]> asedeno.scripts.mit.edu Git - linux.git/blobdiff - drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c
drm/amd/display: update odm mode validation to be in line with policy
[linux.git] / drivers / gpu / drm / amd / display / dc / dcn20 / dcn20_resource.c
index b949e202d6cb71c468daa17d3a8e80b7e018e3be..f57c686398feda7a5796b9ebfdfa747734cb30a9 100644 (file)
@@ -82,6 +82,7 @@
 
 #include "amdgpu_socbb.h"
 
+/* NV12 SOC BB is currently in FW, mark SW bounding box invalid. */
 #define SOC_BOUNDING_BOX_VALID false
 #define DC_LOGGER_INIT(logger)
 
@@ -156,8 +157,119 @@ struct _vcs_dpi_ip_params_st dcn2_0_ip = {
        .xfc_fill_constant_bytes = 0,
 };
 
-struct _vcs_dpi_soc_bounding_box_st dcn2_0_soc = { 0 };
-
+struct _vcs_dpi_soc_bounding_box_st dcn2_0_soc = {
+       /* Defaults that get patched on driver load from firmware. */
+       .clock_limits = {
+                       {
+                               .state = 0,
+                               .dcfclk_mhz = 560.0,
+                               .fabricclk_mhz = 560.0,
+                               .dispclk_mhz = 513.0,
+                               .dppclk_mhz = 513.0,
+                               .phyclk_mhz = 540.0,
+                               .socclk_mhz = 560.0,
+                               .dscclk_mhz = 171.0,
+                               .dram_speed_mts = 8960.0,
+                       },
+                       {
+                               .state = 1,
+                               .dcfclk_mhz = 694.0,
+                               .fabricclk_mhz = 694.0,
+                               .dispclk_mhz = 642.0,
+                               .dppclk_mhz = 642.0,
+                               .phyclk_mhz = 600.0,
+                               .socclk_mhz = 694.0,
+                               .dscclk_mhz = 214.0,
+                               .dram_speed_mts = 11104.0,
+                       },
+                       {
+                               .state = 2,
+                               .dcfclk_mhz = 875.0,
+                               .fabricclk_mhz = 875.0,
+                               .dispclk_mhz = 734.0,
+                               .dppclk_mhz = 734.0,
+                               .phyclk_mhz = 810.0,
+                               .socclk_mhz = 875.0,
+                               .dscclk_mhz = 245.0,
+                               .dram_speed_mts = 14000.0,
+                       },
+                       {
+                               .state = 3,
+                               .dcfclk_mhz = 1000.0,
+                               .fabricclk_mhz = 1000.0,
+                               .dispclk_mhz = 1100.0,
+                               .dppclk_mhz = 1100.0,
+                               .phyclk_mhz = 810.0,
+                               .socclk_mhz = 1000.0,
+                               .dscclk_mhz = 367.0,
+                               .dram_speed_mts = 16000.0,
+                       },
+                       {
+                               .state = 4,
+                               .dcfclk_mhz = 1200.0,
+                               .fabricclk_mhz = 1200.0,
+                               .dispclk_mhz = 1284.0,
+                               .dppclk_mhz = 1284.0,
+                               .phyclk_mhz = 810.0,
+                               .socclk_mhz = 1200.0,
+                               .dscclk_mhz = 428.0,
+                               .dram_speed_mts = 16000.0,
+                       },
+                       /*Extra state, no dispclk ramping*/
+                       {
+                               .state = 5,
+                               .dcfclk_mhz = 1200.0,
+                               .fabricclk_mhz = 1200.0,
+                               .dispclk_mhz = 1284.0,
+                               .dppclk_mhz = 1284.0,
+                               .phyclk_mhz = 810.0,
+                               .socclk_mhz = 1200.0,
+                               .dscclk_mhz = 428.0,
+                               .dram_speed_mts = 16000.0,
+                       },
+               },
+       .num_states = 5,
+       .sr_exit_time_us = 8.6,
+       .sr_enter_plus_exit_time_us = 10.9,
+       .urgent_latency_us = 4.0,
+       .urgent_latency_pixel_data_only_us = 4.0,
+       .urgent_latency_pixel_mixed_with_vm_data_us = 4.0,
+       .urgent_latency_vm_data_only_us = 4.0,
+       .urgent_out_of_order_return_per_channel_pixel_only_bytes = 4096,
+       .urgent_out_of_order_return_per_channel_pixel_and_vm_bytes = 4096,
+       .urgent_out_of_order_return_per_channel_vm_only_bytes = 4096,
+       .pct_ideal_dram_sdp_bw_after_urgent_pixel_only = 40.0,
+       .pct_ideal_dram_sdp_bw_after_urgent_pixel_and_vm = 40.0,
+       .pct_ideal_dram_sdp_bw_after_urgent_vm_only = 40.0,
+       .max_avg_sdp_bw_use_normal_percent = 40.0,
+       .max_avg_dram_bw_use_normal_percent = 40.0,
+       .writeback_latency_us = 12.0,
+       .ideal_dram_bw_after_urgent_percent = 40.0,
+       .max_request_size_bytes = 256,
+       .dram_channel_width_bytes = 2,
+       .fabric_datapath_to_dcn_data_return_bytes = 64,
+       .dcn_downspread_percent = 0.5,
+       .downspread_percent = 0.38,
+       .dram_page_open_time_ns = 50.0,
+       .dram_rw_turnaround_time_ns = 17.5,
+       .dram_return_buffer_per_channel_bytes = 8192,
+       .round_trip_ping_latency_dcfclk_cycles = 131,
+       .urgent_out_of_order_return_per_channel_bytes = 256,
+       .channel_interleave_bytes = 256,
+       .num_banks = 8,
+       .num_chans = 16,
+       .vmm_page_size_bytes = 4096,
+       .dram_clock_change_latency_us = 404.0,
+       .dummy_pstate_latency_us = 5.0,
+       .writeback_dram_clock_change_latency_us = 23.0,
+       .return_bus_width_bytes = 64,
+       .dispclk_dppclk_vco_speed_mhz = 3850,
+       .xfc_bus_transport_time_us = 20,
+       .xfc_xbuf_latency_tolerance_us = 4,
+       .use_urgent_burst_bw = 0
+};
+
+struct _vcs_dpi_soc_bounding_box_st dcn2_0_nv12_soc = { 0 };
 
 #ifndef mmDP0_DP_DPHY_INTERNAL_CTRL
        #define mmDP0_DP_DPHY_INTERNAL_CTRL             0x210f
@@ -314,7 +426,7 @@ static const struct dce_audio_shift audio_shift = {
                DCE120_AUD_COMMON_MASK_SH_LIST(__SHIFT)
 };
 
-static const struct dce_aduio_mask audio_mask = {
+static const struct dce_audio_mask audio_mask = {
                DCE120_AUD_COMMON_MASK_SH_LIST(_MASK)
 };
 
@@ -695,6 +807,16 @@ static const struct dc_plane_cap plane_cap = {
                        .fp16 = 1
        }
 };
+static const struct resource_caps res_cap_nv14 = {
+               .num_timing_generator = 5,
+               .num_opp = 5,
+               .num_video_plane = 5,
+               .num_audio = 6,
+               .num_stream_encoder = 5,
+               .num_pll = 5,
+               .num_dwb = 0,
+               .num_ddc = 5,
+};
 
 static const struct dc_debug_options debug_defaults_drv = {
                .disable_dmcu = true,
@@ -1197,7 +1319,11 @@ static void get_pixel_clock_parameters(
        struct pixel_clk_params *pixel_clk_params)
 {
        const struct dc_stream_state *stream = pipe_ctx->stream;
-       bool odm_combine = dc_res_get_odm_bottom_pipe(pipe_ctx) != NULL;
+       struct pipe_ctx *odm_pipe;
+       int opp_cnt = 1;
+
+       for (odm_pipe = pipe_ctx->next_odm_pipe; odm_pipe; odm_pipe = odm_pipe->next_odm_pipe)
+               opp_cnt++;
 
        pixel_clk_params->requested_pix_clk_100hz = stream->timing.pix_clk_100hz;
        pixel_clk_params->encoder_object_id = stream->link->link_enc->id;
@@ -1215,7 +1341,9 @@ static void get_pixel_clock_parameters(
        if (stream->timing.pixel_encoding == PIXEL_ENCODING_YCBCR422)
                pixel_clk_params->color_depth = COLOR_DEPTH_888;
 
-       if (optc1_is_two_pixels_per_containter(&stream->timing) || odm_combine)
+       if (opp_cnt == 4)
+               pixel_clk_params->requested_pix_clk_100hz /= 4;
+       else if (optc1_is_two_pixels_per_containter(&stream->timing) || opp_cnt == 2)
                pixel_clk_params->requested_pix_clk_100hz /= 2;
 
        if (stream->timing.timing_3d_format == TIMING_3D_FORMAT_HW_FRAME_PACKING)
@@ -1359,22 +1487,16 @@ static enum dc_status remove_dsc_from_stream_resource(struct dc *dc,
        for (i = 0; i < MAX_PIPES; i++) {
                if (new_ctx->res_ctx.pipe_ctx[i].stream == dc_stream && !new_ctx->res_ctx.pipe_ctx[i].top_pipe) {
                        pipe_ctx = &new_ctx->res_ctx.pipe_ctx[i];
-                       break;
+
+                       if (pipe_ctx->stream_res.dsc)
+                               release_dsc(&new_ctx->res_ctx, dc->res_pool, &pipe_ctx->stream_res.dsc);
                }
        }
 
        if (!pipe_ctx)
                return DC_ERROR_UNEXPECTED;
-
-       if (pipe_ctx->stream_res.dsc) {
-               struct pipe_ctx *odm_pipe = dc_res_get_odm_bottom_pipe(pipe_ctx);
-
-               release_dsc(&new_ctx->res_ctx, dc->res_pool, &pipe_ctx->stream_res.dsc);
-               if (odm_pipe)
-                       release_dsc(&new_ctx->res_ctx, dc->res_pool, &odm_pipe->stream_res.dsc);
-       }
-
-       return DC_OK;
+       else
+               return DC_OK;
 }
 #endif
 
@@ -1473,17 +1595,92 @@ static void swizzle_to_dml_params(
        }
 }
 
-static bool dcn20_split_stream_for_combine(
+static bool dcn20_split_stream_for_odm(
+               struct resource_context *res_ctx,
+               const struct resource_pool *pool,
+               struct pipe_ctx *prev_odm_pipe,
+               struct pipe_ctx *next_odm_pipe)
+{
+       int pipe_idx = next_odm_pipe->pipe_idx;
+
+       *next_odm_pipe = *prev_odm_pipe;
+
+       next_odm_pipe->pipe_idx = pipe_idx;
+       next_odm_pipe->plane_res.mi = pool->mis[next_odm_pipe->pipe_idx];
+       next_odm_pipe->plane_res.hubp = pool->hubps[next_odm_pipe->pipe_idx];
+       next_odm_pipe->plane_res.ipp = pool->ipps[next_odm_pipe->pipe_idx];
+       next_odm_pipe->plane_res.xfm = pool->transforms[next_odm_pipe->pipe_idx];
+       next_odm_pipe->plane_res.dpp = pool->dpps[next_odm_pipe->pipe_idx];
+       next_odm_pipe->plane_res.mpcc_inst = pool->dpps[next_odm_pipe->pipe_idx]->inst;
+#ifdef CONFIG_DRM_AMD_DC_DSC_SUPPORT
+       next_odm_pipe->stream_res.dsc = NULL;
+#endif
+       if (prev_odm_pipe->next_odm_pipe && prev_odm_pipe->next_odm_pipe != next_odm_pipe) {
+               ASSERT(!next_odm_pipe->next_odm_pipe);
+               next_odm_pipe->next_odm_pipe = prev_odm_pipe->next_odm_pipe;
+               next_odm_pipe->next_odm_pipe->prev_odm_pipe = next_odm_pipe;
+       }
+       prev_odm_pipe->next_odm_pipe = next_odm_pipe;
+       next_odm_pipe->prev_odm_pipe = prev_odm_pipe;
+       ASSERT(next_odm_pipe->top_pipe == NULL);
+
+       if (prev_odm_pipe->plane_state) {
+               struct scaler_data *sd = &prev_odm_pipe->plane_res.scl_data;
+               int new_width;
+
+               /* HACTIVE halved for odm combine */
+               sd->h_active /= 2;
+               /* Calculate new vp and recout for left pipe */
+               /* Need at least 16 pixels width per side */
+               if (sd->recout.x + 16 >= sd->h_active)
+                       return false;
+               new_width = sd->h_active - sd->recout.x;
+               sd->viewport.width -= dc_fixpt_floor(dc_fixpt_mul_int(
+                               sd->ratios.horz, sd->recout.width - new_width));
+               sd->viewport_c.width -= dc_fixpt_floor(dc_fixpt_mul_int(
+                               sd->ratios.horz_c, sd->recout.width - new_width));
+               sd->recout.width = new_width;
+
+               /* Calculate new vp and recout for right pipe */
+               sd = &next_odm_pipe->plane_res.scl_data;
+               /* HACTIVE halved for odm combine */
+               sd->h_active /= 2;
+               /* Need at least 16 pixels width per side */
+               if (new_width <= 16)
+                       return false;
+               new_width = sd->recout.width + sd->recout.x - sd->h_active;
+               sd->viewport.width -= dc_fixpt_floor(dc_fixpt_mul_int(
+                               sd->ratios.horz, sd->recout.width - new_width));
+               sd->viewport_c.width -= dc_fixpt_floor(dc_fixpt_mul_int(
+                               sd->ratios.horz_c, sd->recout.width - new_width));
+               sd->recout.width = new_width;
+               sd->viewport.x += dc_fixpt_floor(dc_fixpt_mul_int(
+                               sd->ratios.horz, sd->h_active - sd->recout.x));
+               sd->viewport_c.x += dc_fixpt_floor(dc_fixpt_mul_int(
+                               sd->ratios.horz_c, sd->h_active - sd->recout.x));
+               sd->recout.x = 0;
+       }
+       next_odm_pipe->stream_res.opp = pool->opps[next_odm_pipe->pipe_idx];
+#ifdef CONFIG_DRM_AMD_DC_DSC_SUPPORT
+       if (next_odm_pipe->stream->timing.flags.DSC == 1) {
+               acquire_dsc(res_ctx, pool, &next_odm_pipe->stream_res.dsc);
+               ASSERT(next_odm_pipe->stream_res.dsc);
+               if (next_odm_pipe->stream_res.dsc == NULL)
+                       return false;
+       }
+#endif
+
+       return true;
+}
+
+static void dcn20_split_stream_for_mpc(
                struct resource_context *res_ctx,
                const struct resource_pool *pool,
                struct pipe_ctx *primary_pipe,
-               struct pipe_ctx *secondary_pipe,
-               bool is_odm_combine)
+               struct pipe_ctx *secondary_pipe)
 {
        int pipe_idx = secondary_pipe->pipe_idx;
-       struct scaler_data *sd = &primary_pipe->plane_res.scl_data;
        struct pipe_ctx *sec_bot_pipe = secondary_pipe->bottom_pipe;
-       int new_width;
 
        *secondary_pipe = *primary_pipe;
        secondary_pipe->bottom_pipe = sec_bot_pipe;
@@ -1506,57 +1703,9 @@ static bool dcn20_split_stream_for_combine(
        primary_pipe->bottom_pipe = secondary_pipe;
        secondary_pipe->top_pipe = primary_pipe;
 
-       if (is_odm_combine) {
-               if (primary_pipe->plane_state) {
-                       /* HACTIVE halved for odm combine */
-                       sd->h_active /= 2;
-                       /* Copy scl_data to secondary pipe */
-                       secondary_pipe->plane_res.scl_data = *sd;
-
-                       /* Calculate new vp and recout for left pipe */
-                       /* Need at least 16 pixels width per side */
-                       if (sd->recout.x + 16 >= sd->h_active)
-                               return false;
-                       new_width = sd->h_active - sd->recout.x;
-                       sd->viewport.width -= dc_fixpt_floor(dc_fixpt_mul_int(
-                                       sd->ratios.horz, sd->recout.width - new_width));
-                       sd->viewport_c.width -= dc_fixpt_floor(dc_fixpt_mul_int(
-                                       sd->ratios.horz_c, sd->recout.width - new_width));
-                       sd->recout.width = new_width;
-
-                       /* Calculate new vp and recout for right pipe */
-                       sd = &secondary_pipe->plane_res.scl_data;
-                       new_width = sd->recout.width + sd->recout.x - sd->h_active;
-                       /* Need at least 16 pixels width per side */
-                       if (new_width <= 16)
-                               return false;
-                       sd->viewport.width -= dc_fixpt_floor(dc_fixpt_mul_int(
-                                       sd->ratios.horz, sd->recout.width - new_width));
-                       sd->viewport_c.width -= dc_fixpt_floor(dc_fixpt_mul_int(
-                                       sd->ratios.horz_c, sd->recout.width - new_width));
-                       sd->recout.width = new_width;
-                       sd->viewport.x += dc_fixpt_floor(dc_fixpt_mul_int(
-                                       sd->ratios.horz, sd->h_active - sd->recout.x));
-                       sd->viewport_c.x += dc_fixpt_floor(dc_fixpt_mul_int(
-                                       sd->ratios.horz_c, sd->h_active - sd->recout.x));
-                       sd->recout.x = 0;
-               }
-               secondary_pipe->stream_res.opp = pool->opps[secondary_pipe->pipe_idx];
-#ifdef CONFIG_DRM_AMD_DC_DSC_SUPPORT
-               if (secondary_pipe->stream->timing.flags.DSC == 1) {
-                       acquire_dsc(res_ctx, pool, &secondary_pipe->stream_res.dsc);
-                       ASSERT(secondary_pipe->stream_res.dsc);
-                       if (secondary_pipe->stream_res.dsc == NULL)
-                               return false;
-               }
-#endif
-       } else {
-               ASSERT(primary_pipe->plane_state);
-               resource_build_scaling_params(primary_pipe);
-               resource_build_scaling_params(secondary_pipe);
-       }
-
-       return true;
+       ASSERT(primary_pipe->plane_state);
+       resource_build_scaling_params(primary_pipe);
+       resource_build_scaling_params(secondary_pipe);
 }
 
 void dcn20_populate_dml_writeback_from_context(
@@ -1610,7 +1759,7 @@ int dcn20_populate_dml_pipes_from_context(
                        pipe_cnt = i;
                        continue;
                }
-               if (!resource_are_streams_timing_synchronizable(
+               if (dc->debug.disable_timing_sync || !resource_are_streams_timing_synchronizable(
                                res_ctx->pipe_ctx[pipe_cnt].stream,
                                res_ctx->pipe_ctx[i].stream)) {
                        synchronized_vblank = false;
@@ -1669,6 +1818,19 @@ int dcn20_populate_dml_pipes_from_context(
                pipes[pipe_cnt].dout.dp_lanes = 4;
                pipes[pipe_cnt].pipe.dest.vtotal_min = res_ctx->pipe_ctx[i].stream->adjust.v_total_min;
                pipes[pipe_cnt].pipe.dest.vtotal_max = res_ctx->pipe_ctx[i].stream->adjust.v_total_max;
+               pipes[pipe_cnt].pipe.dest.odm_combine = res_ctx->pipe_ctx[i].prev_odm_pipe
+                                                       || res_ctx->pipe_ctx[i].next_odm_pipe;
+               pipes[pipe_cnt].pipe.src.hsplit_grp = res_ctx->pipe_ctx[i].pipe_idx;
+               if (res_ctx->pipe_ctx[i].top_pipe && res_ctx->pipe_ctx[i].top_pipe->plane_state
+                               == res_ctx->pipe_ctx[i].plane_state)
+                       pipes[pipe_cnt].pipe.src.hsplit_grp = res_ctx->pipe_ctx[i].top_pipe->pipe_idx;
+               else if (res_ctx->pipe_ctx[i].prev_odm_pipe) {
+                       struct pipe_ctx *first_pipe = res_ctx->pipe_ctx[i].prev_odm_pipe;
+
+                       while (first_pipe->prev_odm_pipe)
+                               first_pipe = first_pipe->prev_odm_pipe;
+                       pipes[pipe_cnt].pipe.src.hsplit_grp = first_pipe->pipe_idx;
+               }
 
                switch (res_ctx->pipe_ctx[i].stream->signal) {
                case SIGNAL_TYPE_DISPLAY_PORT_MST:
@@ -1721,7 +1883,6 @@ int dcn20_populate_dml_pipes_from_context(
                        break;
                }
 
-
                switch (res_ctx->pipe_ctx[i].stream->timing.pixel_encoding) {
                case PIXEL_ENCODING_RGB:
                case PIXEL_ENCODING_YCBCR444:
@@ -1743,10 +1904,6 @@ int dcn20_populate_dml_pipes_from_context(
                        pipes[pipe_cnt].dout.output_format = dm_444;
                        pipes[pipe_cnt].dout.output_bpp = output_bpc * 3;
                }
-               pipes[pipe_cnt].pipe.src.hsplit_grp = res_ctx->pipe_ctx[i].pipe_idx;
-               if (res_ctx->pipe_ctx[i].top_pipe && res_ctx->pipe_ctx[i].top_pipe->plane_state
-                               == res_ctx->pipe_ctx[i].plane_state)
-                       pipes[pipe_cnt].pipe.src.hsplit_grp = res_ctx->pipe_ctx[i].top_pipe->pipe_idx;
 
                /* todo: default max for now, until there is logic reflecting this in dc*/
                pipes[pipe_cnt].dout.output_bpc = 12;
@@ -1795,14 +1952,6 @@ int dcn20_populate_dml_pipes_from_context(
                                        && res_ctx->pipe_ctx[i].bottom_pipe->plane_state == pln)
                                        || (res_ctx->pipe_ctx[i].top_pipe
                                        && res_ctx->pipe_ctx[i].top_pipe->plane_state == pln);
-                       pipes[pipe_cnt].pipe.dest.odm_combine = (res_ctx->pipe_ctx[i].bottom_pipe
-                                       && res_ctx->pipe_ctx[i].bottom_pipe->plane_state == pln
-                                       && res_ctx->pipe_ctx[i].bottom_pipe->stream_res.opp
-                                               != res_ctx->pipe_ctx[i].stream_res.opp)
-                               || (res_ctx->pipe_ctx[i].top_pipe
-                                       && res_ctx->pipe_ctx[i].top_pipe->plane_state == pln
-                                       && res_ctx->pipe_ctx[i].top_pipe->stream_res.opp
-                                               != res_ctx->pipe_ctx[i].stream_res.opp);
                        pipes[pipe_cnt].pipe.src.source_scan = pln->rotation == ROTATION_ANGLE_90
                                        || pln->rotation == ROTATION_ANGLE_270 ? dm_vert : dm_horz;
                        pipes[pipe_cnt].pipe.src.viewport_y_y = scl->viewport.y;
@@ -1812,13 +1961,13 @@ int dcn20_populate_dml_pipes_from_context(
                        pipes[pipe_cnt].pipe.src.viewport_height = scl->viewport.height;
                        pipes[pipe_cnt].pipe.src.viewport_height_c = scl->viewport_c.height;
                        if (pln->format >= SURFACE_PIXEL_FORMAT_VIDEO_BEGIN) {
-                               pipes[pipe_cnt].pipe.src.data_pitch = pln->plane_size.video.luma_pitch;
-                               pipes[pipe_cnt].pipe.src.data_pitch_c = pln->plane_size.video.chroma_pitch;
-                               pipes[pipe_cnt].pipe.src.meta_pitch = pln->dcc.video.meta_pitch_l;
-                               pipes[pipe_cnt].pipe.src.meta_pitch_c = pln->dcc.video.meta_pitch_c;
+                               pipes[pipe_cnt].pipe.src.data_pitch = pln->plane_size.surface_pitch;
+                               pipes[pipe_cnt].pipe.src.data_pitch_c = pln->plane_size.chroma_pitch;
+                               pipes[pipe_cnt].pipe.src.meta_pitch = pln->dcc.meta_pitch;
+                               pipes[pipe_cnt].pipe.src.meta_pitch_c = pln->dcc.meta_pitch_c;
                        } else {
-                               pipes[pipe_cnt].pipe.src.data_pitch = pln->plane_size.grph.surface_pitch;
-                               pipes[pipe_cnt].pipe.src.meta_pitch = pln->dcc.grph.meta_pitch;
+                               pipes[pipe_cnt].pipe.src.data_pitch = pln->plane_size.surface_pitch;
+                               pipes[pipe_cnt].pipe.src.meta_pitch = pln->dcc.meta_pitch;
                        }
                        pipes[pipe_cnt].pipe.src.dcc = pln->dcc.enable;
                        pipes[pipe_cnt].pipe.dest.recout_width = scl->recout.width;
@@ -1986,20 +2135,24 @@ static bool dcn20_validate_dsc(struct dc *dc, struct dc_state *new_ctx)
                struct pipe_ctx *pipe_ctx = &new_ctx->res_ctx.pipe_ctx[i];
                struct dc_stream_state *stream = pipe_ctx->stream;
                struct dsc_config dsc_cfg;
+               struct pipe_ctx *odm_pipe;
+               int opp_cnt = 1;
+
+               for (odm_pipe = pipe_ctx->next_odm_pipe; odm_pipe; odm_pipe = odm_pipe->next_odm_pipe)
+                       opp_cnt++;
 
                /* Only need to validate top pipe */
-               if (pipe_ctx->top_pipe || !stream || !stream->timing.flags.DSC)
+               if (pipe_ctx->top_pipe || pipe_ctx->prev_odm_pipe || !stream || !stream->timing.flags.DSC)
                        continue;
 
-               dsc_cfg.pic_width = stream->timing.h_addressable + stream->timing.h_border_left
-                               + stream->timing.h_border_right;
+               dsc_cfg.pic_width = (stream->timing.h_addressable + stream->timing.h_border_left
+                               + stream->timing.h_border_right) / opp_cnt;
                dsc_cfg.pic_height = stream->timing.v_addressable + stream->timing.v_border_top
                                + stream->timing.v_border_bottom;
-               if (dc_res_get_odm_bottom_pipe(pipe_ctx))
-                       dsc_cfg.pic_width /= 2;
                dsc_cfg.pixel_encoding = stream->timing.pixel_encoding;
                dsc_cfg.color_depth = stream->timing.display_color_depth;
                dsc_cfg.dc_dsc_cfg = stream->timing.dsc_cfg;
+               dsc_cfg.dc_dsc_cfg.num_slices_h /= opp_cnt;
 
                if (!pipe_ctx->stream_res.dsc->funcs->dsc_validate_stream(pipe_ctx->stream_res.dsc, &dsc_cfg))
                        return false;
@@ -2008,31 +2161,134 @@ static bool dcn20_validate_dsc(struct dc *dc, struct dc_state *new_ctx)
 }
 #endif
 
-bool dcn20_validate_bandwidth(struct dc *dc, struct dc_state *context,
-               bool fast_validate)
+static struct pipe_ctx *dcn20_find_secondary_pipe(struct dc *dc,
+               struct resource_context *res_ctx,
+               const struct resource_pool *pool,
+               const struct pipe_ctx *primary_pipe)
 {
-       bool out = false;
+       struct pipe_ctx *secondary_pipe = NULL;
+
+       if (dc && primary_pipe) {
+               int j;
+               int preferred_pipe_idx = 0;
+
+               /* first check the prev dc state:
+                * if this primary pipe has a bottom pipe in prev. state
+                * and if the bottom pipe is still available (which it should be),
+                * pick that pipe as secondary
+                * Same logic applies for ODM pipes. Since mpo is not allowed with odm
+                * check in else case.
+                */
+               if (dc->current_state->res_ctx.pipe_ctx[primary_pipe->pipe_idx].bottom_pipe) {
+                       preferred_pipe_idx = dc->current_state->res_ctx.pipe_ctx[primary_pipe->pipe_idx].bottom_pipe->pipe_idx;
+                       if (res_ctx->pipe_ctx[preferred_pipe_idx].stream == NULL) {
+                               secondary_pipe = &res_ctx->pipe_ctx[preferred_pipe_idx];
+                               secondary_pipe->pipe_idx = preferred_pipe_idx;
+                       }
+               } else if (dc->current_state->res_ctx.pipe_ctx[primary_pipe->pipe_idx].next_odm_pipe) {
+                       preferred_pipe_idx = dc->current_state->res_ctx.pipe_ctx[primary_pipe->pipe_idx].next_odm_pipe->pipe_idx;
+                       if (res_ctx->pipe_ctx[preferred_pipe_idx].stream == NULL) {
+                               secondary_pipe = &res_ctx->pipe_ctx[preferred_pipe_idx];
+                               secondary_pipe->pipe_idx = preferred_pipe_idx;
+                       }
+               }
 
-       BW_VAL_TRACE_SETUP();
+               /*
+                * if this primary pipe does not have a bottom pipe in prev. state
+                * start backward and find a pipe that did not used to be a bottom pipe in
+                * prev. dc state. This way we make sure we keep the same assignment as
+                * last state and will not have to reprogram every pipe
+                */
+               if (secondary_pipe == NULL) {
+                       for (j = dc->res_pool->pipe_count - 1; j >= 0; j--) {
+                               if (dc->current_state->res_ctx.pipe_ctx[j].top_pipe == NULL) {
+                                       preferred_pipe_idx = j;
+
+                                       if (res_ctx->pipe_ctx[preferred_pipe_idx].stream == NULL) {
+                                               secondary_pipe = &res_ctx->pipe_ctx[preferred_pipe_idx];
+                                               secondary_pipe->pipe_idx = preferred_pipe_idx;
+                                               break;
+                                       }
+                               }
+                       }
+               }
+               /*
+                * We should never hit this assert unless assignments are shuffled around
+                * if this happens we will prob. hit a vsync tdr
+                */
+               ASSERT(secondary_pipe);
+               /*
+                * search backwards for the second pipe to keep pipe
+                * assignment more consistent
+                */
+               if (secondary_pipe == NULL) {
+                       for (j = dc->res_pool->pipe_count - 1; j >= 0; j--) {
+                               preferred_pipe_idx = j;
+
+                               if (res_ctx->pipe_ctx[preferred_pipe_idx].stream == NULL) {
+                                       secondary_pipe = &res_ctx->pipe_ctx[preferred_pipe_idx];
+                                       secondary_pipe->pipe_idx = preferred_pipe_idx;
+                                       break;
+                               }
+                       }
+               }
+       }
+
+       return secondary_pipe;
+}
+
+bool dcn20_fast_validate_bw(
+               struct dc *dc,
+               struct dc_state *context,
+               display_e2e_pipe_params_st *pipes,
+               int *pipe_cnt_out,
+               int *pipe_split_from,
+               int *vlevel_out)
+{
+       bool out = false;
 
        int pipe_cnt, i, pipe_idx, vlevel, vlevel_unsplit;
-       int pipe_split_from[MAX_PIPES];
-       bool odm_capable = context->bw_ctx.dml.ip.odm_capable;
        bool force_split = false;
-#ifdef CONFIG_DRM_AMD_DC_DSC_SUPPORT
-       bool failed_non_odm_dsc = false;
-#endif
        int split_threshold = dc->res_pool->pipe_count / 2;
        bool avoid_split = dc->debug.pipe_split_policy != MPC_SPLIT_DYNAMIC;
-       display_e2e_pipe_params_st *pipes = kzalloc(dc->res_pool->pipe_count * sizeof(display_e2e_pipe_params_st), GFP_KERNEL);
-       DC_LOGGER_INIT(dc->ctx->logger);
 
-       BW_VAL_TRACE_COUNT();
 
        ASSERT(pipes);
        if (!pipes)
                return false;
 
+       /* merge previously split odm pipes since mode support needs to make the decision */
+       for (i = 0; i < dc->res_pool->pipe_count; i++) {
+               struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
+               struct pipe_ctx *odm_pipe = pipe->next_odm_pipe;
+
+               if (pipe->prev_odm_pipe)
+                       continue;
+
+               pipe->next_odm_pipe = NULL;
+               while (odm_pipe) {
+                       struct pipe_ctx *next_odm_pipe = odm_pipe->next_odm_pipe;
+
+                       odm_pipe->plane_state = NULL;
+                       odm_pipe->stream = NULL;
+                       odm_pipe->top_pipe = NULL;
+                       odm_pipe->bottom_pipe = NULL;
+                       odm_pipe->prev_odm_pipe = NULL;
+                       odm_pipe->next_odm_pipe = NULL;
+#ifdef CONFIG_DRM_AMD_DC_DSC_SUPPORT
+                       if (odm_pipe->stream_res.dsc)
+                               release_dsc(&context->res_ctx, dc->res_pool, &odm_pipe->stream_res.dsc);
+#endif
+                       /* Clear plane_res and stream_res */
+                       memset(&odm_pipe->plane_res, 0, sizeof(odm_pipe->plane_res));
+                       memset(&odm_pipe->stream_res, 0, sizeof(odm_pipe->stream_res));
+                       odm_pipe = next_odm_pipe;
+               }
+               if (pipe->plane_state)
+                       resource_build_scaling_params(pipe);
+       }
+
+       /* merge previously mpc split pipes since mode support needs to make the decision */
        for (i = 0; i < dc->res_pool->pipe_count; i++) {
                struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
                struct pipe_ctx *hsplit_pipe = pipe->bottom_pipe;
@@ -2040,7 +2296,6 @@ bool dcn20_validate_bandwidth(struct dc *dc, struct dc_state *context,
                if (!hsplit_pipe || hsplit_pipe->plane_state != pipe->plane_state)
                        continue;
 
-               /* merge previously split pipe since mode support needs to make the decision */
                pipe->bottom_pipe = hsplit_pipe->bottom_pipe;
                if (hsplit_pipe->bottom_pipe)
                        hsplit_pipe->bottom_pipe->top_pipe = pipe;
@@ -2048,10 +2303,7 @@ bool dcn20_validate_bandwidth(struct dc *dc, struct dc_state *context,
                hsplit_pipe->stream = NULL;
                hsplit_pipe->top_pipe = NULL;
                hsplit_pipe->bottom_pipe = NULL;
-#ifdef CONFIG_DRM_AMD_DC_DSC_SUPPORT
-               if (hsplit_pipe->stream_res.dsc && hsplit_pipe->stream_res.dsc != pipe->stream_res.dsc)
-                       release_dsc(&context->res_ctx, dc->res_pool, &hsplit_pipe->stream_res.dsc);
-#endif
+
                /* Clear plane_res and stream_res */
                memset(&hsplit_pipe->plane_res, 0, sizeof(hsplit_pipe->plane_res));
                memset(&hsplit_pipe->stream_res, 0, sizeof(hsplit_pipe->stream_res));
@@ -2066,30 +2318,15 @@ bool dcn20_validate_bandwidth(struct dc *dc, struct dc_state *context,
                pipe_cnt = dcn20_populate_dml_pipes_from_context(dc,
                        &context->res_ctx, pipes);
 
+       *pipe_cnt_out = pipe_cnt;
+
        if (!pipe_cnt) {
-               BW_VAL_TRACE_SKIP(pass);
                out = true;
                goto validate_out;
        }
 
-       context->bw_ctx.dml.ip.odm_capable = 0;
-
        vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, pipe_cnt);
 
-       context->bw_ctx.dml.ip.odm_capable = odm_capable;
-
-#ifdef CONFIG_DRM_AMD_DC_DSC_SUPPORT
-       /* 1 dsc per stream dsc validation */
-       if (vlevel <= context->bw_ctx.dml.soc.num_states)
-               if (!dcn20_validate_dsc(dc, context)) {
-                       failed_non_odm_dsc = true;
-                       vlevel = context->bw_ctx.dml.soc.num_states + 1;
-               }
-#endif
-
-       if (vlevel > context->bw_ctx.dml.soc.num_states && odm_capable)
-               vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, pipe_cnt);
-
        if (vlevel > context->bw_ctx.dml.soc.num_states)
                goto validate_fail;
 
@@ -2160,17 +2397,12 @@ bool dcn20_validate_bandwidth(struct dc *dc, struct dc_state *context,
                }
                if (force_split && context->bw_ctx.dml.vba.NoOfDPP[vlevel][context->bw_ctx.dml.vba.maxMpcComb][pipe_idx] == 1)
                        context->bw_ctx.dml.vba.RequiredDPPCLK[vlevel][context->bw_ctx.dml.vba.maxMpcComb][pipe_idx] /= 2;
-               if (dc->config.forced_clocks == true) {
-                       context->bw_ctx.dml.vba.RequiredDPPCLK[vlevel][context->bw_ctx.dml.vba.maxMpcComb][pipe_idx] =
-                                       context->bw_ctx.dml.soc.clock_limits[0].dppclk_mhz;
-               }
                if (!pipe->top_pipe && !pipe->plane_state && context->bw_ctx.dml.vba.ODMCombineEnabled[pipe_idx]) {
-                       hsplit_pipe = find_idle_secondary_pipe(&context->res_ctx, dc->res_pool, pipe);
+                       hsplit_pipe = dcn20_find_secondary_pipe(dc, &context->res_ctx, dc->res_pool, pipe);
                        ASSERT(hsplit_pipe);
-                       if (!dcn20_split_stream_for_combine(
+                       if (!dcn20_split_stream_for_odm(
                                        &context->res_ctx, dc->res_pool,
-                                       pipe, hsplit_pipe,
-                                       true))
+                                       pipe, hsplit_pipe))
                                goto validate_fail;
                        pipe_split_from[hsplit_pipe->pipe_idx] = pipe_idx;
                        dcn20_build_mapped_resource(dc, context, pipe->stream);
@@ -2206,16 +2438,21 @@ bool dcn20_validate_bandwidth(struct dc *dc, struct dc_state *context,
                if (need_split3d || need_split || force_split) {
                        if (!hsplit_pipe || hsplit_pipe->plane_state != pipe->plane_state) {
                                /* pipe not split previously needs split */
-                               hsplit_pipe = find_idle_secondary_pipe(&context->res_ctx, dc->res_pool, pipe);
+                               hsplit_pipe = dcn20_find_secondary_pipe(dc, &context->res_ctx, dc->res_pool, pipe);
                                ASSERT(hsplit_pipe || force_split);
                                if (!hsplit_pipe)
                                        continue;
 
-                               if (!dcn20_split_stream_for_combine(
+                               if (context->bw_ctx.dml.vba.ODMCombineEnabled[pipe_idx]) {
+                                       if (!dcn20_split_stream_for_odm(
+                                                       &context->res_ctx, dc->res_pool,
+                                                       pipe, hsplit_pipe))
+                                               goto validate_fail;
+                                       dcn20_build_mapped_resource(dc, context, pipe->stream);
+                               } else
+                                       dcn20_split_stream_for_mpc(
                                                &context->res_ctx, dc->res_pool,
-                                               pipe, hsplit_pipe,
-                                               context->bw_ctx.dml.vba.ODMCombineEnabled[pipe_idx]))
-                                       goto validate_fail;
+                                               pipe, hsplit_pipe);
                                pipe_split_from[hsplit_pipe->pipe_idx] = pipe_idx;
                        }
                } else if (hsplit_pipe && hsplit_pipe->plane_state == pipe->plane_state) {
@@ -2225,20 +2462,33 @@ bool dcn20_validate_bandwidth(struct dc *dc, struct dc_state *context,
        }
 #ifdef CONFIG_DRM_AMD_DC_DSC_SUPPORT
        /* Actual dsc count per stream dsc validation*/
-       if (failed_non_odm_dsc && !dcn20_validate_dsc(dc, context)) {
+       if (!dcn20_validate_dsc(dc, context)) {
                context->bw_ctx.dml.vba.ValidationStatus[context->bw_ctx.dml.vba.soc.num_states] =
                                DML_FAIL_DSC_VALIDATION_FAILURE;
                goto validate_fail;
        }
 #endif
 
-       BW_VAL_TRACE_END_VOLTAGE_LEVEL();
+       *vlevel_out = vlevel;
 
-       if (fast_validate) {
-               BW_VAL_TRACE_SKIP(fast);
-               out = true;
-               goto validate_out;
-       }
+       out = true;
+       goto validate_out;
+
+validate_fail:
+       out = false;
+
+validate_out:
+       return out;
+}
+
+void dcn20_calculate_wm(
+               struct dc *dc, struct dc_state *context,
+               display_e2e_pipe_params_st *pipes,
+               int *out_pipe_cnt,
+               int *pipe_split_from,
+               int vlevel)
+{
+       int pipe_cnt, i, pipe_idx;
 
        for (i = 0, pipe_idx = 0, pipe_cnt = 0; i < dc->res_pool->pipe_count; i++) {
                if (!context->res_ctx.pipe_ctx[i].stream)
@@ -2265,10 +2515,16 @@ bool dcn20_validate_bandwidth(struct dc *dc, struct dc_state *context,
                        else
                                pipes[pipe_cnt].pipe.dest.odm_combine = 0;
                }
+
                if (dc->config.forced_clocks) {
                        pipes[pipe_cnt].clks_cfg.dispclk_mhz = context->bw_ctx.dml.soc.clock_limits[0].dispclk_mhz;
                        pipes[pipe_cnt].clks_cfg.dppclk_mhz = context->bw_ctx.dml.soc.clock_limits[0].dppclk_mhz;
                }
+               if (dc->debug.min_disp_clk_khz > pipes[pipe_cnt].clks_cfg.dispclk_mhz * 1000)
+                       pipes[pipe_cnt].clks_cfg.dispclk_mhz = dc->debug.min_disp_clk_khz / 1000.0;
+               if (dc->debug.min_dpp_clk_khz > pipes[pipe_cnt].clks_cfg.dppclk_mhz * 1000)
+                       pipes[pipe_cnt].clks_cfg.dppclk_mhz = dc->debug.min_dpp_clk_khz / 1000.0;
+
                pipe_cnt++;
        }
 
@@ -2281,6 +2537,8 @@ bool dcn20_validate_bandwidth(struct dc *dc, struct dc_state *context,
                                &context->res_ctx, pipes);
        }
 
+       *out_pipe_cnt = pipe_cnt;
+
        pipes[0].clks_cfg.voltage = vlevel;
        pipes[0].clks_cfg.dcfclk_mhz = context->bw_ctx.dml.soc.clock_limits[vlevel].dcfclk_mhz;
        pipes[0].clks_cfg.socclk_mhz = context->bw_ctx.dml.soc.clock_limits[vlevel].socclk_mhz;
@@ -2327,6 +2585,17 @@ bool dcn20_validate_bandwidth(struct dc *dc, struct dc_state *context,
        context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
        context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.pstate_change_ns = get_wm_dram_clock_change(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
        context->bw_ctx.bw.dcn.watermarks.a.pte_meta_urgent_ns = get_wm_memory_trip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
+}
+
+void dcn20_calculate_dlg_params(
+               struct dc *dc, struct dc_state *context,
+               display_e2e_pipe_params_st *pipes,
+               int pipe_cnt,
+               int vlevel)
+{
+       int i, j, pipe_idx, pipe_idx_unsplit;
+       bool visited[MAX_PIPES] = { 0 };
+
        /* Writeback MCIF_WB arbitration parameters */
        dc->res_pool->funcs->set_mcif_arb_params(dc, context, pipes, pipe_cnt);
 
@@ -2341,26 +2610,63 @@ bool dcn20_validate_bandwidth(struct dc *dc, struct dc_state *context,
                                                        != dm_dram_clock_change_unsupported;
        context->bw_ctx.bw.dcn.clk.dppclk_khz = 0;
 
-       BW_VAL_TRACE_END_WATERMARKS();
+       /*
+        * An artifact of dml pipe split/odm is that pipes get merged back together for
+        * calculation. Therefore we need to only extract for first pipe in ascending index order
+        * and copy into the other split half.
+        */
+       for (i = 0, pipe_idx = 0, pipe_idx_unsplit = 0; i < dc->res_pool->pipe_count; i++) {
+               if (!context->res_ctx.pipe_ctx[i].stream)
+                       continue;
+
+               if (!visited[pipe_idx]) {
+                       display_pipe_source_params_st *src = &pipes[pipe_idx].pipe.src;
+                       display_pipe_dest_params_st *dst = &pipes[pipe_idx].pipe.dest;
+
+                       dst->vstartup_start = context->bw_ctx.dml.vba.VStartup[pipe_idx_unsplit];
+                       dst->vupdate_offset = context->bw_ctx.dml.vba.VUpdateOffsetPix[pipe_idx_unsplit];
+                       dst->vupdate_width = context->bw_ctx.dml.vba.VUpdateWidthPix[pipe_idx_unsplit];
+                       dst->vready_offset = context->bw_ctx.dml.vba.VReadyOffsetPix[pipe_idx_unsplit];
+                       /*
+                        * j iterates inside pipes array, unlike i which iterates inside
+                        * pipe_ctx array
+                        */
+                       if (src->is_hsplit)
+                               for (j = pipe_idx + 1; j < pipe_cnt; j++) {
+                                       display_pipe_source_params_st *src_j = &pipes[j].pipe.src;
+                                       display_pipe_dest_params_st *dst_j = &pipes[j].pipe.dest;
+
+                                       if (src_j->is_hsplit && !visited[j]
+                                                       && src->hsplit_grp == src_j->hsplit_grp) {
+                                               dst_j->vstartup_start = context->bw_ctx.dml.vba.VStartup[pipe_idx_unsplit];
+                                               dst_j->vupdate_offset = context->bw_ctx.dml.vba.VUpdateOffsetPix[pipe_idx_unsplit];
+                                               dst_j->vupdate_width = context->bw_ctx.dml.vba.VUpdateWidthPix[pipe_idx_unsplit];
+                                               dst_j->vready_offset = context->bw_ctx.dml.vba.VReadyOffsetPix[pipe_idx_unsplit];
+                                               visited[j] = true;
+                                       }
+                               }
+                       visited[pipe_idx] = true;
+                       pipe_idx_unsplit++;
+               }
+               pipe_idx++;
+       }
 
        for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
                if (!context->res_ctx.pipe_ctx[i].stream)
                        continue;
-               pipes[pipe_idx].pipe.dest.vstartup_start = context->bw_ctx.dml.vba.VStartup[pipe_idx];
-               pipes[pipe_idx].pipe.dest.vupdate_offset = context->bw_ctx.dml.vba.VUpdateOffsetPix[pipe_idx];
-               pipes[pipe_idx].pipe.dest.vupdate_width = context->bw_ctx.dml.vba.VUpdateWidthPix[pipe_idx];
-               pipes[pipe_idx].pipe.dest.vready_offset = context->bw_ctx.dml.vba.VReadyOffsetPix[pipe_idx];
                if (context->bw_ctx.bw.dcn.clk.dppclk_khz < pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000)
                        context->bw_ctx.bw.dcn.clk.dppclk_khz = pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000;
                context->res_ctx.pipe_ctx[i].plane_res.bw.dppclk_khz =
                                                pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000;
-#ifdef CONFIG_DRM_AMD_DC_DSC_SUPPORT
-               context->res_ctx.pipe_ctx[i].stream_res.dscclk_khz =
-                               context->bw_ctx.dml.vba.DSCCLK_calculated[pipe_idx] * 1000;
-#endif
+               ASSERT(visited[pipe_idx]);
                context->res_ctx.pipe_ctx[i].pipe_dlg_param = pipes[pipe_idx].pipe.dest;
                pipe_idx++;
        }
+       /*save a original dppclock copy*/
+       context->bw_ctx.bw.dcn.clk.bw_dppclk_khz = context->bw_ctx.bw.dcn.clk.dppclk_khz;
+       context->bw_ctx.bw.dcn.clk.bw_dispclk_khz = context->bw_ctx.bw.dcn.clk.dispclk_khz;
+       context->bw_ctx.bw.dcn.clk.max_supported_dppclk_khz = context->bw_ctx.dml.soc.clock_limits[vlevel].dppclk_mhz * 1000;
+       context->bw_ctx.bw.dcn.clk.max_supported_dispclk_khz = context->bw_ctx.dml.soc.clock_limits[vlevel].dispclk_mhz * 1000;
 
        for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
                bool cstate_en = context->bw_ctx.dml.vba.PrefetchMode[vlevel][context->bw_ctx.dml.vba.maxMpcComb] != 2;
@@ -2383,8 +2689,43 @@ bool dcn20_validate_bandwidth(struct dc *dc, struct dc_state *context,
                                pipes[pipe_idx].pipe);
                pipe_idx++;
        }
+}
+
+static bool dcn20_validate_bandwidth_internal(struct dc *dc, struct dc_state *context,
+               bool fast_validate)
+{
+       bool out = false;
+
+       BW_VAL_TRACE_SETUP();
+
+       int vlevel = 0;
+       int pipe_split_from[MAX_PIPES];
+       int pipe_cnt = 0;
+       display_e2e_pipe_params_st *pipes = kzalloc(dc->res_pool->pipe_count * sizeof(display_e2e_pipe_params_st), GFP_KERNEL);
+       DC_LOGGER_INIT(dc->ctx->logger);
+
+       BW_VAL_TRACE_COUNT();
+
+       out = dcn20_fast_validate_bw(dc, context, pipes, &pipe_cnt, pipe_split_from, &vlevel);
+
+       if (pipe_cnt == 0)
+               goto validate_out;
+
+       if (!out)
+               goto validate_fail;
+
+       BW_VAL_TRACE_END_VOLTAGE_LEVEL();
+
+       if (fast_validate) {
+               BW_VAL_TRACE_SKIP(fast);
+               goto validate_out;
+       }
+
+       dcn20_calculate_wm(dc, context, pipes, &pipe_cnt, pipe_split_from, vlevel);
+       dcn20_calculate_dlg_params(dc, context, pipes, pipe_cnt, vlevel);
+
+       BW_VAL_TRACE_END_WATERMARKS();
 
-       out = true;
        goto validate_out;
 
 validate_fail:
@@ -2402,6 +2743,49 @@ bool dcn20_validate_bandwidth(struct dc *dc, struct dc_state *context,
        return out;
 }
 
+
+bool dcn20_validate_bandwidth(struct dc *dc, struct dc_state *context,
+               bool fast_validate)
+{
+       bool voltage_supported = false;
+       bool full_pstate_supported = false;
+       bool dummy_pstate_supported = false;
+       double p_state_latency_us = context->bw_ctx.dml.soc.dram_clock_change_latency_us;
+
+       if (fast_validate)
+               return dcn20_validate_bandwidth_internal(dc, context, true);
+
+
+       // Best case, we support full UCLK switch latency
+       voltage_supported = dcn20_validate_bandwidth_internal(dc, context, false);
+       full_pstate_supported = context->bw_ctx.bw.dcn.clk.p_state_change_support;
+
+       if (context->bw_ctx.dml.soc.dummy_pstate_latency_us == 0 ||
+               (voltage_supported && full_pstate_supported)) {
+               context->bw_ctx.bw.dcn.clk.p_state_change_support = true;
+               goto restore_dml_state;
+       }
+
+       // Fallback: Try to only support G6 temperature read latency
+       context->bw_ctx.dml.soc.dram_clock_change_latency_us = context->bw_ctx.dml.soc.dummy_pstate_latency_us;
+
+       voltage_supported = dcn20_validate_bandwidth_internal(dc, context, false);
+       dummy_pstate_supported = context->bw_ctx.bw.dcn.clk.p_state_change_support;
+
+       if (voltage_supported && dummy_pstate_supported) {
+               context->bw_ctx.bw.dcn.clk.p_state_change_support = false;
+               goto restore_dml_state;
+       }
+
+       // ERROR: fallback is supposed to always work.
+       ASSERT(false);
+
+restore_dml_state:
+       context->bw_ctx.dml.soc.dram_clock_change_latency_us = p_state_latency_us;
+
+       return voltage_supported;
+}
+
 struct pipe_ctx *dcn20_acquire_idle_pipe_for_layer(
                struct dc_state *state,
                const struct resource_pool *pool,
@@ -2576,9 +2960,6 @@ static void cap_soc_clocks(
                                                && max_clocks.uClockInKhz != 0)
                        bb->clock_limits[i].dram_speed_mts = (max_clocks.uClockInKhz / 1000) * 16;
 
-               // HACK: Force every uclk to max for now to "disable" uclk switching.
-               bb->clock_limits[i].dram_speed_mts = (max_clocks.uClockInKhz / 1000) * 16;
-
                if ((bb->clock_limits[i].fabricclk_mhz > (max_clocks.fabricClockInKhz / 1000))
                                                && max_clocks.fabricClockInKhz != 0)
                        bb->clock_limits[i].fabricclk_mhz = (max_clocks.fabricClockInKhz / 1000);
@@ -2674,6 +3055,10 @@ static void update_bounding_box(struct dc *dc, struct _vcs_dpi_soc_bounding_box_
                num_calculated_states++;
        }
 
+       calculated_states[num_calculated_states - 1].socclk_mhz = max_clocks->socClockInKhz / 1000;
+       calculated_states[num_calculated_states - 1].fabricclk_mhz = max_clocks->socClockInKhz / 1000;
+       calculated_states[num_calculated_states - 1].dcfclk_mhz = max_clocks->dcfClockInKhz / 1000;
+
        memcpy(bb->clock_limits, calculated_states, sizeof(bb->clock_limits));
        bb->num_states = num_calculated_states;
 
@@ -2711,6 +3096,27 @@ static void patch_bounding_box(struct dc *dc, struct _vcs_dpi_soc_bounding_box_s
        kernel_fpu_end();
 }
 
+static struct _vcs_dpi_soc_bounding_box_st *get_asic_rev_soc_bb(
+       uint32_t hw_internal_rev)
+{
+       if (ASICREV_IS_NAVI12_P(hw_internal_rev))
+               return &dcn2_0_nv12_soc;
+
+       return &dcn2_0_soc;
+}
+
+static struct _vcs_dpi_ip_params_st *get_asic_rev_ip_params(
+       uint32_t hw_internal_rev)
+{
+       /* NV12 and NV10 */
+       return &dcn2_0_ip;
+}
+
+static enum dml_project get_dml_project_version(uint32_t hw_internal_rev)
+{
+       return DML_PROJECT_NAVI10v2;
+}
+
 #define fixed16_to_double(x) (((double) x) / ((double) (1 << 16)))
 #define fixed16_to_double_to_cpu(x) fixed16_to_double(le32_to_cpu(x))
 
@@ -2718,6 +3124,11 @@ static bool init_soc_bounding_box(struct dc *dc,
                                  struct dcn20_resource_pool *pool)
 {
        const struct gpu_info_soc_bounding_box_v1_0 *bb = dc->soc_bounding_box;
+       struct _vcs_dpi_soc_bounding_box_st *loaded_bb =
+                       get_asic_rev_soc_bb(dc->ctx->asic_id.hw_internal_rev);
+       struct _vcs_dpi_ip_params_st *loaded_ip =
+                       get_asic_rev_ip_params(dc->ctx->asic_id.hw_internal_rev);
+
        DC_LOGGER_INIT(dc->ctx->logger);
 
        if (!bb && !SOC_BOUNDING_BOX_VALID) {
@@ -2728,103 +3139,103 @@ static bool init_soc_bounding_box(struct dc *dc,
        if (bb && !SOC_BOUNDING_BOX_VALID) {
                int i;
 
-               dcn2_0_soc.sr_exit_time_us =
+               dcn2_0_nv12_soc.sr_exit_time_us =
                                fixed16_to_double_to_cpu(bb->sr_exit_time_us);
-               dcn2_0_soc.sr_enter_plus_exit_time_us =
+               dcn2_0_nv12_soc.sr_enter_plus_exit_time_us =
                                fixed16_to_double_to_cpu(bb->sr_enter_plus_exit_time_us);
-               dcn2_0_soc.urgent_latency_us =
+               dcn2_0_nv12_soc.urgent_latency_us =
                                fixed16_to_double_to_cpu(bb->urgent_latency_us);
-               dcn2_0_soc.urgent_latency_pixel_data_only_us =
+               dcn2_0_nv12_soc.urgent_latency_pixel_data_only_us =
                                fixed16_to_double_to_cpu(bb->urgent_latency_pixel_data_only_us);
-               dcn2_0_soc.urgent_latency_pixel_mixed_with_vm_data_us =
+               dcn2_0_nv12_soc.urgent_latency_pixel_mixed_with_vm_data_us =
                                fixed16_to_double_to_cpu(bb->urgent_latency_pixel_mixed_with_vm_data_us);
-               dcn2_0_soc.urgent_latency_vm_data_only_us =
+               dcn2_0_nv12_soc.urgent_latency_vm_data_only_us =
                                fixed16_to_double_to_cpu(bb->urgent_latency_vm_data_only_us);
-               dcn2_0_soc.urgent_out_of_order_return_per_channel_pixel_only_bytes =
+               dcn2_0_nv12_soc.urgent_out_of_order_return_per_channel_pixel_only_bytes =
                                le32_to_cpu(bb->urgent_out_of_order_return_per_channel_pixel_only_bytes);
-               dcn2_0_soc.urgent_out_of_order_return_per_channel_pixel_and_vm_bytes =
+               dcn2_0_nv12_soc.urgent_out_of_order_return_per_channel_pixel_and_vm_bytes =
                                le32_to_cpu(bb->urgent_out_of_order_return_per_channel_pixel_and_vm_bytes);
-               dcn2_0_soc.urgent_out_of_order_return_per_channel_vm_only_bytes =
+               dcn2_0_nv12_soc.urgent_out_of_order_return_per_channel_vm_only_bytes =
                                le32_to_cpu(bb->urgent_out_of_order_return_per_channel_vm_only_bytes);
-               dcn2_0_soc.pct_ideal_dram_sdp_bw_after_urgent_pixel_only =
+               dcn2_0_nv12_soc.pct_ideal_dram_sdp_bw_after_urgent_pixel_only =
                                fixed16_to_double_to_cpu(bb->pct_ideal_dram_sdp_bw_after_urgent_pixel_only);
-               dcn2_0_soc.pct_ideal_dram_sdp_bw_after_urgent_pixel_and_vm =
+               dcn2_0_nv12_soc.pct_ideal_dram_sdp_bw_after_urgent_pixel_and_vm =
                                fixed16_to_double_to_cpu(bb->pct_ideal_dram_sdp_bw_after_urgent_pixel_and_vm);
-               dcn2_0_soc.pct_ideal_dram_sdp_bw_after_urgent_vm_only =
+               dcn2_0_nv12_soc.pct_ideal_dram_sdp_bw_after_urgent_vm_only =
                                fixed16_to_double_to_cpu(bb->pct_ideal_dram_sdp_bw_after_urgent_vm_only);
-               dcn2_0_soc.max_avg_sdp_bw_use_normal_percent =
+               dcn2_0_nv12_soc.max_avg_sdp_bw_use_normal_percent =
                                fixed16_to_double_to_cpu(bb->max_avg_sdp_bw_use_normal_percent);
-               dcn2_0_soc.max_avg_dram_bw_use_normal_percent =
+               dcn2_0_nv12_soc.max_avg_dram_bw_use_normal_percent =
                                fixed16_to_double_to_cpu(bb->max_avg_dram_bw_use_normal_percent);
-               dcn2_0_soc.writeback_latency_us =
+               dcn2_0_nv12_soc.writeback_latency_us =
                                fixed16_to_double_to_cpu(bb->writeback_latency_us);
-               dcn2_0_soc.ideal_dram_bw_after_urgent_percent =
+               dcn2_0_nv12_soc.ideal_dram_bw_after_urgent_percent =
                                fixed16_to_double_to_cpu(bb->ideal_dram_bw_after_urgent_percent);
-               dcn2_0_soc.max_request_size_bytes =
+               dcn2_0_nv12_soc.max_request_size_bytes =
                                le32_to_cpu(bb->max_request_size_bytes);
-               dcn2_0_soc.dram_channel_width_bytes =
+               dcn2_0_nv12_soc.dram_channel_width_bytes =
                                le32_to_cpu(bb->dram_channel_width_bytes);
-               dcn2_0_soc.fabric_datapath_to_dcn_data_return_bytes =
+               dcn2_0_nv12_soc.fabric_datapath_to_dcn_data_return_bytes =
                                le32_to_cpu(bb->fabric_datapath_to_dcn_data_return_bytes);
-               dcn2_0_soc.dcn_downspread_percent =
+               dcn2_0_nv12_soc.dcn_downspread_percent =
                                fixed16_to_double_to_cpu(bb->dcn_downspread_percent);
-               dcn2_0_soc.downspread_percent =
+               dcn2_0_nv12_soc.downspread_percent =
                                fixed16_to_double_to_cpu(bb->downspread_percent);
-               dcn2_0_soc.dram_page_open_time_ns =
+               dcn2_0_nv12_soc.dram_page_open_time_ns =
                                fixed16_to_double_to_cpu(bb->dram_page_open_time_ns);
-               dcn2_0_soc.dram_rw_turnaround_time_ns =
+               dcn2_0_nv12_soc.dram_rw_turnaround_time_ns =
                                fixed16_to_double_to_cpu(bb->dram_rw_turnaround_time_ns);
-               dcn2_0_soc.dram_return_buffer_per_channel_bytes =
+               dcn2_0_nv12_soc.dram_return_buffer_per_channel_bytes =
                                le32_to_cpu(bb->dram_return_buffer_per_channel_bytes);
-               dcn2_0_soc.round_trip_ping_latency_dcfclk_cycles =
+               dcn2_0_nv12_soc.round_trip_ping_latency_dcfclk_cycles =
                                le32_to_cpu(bb->round_trip_ping_latency_dcfclk_cycles);
-               dcn2_0_soc.urgent_out_of_order_return_per_channel_bytes =
+               dcn2_0_nv12_soc.urgent_out_of_order_return_per_channel_bytes =
                                le32_to_cpu(bb->urgent_out_of_order_return_per_channel_bytes);
-               dcn2_0_soc.channel_interleave_bytes =
+               dcn2_0_nv12_soc.channel_interleave_bytes =
                                le32_to_cpu(bb->channel_interleave_bytes);
-               dcn2_0_soc.num_banks =
+               dcn2_0_nv12_soc.num_banks =
                                le32_to_cpu(bb->num_banks);
-               dcn2_0_soc.num_chans =
+               dcn2_0_nv12_soc.num_chans =
                                le32_to_cpu(bb->num_chans);
-               dcn2_0_soc.vmm_page_size_bytes =
+               dcn2_0_nv12_soc.vmm_page_size_bytes =
                                le32_to_cpu(bb->vmm_page_size_bytes);
-               dcn2_0_soc.dram_clock_change_latency_us =
+               dcn2_0_nv12_soc.dram_clock_change_latency_us =
                                fixed16_to_double_to_cpu(bb->dram_clock_change_latency_us);
                // HACK!! Lower uclock latency switch time so we don't switch
-               dcn2_0_soc.dram_clock_change_latency_us = 10;
-               dcn2_0_soc.writeback_dram_clock_change_latency_us =
+               dcn2_0_nv12_soc.dram_clock_change_latency_us = 10;
+               dcn2_0_nv12_soc.writeback_dram_clock_change_latency_us =
                                fixed16_to_double_to_cpu(bb->writeback_dram_clock_change_latency_us);
-               dcn2_0_soc.return_bus_width_bytes =
+               dcn2_0_nv12_soc.return_bus_width_bytes =
                                le32_to_cpu(bb->return_bus_width_bytes);
-               dcn2_0_soc.dispclk_dppclk_vco_speed_mhz =
+               dcn2_0_nv12_soc.dispclk_dppclk_vco_speed_mhz =
                                le32_to_cpu(bb->dispclk_dppclk_vco_speed_mhz);
-               dcn2_0_soc.xfc_bus_transport_time_us =
+               dcn2_0_nv12_soc.xfc_bus_transport_time_us =
                                le32_to_cpu(bb->xfc_bus_transport_time_us);
-               dcn2_0_soc.xfc_xbuf_latency_tolerance_us =
+               dcn2_0_nv12_soc.xfc_xbuf_latency_tolerance_us =
                                le32_to_cpu(bb->xfc_xbuf_latency_tolerance_us);
-               dcn2_0_soc.use_urgent_burst_bw =
+               dcn2_0_nv12_soc.use_urgent_burst_bw =
                                le32_to_cpu(bb->use_urgent_burst_bw);
-               dcn2_0_soc.num_states =
+               dcn2_0_nv12_soc.num_states =
                                le32_to_cpu(bb->num_states);
 
-               for (i = 0; i < dcn2_0_soc.num_states; i++) {
-                       dcn2_0_soc.clock_limits[i].state =
+               for (i = 0; i < dcn2_0_nv12_soc.num_states; i++) {
+                       dcn2_0_nv12_soc.clock_limits[i].state =
                                        le32_to_cpu(bb->clock_limits[i].state);
-                       dcn2_0_soc.clock_limits[i].dcfclk_mhz =
+                       dcn2_0_nv12_soc.clock_limits[i].dcfclk_mhz =
                                        fixed16_to_double_to_cpu(bb->clock_limits[i].dcfclk_mhz);
-                       dcn2_0_soc.clock_limits[i].fabricclk_mhz =
+                       dcn2_0_nv12_soc.clock_limits[i].fabricclk_mhz =
                                        fixed16_to_double_to_cpu(bb->clock_limits[i].fabricclk_mhz);
-                       dcn2_0_soc.clock_limits[i].dispclk_mhz =
+                       dcn2_0_nv12_soc.clock_limits[i].dispclk_mhz =
                                        fixed16_to_double_to_cpu(bb->clock_limits[i].dispclk_mhz);
-                       dcn2_0_soc.clock_limits[i].dppclk_mhz =
+                       dcn2_0_nv12_soc.clock_limits[i].dppclk_mhz =
                                        fixed16_to_double_to_cpu(bb->clock_limits[i].dppclk_mhz);
-                       dcn2_0_soc.clock_limits[i].phyclk_mhz =
+                       dcn2_0_nv12_soc.clock_limits[i].phyclk_mhz =
                                        fixed16_to_double_to_cpu(bb->clock_limits[i].phyclk_mhz);
-                       dcn2_0_soc.clock_limits[i].socclk_mhz =
+                       dcn2_0_nv12_soc.clock_limits[i].socclk_mhz =
                                        fixed16_to_double_to_cpu(bb->clock_limits[i].socclk_mhz);
-                       dcn2_0_soc.clock_limits[i].dscclk_mhz =
+                       dcn2_0_nv12_soc.clock_limits[i].dscclk_mhz =
                                        fixed16_to_double_to_cpu(bb->clock_limits[i].dscclk_mhz);
-                       dcn2_0_soc.clock_limits[i].dram_speed_mts =
+                       dcn2_0_nv12_soc.clock_limits[i].dram_speed_mts =
                                        fixed16_to_double_to_cpu(bb->clock_limits[i].dram_speed_mts);
                }
        }
@@ -2833,7 +3244,6 @@ static bool init_soc_bounding_box(struct dc *dc,
                struct pp_smu_nv_clock_table max_clocks = {0};
                unsigned int uclk_states[8] = {0};
                unsigned int num_states = 0;
-               int i;
                enum pp_smu_status status;
                bool clock_limits_available = false;
                bool uclk_states_available = false;
@@ -2855,19 +3265,15 @@ static bool init_soc_bounding_box(struct dc *dc,
                        clock_limits_available = (status == PP_SMU_RESULT_OK);
                }
 
-               // HACK: Use the max uclk_states value for all elements.
-               for (i = 0; i < num_states; i++)
-                       uclk_states[i] = uclk_states[num_states - 1];
-
                if (clock_limits_available && uclk_states_available && num_states)
-                       update_bounding_box(dc, &dcn2_0_soc, &max_clocks, uclk_states, num_states);
+                       update_bounding_box(dc, loaded_bb, &max_clocks, uclk_states, num_states);
                else if (clock_limits_available)
-                       cap_soc_clocks(&dcn2_0_soc, max_clocks);
+                       cap_soc_clocks(loaded_bb, max_clocks);
        }
 
-       dcn2_0_ip.max_num_otg = pool->base.res_cap->num_timing_generator;
-       dcn2_0_ip.max_num_dpp = pool->base.pipe_count;
-       patch_bounding_box(dc, &dcn2_0_soc);
+       loaded_ip->max_num_otg = pool->base.res_cap->num_timing_generator;
+       loaded_ip->max_num_dpp = pool->base.pipe_count;
+       patch_bounding_box(dc, loaded_bb);
 
        return true;
 }
@@ -2880,19 +3286,30 @@ static bool construct(
        int i;
        struct dc_context *ctx = dc->ctx;
        struct irq_service_init_data init_data;
+       struct _vcs_dpi_soc_bounding_box_st *loaded_bb =
+                       get_asic_rev_soc_bb(ctx->asic_id.hw_internal_rev);
+       struct _vcs_dpi_ip_params_st *loaded_ip =
+                       get_asic_rev_ip_params(ctx->asic_id.hw_internal_rev);
+       enum dml_project dml_project_version =
+                       get_dml_project_version(ctx->asic_id.hw_internal_rev);
 
        ctx->dc_bios->regs = &bios_regs;
-
-       pool->base.res_cap = &res_cap_nv10;
        pool->base.funcs = &dcn20_res_pool_funcs;
 
+       if (ASICREV_IS_NAVI14_M(ctx->asic_id.hw_internal_rev)) {
+               pool->base.res_cap = &res_cap_nv14;
+               pool->base.pipe_count = 5;
+               pool->base.mpcc_count = 5;
+       } else {
+               pool->base.res_cap = &res_cap_nv10;
+               pool->base.pipe_count = 6;
+               pool->base.mpcc_count = 6;
+       }
        /*************************************************
         *  Resource + asic cap harcoding                *
         *************************************************/
        pool->base.underlay_pipe_index = NO_UNDERLAY_PIPE;
 
-       pool->base.pipe_count = 6;
-       pool->base.mpcc_count = 6;
        dc->caps.max_downscale_ratio = 200;
        dc->caps.i2c_speed_in_khz = 100;
        dc->caps.max_cursor_size = 256;
@@ -2998,7 +3415,7 @@ static bool construct(
                goto create_fail;
        }
 
-       dml_init_instance(&dc->dml, &dcn2_0_soc, &dcn2_0_ip, DML_PROJECT_NAVI10);
+       dml_init_instance(&dc->dml, loaded_bb, loaded_ip, dml_project_version);
 
        if (!dc->debug.disable_pplib_wm_range) {
                struct pp_smu_wm_range_sets ranges = {0};
@@ -3006,7 +3423,7 @@ static bool construct(
 
                ranges.num_reader_wm_sets = 0;
 
-               if (dcn2_0_soc.num_states == 1) {
+               if (loaded_bb->num_states == 1) {
                        ranges.reader_wm_sets[0].wm_inst = i;
                        ranges.reader_wm_sets[0].min_drain_clk_mhz = PP_SMU_WM_SET_RANGE_CLK_UNCONSTRAINED_MIN;
                        ranges.reader_wm_sets[0].max_drain_clk_mhz = PP_SMU_WM_SET_RANGE_CLK_UNCONSTRAINED_MAX;
@@ -3014,13 +3431,13 @@ static bool construct(
                        ranges.reader_wm_sets[0].max_fill_clk_mhz = PP_SMU_WM_SET_RANGE_CLK_UNCONSTRAINED_MAX;
 
                        ranges.num_reader_wm_sets = 1;
-               } else if (dcn2_0_soc.num_states > 1) {
-                       for (i = 0; i < 4 && i < dcn2_0_soc.num_states; i++) {
+               } else if (loaded_bb->num_states > 1) {
+                       for (i = 0; i < 4 && i < loaded_bb->num_states; i++) {
                                ranges.reader_wm_sets[i].wm_inst = i;
                                ranges.reader_wm_sets[i].min_drain_clk_mhz = PP_SMU_WM_SET_RANGE_CLK_UNCONSTRAINED_MIN;
                                ranges.reader_wm_sets[i].max_drain_clk_mhz = PP_SMU_WM_SET_RANGE_CLK_UNCONSTRAINED_MAX;
-                               ranges.reader_wm_sets[i].min_fill_clk_mhz = (i > 0) ? (dcn2_0_soc.clock_limits[i - 1].dram_speed_mts / 16) + 1 : 0;
-                               ranges.reader_wm_sets[i].max_fill_clk_mhz = dcn2_0_soc.clock_limits[i].dram_speed_mts / 16;
+                               ranges.reader_wm_sets[i].min_fill_clk_mhz = (i > 0) ? (loaded_bb->clock_limits[i - 1].dram_speed_mts / 16) + 1 : 0;
+                               ranges.reader_wm_sets[i].max_fill_clk_mhz = loaded_bb->clock_limits[i].dram_speed_mts / 16;
 
                                ranges.num_reader_wm_sets = i + 1;
                        }