABLE_CM_LOCATION_BEFORE, mpcc_id);
-	pipe_ctx->plane_state->mcm_location = MPCC_MOVABLE_CM_LOCATION_BEFORE;
-	// 1D LUT
-	if (plane_state->cm.blend_func.type == TF_TYPE_HWPWL)
-		lut_params = &plane_state->cm.blend_func.pwl;
-	else if (plane_state->cm.blend_func.type == TF_TYPE_DISTRIBUTED_POINTS) {
-		rval = cm3_helper_translate_curve_to_hw_format(plane_state->ctx,
-				&plane_state->cm.blend_func,
-				&dpp_base->regamma_params, false);
-		lut_params = rval ? &dpp_base->regamma_params : NULL;
-	}
-	result = mpc->funcs->program_1dlut(mpc, lut_params, mpcc_id);
-	lut_params = NULL;
-
-	// Shaper
-	if (plane_state->cm.shaper_func.type == TF_TYPE_HWPWL)
-		lut_params = &plane_state->cm.shaper_func.pwl;
-	else if (plane_state->cm.shaper_func.type == TF_TYPE_DISTRIBUTED_POINTS) {
-		// TODO: dpp_base replace
-		rval = cm3_helper_translate_curve_to_hw_format(plane_state->ctx,
-				&plane_state->cm.shaper_func,
-				&dpp_base->shaper_params, true);
-		lut_params = rval ? &dpp_base->shaper_params : NULL;
-	}
-	result &= mpc->funcs->program_shaper(mpc, lut_params, mpcc_id);
+	/* MCM */
+	result = dcn401_set_mcm_luts(pipe_ctx, plane_state);
 
-	// 3D
-	if (mpc->funcs->program_3dlut) {
-		if (plane_state->cm.lut3d_func.state.bits.initialized == 1)
-			result &= mpc->funcs->program_3dlut(mpc, &plane_state->cm.lut3d_func.lut_3d, mpcc_id);
-		else
-			result &= mpc->funcs->program_3dlut(mpc, NULL, mpcc_id);
+	/* RMCM */
+	{
+		struct dc *dc = pipe_ctx->plane_res.hubp->ctx->dc;
+		struct hubp *hubp = pipe_ctx->plane_res.hubp;
+		const struct dc_plane_cm *cm = &plane_state->cm;
+		struct mpc *mpc = dc->res_pool->mpc;
+		int mpcc_id = hubp->inst;
+
+		if (cm->flags.bits.rmcm_enable && cm->flags.bits.lut3d_dma_enable) {
+			/* TODO - move RMCM to its own block */
+			dcn42_program_rmcm_luts(
+				hubp,
+				pipe_ctx,
+				cm,
+				mpc,
+				mpcc_id);
+		}
 	}
 
 	return result;
diff --git a/drivers/gpu/drm/amd/display/dc/hwss/dcn42/dcn42_hwseq.h b/drivers/gpu/drm/amd/display/dc/hwss/dcn42/dcn42_hwseq.h
index c469e7535114..c4cfeed45b19 100644
--- a/drivers/gpu/drm/amd/display/dc/hwss/dcn42/dcn42_hwseq.h
+++ b/drivers/gpu/drm/amd/display/dc/hwss/dcn42/dcn42_hwseq.h
@@ -18,17 +18,11 @@ void dcn42_program_cm_hist(
 bool dcn42_set_mcm_luts(struct pipe_ctx *pipe_ctx,
 	const struct dc_plane_state *plane_state);
 
-void dcn42_populate_mcm_luts(struct dc *dc,
-		struct pipe_ctx *pipe_ctx,
-		const struct dc_plane_cm *cm,
-		bool lut_bank_a);
-
 bool dcn42_program_rmcm_luts(
 	struct hubp *hubp,
 	struct pipe_ctx *pipe_ctx,
 	const struct dc_plane_cm *cm,
 	struct mpc *mpc,
-	bool lut_bank_a,
 	int mpcc_id);
 void dcn42_hardware_release(struct dc *dc);
 
diff --git a/drivers/gpu/drm/amd/display/dc/hwss/dcn42/dcn42_init.c b/drivers/gpu/drm/amd/display/dc/hwss/dcn42/dcn42_init.c
index 49c13611a518..ba12a97a220f 100644
--- a/drivers/gpu/drm/amd/display/dc/hwss/dcn42/dcn42_init.c
+++ b/drivers/gpu/drm/amd/display/dc/hwss/dcn42/dcn42_init.c
@@ -155,8 +155,6 @@ static const struct hwseq_private_funcs dcn42_private_funcs = {
 	.calculate_dccg_k1_k2_values = NULL,
 	.apply_single_controller_ctx_to_hw = dce110_apply_single_controller_ctx_to_hw,
 	.reset_back_end_for_pipe = dcn401_reset_back_end_for_pipe,
-	.populate_mcm_luts = NULL,
-	.perform_3dlut_wa_unlock = dcn401_perform_3dlut_wa_unlock,
 	.program_cm_hist = dcn42_program_cm_hist,
 	.dpp_root_clock_control = dcn35_dpp_root_clock_control,
 	.dpstream_root_clock_control = dcn35_dpstream_root_clock_control,
diff --git a/drivers/gpu/drm/amd/display/dc/hwss/dcn60/dcn60_init.c b/drivers/gpu/drm/amd/display/dc/hwss/dcn60/dcn60_init.c
index 5f4f32d516d5..1c908964153b 100644
--- a/drivers/gpu/drm/amd/display/dc/hwss/dcn60/dcn60_init.c
+++ b/drivers/gpu/drm/amd/display/dc/hwss/dcn60/dcn60_init.c
@@ -166,8 +166,6 @@ static const struct hwseq_private_funcs dcn60_private_funcs = {
 	.calculate_dccg_k1_k2_values = NULL,
 	.apply_single_controller_ctx_to_hw = dcn60_apply_single_controller_ctx_to_hw,
 	.reset_back_end_for_pipe = dcn401_reset_back_end_for_pipe,
-	.populate_mcm_luts = NULL,
-	.perform_3dlut_wa_unlock = dcn401_perform_3dlut_wa_unlock,
 	.program_cm_hist = dcn42_program_cm_hist,
 	.program_pipe = dcn401_program_pipe,
 	.program_pipe_sequence = dcn401_program_pipe_sequence,
diff --git a/drivers/gpu/drm/amd/display/dc/hwss/hw_sequencer.h b/drivers/gpu/drm/amd/display/dc/hwss/hw_sequencer.h
index c4296b8390f2..cdf2562a9a40 100644
--- a/drivers/gpu/drm/amd/display/dc/hwss/hw_sequencer.h
+++ b/drivers/gpu/drm/amd/display/dc/hwss/hw_sequencer.h
@@ -1506,7 +1506,7 @@ struct hw_sequencer_funcs {
 	void (*program_output_csc)(struct dc *dc, struct pipe_ctx *pipe_ctx,
 			enum dc_color_space colorspace,
 			uint16_t *matrix, int opp_id);
-	void (*trigger_3dlut_dma_load)(struct dc *dc, struct pipe_ctx *pipe_ctx);
+	void (*trigger_3dlut_dma_load)(struct pipe_ctx *pipe_ctx);
 
 	/* VM Related */
 	int (*init_sys_ctx)(struct dce_hwseq *hws,
diff --git a/drivers/gpu/drm/amd/display/dc/hwss/hw_sequencer_private.h b/drivers/gpu/drm/amd/display/dc/hwss/hw_sequencer_private.h
index 31ace62a37d9..0f7881a21462 100644
--- a/drivers/gpu/drm/amd/display/dc/hwss/hw_sequencer_private.h
+++ b/drivers/gpu/drm/amd/display/dc/hwss/hw_sequencer_private.h
@@ -58,7 +58,6 @@ struct dc_state;
 struct dc_stream_status;
 struct dc_writeback_info;
 struct dchub_init_data;
-struct dc_plane_cm;
 struct dc_static_screen_params;
 struct resource_pool;
 struct resource_context;
@@ -218,10 +217,6 @@ struct hwseq_private_funcs {
 	void (*reset_back_end_for_pipe)(struct dc *dc,
 			struct pipe_ctx *pipe_ctx,
 			struct dc_state *context);
-	void (*populate_mcm_luts)(struct dc *dc,
-			struct pipe_ctx *pipe_ctx,
-			const struct dc_plane_cm *cm,
-			bool lut_bank_a);
 	void (*perform_3dlut_wa_unlock)(struct pipe_ctx *pipe_ctx);
 	void (*wait_for_pipe_update_if_needed)(struct dc *dc, struct pipe_ctx *pipe_ctx, bool is_surface_update_only);
 	void (*set_wait_for_update_needed_for_pipe)(struct dc *dc, struct pipe_ctx *pipe_ctx);
diff --git a/drivers/gpu/drm/amd/display/dc/inc/hw/hubp.h b/drivers/gpu/drm/amd/display/dc/inc/hw/hubp.h
index 6d6eda0e7e9d..9dc9317c28a4 100644
--- a/drivers/gpu/drm/amd/display/dc/inc/hw/hubp.h
+++ b/drivers/gpu/drm/amd/display/dc/inc/hw/hubp.h
@@ -89,7 +89,7 @@ enum hubp_3dlut_fl_addressing_mode {
 enum hubp_3dlut_fl_width {
 	hubp_3dlut_fl_width_17 = 17,
 	hubp_3dlut_fl_width_33 = 33,
-	hubp_3dlut_fl_width_transformed    = 4916, //mpc default
+	hubp_3dlut_fl_width_17_transformed    = 4916, //mpc default
 };
 
 enum hubp_3dlut_fl_crossbar_bit_slice {
@@ -99,22 +99,6 @@ enum hubp_3dlut_fl_crossbar_bit_slice {
 	hubp_3dlut_fl_crossbar_bit_slice_48_63 = 3
 };
 
-struct hubp_fl_3dlut_config {
-	bool enabled;
-	enum hubp_3dlut_fl_width width;
-	enum hubp_3dlut_fl_mode mode;
-	enum hubp_3dlut_fl_format format;
-	uint16_t bias;
-	uint16_t scale;
-	struct dc_plane_address address;
-	enum hubp_3dlut_fl_addressing_mode addr_mode;
-	enum dc_cm_lut_swizzle layout;
-	uint8_t protection_bits;
-	enum hubp_3dlut_fl_crossbar_bit_slice crossbar_bit_slice_y_g;
-	enum hubp_3dlut_fl_crossbar_bit_slice crossbar_bit_slice_cb_b;
-	enum hubp_3dlut_fl_crossbar_bit_slice crossbar_bit_slice_cr_r;
-};
-
 struct hubp {
 	const struct hubp_funcs *funcs;
 	struct dc_context *ctx;
@@ -289,24 +273,15 @@ struct hubp_funcs {
 
 	void (*hubp_wait_pipe_read_start)(struct hubp *hubp);
 	void (*hubp_program_mcache_id_and_split_coordinate)(struct hubp *hubp, struct dml2_hubp_pipe_mcache_regs *mcache_regs);
-	void (*hubp_update_3dlut_fl_bias_scale)(struct hubp *hubp, uint16_t bias, uint16_t scale);
-	void (*hubp_program_3dlut_fl_mode)(struct hubp *hubp,
-			enum hubp_3dlut_fl_mode mode);
-	void (*hubp_program_3dlut_fl_format)(struct hubp *hubp,
-			enum hubp_3dlut_fl_format format);
 	void (*hubp_program_3dlut_fl_addr)(struct hubp *hubp,
-		const struct dc_plane_address address);
+		const struct dc_plane_address *address);
+	void (*hubp_program_3dlut_fl_config)(struct hubp *hubp,
+		const struct dc_3dlut_dma *config);
 	void (*hubp_program_3dlut_fl_dlg_param)(struct hubp *hubp, int refcyc_per_3dlut_group);
 	void (*hubp_enable_3dlut_fl)(struct hubp *hubp, bool enable);
-	void (*hubp_program_3dlut_fl_addressing_mode)(struct hubp *hubp, enum hubp_3dlut_fl_addressing_mode addr_mode);
-	void (*hubp_program_3dlut_fl_width)(struct hubp *hubp, enum hubp_3dlut_fl_width width);
-	void (*hubp_program_3dlut_fl_tmz_protected)(struct hubp *hubp, uint8_t protection_bits);
 	void (*hubp_program_3dlut_fl_crossbar)(struct hubp *hubp,
-			enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_y_g,
-			enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_cb_b,
-			enum hubp_3dlut_fl_crossbar_bit_slice bit_slice_cr_r);
+			enum dc_cm_lut_pixel_format format);
 	uint32_t (*hubp_get_3dlut_fl_done)(struct hubp *hubp);
-	void (*hubp_program_3dlut_fl_config)(struct hubp *hubp, struct hubp_fl_3dlut_config *cfg);
 	void (*hubp_clear_tiling)(struct hubp *hubp);
 	uint32_t (*hubp_get_current_read_line)(struct hubp *hubp);
 	uint32_t (*hubp_get_det_config_error)(struct hubp *hubp);
diff --git a/drivers/gpu/drm/amd/display/dc/inc/hw/mpc.h b/drivers/gpu/drm/amd/display/dc/inc/hw/mpc.h
index 54eb2eba68bf..d863f4bdccad 100644
--- a/drivers/gpu/drm/amd/display/dc/inc/hw/mpc.h
+++ b/drivers/gpu/drm/amd/display/dc/inc/hw/mpc.h
@@ -54,6 +54,7 @@
 #include "dc_hw_types.h"
 #include "hw_shared.h"
 #include "transform.h"
+#include "dc_types.h"
 
 #define MAX_MPCC 6
 #define MAX_OPP 6
@@ -101,13 +102,6 @@ enum mpcc_movable_cm_location {
 	MPCC_MOVABLE_CM_LOCATION_AFTER,
 };
 
-enum MCM_LUT_XABLE {
-	MCM_LUT_DISABLE,
-	MCM_LUT_DISABLED = MCM_LUT_DISABLE,
-	MCM_LUT_ENABLE,
-	MCM_LUT_ENABLED = MCM_LUT_ENABLE,
-};
-
 enum MCM_LUT_ID {
 	MCM_LUT_3DLUT,
 	MCM_LUT_1DLUT,
@@ -116,7 +110,7 @@ enum MCM_LUT_ID {
 
 struct mpc_fl_3dlut_config {
 	bool enabled;
-	uint16_t width;
+	enum dc_cm_lut_size size;
 	bool select_lut_bank_a;
 	uint16_t bit_depth;
 	int hubp_index;
@@ -1075,8 +1069,11 @@ struct mpc_funcs {
 	*
 	* void
 	*/
-	void (*populate_lut)(struct mpc *mpc, const enum MCM_LUT_ID id, const union mcm_lut_params params,
-			bool lut_bank_a, int mpcc_id);
+	void (*populate_lut)(struct mpc *mpc,
+			const enum MCM_LUT_ID id,
+			const union mcm_lut_params *params,
+			const bool lut_bank_a,
+			const int mpcc_id);
 
 	/**
 	* @program_lut_read_write_control:
@@ -1087,13 +1084,18 @@ struct mpc_funcs {
 	* - [in/out] mpc - MPC context.
 	* - [in] id
 	* - [in] lut_bank_a
+	* - [in] bit_depth
 	* - [in] mpcc_id
 	*
 	* Return:
 	*
 	* void
 	*/
-	void (*program_lut_read_write_control)(struct mpc *mpc, const enum MCM_LUT_ID id, bool lut_bank_a, int mpcc_id);
+	void (*program_lut_read_write_control)(struct mpc *mpc,
+		const enum MCM_LUT_ID id,
+		const bool lut_bank_a,
+		const unsigned int bit_depth,
+		const int mpcc_id);
 
 	/**
 	* @program_lut_mode:
@@ -1103,33 +1105,44 @@ struct mpc_funcs {
 	* Parameters:
 	* - [in/out] mpc - MPC context.
 	* - [in] id
-	* - [in] xable
+	* - [in] enable
 	* - [in] lut_bank_a
+	* - [in] size
 	* - [in] mpcc_id
 	*
 	* Return:
 	*
 	* void
 	*/
-	void (*program_lut_mode)(struct mpc *mpc, const enum MCM_LUT_ID id, const enum MCM_LUT_XABLE xable,
-			bool lut_bank_a, int mpcc_id);
+	void (*program_lut_mode)(struct mpc *mpc,
+			const enum MCM_LUT_ID id,
+			const bool enable,
+			const bool lut_bank_a,
+			const enum dc_cm_lut_size size,
+			const int mpcc_id);
 
-	/**
-	 * @mcm:
-	 *
-	 * MPC MCM new HW sequential programming functions
-	 */
-	struct {
-		void (*program_3dlut_size)(struct mpc *mpc, uint32_t width, int mpcc_id);
-		void (*program_bias_scale)(struct mpc *mpc, uint16_t bias, uint16_t scale, int mpcc_id);
-		void (*program_bit_depth)(struct mpc *mpc, uint16_t bit_depth, int mpcc_id);
-		bool (*is_config_supported)(uint32_t width);
-		void (*program_lut_read_write_control)(struct mpc *mpc, const enum MCM_LUT_ID id,
-			bool lut_bank_a, bool enabled, int mpcc_id);
 
-		void (*populate_lut)(struct mpc *mpc, const union mcm_lut_params params,
-			bool lut_bank_a, int mpcc_id);
-	} mcm;
+	/**
+	* @get_lut_mode:
+	*
+	* Obtains enablement and ram bank status.
+	*
+	* Parameters:
+	* - [in/out] mpc - MPC context.
+	* - [in] id
+	* - [in] mpcc_id
+	* - [out] enable
+	* - [out] lut_bank_a
+	*
+	* Return:
+	*
+	* void
+	*/
+	void (*get_lut_mode)(struct mpc *mpc,
+			const enum MCM_LUT_ID id,
+			const int mpcc_id,
+			bool *enable,
+			bool *lut_bank_a);
 
 	/**
 	 * @rmcm:
@@ -1142,9 +1155,11 @@ struct mpc_funcs {
 		void (*update_3dlut_fast_load_select)(struct mpc *mpc, int mpcc_id, int hubp_idx);
 		void (*program_lut_read_write_control)(struct mpc *mpc, const enum MCM_LUT_ID id,
 			bool lut_bank_a, bool enabled, int mpcc_id);
-		void (*program_lut_mode)(struct mpc *mpc, const enum MCM_LUT_XABLE xable,
-			bool lut_bank_a, int mpcc_id);
-		void (*program_3dlut_size)(struct mpc *mpc, uint32_t width, int mpcc_id);
+		void (*program_lut_mode)(struct mpc *mpc,
+			bool enable,
+			bool lut_bank_a,
+			int mpcc_id);
+		void (*program_3dlut_size)(struct mpc *mpc, const enum dc_cm_lut_size size, int mpcc_id);
 		void (*program_bias_scale)(struct mpc *mpc, uint16_t bias, uint16_t scale, int mpcc_id);
 		void (*program_bit_depth)(struct mpc *mpc, uint16_t bit_depth, int mpcc_id);
 		bool (*is_config_supported)(uint32_t width);
@@ -1152,6 +1167,8 @@ struct mpc_funcs {
 		void (*power_on_shaper_3dlut)(struct mpc *mpc, uint32_t mpcc_id, bool power_on);
 		void (*populate_lut)(struct mpc *mpc, const union mcm_lut_params params,
 			bool lut_bank_a, int mpcc_id);
+		void (*get_3dlut_mode)(struct mpc *mpc, int mpcc_id,
+			bool *enable, bool *lut_bank_a);
 	} rmcm;
 };
 
diff --git a/drivers/gpu/drm/amd/display/dc/mpc/dcn401/dcn401_mpc.c b/drivers/gpu/drm/amd/display/dc/mpc/dcn401/dcn401_mpc.c
index d8a4e604a34e..df92d8008fe6 100644
--- a/drivers/gpu/drm/amd/display/dc/mpc/dcn401/dcn401_mpc.c
+++ b/drivers/gpu/drm/amd/display/dc/mpc/dcn401/dcn401_mpc.c
@@ -73,56 +73,15 @@ void mpc401_set_movable_cm_location(struct mpc *mpc, enum mpcc_movable_cm_locati
 	}
 }
 
-static enum dc_lut_mode get3dlut_config(
-			struct mpc *mpc,
-			bool *is_17x17x17,
-			bool *is_12bits_color_channel,
-			int mpcc_id)
-{
-	uint32_t i_mode, i_enable_10bits, lut_size;
-	enum dc_lut_mode mode;
-	struct dcn401_mpc *mpc401 = TO_DCN401_MPC(mpc);
-
-	REG_GET(MPCC_MCM_3DLUT_MODE[mpcc_id],
-			MPCC_MCM_3DLUT_MODE_CURRENT,  &i_mode);
-
-	REG_GET(MPCC_MCM_3DLUT_READ_WRITE_CONTROL[mpcc_id],
-			MPCC_MCM_3DLUT_30BIT_EN, &i_enable_10bits);
-
-	switch (i_mode) {
-	case 0:
-		mode = LUT_BYPASS;
-		break;
-	case 1:
-		mode = LUT_RAM_A;
-		break;
-	case 2:
-		mode = LUT_RAM_B;
-		break;
-	default:
-		mode = LUT_BYPASS;
-		break;
-	}
-	if (i_enable_10bits > 0)
-		*is_12bits_color_channel = false;
-	else
-		*is_12bits_color_channel = true;
-
-	REG_GET(MPCC_MCM_3DLUT_MODE[mpcc_id], MPCC_MCM_3DLUT_SIZE, &lut_size);
-
-	if (lut_size == 0)
-		*is_17x17x17 = true;
-	else
-		*is_17x17x17 = false;
-
-	return mode;
-}
-
-void mpc401_populate_lut(struct mpc *mpc, const enum MCM_LUT_ID id, const union mcm_lut_params params, bool lut_bank_a, int mpcc_id)
+void mpc401_populate_lut(struct mpc *mpc,
+		const enum MCM_LUT_ID id,
+		const union mcm_lut_params *params,
+		const bool lut_bank_a,
+		const int mpcc_id)
 {
 	const enum dc_lut_mode next_mode = lut_bank_a ? LUT_RAM_A : LUT_RAM_B;
-	const struct pwl_params *lut1d = params.pwl;
-	const struct pwl_params *lut_shaper = params.pwl;
+	const struct pwl_params *lut1d = params->pwl;
+	const struct pwl_params *lut_shaper = params->pwl;
 	bool is_17x17x17;
 	bool is_12bits_color_channel;
 	const struct dc_rgb *lut0;
@@ -131,7 +90,7 @@ void mpc401_populate_lut(struct mpc *mpc, const enum MCM_LUT_ID id, const union
 	const struct dc_rgb *lut3;
 	int lut_size0;
 	int lut_size;
-	const struct tetrahedral_params *lut3d = params.lut3d;
+	const struct tetrahedral_params *lut3d = params->lut3d;
 
 	switch (id) {
 	case MCM_LUT_1DLUT:
@@ -174,8 +133,6 @@ void mpc401_populate_lut(struct mpc *mpc, const enum MCM_LUT_ID id, const union
 
 		mpc32_power_on_shaper_3dlut(mpc, mpcc_id, true);
 
-		get3dlut_config(mpc, &is_17x17x17, &is_12bits_color_channel, mpcc_id);
-
 		is_17x17x17 = !lut3d->use_tetrahedral_9;
 		is_12bits_color_channel = lut3d->use_12bits;
 		if (is_17x17x17) {
@@ -198,8 +155,6 @@ void mpc401_populate_lut(struct mpc *mpc, const enum MCM_LUT_ID id, const union
 					sizeof(lut3d->tetrahedral_9.lut1[0]);
 			}
 
-		mpc32_select_3dlut_ram(mpc, next_mode,
-					is_12bits_color_channel, mpcc_id);
 		mpc32_select_3dlut_ram_mask(mpc, 0x1, mpcc_id);
 		if (is_12bits_color_channel)
 			mpc32_set3dlut_ram12(mpc, lut0, lut_size0, mpcc_id);
@@ -232,46 +187,69 @@ void mpc401_populate_lut(struct mpc *mpc, const enum MCM_LUT_ID id, const union
 
 }
 
+static uint32_t mpc401_cm_lut_size_to_3dlut_size(const enum dc_cm_lut_size cm_size)
+{
+	uint32_t size = 0;
+
+	switch (cm_size) {
+	case CM_LUT_SIZE_999:
+		size = 1;
+		break;
+	case CM_LUT_SIZE_171717:
+		size = 0;
+		break;
+	default:
+		/* invalid LUT size */
+		ASSERT(false);
+		size = 0;
+		break;
+	}
+
+	return size;
+}
+
 void mpc401_program_lut_mode(
 		struct mpc *mpc,
 		const enum MCM_LUT_ID id,
-		const enum MCM_LUT_XABLE xable,
-		bool lut_bank_a,
-		int mpcc_id)
+		const bool enable,
+		const bool lut_bank_a,
+		const enum dc_cm_lut_size size,
+		const int mpcc_id)
 {
+	uint32_t lut_size;
 	struct dcn401_mpc *mpc401 = TO_DCN401_MPC(mpc);
 
 	switch (id) {
 	case MCM_LUT_3DLUT:
-		switch (xable) {
-		case MCM_LUT_DISABLE:
+		if (enable) {
+			lut_size = mpc401_cm_lut_size_to_3dlut_size(size);
+			REG_UPDATE_2(MPCC_MCM_3DLUT_MODE[mpcc_id],
+					MPCC_MCM_3DLUT_MODE, lut_bank_a ? 1 : 2,
+					MPCC_MCM_3DLUT_SIZE, lut_size);
+		} else {
+			if (mpc->ctx->dc->debug.enable_mem_low_power.bits.mpc)
+				mpc32_power_on_shaper_3dlut(mpc, mpcc_id, false);
 			REG_UPDATE(MPCC_MCM_3DLUT_MODE[mpcc_id], MPCC_MCM_3DLUT_MODE, 0);
-			break;
-		case MCM_LUT_ENABLE:
-			REG_UPDATE(MPCC_MCM_3DLUT_MODE[mpcc_id], MPCC_MCM_3DLUT_MODE, lut_bank_a ? 1 : 2);
-			break;
 		}
 		break;
 	case MCM_LUT_SHAPER:
-		switch (xable) {
-		case MCM_LUT_DISABLE:
-			REG_UPDATE(MPCC_MCM_SHAPER_CONTROL[mpcc_id], MPCC_MCM_SHAPER_LUT_MODE, 0);
-			break;
-		case MCM_LUT_ENABLE:
+		if (enable) {
 			REG_UPDATE(MPCC_MCM_SHAPER_CONTROL[mpcc_id], MPCC_MCM_SHAPER_LUT_MODE, lut_bank_a ? 1 : 2);
-			break;
+		} else {
+			if (mpc->ctx->dc->debug.enable_mem_low_power.bits.mpc)
+				mpc32_power_on_shaper_3dlut(mpc, mpcc_id, false);
+			REG_UPDATE(MPCC_MCM_SHAPER_CONTROL[mpcc_id], MPCC_MCM_SHAPER_LUT_MODE, 0);
 		}
 		break;
 	case MCM_LUT_1DLUT:
-		switch (xable) {
-		case MCM_LUT_DISABLE:
-			REG_UPDATE(MPCC_MCM_1DLUT_CONTROL[mpcc_id],
-					MPCC_MCM_1DLUT_MODE, 0);
-			break;
-		case MCM_LUT_ENABLE:
+		if (enable) {
 			REG_UPDATE(MPCC_MCM_1DLUT_CONTROL[mpcc_id],
 					MPCC_MCM_1DLUT_MODE, 2);
-			break;
+		} else {
+			if (mpc->ctx->dc->debug.enable_mem_low_power.bits.mpc)
+				mpc32_power_on_blnd_lut(mpc, mpcc_id, false);
+			REG_UPDATE(MPCC_MCM_1DLUT_CONTROL[mpcc_id],
+					MPCC_MCM_1DLUT_MODE, 0);
 		}
 		REG_UPDATE(MPCC_MCM_1DLUT_CONTROL[mpcc_id],
 				MPCC_MCM_1DLUT_SELECT, lut_bank_a ? 0 : 1);
@@ -279,14 +257,20 @@ void mpc401_program_lut_mode(
 	}
 }
 
-void mpc401_program_lut_read_write_control(struct mpc *mpc, const enum MCM_LUT_ID id, bool lut_bank_a, int mpcc_id)
+void mpc401_program_lut_read_write_control(struct mpc *mpc,
+		const enum MCM_LUT_ID id,
+		const bool lut_bank_a,
+		const unsigned int bit_depth,
+		const int mpcc_id)
 {
 	struct dcn401_mpc *mpc401 = TO_DCN401_MPC(mpc);
 
 	switch (id) {
 	case MCM_LUT_3DLUT:
 		mpc32_select_3dlut_ram_mask(mpc, 0xf, mpcc_id);
-		REG_UPDATE(MPCC_MCM_3DLUT_READ_WRITE_CONTROL[mpcc_id], MPCC_MCM_3DLUT_RAM_SEL, lut_bank_a ? 0 : 1);
+		REG_UPDATE_2(MPCC_MCM_3DLUT_READ_WRITE_CONTROL[mpcc_id],
+				MPCC_MCM_3DLUT_30BIT_EN, (bit_depth == 10) ? 1 : 0,
+				MPCC_MCM_3DLUT_RAM_SEL, lut_bank_a ? 0 : 1);
 		break;
 	case MCM_LUT_SHAPER:
 		mpc32_configure_shaper_lut(mpc, lut_bank_a, mpcc_id);
@@ -605,6 +589,44 @@ void mpc401_get_gamut_remap(struct mpc *mpc,
 			coef_format, ARRAY_SIZE(arr_reg_val));
 }
 
+void mpc401_get_lut_mode(struct mpc *mpc,
+		const enum MCM_LUT_ID id,
+		const int mpcc_id,
+		bool *enable,
+		bool *lut_bank_a)
+{
+	struct dcn401_mpc *mpc401 = TO_DCN401_MPC(mpc);
+
+	uint32_t lut_mode = 0;
+	uint32_t lut_select = 0;
+
+	*enable = false;
+	*lut_bank_a = true;
+
+	switch (id) {
+	case MCM_LUT_SHAPER:
+		REG_GET(MPCC_MCM_SHAPER_CONTROL[mpcc_id],
+				MPCC_MCM_SHAPER_MODE_CURRENT, &lut_mode);
+		*enable = lut_mode != 0;
+		*lut_bank_a = lut_mode != 2;
+		break;
+	case MCM_LUT_1DLUT:
+		REG_GET_2(MPCC_MCM_1DLUT_CONTROL[mpcc_id],
+				MPCC_MCM_1DLUT_MODE_CURRENT, &lut_mode,
+				MPCC_MCM_1DLUT_SELECT_CURRENT, &lut_select);
+		*enable = lut_mode != 0;
+		*lut_bank_a = lut_mode == 0 || lut_select == 0;
+		break;
+	case MCM_LUT_3DLUT:
+	default:
+		REG_GET(MPCC_MCM_3DLUT_MODE[mpcc_id],
+				MPCC_MCM_3DLUT_MODE_CURRENT, &lut_mode);
+		*enable = lut_mode != 0;
+		*lut_bank_a = lut_mode != 2;
+		break;
+	}
+}
+
 static const struct mpc_funcs dcn401_mpc_funcs = {
 	.read_mpcc_state = mpc1_read_mpcc_state,
 	.insert_plane = mpc1_insert_plane,
@@ -643,6 +665,7 @@ static const struct mpc_funcs dcn401_mpc_funcs = {
 	.populate_lut = mpc401_populate_lut,
 	.program_lut_read_write_control = mpc401_program_lut_read_write_control,
 	.program_lut_mode = mpc401_program_lut_mode,
+	.get_lut_mode = mpc401_get_lut_mode,
 };
 
 
diff --git a/drivers/gpu/drm/amd/display/dc/mpc/dcn401/dcn401_mpc.h b/drivers/gpu/drm/amd/display/dc/mpc/dcn401/dcn401_mpc.h
index 3d00d9497ed2..75fa05ed2b22 100644
--- a/drivers/gpu/drm/amd/display/dc/mpc/dcn401/dcn401_mpc.h
+++ b/drivers/gpu/drm/amd/display/dc/mpc/dcn401/dcn401_mpc.h
@@ -206,21 +206,32 @@ void dcn401_mpc_construct(struct dcn401_mpc *mpc401,
 	int num_rmu);
 
 void mpc401_set_movable_cm_location(struct mpc *mpc, enum mpcc_movable_cm_location location, int mpcc_id);
-void mpc401_populate_lut(struct mpc *mpc, const enum MCM_LUT_ID id, const union mcm_lut_params params,
-		bool lut_bank_a, int mpcc_id);
+void mpc401_populate_lut(struct mpc *mpc,
+		const enum MCM_LUT_ID id,
+		const union mcm_lut_params *params,
+		bool lut_bank_a,
+		int mpcc_id);
 
 void mpc401_program_lut_mode(
 		struct mpc *mpc,
 		const enum MCM_LUT_ID id,
-		const enum MCM_LUT_XABLE xable,
-		bool lut_bank_a,
-		int mpcc_id);
+		const bool enable,
+		const bool lut_bank_a,
+		const enum dc_cm_lut_size size,
+		const int mpcc_id);
+
+void mpc401_get_lut_mode(struct mpc *mpc,
+		const enum MCM_LUT_ID id,
+		const int mpcc_id,
+		bool *enable,
+		bool *lut_bank_a);
 
 void mpc401_program_lut_read_write_control(
 		struct mpc *mpc,
 		const enum MCM_LUT_ID id,
-		bool lut_bank_a,
-		int mpcc_id);
+		const bool lut_bank_a,
+		const unsigned int bit_depth,
+		const int mpcc_id);
 
 void mpc401_set_gamut_remap(
 	struct mpc *mpc,
diff --git a/drivers/gpu/drm/amd/display/dc/mpc/dcn42/dcn42_mpc.c b/drivers/gpu/drm/amd/display/dc/mpc/dcn42/dcn42_mpc.c
index 38c0e8f96d40..5a888e1e1cc5 100644
--- a/drivers/gpu/drm/amd/display/dc/mpc/dcn42/dcn42_mpc.c
+++ b/drivers/gpu/drm/amd/display/dc/mpc/dcn42/dcn42_mpc.c
@@ -63,154 +63,6 @@ void mpc42_update_blending(
 	mpcc->blnd_cfg = *blnd_cfg;
 }
 
-/* Shaper functions */
-void mpc42_power_on_shaper_3dlut(
-	struct mpc *mpc,
-	uint32_t mpcc_id,
-	bool power_on)
-{
-	uint32_t power_status_shaper = 2;
-	uint32_t power_status_3dlut  = 2;
-	struct dcn42_mpc *mpc42 = TO_DCN42_MPC(mpc);
-	int max_retries = 10;
-
-	REG_SET(MPCC_MCM_MEM_PWR_CTRL[mpcc_id], 0,
-		MPCC_MCM_3DLUT_MEM_PWR_DIS, power_on == true ? 1:0);
-	REG_SET(MPCC_MCM_MEM_PWR_CTRL[mpcc_id], 0,
-		MPCC_MCM_SHAPER_MEM_PWR_DIS, power_on == true ? 1:0);
-	/* wait for memory to fully power up */
-	if (power_on && mpc->ctx->dc->debug.enable_mem_low_power.bits.mpc) {
-		REG_WAIT(MPCC_MCM_MEM_PWR_CTRL[mpcc_id], MPCC_MCM_SHAPER_MEM_PWR_STATE, 0, 1, max_retries);
-		REG_WAIT(MPCC_MCM_MEM_PWR_CTRL[mpcc_id], MPCC_MCM_3DLUT_MEM_PWR_STATE, 0, 1, max_retries);
-	}
-
-	/*read status is not mandatory, it is just for debugging*/
-	REG_GET(MPCC_MCM_MEM_PWR_CTRL[mpcc_id], MPCC_MCM_SHAPER_MEM_PWR_STATE, &power_status_shaper);
-	REG_GET(MPCC_MCM_MEM_PWR_CTRL[mpcc_id], MPCC_MCM_3DLUT_MEM_PWR_STATE, &power_status_3dlut);
-
-	if (power_status_shaper != 0 && power_on == true)
-		BREAK_TO_DEBUGGER();
-
-	if (power_status_3dlut != 0 && power_on == true)
-		BREAK_TO_DEBUGGER();
-}
-
-void mpc42_configure_shaper_lut(
-	struct mpc *mpc,
-	bool is_ram_a,
-	uint32_t mpcc_id)
-{
-	struct dcn42_mpc *mpc42 = TO_DCN42_MPC(mpc);
-
-	REG_UPDATE(MPCC_MCM_SHAPER_SCALE_G_B[mpcc_id],
-		MPCC_MCM_SHAPER_SCALE_B, 0x7000);
-	REG_UPDATE(MPCC_MCM_SHAPER_SCALE_G_B[mpcc_id],
-		MPCC_MCM_SHAPER_SCALE_G, 0x7000);
-	REG_UPDATE(MPCC_MCM_SHAPER_SCALE_R[mpcc_id],
-		MPCC_MCM_SHAPER_SCALE_R, 0x7000);
-	REG_UPDATE(MPCC_MCM_SHAPER_LUT_WRITE_EN_MASK[mpcc_id],
-			MPCC_MCM_SHAPER_LUT_WRITE_EN_MASK, 7);
-	REG_UPDATE(MPCC_MCM_SHAPER_LUT_WRITE_EN_MASK[mpcc_id],
-			MPCC_MCM_SHAPER_LUT_WRITE_SEL, is_ram_a == true ? 0:1);
-	REG_SET(MPCC_MCM_SHAPER_LUT_INDEX[mpcc_id], 0, MPCC_MCM_SHAPER_LUT_INDEX, 0);
-}
-
-
-void mpc42_program_3dlut_size(struct mpc *mpc, uint32_t width, int mpcc_id)
-{
-	struct dcn42_mpc *mpc42 = TO_DCN42_MPC(mpc);
-	uint32_t size = 0xff;
-
-	REG_GET(MPCC_MCM_3DLUT_MODE[mpcc_id], MPCC_MCM_3DLUT_SIZE, &size);
-
-	REG_UPDATE(MPCC_MCM_3DLUT_MODE[mpcc_id], MPCC_MCM_3DLUT_SIZE,
-		(width == 33) ? 2 :
-		(width == 17) ? 0 : 2);
-
-	REG_GET(MPCC_MCM_3DLUT_MODE[mpcc_id], MPCC_MCM_3DLUT_SIZE, &size);
-}
-
-void mpc42_program_3dlut_fl_bias_scale(struct mpc *mpc, uint16_t bias, uint16_t scale, int mpcc_id)
-{
-	struct dcn42_mpc *mpc42 = TO_DCN42_MPC(mpc);
-
-	REG_UPDATE_2(MPCC_MCM_3DLUT_OUT_OFFSET_R[mpcc_id],
-		MPCC_MCM_3DLUT_OUT_OFFSET_R, bias,
-		MPCC_MCM_3DLUT_OUT_SCALE_R, scale);
-
-	REG_UPDATE_2(MPCC_MCM_3DLUT_OUT_OFFSET_G[mpcc_id],
-		MPCC_MCM_3DLUT_OUT_OFFSET_G, bias,
-		MPCC_MCM_3DLUT_OUT_SCALE_G, scale);
-
-	REG_UPDATE_2(MPCC_MCM_3DLUT_OUT_OFFSET_B[mpcc_id],
-		MPCC_MCM_3DLUT_OUT_OFFSET_B, bias,
-		MPCC_MCM_3DLUT_OUT_SCALE_B, scale);
-}
-
-void mpc42_program_bit_depth(struct mpc *mpc, uint16_t bit_depth, int mpcc_id)
-{
-	struct dcn42_mpc *mpc42 = TO_DCN42_MPC(mpc);
-
-	REG_UPDATE(MPCC_MCM_3DLUT_READ_WRITE_CONTROL[mpcc_id], MPCC_MCM_3DLUT_WRITE_EN_MASK, 0xF);
-
-	//program bit_depth
-	REG_UPDATE(MPCC_MCM_3DLUT_READ_WRITE_CONTROL[mpcc_id],
-		MPCC_MCM_3DLUT_30BIT_EN,
-		(bit_depth == 10) ? 1 : 0);
-}
-
-bool mpc42_is_config_supported(uint32_t width)
-{
-	if (width == 17)
-		return true;
-
-	return false;
-}
-
-void mpc42_populate_lut(struct mpc *mpc, const union mcm_lut_params params,
-	bool lut_bank_a, int mpcc_id)
-{
-	const enum dc_lut_mode next_mode = lut_bank_a ? LUT_RAM_A : LUT_RAM_B;
-	const struct pwl_params *lut_shaper = params.pwl;
-
-	if (lut_shaper == NULL)
-		return;
-	if (mpc->ctx->dc->debug.enable_mem_low_power.bits.mpc)
-		mpc42_power_on_shaper_3dlut(mpc, mpcc_id, true);
-
-	mpc42_configure_shaper_lut(mpc, next_mode == LUT_RAM_A, mpcc_id);
-
-	if (next_mode == LUT_RAM_A)
-		mpc32_program_shaper_luta_settings(mpc, lut_shaper, mpcc_id);
-	else
-		mpc32_program_shaper_lutb_settings(mpc, lut_shaper, mpcc_id);
-
-	mpc32_program_shaper_lut(
-			mpc, lut_shaper->rgb_resulted, lut_shaper->hw_points_num, mpcc_id);
-
-	mpc42_power_on_shaper_3dlut(mpc, mpcc_id, false);
-}
-
-void mpc42_program_lut_read_write_control(struct mpc *mpc, const enum MCM_LUT_ID id,
-	bool lut_bank_a, bool enabled, int mpcc_id)
-{
-	struct dcn42_mpc *mpc42 = TO_DCN42_MPC(mpc);
-
-	switch (id) {
-	case MCM_LUT_3DLUT:
-		REG_UPDATE(MPCC_MCM_3DLUT_MODE[mpcc_id], MPCC_MCM_3DLUT_MODE,
-			(!enabled) ? 0 :
-			(lut_bank_a) ? 1 : 2);
-		REG_UPDATE(MPCC_MCM_3DLUT_READ_WRITE_CONTROL[mpcc_id], MPCC_MCM_3DLUT_RAM_SEL, lut_bank_a ? 0 : 1);
-		break;
-	case MCM_LUT_SHAPER:
-		mpc32_configure_shaper_lut(mpc, lut_bank_a, mpcc_id);
-		break;
-	default:
-		break;
-	}
-}
-
 /* RMCM Shaper functions */
 void mpc42_power_on_rmcm_shaper_3dlut(
 	struct mpc *mpc,
@@ -674,32 +526,47 @@ void mpc42_program_rmcm_lut_read_write_control(struct mpc *mpc, const enum MCM_L
 	}
 }
 
-void mpc42_program_lut_mode(struct mpc *mpc, const enum MCM_LUT_XABLE xable,
-	bool lut_bank_a, int mpcc_id)
+void mpc42_program_lut_mode(struct mpc *mpc,
+	bool enable,
+	bool lut_bank_a,
+	int mpcc_id)
 {
 	struct dcn42_mpc *mpc42 = TO_DCN42_MPC(mpc);
 
-	switch (xable) {
-	case MCM_LUT_DISABLE:
+	if (enable) {
+		REG_UPDATE(MPC_RMCM_SHAPER_CONTROL[mpcc_id], MPC_RMCM_SHAPER_LUT_MODE, lut_bank_a ? 1 : 2);
+	} else {
 		REG_UPDATE(MPC_RMCM_SHAPER_CONTROL[mpcc_id], MPC_RMCM_SHAPER_LUT_MODE, 0);
+	}
+}
+
+static uint32_t mpc42_get_rmcm_3dlut_width(
+		const enum dc_cm_lut_size size)
+{
+	uint32_t width = 0;
+
+	switch (size) {
+	case CM_LUT_SIZE_333333:
+		width = 2;
 		break;
-	case MCM_LUT_ENABLE:
-		REG_UPDATE(MPC_RMCM_SHAPER_CONTROL[mpcc_id], MPC_RMCM_SHAPER_LUT_MODE, lut_bank_a ? 1 : 2);
+	case CM_LUT_SIZE_171717:
+	default:
+		width = 0;
 		break;
 	}
+
+	return width;
 }
 
-void mpc42_program_rmcm_3dlut_size(struct mpc *mpc, uint32_t width, int mpcc_id)
+void mpc42_program_rmcm_3dlut_size(struct mpc *mpc,
+		const enum dc_cm_lut_size size,
+		int mpcc_id)
 {
 	struct dcn42_mpc *mpc42 = TO_DCN42_MPC(mpc);
-	uint32_t size = 0xff;
-
-	REG_GET(MPC_RMCM_3DLUT_MODE[mpcc_id], MPC_RMCM_3DLUT_SIZE, &size);
-
-	REG_UPDATE(MPC_RMCM_3DLUT_MODE[mpcc_id], MPC_RMCM_3DLUT_SIZE,
-		(width == 33) ? 2 : 0);
+	uint32_t width = mpc42_get_rmcm_3dlut_width(size);
 
-	REG_GET(MPC_RMCM_3DLUT_MODE[mpcc_id], MPC_RMCM_3DLUT_SIZE, &size);
+	REG_UPDATE(MPC_RMCM_3DLUT_MODE[mpcc_id],
+			MPC_RMCM_3DLUT_SIZE, width);
 }
 
 void mpc42_program_rmcm_3dlut_fast_load_bias_scale(struct mpc *mpc, uint16_t bias, uint16_t scale, int mpcc_id)
@@ -731,14 +598,6 @@ void mpc42_program_rmcm_bit_depth(struct mpc *mpc, uint16_t bit_depth, int mpcc_
 		(bit_depth == 10) ? 1 : 0);
 }
 
-bool mpc42_is_rmcm_config_supported(uint32_t width)
-{
-	if (width == 17 || width == 33)
-		return true;
-
-	return false;
-}
-
 void mpc42_set_fl_config(
 	struct mpc *mpc,
 	struct mpc_fl_3dlut_config *cfg,
@@ -746,6 +605,7 @@ void mpc42_set_fl_config(
 {
 	struct dcn42_mpc *mpc42 = TO_DCN42_MPC(mpc);
 
+	uint32_t width = mpc42_get_rmcm_3dlut_width(cfg->size);
 	/*
 	From: Jie Zhou
 
@@ -786,7 +646,7 @@ void mpc42_set_fl_config(
 
 	//width
 	REG_UPDATE_2(MPC_RMCM_3DLUT_MODE[mpcc_id],
-		MPC_RMCM_3DLUT_SIZE, (cfg->width == 33) ? 2 : 0,
+		MPC_RMCM_3DLUT_SIZE, width,
 		MPC_RMCM_3DLUT_MODE, (!cfg->enabled) ? 0 : (cfg->select_lut_bank_a) ? 1 : 2);
 
 	//connect to hubp
@@ -799,181 +659,23 @@ void mpc42_set_fl_config(
 	REG_UPDATE(MPC_RMCM_CNTL[mpcc_id], MPC_RMCM_CNTL, cfg->enabled ? 0 : 0xF);
 }
 
-//static void rmcm_program_gamut_remap(
-//	struct mpc *mpc,
-//	unsigned int mpcc_id,
-//	const uint16_t *regval,
-//	enum mpcc_gamut_remap_id gamut_remap_block_id,
-//	enum mpcc_gamut_remap_mode_select mode_select)
-//{
-//	struct color_matrices_reg gamut_regs;
-//	struct dcn42_mpc *mpc42 = TO_DCN42_MPC(mpc);
-//
-//	if (gamut_remap_block_id == MPCC_OGAM_GAMUT_REMAP ||
-//		gamut_remap_block_id == MPCC_MCM_FIRST_GAMUT_REMAP ||
-//		gamut_remap_block_id == MPCC_MCM_SECOND_GAMUT_REMAP) {
-//		mpc_program_gamut_remap(mpc, mpcc_id, regval, gamut_remap_block_id, mode_select);
-//		return;
-//	}
-//	if (gamut_remap_block_id == MPCC_OGAM_GAMUT_REMAP) {
-//
-//		if (regval == NULL || mode_select == MPCC_GAMUT_REMAP_MODE_SELECT_0) {
-//			REG_SET(MPC_RMCM_GAMUT_REMAP_MODE[mpcc_id], 0,
-//				MPC_RMCM_GAMUT_REMAP_MODE, mode_select);
-//			return;
-//		}
-//
-//		gamut_regs.shifts.csc_c11 = mpc42->mpc_shift->MPCC_GAMUT_REMAP_C11_A;
-//		gamut_regs.masks.csc_c11 = mpc42->mpc_mask->MPCC_GAMUT_REMAP_C11_A;
-//		gamut_regs.shifts.csc_c12 = mpc42->mpc_shift->MPCC_GAMUT_REMAP_C12_A;
-//		gamut_regs.masks.csc_c12 = mpc42->mpc_mask->MPCC_GAMUT_REMAP_C12_A;
-//
-//		switch (mode_select) {
-//		case MPCC_GAMUT_REMAP_MODE_SELECT_1:
-//			gamut_regs.csc_c11_c12 = REG(MPC_RMCM_GAMUT_REMAP_C11_C12_A[mpcc_id]);
-//			gamut_regs.csc_c33_c34 = REG(MPC_RMCM_GAMUT_REMAP_C33_C34_A[mpcc_id]);
-//			break;
-//		case MPCC_GAMUT_REMAP_MODE_SELECT_2:
-//			gamut_regs.csc_c11_c12 = REG(MPC_RMCM_GAMUT_REMAP_C11_C12_B[mpcc_id]);
-//			gamut_regs.csc_c33_c34 = REG(MPC_RMCM_GAMUT_REMAP_C33_C34_B[mpcc_id]);
-//			break;
-//		default:
-//			break;
-//		}
-//
-//		cm_helper_program_color_matrices(
-//			mpc->ctx,
-//			regval,
-//			&gamut_regs);
-//
-//		//select coefficient set to use, set A (MODE_1) or set B (MODE_2)
-//		REG_SET(MPC_RMCM_GAMUT_REMAP_MODE[mpcc_id], 0, MPC_RMCM_GAMUT_REMAP_MODE, mode_select);
-//	}
-//}
-
-//static bool is_mpc_legacy_gamut_id(enum mpcc_gamut_remap_id gamut_remap_block_id)
-//{
-//	if (gamut_remap_block_id == MPCC_OGAM_GAMUT_REMAP ||
-//		gamut_remap_block_id == MPCC_MCM_FIRST_GAMUT_REMAP ||
-//		gamut_remap_block_id == MPCC_MCM_SECOND_GAMUT_REMAP) {
-//		return true;
-//	}
-//	return false;
-//}
-//static void program_gamut_remap(
-//	struct mpc *mpc,
-//	unsigned int mpcc_id,
-//	const uint16_t *regval,
-//	enum mpcc_gamut_remap_id gamut_remap_block_id,
-//	enum mpcc_gamut_remap_mode_select mode_select)
-//{
-//	if (is_mpc_legacy_gamut_id(gamut_remap_block_id))
-//		mpc_program_gamut_remap(mpc, mpcc_id, regval, gamut_remap_block_id, mode_select);
-//	else
-//		rmcm_program_gamut_remap(mpc, mpcc_id, regval, gamut_remap_block_id, mode_select);
-//}
-
-//void mpc42_set_gamut_remap(
-//	struct mpc *mpc,
-//	int mpcc_id,
-//	const struct mpc_grph_gamut_adjustment *adjust)
-//{
-//	struct dcn42_mpc *mpc42 = TO_DCN42_MPC(mpc);
-//	unsigned int i = 0;
-//	uint32_t mode_select = 0;
-//
-//	if (adjust->gamut_adjust_type != GRAPHICS_GAMUT_ADJUST_TYPE_SW) {
-//		/* Bypass / Disable if type is bypass or hw */
-//		program_gamut_remap(mpc, mpcc_id, NULL,
-//			adjust->mpcc_gamut_remap_block_id, MPCC_GAMUT_REMAP_MODE_SELECT_0);
-//	} else {
-//		struct fixed31_32 arr_matrix[12];
-//		uint16_t arr_reg_val[12];
-//
-//		for (i = 0; i < 12; i++)
-//			arr_matrix[i] = adjust->temperature_matrix[i];
-//
-//		convert_float_matrix(arr_reg_val, arr_matrix, 12);
-//
-//		if (is_mpc_legacy_gamut_id(adjust->mpcc_gamut_remap_block_id))
-//			REG_GET(MPCC_GAMUT_REMAP_MODE[mpcc_id],
-//				MPCC_GAMUT_REMAP_MODE_CURRENT, &mode_select);
-//		else
-//			REG_GET(MPC_RMCM_GAMUT_REMAP_MODE[mpcc_id],
-//				MPC_RMCM_GAMUT_REMAP_MODE_CURRENT, &mode_select);
-//
-//		//If current set in use not set A (MODE_1), then use set A, otherwise use set B
-//		if (mode_select != MPCC_GAMUT_REMAP_MODE_SELECT_1)
-//			mode_select = MPCC_GAMUT_REMAP_MODE_SELECT_1;
-//		else
-//			mode_select = MPCC_GAMUT_REMAP_MODE_SELECT_2;
-//
-//		program_gamut_remap(mpc, mpcc_id, arr_reg_val,
-//			adjust->mpcc_gamut_remap_block_id, mode_select);
-//	}
-//}
-
-//static void read_gamut_remap(struct mpc *mpc,
-//	int mpcc_id,
-//	uint16_t *regval,
-//	enum mpcc_gamut_remap_id gamut_remap_block_id,
-//	uint32_t *mode_select)
-//{
-//	struct color_matrices_reg gamut_regs = {0};
-//	struct dcn42_mpc *mpc42 = TO_DCN42_MPC(mpc);
-//
-//	if (is_mpc_legacy_gamut_id(gamut_remap_block_id)) {
-//		mpc_read_gamut_remap(mpc, mpcc_id, regval, gamut_remap_block_id, mode_select);
-//	}
-//	if (gamut_remap_block_id == MPCC_RMCM_GAMUT_REMAP) {
-//		//current coefficient set in use
-//		REG_GET(MPC_RMCM_GAMUT_REMAP_MODE[mpcc_id], MPC_RMCM_GAMUT_REMAP_MODE, mode_select);
-//
-//		gamut_regs.shifts.csc_c11 = mpc42->mpc_shift->MPCC_GAMUT_REMAP_C11_A;
-//		gamut_regs.masks.csc_c11 = mpc42->mpc_mask->MPCC_GAMUT_REMAP_C11_A;
-//		gamut_regs.shifts.csc_c12 = mpc42->mpc_shift->MPCC_GAMUT_REMAP_C12_A;
-//		gamut_regs.masks.csc_c12 = mpc42->mpc_mask->MPCC_GAMUT_REMAP_C12_A;
-//
-//		switch (*mode_select) {
-//		case MPCC_GAMUT_REMAP_MODE_SELECT_1:
-//			gamut_regs.csc_c11_c12 = REG(MPC_RMCM_GAMUT_REMAP_C11_C12_A[mpcc_id]);
-//			gamut_regs.csc_c33_c34 = REG(MPC_RMCM_GAMUT_REMAP_C33_C34_A[mpcc_id]);
-//			break;
-//		case MPCC_GAMUT_REMAP_MODE_SELECT_2:
-//			gamut_regs.csc_c11_c12 = REG(MPC_RMCM_GAMUT_REMAP_C11_C12_B[mpcc_id]);
-//			gamut_regs.csc_c33_c34 = REG(MPC_RMCM_GAMUT_REMAP_C33_C34_B[mpcc_id]);
-//			break;
-//		default:
-//			break;
-//		}
-//	}
-//
-//	if (*mode_select != MPCC_GAMUT_REMAP_MODE_SELECT_0) {
-//		cm_helper_read_color_matrices(
-//			mpc42->base.ctx,
-//			regval,
-//			&gamut_regs);
-//	}
-//}
-
-//void mpc42_get_gamut_remap(struct mpc *mpc,
-//	int mpcc_id,
-//	struct mpc_grph_gamut_adjustment *adjust)
-//{
-//	uint16_t arr_reg_val[12] = {0};
-//	uint32_t mode_select;
-//
-//	read_gamut_remap(mpc, mpcc_id, arr_reg_val, adjust->mpcc_gamut_remap_block_id, &mode_select);
-//
-//	if (mode_select == MPCC_GAMUT_REMAP_MODE_SELECT_0) {
-//		adjust->gamut_adjust_type = GRAPHICS_GAMUT_ADJUST_TYPE_BYPASS;
-//		return;
-//	}
-//
-//	adjust->gamut_adjust_type = GRAPHICS_GAMUT_ADJUST_TYPE_SW;
-//	convert_hw_matrix(adjust->temperature_matrix,
-//		arr_reg_val, ARRAY_SIZE(arr_reg_val));
-//}
+void mpc42_get_rmcm_3dlut_mode(
+	struct mpc *mpc,
+	int mpcc_id,
+	bool *enable,
+	bool *lut_bank_a)
+{
+	struct dcn42_mpc *mpc42 = TO_DCN42_MPC(mpc);
+	uint32_t mode_current   = 0;
+
+	REG_GET(MPC_RMCM_3DLUT_MODE[mpcc_id], MPC_RMCM_3DLUT_MODE_CURRENT, &mode_current);
+
+	/* MPC_RMCM_3DLUT_MODE encoding:
+	 *   0 -> disabled, 1 -> bank A, 2 -> bank B
+	 */
+	*enable     = mode_current != 0;
+	*lut_bank_a = mode_current != 2;
+}
 
 void mpc42_read_mpcc_state(
 		struct mpc *mpc,
@@ -1071,14 +773,7 @@ static const struct mpc_funcs dcn42_mpc_funcs = {
 	.populate_lut = mpc401_populate_lut,
 	.program_lut_read_write_control = mpc401_program_lut_read_write_control,
 	.program_lut_mode = mpc401_program_lut_mode,
-	.mcm = {
-		.program_lut_read_write_control = mpc42_program_lut_read_write_control,
-		.program_3dlut_size = mpc42_program_3dlut_size,
-		.program_bias_scale = mpc42_program_3dlut_fl_bias_scale,
-		.program_bit_depth = mpc42_program_bit_depth,
-		.is_config_supported = mpc42_is_config_supported,
-		.populate_lut = mpc42_populate_lut,
-	},
+	.get_lut_mode = mpc401_get_lut_mode,
 	.rmcm = {
 		.enable_3dlut_fl = mpc42_enable_3dlut_fl,
 		.update_3dlut_fast_load_select = mpc42_update_3dlut_fast_load_select,
@@ -1087,10 +782,10 @@ static const struct mpc_funcs dcn42_mpc_funcs = {
 		.program_3dlut_size = mpc42_program_rmcm_3dlut_size,
 		.program_bias_scale = mpc42_program_rmcm_3dlut_fast_load_bias_scale,
 		.program_bit_depth = mpc42_program_rmcm_bit_depth,
-		.is_config_supported = mpc42_is_rmcm_config_supported,
 		.power_on_shaper_3dlut = mpc42_power_on_rmcm_shaper_3dlut,
 		.populate_lut = mpc42_populate_rmcm_lut,
 		.fl_3dlut_configure = mpc42_set_fl_config,
+		.get_3dlut_mode = mpc42_get_rmcm_3dlut_mode,
 	},
 };
 
diff --git a/drivers/gpu/drm/amd/display/dc/mpc/dcn42/dcn42_mpc.h b/drivers/gpu/drm/amd/display/dc/mpc/dcn42/dcn42_mpc.h
index 12a12c28e553..af3b90af5396 100644
--- a/drivers/gpu/drm/amd/display/dc/mpc/dcn42/dcn42_mpc.h
+++ b/drivers/gpu/drm/amd/display/dc/mpc/dcn42/dcn42_mpc.h
@@ -884,49 +884,6 @@ void dcn42_mpc_construct(struct dcn42_mpc *mpc401,
 
 void mpc42_init_mpcc(struct mpcc *mpcc, int mpcc_inst);
 
-void mpc42_program_shaper_lutb_settings(
-	struct mpc *mpc,
-	const struct pwl_params *params,
-	uint32_t mpcc_id);
-void mpc42_program_shaper_luta_settings(
-		struct mpc *mpc,
-		const struct pwl_params *params,
-		uint32_t mpcc_id);
-void mpc42_configure_shaper_lut(
-		struct mpc *mpc,
-		bool is_ram_a,
-		uint32_t mpcc_id);
-void mpc42_power_on_shaper_3dlut(
-		struct mpc *mpc,
-		uint32_t mpcc_id,
-		bool power_on);
-void mpc42_program_3dlut_size(
-		struct mpc *mpc,
-		uint32_t width,
-		int mpcc_id);
-void mpc42_program_3dlut_fl_bias_scale(
-		struct mpc *mpc,
-		uint16_t bias,
-		uint16_t scale,
-		int mpcc_id);
-void mpc42_program_bit_depth(
-		struct mpc *mpc,
-		uint16_t bit_depth,
-		int mpcc_id);
-void mpc42_populate_lut(
-		struct mpc *mpc,
-		const union mcm_lut_params params,
-		bool lut_bank_a,
-		int mpcc_id);
-void mpc42_program_lut_read_write_control(
-		struct mpc *mpc,
-		const enum MCM_LUT_ID id,
-		bool lut_bank_a,
-		bool enabled,
-		int mpcc_id);
-
-bool mpc42_is_config_supported(uint32_t width);
-
 /* RMCM */
 void mpc42_program_rmcm_shaper_lut(
 		struct mpc *mpc,
@@ -970,12 +927,12 @@ void mpc42_program_rmcm_lut_read_write_control(
 		int mpcc_id);
 void mpc42_program_lut_mode(
 		struct mpc *mpc,
-		const enum MCM_LUT_XABLE xable,
+		bool enable,
 		bool lut_bank_a,
 		int mpcc_id);
 void mpc42_program_rmcm_3dlut_size(
 		struct mpc *mpc,
-		uint32_t width,
+		const enum dc_cm_lut_size size,
 		int mpcc_id);
 void mpc42_program_rmcm_3dlut_fast_load_bias_scale(
 		struct mpc *mpc,
@@ -987,13 +944,17 @@ void mpc42_program_rmcm_bit_depth(
 		uint16_t bit_depth,
 		int mpcc_id);
 
-bool mpc42_is_rmcm_config_supported(uint32_t width);
-
 void mpc42_set_fl_config(
 	struct mpc *mpc,
 	struct mpc_fl_3dlut_config *cfg,
 	int mpcc_id);
 
+void mpc42_get_rmcm_3dlut_mode(
+	struct mpc *mpc,
+	int mpcc_id,
+	bool *enable,
+	bool *lut_bank_a);
+
 void mpc42_read_mpcc_state(
 		struct mpc *mpc,
 		int mpcc_inst,
diff --git a/drivers/gpu/drm/amd/display/dc/mpc/dcn60/dcn60_mpc.c b/drivers/gpu/drm/amd/display/dc/mpc/dcn60/dcn60_mpc.c
index f8a878ead67d..2c0b92854a9c 100644
--- a/drivers/gpu/drm/amd/display/dc/mpc/dcn60/dcn60_mpc.c
+++ b/drivers/gpu/drm/amd/display/dc/mpc/dcn60/dcn60_mpc.c
@@ -139,165 +139,123 @@ void mpc60_program_rmcm_lut_read_write_control(struct mpc *mpc, const enum MCM_L
 	}
 }
 
-void mpc60_select_3dlut_ram(
-	struct mpc *mpc,
-	enum dc_lut_mode mode,
-	bool is_color_channel_12bits,
-	uint32_t mpcc_id)
+static void mpc60_program_lut_read_write_control(struct mpc *mpc,
+		const enum MCM_LUT_ID id,
+		const bool lut_bank_a,
+		const unsigned int bit_depth,
+		const int mpcc_id)
 {
-	(void)mode;
 	struct dcn60_mpc *mpc60 = TO_DCN60_MPC(mpc);
 
-	REG_UPDATE(MPCC_MCM_3DLUT_READ_WRITE_CONTROL[mpcc_id],
-		MPCC_MCM_3DLUT_30BIT_EN, is_color_channel_12bits == true ? 0 : 1);
+	switch (id) {
+	case MCM_LUT_3DLUT:
+		mpc32_select_3dlut_ram_mask(mpc, 0xf, mpcc_id);
+		REG_UPDATE(MPCC_MCM_3DLUT_READ_WRITE_CONTROL[mpcc_id],
+				MPCC_MCM_3DLUT_30BIT_EN, (bit_depth == 10) ? 1 : 0);
+		break;
+	case MCM_LUT_SHAPER:
+		mpc32_configure_shaper_lut(mpc, lut_bank_a, mpcc_id);
+		break;
+	case MCM_LUT_1DLUT:
+		mpc32_configure_post1dlut(mpc, lut_bank_a, mpcc_id);
+		break;
+	}
 }
 
-static enum dc_lut_mode get3dlut_config(
-	struct mpc *mpc,
-	bool *is_17x17x17,
-	bool *is_12bits_color_channel,
-	int mpcc_id)
+static uint32_t mpc60_cm_lut_size_to_3dlut_size(const enum dc_cm_lut_size cm_size)
 {
-	uint32_t i_mode, i_enable_10bits, lut_size;
-	enum dc_lut_mode mode;
-	struct dcn60_mpc *mpc60 = TO_DCN60_MPC(mpc);
+	uint32_t size = 0;
 
-	REG_GET(MPCC_MCM_3DLUT_MODE[mpcc_id],
-		MPCC_MCM_3DLUT_MODE_CURRENT, &i_mode);
-
-	REG_GET(MPCC_MCM_3DLUT_READ_WRITE_CONTROL[mpcc_id],
-		MPCC_MCM_3DLUT_30BIT_EN, &i_enable_10bits);
-
-	switch (i_mode) {
-	case 0:
-		mode = LUT_BYPASS;
-		break;
-	case 1:
-		mode = LUT_RAM_A;
+	switch (cm_size) {
+	case CM_LUT_SIZE_999:
+		size = 1;
 		break;
-	case 2:
-		mode = LUT_RAM_B;
+	case CM_LUT_SIZE_171717:
+		size = 0;
 		break;
 	default:
-		mode = LUT_BYPASS;
+		/* invalid LUT size for MCM */
+		ASSERT(false);
+		size = 0;
 		break;
 	}
-	if (i_enable_10bits > 0)
-		*is_12bits_color_channel = false;
-	else
-		*is_12bits_color_channel = true;
-
-	REG_GET(MPCC_MCM_3DLUT_MODE[mpcc_id], MPCC_MCM_3DLUT_SIZE, &lut_size);
-
-	if (lut_size == 0)
-		*is_17x17x17 = true;
-	else
-		*is_17x17x17 = false;
-
-	return mode;
-}
-
-bool mpc60_program_3dlut(
-	struct mpc *mpc,
-	const struct tetrahedral_params *params,
-	int mpcc_id)
-{
-	enum dc_lut_mode mode;
-	bool is_17x17x17;
-	bool is_12bits_color_channel;
-	const struct dc_rgb *lut0;
-	const struct dc_rgb *lut1;
-	const struct dc_rgb *lut2;
-	const struct dc_rgb *lut3;
-	int lut_size0;
-	int lut_size;
-
-	if (params == NULL) {
-		mpc32_set_3dlut_mode(mpc, LUT_BYPASS, false, false, mpcc_id);
-		return false;
-	}
-	mpc32_power_on_shaper_3dlut(mpc, mpcc_id, true);
-
-	mode = get3dlut_config(mpc, &is_17x17x17, &is_12bits_color_channel, mpcc_id);
-
-	if (mode == LUT_BYPASS || mode == LUT_RAM_B)
-		mode = LUT_RAM_A;
-	else
-		mode = LUT_RAM_B;
-
-	is_17x17x17 = !params->use_tetrahedral_9;
-	is_12bits_color_channel = params->use_12bits;
-	if (is_17x17x17) {
-		lut0 = params->tetrahedral_17.lut0;
-		lut1 = params->tetrahedral_17.lut1;
-		lut2 = params->tetrahedral_17.lut2;
-		lut3 = params->tetrahedral_17.lut3;
-		lut_size0 = sizeof(params->tetrahedral_17.lut0) /
-			sizeof(params->tetrahedral_17.lut0[0]);
-		lut_size = sizeof(params->tetrahedral_17.lut1) /
-			sizeof(params->tetrahedral_17.lut1[0]);
-	} else {
-		lut0 = params->tetrahedral_9.lut0;
-		lut1 = params->tetrahedral_9.lut1;
-		lut2 = params->tetrahedral_9.lut2;
-		lut3 = params->tetrahedral_9.lut3;
-		lut_size0 = sizeof(params->tetrahedral_9.lut0) /
-			sizeof(params->tetrahedral_9.lut0[0]);
-		lut_size = sizeof(params->tetrahedral_9.lut1) /
-			sizeof(params->tetrahedral_9.lut1[0]);
-	}
 
-	mpc60_select_3dlut_ram(mpc, mode,
-				is_12bits_color_channel, mpcc_id);
-	mpc32_select_3dlut_ram_mask(mpc, 0x1, mpcc_id);
-	if (is_12bits_color_channel)
-		mpc32_set3dlut_ram12(mpc, lut0, lut_size0, mpcc_id);
-	else
-		mpc32_set3dlut_ram10(mpc, lut0, lut_size0, mpcc_id);
-
-	mpc32_select_3dlut_ram_mask(mpc, 0x2, mpcc_id);
-	if (is_12bits_color_channel)
-		mpc32_set3dlut_ram12(mpc, lut1, lut_size, mpcc_id);
-	else
-		mpc32_set3dlut_ram10(mpc, lut1, lut_size, mpcc_id);
-
-	mpc32_select_3dlut_ram_mask(mpc, 0x4, mpcc_id);
-	if (is_12bits_color_channel)
-		mpc32_set3dlut_ram12(mpc, lut2, lut_size, mpcc_id);
-	else
-		mpc32_set3dlut_ram10(mpc, lut2, lut_size, mpcc_id);
-
-	mpc32_select_3dlut_ram_mask(mpc, 0x8, mpcc_id);
-	if (is_12bits_color_channel)
-		mpc32_set3dlut_ram12(mpc, lut3, lut_size, mpcc_id);
-	else
-		mpc32_set3dlut_ram10(mpc, lut3, lut_size, mpcc_id);
-
-	mpc32_set_3dlut_mode(mpc, mode, is_12bits_color_channel,
-		is_17x17x17, mpcc_id);
-
-	if (mpc->ctx->dc->debug.enable_mem_low_power.bits.mpc)
-		mpc32_power_on_shaper_3dlut(mpc, mpcc_id, false);
-
-	return true;
+	return size;
 }
 
-void mpc60_program_lut_read_write_control(struct mpc *mpc, const enum MCM_LUT_ID id, bool lut_bank_a, int mpcc_id)
+static void mpc60_program_lut_mode(
+		struct mpc *mpc,
+		const enum MCM_LUT_ID id,
+		const bool enable,
+		const bool lut_bank_a,
+		const enum dc_cm_lut_size size,
+		const int mpcc_id)
 {
+	uint32_t lut_size;
+	struct dcn60_mpc *mpc60 = TO_DCN60_MPC(mpc);
 
 	switch (id) {
 	case MCM_LUT_3DLUT:
-		mpc32_select_3dlut_ram_mask(mpc, 0xf, mpcc_id);
+		if (enable) {
+			lut_size = mpc60_cm_lut_size_to_3dlut_size(size);
+			REG_UPDATE_2(MPCC_MCM_3DLUT_MODE[mpcc_id],
+					MPCC_MCM_3DLUT_MODE, 1,
+					MPCC_MCM_3DLUT_SIZE, lut_size);
+		} else {
+			if (mpc->ctx->dc->debug.enable_mem_low_power.bits.mpc)
+				mpc32_power_on_shaper_3dlut(mpc, mpcc_id, false);
+			REG_UPDATE(MPCC_MCM_3DLUT_MODE[mpcc_id], MPCC_MCM_3DLUT_MODE, 0);
+		}
 		break;
 	case MCM_LUT_SHAPER:
-		mpc32_configure_shaper_lut(mpc, lut_bank_a, mpcc_id);
+		if (enable) {
+			REG_UPDATE(MPCC_MCM_SHAPER_CONTROL[mpcc_id], MPCC_MCM_SHAPER_LUT_MODE, lut_bank_a ? 1 : 2);
+		} else {
+			if (mpc->ctx->dc->debug.enable_mem_low_power.bits.mpc)
+				mpc32_power_on_shaper_3dlut(mpc, mpcc_id, false);
+			REG_UPDATE(MPCC_MCM_SHAPER_CONTROL[mpcc_id], MPCC_MCM_SHAPER_LUT_MODE, 0);
+		}
 		break;
 	case MCM_LUT_1DLUT:
-		mpc32_configure_post1dlut(mpc, lut_bank_a, mpcc_id);
+		if (enable) {
+			REG_UPDATE(MPCC_MCM_1DLUT_CONTROL[mpcc_id],
+					MPCC_MCM_1DLUT_MODE, 2);
+		} else {
+			if (mpc->ctx->dc->debug.enable_mem_low_power.bits.mpc)
+				mpc32_power_on_blnd_lut(mpc, mpcc_id, false);
+			REG_UPDATE(MPCC_MCM_1DLUT_CONTROL[mpcc_id],
+					MPCC_MCM_1DLUT_MODE, 0);
+		}
+		REG_UPDATE(MPCC_MCM_1DLUT_CONTROL[mpcc_id],
+				MPCC_MCM_1DLUT_SELECT, lut_bank_a ? 0 : 1);
 		break;
 	}
 }
 
+static bool mpc60_program_3dlut(
+	struct mpc *mpc,
+	const struct tetrahedral_params *params,
+	int mpcc_id)
+{
+	union mcm_lut_params lut_params = { 0 };
+	lut_params.lut3d = params;
+
+	mpc60_program_lut_read_write_control(mpc,
+			MCM_LUT_3DLUT,
+			true,
+			params->use_12bits ? 12 : 10,
+			mpcc_id);
+	mpc401_populate_lut(mpc, MCM_LUT_3DLUT, &lut_params, true, mpcc_id);
+	mpc60_program_lut_mode(mpc,
+		MCM_LUT_3DLUT,
+		true,
+		true,
+		params->use_tetrahedral_9 ? CM_LUT_SIZE_999 : CM_LUT_SIZE_171717,
+		mpcc_id);
+
+	return true;
+}
+
 static const struct mpc_funcs dcn60_mpc_funcs = {
 	.read_mpcc_state = mpc1_read_mpcc_state,
 	.insert_plane = mpc60_insert_plane,
@@ -328,15 +286,8 @@ static const struct mpc_funcs dcn60_mpc_funcs = {
 	.get_3dlut_fast_load_status = mpc401_get_3dlut_fast_load_status,
 	.populate_lut = mpc401_populate_lut,
 	.program_lut_read_write_control = mpc60_program_lut_read_write_control,
-	.program_lut_mode = mpc401_program_lut_mode,
-	.mcm = {
-		.program_lut_read_write_control = mpc42_program_lut_read_write_control,
-		.program_3dlut_size = mpc42_program_3dlut_size,
-		.program_bias_scale = mpc42_program_3dlut_fl_bias_scale,
-		.program_bit_depth = mpc42_program_bit_depth,
-		.is_config_supported = mpc42_is_config_supported,
-		.populate_lut = mpc42_populate_lut,
-	},
+	.program_lut_mode = mpc60_program_lut_mode,
+	.get_lut_mode = mpc401_get_lut_mode,
 	.rmcm = {
 		.enable_3dlut_fl = mpc42_enable_3dlut_fl,
 		.update_3dlut_fast_load_select = mpc42_update_3dlut_fast_load_select,
@@ -345,9 +296,9 @@ static const struct mpc_funcs dcn60_mpc_funcs = {
 		.program_3dlut_size = mpc42_program_rmcm_3dlut_size,
 		.program_bias_scale = mpc42_program_rmcm_3dlut_fast_load_bias_scale,
 		.program_bit_depth = mpc42_program_rmcm_bit_depth,
-		.is_config_supported = mpc42_is_rmcm_config_supported,
 		.power_on_shaper_3dlut = mpc42_power_on_rmcm_shaper_3dlut,
 		.populate_lut = mpc42_populate_rmcm_lut,
+		.get_3dlut_mode = mpc42_get_rmcm_3dlut_mode,
 	},
 };
 
diff --git a/drivers/gpu/drm/amd/display/dc/mpc/dcn60/dcn60_mpc.h b/drivers/gpu/drm/amd/display/dc/mpc/dcn60/dcn60_mpc.h
index 76d5c721f742..640a322d15c1 100644
--- a/drivers/gpu/drm/amd/display/dc/mpc/dcn60/dcn60_mpc.h
+++ b/drivers/gpu/drm/amd/display/dc/mpc/dcn60/dcn60_mpc.h
@@ -469,19 +469,4 @@ void mpc60_program_rmcm_lut_read_write_control(struct mpc *mpc,
 	bool lut_bank_a,
 	bool enabled,
 	int mpcc_id);
-
-void mpc60_select_3dlut_ram(struct mpc *mpc,
-	enum dc_lut_mode mode,
-	bool is_color_channel_12bits,
-	uint32_t mpcc_id);
-
-bool mpc60_program_3dlut(struct mpc *mpc,
-	const struct tetrahedral_params *params,
-	int mpcc_id);
-
-void mpc60_program_lut_read_write_control(struct mpc *mpc,
-	const enum MCM_LUT_ID id,
-	bool lut_bank_a,
-	int mpcc_id);
-
 #endif /* DCN60_MPC_H_ */
diff --git a/drivers/gpu/drm/amd/display/dc/resource/dcn42/dcn42_resource.c b/drivers/gpu/drm/amd/display/dc/resource/dcn42/dcn42_resource.c
index 6296330e2450..6370d3903eb5 100644
--- a/drivers/gpu/drm/amd/display/dc/resource/dcn42/dcn42_resource.c
+++ b/drivers/gpu/drm/amd/display/dc/resource/dcn42/dcn42_resource.c
@@ -704,6 +704,7 @@ static const struct resource_caps res_cap_dcn42 = {
 	.num_vmid = 16,
 	.num_mpc_3dlut = 2,
 	.num_dsc = 4,
+	.num_rmcm = 2,
 };
 
 static const struct dc_plane_cap plane_cap = {
diff --git a/drivers/gpu/drm/amd/display/dc/resource/dcn60/dcn60_resource.c b/drivers/gpu/drm/amd/display/dc/resource/dcn60/dcn60_resource.c
index b441d6d2d36b..4b7668abf4dc 100644
--- a/drivers/gpu/drm/amd/display/dc/resource/dcn60/dcn60_resource.c
+++ b/drivers/gpu/drm/amd/display/dc/resource/dcn60/dcn60_resource.c
@@ -584,6 +584,7 @@ static const struct resource_caps res_cap_dcn6_0 = {
 	.num_mpc_3dlut = 4,
 	.num_dsc = 4,
 	.num_aux = 4,
+	.num_rmcm = 2,
 };
 
 static const struct dc_plane_cap plane_cap = {
-- 
2.34.1