dcn303_fpu.c (13839B)
1/* 2 * Copyright 2019-2021 Advanced Micro Devices, Inc. 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 * 22 * Authors: AMD 23 * 24 */ 25#include "resource.h" 26#include "clk_mgr.h" 27#include "dcn20/dcn20_resource.h" 28#include "dcn303/dcn303_resource.h" 29 30#include "dml/dcn20/dcn20_fpu.h" 31#include "dcn303_fpu.h" 32 33struct _vcs_dpi_ip_params_st dcn3_03_ip = { 34 .use_min_dcfclk = 0, 35 .clamp_min_dcfclk = 0, 36 .odm_capable = 1, 37 .gpuvm_enable = 1, 38 .hostvm_enable = 0, 39 .gpuvm_max_page_table_levels = 4, 40 .hostvm_max_page_table_levels = 4, 41 .hostvm_cached_page_table_levels = 0, 42 .pte_group_size_bytes = 2048, 43 .num_dsc = 2, 44 .rob_buffer_size_kbytes = 184, 45 .det_buffer_size_kbytes = 184, 46 .dpte_buffer_size_in_pte_reqs_luma = 64, 47 .dpte_buffer_size_in_pte_reqs_chroma = 34, 48 .pde_proc_buffer_size_64k_reqs = 48, 49 .dpp_output_buffer_pixels = 2560, 50 .opp_output_buffer_lines = 1, 51 .pixel_chunk_size_kbytes = 8, 52 .pte_enable = 1, 53 .max_page_table_levels = 2, 54 .pte_chunk_size_kbytes = 2, // ? 55 .meta_chunk_size_kbytes = 2, 56 .writeback_chunk_size_kbytes = 8, 57 .line_buffer_size_bits = 789504, 58 .is_line_buffer_bpp_fixed = 0, // ? 59 .line_buffer_fixed_bpp = 0, // ? 60 .dcc_supported = true, 61 .writeback_interface_buffer_size_kbytes = 90, 62 .writeback_line_buffer_buffer_size = 0, 63 .max_line_buffer_lines = 12, 64 .writeback_luma_buffer_size_kbytes = 12, // writeback_line_buffer_buffer_size = 656640 65 .writeback_chroma_buffer_size_kbytes = 8, 66 .writeback_chroma_line_buffer_width_pixels = 4, 67 .writeback_max_hscl_ratio = 1, 68 .writeback_max_vscl_ratio = 1, 69 .writeback_min_hscl_ratio = 1, 70 .writeback_min_vscl_ratio = 1, 71 .writeback_max_hscl_taps = 1, 72 .writeback_max_vscl_taps = 1, 73 .writeback_line_buffer_luma_buffer_size = 0, 74 .writeback_line_buffer_chroma_buffer_size = 14643, 75 .cursor_buffer_size = 8, 76 .cursor_chunk_size = 2, 77 .max_num_otg = 2, 78 .max_num_dpp = 2, 79 .max_num_wb = 1, 80 .max_dchub_pscl_bw_pix_per_clk = 4, 81 .max_pscl_lb_bw_pix_per_clk = 2, 82 .max_lb_vscl_bw_pix_per_clk = 4, 83 .max_vscl_hscl_bw_pix_per_clk = 4, 84 .max_hscl_ratio = 6, 85 .max_vscl_ratio = 6, 86 .hscl_mults = 4, 87 .vscl_mults = 4, 88 .max_hscl_taps = 8, 89 .max_vscl_taps = 8, 90 .dispclk_ramp_margin_percent = 1, 91 .underscan_factor = 1.11, 92 .min_vblank_lines = 32, 93 .dppclk_delay_subtotal = 46, 94 .dynamic_metadata_vm_enabled = true, 95 .dppclk_delay_scl_lb_only = 16, 96 .dppclk_delay_scl = 50, 97 .dppclk_delay_cnvc_formatter = 27, 98 .dppclk_delay_cnvc_cursor = 6, 99 .dispclk_delay_subtotal = 119, 100 .dcfclk_cstate_latency = 5.2, // SRExitTime 101 .max_inter_dcn_tile_repeaters = 8, 102 .max_num_hdmi_frl_outputs = 1, 103 .odm_combine_4to1_supported = false, 104 105 .xfc_supported = false, 106 .xfc_fill_bw_overhead_percent = 10.0, 107 .xfc_fill_constant_bytes = 0, 108 .gfx7_compat_tiling_supported = 0, 109 .number_of_cursors = 1, 110}; 111 112struct _vcs_dpi_soc_bounding_box_st dcn3_03_soc = { 113 .clock_limits = { 114 { 115 .state = 0, 116 .dispclk_mhz = 562.0, 117 .dppclk_mhz = 300.0, 118 .phyclk_mhz = 300.0, 119 .phyclk_d18_mhz = 667.0, 120 .dscclk_mhz = 405.6, 121 }, 122 }, 123 124 .min_dcfclk = 500.0, /* TODO: set this to actual min DCFCLK */ 125 .num_states = 1, 126 .sr_exit_time_us = 35.5, 127 .sr_enter_plus_exit_time_us = 40, 128 .urgent_latency_us = 4.0, 129 .urgent_latency_pixel_data_only_us = 4.0, 130 .urgent_latency_pixel_mixed_with_vm_data_us = 4.0, 131 .urgent_latency_vm_data_only_us = 4.0, 132 .urgent_out_of_order_return_per_channel_pixel_only_bytes = 4096, 133 .urgent_out_of_order_return_per_channel_pixel_and_vm_bytes = 4096, 134 .urgent_out_of_order_return_per_channel_vm_only_bytes = 4096, 135 .pct_ideal_dram_sdp_bw_after_urgent_pixel_only = 80.0, 136 .pct_ideal_dram_sdp_bw_after_urgent_pixel_and_vm = 60.0, 137 .pct_ideal_dram_sdp_bw_after_urgent_vm_only = 40.0, 138 .max_avg_sdp_bw_use_normal_percent = 60.0, 139 .max_avg_dram_bw_use_normal_percent = 40.0, 140 .writeback_latency_us = 12.0, 141 .max_request_size_bytes = 256, 142 .fabric_datapath_to_dcn_data_return_bytes = 64, 143 .dcn_downspread_percent = 0.5, 144 .downspread_percent = 0.38, 145 .dram_page_open_time_ns = 50.0, 146 .dram_rw_turnaround_time_ns = 17.5, 147 .dram_return_buffer_per_channel_bytes = 8192, 148 .round_trip_ping_latency_dcfclk_cycles = 156, 149 .urgent_out_of_order_return_per_channel_bytes = 4096, 150 .channel_interleave_bytes = 256, 151 .num_banks = 8, 152 .gpuvm_min_page_size_bytes = 4096, 153 .hostvm_min_page_size_bytes = 4096, 154 .dram_clock_change_latency_us = 404, 155 .dummy_pstate_latency_us = 5, 156 .writeback_dram_clock_change_latency_us = 23.0, 157 .return_bus_width_bytes = 64, 158 .dispclk_dppclk_vco_speed_mhz = 3650, 159 .xfc_bus_transport_time_us = 20, // ? 160 .xfc_xbuf_latency_tolerance_us = 4, // ? 161 .use_urgent_burst_bw = 1, // ? 162 .do_urgent_latency_adjustment = true, 163 .urgent_latency_adjustment_fabric_clock_component_us = 1.0, 164 .urgent_latency_adjustment_fabric_clock_reference_mhz = 1000, 165}; 166 167static void dcn303_get_optimal_dcfclk_fclk_for_uclk(unsigned int uclk_mts, 168 unsigned int *optimal_dcfclk, 169 unsigned int *optimal_fclk) 170{ 171 double bw_from_dram, bw_from_dram1, bw_from_dram2; 172 173 bw_from_dram1 = uclk_mts * dcn3_03_soc.num_chans * 174 dcn3_03_soc.dram_channel_width_bytes * (dcn3_03_soc.max_avg_dram_bw_use_normal_percent / 100); 175 bw_from_dram2 = uclk_mts * dcn3_03_soc.num_chans * 176 dcn3_03_soc.dram_channel_width_bytes * (dcn3_03_soc.max_avg_sdp_bw_use_normal_percent / 100); 177 178 bw_from_dram = (bw_from_dram1 < bw_from_dram2) ? bw_from_dram1 : bw_from_dram2; 179 180 if (optimal_fclk) 181 *optimal_fclk = bw_from_dram / 182 (dcn3_03_soc.fabric_datapath_to_dcn_data_return_bytes * 183 (dcn3_03_soc.max_avg_sdp_bw_use_normal_percent / 100)); 184 185 if (optimal_dcfclk) 186 *optimal_dcfclk = bw_from_dram / 187 (dcn3_03_soc.return_bus_width_bytes * (dcn3_03_soc.max_avg_sdp_bw_use_normal_percent / 100)); 188} 189 190 191void dcn303_fpu_update_bw_bounding_box(struct dc *dc, struct clk_bw_params *bw_params) 192{ 193 unsigned int i, j; 194 unsigned int num_states = 0; 195 196 unsigned int dcfclk_mhz[DC__VOLTAGE_STATES] = {0}; 197 unsigned int dram_speed_mts[DC__VOLTAGE_STATES] = {0}; 198 unsigned int optimal_uclk_for_dcfclk_sta_targets[DC__VOLTAGE_STATES] = {0}; 199 unsigned int optimal_dcfclk_for_uclk[DC__VOLTAGE_STATES] = {0}; 200 201 unsigned int dcfclk_sta_targets[DC__VOLTAGE_STATES] = {694, 875, 1000, 1200}; 202 unsigned int num_dcfclk_sta_targets = 4; 203 unsigned int num_uclk_states; 204 205 dc_assert_fp_enabled(); 206 207 if (dc->ctx->dc_bios->vram_info.num_chans) 208 dcn3_03_soc.num_chans = dc->ctx->dc_bios->vram_info.num_chans; 209 210 if (dc->ctx->dc_bios->vram_info.dram_channel_width_bytes) 211 dcn3_03_soc.dram_channel_width_bytes = dc->ctx->dc_bios->vram_info.dram_channel_width_bytes; 212 213 dcn3_03_soc.dispclk_dppclk_vco_speed_mhz = dc->clk_mgr->dentist_vco_freq_khz / 1000.0; 214 dc->dml.soc.dispclk_dppclk_vco_speed_mhz = dc->clk_mgr->dentist_vco_freq_khz / 1000.0; 215 216 if (bw_params->clk_table.entries[0].memclk_mhz) { 217 int max_dcfclk_mhz = 0, max_dispclk_mhz = 0, max_dppclk_mhz = 0, max_phyclk_mhz = 0; 218 219 for (i = 0; i < MAX_NUM_DPM_LVL; i++) { 220 if (bw_params->clk_table.entries[i].dcfclk_mhz > max_dcfclk_mhz) 221 max_dcfclk_mhz = bw_params->clk_table.entries[i].dcfclk_mhz; 222 if (bw_params->clk_table.entries[i].dispclk_mhz > max_dispclk_mhz) 223 max_dispclk_mhz = bw_params->clk_table.entries[i].dispclk_mhz; 224 if (bw_params->clk_table.entries[i].dppclk_mhz > max_dppclk_mhz) 225 max_dppclk_mhz = bw_params->clk_table.entries[i].dppclk_mhz; 226 if (bw_params->clk_table.entries[i].phyclk_mhz > max_phyclk_mhz) 227 max_phyclk_mhz = bw_params->clk_table.entries[i].phyclk_mhz; 228 } 229 if (!max_dcfclk_mhz) 230 max_dcfclk_mhz = dcn3_03_soc.clock_limits[0].dcfclk_mhz; 231 if (!max_dispclk_mhz) 232 max_dispclk_mhz = dcn3_03_soc.clock_limits[0].dispclk_mhz; 233 if (!max_dppclk_mhz) 234 max_dppclk_mhz = dcn3_03_soc.clock_limits[0].dppclk_mhz; 235 if (!max_phyclk_mhz) 236 max_phyclk_mhz = dcn3_03_soc.clock_limits[0].phyclk_mhz; 237 238 if (max_dcfclk_mhz > dcfclk_sta_targets[num_dcfclk_sta_targets-1]) { 239 dcfclk_sta_targets[num_dcfclk_sta_targets] = max_dcfclk_mhz; 240 num_dcfclk_sta_targets++; 241 } else if (max_dcfclk_mhz < dcfclk_sta_targets[num_dcfclk_sta_targets-1]) { 242 for (i = 0; i < num_dcfclk_sta_targets; i++) { 243 if (dcfclk_sta_targets[i] > max_dcfclk_mhz) { 244 dcfclk_sta_targets[i] = max_dcfclk_mhz; 245 break; 246 } 247 } 248 /* Update size of array since we "removed" duplicates */ 249 num_dcfclk_sta_targets = i + 1; 250 } 251 252 num_uclk_states = bw_params->clk_table.num_entries; 253 254 /* Calculate optimal dcfclk for each uclk */ 255 for (i = 0; i < num_uclk_states; i++) { 256 dcn303_get_optimal_dcfclk_fclk_for_uclk(bw_params->clk_table.entries[i].memclk_mhz * 16, 257 &optimal_dcfclk_for_uclk[i], NULL); 258 if (optimal_dcfclk_for_uclk[i] < bw_params->clk_table.entries[0].dcfclk_mhz) 259 optimal_dcfclk_for_uclk[i] = bw_params->clk_table.entries[0].dcfclk_mhz; 260 } 261 262 /* Calculate optimal uclk for each dcfclk sta target */ 263 for (i = 0; i < num_dcfclk_sta_targets; i++) { 264 for (j = 0; j < num_uclk_states; j++) { 265 if (dcfclk_sta_targets[i] < optimal_dcfclk_for_uclk[j]) { 266 optimal_uclk_for_dcfclk_sta_targets[i] = 267 bw_params->clk_table.entries[j].memclk_mhz * 16; 268 break; 269 } 270 } 271 } 272 273 i = 0; 274 j = 0; 275 /* create the final dcfclk and uclk table */ 276 while (i < num_dcfclk_sta_targets && j < num_uclk_states && num_states < DC__VOLTAGE_STATES) { 277 if (dcfclk_sta_targets[i] < optimal_dcfclk_for_uclk[j] && i < num_dcfclk_sta_targets) { 278 dcfclk_mhz[num_states] = dcfclk_sta_targets[i]; 279 dram_speed_mts[num_states++] = optimal_uclk_for_dcfclk_sta_targets[i++]; 280 } else { 281 if (j < num_uclk_states && optimal_dcfclk_for_uclk[j] <= max_dcfclk_mhz) { 282 dcfclk_mhz[num_states] = optimal_dcfclk_for_uclk[j]; 283 dram_speed_mts[num_states++] = 284 bw_params->clk_table.entries[j++].memclk_mhz * 16; 285 } else { 286 j = num_uclk_states; 287 } 288 } 289 } 290 291 while (i < num_dcfclk_sta_targets && num_states < DC__VOLTAGE_STATES) { 292 dcfclk_mhz[num_states] = dcfclk_sta_targets[i]; 293 dram_speed_mts[num_states++] = optimal_uclk_for_dcfclk_sta_targets[i++]; 294 } 295 296 while (j < num_uclk_states && num_states < DC__VOLTAGE_STATES && 297 optimal_dcfclk_for_uclk[j] <= max_dcfclk_mhz) { 298 dcfclk_mhz[num_states] = optimal_dcfclk_for_uclk[j]; 299 dram_speed_mts[num_states++] = bw_params->clk_table.entries[j++].memclk_mhz * 16; 300 } 301 302 dcn3_03_soc.num_states = num_states; 303 for (i = 0; i < dcn3_03_soc.num_states; i++) { 304 dcn3_03_soc.clock_limits[i].state = i; 305 dcn3_03_soc.clock_limits[i].dcfclk_mhz = dcfclk_mhz[i]; 306 dcn3_03_soc.clock_limits[i].fabricclk_mhz = dcfclk_mhz[i]; 307 dcn3_03_soc.clock_limits[i].dram_speed_mts = dram_speed_mts[i]; 308 309 /* Fill all states with max values of all other clocks */ 310 dcn3_03_soc.clock_limits[i].dispclk_mhz = max_dispclk_mhz; 311 dcn3_03_soc.clock_limits[i].dppclk_mhz = max_dppclk_mhz; 312 dcn3_03_soc.clock_limits[i].phyclk_mhz = max_phyclk_mhz; 313 /* Populate from bw_params for DTBCLK, SOCCLK */ 314 if (!bw_params->clk_table.entries[i].dtbclk_mhz && i > 0) 315 dcn3_03_soc.clock_limits[i].dtbclk_mhz = dcn3_03_soc.clock_limits[i-1].dtbclk_mhz; 316 else 317 dcn3_03_soc.clock_limits[i].dtbclk_mhz = bw_params->clk_table.entries[i].dtbclk_mhz; 318 if (!bw_params->clk_table.entries[i].socclk_mhz && i > 0) 319 dcn3_03_soc.clock_limits[i].socclk_mhz = dcn3_03_soc.clock_limits[i-1].socclk_mhz; 320 else 321 dcn3_03_soc.clock_limits[i].socclk_mhz = bw_params->clk_table.entries[i].socclk_mhz; 322 /* These clocks cannot come from bw_params, always fill from dcn3_03_soc[1] */ 323 /* FCLK, PHYCLK_D18, DSCCLK */ 324 dcn3_03_soc.clock_limits[i].phyclk_d18_mhz = dcn3_03_soc.clock_limits[0].phyclk_d18_mhz; 325 dcn3_03_soc.clock_limits[i].dscclk_mhz = dcn3_03_soc.clock_limits[0].dscclk_mhz; 326 } 327 328 if (dcn3_03_soc.num_chans <= 4) { 329 for (i = 0; i < dcn3_03_soc.num_states; i++) { 330 if (dcn3_03_soc.clock_limits[i].dram_speed_mts > 1700) 331 break; 332 333 if (dcn3_03_soc.clock_limits[i].dram_speed_mts >= 1500) { 334 dcn3_03_soc.clock_limits[i].dcfclk_mhz = 100; 335 dcn3_03_soc.clock_limits[i].fabricclk_mhz = 100; 336 } 337 } 338 } 339 340 /* re-init DML with updated bb */ 341 dml_init_instance(&dc->dml, &dcn3_03_soc, &dcn3_03_ip, DML_PROJECT_DCN30); 342 if (dc->current_state) 343 dml_init_instance(&dc->current_state->bw_ctx.dml, &dcn3_03_soc, &dcn3_03_ip, DML_PROJECT_DCN30); 344 } 345} 346 347void dcn303_fpu_init_soc_bounding_box(struct bp_soc_bb_info bb_info) 348{ 349 dc_assert_fp_enabled(); 350 351 if (bb_info.dram_clock_change_latency_100ns > 0) 352 dcn3_03_soc.dram_clock_change_latency_us = 353 bb_info.dram_clock_change_latency_100ns * 10; 354 355 if (bb_info.dram_sr_enter_exit_latency_100ns > 0) 356 dcn3_03_soc.sr_enter_plus_exit_time_us = 357 bb_info.dram_sr_enter_exit_latency_100ns * 10; 358 359 if (bb_info.dram_sr_exit_latency_100ns > 0) 360 dcn3_03_soc.sr_exit_time_us = 361 bb_info.dram_sr_exit_latency_100ns * 10; 362}