1 // SPDX-License-Identifier: MIT 2 /* 3 * Copyright 2022 Advanced Micro Devices, Inc. 4 * 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: 11 * 12 * The above copyright notice and this permission notice shall be included in 13 * all copies or substantial portions of the Software. 14 * 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. 22 * 23 * Authors: AMD 24 * 25 */ 26 #include "dcn32_fpu.h" 27 #include "dcn32/dcn32_resource.h" 28 #include "dcn20/dcn20_resource.h" 29 #include "display_mode_vba_util_32.h" 30 #include "dml/dcn32/display_mode_vba_32.h" 31 // We need this includes for WATERMARKS_* defines 32 #include "clk_mgr/dcn32/dcn32_smu13_driver_if.h" 33 #include "dcn30/dcn30_resource.h" 34 #include "link.h" 35 36 #define DC_LOGGER_INIT(logger) 37 38 static const struct subvp_high_refresh_list subvp_high_refresh_list = { 39 .min_refresh = 120, 40 .max_refresh = 165, 41 .res = { 42 {.width = 3840, .height = 2160, }, 43 {.width = 3440, .height = 1440, }, 44 {.width = 2560, .height = 1440, }}, 45 }; 46 47 struct _vcs_dpi_ip_params_st dcn3_2_ip = { 48 .gpuvm_enable = 0, 49 .gpuvm_max_page_table_levels = 4, 50 .hostvm_enable = 0, 51 .rob_buffer_size_kbytes = 128, 52 .det_buffer_size_kbytes = DCN3_2_DEFAULT_DET_SIZE, 53 .config_return_buffer_size_in_kbytes = 1280, 54 .compressed_buffer_segment_size_in_kbytes = 64, 55 .meta_fifo_size_in_kentries = 22, 56 .zero_size_buffer_entries = 512, 57 .compbuf_reserved_space_64b = 256, 58 .compbuf_reserved_space_zs = 64, 59 .dpp_output_buffer_pixels = 2560, 60 .opp_output_buffer_lines = 1, 61 .pixel_chunk_size_kbytes = 8, 62 .alpha_pixel_chunk_size_kbytes = 4, 63 .min_pixel_chunk_size_bytes = 1024, 64 .dcc_meta_buffer_size_bytes = 6272, 65 .meta_chunk_size_kbytes = 2, 66 .min_meta_chunk_size_bytes = 256, 67 .writeback_chunk_size_kbytes = 8, 68 .ptoi_supported = false, 69 .num_dsc = 4, 70 .maximum_dsc_bits_per_component = 12, 71 .maximum_pixels_per_line_per_dsc_unit = 6016, 72 .dsc422_native_support = true, 73 .is_line_buffer_bpp_fixed = true, 74 .line_buffer_fixed_bpp = 57, 75 .line_buffer_size_bits = 1171920, 76 .max_line_buffer_lines = 32, 77 .writeback_interface_buffer_size_kbytes = 90, 78 .max_num_dpp = 4, 79 .max_num_otg = 4, 80 .max_num_hdmi_frl_outputs = 1, 81 .max_num_wb = 1, 82 .max_dchub_pscl_bw_pix_per_clk = 4, 83 .max_pscl_lb_bw_pix_per_clk = 2, 84 .max_lb_vscl_bw_pix_per_clk = 4, 85 .max_vscl_hscl_bw_pix_per_clk = 4, 86 .max_hscl_ratio = 6, 87 .max_vscl_ratio = 6, 88 .max_hscl_taps = 8, 89 .max_vscl_taps = 8, 90 .dpte_buffer_size_in_pte_reqs_luma = 64, 91 .dpte_buffer_size_in_pte_reqs_chroma = 34, 92 .dispclk_ramp_margin_percent = 1, 93 .max_inter_dcn_tile_repeaters = 8, 94 .cursor_buffer_size = 16, 95 .cursor_chunk_size = 2, 96 .writeback_line_buffer_buffer_size = 0, 97 .writeback_min_hscl_ratio = 1, 98 .writeback_min_vscl_ratio = 1, 99 .writeback_max_hscl_ratio = 1, 100 .writeback_max_vscl_ratio = 1, 101 .writeback_max_hscl_taps = 1, 102 .writeback_max_vscl_taps = 1, 103 .dppclk_delay_subtotal = 47, 104 .dppclk_delay_scl = 50, 105 .dppclk_delay_scl_lb_only = 16, 106 .dppclk_delay_cnvc_formatter = 28, 107 .dppclk_delay_cnvc_cursor = 6, 108 .dispclk_delay_subtotal = 125, 109 .dynamic_metadata_vm_enabled = false, 110 .odm_combine_4to1_supported = false, 111 .dcc_supported = true, 112 .max_num_dp2p0_outputs = 2, 113 .max_num_dp2p0_streams = 4, 114 }; 115 116 struct _vcs_dpi_soc_bounding_box_st dcn3_2_soc = { 117 .clock_limits = { 118 { 119 .state = 0, 120 .dcfclk_mhz = 1564.0, 121 .fabricclk_mhz = 2500.0, 122 .dispclk_mhz = 2150.0, 123 .dppclk_mhz = 2150.0, 124 .phyclk_mhz = 810.0, 125 .phyclk_d18_mhz = 667.0, 126 .phyclk_d32_mhz = 625.0, 127 .socclk_mhz = 1200.0, 128 .dscclk_mhz = 716.667, 129 .dram_speed_mts = 18000.0, 130 .dtbclk_mhz = 1564.0, 131 }, 132 }, 133 .num_states = 1, 134 .sr_exit_time_us = 42.97, 135 .sr_enter_plus_exit_time_us = 49.94, 136 .sr_exit_z8_time_us = 285.0, 137 .sr_enter_plus_exit_z8_time_us = 320, 138 .writeback_latency_us = 12.0, 139 .round_trip_ping_latency_dcfclk_cycles = 263, 140 .urgent_latency_pixel_data_only_us = 4.0, 141 .urgent_latency_pixel_mixed_with_vm_data_us = 4.0, 142 .urgent_latency_vm_data_only_us = 4.0, 143 .fclk_change_latency_us = 25, 144 .usr_retraining_latency_us = 2, 145 .smn_latency_us = 2, 146 .mall_allocated_for_dcn_mbytes = 64, 147 .urgent_out_of_order_return_per_channel_pixel_only_bytes = 4096, 148 .urgent_out_of_order_return_per_channel_pixel_and_vm_bytes = 4096, 149 .urgent_out_of_order_return_per_channel_vm_only_bytes = 4096, 150 .pct_ideal_sdp_bw_after_urgent = 100.0, 151 .pct_ideal_fabric_bw_after_urgent = 67.0, 152 .pct_ideal_dram_sdp_bw_after_urgent_pixel_only = 20.0, 153 .pct_ideal_dram_sdp_bw_after_urgent_pixel_and_vm = 60.0, // N/A, for now keep as is until DML implemented 154 .pct_ideal_dram_sdp_bw_after_urgent_vm_only = 30.0, // N/A, for now keep as is until DML implemented 155 .pct_ideal_dram_bw_after_urgent_strobe = 67.0, 156 .max_avg_sdp_bw_use_normal_percent = 80.0, 157 .max_avg_fabric_bw_use_normal_percent = 60.0, 158 .max_avg_dram_bw_use_normal_strobe_percent = 50.0, 159 .max_avg_dram_bw_use_normal_percent = 15.0, 160 .num_chans = 24, 161 .dram_channel_width_bytes = 2, 162 .fabric_datapath_to_dcn_data_return_bytes = 64, 163 .return_bus_width_bytes = 64, 164 .downspread_percent = 0.38, 165 .dcn_downspread_percent = 0.5, 166 .dram_clock_change_latency_us = 400, 167 .dispclk_dppclk_vco_speed_mhz = 4300.0, 168 .do_urgent_latency_adjustment = true, 169 .urgent_latency_adjustment_fabric_clock_component_us = 1.0, 170 .urgent_latency_adjustment_fabric_clock_reference_mhz = 3000, 171 }; 172 173 void dcn32_build_wm_range_table_fpu(struct clk_mgr_internal *clk_mgr) 174 { 175 /* defaults */ 176 double pstate_latency_us = clk_mgr->base.ctx->dc->dml.soc.dram_clock_change_latency_us; 177 double fclk_change_latency_us = clk_mgr->base.ctx->dc->dml.soc.fclk_change_latency_us; 178 double sr_exit_time_us = clk_mgr->base.ctx->dc->dml.soc.sr_exit_time_us; 179 double sr_enter_plus_exit_time_us = clk_mgr->base.ctx->dc->dml.soc.sr_enter_plus_exit_time_us; 180 /* For min clocks use as reported by PM FW and report those as min */ 181 uint16_t min_uclk_mhz = clk_mgr->base.bw_params->clk_table.entries[0].memclk_mhz; 182 uint16_t min_dcfclk_mhz = clk_mgr->base.bw_params->clk_table.entries[0].dcfclk_mhz; 183 uint16_t setb_min_uclk_mhz = min_uclk_mhz; 184 uint16_t dcfclk_mhz_for_the_second_state = clk_mgr->base.ctx->dc->dml.soc.clock_limits[2].dcfclk_mhz; 185 186 dc_assert_fp_enabled(); 187 188 /* For Set B ranges use min clocks state 2 when available, and report those to PM FW */ 189 if (dcfclk_mhz_for_the_second_state) 190 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.min_dcfclk = dcfclk_mhz_for_the_second_state; 191 else 192 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.min_dcfclk = clk_mgr->base.bw_params->clk_table.entries[0].dcfclk_mhz; 193 194 if (clk_mgr->base.bw_params->clk_table.entries[2].memclk_mhz) 195 setb_min_uclk_mhz = clk_mgr->base.bw_params->clk_table.entries[2].memclk_mhz; 196 197 /* Set A - Normal - default values */ 198 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].valid = true; 199 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us = pstate_latency_us; 200 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].dml_input.fclk_change_latency_us = fclk_change_latency_us; 201 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].dml_input.sr_exit_time_us = sr_exit_time_us; 202 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].dml_input.sr_enter_plus_exit_time_us = sr_enter_plus_exit_time_us; 203 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.wm_type = WATERMARKS_CLOCK_RANGE; 204 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.min_dcfclk = min_dcfclk_mhz; 205 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.max_dcfclk = 0xFFFF; 206 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.min_uclk = min_uclk_mhz; 207 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.max_uclk = 0xFFFF; 208 209 /* Set B - Performance - higher clocks, using DPM[2] DCFCLK and UCLK */ 210 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].valid = true; 211 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].dml_input.pstate_latency_us = pstate_latency_us; 212 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].dml_input.fclk_change_latency_us = fclk_change_latency_us; 213 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].dml_input.sr_exit_time_us = sr_exit_time_us; 214 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].dml_input.sr_enter_plus_exit_time_us = sr_enter_plus_exit_time_us; 215 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.wm_type = WATERMARKS_CLOCK_RANGE; 216 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.max_dcfclk = 0xFFFF; 217 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.min_uclk = setb_min_uclk_mhz; 218 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.max_uclk = 0xFFFF; 219 220 /* Set C - Dummy P-State - P-State latency set to "dummy p-state" value */ 221 /* 'DalDummyClockChangeLatencyNs' registry key option set to 0x7FFFFFFF can be used to disable Set C for dummy p-state */ 222 if (clk_mgr->base.ctx->dc->bb_overrides.dummy_clock_change_latency_ns != 0x7FFFFFFF) { 223 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].valid = true; 224 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].dml_input.pstate_latency_us = 50; 225 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].dml_input.fclk_change_latency_us = fclk_change_latency_us; 226 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].dml_input.sr_exit_time_us = sr_exit_time_us; 227 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].dml_input.sr_enter_plus_exit_time_us = sr_enter_plus_exit_time_us; 228 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.wm_type = WATERMARKS_DUMMY_PSTATE; 229 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.min_dcfclk = min_dcfclk_mhz; 230 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.max_dcfclk = 0xFFFF; 231 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.min_uclk = min_uclk_mhz; 232 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.max_uclk = 0xFFFF; 233 clk_mgr->base.bw_params->dummy_pstate_table[0].dram_speed_mts = clk_mgr->base.bw_params->clk_table.entries[0].memclk_mhz * 16; 234 clk_mgr->base.bw_params->dummy_pstate_table[0].dummy_pstate_latency_us = 50; 235 clk_mgr->base.bw_params->dummy_pstate_table[1].dram_speed_mts = clk_mgr->base.bw_params->clk_table.entries[1].memclk_mhz * 16; 236 clk_mgr->base.bw_params->dummy_pstate_table[1].dummy_pstate_latency_us = 9; 237 clk_mgr->base.bw_params->dummy_pstate_table[2].dram_speed_mts = clk_mgr->base.bw_params->clk_table.entries[2].memclk_mhz * 16; 238 clk_mgr->base.bw_params->dummy_pstate_table[2].dummy_pstate_latency_us = 8; 239 clk_mgr->base.bw_params->dummy_pstate_table[3].dram_speed_mts = clk_mgr->base.bw_params->clk_table.entries[3].memclk_mhz * 16; 240 clk_mgr->base.bw_params->dummy_pstate_table[3].dummy_pstate_latency_us = 5; 241 } 242 /* Set D - MALL - SR enter and exit time specific to MALL, TBD after bringup or later phase for now use DRAM values / 2 */ 243 /* For MALL DRAM clock change latency is N/A, for watermak calculations use lowest value dummy P state latency */ 244 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].valid = true; 245 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].dml_input.pstate_latency_us = clk_mgr->base.bw_params->dummy_pstate_table[3].dummy_pstate_latency_us; 246 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].dml_input.fclk_change_latency_us = fclk_change_latency_us; 247 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].dml_input.sr_exit_time_us = sr_exit_time_us / 2; // TBD 248 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].dml_input.sr_enter_plus_exit_time_us = sr_enter_plus_exit_time_us / 2; // TBD 249 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.wm_type = WATERMARKS_MALL; 250 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.min_dcfclk = min_dcfclk_mhz; 251 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.max_dcfclk = 0xFFFF; 252 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.min_uclk = min_uclk_mhz; 253 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.max_uclk = 0xFFFF; 254 } 255 256 /* 257 * Finds dummy_latency_index when MCLK switching using firmware based 258 * vblank stretch is enabled. This function will iterate through the 259 * table of dummy pstate latencies until the lowest value that allows 260 * dm_allow_self_refresh_and_mclk_switch to happen is found 261 */ 262 int dcn32_find_dummy_latency_index_for_fw_based_mclk_switch(struct dc *dc, 263 struct dc_state *context, 264 display_e2e_pipe_params_st *pipes, 265 int pipe_cnt, 266 int vlevel) 267 { 268 const int max_latency_table_entries = 4; 269 struct vba_vars_st *vba = &context->bw_ctx.dml.vba; 270 int dummy_latency_index = 0; 271 enum clock_change_support temp_clock_change_support = vba->DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb]; 272 273 dc_assert_fp_enabled(); 274 275 while (dummy_latency_index < max_latency_table_entries) { 276 if (temp_clock_change_support != dm_dram_clock_change_unsupported) 277 vba->DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] = temp_clock_change_support; 278 context->bw_ctx.dml.soc.dram_clock_change_latency_us = 279 dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us; 280 dcn32_internal_validate_bw(dc, context, pipes, &pipe_cnt, &vlevel, false); 281 282 /* for subvp + DRR case, if subvp pipes are still present we support pstate */ 283 if (vba->DRAMClockChangeSupport[vlevel][vba->maxMpcComb] == dm_dram_clock_change_unsupported && 284 dcn32_subvp_in_use(dc, context)) 285 vba->DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] = temp_clock_change_support; 286 287 if (vlevel < context->bw_ctx.dml.vba.soc.num_states && 288 vba->DRAMClockChangeSupport[vlevel][vba->maxMpcComb] != dm_dram_clock_change_unsupported) 289 break; 290 291 dummy_latency_index++; 292 } 293 294 if (dummy_latency_index == max_latency_table_entries) { 295 ASSERT(dummy_latency_index != max_latency_table_entries); 296 /* If the execution gets here, it means dummy p_states are 297 * not possible. This should never happen and would mean 298 * something is severely wrong. 299 * Here we reset dummy_latency_index to 3, because it is 300 * better to have underflows than system crashes. 301 */ 302 dummy_latency_index = max_latency_table_entries - 1; 303 } 304 305 return dummy_latency_index; 306 } 307 308 /** 309 * dcn32_helper_populate_phantom_dlg_params - Get DLG params for phantom pipes 310 * and populate pipe_ctx with those params. 311 * @dc: [in] current dc state 312 * @context: [in] new dc state 313 * @pipes: [in] DML pipe params array 314 * @pipe_cnt: [in] DML pipe count 315 * 316 * This function must be called AFTER the phantom pipes are added to context 317 * and run through DML (so that the DLG params for the phantom pipes can be 318 * populated), and BEFORE we program the timing for the phantom pipes. 319 */ 320 void dcn32_helper_populate_phantom_dlg_params(struct dc *dc, 321 struct dc_state *context, 322 display_e2e_pipe_params_st *pipes, 323 int pipe_cnt) 324 { 325 uint32_t i, pipe_idx; 326 327 dc_assert_fp_enabled(); 328 329 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) { 330 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i]; 331 332 if (!pipe->stream) 333 continue; 334 335 if (pipe->plane_state && pipe->stream->mall_stream_config.type == SUBVP_PHANTOM) { 336 pipes[pipe_idx].pipe.dest.vstartup_start = 337 get_vstartup(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx); 338 pipes[pipe_idx].pipe.dest.vupdate_offset = 339 get_vupdate_offset(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx); 340 pipes[pipe_idx].pipe.dest.vupdate_width = 341 get_vupdate_width(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx); 342 pipes[pipe_idx].pipe.dest.vready_offset = 343 get_vready_offset(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx); 344 pipe->pipe_dlg_param = pipes[pipe_idx].pipe.dest; 345 } 346 pipe_idx++; 347 } 348 } 349 350 /** 351 * dcn32_predict_pipe_split - Predict if pipe split will occur for a given DML pipe 352 * @context: [in] New DC state to be programmed 353 * @pipe_e2e: [in] DML pipe end to end context 354 * 355 * This function takes in a DML pipe (pipe_e2e) and predicts if pipe split is required (both 356 * ODM and MPC). For pipe split, ODM combine is determined by the ODM mode, and MPC combine is 357 * determined by DPPClk requirements 358 * 359 * This function follows the same policy as DML: 360 * - Check for ODM combine requirements / policy first 361 * - MPC combine is only chosen if there is no ODM combine requirements / policy in place, and 362 * MPC is required 363 * 364 * Return: Number of splits expected (1 for 2:1 split, 3 for 4:1 split, 0 for no splits). 365 */ 366 uint8_t dcn32_predict_pipe_split(struct dc_state *context, 367 display_e2e_pipe_params_st *pipe_e2e) 368 { 369 double pscl_throughput; 370 double pscl_throughput_chroma; 371 double dpp_clk_single_dpp, clock; 372 double clk_frequency = 0.0; 373 double vco_speed = context->bw_ctx.dml.soc.dispclk_dppclk_vco_speed_mhz; 374 bool total_available_pipes_support = false; 375 uint32_t number_of_dpp = 0; 376 enum odm_combine_mode odm_mode = dm_odm_combine_mode_disabled; 377 double req_dispclk_per_surface = 0; 378 uint8_t num_splits = 0; 379 380 dc_assert_fp_enabled(); 381 382 dml32_CalculateODMMode(context->bw_ctx.dml.ip.maximum_pixels_per_line_per_dsc_unit, 383 pipe_e2e->pipe.dest.hactive, 384 pipe_e2e->dout.output_format, 385 pipe_e2e->dout.output_type, 386 pipe_e2e->pipe.dest.odm_combine_policy, 387 context->bw_ctx.dml.soc.clock_limits[context->bw_ctx.dml.soc.num_states - 1].dispclk_mhz, 388 context->bw_ctx.dml.soc.clock_limits[context->bw_ctx.dml.soc.num_states - 1].dispclk_mhz, 389 pipe_e2e->dout.dsc_enable != 0, 390 0, /* TotalNumberOfActiveDPP can be 0 since we're predicting pipe split requirement */ 391 context->bw_ctx.dml.ip.max_num_dpp, 392 pipe_e2e->pipe.dest.pixel_rate_mhz, 393 context->bw_ctx.dml.soc.dcn_downspread_percent, 394 context->bw_ctx.dml.ip.dispclk_ramp_margin_percent, 395 context->bw_ctx.dml.soc.dispclk_dppclk_vco_speed_mhz, 396 pipe_e2e->dout.dsc_slices, 397 /* Output */ 398 &total_available_pipes_support, 399 &number_of_dpp, 400 &odm_mode, 401 &req_dispclk_per_surface); 402 403 dml32_CalculateSinglePipeDPPCLKAndSCLThroughput(pipe_e2e->pipe.scale_ratio_depth.hscl_ratio, 404 pipe_e2e->pipe.scale_ratio_depth.hscl_ratio_c, 405 pipe_e2e->pipe.scale_ratio_depth.vscl_ratio, 406 pipe_e2e->pipe.scale_ratio_depth.vscl_ratio_c, 407 context->bw_ctx.dml.ip.max_dchub_pscl_bw_pix_per_clk, 408 context->bw_ctx.dml.ip.max_pscl_lb_bw_pix_per_clk, 409 pipe_e2e->pipe.dest.pixel_rate_mhz, 410 pipe_e2e->pipe.src.source_format, 411 pipe_e2e->pipe.scale_taps.htaps, 412 pipe_e2e->pipe.scale_taps.htaps_c, 413 pipe_e2e->pipe.scale_taps.vtaps, 414 pipe_e2e->pipe.scale_taps.vtaps_c, 415 /* Output */ 416 &pscl_throughput, &pscl_throughput_chroma, 417 &dpp_clk_single_dpp); 418 419 clock = dpp_clk_single_dpp * (1 + context->bw_ctx.dml.soc.dcn_downspread_percent / 100); 420 421 if (clock > 0) 422 clk_frequency = vco_speed * 4.0 / ((int)(vco_speed * 4.0) / clock); 423 424 if (odm_mode == dm_odm_combine_mode_2to1) 425 num_splits = 1; 426 else if (odm_mode == dm_odm_combine_mode_4to1) 427 num_splits = 3; 428 else if (clk_frequency > context->bw_ctx.dml.soc.clock_limits[context->bw_ctx.dml.soc.num_states - 1].dppclk_mhz) 429 num_splits = 1; 430 431 return num_splits; 432 } 433 434 static float calculate_net_bw_in_kbytes_sec(struct _vcs_dpi_voltage_scaling_st *entry) 435 { 436 float memory_bw_kbytes_sec; 437 float fabric_bw_kbytes_sec; 438 float sdp_bw_kbytes_sec; 439 float limiting_bw_kbytes_sec; 440 441 memory_bw_kbytes_sec = entry->dram_speed_mts * 442 dcn3_2_soc.num_chans * 443 dcn3_2_soc.dram_channel_width_bytes * 444 ((float)dcn3_2_soc.pct_ideal_dram_sdp_bw_after_urgent_pixel_only / 100); 445 446 fabric_bw_kbytes_sec = entry->fabricclk_mhz * 447 dcn3_2_soc.return_bus_width_bytes * 448 ((float)dcn3_2_soc.pct_ideal_fabric_bw_after_urgent / 100); 449 450 sdp_bw_kbytes_sec = entry->dcfclk_mhz * 451 dcn3_2_soc.return_bus_width_bytes * 452 ((float)dcn3_2_soc.pct_ideal_sdp_bw_after_urgent / 100); 453 454 limiting_bw_kbytes_sec = memory_bw_kbytes_sec; 455 456 if (fabric_bw_kbytes_sec < limiting_bw_kbytes_sec) 457 limiting_bw_kbytes_sec = fabric_bw_kbytes_sec; 458 459 if (sdp_bw_kbytes_sec < limiting_bw_kbytes_sec) 460 limiting_bw_kbytes_sec = sdp_bw_kbytes_sec; 461 462 return limiting_bw_kbytes_sec; 463 } 464 465 static void get_optimal_ntuple(struct _vcs_dpi_voltage_scaling_st *entry) 466 { 467 if (entry->dcfclk_mhz > 0) { 468 float bw_on_sdp = entry->dcfclk_mhz * dcn3_2_soc.return_bus_width_bytes * ((float)dcn3_2_soc.pct_ideal_sdp_bw_after_urgent / 100); 469 470 entry->fabricclk_mhz = bw_on_sdp / (dcn3_2_soc.return_bus_width_bytes * ((float)dcn3_2_soc.pct_ideal_fabric_bw_after_urgent / 100)); 471 entry->dram_speed_mts = bw_on_sdp / (dcn3_2_soc.num_chans * 472 dcn3_2_soc.dram_channel_width_bytes * ((float)dcn3_2_soc.pct_ideal_dram_sdp_bw_after_urgent_pixel_only / 100)); 473 } else if (entry->fabricclk_mhz > 0) { 474 float bw_on_fabric = entry->fabricclk_mhz * dcn3_2_soc.return_bus_width_bytes * ((float)dcn3_2_soc.pct_ideal_fabric_bw_after_urgent / 100); 475 476 entry->dcfclk_mhz = bw_on_fabric / (dcn3_2_soc.return_bus_width_bytes * ((float)dcn3_2_soc.pct_ideal_sdp_bw_after_urgent / 100)); 477 entry->dram_speed_mts = bw_on_fabric / (dcn3_2_soc.num_chans * 478 dcn3_2_soc.dram_channel_width_bytes * ((float)dcn3_2_soc.pct_ideal_dram_sdp_bw_after_urgent_pixel_only / 100)); 479 } else if (entry->dram_speed_mts > 0) { 480 float bw_on_dram = entry->dram_speed_mts * dcn3_2_soc.num_chans * 481 dcn3_2_soc.dram_channel_width_bytes * ((float)dcn3_2_soc.pct_ideal_dram_sdp_bw_after_urgent_pixel_only / 100); 482 483 entry->fabricclk_mhz = bw_on_dram / (dcn3_2_soc.return_bus_width_bytes * ((float)dcn3_2_soc.pct_ideal_fabric_bw_after_urgent / 100)); 484 entry->dcfclk_mhz = bw_on_dram / (dcn3_2_soc.return_bus_width_bytes * ((float)dcn3_2_soc.pct_ideal_sdp_bw_after_urgent / 100)); 485 } 486 } 487 488 void insert_entry_into_table_sorted(struct _vcs_dpi_voltage_scaling_st *table, 489 unsigned int *num_entries, 490 struct _vcs_dpi_voltage_scaling_st *entry) 491 { 492 int i = 0; 493 int index = 0; 494 float net_bw_of_new_state = 0; 495 496 dc_assert_fp_enabled(); 497 498 get_optimal_ntuple(entry); 499 500 if (*num_entries == 0) { 501 table[0] = *entry; 502 (*num_entries)++; 503 } else { 504 net_bw_of_new_state = calculate_net_bw_in_kbytes_sec(entry); 505 while (net_bw_of_new_state > calculate_net_bw_in_kbytes_sec(&table[index])) { 506 index++; 507 if (index >= *num_entries) 508 break; 509 } 510 511 for (i = *num_entries; i > index; i--) 512 table[i] = table[i - 1]; 513 514 table[index] = *entry; 515 (*num_entries)++; 516 } 517 } 518 519 /** 520 * dcn32_set_phantom_stream_timing - Set timing params for the phantom stream 521 * @dc: current dc state 522 * @context: new dc state 523 * @ref_pipe: Main pipe for the phantom stream 524 * @phantom_stream: target phantom stream state 525 * @pipes: DML pipe params 526 * @pipe_cnt: number of DML pipes 527 * @dc_pipe_idx: DC pipe index for the main pipe (i.e. ref_pipe) 528 * 529 * Set timing params of the phantom stream based on calculated output from DML. 530 * This function first gets the DML pipe index using the DC pipe index, then 531 * calls into DML (get_subviewport_lines_needed_in_mall) to get the number of 532 * lines required for SubVP MCLK switching and assigns to the phantom stream 533 * accordingly. 534 * 535 * - The number of SubVP lines calculated in DML does not take into account 536 * FW processing delays and required pstate allow width, so we must include 537 * that separately. 538 * 539 * - Set phantom backporch = vstartup of main pipe 540 */ 541 void dcn32_set_phantom_stream_timing(struct dc *dc, 542 struct dc_state *context, 543 struct pipe_ctx *ref_pipe, 544 struct dc_stream_state *phantom_stream, 545 display_e2e_pipe_params_st *pipes, 546 unsigned int pipe_cnt, 547 unsigned int dc_pipe_idx) 548 { 549 unsigned int i, pipe_idx; 550 struct pipe_ctx *pipe; 551 uint32_t phantom_vactive, phantom_bp, pstate_width_fw_delay_lines; 552 unsigned int num_dpp; 553 unsigned int vlevel = context->bw_ctx.dml.vba.VoltageLevel; 554 unsigned int dcfclk = context->bw_ctx.dml.vba.DCFCLKState[vlevel][context->bw_ctx.dml.vba.maxMpcComb]; 555 unsigned int socclk = context->bw_ctx.dml.vba.SOCCLKPerState[vlevel]; 556 struct vba_vars_st *vba = &context->bw_ctx.dml.vba; 557 struct dc_stream_state *main_stream = ref_pipe->stream; 558 559 dc_assert_fp_enabled(); 560 561 // Find DML pipe index (pipe_idx) using dc_pipe_idx 562 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) { 563 pipe = &context->res_ctx.pipe_ctx[i]; 564 565 if (!pipe->stream) 566 continue; 567 568 if (i == dc_pipe_idx) 569 break; 570 571 pipe_idx++; 572 } 573 574 // Calculate lines required for pstate allow width and FW processing delays 575 pstate_width_fw_delay_lines = ((double)(dc->caps.subvp_fw_processing_delay_us + 576 dc->caps.subvp_pstate_allow_width_us) / 1000000) * 577 (ref_pipe->stream->timing.pix_clk_100hz * 100) / 578 (double)ref_pipe->stream->timing.h_total; 579 580 // Update clks_cfg for calling into recalculate 581 pipes[0].clks_cfg.voltage = vlevel; 582 pipes[0].clks_cfg.dcfclk_mhz = dcfclk; 583 pipes[0].clks_cfg.socclk_mhz = socclk; 584 585 // DML calculation for MALL region doesn't take into account FW delay 586 // and required pstate allow width for multi-display cases 587 /* Add 16 lines margin to the MALL REGION because SUB_VP_START_LINE must be aligned 588 * to 2 swaths (i.e. 16 lines) 589 */ 590 phantom_vactive = get_subviewport_lines_needed_in_mall(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx) + 591 pstate_width_fw_delay_lines + dc->caps.subvp_swath_height_margin_lines; 592 593 // W/A for DCC corruption with certain high resolution timings. 594 // Determing if pipesplit is used. If so, add meta_row_height to the phantom vactive. 595 num_dpp = vba->NoOfDPP[vba->VoltageLevel][vba->maxMpcComb][vba->pipe_plane[pipe_idx]]; 596 phantom_vactive += num_dpp > 1 ? vba->meta_row_height[vba->pipe_plane[pipe_idx]] : 0; 597 598 /* dc->debug.subvp_extra_lines 0 by default*/ 599 phantom_vactive += dc->debug.subvp_extra_lines; 600 601 // For backporch of phantom pipe, use vstartup of the main pipe 602 phantom_bp = get_vstartup(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx); 603 604 phantom_stream->dst.y = 0; 605 phantom_stream->dst.height = phantom_vactive; 606 /* When scaling, DML provides the end to end required number of lines for MALL. 607 * dst.height is always correct for this case, but src.height is not which causes a 608 * delta between main and phantom pipe scaling outputs. Need to adjust src.height on 609 * phantom for this case. 610 */ 611 phantom_stream->src.y = 0; 612 phantom_stream->src.height = (double)phantom_vactive * (double)main_stream->src.height / (double)main_stream->dst.height; 613 614 phantom_stream->timing.v_addressable = phantom_vactive; 615 phantom_stream->timing.v_front_porch = 1; 616 phantom_stream->timing.v_total = phantom_stream->timing.v_addressable + 617 phantom_stream->timing.v_front_porch + 618 phantom_stream->timing.v_sync_width + 619 phantom_bp; 620 phantom_stream->timing.flags.DSC = 0; // Don't need DSC for phantom timing 621 } 622 623 /** 624 * dcn32_get_num_free_pipes - Calculate number of free pipes 625 * @dc: current dc state 626 * @context: new dc state 627 * 628 * This function assumes that a "used" pipe is a pipe that has 629 * both a stream and a plane assigned to it. 630 * 631 * Return: Number of free pipes available in the context 632 */ 633 static unsigned int dcn32_get_num_free_pipes(struct dc *dc, struct dc_state *context) 634 { 635 unsigned int i; 636 unsigned int free_pipes = 0; 637 unsigned int num_pipes = 0; 638 639 for (i = 0; i < dc->res_pool->pipe_count; i++) { 640 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i]; 641 642 if (pipe->stream && !pipe->top_pipe) { 643 while (pipe) { 644 num_pipes++; 645 pipe = pipe->bottom_pipe; 646 } 647 } 648 } 649 650 free_pipes = dc->res_pool->pipe_count - num_pipes; 651 return free_pipes; 652 } 653 654 /** 655 * dcn32_assign_subvp_pipe - Function to decide which pipe will use Sub-VP. 656 * @dc: current dc state 657 * @context: new dc state 658 * @index: [out] dc pipe index for the pipe chosen to have phantom pipes assigned 659 * 660 * We enter this function if we are Sub-VP capable (i.e. enough pipes available) 661 * and regular P-State switching (i.e. VACTIVE/VBLANK) is not supported, or if 662 * we are forcing SubVP P-State switching on the current config. 663 * 664 * The number of pipes used for the chosen surface must be less than or equal to the 665 * number of free pipes available. 666 * 667 * In general we choose surfaces with the longest frame time first (better for SubVP + VBLANK). 668 * For multi-display cases the ActiveDRAMClockChangeMargin doesn't provide enough info on its own 669 * for determining which should be the SubVP pipe (need a way to determine if a pipe / plane doesn't 670 * support MCLK switching naturally [i.e. ACTIVE or VBLANK]). 671 * 672 * Return: True if a valid pipe assignment was found for Sub-VP. Otherwise false. 673 */ 674 static bool dcn32_assign_subvp_pipe(struct dc *dc, 675 struct dc_state *context, 676 unsigned int *index) 677 { 678 unsigned int i, pipe_idx; 679 unsigned int max_frame_time = 0; 680 bool valid_assignment_found = false; 681 unsigned int free_pipes = dcn32_get_num_free_pipes(dc, context); 682 bool current_assignment_freesync = false; 683 struct vba_vars_st *vba = &context->bw_ctx.dml.vba; 684 685 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) { 686 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i]; 687 unsigned int num_pipes = 0; 688 unsigned int refresh_rate = 0; 689 690 if (!pipe->stream) 691 continue; 692 693 // Round up 694 refresh_rate = (pipe->stream->timing.pix_clk_100hz * 100 + 695 pipe->stream->timing.v_total * pipe->stream->timing.h_total - 1) 696 / (double)(pipe->stream->timing.v_total * pipe->stream->timing.h_total); 697 /* SubVP pipe candidate requirements: 698 * - Refresh rate < 120hz 699 * - Not able to switch in vactive naturally (switching in active means the 700 * DET provides enough buffer to hide the P-State switch latency -- trying 701 * to combine this with SubVP can cause issues with the scheduling). 702 * - Not TMZ surface 703 */ 704 if (pipe->plane_state && !pipe->top_pipe && !dcn32_is_center_timing(pipe) && 705 (!dcn32_is_psr_capable(pipe) || (context->stream_count == 1 && dc->caps.dmub_caps.subvp_psr)) && 706 pipe->stream->mall_stream_config.type == SUBVP_NONE && 707 (refresh_rate < 120 || dcn32_allow_subvp_high_refresh_rate(dc, context, pipe)) && 708 !pipe->plane_state->address.tmz_surface && 709 (vba->ActiveDRAMClockChangeLatencyMarginPerState[vba->VoltageLevel][vba->maxMpcComb][vba->pipe_plane[pipe_idx]] <= 0 || 710 (vba->ActiveDRAMClockChangeLatencyMarginPerState[vba->VoltageLevel][vba->maxMpcComb][vba->pipe_plane[pipe_idx]] > 0 && 711 dcn32_allow_subvp_with_active_margin(pipe)))) { 712 while (pipe) { 713 num_pipes++; 714 pipe = pipe->bottom_pipe; 715 } 716 717 pipe = &context->res_ctx.pipe_ctx[i]; 718 if (num_pipes <= free_pipes) { 719 struct dc_stream_state *stream = pipe->stream; 720 unsigned int frame_us = (stream->timing.v_total * stream->timing.h_total / 721 (double)(stream->timing.pix_clk_100hz * 100)) * 1000000; 722 if (frame_us > max_frame_time && !stream->ignore_msa_timing_param) { 723 *index = i; 724 max_frame_time = frame_us; 725 valid_assignment_found = true; 726 current_assignment_freesync = false; 727 /* For the 2-Freesync display case, still choose the one with the 728 * longest frame time 729 */ 730 } else if (stream->ignore_msa_timing_param && (!valid_assignment_found || 731 (current_assignment_freesync && frame_us > max_frame_time))) { 732 *index = i; 733 valid_assignment_found = true; 734 current_assignment_freesync = true; 735 } 736 } 737 } 738 pipe_idx++; 739 } 740 return valid_assignment_found; 741 } 742 743 /** 744 * dcn32_enough_pipes_for_subvp - Function to check if there are "enough" pipes for SubVP. 745 * @dc: current dc state 746 * @context: new dc state 747 * 748 * This function returns true if there are enough free pipes 749 * to create the required phantom pipes for any given stream 750 * (that does not already have phantom pipe assigned). 751 * 752 * e.g. For a 2 stream config where the first stream uses one 753 * pipe and the second stream uses 2 pipes (i.e. pipe split), 754 * this function will return true because there is 1 remaining 755 * pipe which can be used as the phantom pipe for the non pipe 756 * split pipe. 757 * 758 * Return: 759 * True if there are enough free pipes to assign phantom pipes to at least one 760 * stream that does not already have phantom pipes assigned. Otherwise false. 761 */ 762 static bool dcn32_enough_pipes_for_subvp(struct dc *dc, struct dc_state *context) 763 { 764 unsigned int i, split_cnt, free_pipes; 765 unsigned int min_pipe_split = dc->res_pool->pipe_count + 1; // init as max number of pipes + 1 766 bool subvp_possible = false; 767 768 for (i = 0; i < dc->res_pool->pipe_count; i++) { 769 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i]; 770 771 // Find the minimum pipe split count for non SubVP pipes 772 if (pipe->stream && !pipe->top_pipe && 773 pipe->stream->mall_stream_config.type == SUBVP_NONE) { 774 split_cnt = 0; 775 while (pipe) { 776 split_cnt++; 777 pipe = pipe->bottom_pipe; 778 } 779 780 if (split_cnt < min_pipe_split) 781 min_pipe_split = split_cnt; 782 } 783 } 784 785 free_pipes = dcn32_get_num_free_pipes(dc, context); 786 787 // SubVP only possible if at least one pipe is being used (i.e. free_pipes 788 // should not equal to the pipe_count) 789 if (free_pipes >= min_pipe_split && free_pipes < dc->res_pool->pipe_count) 790 subvp_possible = true; 791 792 return subvp_possible; 793 } 794 795 /** 796 * subvp_subvp_schedulable - Determine if SubVP + SubVP config is schedulable 797 * @dc: current dc state 798 * @context: new dc state 799 * 800 * High level algorithm: 801 * 1. Find longest microschedule length (in us) between the two SubVP pipes 802 * 2. Check if the worst case overlap (VBLANK in middle of ACTIVE) for both 803 * pipes still allows for the maximum microschedule to fit in the active 804 * region for both pipes. 805 * 806 * Return: True if the SubVP + SubVP config is schedulable, false otherwise 807 */ 808 static bool subvp_subvp_schedulable(struct dc *dc, struct dc_state *context) 809 { 810 struct pipe_ctx *subvp_pipes[2]; 811 struct dc_stream_state *phantom = NULL; 812 uint32_t microschedule_lines = 0; 813 uint32_t index = 0; 814 uint32_t i; 815 uint32_t max_microschedule_us = 0; 816 int32_t vactive1_us, vactive2_us, vblank1_us, vblank2_us; 817 818 for (i = 0; i < dc->res_pool->pipe_count; i++) { 819 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i]; 820 uint32_t time_us = 0; 821 822 /* Loop to calculate the maximum microschedule time between the two SubVP pipes, 823 * and also to store the two main SubVP pipe pointers in subvp_pipes[2]. 824 */ 825 if (pipe->stream && pipe->plane_state && !pipe->top_pipe && 826 pipe->stream->mall_stream_config.type == SUBVP_MAIN) { 827 phantom = pipe->stream->mall_stream_config.paired_stream; 828 microschedule_lines = (phantom->timing.v_total - phantom->timing.v_front_porch) + 829 phantom->timing.v_addressable; 830 831 // Round up when calculating microschedule time (+ 1 at the end) 832 time_us = (microschedule_lines * phantom->timing.h_total) / 833 (double)(phantom->timing.pix_clk_100hz * 100) * 1000000 + 834 dc->caps.subvp_prefetch_end_to_mall_start_us + 835 dc->caps.subvp_fw_processing_delay_us + 1; 836 if (time_us > max_microschedule_us) 837 max_microschedule_us = time_us; 838 839 subvp_pipes[index] = pipe; 840 index++; 841 842 // Maximum 2 SubVP pipes 843 if (index == 2) 844 break; 845 } 846 } 847 vactive1_us = ((subvp_pipes[0]->stream->timing.v_addressable * subvp_pipes[0]->stream->timing.h_total) / 848 (double)(subvp_pipes[0]->stream->timing.pix_clk_100hz * 100)) * 1000000; 849 vactive2_us = ((subvp_pipes[1]->stream->timing.v_addressable * subvp_pipes[1]->stream->timing.h_total) / 850 (double)(subvp_pipes[1]->stream->timing.pix_clk_100hz * 100)) * 1000000; 851 vblank1_us = (((subvp_pipes[0]->stream->timing.v_total - subvp_pipes[0]->stream->timing.v_addressable) * 852 subvp_pipes[0]->stream->timing.h_total) / 853 (double)(subvp_pipes[0]->stream->timing.pix_clk_100hz * 100)) * 1000000; 854 vblank2_us = (((subvp_pipes[1]->stream->timing.v_total - subvp_pipes[1]->stream->timing.v_addressable) * 855 subvp_pipes[1]->stream->timing.h_total) / 856 (double)(subvp_pipes[1]->stream->timing.pix_clk_100hz * 100)) * 1000000; 857 858 if ((vactive1_us - vblank2_us) / 2 > max_microschedule_us && 859 (vactive2_us - vblank1_us) / 2 > max_microschedule_us) 860 return true; 861 862 return false; 863 } 864 865 /** 866 * subvp_drr_schedulable - Determine if SubVP + DRR config is schedulable 867 * @dc: current dc state 868 * @context: new dc state 869 * @drr_pipe: DRR pipe_ctx for the SubVP + DRR config 870 * 871 * High level algorithm: 872 * 1. Get timing for SubVP pipe, phantom pipe, and DRR pipe 873 * 2. Determine the frame time for the DRR display when adding required margin for MCLK switching 874 * (the margin is equal to the MALL region + DRR margin (500us)) 875 * 3.If (SubVP Active - Prefetch > Stretched DRR frame + max(MALL region, Stretched DRR frame)) 876 * then report the configuration as supported 877 * 878 * Return: True if the SubVP + DRR config is schedulable, false otherwise 879 */ 880 static bool subvp_drr_schedulable(struct dc *dc, struct dc_state *context, struct pipe_ctx *drr_pipe) 881 { 882 bool schedulable = false; 883 uint32_t i; 884 struct pipe_ctx *pipe = NULL; 885 struct dc_crtc_timing *main_timing = NULL; 886 struct dc_crtc_timing *phantom_timing = NULL; 887 struct dc_crtc_timing *drr_timing = NULL; 888 int16_t prefetch_us = 0; 889 int16_t mall_region_us = 0; 890 int16_t drr_frame_us = 0; // nominal frame time 891 int16_t subvp_active_us = 0; 892 int16_t stretched_drr_us = 0; 893 int16_t drr_stretched_vblank_us = 0; 894 int16_t max_vblank_mallregion = 0; 895 const struct dc_config *config = &dc->config; 896 897 if (config->disable_subvp_drr) 898 return false; 899 900 // Find SubVP pipe 901 for (i = 0; i < dc->res_pool->pipe_count; i++) { 902 pipe = &context->res_ctx.pipe_ctx[i]; 903 904 // We check for master pipe, but it shouldn't matter since we only need 905 // the pipe for timing info (stream should be same for any pipe splits) 906 if (!pipe->stream || !pipe->plane_state || pipe->top_pipe || pipe->prev_odm_pipe) 907 continue; 908 909 // Find the SubVP pipe 910 if (pipe->stream->mall_stream_config.type == SUBVP_MAIN) 911 break; 912 } 913 914 main_timing = &pipe->stream->timing; 915 phantom_timing = &pipe->stream->mall_stream_config.paired_stream->timing; 916 drr_timing = &drr_pipe->stream->timing; 917 prefetch_us = (phantom_timing->v_total - phantom_timing->v_front_porch) * phantom_timing->h_total / 918 (double)(phantom_timing->pix_clk_100hz * 100) * 1000000 + 919 dc->caps.subvp_prefetch_end_to_mall_start_us; 920 subvp_active_us = main_timing->v_addressable * main_timing->h_total / 921 (double)(main_timing->pix_clk_100hz * 100) * 1000000; 922 drr_frame_us = drr_timing->v_total * drr_timing->h_total / 923 (double)(drr_timing->pix_clk_100hz * 100) * 1000000; 924 // P-State allow width and FW delays already included phantom_timing->v_addressable 925 mall_region_us = phantom_timing->v_addressable * phantom_timing->h_total / 926 (double)(phantom_timing->pix_clk_100hz * 100) * 1000000; 927 stretched_drr_us = drr_frame_us + mall_region_us + SUBVP_DRR_MARGIN_US; 928 drr_stretched_vblank_us = (drr_timing->v_total - drr_timing->v_addressable) * drr_timing->h_total / 929 (double)(drr_timing->pix_clk_100hz * 100) * 1000000 + (stretched_drr_us - drr_frame_us); 930 max_vblank_mallregion = drr_stretched_vblank_us > mall_region_us ? drr_stretched_vblank_us : mall_region_us; 931 932 /* We consider SubVP + DRR schedulable if the stretched frame duration of the DRR display (i.e. the 933 * highest refresh rate + margin that can support UCLK P-State switch) passes the static analysis 934 * for VBLANK: (VACTIVE region of the SubVP pipe can fit the MALL prefetch, VBLANK frame time, 935 * and the max of (VBLANK blanking time, MALL region)). 936 */ 937 if (stretched_drr_us < (1 / (double)drr_timing->min_refresh_in_uhz) * 1000000 * 1000000 && 938 subvp_active_us - prefetch_us - stretched_drr_us - max_vblank_mallregion > 0) 939 schedulable = true; 940 941 return schedulable; 942 } 943 944 945 /** 946 * subvp_vblank_schedulable - Determine if SubVP + VBLANK config is schedulable 947 * @dc: current dc state 948 * @context: new dc state 949 * 950 * High level algorithm: 951 * 1. Get timing for SubVP pipe, phantom pipe, and VBLANK pipe 952 * 2. If (SubVP Active - Prefetch > Vblank Frame Time + max(MALL region, Vblank blanking time)) 953 * then report the configuration as supported 954 * 3. If the VBLANK display is DRR, then take the DRR static schedulability path 955 * 956 * Return: True if the SubVP + VBLANK/DRR config is schedulable, false otherwise 957 */ 958 static bool subvp_vblank_schedulable(struct dc *dc, struct dc_state *context) 959 { 960 struct pipe_ctx *pipe = NULL; 961 struct pipe_ctx *subvp_pipe = NULL; 962 bool found = false; 963 bool schedulable = false; 964 uint32_t i = 0; 965 uint8_t vblank_index = 0; 966 uint16_t prefetch_us = 0; 967 uint16_t mall_region_us = 0; 968 uint16_t vblank_frame_us = 0; 969 uint16_t subvp_active_us = 0; 970 uint16_t vblank_blank_us = 0; 971 uint16_t max_vblank_mallregion = 0; 972 struct dc_crtc_timing *main_timing = NULL; 973 struct dc_crtc_timing *phantom_timing = NULL; 974 struct dc_crtc_timing *vblank_timing = NULL; 975 976 /* For SubVP + VBLANK/DRR cases, we assume there can only be 977 * a single VBLANK/DRR display. If DML outputs SubVP + VBLANK 978 * is supported, it is either a single VBLANK case or two VBLANK 979 * displays which are synchronized (in which case they have identical 980 * timings). 981 */ 982 for (i = 0; i < dc->res_pool->pipe_count; i++) { 983 pipe = &context->res_ctx.pipe_ctx[i]; 984 985 // We check for master pipe, but it shouldn't matter since we only need 986 // the pipe for timing info (stream should be same for any pipe splits) 987 if (!pipe->stream || !pipe->plane_state || pipe->top_pipe || pipe->prev_odm_pipe) 988 continue; 989 990 if (!found && pipe->stream->mall_stream_config.type == SUBVP_NONE) { 991 // Found pipe which is not SubVP or Phantom (i.e. the VBLANK pipe). 992 vblank_index = i; 993 found = true; 994 } 995 996 if (!subvp_pipe && pipe->stream->mall_stream_config.type == SUBVP_MAIN) 997 subvp_pipe = pipe; 998 } 999 // Use ignore_msa_timing_param and VRR active, or Freesync flag to identify as DRR On 1000 if (found && context->res_ctx.pipe_ctx[vblank_index].stream->ignore_msa_timing_param && 1001 (context->res_ctx.pipe_ctx[vblank_index].stream->allow_freesync || 1002 context->res_ctx.pipe_ctx[vblank_index].stream->vrr_active_variable)) { 1003 // SUBVP + DRR case -- only allowed if run through DRR validation path 1004 schedulable = false; 1005 } else if (found) { 1006 main_timing = &subvp_pipe->stream->timing; 1007 phantom_timing = &subvp_pipe->stream->mall_stream_config.paired_stream->timing; 1008 vblank_timing = &context->res_ctx.pipe_ctx[vblank_index].stream->timing; 1009 // Prefetch time is equal to VACTIVE + BP + VSYNC of the phantom pipe 1010 // Also include the prefetch end to mallstart delay time 1011 prefetch_us = (phantom_timing->v_total - phantom_timing->v_front_porch) * phantom_timing->h_total / 1012 (double)(phantom_timing->pix_clk_100hz * 100) * 1000000 + 1013 dc->caps.subvp_prefetch_end_to_mall_start_us; 1014 // P-State allow width and FW delays already included phantom_timing->v_addressable 1015 mall_region_us = phantom_timing->v_addressable * phantom_timing->h_total / 1016 (double)(phantom_timing->pix_clk_100hz * 100) * 1000000; 1017 vblank_frame_us = vblank_timing->v_total * vblank_timing->h_total / 1018 (double)(vblank_timing->pix_clk_100hz * 100) * 1000000; 1019 vblank_blank_us = (vblank_timing->v_total - vblank_timing->v_addressable) * vblank_timing->h_total / 1020 (double)(vblank_timing->pix_clk_100hz * 100) * 1000000; 1021 subvp_active_us = main_timing->v_addressable * main_timing->h_total / 1022 (double)(main_timing->pix_clk_100hz * 100) * 1000000; 1023 max_vblank_mallregion = vblank_blank_us > mall_region_us ? vblank_blank_us : mall_region_us; 1024 1025 // Schedulable if VACTIVE region of the SubVP pipe can fit the MALL prefetch, VBLANK frame time, 1026 // and the max of (VBLANK blanking time, MALL region) 1027 // TODO: Possibly add some margin (i.e. the below conditions should be [...] > X instead of [...] > 0) 1028 if (subvp_active_us - prefetch_us - vblank_frame_us - max_vblank_mallregion > 0) 1029 schedulable = true; 1030 } 1031 return schedulable; 1032 } 1033 1034 /** 1035 * subvp_validate_static_schedulability - Check which SubVP case is calculated 1036 * and handle static analysis based on the case. 1037 * @dc: current dc state 1038 * @context: new dc state 1039 * @vlevel: Voltage level calculated by DML 1040 * 1041 * Three cases: 1042 * 1. SubVP + SubVP 1043 * 2. SubVP + VBLANK (DRR checked internally) 1044 * 3. SubVP + VACTIVE (currently unsupported) 1045 * 1046 * Return: True if statically schedulable, false otherwise 1047 */ 1048 static bool subvp_validate_static_schedulability(struct dc *dc, 1049 struct dc_state *context, 1050 int vlevel) 1051 { 1052 bool schedulable = true; // true by default for single display case 1053 struct vba_vars_st *vba = &context->bw_ctx.dml.vba; 1054 uint32_t i, pipe_idx; 1055 uint8_t subvp_count = 0; 1056 uint8_t vactive_count = 0; 1057 1058 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) { 1059 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i]; 1060 1061 if (!pipe->stream) 1062 continue; 1063 1064 if (pipe->plane_state && !pipe->top_pipe && 1065 pipe->stream->mall_stream_config.type == SUBVP_MAIN) 1066 subvp_count++; 1067 1068 // Count how many planes that aren't SubVP/phantom are capable of VACTIVE 1069 // switching (SubVP + VACTIVE unsupported). In situations where we force 1070 // SubVP for a VACTIVE plane, we don't want to increment the vactive_count. 1071 if (vba->ActiveDRAMClockChangeLatencyMargin[vba->pipe_plane[pipe_idx]] > 0 && 1072 pipe->stream->mall_stream_config.type == SUBVP_NONE) { 1073 vactive_count++; 1074 } 1075 pipe_idx++; 1076 } 1077 1078 if (subvp_count == 2) { 1079 // Static schedulability check for SubVP + SubVP case 1080 schedulable = subvp_subvp_schedulable(dc, context); 1081 } else if (vba->DRAMClockChangeSupport[vlevel][vba->maxMpcComb] == dm_dram_clock_change_vblank_w_mall_sub_vp) { 1082 // Static schedulability check for SubVP + VBLANK case. Also handle the case where 1083 // DML outputs SubVP + VBLANK + VACTIVE (DML will report as SubVP + VBLANK) 1084 if (vactive_count > 0) 1085 schedulable = false; 1086 else 1087 schedulable = subvp_vblank_schedulable(dc, context); 1088 } else if (vba->DRAMClockChangeSupport[vlevel][vba->maxMpcComb] == dm_dram_clock_change_vactive_w_mall_sub_vp && 1089 vactive_count > 0) { 1090 // For single display SubVP cases, DML will output dm_dram_clock_change_vactive_w_mall_sub_vp by default. 1091 // We tell the difference between SubVP vs. SubVP + VACTIVE by checking the vactive_count. 1092 // SubVP + VACTIVE currently unsupported 1093 schedulable = false; 1094 } 1095 return schedulable; 1096 } 1097 1098 static void dcn32_full_validate_bw_helper(struct dc *dc, 1099 struct dc_state *context, 1100 display_e2e_pipe_params_st *pipes, 1101 int *vlevel, 1102 int *split, 1103 bool *merge, 1104 int *pipe_cnt) 1105 { 1106 struct vba_vars_st *vba = &context->bw_ctx.dml.vba; 1107 unsigned int dc_pipe_idx = 0; 1108 int i = 0; 1109 bool found_supported_config = false; 1110 struct pipe_ctx *pipe = NULL; 1111 uint32_t non_subvp_pipes = 0; 1112 bool drr_pipe_found = false; 1113 uint32_t drr_pipe_index = 0; 1114 1115 dc_assert_fp_enabled(); 1116 1117 /* 1118 * DML favors voltage over p-state, but we're more interested in 1119 * supporting p-state over voltage. We can't support p-state in 1120 * prefetch mode > 0 so try capping the prefetch mode to start. 1121 * Override present for testing. 1122 */ 1123 if (dc->debug.dml_disallow_alternate_prefetch_modes) 1124 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final = 1125 dm_prefetch_support_uclk_fclk_and_stutter; 1126 else 1127 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final = 1128 dm_prefetch_support_uclk_fclk_and_stutter_if_possible; 1129 1130 *vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, *pipe_cnt); 1131 /* This may adjust vlevel and maxMpcComb */ 1132 if (*vlevel < context->bw_ctx.dml.soc.num_states) { 1133 *vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, *vlevel, split, merge); 1134 vba->VoltageLevel = *vlevel; 1135 } 1136 1137 /* Conditions for setting up phantom pipes for SubVP: 1138 * 1. Not force disable SubVP 1139 * 2. Full update (i.e. !fast_validate) 1140 * 3. Enough pipes are available to support SubVP (TODO: Which pipes will use VACTIVE / VBLANK / SUBVP?) 1141 * 4. Display configuration passes validation 1142 * 5. (Config doesn't support MCLK in VACTIVE/VBLANK || dc->debug.force_subvp_mclk_switch) 1143 */ 1144 if (!dc->debug.force_disable_subvp && !dc->caps.dmub_caps.gecc_enable && dcn32_all_pipes_have_stream_and_plane(dc, context) && 1145 !dcn32_mpo_in_use(context) && !dcn32_any_surfaces_rotated(dc, context) && 1146 (*vlevel == context->bw_ctx.dml.soc.num_states || 1147 vba->DRAMClockChangeSupport[*vlevel][vba->maxMpcComb] == dm_dram_clock_change_unsupported || 1148 dc->debug.force_subvp_mclk_switch)) { 1149 1150 dcn32_merge_pipes_for_subvp(dc, context); 1151 memset(merge, 0, MAX_PIPES * sizeof(bool)); 1152 1153 /* to re-initialize viewport after the pipe merge */ 1154 for (i = 0; i < dc->res_pool->pipe_count; i++) { 1155 struct pipe_ctx *pipe_ctx = &context->res_ctx.pipe_ctx[i]; 1156 1157 if (!pipe_ctx->plane_state || !pipe_ctx->stream) 1158 continue; 1159 1160 resource_build_scaling_params(pipe_ctx); 1161 } 1162 1163 while (!found_supported_config && dcn32_enough_pipes_for_subvp(dc, context) && 1164 dcn32_assign_subvp_pipe(dc, context, &dc_pipe_idx)) { 1165 /* For the case where *vlevel = num_states, bandwidth validation has failed for this config. 1166 * Adding phantom pipes won't change the validation result, so change the DML input param 1167 * for P-State support before adding phantom pipes and recalculating the DML result. 1168 * However, this case is only applicable for SubVP + DRR cases because the prefetch mode 1169 * will not allow for switch in VBLANK. The DRR display must have it's VBLANK stretched 1170 * enough to support MCLK switching. 1171 */ 1172 if (*vlevel == context->bw_ctx.dml.soc.num_states && 1173 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final == 1174 dm_prefetch_support_uclk_fclk_and_stutter) { 1175 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final = 1176 dm_prefetch_support_fclk_and_stutter; 1177 /* There are params (such as FabricClock) that need to be recalculated 1178 * after validation fails (otherwise it will be 0). Calculation for 1179 * phantom vactive requires call into DML, so we must ensure all the 1180 * vba params are valid otherwise we'll get incorrect phantom vactive. 1181 */ 1182 *vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, *pipe_cnt); 1183 } 1184 1185 dc->res_pool->funcs->add_phantom_pipes(dc, context, pipes, *pipe_cnt, dc_pipe_idx); 1186 1187 *pipe_cnt = dc->res_pool->funcs->populate_dml_pipes(dc, context, pipes, false); 1188 // Populate dppclk to trigger a recalculate in dml_get_voltage_level 1189 // so the phantom pipe DLG params can be assigned correctly. 1190 pipes[0].clks_cfg.dppclk_mhz = get_dppclk_calculated(&context->bw_ctx.dml, pipes, *pipe_cnt, 0); 1191 *vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, *pipe_cnt); 1192 1193 /* Check that vlevel requested supports pstate or not 1194 * if not, select the lowest vlevel that supports it 1195 */ 1196 for (i = *vlevel; i < context->bw_ctx.dml.soc.num_states; i++) { 1197 if (vba->DRAMClockChangeSupport[i][vba->maxMpcComb] != dm_dram_clock_change_unsupported) { 1198 *vlevel = i; 1199 break; 1200 } 1201 } 1202 1203 if (*vlevel < context->bw_ctx.dml.soc.num_states && 1204 vba->DRAMClockChangeSupport[*vlevel][vba->maxMpcComb] != dm_dram_clock_change_unsupported 1205 && subvp_validate_static_schedulability(dc, context, *vlevel)) { 1206 found_supported_config = true; 1207 } else if (*vlevel < context->bw_ctx.dml.soc.num_states) { 1208 /* Case where 1 SubVP is added, and DML reports MCLK unsupported or DRR is allowed. 1209 * This handles the case for SubVP + DRR, where the DRR display does not support MCLK 1210 * switch at it's native refresh rate / timing, or DRR is allowed for the non-subvp 1211 * display. 1212 */ 1213 for (i = 0; i < dc->res_pool->pipe_count; i++) { 1214 pipe = &context->res_ctx.pipe_ctx[i]; 1215 if (pipe->stream && pipe->plane_state && !pipe->top_pipe && 1216 pipe->stream->mall_stream_config.type == SUBVP_NONE) { 1217 non_subvp_pipes++; 1218 // Use ignore_msa_timing_param flag to identify as DRR 1219 if (pipe->stream->ignore_msa_timing_param && pipe->stream->allow_freesync) { 1220 drr_pipe_found = true; 1221 drr_pipe_index = i; 1222 } 1223 } 1224 } 1225 // If there is only 1 remaining non SubVP pipe that is DRR, check static 1226 // schedulability for SubVP + DRR. 1227 if (non_subvp_pipes == 1 && drr_pipe_found) { 1228 /* find lowest vlevel that supports the config */ 1229 for (i = *vlevel; i >= 0; i--) { 1230 if (vba->ModeSupport[i][vba->maxMpcComb]) { 1231 *vlevel = i; 1232 } else { 1233 break; 1234 } 1235 } 1236 1237 found_supported_config = subvp_drr_schedulable(dc, context, 1238 &context->res_ctx.pipe_ctx[drr_pipe_index]); 1239 } 1240 } 1241 } 1242 1243 // If SubVP pipe config is unsupported (or cannot be used for UCLK switching) 1244 // remove phantom pipes and repopulate dml pipes 1245 if (!found_supported_config) { 1246 dc->res_pool->funcs->remove_phantom_pipes(dc, context, false); 1247 vba->DRAMClockChangeSupport[*vlevel][vba->maxMpcComb] = dm_dram_clock_change_unsupported; 1248 *pipe_cnt = dc->res_pool->funcs->populate_dml_pipes(dc, context, pipes, false); 1249 1250 *vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, *pipe_cnt); 1251 /* This may adjust vlevel and maxMpcComb */ 1252 if (*vlevel < context->bw_ctx.dml.soc.num_states) { 1253 *vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, *vlevel, split, merge); 1254 vba->VoltageLevel = *vlevel; 1255 } 1256 } else { 1257 // Most populate phantom DLG params before programming hardware / timing for phantom pipe 1258 dcn32_helper_populate_phantom_dlg_params(dc, context, pipes, *pipe_cnt); 1259 1260 /* Call validate_apply_pipe_split flags after calling DML getters for 1261 * phantom dlg params, or some of the VBA params indicating pipe split 1262 * can be overwritten by the getters. 1263 * 1264 * When setting up SubVP config, all pipes are merged before attempting to 1265 * add phantom pipes. If pipe split (ODM / MPC) is required, both the main 1266 * and phantom pipes will be split in the regular pipe splitting sequence. 1267 */ 1268 memset(split, 0, MAX_PIPES * sizeof(int)); 1269 memset(merge, 0, MAX_PIPES * sizeof(bool)); 1270 *vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, *vlevel, split, merge); 1271 vba->VoltageLevel = *vlevel; 1272 // Note: We can't apply the phantom pipes to hardware at this time. We have to wait 1273 // until driver has acquired the DMCUB lock to do it safely. 1274 } 1275 } 1276 } 1277 1278 static bool is_dtbclk_required(struct dc *dc, struct dc_state *context) 1279 { 1280 int i; 1281 1282 for (i = 0; i < dc->res_pool->pipe_count; i++) { 1283 if (!context->res_ctx.pipe_ctx[i].stream) 1284 continue; 1285 if (dc->link_srv->dp_is_128b_132b_signal(&context->res_ctx.pipe_ctx[i])) 1286 return true; 1287 } 1288 return false; 1289 } 1290 1291 static void dcn20_adjust_freesync_v_startup(const struct dc_crtc_timing *dc_crtc_timing, int *vstartup_start) 1292 { 1293 struct dc_crtc_timing patched_crtc_timing; 1294 uint32_t asic_blank_end = 0; 1295 uint32_t asic_blank_start = 0; 1296 uint32_t newVstartup = 0; 1297 1298 patched_crtc_timing = *dc_crtc_timing; 1299 1300 if (patched_crtc_timing.flags.INTERLACE == 1) { 1301 if (patched_crtc_timing.v_front_porch < 2) 1302 patched_crtc_timing.v_front_porch = 2; 1303 } else { 1304 if (patched_crtc_timing.v_front_porch < 1) 1305 patched_crtc_timing.v_front_porch = 1; 1306 } 1307 1308 /* blank_start = frame end - front porch */ 1309 asic_blank_start = patched_crtc_timing.v_total - 1310 patched_crtc_timing.v_front_porch; 1311 1312 /* blank_end = blank_start - active */ 1313 asic_blank_end = asic_blank_start - 1314 patched_crtc_timing.v_border_bottom - 1315 patched_crtc_timing.v_addressable - 1316 patched_crtc_timing.v_border_top; 1317 1318 newVstartup = asic_blank_end + (patched_crtc_timing.v_total - asic_blank_start); 1319 1320 *vstartup_start = ((newVstartup > *vstartup_start) ? newVstartup : *vstartup_start); 1321 } 1322 1323 static void dcn32_calculate_dlg_params(struct dc *dc, struct dc_state *context, 1324 display_e2e_pipe_params_st *pipes, 1325 int pipe_cnt, int vlevel) 1326 { 1327 int i, pipe_idx, active_hubp_count = 0; 1328 bool usr_retraining_support = false; 1329 bool unbounded_req_enabled = false; 1330 1331 dc_assert_fp_enabled(); 1332 1333 /* Writeback MCIF_WB arbitration parameters */ 1334 dc->res_pool->funcs->set_mcif_arb_params(dc, context, pipes, pipe_cnt); 1335 1336 context->bw_ctx.bw.dcn.clk.dispclk_khz = context->bw_ctx.dml.vba.DISPCLK * 1000; 1337 context->bw_ctx.bw.dcn.clk.dcfclk_khz = context->bw_ctx.dml.vba.DCFCLK * 1000; 1338 context->bw_ctx.bw.dcn.clk.socclk_khz = context->bw_ctx.dml.vba.SOCCLK * 1000; 1339 context->bw_ctx.bw.dcn.clk.dramclk_khz = context->bw_ctx.dml.vba.DRAMSpeed * 1000 / 16; 1340 context->bw_ctx.bw.dcn.clk.dcfclk_deep_sleep_khz = context->bw_ctx.dml.vba.DCFCLKDeepSleep * 1000; 1341 context->bw_ctx.bw.dcn.clk.fclk_khz = context->bw_ctx.dml.vba.FabricClock * 1000; 1342 context->bw_ctx.bw.dcn.clk.p_state_change_support = 1343 context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] 1344 != dm_dram_clock_change_unsupported; 1345 1346 /* Pstate change might not be supported by hardware, but it might be 1347 * possible with firmware driven vertical blank stretching. 1348 */ 1349 context->bw_ctx.bw.dcn.clk.p_state_change_support |= context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching; 1350 1351 context->bw_ctx.bw.dcn.clk.dppclk_khz = 0; 1352 context->bw_ctx.bw.dcn.clk.dtbclk_en = is_dtbclk_required(dc, context); 1353 context->bw_ctx.bw.dcn.clk.ref_dtbclk_khz = context->bw_ctx.dml.vba.DTBCLKPerState[vlevel] * 1000; 1354 if (context->bw_ctx.dml.vba.FCLKChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] == dm_fclock_change_unsupported) 1355 context->bw_ctx.bw.dcn.clk.fclk_p_state_change_support = false; 1356 else 1357 context->bw_ctx.bw.dcn.clk.fclk_p_state_change_support = true; 1358 1359 usr_retraining_support = context->bw_ctx.dml.vba.USRRetrainingSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb]; 1360 ASSERT(usr_retraining_support); 1361 1362 if (context->bw_ctx.bw.dcn.clk.dispclk_khz < dc->debug.min_disp_clk_khz) 1363 context->bw_ctx.bw.dcn.clk.dispclk_khz = dc->debug.min_disp_clk_khz; 1364 1365 unbounded_req_enabled = get_unbounded_request_enabled(&context->bw_ctx.dml, pipes, pipe_cnt); 1366 1367 if (unbounded_req_enabled && pipe_cnt > 1) { 1368 // Unbounded requesting should not ever be used when more than 1 pipe is enabled. 1369 ASSERT(false); 1370 unbounded_req_enabled = false; 1371 } 1372 1373 context->bw_ctx.bw.dcn.mall_ss_size_bytes = 0; 1374 context->bw_ctx.bw.dcn.mall_ss_psr_active_size_bytes = 0; 1375 context->bw_ctx.bw.dcn.mall_subvp_size_bytes = 0; 1376 1377 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) { 1378 if (!context->res_ctx.pipe_ctx[i].stream) 1379 continue; 1380 if (context->res_ctx.pipe_ctx[i].plane_state) 1381 active_hubp_count++; 1382 pipes[pipe_idx].pipe.dest.vstartup_start = get_vstartup(&context->bw_ctx.dml, pipes, pipe_cnt, 1383 pipe_idx); 1384 pipes[pipe_idx].pipe.dest.vupdate_offset = get_vupdate_offset(&context->bw_ctx.dml, pipes, pipe_cnt, 1385 pipe_idx); 1386 pipes[pipe_idx].pipe.dest.vupdate_width = get_vupdate_width(&context->bw_ctx.dml, pipes, pipe_cnt, 1387 pipe_idx); 1388 pipes[pipe_idx].pipe.dest.vready_offset = get_vready_offset(&context->bw_ctx.dml, pipes, pipe_cnt, 1389 pipe_idx); 1390 1391 if (context->res_ctx.pipe_ctx[i].stream->mall_stream_config.type == SUBVP_PHANTOM) { 1392 // Phantom pipe requires that DET_SIZE = 0 and no unbounded requests 1393 context->res_ctx.pipe_ctx[i].det_buffer_size_kb = 0; 1394 context->res_ctx.pipe_ctx[i].unbounded_req = false; 1395 } else { 1396 context->res_ctx.pipe_ctx[i].det_buffer_size_kb = get_det_buffer_size_kbytes(&context->bw_ctx.dml, pipes, pipe_cnt, 1397 pipe_idx); 1398 context->res_ctx.pipe_ctx[i].unbounded_req = unbounded_req_enabled; 1399 } 1400 1401 if (context->bw_ctx.bw.dcn.clk.dppclk_khz < pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000) 1402 context->bw_ctx.bw.dcn.clk.dppclk_khz = pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000; 1403 if (context->res_ctx.pipe_ctx[i].plane_state) 1404 context->res_ctx.pipe_ctx[i].plane_res.bw.dppclk_khz = pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000; 1405 else 1406 context->res_ctx.pipe_ctx[i].plane_res.bw.dppclk_khz = 0; 1407 context->res_ctx.pipe_ctx[i].pipe_dlg_param = pipes[pipe_idx].pipe.dest; 1408 1409 context->res_ctx.pipe_ctx[i].surface_size_in_mall_bytes = get_surface_size_in_mall(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx); 1410 1411 /* MALL Allocation Sizes */ 1412 /* count from active, top pipes per plane only */ 1413 if (context->res_ctx.pipe_ctx[i].stream && context->res_ctx.pipe_ctx[i].plane_state && 1414 (context->res_ctx.pipe_ctx[i].top_pipe == NULL || 1415 context->res_ctx.pipe_ctx[i].plane_state != context->res_ctx.pipe_ctx[i].top_pipe->plane_state) && 1416 context->res_ctx.pipe_ctx[i].prev_odm_pipe == NULL) { 1417 /* SS: all active surfaces stored in MALL */ 1418 if (context->res_ctx.pipe_ctx[i].stream->mall_stream_config.type != SUBVP_PHANTOM) { 1419 context->bw_ctx.bw.dcn.mall_ss_size_bytes += context->res_ctx.pipe_ctx[i].surface_size_in_mall_bytes; 1420 1421 if (context->res_ctx.pipe_ctx[i].stream->link->psr_settings.psr_version == DC_PSR_VERSION_UNSUPPORTED) { 1422 /* SS PSR On: all active surfaces part of streams not supporting PSR stored in MALL */ 1423 context->bw_ctx.bw.dcn.mall_ss_psr_active_size_bytes += context->res_ctx.pipe_ctx[i].surface_size_in_mall_bytes; 1424 } 1425 } else { 1426 /* SUBVP: phantom surfaces only stored in MALL */ 1427 context->bw_ctx.bw.dcn.mall_subvp_size_bytes += context->res_ctx.pipe_ctx[i].surface_size_in_mall_bytes; 1428 } 1429 } 1430 1431 if (context->res_ctx.pipe_ctx[i].stream->adaptive_sync_infopacket.valid) 1432 dcn20_adjust_freesync_v_startup( 1433 &context->res_ctx.pipe_ctx[i].stream->timing, 1434 &context->res_ctx.pipe_ctx[i].pipe_dlg_param.vstartup_start); 1435 1436 pipe_idx++; 1437 } 1438 /* If DCN isn't making memory requests we can allow pstate change and lower clocks */ 1439 if (!active_hubp_count) { 1440 context->bw_ctx.bw.dcn.clk.socclk_khz = 0; 1441 context->bw_ctx.bw.dcn.clk.dppclk_khz = 0; 1442 context->bw_ctx.bw.dcn.clk.dcfclk_khz = 0; 1443 context->bw_ctx.bw.dcn.clk.dcfclk_deep_sleep_khz = 0; 1444 context->bw_ctx.bw.dcn.clk.dramclk_khz = 0; 1445 context->bw_ctx.bw.dcn.clk.fclk_khz = 0; 1446 context->bw_ctx.bw.dcn.clk.p_state_change_support = true; 1447 context->bw_ctx.bw.dcn.clk.fclk_p_state_change_support = true; 1448 } 1449 /*save a original dppclock copy*/ 1450 context->bw_ctx.bw.dcn.clk.bw_dppclk_khz = context->bw_ctx.bw.dcn.clk.dppclk_khz; 1451 context->bw_ctx.bw.dcn.clk.bw_dispclk_khz = context->bw_ctx.bw.dcn.clk.dispclk_khz; 1452 context->bw_ctx.bw.dcn.clk.max_supported_dppclk_khz = context->bw_ctx.dml.soc.clock_limits[vlevel].dppclk_mhz 1453 * 1000; 1454 context->bw_ctx.bw.dcn.clk.max_supported_dispclk_khz = context->bw_ctx.dml.soc.clock_limits[vlevel].dispclk_mhz 1455 * 1000; 1456 1457 context->bw_ctx.bw.dcn.clk.num_ways = dcn32_helper_calculate_num_ways_for_subvp(dc, context); 1458 1459 context->bw_ctx.bw.dcn.compbuf_size_kb = context->bw_ctx.dml.ip.config_return_buffer_size_in_kbytes; 1460 1461 for (i = 0; i < dc->res_pool->pipe_count; i++) { 1462 if (context->res_ctx.pipe_ctx[i].stream) 1463 context->bw_ctx.bw.dcn.compbuf_size_kb -= context->res_ctx.pipe_ctx[i].det_buffer_size_kb; 1464 } 1465 1466 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) { 1467 1468 if (!context->res_ctx.pipe_ctx[i].stream) 1469 continue; 1470 1471 context->bw_ctx.dml.funcs.rq_dlg_get_dlg_reg_v2(&context->bw_ctx.dml, 1472 &context->res_ctx.pipe_ctx[i].dlg_regs, &context->res_ctx.pipe_ctx[i].ttu_regs, pipes, 1473 pipe_cnt, pipe_idx); 1474 1475 context->bw_ctx.dml.funcs.rq_dlg_get_rq_reg_v2(&context->res_ctx.pipe_ctx[i].rq_regs, 1476 &context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx); 1477 pipe_idx++; 1478 } 1479 } 1480 1481 static struct pipe_ctx *dcn32_find_split_pipe( 1482 struct dc *dc, 1483 struct dc_state *context, 1484 int old_index) 1485 { 1486 struct pipe_ctx *pipe = NULL; 1487 int i; 1488 1489 if (old_index >= 0 && context->res_ctx.pipe_ctx[old_index].stream == NULL) { 1490 pipe = &context->res_ctx.pipe_ctx[old_index]; 1491 pipe->pipe_idx = old_index; 1492 } 1493 1494 if (!pipe) 1495 for (i = dc->res_pool->pipe_count - 1; i >= 0; i--) { 1496 if (dc->current_state->res_ctx.pipe_ctx[i].top_pipe == NULL 1497 && dc->current_state->res_ctx.pipe_ctx[i].prev_odm_pipe == NULL) { 1498 if (context->res_ctx.pipe_ctx[i].stream == NULL) { 1499 pipe = &context->res_ctx.pipe_ctx[i]; 1500 pipe->pipe_idx = i; 1501 break; 1502 } 1503 } 1504 } 1505 1506 /* 1507 * May need to fix pipes getting tossed from 1 opp to another on flip 1508 * Add for debugging transient underflow during topology updates: 1509 * ASSERT(pipe); 1510 */ 1511 if (!pipe) 1512 for (i = dc->res_pool->pipe_count - 1; i >= 0; i--) { 1513 if (context->res_ctx.pipe_ctx[i].stream == NULL) { 1514 pipe = &context->res_ctx.pipe_ctx[i]; 1515 pipe->pipe_idx = i; 1516 break; 1517 } 1518 } 1519 1520 return pipe; 1521 } 1522 1523 static bool dcn32_split_stream_for_mpc_or_odm( 1524 const struct dc *dc, 1525 struct resource_context *res_ctx, 1526 struct pipe_ctx *pri_pipe, 1527 struct pipe_ctx *sec_pipe, 1528 bool odm) 1529 { 1530 int pipe_idx = sec_pipe->pipe_idx; 1531 const struct resource_pool *pool = dc->res_pool; 1532 1533 DC_LOGGER_INIT(dc->ctx->logger); 1534 1535 if (odm && pri_pipe->plane_state) { 1536 /* ODM + window MPO, where MPO window is on left half only */ 1537 if (pri_pipe->plane_state->clip_rect.x + pri_pipe->plane_state->clip_rect.width <= 1538 pri_pipe->stream->src.x + pri_pipe->stream->src.width/2) { 1539 1540 DC_LOG_SCALER("%s - ODM + window MPO(left). pri_pipe:%d\n", 1541 __func__, 1542 pri_pipe->pipe_idx); 1543 return true; 1544 } 1545 1546 /* ODM + window MPO, where MPO window is on right half only */ 1547 if (pri_pipe->plane_state->clip_rect.x >= pri_pipe->stream->src.x + pri_pipe->stream->src.width/2) { 1548 1549 DC_LOG_SCALER("%s - ODM + window MPO(right). pri_pipe:%d\n", 1550 __func__, 1551 pri_pipe->pipe_idx); 1552 return true; 1553 } 1554 } 1555 1556 *sec_pipe = *pri_pipe; 1557 1558 sec_pipe->pipe_idx = pipe_idx; 1559 sec_pipe->plane_res.mi = pool->mis[pipe_idx]; 1560 sec_pipe->plane_res.hubp = pool->hubps[pipe_idx]; 1561 sec_pipe->plane_res.ipp = pool->ipps[pipe_idx]; 1562 sec_pipe->plane_res.xfm = pool->transforms[pipe_idx]; 1563 sec_pipe->plane_res.dpp = pool->dpps[pipe_idx]; 1564 sec_pipe->plane_res.mpcc_inst = pool->dpps[pipe_idx]->inst; 1565 sec_pipe->stream_res.dsc = NULL; 1566 if (odm) { 1567 if (pri_pipe->next_odm_pipe) { 1568 ASSERT(pri_pipe->next_odm_pipe != sec_pipe); 1569 sec_pipe->next_odm_pipe = pri_pipe->next_odm_pipe; 1570 sec_pipe->next_odm_pipe->prev_odm_pipe = sec_pipe; 1571 } 1572 if (pri_pipe->top_pipe && pri_pipe->top_pipe->next_odm_pipe) { 1573 pri_pipe->top_pipe->next_odm_pipe->bottom_pipe = sec_pipe; 1574 sec_pipe->top_pipe = pri_pipe->top_pipe->next_odm_pipe; 1575 } 1576 if (pri_pipe->bottom_pipe && pri_pipe->bottom_pipe->next_odm_pipe) { 1577 pri_pipe->bottom_pipe->next_odm_pipe->top_pipe = sec_pipe; 1578 sec_pipe->bottom_pipe = pri_pipe->bottom_pipe->next_odm_pipe; 1579 } 1580 pri_pipe->next_odm_pipe = sec_pipe; 1581 sec_pipe->prev_odm_pipe = pri_pipe; 1582 ASSERT(sec_pipe->top_pipe == NULL); 1583 1584 if (!sec_pipe->top_pipe) 1585 sec_pipe->stream_res.opp = pool->opps[pipe_idx]; 1586 else 1587 sec_pipe->stream_res.opp = sec_pipe->top_pipe->stream_res.opp; 1588 if (sec_pipe->stream->timing.flags.DSC == 1) { 1589 dcn20_acquire_dsc(dc, res_ctx, &sec_pipe->stream_res.dsc, pipe_idx); 1590 ASSERT(sec_pipe->stream_res.dsc); 1591 if (sec_pipe->stream_res.dsc == NULL) 1592 return false; 1593 } 1594 } else { 1595 if (pri_pipe->bottom_pipe) { 1596 ASSERT(pri_pipe->bottom_pipe != sec_pipe); 1597 sec_pipe->bottom_pipe = pri_pipe->bottom_pipe; 1598 sec_pipe->bottom_pipe->top_pipe = sec_pipe; 1599 } 1600 pri_pipe->bottom_pipe = sec_pipe; 1601 sec_pipe->top_pipe = pri_pipe; 1602 1603 ASSERT(pri_pipe->plane_state); 1604 } 1605 1606 return true; 1607 } 1608 1609 bool dcn32_internal_validate_bw(struct dc *dc, 1610 struct dc_state *context, 1611 display_e2e_pipe_params_st *pipes, 1612 int *pipe_cnt_out, 1613 int *vlevel_out, 1614 bool fast_validate) 1615 { 1616 bool out = false; 1617 bool repopulate_pipes = false; 1618 int split[MAX_PIPES] = { 0 }; 1619 bool merge[MAX_PIPES] = { false }; 1620 bool newly_split[MAX_PIPES] = { false }; 1621 int pipe_cnt, i, pipe_idx; 1622 int vlevel = context->bw_ctx.dml.soc.num_states; 1623 struct vba_vars_st *vba = &context->bw_ctx.dml.vba; 1624 1625 dc_assert_fp_enabled(); 1626 1627 ASSERT(pipes); 1628 if (!pipes) 1629 return false; 1630 1631 // For each full update, remove all existing phantom pipes first 1632 dc->res_pool->funcs->remove_phantom_pipes(dc, context, fast_validate); 1633 1634 dc->res_pool->funcs->update_soc_for_wm_a(dc, context); 1635 1636 pipe_cnt = dc->res_pool->funcs->populate_dml_pipes(dc, context, pipes, fast_validate); 1637 1638 if (!pipe_cnt) { 1639 out = true; 1640 goto validate_out; 1641 } 1642 1643 dml_log_pipe_params(&context->bw_ctx.dml, pipes, pipe_cnt); 1644 context->bw_ctx.dml.soc.max_vratio_pre = dcn32_determine_max_vratio_prefetch(dc, context); 1645 1646 if (!fast_validate) 1647 dcn32_full_validate_bw_helper(dc, context, pipes, &vlevel, split, merge, &pipe_cnt); 1648 1649 if (fast_validate || 1650 (dc->debug.dml_disallow_alternate_prefetch_modes && 1651 (vlevel == context->bw_ctx.dml.soc.num_states || 1652 vba->DRAMClockChangeSupport[vlevel][vba->maxMpcComb] == dm_dram_clock_change_unsupported))) { 1653 /* 1654 * If dml_disallow_alternate_prefetch_modes is false, then we have already 1655 * tried alternate prefetch modes during full validation. 1656 * 1657 * If mode is unsupported or there is no p-state support, then 1658 * fall back to favouring voltage. 1659 * 1660 * If Prefetch mode 0 failed for this config, or passed with Max UCLK, then try 1661 * to support with Prefetch mode 1 (dm_prefetch_support_fclk_and_stutter == 2) 1662 */ 1663 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final = 1664 dm_prefetch_support_none; 1665 1666 context->bw_ctx.dml.validate_max_state = fast_validate; 1667 vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, pipe_cnt); 1668 1669 context->bw_ctx.dml.validate_max_state = false; 1670 1671 if (vlevel < context->bw_ctx.dml.soc.num_states) { 1672 memset(split, 0, sizeof(split)); 1673 memset(merge, 0, sizeof(merge)); 1674 vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, vlevel, split, merge); 1675 // dcn20_validate_apply_pipe_split_flags can modify voltage level outside of DML 1676 vba->VoltageLevel = vlevel; 1677 } 1678 } 1679 1680 dml_log_mode_support_params(&context->bw_ctx.dml); 1681 1682 if (vlevel == context->bw_ctx.dml.soc.num_states) 1683 goto validate_fail; 1684 1685 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) { 1686 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i]; 1687 struct pipe_ctx *mpo_pipe = pipe->bottom_pipe; 1688 1689 if (!pipe->stream) 1690 continue; 1691 1692 if (vba->ODMCombineEnabled[vba->pipe_plane[pipe_idx]] != dm_odm_combine_mode_disabled 1693 && !dc->config.enable_windowed_mpo_odm 1694 && pipe->plane_state && mpo_pipe 1695 && memcmp(&mpo_pipe->plane_res.scl_data.recout, 1696 &pipe->plane_res.scl_data.recout, 1697 sizeof(struct rect)) != 0) { 1698 ASSERT(mpo_pipe->plane_state != pipe->plane_state); 1699 goto validate_fail; 1700 } 1701 pipe_idx++; 1702 } 1703 1704 /* merge pipes if necessary */ 1705 for (i = 0; i < dc->res_pool->pipe_count; i++) { 1706 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i]; 1707 1708 /*skip pipes that don't need merging*/ 1709 if (!merge[i]) 1710 continue; 1711 1712 /* if ODM merge we ignore mpc tree, mpo pipes will have their own flags */ 1713 if (pipe->prev_odm_pipe) { 1714 /*split off odm pipe*/ 1715 pipe->prev_odm_pipe->next_odm_pipe = pipe->next_odm_pipe; 1716 if (pipe->next_odm_pipe) 1717 pipe->next_odm_pipe->prev_odm_pipe = pipe->prev_odm_pipe; 1718 1719 /*2:1ODM+MPC Split MPO to Single Pipe + MPC Split MPO*/ 1720 if (pipe->bottom_pipe) { 1721 if (pipe->bottom_pipe->prev_odm_pipe || pipe->bottom_pipe->next_odm_pipe) { 1722 /*MPC split rules will handle this case*/ 1723 pipe->bottom_pipe->top_pipe = NULL; 1724 } else { 1725 /* when merging an ODM pipes, the bottom MPC pipe must now point to 1726 * the previous ODM pipe and its associated stream assets 1727 */ 1728 if (pipe->prev_odm_pipe->bottom_pipe) { 1729 /* 3 plane MPO*/ 1730 pipe->bottom_pipe->top_pipe = pipe->prev_odm_pipe->bottom_pipe; 1731 pipe->prev_odm_pipe->bottom_pipe->bottom_pipe = pipe->bottom_pipe; 1732 } else { 1733 /* 2 plane MPO*/ 1734 pipe->bottom_pipe->top_pipe = pipe->prev_odm_pipe; 1735 pipe->prev_odm_pipe->bottom_pipe = pipe->bottom_pipe; 1736 } 1737 1738 memcpy(&pipe->bottom_pipe->stream_res, &pipe->bottom_pipe->top_pipe->stream_res, sizeof(struct stream_resource)); 1739 } 1740 } 1741 1742 if (pipe->top_pipe) { 1743 pipe->top_pipe->bottom_pipe = NULL; 1744 } 1745 1746 pipe->bottom_pipe = NULL; 1747 pipe->next_odm_pipe = NULL; 1748 pipe->plane_state = NULL; 1749 pipe->stream = NULL; 1750 pipe->top_pipe = NULL; 1751 pipe->prev_odm_pipe = NULL; 1752 if (pipe->stream_res.dsc) 1753 dcn20_release_dsc(&context->res_ctx, dc->res_pool, &pipe->stream_res.dsc); 1754 memset(&pipe->plane_res, 0, sizeof(pipe->plane_res)); 1755 memset(&pipe->stream_res, 0, sizeof(pipe->stream_res)); 1756 memset(&pipe->link_res, 0, sizeof(pipe->link_res)); 1757 repopulate_pipes = true; 1758 } else if (pipe->top_pipe && pipe->top_pipe->plane_state == pipe->plane_state) { 1759 struct pipe_ctx *top_pipe = pipe->top_pipe; 1760 struct pipe_ctx *bottom_pipe = pipe->bottom_pipe; 1761 1762 top_pipe->bottom_pipe = bottom_pipe; 1763 if (bottom_pipe) 1764 bottom_pipe->top_pipe = top_pipe; 1765 1766 pipe->top_pipe = NULL; 1767 pipe->bottom_pipe = NULL; 1768 pipe->plane_state = NULL; 1769 pipe->stream = NULL; 1770 memset(&pipe->plane_res, 0, sizeof(pipe->plane_res)); 1771 memset(&pipe->stream_res, 0, sizeof(pipe->stream_res)); 1772 memset(&pipe->link_res, 0, sizeof(pipe->link_res)); 1773 repopulate_pipes = true; 1774 } else 1775 ASSERT(0); /* Should never try to merge master pipe */ 1776 1777 } 1778 1779 for (i = 0, pipe_idx = -1; i < dc->res_pool->pipe_count; i++) { 1780 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i]; 1781 struct pipe_ctx *old_pipe = &dc->current_state->res_ctx.pipe_ctx[i]; 1782 struct pipe_ctx *hsplit_pipe = NULL; 1783 bool odm; 1784 int old_index = -1; 1785 1786 if (!pipe->stream || newly_split[i]) 1787 continue; 1788 1789 pipe_idx++; 1790 odm = vba->ODMCombineEnabled[vba->pipe_plane[pipe_idx]] != dm_odm_combine_mode_disabled; 1791 1792 if (!pipe->plane_state && !odm) 1793 continue; 1794 1795 if (split[i]) { 1796 if (odm) { 1797 if (split[i] == 4 && old_pipe->next_odm_pipe && old_pipe->next_odm_pipe->next_odm_pipe) 1798 old_index = old_pipe->next_odm_pipe->next_odm_pipe->pipe_idx; 1799 else if (old_pipe->next_odm_pipe) 1800 old_index = old_pipe->next_odm_pipe->pipe_idx; 1801 } else { 1802 if (split[i] == 4 && old_pipe->bottom_pipe && old_pipe->bottom_pipe->bottom_pipe && 1803 old_pipe->bottom_pipe->bottom_pipe->plane_state == old_pipe->plane_state) 1804 old_index = old_pipe->bottom_pipe->bottom_pipe->pipe_idx; 1805 else if (old_pipe->bottom_pipe && 1806 old_pipe->bottom_pipe->plane_state == old_pipe->plane_state) 1807 old_index = old_pipe->bottom_pipe->pipe_idx; 1808 } 1809 hsplit_pipe = dcn32_find_split_pipe(dc, context, old_index); 1810 ASSERT(hsplit_pipe); 1811 if (!hsplit_pipe) 1812 goto validate_fail; 1813 1814 if (!dcn32_split_stream_for_mpc_or_odm( 1815 dc, &context->res_ctx, 1816 pipe, hsplit_pipe, odm)) 1817 goto validate_fail; 1818 1819 newly_split[hsplit_pipe->pipe_idx] = true; 1820 repopulate_pipes = true; 1821 } 1822 if (split[i] == 4) { 1823 struct pipe_ctx *pipe_4to1; 1824 1825 if (odm && old_pipe->next_odm_pipe) 1826 old_index = old_pipe->next_odm_pipe->pipe_idx; 1827 else if (!odm && old_pipe->bottom_pipe && 1828 old_pipe->bottom_pipe->plane_state == old_pipe->plane_state) 1829 old_index = old_pipe->bottom_pipe->pipe_idx; 1830 else 1831 old_index = -1; 1832 pipe_4to1 = dcn32_find_split_pipe(dc, context, old_index); 1833 ASSERT(pipe_4to1); 1834 if (!pipe_4to1) 1835 goto validate_fail; 1836 if (!dcn32_split_stream_for_mpc_or_odm( 1837 dc, &context->res_ctx, 1838 pipe, pipe_4to1, odm)) 1839 goto validate_fail; 1840 newly_split[pipe_4to1->pipe_idx] = true; 1841 1842 if (odm && old_pipe->next_odm_pipe && old_pipe->next_odm_pipe->next_odm_pipe 1843 && old_pipe->next_odm_pipe->next_odm_pipe->next_odm_pipe) 1844 old_index = old_pipe->next_odm_pipe->next_odm_pipe->next_odm_pipe->pipe_idx; 1845 else if (!odm && old_pipe->bottom_pipe && old_pipe->bottom_pipe->bottom_pipe && 1846 old_pipe->bottom_pipe->bottom_pipe->bottom_pipe && 1847 old_pipe->bottom_pipe->bottom_pipe->bottom_pipe->plane_state == old_pipe->plane_state) 1848 old_index = old_pipe->bottom_pipe->bottom_pipe->bottom_pipe->pipe_idx; 1849 else 1850 old_index = -1; 1851 pipe_4to1 = dcn32_find_split_pipe(dc, context, old_index); 1852 ASSERT(pipe_4to1); 1853 if (!pipe_4to1) 1854 goto validate_fail; 1855 if (!dcn32_split_stream_for_mpc_or_odm( 1856 dc, &context->res_ctx, 1857 hsplit_pipe, pipe_4to1, odm)) 1858 goto validate_fail; 1859 newly_split[pipe_4to1->pipe_idx] = true; 1860 } 1861 if (odm) 1862 dcn20_build_mapped_resource(dc, context, pipe->stream); 1863 } 1864 1865 for (i = 0; i < dc->res_pool->pipe_count; i++) { 1866 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i]; 1867 1868 if (pipe->plane_state) { 1869 if (!resource_build_scaling_params(pipe)) 1870 goto validate_fail; 1871 } 1872 } 1873 1874 /* Actual dsc count per stream dsc validation*/ 1875 if (!dcn20_validate_dsc(dc, context)) { 1876 vba->ValidationStatus[vba->soc.num_states] = DML_FAIL_DSC_VALIDATION_FAILURE; 1877 goto validate_fail; 1878 } 1879 1880 if (repopulate_pipes) { 1881 int flag_max_mpc_comb = vba->maxMpcComb; 1882 int flag_vlevel = vlevel; 1883 int i; 1884 1885 pipe_cnt = dc->res_pool->funcs->populate_dml_pipes(dc, context, pipes, fast_validate); 1886 1887 /* repopulate_pipes = 1 means the pipes were either split or merged. In this case 1888 * we have to re-calculate the DET allocation and run through DML once more to 1889 * ensure all the params are calculated correctly. We do not need to run the 1890 * pipe split check again after this call (pipes are already split / merged). 1891 * */ 1892 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final = 1893 dm_prefetch_support_uclk_fclk_and_stutter_if_possible; 1894 vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, pipe_cnt); 1895 if (vlevel == context->bw_ctx.dml.soc.num_states) { 1896 /* failed after DET size changes */ 1897 goto validate_fail; 1898 } else if (flag_max_mpc_comb == 0 && 1899 flag_max_mpc_comb != context->bw_ctx.dml.vba.maxMpcComb) { 1900 /* check the context constructed with pipe split flags is still valid*/ 1901 bool flags_valid = false; 1902 for (i = flag_vlevel; i < context->bw_ctx.dml.soc.num_states; i++) { 1903 if (vba->ModeSupport[i][flag_max_mpc_comb]) { 1904 vba->maxMpcComb = flag_max_mpc_comb; 1905 vba->VoltageLevel = i; 1906 vlevel = i; 1907 flags_valid = true; 1908 } 1909 } 1910 1911 /* this should never happen */ 1912 if (!flags_valid) 1913 goto validate_fail; 1914 } 1915 } 1916 *vlevel_out = vlevel; 1917 *pipe_cnt_out = pipe_cnt; 1918 1919 out = true; 1920 goto validate_out; 1921 1922 validate_fail: 1923 out = false; 1924 1925 validate_out: 1926 return out; 1927 } 1928 1929 1930 void dcn32_calculate_wm_and_dlg_fpu(struct dc *dc, struct dc_state *context, 1931 display_e2e_pipe_params_st *pipes, 1932 int pipe_cnt, 1933 int vlevel) 1934 { 1935 int i, pipe_idx, vlevel_temp = 0; 1936 double dcfclk = dcn3_2_soc.clock_limits[0].dcfclk_mhz; 1937 double dcfclk_from_validation = context->bw_ctx.dml.vba.DCFCLKState[vlevel][context->bw_ctx.dml.vba.maxMpcComb]; 1938 double dcfclk_from_fw_based_mclk_switching = dcfclk_from_validation; 1939 bool pstate_en = context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] != 1940 dm_dram_clock_change_unsupported; 1941 unsigned int dummy_latency_index = 0; 1942 int maxMpcComb = context->bw_ctx.dml.vba.maxMpcComb; 1943 unsigned int min_dram_speed_mts = context->bw_ctx.dml.vba.DRAMSpeed; 1944 bool subvp_in_use = dcn32_subvp_in_use(dc, context); 1945 unsigned int min_dram_speed_mts_margin; 1946 bool need_fclk_lat_as_dummy = false; 1947 bool is_subvp_p_drr = false; 1948 struct dc_stream_state *fpo_candidate_stream = NULL; 1949 1950 dc_assert_fp_enabled(); 1951 1952 /* need to find dummy latency index for subvp */ 1953 if (subvp_in_use) { 1954 /* Override DRAMClockChangeSupport for SubVP + DRR case where the DRR cannot switch without stretching it's VBLANK */ 1955 if (!pstate_en) { 1956 context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][maxMpcComb] = dm_dram_clock_change_vblank_w_mall_sub_vp; 1957 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final = dm_prefetch_support_fclk_and_stutter; 1958 pstate_en = true; 1959 is_subvp_p_drr = true; 1960 } 1961 dummy_latency_index = dcn32_find_dummy_latency_index_for_fw_based_mclk_switch(dc, 1962 context, pipes, pipe_cnt, vlevel); 1963 1964 /* For DCN32/321 need to validate with fclk pstate change latency equal to dummy so prefetch is 1965 * scheduled correctly to account for dummy pstate. 1966 */ 1967 if (context->bw_ctx.dml.soc.fclk_change_latency_us < dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us) { 1968 need_fclk_lat_as_dummy = true; 1969 context->bw_ctx.dml.soc.fclk_change_latency_us = 1970 dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us; 1971 } 1972 context->bw_ctx.dml.soc.dram_clock_change_latency_us = 1973 dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us; 1974 dcn32_internal_validate_bw(dc, context, pipes, &pipe_cnt, &vlevel, false); 1975 maxMpcComb = context->bw_ctx.dml.vba.maxMpcComb; 1976 if (is_subvp_p_drr) { 1977 context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][maxMpcComb] = dm_dram_clock_change_vblank_w_mall_sub_vp; 1978 } 1979 } 1980 1981 context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching = false; 1982 for (i = 0; i < context->stream_count; i++) { 1983 if (context->streams[i]) 1984 context->streams[i]->fpo_in_use = false; 1985 } 1986 1987 if (!pstate_en || (!dc->debug.disable_fpo_optimizations && 1988 pstate_en && vlevel != 0)) { 1989 /* only when the mclk switch can not be natural, is the fw based vblank stretch attempted */ 1990 fpo_candidate_stream = dcn32_can_support_mclk_switch_using_fw_based_vblank_stretch(dc, context); 1991 if (fpo_candidate_stream) { 1992 fpo_candidate_stream->fpo_in_use = true; 1993 context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching = true; 1994 } 1995 1996 if (context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching) { 1997 dummy_latency_index = dcn32_find_dummy_latency_index_for_fw_based_mclk_switch(dc, 1998 context, pipes, pipe_cnt, vlevel); 1999 2000 /* After calling dcn30_find_dummy_latency_index_for_fw_based_mclk_switch 2001 * we reinstate the original dram_clock_change_latency_us on the context 2002 * and all variables that may have changed up to this point, except the 2003 * newly found dummy_latency_index 2004 */ 2005 context->bw_ctx.dml.soc.dram_clock_change_latency_us = 2006 dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us; 2007 /* For DCN32/321 need to validate with fclk pstate change latency equal to dummy so 2008 * prefetch is scheduled correctly to account for dummy pstate. 2009 */ 2010 if (context->bw_ctx.dml.soc.fclk_change_latency_us < dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us) { 2011 need_fclk_lat_as_dummy = true; 2012 context->bw_ctx.dml.soc.fclk_change_latency_us = 2013 dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us; 2014 } 2015 dcn32_internal_validate_bw(dc, context, pipes, &pipe_cnt, &vlevel_temp, false); 2016 if (vlevel_temp < vlevel) { 2017 vlevel = vlevel_temp; 2018 maxMpcComb = context->bw_ctx.dml.vba.maxMpcComb; 2019 dcfclk_from_fw_based_mclk_switching = context->bw_ctx.dml.vba.DCFCLKState[vlevel][context->bw_ctx.dml.vba.maxMpcComb]; 2020 pstate_en = true; 2021 } else { 2022 /* Restore FCLK latency and re-run validation to go back to original validation 2023 * output if we find that enabling FPO does not give us any benefit (i.e. lower 2024 * voltage level) 2025 */ 2026 context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching = false; 2027 for (i = 0; i < context->stream_count; i++) { 2028 if (context->streams[i]) 2029 context->streams[i]->fpo_in_use = false; 2030 } 2031 context->bw_ctx.dml.soc.fclk_change_latency_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.fclk_change_latency_us; 2032 dcn32_internal_validate_bw(dc, context, pipes, &pipe_cnt, &vlevel, false); 2033 } 2034 } 2035 } 2036 2037 /* Set B: 2038 * For Set B calculations use clocks from clock_limits[2] when available i.e. when SMU is present, 2039 * otherwise use arbitrary low value from spreadsheet for DCFCLK as lower is safer for watermark 2040 * calculations to cover bootup clocks. 2041 * DCFCLK: soc.clock_limits[2] when available 2042 * UCLK: soc.clock_limits[2] when available 2043 */ 2044 if (dcn3_2_soc.num_states > 2) { 2045 vlevel_temp = 2; 2046 dcfclk = dcn3_2_soc.clock_limits[2].dcfclk_mhz; 2047 } else 2048 dcfclk = 615; //DCFCLK Vmin_lv 2049 2050 pipes[0].clks_cfg.voltage = vlevel_temp; 2051 pipes[0].clks_cfg.dcfclk_mhz = dcfclk; 2052 pipes[0].clks_cfg.socclk_mhz = context->bw_ctx.dml.soc.clock_limits[vlevel_temp].socclk_mhz; 2053 2054 if (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_B].valid) { 2055 context->bw_ctx.dml.soc.dram_clock_change_latency_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_B].dml_input.pstate_latency_us; 2056 context->bw_ctx.dml.soc.fclk_change_latency_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_B].dml_input.fclk_change_latency_us; 2057 context->bw_ctx.dml.soc.sr_enter_plus_exit_time_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_B].dml_input.sr_enter_plus_exit_time_us; 2058 context->bw_ctx.dml.soc.sr_exit_time_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_B].dml_input.sr_exit_time_us; 2059 } 2060 context->bw_ctx.bw.dcn.watermarks.b.urgent_ns = get_wm_urgent(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2061 context->bw_ctx.bw.dcn.watermarks.b.cstate_pstate.cstate_enter_plus_exit_ns = get_wm_stutter_enter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2062 context->bw_ctx.bw.dcn.watermarks.b.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2063 context->bw_ctx.bw.dcn.watermarks.b.cstate_pstate.pstate_change_ns = get_wm_dram_clock_change(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2064 context->bw_ctx.bw.dcn.watermarks.b.pte_meta_urgent_ns = get_wm_memory_trip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2065 context->bw_ctx.bw.dcn.watermarks.b.frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2066 context->bw_ctx.bw.dcn.watermarks.b.frac_urg_bw_flip = get_fraction_of_urgent_bandwidth_imm_flip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2067 context->bw_ctx.bw.dcn.watermarks.b.urgent_latency_ns = get_urgent_latency(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2068 context->bw_ctx.bw.dcn.watermarks.b.cstate_pstate.fclk_pstate_change_ns = get_fclk_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2069 context->bw_ctx.bw.dcn.watermarks.b.usr_retraining_ns = get_usr_retraining_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2070 2071 /* Set D: 2072 * All clocks min. 2073 * DCFCLK: Min, as reported by PM FW when available 2074 * UCLK : Min, as reported by PM FW when available 2075 * sr_enter_exit/sr_exit should be lower than used for DRAM (TBD after bringup or later, use as decided in Clk Mgr) 2076 */ 2077 2078 /* 2079 if (dcn3_2_soc.num_states > 2) { 2080 vlevel_temp = 0; 2081 dcfclk = dc->clk_mgr->bw_params->clk_table.entries[0].dcfclk_mhz; 2082 } else 2083 dcfclk = 615; //DCFCLK Vmin_lv 2084 2085 pipes[0].clks_cfg.voltage = vlevel_temp; 2086 pipes[0].clks_cfg.dcfclk_mhz = dcfclk; 2087 pipes[0].clks_cfg.socclk_mhz = context->bw_ctx.dml.soc.clock_limits[vlevel_temp].socclk_mhz; 2088 2089 if (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_D].valid) { 2090 context->bw_ctx.dml.soc.dram_clock_change_latency_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_D].dml_input.pstate_latency_us; 2091 context->bw_ctx.dml.soc.fclk_change_latency_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_D].dml_input.fclk_change_latency_us; 2092 context->bw_ctx.dml.soc.sr_enter_plus_exit_time_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_D].dml_input.sr_enter_plus_exit_time_us; 2093 context->bw_ctx.dml.soc.sr_exit_time_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_D].dml_input.sr_exit_time_us; 2094 } 2095 context->bw_ctx.bw.dcn.watermarks.d.urgent_ns = get_wm_urgent(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2096 context->bw_ctx.bw.dcn.watermarks.d.cstate_pstate.cstate_enter_plus_exit_ns = get_wm_stutter_enter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2097 context->bw_ctx.bw.dcn.watermarks.d.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2098 context->bw_ctx.bw.dcn.watermarks.d.cstate_pstate.pstate_change_ns = get_wm_dram_clock_change(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2099 context->bw_ctx.bw.dcn.watermarks.d.pte_meta_urgent_ns = get_wm_memory_trip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2100 context->bw_ctx.bw.dcn.watermarks.d.frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2101 context->bw_ctx.bw.dcn.watermarks.d.frac_urg_bw_flip = get_fraction_of_urgent_bandwidth_imm_flip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2102 context->bw_ctx.bw.dcn.watermarks.d.urgent_latency_ns = get_urgent_latency(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2103 context->bw_ctx.bw.dcn.watermarks.d.cstate_pstate.fclk_pstate_change_ns = get_fclk_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2104 context->bw_ctx.bw.dcn.watermarks.d.usr_retraining_ns = get_usr_retraining_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2105 */ 2106 2107 /* Set C, for Dummy P-State: 2108 * All clocks min. 2109 * DCFCLK: Min, as reported by PM FW, when available 2110 * UCLK : Min, as reported by PM FW, when available 2111 * pstate latency as per UCLK state dummy pstate latency 2112 */ 2113 2114 // For Set A and Set C use values from validation 2115 pipes[0].clks_cfg.voltage = vlevel; 2116 pipes[0].clks_cfg.dcfclk_mhz = dcfclk_from_validation; 2117 pipes[0].clks_cfg.socclk_mhz = context->bw_ctx.dml.soc.clock_limits[vlevel].socclk_mhz; 2118 2119 if (context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching) { 2120 pipes[0].clks_cfg.dcfclk_mhz = dcfclk_from_fw_based_mclk_switching; 2121 } 2122 2123 if (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_C].valid) { 2124 min_dram_speed_mts = context->bw_ctx.dml.vba.DRAMSpeed; 2125 min_dram_speed_mts_margin = 160; 2126 2127 context->bw_ctx.dml.soc.dram_clock_change_latency_us = 2128 dc->clk_mgr->bw_params->dummy_pstate_table[0].dummy_pstate_latency_us; 2129 2130 if (context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][maxMpcComb] == 2131 dm_dram_clock_change_unsupported) { 2132 int min_dram_speed_mts_offset = dc->clk_mgr->bw_params->clk_table.num_entries_per_clk.num_memclk_levels - 1; 2133 2134 min_dram_speed_mts = 2135 dc->clk_mgr->bw_params->clk_table.entries[min_dram_speed_mts_offset].memclk_mhz * 16; 2136 } 2137 2138 if (!context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching && !subvp_in_use) { 2139 /* find largest table entry that is lower than dram speed, 2140 * but lower than DPM0 still uses DPM0 2141 */ 2142 for (dummy_latency_index = 3; dummy_latency_index > 0; dummy_latency_index--) 2143 if (min_dram_speed_mts + min_dram_speed_mts_margin > 2144 dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dram_speed_mts) 2145 break; 2146 } 2147 2148 context->bw_ctx.dml.soc.dram_clock_change_latency_us = 2149 dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us; 2150 2151 context->bw_ctx.dml.soc.fclk_change_latency_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_C].dml_input.fclk_change_latency_us; 2152 context->bw_ctx.dml.soc.sr_enter_plus_exit_time_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_C].dml_input.sr_enter_plus_exit_time_us; 2153 context->bw_ctx.dml.soc.sr_exit_time_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_C].dml_input.sr_exit_time_us; 2154 } 2155 2156 context->bw_ctx.bw.dcn.watermarks.c.urgent_ns = get_wm_urgent(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2157 context->bw_ctx.bw.dcn.watermarks.c.cstate_pstate.cstate_enter_plus_exit_ns = get_wm_stutter_enter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2158 context->bw_ctx.bw.dcn.watermarks.c.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2159 context->bw_ctx.bw.dcn.watermarks.c.cstate_pstate.pstate_change_ns = get_wm_dram_clock_change(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2160 context->bw_ctx.bw.dcn.watermarks.c.pte_meta_urgent_ns = get_wm_memory_trip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2161 context->bw_ctx.bw.dcn.watermarks.c.frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2162 context->bw_ctx.bw.dcn.watermarks.c.frac_urg_bw_flip = get_fraction_of_urgent_bandwidth_imm_flip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2163 context->bw_ctx.bw.dcn.watermarks.c.urgent_latency_ns = get_urgent_latency(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2164 /* On DCN32/321, PMFW will set PSTATE_CHANGE_TYPE = 1 (FCLK) for UCLK dummy p-state. 2165 * In this case we must program FCLK WM Set C to use the UCLK dummy p-state WM 2166 * value. 2167 */ 2168 context->bw_ctx.bw.dcn.watermarks.c.cstate_pstate.fclk_pstate_change_ns = get_wm_dram_clock_change(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2169 context->bw_ctx.bw.dcn.watermarks.c.usr_retraining_ns = get_usr_retraining_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2170 2171 if ((!pstate_en) && (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_C].valid)) { 2172 /* The only difference between A and C is p-state latency, if p-state is not supported 2173 * with full p-state latency we want to calculate DLG based on dummy p-state latency, 2174 * Set A p-state watermark set to 0 on DCN30, when p-state unsupported, for now keep as DCN30. 2175 */ 2176 context->bw_ctx.bw.dcn.watermarks.a = context->bw_ctx.bw.dcn.watermarks.c; 2177 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.pstate_change_ns = 0; 2178 /* Calculate FCLK p-state change watermark based on FCLK pstate change latency in case 2179 * UCLK p-state is not supported, to avoid underflow in case FCLK pstate is supported 2180 */ 2181 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.fclk_pstate_change_ns = get_fclk_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2182 } else { 2183 /* Set A: 2184 * All clocks min. 2185 * DCFCLK: Min, as reported by PM FW, when available 2186 * UCLK: Min, as reported by PM FW, when available 2187 */ 2188 2189 /* For set A set the correct latency values (i.e. non-dummy values) unconditionally 2190 */ 2191 context->bw_ctx.dml.soc.dram_clock_change_latency_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us; 2192 context->bw_ctx.dml.soc.sr_enter_plus_exit_time_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.sr_enter_plus_exit_time_us; 2193 context->bw_ctx.dml.soc.sr_exit_time_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.sr_exit_time_us; 2194 2195 context->bw_ctx.bw.dcn.watermarks.a.urgent_ns = get_wm_urgent(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2196 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.cstate_enter_plus_exit_ns = get_wm_stutter_enter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2197 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2198 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; 2199 context->bw_ctx.bw.dcn.watermarks.a.pte_meta_urgent_ns = get_wm_memory_trip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2200 context->bw_ctx.bw.dcn.watermarks.a.frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2201 context->bw_ctx.bw.dcn.watermarks.a.frac_urg_bw_flip = get_fraction_of_urgent_bandwidth_imm_flip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2202 context->bw_ctx.bw.dcn.watermarks.a.urgent_latency_ns = get_urgent_latency(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2203 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.fclk_pstate_change_ns = get_fclk_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2204 context->bw_ctx.bw.dcn.watermarks.a.usr_retraining_ns = get_usr_retraining_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; 2205 } 2206 2207 /* Make set D = set A since we do not optimized watermarks for MALL */ 2208 context->bw_ctx.bw.dcn.watermarks.d = context->bw_ctx.bw.dcn.watermarks.a; 2209 2210 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) { 2211 if (!context->res_ctx.pipe_ctx[i].stream) 2212 continue; 2213 2214 pipes[pipe_idx].clks_cfg.dispclk_mhz = get_dispclk_calculated(&context->bw_ctx.dml, pipes, pipe_cnt); 2215 pipes[pipe_idx].clks_cfg.dppclk_mhz = get_dppclk_calculated(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx); 2216 2217 if (dc->config.forced_clocks) { 2218 pipes[pipe_idx].clks_cfg.dispclk_mhz = context->bw_ctx.dml.soc.clock_limits[0].dispclk_mhz; 2219 pipes[pipe_idx].clks_cfg.dppclk_mhz = context->bw_ctx.dml.soc.clock_limits[0].dppclk_mhz; 2220 } 2221 if (dc->debug.min_disp_clk_khz > pipes[pipe_idx].clks_cfg.dispclk_mhz * 1000) 2222 pipes[pipe_idx].clks_cfg.dispclk_mhz = dc->debug.min_disp_clk_khz / 1000.0; 2223 if (dc->debug.min_dpp_clk_khz > pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000) 2224 pipes[pipe_idx].clks_cfg.dppclk_mhz = dc->debug.min_dpp_clk_khz / 1000.0; 2225 2226 pipe_idx++; 2227 } 2228 2229 context->perf_params.stutter_period_us = context->bw_ctx.dml.vba.StutterPeriod; 2230 2231 /* for proper prefetch calculations, if dummy lat > fclk lat, use fclk lat = dummy lat */ 2232 if (need_fclk_lat_as_dummy) 2233 context->bw_ctx.dml.soc.fclk_change_latency_us = 2234 dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us; 2235 2236 dcn32_calculate_dlg_params(dc, context, pipes, pipe_cnt, vlevel); 2237 2238 if (!pstate_en) 2239 /* Restore full p-state latency */ 2240 context->bw_ctx.dml.soc.dram_clock_change_latency_us = 2241 dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us; 2242 2243 /* revert fclk lat changes if required */ 2244 if (need_fclk_lat_as_dummy) 2245 context->bw_ctx.dml.soc.fclk_change_latency_us = 2246 dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.fclk_change_latency_us; 2247 } 2248 2249 static void dcn32_get_optimal_dcfclk_fclk_for_uclk(unsigned int uclk_mts, 2250 unsigned int *optimal_dcfclk, 2251 unsigned int *optimal_fclk) 2252 { 2253 double bw_from_dram, bw_from_dram1, bw_from_dram2; 2254 2255 bw_from_dram1 = uclk_mts * dcn3_2_soc.num_chans * 2256 dcn3_2_soc.dram_channel_width_bytes * (dcn3_2_soc.max_avg_dram_bw_use_normal_percent / 100); 2257 bw_from_dram2 = uclk_mts * dcn3_2_soc.num_chans * 2258 dcn3_2_soc.dram_channel_width_bytes * (dcn3_2_soc.max_avg_sdp_bw_use_normal_percent / 100); 2259 2260 bw_from_dram = (bw_from_dram1 < bw_from_dram2) ? bw_from_dram1 : bw_from_dram2; 2261 2262 if (optimal_fclk) 2263 *optimal_fclk = bw_from_dram / 2264 (dcn3_2_soc.fabric_datapath_to_dcn_data_return_bytes * (dcn3_2_soc.max_avg_sdp_bw_use_normal_percent / 100)); 2265 2266 if (optimal_dcfclk) 2267 *optimal_dcfclk = bw_from_dram / 2268 (dcn3_2_soc.return_bus_width_bytes * (dcn3_2_soc.max_avg_sdp_bw_use_normal_percent / 100)); 2269 } 2270 2271 static void remove_entry_from_table_at_index(struct _vcs_dpi_voltage_scaling_st *table, unsigned int *num_entries, 2272 unsigned int index) 2273 { 2274 int i; 2275 2276 if (*num_entries == 0) 2277 return; 2278 2279 for (i = index; i < *num_entries - 1; i++) { 2280 table[i] = table[i + 1]; 2281 } 2282 memset(&table[--(*num_entries)], 0, sizeof(struct _vcs_dpi_voltage_scaling_st)); 2283 } 2284 2285 void dcn32_patch_dpm_table(struct clk_bw_params *bw_params) 2286 { 2287 int i; 2288 unsigned int max_dcfclk_mhz = 0, max_dispclk_mhz = 0, max_dppclk_mhz = 0, 2289 max_phyclk_mhz = 0, max_dtbclk_mhz = 0, max_fclk_mhz = 0, max_uclk_mhz = 0; 2290 2291 for (i = 0; i < MAX_NUM_DPM_LVL; i++) { 2292 if (bw_params->clk_table.entries[i].dcfclk_mhz > max_dcfclk_mhz) 2293 max_dcfclk_mhz = bw_params->clk_table.entries[i].dcfclk_mhz; 2294 if (bw_params->clk_table.entries[i].fclk_mhz > max_fclk_mhz) 2295 max_fclk_mhz = bw_params->clk_table.entries[i].fclk_mhz; 2296 if (bw_params->clk_table.entries[i].memclk_mhz > max_uclk_mhz) 2297 max_uclk_mhz = bw_params->clk_table.entries[i].memclk_mhz; 2298 if (bw_params->clk_table.entries[i].dispclk_mhz > max_dispclk_mhz) 2299 max_dispclk_mhz = bw_params->clk_table.entries[i].dispclk_mhz; 2300 if (bw_params->clk_table.entries[i].dppclk_mhz > max_dppclk_mhz) 2301 max_dppclk_mhz = bw_params->clk_table.entries[i].dppclk_mhz; 2302 if (bw_params->clk_table.entries[i].phyclk_mhz > max_phyclk_mhz) 2303 max_phyclk_mhz = bw_params->clk_table.entries[i].phyclk_mhz; 2304 if (bw_params->clk_table.entries[i].dtbclk_mhz > max_dtbclk_mhz) 2305 max_dtbclk_mhz = bw_params->clk_table.entries[i].dtbclk_mhz; 2306 } 2307 2308 /* Scan through clock values we currently have and if they are 0, 2309 * then populate it with dcn3_2_soc.clock_limits[] value. 2310 * 2311 * Do it for DCFCLK, DISPCLK, DTBCLK and UCLK as any of those being 2312 * 0, will cause it to skip building the clock table. 2313 */ 2314 if (max_dcfclk_mhz == 0) 2315 bw_params->clk_table.entries[0].dcfclk_mhz = dcn3_2_soc.clock_limits[0].dcfclk_mhz; 2316 if (max_dispclk_mhz == 0) 2317 bw_params->clk_table.entries[0].dispclk_mhz = dcn3_2_soc.clock_limits[0].dispclk_mhz; 2318 if (max_dtbclk_mhz == 0) 2319 bw_params->clk_table.entries[0].dtbclk_mhz = dcn3_2_soc.clock_limits[0].dtbclk_mhz; 2320 if (max_uclk_mhz == 0) 2321 bw_params->clk_table.entries[0].memclk_mhz = dcn3_2_soc.clock_limits[0].dram_speed_mts / 16; 2322 } 2323 2324 static int build_synthetic_soc_states(struct clk_bw_params *bw_params, 2325 struct _vcs_dpi_voltage_scaling_st *table, unsigned int *num_entries) 2326 { 2327 int i, j; 2328 struct _vcs_dpi_voltage_scaling_st entry = {0}; 2329 2330 unsigned int max_dcfclk_mhz = 0, max_dispclk_mhz = 0, max_dppclk_mhz = 0, 2331 max_phyclk_mhz = 0, max_dtbclk_mhz = 0, max_fclk_mhz = 0, max_uclk_mhz = 0; 2332 2333 unsigned int min_dcfclk_mhz = 199, min_fclk_mhz = 299; 2334 2335 static const unsigned int num_dcfclk_stas = 5; 2336 unsigned int dcfclk_sta_targets[DC__VOLTAGE_STATES] = {199, 615, 906, 1324, 1564}; 2337 2338 unsigned int num_uclk_dpms = 0; 2339 unsigned int num_fclk_dpms = 0; 2340 unsigned int num_dcfclk_dpms = 0; 2341 2342 for (i = 0; i < MAX_NUM_DPM_LVL; i++) { 2343 if (bw_params->clk_table.entries[i].dcfclk_mhz > max_dcfclk_mhz) 2344 max_dcfclk_mhz = bw_params->clk_table.entries[i].dcfclk_mhz; 2345 if (bw_params->clk_table.entries[i].fclk_mhz > max_fclk_mhz) 2346 max_fclk_mhz = bw_params->clk_table.entries[i].fclk_mhz; 2347 if (bw_params->clk_table.entries[i].memclk_mhz > max_uclk_mhz) 2348 max_uclk_mhz = bw_params->clk_table.entries[i].memclk_mhz; 2349 if (bw_params->clk_table.entries[i].dispclk_mhz > max_dispclk_mhz) 2350 max_dispclk_mhz = bw_params->clk_table.entries[i].dispclk_mhz; 2351 if (bw_params->clk_table.entries[i].dppclk_mhz > max_dppclk_mhz) 2352 max_dppclk_mhz = bw_params->clk_table.entries[i].dppclk_mhz; 2353 if (bw_params->clk_table.entries[i].phyclk_mhz > max_phyclk_mhz) 2354 max_phyclk_mhz = bw_params->clk_table.entries[i].phyclk_mhz; 2355 if (bw_params->clk_table.entries[i].dtbclk_mhz > max_dtbclk_mhz) 2356 max_dtbclk_mhz = bw_params->clk_table.entries[i].dtbclk_mhz; 2357 2358 if (bw_params->clk_table.entries[i].memclk_mhz > 0) 2359 num_uclk_dpms++; 2360 if (bw_params->clk_table.entries[i].fclk_mhz > 0) 2361 num_fclk_dpms++; 2362 if (bw_params->clk_table.entries[i].dcfclk_mhz > 0) 2363 num_dcfclk_dpms++; 2364 } 2365 2366 if (num_dcfclk_dpms > 0 && bw_params->clk_table.entries[0].fclk_mhz > min_fclk_mhz) 2367 min_fclk_mhz = bw_params->clk_table.entries[0].fclk_mhz; 2368 2369 if (!max_dcfclk_mhz || !max_dispclk_mhz || !max_dtbclk_mhz) 2370 return -1; 2371 2372 if (max_dppclk_mhz == 0) 2373 max_dppclk_mhz = max_dispclk_mhz; 2374 2375 if (max_fclk_mhz == 0) 2376 max_fclk_mhz = max_dcfclk_mhz * dcn3_2_soc.pct_ideal_sdp_bw_after_urgent / dcn3_2_soc.pct_ideal_fabric_bw_after_urgent; 2377 2378 if (max_phyclk_mhz == 0) 2379 max_phyclk_mhz = dcn3_2_soc.clock_limits[0].phyclk_mhz; 2380 2381 *num_entries = 0; 2382 entry.dispclk_mhz = max_dispclk_mhz; 2383 entry.dscclk_mhz = max_dispclk_mhz / 3; 2384 entry.dppclk_mhz = max_dppclk_mhz; 2385 entry.dtbclk_mhz = max_dtbclk_mhz; 2386 entry.phyclk_mhz = max_phyclk_mhz; 2387 entry.phyclk_d18_mhz = dcn3_2_soc.clock_limits[0].phyclk_d18_mhz; 2388 entry.phyclk_d32_mhz = dcn3_2_soc.clock_limits[0].phyclk_d32_mhz; 2389 2390 // Insert all the DCFCLK STAs 2391 for (i = 0; i < num_dcfclk_stas; i++) { 2392 entry.dcfclk_mhz = dcfclk_sta_targets[i]; 2393 entry.fabricclk_mhz = 0; 2394 entry.dram_speed_mts = 0; 2395 2396 insert_entry_into_table_sorted(table, num_entries, &entry); 2397 } 2398 2399 // Insert the max DCFCLK 2400 entry.dcfclk_mhz = max_dcfclk_mhz; 2401 entry.fabricclk_mhz = 0; 2402 entry.dram_speed_mts = 0; 2403 2404 insert_entry_into_table_sorted(table, num_entries, &entry); 2405 2406 // Insert the UCLK DPMS 2407 for (i = 0; i < num_uclk_dpms; i++) { 2408 entry.dcfclk_mhz = 0; 2409 entry.fabricclk_mhz = 0; 2410 entry.dram_speed_mts = bw_params->clk_table.entries[i].memclk_mhz * 16; 2411 2412 insert_entry_into_table_sorted(table, num_entries, &entry); 2413 } 2414 2415 // If FCLK is coarse grained, insert individual DPMs. 2416 if (num_fclk_dpms > 2) { 2417 for (i = 0; i < num_fclk_dpms; i++) { 2418 entry.dcfclk_mhz = 0; 2419 entry.fabricclk_mhz = bw_params->clk_table.entries[i].fclk_mhz; 2420 entry.dram_speed_mts = 0; 2421 2422 insert_entry_into_table_sorted(table, num_entries, &entry); 2423 } 2424 } 2425 // If FCLK fine grained, only insert max 2426 else { 2427 entry.dcfclk_mhz = 0; 2428 entry.fabricclk_mhz = max_fclk_mhz; 2429 entry.dram_speed_mts = 0; 2430 2431 insert_entry_into_table_sorted(table, num_entries, &entry); 2432 } 2433 2434 // At this point, the table contains all "points of interest" based on 2435 // DPMs from PMFW, and STAs. Table is sorted by BW, and all clock 2436 // ratios (by derate, are exact). 2437 2438 // Remove states that require higher clocks than are supported 2439 for (i = *num_entries - 1; i >= 0 ; i--) { 2440 if (table[i].dcfclk_mhz > max_dcfclk_mhz || 2441 table[i].fabricclk_mhz > max_fclk_mhz || 2442 table[i].dram_speed_mts > max_uclk_mhz * 16) 2443 remove_entry_from_table_at_index(table, num_entries, i); 2444 } 2445 2446 // At this point, the table only contains supported points of interest 2447 // it could be used as is, but some states may be redundant due to 2448 // coarse grained nature of some clocks, so we want to round up to 2449 // coarse grained DPMs and remove duplicates. 2450 2451 // Round up UCLKs 2452 for (i = *num_entries - 1; i >= 0 ; i--) { 2453 for (j = 0; j < num_uclk_dpms; j++) { 2454 if (bw_params->clk_table.entries[j].memclk_mhz * 16 >= table[i].dram_speed_mts) { 2455 table[i].dram_speed_mts = bw_params->clk_table.entries[j].memclk_mhz * 16; 2456 break; 2457 } 2458 } 2459 } 2460 2461 // If FCLK is coarse grained, round up to next DPMs 2462 if (num_fclk_dpms > 2) { 2463 for (i = *num_entries - 1; i >= 0 ; i--) { 2464 for (j = 0; j < num_fclk_dpms; j++) { 2465 if (bw_params->clk_table.entries[j].fclk_mhz >= table[i].fabricclk_mhz) { 2466 table[i].fabricclk_mhz = bw_params->clk_table.entries[j].fclk_mhz; 2467 break; 2468 } 2469 } 2470 } 2471 } 2472 // Otherwise, round up to minimum. 2473 else { 2474 for (i = *num_entries - 1; i >= 0 ; i--) { 2475 if (table[i].fabricclk_mhz < min_fclk_mhz) { 2476 table[i].fabricclk_mhz = min_fclk_mhz; 2477 } 2478 } 2479 } 2480 2481 // Round DCFCLKs up to minimum 2482 for (i = *num_entries - 1; i >= 0 ; i--) { 2483 if (table[i].dcfclk_mhz < min_dcfclk_mhz) { 2484 table[i].dcfclk_mhz = min_dcfclk_mhz; 2485 } 2486 } 2487 2488 // Remove duplicate states, note duplicate states are always neighbouring since table is sorted. 2489 i = 0; 2490 while (i < *num_entries - 1) { 2491 if (table[i].dcfclk_mhz == table[i + 1].dcfclk_mhz && 2492 table[i].fabricclk_mhz == table[i + 1].fabricclk_mhz && 2493 table[i].dram_speed_mts == table[i + 1].dram_speed_mts) 2494 remove_entry_from_table_at_index(table, num_entries, i + 1); 2495 else 2496 i++; 2497 } 2498 2499 // Fix up the state indicies 2500 for (i = *num_entries - 1; i >= 0 ; i--) { 2501 table[i].state = i; 2502 } 2503 2504 return 0; 2505 } 2506 2507 /* 2508 * dcn32_update_bw_bounding_box 2509 * 2510 * This would override some dcn3_2 ip_or_soc initial parameters hardcoded from 2511 * spreadsheet with actual values as per dGPU SKU: 2512 * - with passed few options from dc->config 2513 * - with dentist_vco_frequency from Clk Mgr (currently hardcoded, but might 2514 * need to get it from PM FW) 2515 * - with passed latency values (passed in ns units) in dc-> bb override for 2516 * debugging purposes 2517 * - with passed latencies from VBIOS (in 100_ns units) if available for 2518 * certain dGPU SKU 2519 * - with number of DRAM channels from VBIOS (which differ for certain dGPU SKU 2520 * of the same ASIC) 2521 * - clocks levels with passed clk_table entries from Clk Mgr as reported by PM 2522 * FW for different clocks (which might differ for certain dGPU SKU of the 2523 * same ASIC) 2524 */ 2525 void dcn32_update_bw_bounding_box_fpu(struct dc *dc, struct clk_bw_params *bw_params) 2526 { 2527 dc_assert_fp_enabled(); 2528 2529 if (!IS_FPGA_MAXIMUS_DC(dc->ctx->dce_environment)) { 2530 /* Overrides from dc->config options */ 2531 dcn3_2_ip.clamp_min_dcfclk = dc->config.clamp_min_dcfclk; 2532 2533 /* Override from passed dc->bb_overrides if available*/ 2534 if ((int)(dcn3_2_soc.sr_exit_time_us * 1000) != dc->bb_overrides.sr_exit_time_ns 2535 && dc->bb_overrides.sr_exit_time_ns) { 2536 dcn3_2_soc.sr_exit_time_us = dc->bb_overrides.sr_exit_time_ns / 1000.0; 2537 } 2538 2539 if ((int)(dcn3_2_soc.sr_enter_plus_exit_time_us * 1000) 2540 != dc->bb_overrides.sr_enter_plus_exit_time_ns 2541 && dc->bb_overrides.sr_enter_plus_exit_time_ns) { 2542 dcn3_2_soc.sr_enter_plus_exit_time_us = 2543 dc->bb_overrides.sr_enter_plus_exit_time_ns / 1000.0; 2544 } 2545 2546 if ((int)(dcn3_2_soc.urgent_latency_us * 1000) != dc->bb_overrides.urgent_latency_ns 2547 && dc->bb_overrides.urgent_latency_ns) { 2548 dcn3_2_soc.urgent_latency_us = dc->bb_overrides.urgent_latency_ns / 1000.0; 2549 dcn3_2_soc.urgent_latency_pixel_data_only_us = dc->bb_overrides.urgent_latency_ns / 1000.0; 2550 } 2551 2552 if ((int)(dcn3_2_soc.dram_clock_change_latency_us * 1000) 2553 != dc->bb_overrides.dram_clock_change_latency_ns 2554 && dc->bb_overrides.dram_clock_change_latency_ns) { 2555 dcn3_2_soc.dram_clock_change_latency_us = 2556 dc->bb_overrides.dram_clock_change_latency_ns / 1000.0; 2557 } 2558 2559 if ((int)(dcn3_2_soc.fclk_change_latency_us * 1000) 2560 != dc->bb_overrides.fclk_clock_change_latency_ns 2561 && dc->bb_overrides.fclk_clock_change_latency_ns) { 2562 dcn3_2_soc.fclk_change_latency_us = 2563 dc->bb_overrides.fclk_clock_change_latency_ns / 1000; 2564 } 2565 2566 if ((int)(dcn3_2_soc.dummy_pstate_latency_us * 1000) 2567 != dc->bb_overrides.dummy_clock_change_latency_ns 2568 && dc->bb_overrides.dummy_clock_change_latency_ns) { 2569 dcn3_2_soc.dummy_pstate_latency_us = 2570 dc->bb_overrides.dummy_clock_change_latency_ns / 1000.0; 2571 } 2572 2573 /* Override from VBIOS if VBIOS bb_info available */ 2574 if (dc->ctx->dc_bios->funcs->get_soc_bb_info) { 2575 struct bp_soc_bb_info bb_info = {0}; 2576 2577 if (dc->ctx->dc_bios->funcs->get_soc_bb_info(dc->ctx->dc_bios, &bb_info) == BP_RESULT_OK) { 2578 if (bb_info.dram_clock_change_latency_100ns > 0) 2579 dcn3_2_soc.dram_clock_change_latency_us = 2580 bb_info.dram_clock_change_latency_100ns * 10; 2581 2582 if (bb_info.dram_sr_enter_exit_latency_100ns > 0) 2583 dcn3_2_soc.sr_enter_plus_exit_time_us = 2584 bb_info.dram_sr_enter_exit_latency_100ns * 10; 2585 2586 if (bb_info.dram_sr_exit_latency_100ns > 0) 2587 dcn3_2_soc.sr_exit_time_us = 2588 bb_info.dram_sr_exit_latency_100ns * 10; 2589 } 2590 } 2591 2592 /* Override from VBIOS for num_chan */ 2593 if (dc->ctx->dc_bios->vram_info.num_chans) { 2594 dcn3_2_soc.num_chans = dc->ctx->dc_bios->vram_info.num_chans; 2595 dcn3_2_soc.mall_allocated_for_dcn_mbytes = (double)(dcn32_calc_num_avail_chans_for_mall(dc, 2596 dc->ctx->dc_bios->vram_info.num_chans) * dc->caps.mall_size_per_mem_channel); 2597 } 2598 2599 if (dc->ctx->dc_bios->vram_info.dram_channel_width_bytes) 2600 dcn3_2_soc.dram_channel_width_bytes = dc->ctx->dc_bios->vram_info.dram_channel_width_bytes; 2601 } 2602 2603 /* DML DSC delay factor workaround */ 2604 dcn3_2_ip.dsc_delay_factor_wa = dc->debug.dsc_delay_factor_wa_x1000 / 1000.0; 2605 2606 dcn3_2_ip.min_prefetch_in_strobe_us = dc->debug.min_prefetch_in_strobe_ns / 1000.0; 2607 2608 /* Override dispclk_dppclk_vco_speed_mhz from Clk Mgr */ 2609 dcn3_2_soc.dispclk_dppclk_vco_speed_mhz = dc->clk_mgr->dentist_vco_freq_khz / 1000.0; 2610 dc->dml.soc.dispclk_dppclk_vco_speed_mhz = dc->clk_mgr->dentist_vco_freq_khz / 1000.0; 2611 2612 /* Overrides Clock levelsfrom CLK Mgr table entries as reported by PM FW */ 2613 if ((!IS_FPGA_MAXIMUS_DC(dc->ctx->dce_environment)) && (bw_params->clk_table.entries[0].memclk_mhz)) { 2614 if (dc->debug.use_legacy_soc_bb_mechanism) { 2615 unsigned int i = 0, j = 0, num_states = 0; 2616 2617 unsigned int dcfclk_mhz[DC__VOLTAGE_STATES] = {0}; 2618 unsigned int dram_speed_mts[DC__VOLTAGE_STATES] = {0}; 2619 unsigned int optimal_uclk_for_dcfclk_sta_targets[DC__VOLTAGE_STATES] = {0}; 2620 unsigned int optimal_dcfclk_for_uclk[DC__VOLTAGE_STATES] = {0}; 2621 unsigned int min_dcfclk = UINT_MAX; 2622 /* Set 199 as first value in STA target array to have a minimum DCFCLK value. 2623 * For DCN32 we set min to 199 so minimum FCLK DPM0 (300Mhz can be achieved) */ 2624 unsigned int dcfclk_sta_targets[DC__VOLTAGE_STATES] = {199, 615, 906, 1324, 1564}; 2625 unsigned int num_dcfclk_sta_targets = 4, num_uclk_states = 0; 2626 unsigned int max_dcfclk_mhz = 0, max_dispclk_mhz = 0, max_dppclk_mhz = 0, max_phyclk_mhz = 0; 2627 2628 for (i = 0; i < MAX_NUM_DPM_LVL; i++) { 2629 if (bw_params->clk_table.entries[i].dcfclk_mhz > max_dcfclk_mhz) 2630 max_dcfclk_mhz = bw_params->clk_table.entries[i].dcfclk_mhz; 2631 if (bw_params->clk_table.entries[i].dcfclk_mhz != 0 && 2632 bw_params->clk_table.entries[i].dcfclk_mhz < min_dcfclk) 2633 min_dcfclk = bw_params->clk_table.entries[i].dcfclk_mhz; 2634 if (bw_params->clk_table.entries[i].dispclk_mhz > max_dispclk_mhz) 2635 max_dispclk_mhz = bw_params->clk_table.entries[i].dispclk_mhz; 2636 if (bw_params->clk_table.entries[i].dppclk_mhz > max_dppclk_mhz) 2637 max_dppclk_mhz = bw_params->clk_table.entries[i].dppclk_mhz; 2638 if (bw_params->clk_table.entries[i].phyclk_mhz > max_phyclk_mhz) 2639 max_phyclk_mhz = bw_params->clk_table.entries[i].phyclk_mhz; 2640 } 2641 if (min_dcfclk > dcfclk_sta_targets[0]) 2642 dcfclk_sta_targets[0] = min_dcfclk; 2643 if (!max_dcfclk_mhz) 2644 max_dcfclk_mhz = dcn3_2_soc.clock_limits[0].dcfclk_mhz; 2645 if (!max_dispclk_mhz) 2646 max_dispclk_mhz = dcn3_2_soc.clock_limits[0].dispclk_mhz; 2647 if (!max_dppclk_mhz) 2648 max_dppclk_mhz = dcn3_2_soc.clock_limits[0].dppclk_mhz; 2649 if (!max_phyclk_mhz) 2650 max_phyclk_mhz = dcn3_2_soc.clock_limits[0].phyclk_mhz; 2651 2652 if (max_dcfclk_mhz > dcfclk_sta_targets[num_dcfclk_sta_targets-1]) { 2653 // If max DCFCLK is greater than the max DCFCLK STA target, insert into the DCFCLK STA target array 2654 dcfclk_sta_targets[num_dcfclk_sta_targets] = max_dcfclk_mhz; 2655 num_dcfclk_sta_targets++; 2656 } else if (max_dcfclk_mhz < dcfclk_sta_targets[num_dcfclk_sta_targets-1]) { 2657 // If max DCFCLK is less than the max DCFCLK STA target, cap values and remove duplicates 2658 for (i = 0; i < num_dcfclk_sta_targets; i++) { 2659 if (dcfclk_sta_targets[i] > max_dcfclk_mhz) { 2660 dcfclk_sta_targets[i] = max_dcfclk_mhz; 2661 break; 2662 } 2663 } 2664 // Update size of array since we "removed" duplicates 2665 num_dcfclk_sta_targets = i + 1; 2666 } 2667 2668 num_uclk_states = bw_params->clk_table.num_entries; 2669 2670 // Calculate optimal dcfclk for each uclk 2671 for (i = 0; i < num_uclk_states; i++) { 2672 dcn32_get_optimal_dcfclk_fclk_for_uclk(bw_params->clk_table.entries[i].memclk_mhz * 16, 2673 &optimal_dcfclk_for_uclk[i], NULL); 2674 if (optimal_dcfclk_for_uclk[i] < bw_params->clk_table.entries[0].dcfclk_mhz) { 2675 optimal_dcfclk_for_uclk[i] = bw_params->clk_table.entries[0].dcfclk_mhz; 2676 } 2677 } 2678 2679 // Calculate optimal uclk for each dcfclk sta target 2680 for (i = 0; i < num_dcfclk_sta_targets; i++) { 2681 for (j = 0; j < num_uclk_states; j++) { 2682 if (dcfclk_sta_targets[i] < optimal_dcfclk_for_uclk[j]) { 2683 optimal_uclk_for_dcfclk_sta_targets[i] = 2684 bw_params->clk_table.entries[j].memclk_mhz * 16; 2685 break; 2686 } 2687 } 2688 } 2689 2690 i = 0; 2691 j = 0; 2692 // create the final dcfclk and uclk table 2693 while (i < num_dcfclk_sta_targets && j < num_uclk_states && num_states < DC__VOLTAGE_STATES) { 2694 if (dcfclk_sta_targets[i] < optimal_dcfclk_for_uclk[j] && i < num_dcfclk_sta_targets) { 2695 dcfclk_mhz[num_states] = dcfclk_sta_targets[i]; 2696 dram_speed_mts[num_states++] = optimal_uclk_for_dcfclk_sta_targets[i++]; 2697 } else { 2698 if (j < num_uclk_states && optimal_dcfclk_for_uclk[j] <= max_dcfclk_mhz) { 2699 dcfclk_mhz[num_states] = optimal_dcfclk_for_uclk[j]; 2700 dram_speed_mts[num_states++] = bw_params->clk_table.entries[j++].memclk_mhz * 16; 2701 } else { 2702 j = num_uclk_states; 2703 } 2704 } 2705 } 2706 2707 while (i < num_dcfclk_sta_targets && num_states < DC__VOLTAGE_STATES) { 2708 dcfclk_mhz[num_states] = dcfclk_sta_targets[i]; 2709 dram_speed_mts[num_states++] = optimal_uclk_for_dcfclk_sta_targets[i++]; 2710 } 2711 2712 while (j < num_uclk_states && num_states < DC__VOLTAGE_STATES && 2713 optimal_dcfclk_for_uclk[j] <= max_dcfclk_mhz) { 2714 dcfclk_mhz[num_states] = optimal_dcfclk_for_uclk[j]; 2715 dram_speed_mts[num_states++] = bw_params->clk_table.entries[j++].memclk_mhz * 16; 2716 } 2717 2718 dcn3_2_soc.num_states = num_states; 2719 for (i = 0; i < dcn3_2_soc.num_states; i++) { 2720 dcn3_2_soc.clock_limits[i].state = i; 2721 dcn3_2_soc.clock_limits[i].dcfclk_mhz = dcfclk_mhz[i]; 2722 dcn3_2_soc.clock_limits[i].fabricclk_mhz = dcfclk_mhz[i]; 2723 2724 /* Fill all states with max values of all these clocks */ 2725 dcn3_2_soc.clock_limits[i].dispclk_mhz = max_dispclk_mhz; 2726 dcn3_2_soc.clock_limits[i].dppclk_mhz = max_dppclk_mhz; 2727 dcn3_2_soc.clock_limits[i].phyclk_mhz = max_phyclk_mhz; 2728 dcn3_2_soc.clock_limits[i].dscclk_mhz = max_dispclk_mhz / 3; 2729 2730 /* Populate from bw_params for DTBCLK, SOCCLK */ 2731 if (i > 0) { 2732 if (!bw_params->clk_table.entries[i].dtbclk_mhz) { 2733 dcn3_2_soc.clock_limits[i].dtbclk_mhz = dcn3_2_soc.clock_limits[i-1].dtbclk_mhz; 2734 } else { 2735 dcn3_2_soc.clock_limits[i].dtbclk_mhz = bw_params->clk_table.entries[i].dtbclk_mhz; 2736 } 2737 } else if (bw_params->clk_table.entries[i].dtbclk_mhz) { 2738 dcn3_2_soc.clock_limits[i].dtbclk_mhz = bw_params->clk_table.entries[i].dtbclk_mhz; 2739 } 2740 2741 if (!bw_params->clk_table.entries[i].socclk_mhz && i > 0) 2742 dcn3_2_soc.clock_limits[i].socclk_mhz = dcn3_2_soc.clock_limits[i-1].socclk_mhz; 2743 else 2744 dcn3_2_soc.clock_limits[i].socclk_mhz = bw_params->clk_table.entries[i].socclk_mhz; 2745 2746 if (!dram_speed_mts[i] && i > 0) 2747 dcn3_2_soc.clock_limits[i].dram_speed_mts = dcn3_2_soc.clock_limits[i-1].dram_speed_mts; 2748 else 2749 dcn3_2_soc.clock_limits[i].dram_speed_mts = dram_speed_mts[i]; 2750 2751 /* These clocks cannot come from bw_params, always fill from dcn3_2_soc[0] */ 2752 /* PHYCLK_D18, PHYCLK_D32 */ 2753 dcn3_2_soc.clock_limits[i].phyclk_d18_mhz = dcn3_2_soc.clock_limits[0].phyclk_d18_mhz; 2754 dcn3_2_soc.clock_limits[i].phyclk_d32_mhz = dcn3_2_soc.clock_limits[0].phyclk_d32_mhz; 2755 } 2756 } else { 2757 build_synthetic_soc_states(bw_params, dcn3_2_soc.clock_limits, &dcn3_2_soc.num_states); 2758 } 2759 2760 /* Re-init DML with updated bb */ 2761 dml_init_instance(&dc->dml, &dcn3_2_soc, &dcn3_2_ip, DML_PROJECT_DCN32); 2762 if (dc->current_state) 2763 dml_init_instance(&dc->current_state->bw_ctx.dml, &dcn3_2_soc, &dcn3_2_ip, DML_PROJECT_DCN32); 2764 } 2765 } 2766 2767 void dcn32_zero_pipe_dcc_fraction(display_e2e_pipe_params_st *pipes, 2768 int pipe_cnt) 2769 { 2770 dc_assert_fp_enabled(); 2771 2772 pipes[pipe_cnt].pipe.src.dcc_fraction_of_zs_req_luma = 0; 2773 pipes[pipe_cnt].pipe.src.dcc_fraction_of_zs_req_chroma = 0; 2774 } 2775 2776 bool dcn32_allow_subvp_with_active_margin(struct pipe_ctx *pipe) 2777 { 2778 bool allow = false; 2779 uint32_t refresh_rate = 0; 2780 2781 /* Allow subvp on displays that have active margin for 2560x1440@60hz displays 2782 * only for now. There must be no scaling as well. 2783 * 2784 * For now we only enable on 2560x1440@60hz displays to enable 4K60 + 1440p60 configs 2785 * for p-state switching. 2786 */ 2787 if (pipe->stream && pipe->plane_state) { 2788 refresh_rate = (pipe->stream->timing.pix_clk_100hz * 100 + 2789 pipe->stream->timing.v_total * pipe->stream->timing.h_total - 1) 2790 / (double)(pipe->stream->timing.v_total * pipe->stream->timing.h_total); 2791 if (pipe->stream->timing.v_addressable == 1440 && 2792 pipe->stream->timing.h_addressable == 2560 && 2793 refresh_rate >= 55 && refresh_rate <= 65 && 2794 pipe->plane_state->src_rect.height == 1440 && 2795 pipe->plane_state->src_rect.width == 2560 && 2796 pipe->plane_state->dst_rect.height == 1440 && 2797 pipe->plane_state->dst_rect.width == 2560) 2798 allow = true; 2799 } 2800 return allow; 2801 } 2802 2803 /** 2804 * ************************************************************************************************ 2805 * dcn32_allow_subvp_high_refresh_rate: Determine if the high refresh rate config will allow subvp 2806 * 2807 * @param [in]: dc: Current DC state 2808 * @param [in]: context: New DC state to be programmed 2809 * @param [in]: pipe: Pipe to be considered for use in subvp 2810 * 2811 * On high refresh rate display configs, we will allow subvp under the following conditions: 2812 * 1. Resolution is 3840x2160, 3440x1440, or 2560x1440 2813 * 2. Refresh rate is between 120hz - 165hz 2814 * 3. No scaling 2815 * 4. Freesync is inactive 2816 * 5. For single display cases, freesync must be disabled 2817 * 2818 * @return: True if pipe can be used for subvp, false otherwise 2819 * 2820 * ************************************************************************************************ 2821 */ 2822 bool dcn32_allow_subvp_high_refresh_rate(struct dc *dc, struct dc_state *context, struct pipe_ctx *pipe) 2823 { 2824 bool allow = false; 2825 uint32_t refresh_rate = 0; 2826 uint32_t min_refresh = subvp_high_refresh_list.min_refresh; 2827 uint32_t max_refresh = subvp_high_refresh_list.max_refresh; 2828 uint32_t i; 2829 2830 if (!dc->debug.disable_subvp_high_refresh && pipe->stream && 2831 pipe->plane_state && !pipe->stream->vrr_active_variable) { 2832 refresh_rate = (pipe->stream->timing.pix_clk_100hz * 100 + 2833 pipe->stream->timing.v_total * pipe->stream->timing.h_total - 1) 2834 / (double)(pipe->stream->timing.v_total * pipe->stream->timing.h_total); 2835 if (refresh_rate >= min_refresh && refresh_rate <= max_refresh) { 2836 for (i = 0; i < SUBVP_HIGH_REFRESH_LIST_LEN; i++) { 2837 uint32_t width = subvp_high_refresh_list.res[i].width; 2838 uint32_t height = subvp_high_refresh_list.res[i].height; 2839 2840 if (dcn32_check_native_scaling_for_res(pipe, width, height)) { 2841 if ((context->stream_count == 1 && !pipe->stream->allow_freesync) || context->stream_count > 1) { 2842 allow = true; 2843 break; 2844 } 2845 } 2846 } 2847 } 2848 } 2849 return allow; 2850 } 2851 2852 /** 2853 * ******************************************************************************************* 2854 * dcn32_determine_max_vratio_prefetch: Determine max Vratio for prefetch by driver policy 2855 * 2856 * @param [in]: dc: Current DC state 2857 * @param [in]: context: New DC state to be programmed 2858 * 2859 * @return: Max vratio for prefetch 2860 * 2861 * ******************************************************************************************* 2862 */ 2863 double dcn32_determine_max_vratio_prefetch(struct dc *dc, struct dc_state *context) 2864 { 2865 double max_vratio_pre = __DML_MAX_BW_RATIO_PRE__; // Default value is 4 2866 int i; 2867 2868 /* For single display MPO configs, allow the max vratio to be 8 2869 * if any plane is YUV420 format 2870 */ 2871 if (context->stream_count == 1 && context->stream_status[0].plane_count > 1) { 2872 for (i = 0; i < context->stream_status[0].plane_count; i++) { 2873 if (context->stream_status[0].plane_states[i]->format == SURFACE_PIXEL_FORMAT_VIDEO_420_YCbCr || 2874 context->stream_status[0].plane_states[i]->format == SURFACE_PIXEL_FORMAT_VIDEO_420_YCrCb) { 2875 max_vratio_pre = __DML_MAX_VRATIO_PRE__; 2876 } 2877 } 2878 } 2879 return max_vratio_pre; 2880 } 2881 2882 /** 2883 * dcn32_assign_fpo_vactive_candidate - Assign the FPO stream candidate for FPO + VActive case 2884 * 2885 * This function chooses the FPO candidate stream for FPO + VActive cases (2 stream config). 2886 * For FPO + VAtive cases, the assumption is that one display has ActiveMargin > 0, and the 2887 * other display has ActiveMargin <= 0. This function will choose the pipe/stream that has 2888 * ActiveMargin <= 0 to be the FPO stream candidate if found. 2889 * 2890 * 2891 * @param [in]: dc - current dc state 2892 * @param [in]: context - new dc state 2893 * @param [out]: fpo_candidate_stream - pointer to FPO stream candidate if one is found 2894 * 2895 * Return: void 2896 */ 2897 void dcn32_assign_fpo_vactive_candidate(struct dc *dc, const struct dc_state *context, struct dc_stream_state **fpo_candidate_stream) 2898 { 2899 unsigned int i, pipe_idx; 2900 const struct vba_vars_st *vba = &context->bw_ctx.dml.vba; 2901 2902 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) { 2903 const struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i]; 2904 2905 if (!pipe->stream) 2906 continue; 2907 2908 if (vba->ActiveDRAMClockChangeLatencyMarginPerState[vba->VoltageLevel][vba->maxMpcComb][vba->pipe_plane[pipe_idx]] <= 0) { 2909 *fpo_candidate_stream = pipe->stream; 2910 break; 2911 } 2912 pipe_idx++; 2913 } 2914 } 2915 2916 /** 2917 * dcn32_find_vactive_pipe - Determines if the config has a pipe that can switch in VACTIVE 2918 * 2919 * @param [in]: dc - current dc state 2920 * @param [in]: context - new dc state 2921 * @param [in]: vactive_margin_req_us - The vactive marign required for a vactive pipe to be 2922 * considered "found" 2923 * 2924 * Return: True if VACTIVE display is found, false otherwise 2925 */ 2926 bool dcn32_find_vactive_pipe(struct dc *dc, const struct dc_state *context, uint32_t vactive_margin_req_us) 2927 { 2928 unsigned int i, pipe_idx; 2929 const struct vba_vars_st *vba = &context->bw_ctx.dml.vba; 2930 bool vactive_found = false; 2931 2932 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) { 2933 const struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i]; 2934 2935 if (!pipe->stream) 2936 continue; 2937 2938 if (vba->ActiveDRAMClockChangeLatencyMarginPerState[vba->VoltageLevel][vba->maxMpcComb][vba->pipe_plane[pipe_idx]] >= vactive_margin_req_us) { 2939 vactive_found = true; 2940 break; 2941 } 2942 pipe_idx++; 2943 } 2944 return vactive_found; 2945 } 2946 2947 void dcn32_set_clock_limits(const struct _vcs_dpi_soc_bounding_box_st *soc_bb) 2948 { 2949 dc_assert_fp_enabled(); 2950 dcn3_2_soc.clock_limits[0].dcfclk_mhz = 1200.0; 2951 } 2952 2953 void dcn32_override_min_req_memclk(struct dc *dc, struct dc_state *context) 2954 { 2955 // WA: restrict FPO and SubVP to use first non-strobe mode (DCN32 BW issue) 2956 if ((context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching || dcn32_subvp_in_use(dc, context)) && 2957 dc->dml.soc.num_chans <= 8) { 2958 int num_mclk_levels = dc->clk_mgr->bw_params->clk_table.num_entries_per_clk.num_memclk_levels; 2959 2960 if (context->bw_ctx.dml.vba.DRAMSpeed <= dc->clk_mgr->bw_params->clk_table.entries[0].memclk_mhz * 16 && 2961 num_mclk_levels > 1) { 2962 context->bw_ctx.dml.vba.DRAMSpeed = dc->clk_mgr->bw_params->clk_table.entries[1].memclk_mhz * 16; 2963 context->bw_ctx.bw.dcn.clk.dramclk_khz = context->bw_ctx.dml.vba.DRAMSpeed * 1000 / 16; 2964 } 2965 } 2966 } 2967