1 // SPDX-License-Identifier: MIT
3 * Copyright 2022 Advanced Micro Devices, Inc.
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice shall be included in
13 * all copies or substantial portions of the Software.
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
19 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
20 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
21 * OTHER DEALINGS IN THE SOFTWARE.
29 #include "dcn31/dcn31_hubbub.h"
30 #include "dcn314_fpu.h"
31 #include "dml/dcn20/dcn20_fpu.h"
32 #include "dml/dcn31/dcn31_fpu.h"
33 #include "dml/display_mode_vba.h"
35 struct _vcs_dpi_ip_params_st dcn3_14_ip = {
36 .VBlankNomDefaultUS = 668,
38 .gpuvm_max_page_table_levels = 1,
40 .hostvm_max_page_table_levels = 2,
41 .rob_buffer_size_kbytes = 64,
42 .det_buffer_size_kbytes = DCN3_14_DEFAULT_DET_SIZE,
43 .config_return_buffer_size_in_kbytes = 1792,
44 .compressed_buffer_segment_size_in_kbytes = 64,
45 .meta_fifo_size_in_kentries = 32,
46 .zero_size_buffer_entries = 512,
47 .compbuf_reserved_space_64b = 256,
48 .compbuf_reserved_space_zs = 64,
49 .dpp_output_buffer_pixels = 2560,
50 .opp_output_buffer_lines = 1,
51 .pixel_chunk_size_kbytes = 8,
52 .meta_chunk_size_kbytes = 2,
53 .min_meta_chunk_size_bytes = 256,
54 .writeback_chunk_size_kbytes = 8,
55 .ptoi_supported = false,
57 .maximum_dsc_bits_per_component = 10,
58 .dsc422_native_support = false,
59 .is_line_buffer_bpp_fixed = true,
60 .line_buffer_fixed_bpp = 48,
61 .line_buffer_size_bits = 789504,
62 .max_line_buffer_lines = 12,
63 .writeback_interface_buffer_size_kbytes = 90,
66 .max_num_hdmi_frl_outputs = 1,
68 .max_dchub_pscl_bw_pix_per_clk = 4,
69 .max_pscl_lb_bw_pix_per_clk = 2,
70 .max_lb_vscl_bw_pix_per_clk = 4,
71 .max_vscl_hscl_bw_pix_per_clk = 4,
76 .dpte_buffer_size_in_pte_reqs_luma = 64,
77 .dpte_buffer_size_in_pte_reqs_chroma = 34,
78 .dispclk_ramp_margin_percent = 1,
79 .max_inter_dcn_tile_repeaters = 8,
80 .cursor_buffer_size = 16,
81 .cursor_chunk_size = 2,
82 .writeback_line_buffer_buffer_size = 0,
83 .writeback_min_hscl_ratio = 1,
84 .writeback_min_vscl_ratio = 1,
85 .writeback_max_hscl_ratio = 1,
86 .writeback_max_vscl_ratio = 1,
87 .writeback_max_hscl_taps = 1,
88 .writeback_max_vscl_taps = 1,
89 .dppclk_delay_subtotal = 46,
90 .dppclk_delay_scl = 50,
91 .dppclk_delay_scl_lb_only = 16,
92 .dppclk_delay_cnvc_formatter = 27,
93 .dppclk_delay_cnvc_cursor = 6,
94 .dispclk_delay_subtotal = 119,
95 .dynamic_metadata_vm_enabled = false,
96 .odm_combine_4to1_supported = false,
97 .dcc_supported = true,
100 static struct _vcs_dpi_soc_bounding_box_st dcn3_14_soc = {
101 /*TODO: correct dispclk/dppclk voltage level determination*/
105 .dispclk_mhz = 1200.0,
106 .dppclk_mhz = 1200.0,
108 .phyclk_d18_mhz = 667.0,
114 .dispclk_mhz = 1200.0,
115 .dppclk_mhz = 1200.0,
117 .phyclk_d18_mhz = 667.0,
123 .dispclk_mhz = 1200.0,
124 .dppclk_mhz = 1200.0,
126 .phyclk_d18_mhz = 667.0,
132 .dispclk_mhz = 1200.0,
133 .dppclk_mhz = 1200.0,
135 .phyclk_d18_mhz = 667.0,
141 .dispclk_mhz = 1200.0,
142 .dppclk_mhz = 1200.0,
144 .phyclk_d18_mhz = 667.0,
150 .sr_exit_time_us = 16.5,
151 .sr_enter_plus_exit_time_us = 18.5,
152 .sr_exit_z8_time_us = 268.0,
153 .sr_enter_plus_exit_z8_time_us = 393.0,
154 .writeback_latency_us = 12.0,
155 .dram_channel_width_bytes = 4,
156 .round_trip_ping_latency_dcfclk_cycles = 106,
157 .urgent_latency_pixel_data_only_us = 4.0,
158 .urgent_latency_pixel_mixed_with_vm_data_us = 4.0,
159 .urgent_latency_vm_data_only_us = 4.0,
160 .urgent_out_of_order_return_per_channel_pixel_only_bytes = 4096,
161 .urgent_out_of_order_return_per_channel_pixel_and_vm_bytes = 4096,
162 .urgent_out_of_order_return_per_channel_vm_only_bytes = 4096,
163 .pct_ideal_sdp_bw_after_urgent = 80.0,
164 .pct_ideal_dram_sdp_bw_after_urgent_pixel_only = 65.0,
165 .pct_ideal_dram_sdp_bw_after_urgent_pixel_and_vm = 60.0,
166 .pct_ideal_dram_sdp_bw_after_urgent_vm_only = 30.0,
167 .max_avg_sdp_bw_use_normal_percent = 60.0,
168 .max_avg_dram_bw_use_normal_percent = 60.0,
169 .fabric_datapath_to_dcn_data_return_bytes = 32,
170 .return_bus_width_bytes = 64,
171 .downspread_percent = 0.38,
172 .dcn_downspread_percent = 0.5,
173 .gpuvm_min_page_size_bytes = 4096,
174 .hostvm_min_page_size_bytes = 4096,
175 .do_urgent_latency_adjustment = false,
176 .urgent_latency_adjustment_fabric_clock_component_us = 0,
177 .urgent_latency_adjustment_fabric_clock_reference_mhz = 0,
181 void dcn314_update_bw_bounding_box_fpu(struct dc *dc, struct clk_bw_params *bw_params)
183 struct clk_limit_table *clk_table = &bw_params->clk_table;
184 struct _vcs_dpi_voltage_scaling_st *clock_limits =
185 dcn3_14_soc.clock_limits;
186 unsigned int i, closest_clk_lvl;
187 int max_dispclk_mhz = 0, max_dppclk_mhz = 0;
190 dc_assert_fp_enabled();
192 // Default clock levels are used for diags, which may lead to overclocking.
193 if (!IS_DIAG_DC(dc->ctx->dce_environment) && dc->config.use_default_clock_table == false) {
195 dcn3_14_ip.max_num_otg = dc->res_pool->res_cap->num_timing_generator;
196 dcn3_14_ip.max_num_dpp = dc->res_pool->pipe_count;
198 if (bw_params->dram_channel_width_bytes > 0)
199 dcn3_14_soc.dram_channel_width_bytes = bw_params->dram_channel_width_bytes;
201 if (bw_params->num_channels > 0)
202 dcn3_14_soc.num_chans = bw_params->num_channels;
204 ASSERT(dcn3_14_soc.num_chans);
205 ASSERT(clk_table->num_entries);
207 /* Prepass to find max clocks independent of voltage level. */
208 for (i = 0; i < clk_table->num_entries; ++i) {
209 if (clk_table->entries[i].dispclk_mhz > max_dispclk_mhz)
210 max_dispclk_mhz = clk_table->entries[i].dispclk_mhz;
211 if (clk_table->entries[i].dppclk_mhz > max_dppclk_mhz)
212 max_dppclk_mhz = clk_table->entries[i].dppclk_mhz;
215 for (i = 0; i < clk_table->num_entries; i++) {
217 for (closest_clk_lvl = 0, j = dcn3_14_soc.num_states - 1; j >= 0; j--) {
218 if ((unsigned int) dcn3_14_soc.clock_limits[j].dcfclk_mhz <= clk_table->entries[i].dcfclk_mhz) {
223 if (clk_table->num_entries == 1) {
224 /*smu gives one DPM level, let's take the highest one*/
225 closest_clk_lvl = dcn3_14_soc.num_states - 1;
228 clock_limits[i].state = i;
230 /* Clocks dependent on voltage level. */
231 clock_limits[i].dcfclk_mhz = clk_table->entries[i].dcfclk_mhz;
232 if (clk_table->num_entries == 1 &&
233 clock_limits[i].dcfclk_mhz < dcn3_14_soc.clock_limits[closest_clk_lvl].dcfclk_mhz) {
234 /*SMU fix not released yet*/
235 clock_limits[i].dcfclk_mhz = dcn3_14_soc.clock_limits[closest_clk_lvl].dcfclk_mhz;
237 clock_limits[i].fabricclk_mhz = clk_table->entries[i].fclk_mhz;
238 clock_limits[i].socclk_mhz = clk_table->entries[i].socclk_mhz;
240 if (clk_table->entries[i].memclk_mhz && clk_table->entries[i].wck_ratio)
241 clock_limits[i].dram_speed_mts = clk_table->entries[i].memclk_mhz * 2 * clk_table->entries[i].wck_ratio;
243 /* Clocks independent of voltage level. */
244 clock_limits[i].dispclk_mhz = max_dispclk_mhz ? max_dispclk_mhz :
245 dcn3_14_soc.clock_limits[closest_clk_lvl].dispclk_mhz;
247 clock_limits[i].dppclk_mhz = max_dppclk_mhz ? max_dppclk_mhz :
248 dcn3_14_soc.clock_limits[closest_clk_lvl].dppclk_mhz;
250 clock_limits[i].dram_bw_per_chan_gbps = dcn3_14_soc.clock_limits[closest_clk_lvl].dram_bw_per_chan_gbps;
251 clock_limits[i].dscclk_mhz = dcn3_14_soc.clock_limits[closest_clk_lvl].dscclk_mhz;
252 clock_limits[i].dtbclk_mhz = dcn3_14_soc.clock_limits[closest_clk_lvl].dtbclk_mhz;
253 clock_limits[i].phyclk_d18_mhz = dcn3_14_soc.clock_limits[closest_clk_lvl].phyclk_d18_mhz;
254 clock_limits[i].phyclk_mhz = dcn3_14_soc.clock_limits[closest_clk_lvl].phyclk_mhz;
256 for (i = 0; i < clk_table->num_entries; i++)
257 dcn3_14_soc.clock_limits[i] = clock_limits[i];
258 if (clk_table->num_entries) {
259 dcn3_14_soc.num_states = clk_table->num_entries;
263 if (max_dispclk_mhz) {
264 dcn3_14_soc.dispclk_dppclk_vco_speed_mhz = max_dispclk_mhz * 2;
265 dc->dml.soc.dispclk_dppclk_vco_speed_mhz = max_dispclk_mhz * 2;
268 dcn20_patch_bounding_box(dc, &dcn3_14_soc);
270 if (!IS_FPGA_MAXIMUS_DC(dc->ctx->dce_environment))
271 dml_init_instance(&dc->dml, &dcn3_14_soc, &dcn3_14_ip, DML_PROJECT_DCN314);
273 dml_init_instance(&dc->dml, &dcn3_14_soc, &dcn3_14_ip, DML_PROJECT_DCN31_FPGA);
276 static bool is_dual_plane(enum surface_pixel_format format)
278 return format >= SURFACE_PIXEL_FORMAT_VIDEO_BEGIN || format == SURFACE_PIXEL_FORMAT_GRPH_RGBE_ALPHA;
281 int dcn314_populate_dml_pipes_from_context_fpu(struct dc *dc, struct dc_state *context,
282 display_e2e_pipe_params_st *pipes,
286 struct resource_context *res_ctx = &context->res_ctx;
287 struct pipe_ctx *pipe;
288 bool upscaled = false;
289 bool isFreesyncVideo = false;
291 dc_assert_fp_enabled();
293 dcn31x_populate_dml_pipes_from_context(dc, context, pipes, fast_validate);
295 for (i = 0, pipe_cnt = 0; i < dc->res_pool->pipe_count; i++) {
296 struct dc_crtc_timing *timing;
298 if (!res_ctx->pipe_ctx[i].stream)
300 pipe = &res_ctx->pipe_ctx[i];
301 timing = &pipe->stream->timing;
303 isFreesyncVideo = pipe->stream->adjust.v_total_max == pipe->stream->adjust.v_total_min;
304 isFreesyncVideo = isFreesyncVideo && pipe->stream->adjust.v_total_min > timing->v_total;
306 if (!isFreesyncVideo) {
307 pipes[pipe_cnt].pipe.dest.vblank_nom =
308 dcn3_14_ip.VBlankNomDefaultUS / (timing->h_total / (timing->pix_clk_100hz / 10000.0));
310 pipes[pipe_cnt].pipe.dest.vtotal = pipe->stream->adjust.v_total_min;
311 pipes[pipe_cnt].pipe.dest.vblank_nom = timing->v_total - pipes[pipe_cnt].pipe.dest.vactive;
314 if (pipe->plane_state &&
315 (pipe->plane_state->src_rect.height < pipe->plane_state->dst_rect.height ||
316 pipe->plane_state->src_rect.width < pipe->plane_state->dst_rect.width))
319 /* Apply HostVM policy - either based on hypervisor globally enabled, or rIOMMU active */
320 if (dc->debug.dml_hostvm_override == DML_HOSTVM_NO_OVERRIDE)
321 pipes[i].pipe.src.hostvm = dc->vm_pa_config.is_hvm_enabled || dc->res_pool->hubbub->riommu_active;
324 * Immediate flip can be set dynamically after enabling the plane.
325 * We need to require support for immediate flip or underflow can be
326 * intermittently experienced depending on peak b/w requirements.
328 pipes[pipe_cnt].pipe.src.immediate_flip = true;
330 pipes[pipe_cnt].pipe.src.unbounded_req_mode = false;
331 pipes[pipe_cnt].pipe.src.dcc_fraction_of_zs_req_luma = 0;
332 pipes[pipe_cnt].pipe.src.dcc_fraction_of_zs_req_chroma = 0;
333 pipes[pipe_cnt].pipe.dest.vfront_porch = timing->v_front_porch;
334 pipes[pipe_cnt].pipe.src.dcc_rate = 3;
335 pipes[pipe_cnt].dout.dsc_input_bpc = 0;
337 if (pipes[pipe_cnt].dout.dsc_enable) {
338 switch (timing->display_color_depth) {
339 case COLOR_DEPTH_888:
340 pipes[pipe_cnt].dout.dsc_input_bpc = 8;
342 case COLOR_DEPTH_101010:
343 pipes[pipe_cnt].dout.dsc_input_bpc = 10;
345 case COLOR_DEPTH_121212:
346 pipes[pipe_cnt].dout.dsc_input_bpc = 12;
356 context->bw_ctx.dml.ip.det_buffer_size_kbytes = DCN3_14_DEFAULT_DET_SIZE;
358 dc->config.enable_4to1MPC = false;
359 if (pipe_cnt == 1 && pipe->plane_state
360 && pipe->plane_state->rotation == ROTATION_ANGLE_0 && !dc->debug.disable_z9_mpc) {
361 if (is_dual_plane(pipe->plane_state->format)
362 && pipe->plane_state->src_rect.width <= 1920 && pipe->plane_state->src_rect.height <= 1080) {
363 dc->config.enable_4to1MPC = true;
364 } else if (!is_dual_plane(pipe->plane_state->format) && pipe->plane_state->src_rect.width <= 5120) {
365 /* Limit to 5k max to avoid forced pipe split when there is not enough detile for swath */
366 context->bw_ctx.dml.ip.det_buffer_size_kbytes = 192;
367 pipes[0].pipe.src.unbounded_req_mode = true;
369 } else if (context->stream_count >= dc->debug.crb_alloc_policy_min_disp_count
370 && dc->debug.crb_alloc_policy > DET_SIZE_DEFAULT) {
371 context->bw_ctx.dml.ip.det_buffer_size_kbytes = dc->debug.crb_alloc_policy * 64;
372 } else if (context->stream_count >= 3 && upscaled) {
373 context->bw_ctx.dml.ip.det_buffer_size_kbytes = 192;
376 for (i = 0; i < dc->res_pool->pipe_count; i++) {
377 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
382 if (pipe->stream->signal == SIGNAL_TYPE_EDP && dc->debug.seamless_boot_odm_combine &&
383 pipe->stream->apply_seamless_boot_optimization) {
385 if (pipe->stream->apply_boot_odm_mode == dm_odm_combine_policy_2to1) {
386 context->bw_ctx.dml.vba.ODMCombinePolicy = dm_odm_combine_policy_2to1;