drm/amd/display: add dwb support to dml2

[why]
dwb was not POR previosly.
now need to enable dwb in dml2.

Limitation:
HW DML assumes only one DWB
one set of watermark for all 4 watermark sets
one stream has one DWB only.
WB scaling dml input has one set of scaling tap.
(no chroma so far)

needs to follow up
Reviewed-by: Chris Park <chris.park@amd.com>
Acked-by: Hamza Mahfooz <hamza.mahfooz@amd.com>
Signed-off-by: Charlene Liu <charlene.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/display/dc/dml2/display_mode_core.c

@@ -10214,6 +10214,7 @@ dml_get_var_func(fraction_of_urgent_bandwidth_imm_flip, dml_float_t, mode_lib->m
 dml_get_var_func(urgent_latency, dml_float_t, mode_lib->mp.UrgentLatency);
 dml_get_var_func(clk_dcf_deepsleep, dml_float_t, mode_lib->mp.DCFCLKDeepSleep);
 dml_get_var_func(wm_writeback_dram_clock_change, dml_float_t, mode_lib->mp.Watermark.WritebackDRAMClockChangeWatermark);
+dml_get_var_func(wm_writeback_urgent, dml_float_t, mode_lib->mp.Watermark.WritebackUrgentWatermark);
 dml_get_var_func(stutter_efficiency, dml_float_t, mode_lib->mp.StutterEfficiency);
 dml_get_var_func(stutter_efficiency_no_vblank, dml_float_t, mode_lib->mp.StutterEfficiencyNotIncludingVBlank);
 dml_get_var_func(stutter_efficiency_z8, dml_float_t, mode_lib->mp.Z8StutterEfficiency);

drivers/gpu/drm/amd/display/dc/dml2/display_mode_core.h

@@ -94,6 +94,7 @@ dml_get_var_decl(wm_usr_retraining, dml_float_t);
 dml_get_var_decl(urgent_latency, dml_float_t);
 
 dml_get_var_decl(wm_writeback_dram_clock_change, dml_float_t);
+dml_get_var_decl(wm_writeback_urgent, dml_float_t);
 dml_get_var_decl(stutter_efficiency_no_vblank, dml_float_t);
 dml_get_var_decl(stutter_efficiency, dml_float_t);
 dml_get_var_decl(stutter_efficiency_z8, dml_float_t);

drivers/gpu/drm/amd/display/dc/dml2/display_mode_lib_defines.h

@@ -41,6 +41,7 @@
 #define DCN_DML__VM_PRESENT__1 1
 #define DCN_DML__HOST_VM_PRESENT 1
 #define DCN_DML__HOST_VM_PRESENT__1 1
+#define DCN_DML__DWB 1
 
 #include "dml_depedencies.h"
 
@@ -59,6 +60,7 @@
 #define __DML_NUM_PLANES__                              DCN_DML__NUM_PLANE
 #define __DML_NUM_CURSORS__                             DCN_DML__NUM_CURSOR
 #define __DML_DPP_INVALID__                             0
+#define __DML_NUM_DMB__                                 DCN_DML__DWB
 #define __DML_PIPE_NO_PLANE__                           99
 
 #define __DML_MAX_STATE_ARRAY_SIZE__        DCN_DML__NUM_PWR_STATE

drivers/gpu/drm/amd/display/dc/dml2/dml2_translation_helper.c

@@ -1061,7 +1061,46 @@ static void dml2_populate_pipe_to_plane_index_mapping(struct dml2_context *dml2,
 		plane_index = 0;
 	}
 }
-
+static void populate_dml_writeback_cfg_from_stream_state(struct dml_writeback_cfg_st *out,
+		unsigned int location, const struct dc_stream_state *in)
+{
+	if (in->num_wb_info > 0) {
+		for (int i = 0; i < __DML_NUM_DMB__; i++) {
+			const struct dc_writeback_info *wb_info = &in->writeback_info[i];
+			/*current dml support 1 dwb per stream, limitation*/
+			if (wb_info->wb_enabled) {
+				out->WritebackEnable[location] = wb_info->wb_enabled;
+				out->ActiveWritebacksPerSurface[location] = wb_info->dwb_params.cnv_params.src_width;
+				out->WritebackDestinationWidth[location] = wb_info->dwb_params.dest_width;
+				out->WritebackDestinationHeight[location] = wb_info->dwb_params.dest_height;
+
+				out->WritebackSourceWidth[location] = wb_info->dwb_params.cnv_params.crop_en ?
+					wb_info->dwb_params.cnv_params.crop_width :
+					wb_info->dwb_params.cnv_params.src_width;
+
+				out->WritebackSourceHeight[location] = wb_info->dwb_params.cnv_params.crop_en ?
+					wb_info->dwb_params.cnv_params.crop_height :
+					wb_info->dwb_params.cnv_params.src_height;
+				/*current design does not have chroma scaling, need to follow up*/
+				out->WritebackHTaps[location] = wb_info->dwb_params.scaler_taps.h_taps > 0 ?
+					wb_info->dwb_params.scaler_taps.h_taps : 1;
+				out->WritebackVTaps[location] = wb_info->dwb_params.scaler_taps.v_taps > 0 ?
+					wb_info->dwb_params.scaler_taps.v_taps : 1;
+
+				out->WritebackHRatio[location] = wb_info->dwb_params.cnv_params.crop_en ?
+					(double)wb_info->dwb_params.cnv_params.crop_width /
+						(double)wb_info->dwb_params.dest_width :
+					(double)wb_info->dwb_params.cnv_params.src_width /
+						(double)wb_info->dwb_params.dest_width;
+				out->WritebackVRatio[location] = wb_info->dwb_params.cnv_params.crop_en ?
+					(double)wb_info->dwb_params.cnv_params.crop_height /
+						(double)wb_info->dwb_params.dest_height :
+					(double)wb_info->dwb_params.cnv_params.src_height /
+						(double)wb_info->dwb_params.dest_height;
+			}
+		}
+	}
+}
 void map_dc_state_into_dml_display_cfg(struct dml2_context *dml2, struct dc_state *context, struct dml_display_cfg_st *dml_dispcfg)
 {
 	int i = 0, j = 0, k = 0;
@@ -1106,6 +1145,10 @@ void map_dc_state_into_dml_display_cfg(struct dml2_context *dml2, struct dc_stat
 
 		populate_dml_timing_cfg_from_stream_state(&dml_dispcfg->timing, disp_cfg_stream_location, context->streams[i]);
 		populate_dml_output_cfg_from_stream_state(&dml_dispcfg->output, disp_cfg_stream_location, context->streams[i], current_pipe_context);
+		/*Call site for populate_dml_writeback_cfg_from_stream_state*/
+		populate_dml_writeback_cfg_from_stream_state(&dml_dispcfg->writeback,
+			disp_cfg_stream_location, context->streams[i]);
+
 		switch (context->streams[i]->debug.force_odm_combine_segments) {
 		case 2:
 			dml2->v20.dml_core_ctx.policy.ODMUse[disp_cfg_stream_location] = dml_odm_use_policy_combine_2to1;

drivers/gpu/drm/amd/display/dc/dml2/dml2_utils.c

@@ -404,6 +404,71 @@ void dml2_extract_watermark_set(struct dcn_watermarks *watermark, struct display
 	watermark->cstate_pstate.cstate_exit_z8_ns = dml_get_wm_z8_stutter(dml_core_ctx) * 1000;
 }
 
+unsigned int dml2_calc_max_scaled_time(
+		unsigned int time_per_pixel,
+		enum mmhubbub_wbif_mode mode,
+		unsigned int urgent_watermark)
+{
+	unsigned int time_per_byte = 0;
+	unsigned int total_free_entry = 0xb40;
+	unsigned int buf_lh_capability;
+	unsigned int max_scaled_time;
+
+	if (mode == PACKED_444) /* packed mode 32 bpp */
+		time_per_byte = time_per_pixel/4;
+	else if (mode == PACKED_444_FP16) /* packed mode 64 bpp */
+		time_per_byte = time_per_pixel/8;
+
+	if (time_per_byte == 0)
+		time_per_byte = 1;
+
+	buf_lh_capability = (total_free_entry*time_per_byte*32) >> 6; /* time_per_byte is in u6.6*/
+	max_scaled_time   = buf_lh_capability - urgent_watermark;
+	return max_scaled_time;
+}
+
+void dml2_extract_writeback_wm(struct dc_state *context, struct display_mode_lib_st *dml_core_ctx)
+{
+	int i, j = 0;;
+	struct mcif_arb_params *wb_arb_params = NULL;
+	struct dcn_bw_writeback *bw_writeback = NULL;
+	enum mmhubbub_wbif_mode wbif_mode = PACKED_444_FP16; /*for now*/
+
+	if (context->stream_count != 0) {
+		for (i = 0; i < context->stream_count; i++) {
+			if (context->streams[i]->num_wb_info != 0)
+				j++;
+		}
+	}
+	if (j == 0) /*no dwb */
+		return;
+	for (i = 0; i < __DML_NUM_DMB__; i++) {
+		bw_writeback = &context->bw_ctx.bw.dcn.bw_writeback;
+		wb_arb_params = &context->bw_ctx.bw.dcn.bw_writeback.mcif_wb_arb[i];
+
+		for (j = 0 ; j < 4; j++) {
+			/*current dml only has one set of watermark, need to follow up*/
+			bw_writeback->mcif_wb_arb[i].cli_watermark[j] =
+					dml_get_wm_writeback_urgent(dml_core_ctx) * 1000;
+			bw_writeback->mcif_wb_arb[i].pstate_watermark[j] =
+					dml_get_wm_writeback_dram_clock_change(dml_core_ctx) * 1000;
+		}
+		if (context->res_ctx.pipe_ctx[i].stream->phy_pix_clk != 0) {
+			/* time_per_pixel should be in u6.6 format */
+			bw_writeback->mcif_wb_arb[i].time_per_pixel =
+				(1000000 << 6) / context->res_ctx.pipe_ctx[i].stream->phy_pix_clk;
+		}
+		bw_writeback->mcif_wb_arb[i].slice_lines = 32;
+		bw_writeback->mcif_wb_arb[i].arbitration_slice = 2;
+		bw_writeback->mcif_wb_arb[i].max_scaled_time =
+			dml2_calc_max_scaled_time(wb_arb_params->time_per_pixel,
+					wbif_mode, 	wb_arb_params->cli_watermark[0]);
+		/*not required any more*/
+		bw_writeback->mcif_wb_arb[i].dram_speed_change_duration =
+			dml_get_wm_writeback_dram_clock_change(dml_core_ctx) * 1000;
+
+	}
+}
 void dml2_initialize_det_scratch(struct dml2_context *in_ctx)
 {
 	int i;

drivers/gpu/drm/amd/display/dc/dml2/dml2_utils.h

@@ -40,9 +40,14 @@ void dml2_util_copy_dml_output(struct dml_output_cfg_st *dml_output_array, unsig
 unsigned int dml2_util_get_maximum_odm_combine_for_output(bool force_odm_4to1, enum dml_output_encoder_class encoder, bool dsc_enabled);
 void dml2_copy_clocks_to_dc_state(struct dml2_dcn_clocks *out_clks, struct dc_state *context);
 void dml2_extract_watermark_set(struct dcn_watermarks *watermark, struct display_mode_lib_st *dml_core_ctx);
+void dml2_extract_writeback_wm(struct dc_state *context, struct display_mode_lib_st *dml_core_ctx);
 int dml2_helper_find_dml_pipe_idx_by_stream_id(struct dml2_context *ctx, unsigned int stream_id);
 bool is_dtbclk_required(const struct dc *dc, struct dc_state *context);
 bool dml2_is_stereo_timing(const struct dc_stream_state *stream);
+unsigned int dml2_calc_max_scaled_time(
+		unsigned int time_per_pixel,
+		enum mmhubbub_wbif_mode mode,
+		unsigned int urgent_watermark);
 
 /*
  * dml2_dc_construct_pipes - This function will determine if we need additional pipes based

drivers/gpu/drm/amd/display/dc/dml2/dml2_wrapper.c

@@ -640,6 +640,7 @@ static bool dml2_validate_and_build_resource(const struct dc *in_dc, struct dc_s
 		dml2_extract_watermark_set(&context->bw_ctx.bw.dcn.watermarks.b, &dml2->v20.dml_core_ctx);
 		memcpy(&context->bw_ctx.bw.dcn.watermarks.c, &dml2->v20.g6_temp_read_watermark_set, sizeof(context->bw_ctx.bw.dcn.watermarks.c));
 		dml2_extract_watermark_set(&context->bw_ctx.bw.dcn.watermarks.d, &dml2->v20.dml_core_ctx);
+		dml2_extract_writeback_wm(context, &dml2->v20.dml_core_ctx);
 		//copy for deciding zstate use
 		context->bw_ctx.dml.vba.StutterPeriod = context->bw_ctx.dml2->v20.dml_core_ctx.mp.StutterPeriod;