1 // SPDX-License-Identifier: MIT
3 * Copyright 2022 Advanced Micro Devices, Inc.
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice shall be included in
13 * all copies or substantial portions of the Software.
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
19 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
20 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
21 * OTHER DEALINGS IN THE SOFTWARE.
26 #include "dcn32_fpu.h"
27 #include "dc_link_dp.h"
28 #include "dcn32/dcn32_resource.h"
29 #include "dcn20/dcn20_resource.h"
30 #include "display_mode_vba_util_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"
35 #define DC_LOGGER_INIT(logger)
37 struct _vcs_dpi_ip_params_st dcn3_2_ip = {
39 .gpuvm_max_page_table_levels = 4,
41 .rob_buffer_size_kbytes = 128,
42 .det_buffer_size_kbytes = DCN3_2_DEFAULT_DET_SIZE,
43 .config_return_buffer_size_in_kbytes = 1280,
44 .compressed_buffer_segment_size_in_kbytes = 64,
45 .meta_fifo_size_in_kentries = 22,
46 .zero_size_buffer_entries = 512,
47 .compbuf_reserved_space_64b = 256,
48 .compbuf_reserved_space_zs = 64,
49 .dpp_output_buffer_pixels = 2560,
50 .opp_output_buffer_lines = 1,
51 .pixel_chunk_size_kbytes = 8,
52 .alpha_pixel_chunk_size_kbytes = 4,
53 .min_pixel_chunk_size_bytes = 1024,
54 .dcc_meta_buffer_size_bytes = 6272,
55 .meta_chunk_size_kbytes = 2,
56 .min_meta_chunk_size_bytes = 256,
57 .writeback_chunk_size_kbytes = 8,
58 .ptoi_supported = false,
60 .maximum_dsc_bits_per_component = 12,
61 .maximum_pixels_per_line_per_dsc_unit = 6016,
62 .dsc422_native_support = true,
63 .is_line_buffer_bpp_fixed = true,
64 .line_buffer_fixed_bpp = 57,
65 .line_buffer_size_bits = 1171920,
66 .max_line_buffer_lines = 32,
67 .writeback_interface_buffer_size_kbytes = 90,
70 .max_num_hdmi_frl_outputs = 1,
72 .max_dchub_pscl_bw_pix_per_clk = 4,
73 .max_pscl_lb_bw_pix_per_clk = 2,
74 .max_lb_vscl_bw_pix_per_clk = 4,
75 .max_vscl_hscl_bw_pix_per_clk = 4,
80 .dpte_buffer_size_in_pte_reqs_luma = 64,
81 .dpte_buffer_size_in_pte_reqs_chroma = 34,
82 .dispclk_ramp_margin_percent = 1,
83 .max_inter_dcn_tile_repeaters = 8,
84 .cursor_buffer_size = 16,
85 .cursor_chunk_size = 2,
86 .writeback_line_buffer_buffer_size = 0,
87 .writeback_min_hscl_ratio = 1,
88 .writeback_min_vscl_ratio = 1,
89 .writeback_max_hscl_ratio = 1,
90 .writeback_max_vscl_ratio = 1,
91 .writeback_max_hscl_taps = 1,
92 .writeback_max_vscl_taps = 1,
93 .dppclk_delay_subtotal = 47,
94 .dppclk_delay_scl = 50,
95 .dppclk_delay_scl_lb_only = 16,
96 .dppclk_delay_cnvc_formatter = 28,
97 .dppclk_delay_cnvc_cursor = 6,
98 .dispclk_delay_subtotal = 125,
99 .dynamic_metadata_vm_enabled = false,
100 .odm_combine_4to1_supported = false,
101 .dcc_supported = true,
102 .max_num_dp2p0_outputs = 2,
103 .max_num_dp2p0_streams = 4,
106 struct _vcs_dpi_soc_bounding_box_st dcn3_2_soc = {
110 .dcfclk_mhz = 1564.0,
111 .fabricclk_mhz = 400.0,
112 .dispclk_mhz = 2150.0,
113 .dppclk_mhz = 2150.0,
115 .phyclk_d18_mhz = 667.0,
116 .phyclk_d32_mhz = 625.0,
117 .socclk_mhz = 1200.0,
118 .dscclk_mhz = 716.667,
119 .dram_speed_mts = 16000.0,
120 .dtbclk_mhz = 1564.0,
124 .sr_exit_time_us = 42.97,
125 .sr_enter_plus_exit_time_us = 49.94,
126 .sr_exit_z8_time_us = 285.0,
127 .sr_enter_plus_exit_z8_time_us = 320,
128 .writeback_latency_us = 12.0,
129 .round_trip_ping_latency_dcfclk_cycles = 263,
130 .urgent_latency_pixel_data_only_us = 4.0,
131 .urgent_latency_pixel_mixed_with_vm_data_us = 4.0,
132 .urgent_latency_vm_data_only_us = 4.0,
133 .fclk_change_latency_us = 20,
134 .usr_retraining_latency_us = 2,
136 .mall_allocated_for_dcn_mbytes = 64,
137 .urgent_out_of_order_return_per_channel_pixel_only_bytes = 4096,
138 .urgent_out_of_order_return_per_channel_pixel_and_vm_bytes = 4096,
139 .urgent_out_of_order_return_per_channel_vm_only_bytes = 4096,
140 .pct_ideal_sdp_bw_after_urgent = 100.0,
141 .pct_ideal_fabric_bw_after_urgent = 67.0,
142 .pct_ideal_dram_sdp_bw_after_urgent_pixel_only = 20.0,
143 .pct_ideal_dram_sdp_bw_after_urgent_pixel_and_vm = 60.0, // N/A, for now keep as is until DML implemented
144 .pct_ideal_dram_sdp_bw_after_urgent_vm_only = 30.0, // N/A, for now keep as is until DML implemented
145 .pct_ideal_dram_bw_after_urgent_strobe = 67.0,
146 .max_avg_sdp_bw_use_normal_percent = 80.0,
147 .max_avg_fabric_bw_use_normal_percent = 60.0,
148 .max_avg_dram_bw_use_normal_strobe_percent = 50.0,
149 .max_avg_dram_bw_use_normal_percent = 15.0,
151 .dram_channel_width_bytes = 2,
152 .fabric_datapath_to_dcn_data_return_bytes = 64,
153 .return_bus_width_bytes = 64,
154 .downspread_percent = 0.38,
155 .dcn_downspread_percent = 0.5,
156 .dram_clock_change_latency_us = 400,
157 .dispclk_dppclk_vco_speed_mhz = 4300.0,
158 .do_urgent_latency_adjustment = true,
159 .urgent_latency_adjustment_fabric_clock_component_us = 1.0,
160 .urgent_latency_adjustment_fabric_clock_reference_mhz = 1000,
163 void dcn32_build_wm_range_table_fpu(struct clk_mgr_internal *clk_mgr)
166 double pstate_latency_us = clk_mgr->base.ctx->dc->dml.soc.dram_clock_change_latency_us;
167 double fclk_change_latency_us = clk_mgr->base.ctx->dc->dml.soc.fclk_change_latency_us;
168 double sr_exit_time_us = clk_mgr->base.ctx->dc->dml.soc.sr_exit_time_us;
169 double sr_enter_plus_exit_time_us = clk_mgr->base.ctx->dc->dml.soc.sr_enter_plus_exit_time_us;
170 /* For min clocks use as reported by PM FW and report those as min */
171 uint16_t min_uclk_mhz = clk_mgr->base.bw_params->clk_table.entries[0].memclk_mhz;
172 uint16_t min_dcfclk_mhz = clk_mgr->base.bw_params->clk_table.entries[0].dcfclk_mhz;
173 uint16_t setb_min_uclk_mhz = min_uclk_mhz;
174 uint16_t dcfclk_mhz_for_the_second_state = clk_mgr->base.ctx->dc->dml.soc.clock_limits[2].dcfclk_mhz;
176 dc_assert_fp_enabled();
178 /* For Set B ranges use min clocks state 2 when available, and report those to PM FW */
179 if (dcfclk_mhz_for_the_second_state)
180 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.min_dcfclk = dcfclk_mhz_for_the_second_state;
182 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;
184 if (clk_mgr->base.bw_params->clk_table.entries[2].memclk_mhz)
185 setb_min_uclk_mhz = clk_mgr->base.bw_params->clk_table.entries[2].memclk_mhz;
187 /* Set A - Normal - default values */
188 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].valid = true;
189 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us = pstate_latency_us;
190 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].dml_input.fclk_change_latency_us = fclk_change_latency_us;
191 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].dml_input.sr_exit_time_us = sr_exit_time_us;
192 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;
193 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.wm_type = WATERMARKS_CLOCK_RANGE;
194 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.min_dcfclk = min_dcfclk_mhz;
195 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.max_dcfclk = 0xFFFF;
196 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.min_uclk = min_uclk_mhz;
197 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.max_uclk = 0xFFFF;
199 /* Set B - Performance - higher clocks, using DPM[2] DCFCLK and UCLK */
200 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].valid = true;
201 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].dml_input.pstate_latency_us = pstate_latency_us;
202 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].dml_input.fclk_change_latency_us = fclk_change_latency_us;
203 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].dml_input.sr_exit_time_us = sr_exit_time_us;
204 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;
205 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.wm_type = WATERMARKS_CLOCK_RANGE;
206 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.max_dcfclk = 0xFFFF;
207 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.min_uclk = setb_min_uclk_mhz;
208 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.max_uclk = 0xFFFF;
210 /* Set C - Dummy P-State - P-State latency set to "dummy p-state" value */
211 /* 'DalDummyClockChangeLatencyNs' registry key option set to 0x7FFFFFFF can be used to disable Set C for dummy p-state */
212 if (clk_mgr->base.ctx->dc->bb_overrides.dummy_clock_change_latency_ns != 0x7FFFFFFF) {
213 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].valid = true;
214 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].dml_input.pstate_latency_us = 38;
215 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].dml_input.fclk_change_latency_us = fclk_change_latency_us;
216 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].dml_input.sr_exit_time_us = sr_exit_time_us;
217 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;
218 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.wm_type = WATERMARKS_DUMMY_PSTATE;
219 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.min_dcfclk = min_dcfclk_mhz;
220 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.max_dcfclk = 0xFFFF;
221 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.min_uclk = min_uclk_mhz;
222 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.max_uclk = 0xFFFF;
223 clk_mgr->base.bw_params->dummy_pstate_table[0].dram_speed_mts = clk_mgr->base.bw_params->clk_table.entries[0].memclk_mhz * 16;
224 clk_mgr->base.bw_params->dummy_pstate_table[0].dummy_pstate_latency_us = 38;
225 clk_mgr->base.bw_params->dummy_pstate_table[1].dram_speed_mts = clk_mgr->base.bw_params->clk_table.entries[1].memclk_mhz * 16;
226 clk_mgr->base.bw_params->dummy_pstate_table[1].dummy_pstate_latency_us = 9;
227 clk_mgr->base.bw_params->dummy_pstate_table[2].dram_speed_mts = clk_mgr->base.bw_params->clk_table.entries[2].memclk_mhz * 16;
228 clk_mgr->base.bw_params->dummy_pstate_table[2].dummy_pstate_latency_us = 8;
229 clk_mgr->base.bw_params->dummy_pstate_table[3].dram_speed_mts = clk_mgr->base.bw_params->clk_table.entries[3].memclk_mhz * 16;
230 clk_mgr->base.bw_params->dummy_pstate_table[3].dummy_pstate_latency_us = 5;
232 /* Set D - MALL - SR enter and exit time specific to MALL, TBD after bringup or later phase for now use DRAM values / 2 */
233 /* For MALL DRAM clock change latency is N/A, for watermak calculations use lowest value dummy P state latency */
234 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].valid = true;
235 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;
236 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].dml_input.fclk_change_latency_us = fclk_change_latency_us;
237 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].dml_input.sr_exit_time_us = sr_exit_time_us / 2; // TBD
238 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
239 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.wm_type = WATERMARKS_MALL;
240 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.min_dcfclk = min_dcfclk_mhz;
241 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.max_dcfclk = 0xFFFF;
242 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.min_uclk = min_uclk_mhz;
243 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.max_uclk = 0xFFFF;
247 * Finds dummy_latency_index when MCLK switching using firmware based
248 * vblank stretch is enabled. This function will iterate through the
249 * table of dummy pstate latencies until the lowest value that allows
250 * dm_allow_self_refresh_and_mclk_switch to happen is found
252 int dcn32_find_dummy_latency_index_for_fw_based_mclk_switch(struct dc *dc,
253 struct dc_state *context,
254 display_e2e_pipe_params_st *pipes,
258 const int max_latency_table_entries = 4;
259 const struct vba_vars_st *vba = &context->bw_ctx.dml.vba;
260 int dummy_latency_index = 0;
262 dc_assert_fp_enabled();
264 while (dummy_latency_index < max_latency_table_entries) {
265 context->bw_ctx.dml.soc.dram_clock_change_latency_us =
266 dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us;
267 dcn32_internal_validate_bw(dc, context, pipes, &pipe_cnt, &vlevel, false);
269 if (vlevel < context->bw_ctx.dml.vba.soc.num_states &&
270 vba->DRAMClockChangeSupport[vlevel][vba->maxMpcComb] != dm_dram_clock_change_unsupported)
273 dummy_latency_index++;
276 if (dummy_latency_index == max_latency_table_entries) {
277 ASSERT(dummy_latency_index != max_latency_table_entries);
278 /* If the execution gets here, it means dummy p_states are
279 * not possible. This should never happen and would mean
280 * something is severely wrong.
281 * Here we reset dummy_latency_index to 3, because it is
282 * better to have underflows than system crashes.
284 dummy_latency_index = max_latency_table_entries - 1;
287 return dummy_latency_index;
291 * dcn32_helper_populate_phantom_dlg_params - Get DLG params for phantom pipes
292 * and populate pipe_ctx with those params.
293 * @dc: [in] current dc state
294 * @context: [in] new dc state
295 * @pipes: [in] DML pipe params array
296 * @pipe_cnt: [in] DML pipe count
298 * This function must be called AFTER the phantom pipes are added to context
299 * and run through DML (so that the DLG params for the phantom pipes can be
300 * populated), and BEFORE we program the timing for the phantom pipes.
302 void dcn32_helper_populate_phantom_dlg_params(struct dc *dc,
303 struct dc_state *context,
304 display_e2e_pipe_params_st *pipes,
307 uint32_t i, pipe_idx;
309 dc_assert_fp_enabled();
311 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
312 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
317 if (pipe->plane_state && pipe->stream->mall_stream_config.type == SUBVP_PHANTOM) {
318 pipes[pipe_idx].pipe.dest.vstartup_start =
319 get_vstartup(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx);
320 pipes[pipe_idx].pipe.dest.vupdate_offset =
321 get_vupdate_offset(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx);
322 pipes[pipe_idx].pipe.dest.vupdate_width =
323 get_vupdate_width(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx);
324 pipes[pipe_idx].pipe.dest.vready_offset =
325 get_vready_offset(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx);
326 pipe->pipe_dlg_param = pipes[pipe_idx].pipe.dest;
333 * dcn32_predict_pipe_split - Predict if pipe split will occur for a given DML pipe
334 * @context: [in] New DC state to be programmed
335 * @pipe_e2e: [in] DML pipe end to end context
337 * This function takes in a DML pipe (pipe_e2e) and predicts if pipe split is required (both
338 * ODM and MPC). For pipe split, ODM combine is determined by the ODM mode, and MPC combine is
339 * determined by DPPClk requirements
341 * This function follows the same policy as DML:
342 * - Check for ODM combine requirements / policy first
343 * - MPC combine is only chosen if there is no ODM combine requirements / policy in place, and
346 * Return: Number of splits expected (1 for 2:1 split, 3 for 4:1 split, 0 for no splits).
348 uint8_t dcn32_predict_pipe_split(struct dc_state *context,
349 display_e2e_pipe_params_st *pipe_e2e)
351 double pscl_throughput;
352 double pscl_throughput_chroma;
353 double dpp_clk_single_dpp, clock;
354 double clk_frequency = 0.0;
355 double vco_speed = context->bw_ctx.dml.soc.dispclk_dppclk_vco_speed_mhz;
356 bool total_available_pipes_support = false;
357 uint32_t number_of_dpp = 0;
358 enum odm_combine_mode odm_mode = dm_odm_combine_mode_disabled;
359 double req_dispclk_per_surface = 0;
360 uint8_t num_splits = 0;
362 dc_assert_fp_enabled();
364 dml32_CalculateODMMode(context->bw_ctx.dml.ip.maximum_pixels_per_line_per_dsc_unit,
365 pipe_e2e->pipe.dest.hactive,
366 pipe_e2e->dout.output_format,
367 pipe_e2e->dout.output_type,
368 pipe_e2e->pipe.dest.odm_combine_policy,
369 context->bw_ctx.dml.soc.clock_limits[context->bw_ctx.dml.soc.num_states - 1].dispclk_mhz,
370 context->bw_ctx.dml.soc.clock_limits[context->bw_ctx.dml.soc.num_states - 1].dispclk_mhz,
371 pipe_e2e->dout.dsc_enable != 0,
372 0, /* TotalNumberOfActiveDPP can be 0 since we're predicting pipe split requirement */
373 context->bw_ctx.dml.ip.max_num_dpp,
374 pipe_e2e->pipe.dest.pixel_rate_mhz,
375 context->bw_ctx.dml.soc.dcn_downspread_percent,
376 context->bw_ctx.dml.ip.dispclk_ramp_margin_percent,
377 context->bw_ctx.dml.soc.dispclk_dppclk_vco_speed_mhz,
378 pipe_e2e->dout.dsc_slices,
380 &total_available_pipes_support,
383 &req_dispclk_per_surface);
385 dml32_CalculateSinglePipeDPPCLKAndSCLThroughput(pipe_e2e->pipe.scale_ratio_depth.hscl_ratio,
386 pipe_e2e->pipe.scale_ratio_depth.hscl_ratio_c,
387 pipe_e2e->pipe.scale_ratio_depth.vscl_ratio,
388 pipe_e2e->pipe.scale_ratio_depth.vscl_ratio_c,
389 context->bw_ctx.dml.ip.max_dchub_pscl_bw_pix_per_clk,
390 context->bw_ctx.dml.ip.max_pscl_lb_bw_pix_per_clk,
391 pipe_e2e->pipe.dest.pixel_rate_mhz,
392 pipe_e2e->pipe.src.source_format,
393 pipe_e2e->pipe.scale_taps.htaps,
394 pipe_e2e->pipe.scale_taps.htaps_c,
395 pipe_e2e->pipe.scale_taps.vtaps,
396 pipe_e2e->pipe.scale_taps.vtaps_c,
398 &pscl_throughput, &pscl_throughput_chroma,
399 &dpp_clk_single_dpp);
401 clock = dpp_clk_single_dpp * (1 + context->bw_ctx.dml.soc.dcn_downspread_percent / 100);
404 clk_frequency = vco_speed * 4.0 / ((int)(vco_speed * 4.0) / clock);
406 if (odm_mode == dm_odm_combine_mode_2to1)
408 else if (odm_mode == dm_odm_combine_mode_4to1)
410 else if (clk_frequency > context->bw_ctx.dml.soc.clock_limits[context->bw_ctx.dml.soc.num_states - 1].dppclk_mhz)
416 static float calculate_net_bw_in_kbytes_sec(struct _vcs_dpi_voltage_scaling_st *entry)
418 float memory_bw_kbytes_sec;
419 float fabric_bw_kbytes_sec;
420 float sdp_bw_kbytes_sec;
421 float limiting_bw_kbytes_sec;
423 memory_bw_kbytes_sec = entry->dram_speed_mts *
424 dcn3_2_soc.num_chans *
425 dcn3_2_soc.dram_channel_width_bytes *
426 ((float)dcn3_2_soc.pct_ideal_dram_sdp_bw_after_urgent_pixel_only / 100);
428 fabric_bw_kbytes_sec = entry->fabricclk_mhz *
429 dcn3_2_soc.return_bus_width_bytes *
430 ((float)dcn3_2_soc.pct_ideal_fabric_bw_after_urgent / 100);
432 sdp_bw_kbytes_sec = entry->dcfclk_mhz *
433 dcn3_2_soc.return_bus_width_bytes *
434 ((float)dcn3_2_soc.pct_ideal_sdp_bw_after_urgent / 100);
436 limiting_bw_kbytes_sec = memory_bw_kbytes_sec;
438 if (fabric_bw_kbytes_sec < limiting_bw_kbytes_sec)
439 limiting_bw_kbytes_sec = fabric_bw_kbytes_sec;
441 if (sdp_bw_kbytes_sec < limiting_bw_kbytes_sec)
442 limiting_bw_kbytes_sec = sdp_bw_kbytes_sec;
444 return limiting_bw_kbytes_sec;
447 static void get_optimal_ntuple(struct _vcs_dpi_voltage_scaling_st *entry)
449 if (entry->dcfclk_mhz > 0) {
450 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);
452 entry->fabricclk_mhz = bw_on_sdp / (dcn3_2_soc.return_bus_width_bytes * ((float)dcn3_2_soc.pct_ideal_fabric_bw_after_urgent / 100));
453 entry->dram_speed_mts = bw_on_sdp / (dcn3_2_soc.num_chans *
454 dcn3_2_soc.dram_channel_width_bytes * ((float)dcn3_2_soc.pct_ideal_dram_sdp_bw_after_urgent_pixel_only / 100));
455 } else if (entry->fabricclk_mhz > 0) {
456 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);
458 entry->dcfclk_mhz = bw_on_fabric / (dcn3_2_soc.return_bus_width_bytes * ((float)dcn3_2_soc.pct_ideal_sdp_bw_after_urgent / 100));
459 entry->dram_speed_mts = bw_on_fabric / (dcn3_2_soc.num_chans *
460 dcn3_2_soc.dram_channel_width_bytes * ((float)dcn3_2_soc.pct_ideal_dram_sdp_bw_after_urgent_pixel_only / 100));
461 } else if (entry->dram_speed_mts > 0) {
462 float bw_on_dram = entry->dram_speed_mts * dcn3_2_soc.num_chans *
463 dcn3_2_soc.dram_channel_width_bytes * ((float)dcn3_2_soc.pct_ideal_dram_sdp_bw_after_urgent_pixel_only / 100);
465 entry->fabricclk_mhz = bw_on_dram / (dcn3_2_soc.return_bus_width_bytes * ((float)dcn3_2_soc.pct_ideal_fabric_bw_after_urgent / 100));
466 entry->dcfclk_mhz = bw_on_dram / (dcn3_2_soc.return_bus_width_bytes * ((float)dcn3_2_soc.pct_ideal_sdp_bw_after_urgent / 100));
470 void insert_entry_into_table_sorted(struct _vcs_dpi_voltage_scaling_st *table,
471 unsigned int *num_entries,
472 struct _vcs_dpi_voltage_scaling_st *entry)
476 float net_bw_of_new_state = 0;
478 dc_assert_fp_enabled();
480 get_optimal_ntuple(entry);
482 if (*num_entries == 0) {
486 net_bw_of_new_state = calculate_net_bw_in_kbytes_sec(entry);
487 while (net_bw_of_new_state > calculate_net_bw_in_kbytes_sec(&table[index])) {
489 if (index >= *num_entries)
493 for (i = *num_entries; i > index; i--)
494 table[i] = table[i - 1];
496 table[index] = *entry;
502 * dcn32_set_phantom_stream_timing - Set timing params for the phantom stream
503 * @dc: current dc state
504 * @context: new dc state
505 * @ref_pipe: Main pipe for the phantom stream
506 * @phantom_stream: target phantom stream state
507 * @pipes: DML pipe params
508 * @pipe_cnt: number of DML pipes
509 * @dc_pipe_idx: DC pipe index for the main pipe (i.e. ref_pipe)
511 * Set timing params of the phantom stream based on calculated output from DML.
512 * This function first gets the DML pipe index using the DC pipe index, then
513 * calls into DML (get_subviewport_lines_needed_in_mall) to get the number of
514 * lines required for SubVP MCLK switching and assigns to the phantom stream
517 * - The number of SubVP lines calculated in DML does not take into account
518 * FW processing delays and required pstate allow width, so we must include
521 * - Set phantom backporch = vstartup of main pipe
523 void dcn32_set_phantom_stream_timing(struct dc *dc,
524 struct dc_state *context,
525 struct pipe_ctx *ref_pipe,
526 struct dc_stream_state *phantom_stream,
527 display_e2e_pipe_params_st *pipes,
528 unsigned int pipe_cnt,
529 unsigned int dc_pipe_idx)
531 unsigned int i, pipe_idx;
532 struct pipe_ctx *pipe;
533 uint32_t phantom_vactive, phantom_bp, pstate_width_fw_delay_lines;
534 unsigned int vlevel = context->bw_ctx.dml.vba.VoltageLevel;
535 unsigned int dcfclk = context->bw_ctx.dml.vba.DCFCLKState[vlevel][context->bw_ctx.dml.vba.maxMpcComb];
536 unsigned int socclk = context->bw_ctx.dml.vba.SOCCLKPerState[vlevel];
538 dc_assert_fp_enabled();
540 // Find DML pipe index (pipe_idx) using dc_pipe_idx
541 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
542 pipe = &context->res_ctx.pipe_ctx[i];
547 if (i == dc_pipe_idx)
553 // Calculate lines required for pstate allow width and FW processing delays
554 pstate_width_fw_delay_lines = ((double)(dc->caps.subvp_fw_processing_delay_us +
555 dc->caps.subvp_pstate_allow_width_us) / 1000000) *
556 (ref_pipe->stream->timing.pix_clk_100hz * 100) /
557 (double)ref_pipe->stream->timing.h_total;
559 // Update clks_cfg for calling into recalculate
560 pipes[0].clks_cfg.voltage = vlevel;
561 pipes[0].clks_cfg.dcfclk_mhz = dcfclk;
562 pipes[0].clks_cfg.socclk_mhz = socclk;
564 // DML calculation for MALL region doesn't take into account FW delay
565 // and required pstate allow width for multi-display cases
566 /* Add 16 lines margin to the MALL REGION because SUB_VP_START_LINE must be aligned
567 * to 2 swaths (i.e. 16 lines)
569 phantom_vactive = get_subviewport_lines_needed_in_mall(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx) +
570 pstate_width_fw_delay_lines + dc->caps.subvp_swath_height_margin_lines;
572 // For backporch of phantom pipe, use vstartup of the main pipe
573 phantom_bp = get_vstartup(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx);
575 phantom_stream->dst.y = 0;
576 phantom_stream->dst.height = phantom_vactive;
577 phantom_stream->src.y = 0;
578 phantom_stream->src.height = phantom_vactive;
580 phantom_stream->timing.v_addressable = phantom_vactive;
581 phantom_stream->timing.v_front_porch = 1;
582 phantom_stream->timing.v_total = phantom_stream->timing.v_addressable +
583 phantom_stream->timing.v_front_porch +
584 phantom_stream->timing.v_sync_width +
586 phantom_stream->timing.flags.DSC = 0; // Don't need DSC for phantom timing
590 * dcn32_get_num_free_pipes - Calculate number of free pipes
591 * @dc: current dc state
592 * @context: new dc state
594 * This function assumes that a "used" pipe is a pipe that has
595 * both a stream and a plane assigned to it.
597 * Return: Number of free pipes available in the context
599 static unsigned int dcn32_get_num_free_pipes(struct dc *dc, struct dc_state *context)
602 unsigned int free_pipes = 0;
603 unsigned int num_pipes = 0;
605 for (i = 0; i < dc->res_pool->pipe_count; i++) {
606 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
608 if (pipe->stream && !pipe->top_pipe) {
611 pipe = pipe->bottom_pipe;
616 free_pipes = dc->res_pool->pipe_count - num_pipes;
621 * dcn32_assign_subvp_pipe - Function to decide which pipe will use Sub-VP.
622 * @dc: current dc state
623 * @context: new dc state
624 * @index: [out] dc pipe index for the pipe chosen to have phantom pipes assigned
626 * We enter this function if we are Sub-VP capable (i.e. enough pipes available)
627 * and regular P-State switching (i.e. VACTIVE/VBLANK) is not supported, or if
628 * we are forcing SubVP P-State switching on the current config.
630 * The number of pipes used for the chosen surface must be less than or equal to the
631 * number of free pipes available.
633 * In general we choose surfaces with the longest frame time first (better for SubVP + VBLANK).
634 * For multi-display cases the ActiveDRAMClockChangeMargin doesn't provide enough info on its own
635 * for determining which should be the SubVP pipe (need a way to determine if a pipe / plane doesn't
636 * support MCLK switching naturally [i.e. ACTIVE or VBLANK]).
638 * Return: True if a valid pipe assignment was found for Sub-VP. Otherwise false.
640 static bool dcn32_assign_subvp_pipe(struct dc *dc,
641 struct dc_state *context,
644 unsigned int i, pipe_idx;
645 unsigned int max_frame_time = 0;
646 bool valid_assignment_found = false;
647 unsigned int free_pipes = dcn32_get_num_free_pipes(dc, context);
648 bool current_assignment_freesync = false;
649 struct vba_vars_st *vba = &context->bw_ctx.dml.vba;
651 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
652 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
653 unsigned int num_pipes = 0;
654 unsigned int refresh_rate = 0;
660 refresh_rate = (pipe->stream->timing.pix_clk_100hz * 100 +
661 pipe->stream->timing.v_total * pipe->stream->timing.h_total - 1)
662 / (double)(pipe->stream->timing.v_total * pipe->stream->timing.h_total);
663 /* SubVP pipe candidate requirements:
664 * - Refresh rate < 120hz
665 * - Not able to switch in vactive naturally (switching in active means the
666 * DET provides enough buffer to hide the P-State switch latency -- trying
667 * to combine this with SubVP can cause issues with the scheduling).
670 if (pipe->plane_state && !pipe->top_pipe &&
671 pipe->stream->mall_stream_config.type == SUBVP_NONE && refresh_rate < 120 && !pipe->plane_state->address.tmz_surface &&
672 vba->ActiveDRAMClockChangeLatencyMarginPerState[vba->VoltageLevel][vba->maxMpcComb][vba->pipe_plane[pipe_idx]] <= 0) {
675 pipe = pipe->bottom_pipe;
678 pipe = &context->res_ctx.pipe_ctx[i];
679 if (num_pipes <= free_pipes) {
680 struct dc_stream_state *stream = pipe->stream;
681 unsigned int frame_us = (stream->timing.v_total * stream->timing.h_total /
682 (double)(stream->timing.pix_clk_100hz * 100)) * 1000000;
683 if (frame_us > max_frame_time && !stream->ignore_msa_timing_param) {
685 max_frame_time = frame_us;
686 valid_assignment_found = true;
687 current_assignment_freesync = false;
688 /* For the 2-Freesync display case, still choose the one with the
691 } else if (stream->ignore_msa_timing_param && (!valid_assignment_found ||
692 (current_assignment_freesync && frame_us > max_frame_time))) {
694 valid_assignment_found = true;
695 current_assignment_freesync = true;
701 return valid_assignment_found;
705 * dcn32_enough_pipes_for_subvp - Function to check if there are "enough" pipes for SubVP.
706 * @dc: current dc state
707 * @context: new dc state
709 * This function returns true if there are enough free pipes
710 * to create the required phantom pipes for any given stream
711 * (that does not already have phantom pipe assigned).
713 * e.g. For a 2 stream config where the first stream uses one
714 * pipe and the second stream uses 2 pipes (i.e. pipe split),
715 * this function will return true because there is 1 remaining
716 * pipe which can be used as the phantom pipe for the non pipe
720 * True if there are enough free pipes to assign phantom pipes to at least one
721 * stream that does not already have phantom pipes assigned. Otherwise false.
723 static bool dcn32_enough_pipes_for_subvp(struct dc *dc, struct dc_state *context)
725 unsigned int i, split_cnt, free_pipes;
726 unsigned int min_pipe_split = dc->res_pool->pipe_count + 1; // init as max number of pipes + 1
727 bool subvp_possible = false;
729 for (i = 0; i < dc->res_pool->pipe_count; i++) {
730 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
732 // Find the minimum pipe split count for non SubVP pipes
733 if (pipe->stream && !pipe->top_pipe &&
734 pipe->stream->mall_stream_config.type == SUBVP_NONE) {
738 pipe = pipe->bottom_pipe;
741 if (split_cnt < min_pipe_split)
742 min_pipe_split = split_cnt;
746 free_pipes = dcn32_get_num_free_pipes(dc, context);
748 // SubVP only possible if at least one pipe is being used (i.e. free_pipes
749 // should not equal to the pipe_count)
750 if (free_pipes >= min_pipe_split && free_pipes < dc->res_pool->pipe_count)
751 subvp_possible = true;
753 return subvp_possible;
757 * subvp_subvp_schedulable - Determine if SubVP + SubVP config is schedulable
758 * @dc: current dc state
759 * @context: new dc state
761 * High level algorithm:
762 * 1. Find longest microschedule length (in us) between the two SubVP pipes
763 * 2. Check if the worst case overlap (VBLANK in middle of ACTIVE) for both
764 * pipes still allows for the maximum microschedule to fit in the active
765 * region for both pipes.
767 * Return: True if the SubVP + SubVP config is schedulable, false otherwise
769 static bool subvp_subvp_schedulable(struct dc *dc, struct dc_state *context)
771 struct pipe_ctx *subvp_pipes[2];
772 struct dc_stream_state *phantom = NULL;
773 uint32_t microschedule_lines = 0;
776 uint32_t max_microschedule_us = 0;
777 int32_t vactive1_us, vactive2_us, vblank1_us, vblank2_us;
779 for (i = 0; i < dc->res_pool->pipe_count; i++) {
780 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
781 uint32_t time_us = 0;
783 /* Loop to calculate the maximum microschedule time between the two SubVP pipes,
784 * and also to store the two main SubVP pipe pointers in subvp_pipes[2].
786 if (pipe->stream && pipe->plane_state && !pipe->top_pipe &&
787 pipe->stream->mall_stream_config.type == SUBVP_MAIN) {
788 phantom = pipe->stream->mall_stream_config.paired_stream;
789 microschedule_lines = (phantom->timing.v_total - phantom->timing.v_front_porch) +
790 phantom->timing.v_addressable;
792 // Round up when calculating microschedule time (+ 1 at the end)
793 time_us = (microschedule_lines * phantom->timing.h_total) /
794 (double)(phantom->timing.pix_clk_100hz * 100) * 1000000 +
795 dc->caps.subvp_prefetch_end_to_mall_start_us +
796 dc->caps.subvp_fw_processing_delay_us + 1;
797 if (time_us > max_microschedule_us)
798 max_microschedule_us = time_us;
800 subvp_pipes[index] = pipe;
803 // Maximum 2 SubVP pipes
808 vactive1_us = ((subvp_pipes[0]->stream->timing.v_addressable * subvp_pipes[0]->stream->timing.h_total) /
809 (double)(subvp_pipes[0]->stream->timing.pix_clk_100hz * 100)) * 1000000;
810 vactive2_us = ((subvp_pipes[1]->stream->timing.v_addressable * subvp_pipes[1]->stream->timing.h_total) /
811 (double)(subvp_pipes[1]->stream->timing.pix_clk_100hz * 100)) * 1000000;
812 vblank1_us = (((subvp_pipes[0]->stream->timing.v_total - subvp_pipes[0]->stream->timing.v_addressable) *
813 subvp_pipes[0]->stream->timing.h_total) /
814 (double)(subvp_pipes[0]->stream->timing.pix_clk_100hz * 100)) * 1000000;
815 vblank2_us = (((subvp_pipes[1]->stream->timing.v_total - subvp_pipes[1]->stream->timing.v_addressable) *
816 subvp_pipes[1]->stream->timing.h_total) /
817 (double)(subvp_pipes[1]->stream->timing.pix_clk_100hz * 100)) * 1000000;
819 if ((vactive1_us - vblank2_us) / 2 > max_microschedule_us &&
820 (vactive2_us - vblank1_us) / 2 > max_microschedule_us)
827 * subvp_drr_schedulable - Determine if SubVP + DRR config is schedulable
828 * @dc: current dc state
829 * @context: new dc state
830 * @drr_pipe: DRR pipe_ctx for the SubVP + DRR config
832 * High level algorithm:
833 * 1. Get timing for SubVP pipe, phantom pipe, and DRR pipe
834 * 2. Determine the frame time for the DRR display when adding required margin for MCLK switching
835 * (the margin is equal to the MALL region + DRR margin (500us))
836 * 3.If (SubVP Active - Prefetch > Stretched DRR frame + max(MALL region, Stretched DRR frame))
837 * then report the configuration as supported
839 * Return: True if the SubVP + DRR config is schedulable, false otherwise
841 static bool subvp_drr_schedulable(struct dc *dc, struct dc_state *context, struct pipe_ctx *drr_pipe)
843 bool schedulable = false;
845 struct pipe_ctx *pipe = NULL;
846 struct dc_crtc_timing *main_timing = NULL;
847 struct dc_crtc_timing *phantom_timing = NULL;
848 struct dc_crtc_timing *drr_timing = NULL;
849 int16_t prefetch_us = 0;
850 int16_t mall_region_us = 0;
851 int16_t drr_frame_us = 0; // nominal frame time
852 int16_t subvp_active_us = 0;
853 int16_t stretched_drr_us = 0;
854 int16_t drr_stretched_vblank_us = 0;
855 int16_t max_vblank_mallregion = 0;
858 for (i = 0; i < dc->res_pool->pipe_count; i++) {
859 pipe = &context->res_ctx.pipe_ctx[i];
861 // We check for master pipe, but it shouldn't matter since we only need
862 // the pipe for timing info (stream should be same for any pipe splits)
863 if (!pipe->stream || !pipe->plane_state || pipe->top_pipe || pipe->prev_odm_pipe)
866 // Find the SubVP pipe
867 if (pipe->stream->mall_stream_config.type == SUBVP_MAIN)
871 main_timing = &pipe->stream->timing;
872 phantom_timing = &pipe->stream->mall_stream_config.paired_stream->timing;
873 drr_timing = &drr_pipe->stream->timing;
874 prefetch_us = (phantom_timing->v_total - phantom_timing->v_front_porch) * phantom_timing->h_total /
875 (double)(phantom_timing->pix_clk_100hz * 100) * 1000000 +
876 dc->caps.subvp_prefetch_end_to_mall_start_us;
877 subvp_active_us = main_timing->v_addressable * main_timing->h_total /
878 (double)(main_timing->pix_clk_100hz * 100) * 1000000;
879 drr_frame_us = drr_timing->v_total * drr_timing->h_total /
880 (double)(drr_timing->pix_clk_100hz * 100) * 1000000;
881 // P-State allow width and FW delays already included phantom_timing->v_addressable
882 mall_region_us = phantom_timing->v_addressable * phantom_timing->h_total /
883 (double)(phantom_timing->pix_clk_100hz * 100) * 1000000;
884 stretched_drr_us = drr_frame_us + mall_region_us + SUBVP_DRR_MARGIN_US;
885 drr_stretched_vblank_us = (drr_timing->v_total - drr_timing->v_addressable) * drr_timing->h_total /
886 (double)(drr_timing->pix_clk_100hz * 100) * 1000000 + (stretched_drr_us - drr_frame_us);
887 max_vblank_mallregion = drr_stretched_vblank_us > mall_region_us ? drr_stretched_vblank_us : mall_region_us;
889 /* We consider SubVP + DRR schedulable if the stretched frame duration of the DRR display (i.e. the
890 * highest refresh rate + margin that can support UCLK P-State switch) passes the static analysis
891 * for VBLANK: (VACTIVE region of the SubVP pipe can fit the MALL prefetch, VBLANK frame time,
892 * and the max of (VBLANK blanking time, MALL region)).
894 if (stretched_drr_us < (1 / (double)drr_timing->min_refresh_in_uhz) * 1000000 * 1000000 &&
895 subvp_active_us - prefetch_us - stretched_drr_us - max_vblank_mallregion > 0)
903 * subvp_vblank_schedulable - Determine if SubVP + VBLANK config is schedulable
904 * @dc: current dc state
905 * @context: new dc state
907 * High level algorithm:
908 * 1. Get timing for SubVP pipe, phantom pipe, and VBLANK pipe
909 * 2. If (SubVP Active - Prefetch > Vblank Frame Time + max(MALL region, Vblank blanking time))
910 * then report the configuration as supported
911 * 3. If the VBLANK display is DRR, then take the DRR static schedulability path
913 * Return: True if the SubVP + VBLANK/DRR config is schedulable, false otherwise
915 static bool subvp_vblank_schedulable(struct dc *dc, struct dc_state *context)
917 struct pipe_ctx *pipe = NULL;
918 struct pipe_ctx *subvp_pipe = NULL;
920 bool schedulable = false;
922 uint8_t vblank_index = 0;
923 uint16_t prefetch_us = 0;
924 uint16_t mall_region_us = 0;
925 uint16_t vblank_frame_us = 0;
926 uint16_t subvp_active_us = 0;
927 uint16_t vblank_blank_us = 0;
928 uint16_t max_vblank_mallregion = 0;
929 struct dc_crtc_timing *main_timing = NULL;
930 struct dc_crtc_timing *phantom_timing = NULL;
931 struct dc_crtc_timing *vblank_timing = NULL;
933 /* For SubVP + VBLANK/DRR cases, we assume there can only be
934 * a single VBLANK/DRR display. If DML outputs SubVP + VBLANK
935 * is supported, it is either a single VBLANK case or two VBLANK
936 * displays which are synchronized (in which case they have identical
939 for (i = 0; i < dc->res_pool->pipe_count; i++) {
940 pipe = &context->res_ctx.pipe_ctx[i];
942 // We check for master pipe, but it shouldn't matter since we only need
943 // the pipe for timing info (stream should be same for any pipe splits)
944 if (!pipe->stream || !pipe->plane_state || pipe->top_pipe || pipe->prev_odm_pipe)
947 if (!found && pipe->stream->mall_stream_config.type == SUBVP_NONE) {
948 // Found pipe which is not SubVP or Phantom (i.e. the VBLANK pipe).
953 if (!subvp_pipe && pipe->stream->mall_stream_config.type == SUBVP_MAIN)
956 // Use ignore_msa_timing_param flag to identify as DRR
957 if (found && context->res_ctx.pipe_ctx[vblank_index].stream->ignore_msa_timing_param) {
959 schedulable = subvp_drr_schedulable(dc, context, &context->res_ctx.pipe_ctx[vblank_index]);
961 main_timing = &subvp_pipe->stream->timing;
962 phantom_timing = &subvp_pipe->stream->mall_stream_config.paired_stream->timing;
963 vblank_timing = &context->res_ctx.pipe_ctx[vblank_index].stream->timing;
964 // Prefetch time is equal to VACTIVE + BP + VSYNC of the phantom pipe
965 // Also include the prefetch end to mallstart delay time
966 prefetch_us = (phantom_timing->v_total - phantom_timing->v_front_porch) * phantom_timing->h_total /
967 (double)(phantom_timing->pix_clk_100hz * 100) * 1000000 +
968 dc->caps.subvp_prefetch_end_to_mall_start_us;
969 // P-State allow width and FW delays already included phantom_timing->v_addressable
970 mall_region_us = phantom_timing->v_addressable * phantom_timing->h_total /
971 (double)(phantom_timing->pix_clk_100hz * 100) * 1000000;
972 vblank_frame_us = vblank_timing->v_total * vblank_timing->h_total /
973 (double)(vblank_timing->pix_clk_100hz * 100) * 1000000;
974 vblank_blank_us = (vblank_timing->v_total - vblank_timing->v_addressable) * vblank_timing->h_total /
975 (double)(vblank_timing->pix_clk_100hz * 100) * 1000000;
976 subvp_active_us = main_timing->v_addressable * main_timing->h_total /
977 (double)(main_timing->pix_clk_100hz * 100) * 1000000;
978 max_vblank_mallregion = vblank_blank_us > mall_region_us ? vblank_blank_us : mall_region_us;
980 // Schedulable if VACTIVE region of the SubVP pipe can fit the MALL prefetch, VBLANK frame time,
981 // and the max of (VBLANK blanking time, MALL region)
982 // TODO: Possibly add some margin (i.e. the below conditions should be [...] > X instead of [...] > 0)
983 if (subvp_active_us - prefetch_us - vblank_frame_us - max_vblank_mallregion > 0)
990 * subvp_validate_static_schedulability - Check which SubVP case is calculated
991 * and handle static analysis based on the case.
992 * @dc: current dc state
993 * @context: new dc state
994 * @vlevel: Voltage level calculated by DML
998 * 2. SubVP + VBLANK (DRR checked internally)
999 * 3. SubVP + VACTIVE (currently unsupported)
1001 * Return: True if statically schedulable, false otherwise
1003 static bool subvp_validate_static_schedulability(struct dc *dc,
1004 struct dc_state *context,
1007 bool schedulable = true; // true by default for single display case
1008 struct vba_vars_st *vba = &context->bw_ctx.dml.vba;
1009 uint32_t i, pipe_idx;
1010 uint8_t subvp_count = 0;
1011 uint8_t vactive_count = 0;
1013 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
1014 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
1019 if (pipe->plane_state && !pipe->top_pipe &&
1020 pipe->stream->mall_stream_config.type == SUBVP_MAIN)
1023 // Count how many planes that aren't SubVP/phantom are capable of VACTIVE
1024 // switching (SubVP + VACTIVE unsupported). In situations where we force
1025 // SubVP for a VACTIVE plane, we don't want to increment the vactive_count.
1026 if (vba->ActiveDRAMClockChangeLatencyMargin[vba->pipe_plane[pipe_idx]] > 0 &&
1027 pipe->stream->mall_stream_config.type == SUBVP_NONE) {
1033 if (subvp_count == 2) {
1034 // Static schedulability check for SubVP + SubVP case
1035 schedulable = subvp_subvp_schedulable(dc, context);
1036 } else if (vba->DRAMClockChangeSupport[vlevel][vba->maxMpcComb] == dm_dram_clock_change_vblank_w_mall_sub_vp) {
1037 // Static schedulability check for SubVP + VBLANK case. Also handle the case where
1038 // DML outputs SubVP + VBLANK + VACTIVE (DML will report as SubVP + VBLANK)
1039 if (vactive_count > 0)
1040 schedulable = false;
1042 schedulable = subvp_vblank_schedulable(dc, context);
1043 } else if (vba->DRAMClockChangeSupport[vlevel][vba->maxMpcComb] == dm_dram_clock_change_vactive_w_mall_sub_vp &&
1044 vactive_count > 0) {
1045 // For single display SubVP cases, DML will output dm_dram_clock_change_vactive_w_mall_sub_vp by default.
1046 // We tell the difference between SubVP vs. SubVP + VACTIVE by checking the vactive_count.
1047 // SubVP + VACTIVE currently unsupported
1048 schedulable = false;
1053 static void dcn32_full_validate_bw_helper(struct dc *dc,
1054 struct dc_state *context,
1055 display_e2e_pipe_params_st *pipes,
1061 struct vba_vars_st *vba = &context->bw_ctx.dml.vba;
1062 unsigned int dc_pipe_idx = 0;
1063 bool found_supported_config = false;
1064 struct pipe_ctx *pipe = NULL;
1065 uint32_t non_subvp_pipes = 0;
1066 bool drr_pipe_found = false;
1067 uint32_t drr_pipe_index = 0;
1070 dc_assert_fp_enabled();
1073 * DML favors voltage over p-state, but we're more interested in
1074 * supporting p-state over voltage. We can't support p-state in
1075 * prefetch mode > 0 so try capping the prefetch mode to start.
1076 * Override present for testing.
1078 if (dc->debug.dml_disallow_alternate_prefetch_modes)
1079 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final =
1080 dm_prefetch_support_uclk_fclk_and_stutter;
1082 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final =
1083 dm_prefetch_support_uclk_fclk_and_stutter_if_possible;
1085 *vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, *pipe_cnt);
1086 /* This may adjust vlevel and maxMpcComb */
1087 if (*vlevel < context->bw_ctx.dml.soc.num_states) {
1088 *vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, *vlevel, split, merge);
1089 vba->VoltageLevel = *vlevel;
1092 /* Conditions for setting up phantom pipes for SubVP:
1093 * 1. Not force disable SubVP
1094 * 2. Full update (i.e. !fast_validate)
1095 * 3. Enough pipes are available to support SubVP (TODO: Which pipes will use VACTIVE / VBLANK / SUBVP?)
1096 * 4. Display configuration passes validation
1097 * 5. (Config doesn't support MCLK in VACTIVE/VBLANK || dc->debug.force_subvp_mclk_switch)
1099 if (!dc->debug.force_disable_subvp && dcn32_all_pipes_have_stream_and_plane(dc, context) &&
1100 !dcn32_mpo_in_use(context) && !dcn32_any_surfaces_rotated(dc, context) &&
1101 (*vlevel == context->bw_ctx.dml.soc.num_states ||
1102 vba->DRAMClockChangeSupport[*vlevel][vba->maxMpcComb] == dm_dram_clock_change_unsupported ||
1103 dc->debug.force_subvp_mclk_switch)) {
1105 dcn32_merge_pipes_for_subvp(dc, context);
1106 memset(merge, 0, MAX_PIPES * sizeof(bool));
1108 /* to re-initialize viewport after the pipe merge */
1109 for (i = 0; i < dc->res_pool->pipe_count; i++) {
1110 struct pipe_ctx *pipe_ctx = &context->res_ctx.pipe_ctx[i];
1112 if (!pipe_ctx->plane_state || !pipe_ctx->stream)
1115 resource_build_scaling_params(pipe_ctx);
1118 while (!found_supported_config && dcn32_enough_pipes_for_subvp(dc, context) &&
1119 dcn32_assign_subvp_pipe(dc, context, &dc_pipe_idx)) {
1120 /* For the case where *vlevel = num_states, bandwidth validation has failed for this config.
1121 * Adding phantom pipes won't change the validation result, so change the DML input param
1122 * for P-State support before adding phantom pipes and recalculating the DML result.
1123 * However, this case is only applicable for SubVP + DRR cases because the prefetch mode
1124 * will not allow for switch in VBLANK. The DRR display must have it's VBLANK stretched
1125 * enough to support MCLK switching.
1127 if (*vlevel == context->bw_ctx.dml.soc.num_states &&
1128 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final ==
1129 dm_prefetch_support_uclk_fclk_and_stutter) {
1130 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final =
1131 dm_prefetch_support_stutter;
1132 /* There are params (such as FabricClock) that need to be recalculated
1133 * after validation fails (otherwise it will be 0). Calculation for
1134 * phantom vactive requires call into DML, so we must ensure all the
1135 * vba params are valid otherwise we'll get incorrect phantom vactive.
1137 *vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, *pipe_cnt);
1140 dc->res_pool->funcs->add_phantom_pipes(dc, context, pipes, *pipe_cnt, dc_pipe_idx);
1142 *pipe_cnt = dc->res_pool->funcs->populate_dml_pipes(dc, context, pipes, false);
1143 // Populate dppclk to trigger a recalculate in dml_get_voltage_level
1144 // so the phantom pipe DLG params can be assigned correctly.
1145 pipes[0].clks_cfg.dppclk_mhz = get_dppclk_calculated(&context->bw_ctx.dml, pipes, *pipe_cnt, 0);
1146 *vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, *pipe_cnt);
1148 if (*vlevel < context->bw_ctx.dml.soc.num_states &&
1149 vba->DRAMClockChangeSupport[*vlevel][vba->maxMpcComb] != dm_dram_clock_change_unsupported
1150 && subvp_validate_static_schedulability(dc, context, *vlevel)) {
1151 found_supported_config = true;
1152 } else if (*vlevel < context->bw_ctx.dml.soc.num_states &&
1153 vba->DRAMClockChangeSupport[*vlevel][vba->maxMpcComb] == dm_dram_clock_change_unsupported) {
1154 /* Case where 1 SubVP is added, and DML reports MCLK unsupported. This handles
1155 * the case for SubVP + DRR, where the DRR display does not support MCLK switch
1156 * at it's native refresh rate / timing.
1158 for (i = 0; i < dc->res_pool->pipe_count; i++) {
1159 pipe = &context->res_ctx.pipe_ctx[i];
1160 if (pipe->stream && pipe->plane_state && !pipe->top_pipe &&
1161 pipe->stream->mall_stream_config.type == SUBVP_NONE) {
1163 // Use ignore_msa_timing_param flag to identify as DRR
1164 if (pipe->stream->ignore_msa_timing_param) {
1165 drr_pipe_found = true;
1170 // If there is only 1 remaining non SubVP pipe that is DRR, check static
1171 // schedulability for SubVP + DRR.
1172 if (non_subvp_pipes == 1 && drr_pipe_found) {
1173 found_supported_config = subvp_drr_schedulable(dc, context,
1174 &context->res_ctx.pipe_ctx[drr_pipe_index]);
1179 // If SubVP pipe config is unsupported (or cannot be used for UCLK switching)
1180 // remove phantom pipes and repopulate dml pipes
1181 if (!found_supported_config) {
1182 dc->res_pool->funcs->remove_phantom_pipes(dc, context);
1183 vba->DRAMClockChangeSupport[*vlevel][vba->maxMpcComb] = dm_dram_clock_change_unsupported;
1184 *pipe_cnt = dc->res_pool->funcs->populate_dml_pipes(dc, context, pipes, false);
1186 *vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, *pipe_cnt);
1187 /* This may adjust vlevel and maxMpcComb */
1188 if (*vlevel < context->bw_ctx.dml.soc.num_states) {
1189 *vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, *vlevel, split, merge);
1190 vba->VoltageLevel = *vlevel;
1193 // Most populate phantom DLG params before programming hardware / timing for phantom pipe
1195 dcn32_helper_populate_phantom_dlg_params(dc, context, pipes, *pipe_cnt);
1198 /* Call validate_apply_pipe_split flags after calling DML getters for
1199 * phantom dlg params, or some of the VBA params indicating pipe split
1200 * can be overwritten by the getters.
1202 * When setting up SubVP config, all pipes are merged before attempting to
1203 * add phantom pipes. If pipe split (ODM / MPC) is required, both the main
1204 * and phantom pipes will be split in the regular pipe splitting sequence.
1206 memset(split, 0, MAX_PIPES * sizeof(int));
1207 memset(merge, 0, MAX_PIPES * sizeof(bool));
1208 *vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, *vlevel, split, merge);
1209 vba->VoltageLevel = *vlevel;
1210 // Note: We can't apply the phantom pipes to hardware at this time. We have to wait
1211 // until driver has acquired the DMCUB lock to do it safely.
1216 static bool is_dtbclk_required(struct dc *dc, struct dc_state *context)
1220 for (i = 0; i < dc->res_pool->pipe_count; i++) {
1221 if (!context->res_ctx.pipe_ctx[i].stream)
1223 if (is_dp_128b_132b_signal(&context->res_ctx.pipe_ctx[i]))
1229 static void dcn32_calculate_dlg_params(struct dc *dc, struct dc_state *context,
1230 display_e2e_pipe_params_st *pipes,
1231 int pipe_cnt, int vlevel)
1234 bool usr_retraining_support = false;
1235 bool unbounded_req_enabled = false;
1237 dc_assert_fp_enabled();
1239 /* Writeback MCIF_WB arbitration parameters */
1240 dc->res_pool->funcs->set_mcif_arb_params(dc, context, pipes, pipe_cnt);
1242 context->bw_ctx.bw.dcn.clk.dispclk_khz = context->bw_ctx.dml.vba.DISPCLK * 1000;
1243 context->bw_ctx.bw.dcn.clk.dcfclk_khz = context->bw_ctx.dml.vba.DCFCLK * 1000;
1244 context->bw_ctx.bw.dcn.clk.socclk_khz = context->bw_ctx.dml.vba.SOCCLK * 1000;
1245 context->bw_ctx.bw.dcn.clk.dramclk_khz = context->bw_ctx.dml.vba.DRAMSpeed * 1000 / 16;
1246 context->bw_ctx.bw.dcn.clk.dcfclk_deep_sleep_khz = context->bw_ctx.dml.vba.DCFCLKDeepSleep * 1000;
1247 context->bw_ctx.bw.dcn.clk.fclk_khz = context->bw_ctx.dml.vba.FabricClock * 1000;
1248 context->bw_ctx.bw.dcn.clk.p_state_change_support =
1249 context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb]
1250 != dm_dram_clock_change_unsupported;
1251 context->bw_ctx.bw.dcn.clk.num_ways = dcn32_helper_calculate_num_ways_for_subvp(dc, context);
1253 context->bw_ctx.bw.dcn.clk.dppclk_khz = 0;
1254 context->bw_ctx.bw.dcn.clk.dtbclk_en = is_dtbclk_required(dc, context);
1255 context->bw_ctx.bw.dcn.clk.ref_dtbclk_khz = context->bw_ctx.dml.vba.DTBCLKPerState[vlevel] * 1000;
1256 if (context->bw_ctx.dml.vba.FCLKChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] == dm_fclock_change_unsupported)
1257 context->bw_ctx.bw.dcn.clk.fclk_p_state_change_support = false;
1259 context->bw_ctx.bw.dcn.clk.fclk_p_state_change_support = true;
1261 usr_retraining_support = context->bw_ctx.dml.vba.USRRetrainingSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb];
1262 ASSERT(usr_retraining_support);
1264 if (context->bw_ctx.bw.dcn.clk.dispclk_khz < dc->debug.min_disp_clk_khz)
1265 context->bw_ctx.bw.dcn.clk.dispclk_khz = dc->debug.min_disp_clk_khz;
1267 unbounded_req_enabled = get_unbounded_request_enabled(&context->bw_ctx.dml, pipes, pipe_cnt);
1269 if (unbounded_req_enabled && pipe_cnt > 1) {
1270 // Unbounded requesting should not ever be used when more than 1 pipe is enabled.
1272 unbounded_req_enabled = false;
1275 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
1276 if (!context->res_ctx.pipe_ctx[i].stream)
1278 pipes[pipe_idx].pipe.dest.vstartup_start = get_vstartup(&context->bw_ctx.dml, pipes, pipe_cnt,
1280 pipes[pipe_idx].pipe.dest.vupdate_offset = get_vupdate_offset(&context->bw_ctx.dml, pipes, pipe_cnt,
1282 pipes[pipe_idx].pipe.dest.vupdate_width = get_vupdate_width(&context->bw_ctx.dml, pipes, pipe_cnt,
1284 pipes[pipe_idx].pipe.dest.vready_offset = get_vready_offset(&context->bw_ctx.dml, pipes, pipe_cnt,
1287 if (context->res_ctx.pipe_ctx[i].stream->mall_stream_config.type == SUBVP_PHANTOM) {
1288 // Phantom pipe requires that DET_SIZE = 0 and no unbounded requests
1289 context->res_ctx.pipe_ctx[i].det_buffer_size_kb = 0;
1290 context->res_ctx.pipe_ctx[i].unbounded_req = false;
1292 context->res_ctx.pipe_ctx[i].det_buffer_size_kb = get_det_buffer_size_kbytes(&context->bw_ctx.dml, pipes, pipe_cnt,
1294 context->res_ctx.pipe_ctx[i].unbounded_req = unbounded_req_enabled;
1297 if (context->bw_ctx.bw.dcn.clk.dppclk_khz < pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000)
1298 context->bw_ctx.bw.dcn.clk.dppclk_khz = pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000;
1299 context->res_ctx.pipe_ctx[i].plane_res.bw.dppclk_khz = pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000;
1300 context->res_ctx.pipe_ctx[i].pipe_dlg_param = pipes[pipe_idx].pipe.dest;
1303 /*save a original dppclock copy*/
1304 context->bw_ctx.bw.dcn.clk.bw_dppclk_khz = context->bw_ctx.bw.dcn.clk.dppclk_khz;
1305 context->bw_ctx.bw.dcn.clk.bw_dispclk_khz = context->bw_ctx.bw.dcn.clk.dispclk_khz;
1306 context->bw_ctx.bw.dcn.clk.max_supported_dppclk_khz = context->bw_ctx.dml.soc.clock_limits[vlevel].dppclk_mhz
1308 context->bw_ctx.bw.dcn.clk.max_supported_dispclk_khz = context->bw_ctx.dml.soc.clock_limits[vlevel].dispclk_mhz
1311 context->bw_ctx.bw.dcn.compbuf_size_kb = context->bw_ctx.dml.ip.config_return_buffer_size_in_kbytes;
1313 for (i = 0; i < dc->res_pool->pipe_count; i++) {
1314 if (context->res_ctx.pipe_ctx[i].stream)
1315 context->bw_ctx.bw.dcn.compbuf_size_kb -= context->res_ctx.pipe_ctx[i].det_buffer_size_kb;
1318 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
1320 if (!context->res_ctx.pipe_ctx[i].stream)
1323 context->bw_ctx.dml.funcs.rq_dlg_get_dlg_reg_v2(&context->bw_ctx.dml,
1324 &context->res_ctx.pipe_ctx[i].dlg_regs, &context->res_ctx.pipe_ctx[i].ttu_regs, pipes,
1325 pipe_cnt, pipe_idx);
1327 context->bw_ctx.dml.funcs.rq_dlg_get_rq_reg_v2(&context->res_ctx.pipe_ctx[i].rq_regs,
1328 &context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx);
1333 static struct pipe_ctx *dcn32_find_split_pipe(
1335 struct dc_state *context,
1338 struct pipe_ctx *pipe = NULL;
1341 if (old_index >= 0 && context->res_ctx.pipe_ctx[old_index].stream == NULL) {
1342 pipe = &context->res_ctx.pipe_ctx[old_index];
1343 pipe->pipe_idx = old_index;
1347 for (i = dc->res_pool->pipe_count - 1; i >= 0; i--) {
1348 if (dc->current_state->res_ctx.pipe_ctx[i].top_pipe == NULL
1349 && dc->current_state->res_ctx.pipe_ctx[i].prev_odm_pipe == NULL) {
1350 if (context->res_ctx.pipe_ctx[i].stream == NULL) {
1351 pipe = &context->res_ctx.pipe_ctx[i];
1359 * May need to fix pipes getting tossed from 1 opp to another on flip
1360 * Add for debugging transient underflow during topology updates:
1364 for (i = dc->res_pool->pipe_count - 1; i >= 0; i--) {
1365 if (context->res_ctx.pipe_ctx[i].stream == NULL) {
1366 pipe = &context->res_ctx.pipe_ctx[i];
1375 static bool dcn32_split_stream_for_mpc_or_odm(
1376 const struct dc *dc,
1377 struct resource_context *res_ctx,
1378 struct pipe_ctx *pri_pipe,
1379 struct pipe_ctx *sec_pipe,
1382 int pipe_idx = sec_pipe->pipe_idx;
1383 const struct resource_pool *pool = dc->res_pool;
1385 DC_LOGGER_INIT(dc->ctx->logger);
1387 if (odm && pri_pipe->plane_state) {
1388 /* ODM + window MPO, where MPO window is on left half only */
1389 if (pri_pipe->plane_state->clip_rect.x + pri_pipe->plane_state->clip_rect.width <=
1390 pri_pipe->stream->src.x + pri_pipe->stream->src.width/2) {
1392 DC_LOG_SCALER("%s - ODM + window MPO(left). pri_pipe:%d\n",
1394 pri_pipe->pipe_idx);
1398 /* ODM + window MPO, where MPO window is on right half only */
1399 if (pri_pipe->plane_state->clip_rect.x >= pri_pipe->stream->src.x + pri_pipe->stream->src.width/2) {
1401 DC_LOG_SCALER("%s - ODM + window MPO(right). pri_pipe:%d\n",
1403 pri_pipe->pipe_idx);
1408 *sec_pipe = *pri_pipe;
1410 sec_pipe->pipe_idx = pipe_idx;
1411 sec_pipe->plane_res.mi = pool->mis[pipe_idx];
1412 sec_pipe->plane_res.hubp = pool->hubps[pipe_idx];
1413 sec_pipe->plane_res.ipp = pool->ipps[pipe_idx];
1414 sec_pipe->plane_res.xfm = pool->transforms[pipe_idx];
1415 sec_pipe->plane_res.dpp = pool->dpps[pipe_idx];
1416 sec_pipe->plane_res.mpcc_inst = pool->dpps[pipe_idx]->inst;
1417 sec_pipe->stream_res.dsc = NULL;
1419 if (pri_pipe->next_odm_pipe) {
1420 ASSERT(pri_pipe->next_odm_pipe != sec_pipe);
1421 sec_pipe->next_odm_pipe = pri_pipe->next_odm_pipe;
1422 sec_pipe->next_odm_pipe->prev_odm_pipe = sec_pipe;
1424 if (pri_pipe->top_pipe && pri_pipe->top_pipe->next_odm_pipe) {
1425 pri_pipe->top_pipe->next_odm_pipe->bottom_pipe = sec_pipe;
1426 sec_pipe->top_pipe = pri_pipe->top_pipe->next_odm_pipe;
1428 if (pri_pipe->bottom_pipe && pri_pipe->bottom_pipe->next_odm_pipe) {
1429 pri_pipe->bottom_pipe->next_odm_pipe->top_pipe = sec_pipe;
1430 sec_pipe->bottom_pipe = pri_pipe->bottom_pipe->next_odm_pipe;
1432 pri_pipe->next_odm_pipe = sec_pipe;
1433 sec_pipe->prev_odm_pipe = pri_pipe;
1434 ASSERT(sec_pipe->top_pipe == NULL);
1436 if (!sec_pipe->top_pipe)
1437 sec_pipe->stream_res.opp = pool->opps[pipe_idx];
1439 sec_pipe->stream_res.opp = sec_pipe->top_pipe->stream_res.opp;
1440 if (sec_pipe->stream->timing.flags.DSC == 1) {
1441 dcn20_acquire_dsc(dc, res_ctx, &sec_pipe->stream_res.dsc, pipe_idx);
1442 ASSERT(sec_pipe->stream_res.dsc);
1443 if (sec_pipe->stream_res.dsc == NULL)
1447 if (pri_pipe->bottom_pipe) {
1448 ASSERT(pri_pipe->bottom_pipe != sec_pipe);
1449 sec_pipe->bottom_pipe = pri_pipe->bottom_pipe;
1450 sec_pipe->bottom_pipe->top_pipe = sec_pipe;
1452 pri_pipe->bottom_pipe = sec_pipe;
1453 sec_pipe->top_pipe = pri_pipe;
1455 ASSERT(pri_pipe->plane_state);
1461 bool dcn32_internal_validate_bw(struct dc *dc,
1462 struct dc_state *context,
1463 display_e2e_pipe_params_st *pipes,
1469 bool repopulate_pipes = false;
1470 int split[MAX_PIPES] = { 0 };
1471 bool merge[MAX_PIPES] = { false };
1472 bool newly_split[MAX_PIPES] = { false };
1473 int pipe_cnt, i, pipe_idx;
1474 int vlevel = context->bw_ctx.dml.soc.num_states;
1475 struct vba_vars_st *vba = &context->bw_ctx.dml.vba;
1477 dc_assert_fp_enabled();
1483 // For each full update, remove all existing phantom pipes first
1484 dc->res_pool->funcs->remove_phantom_pipes(dc, context);
1486 dc->res_pool->funcs->update_soc_for_wm_a(dc, context);
1488 pipe_cnt = dc->res_pool->funcs->populate_dml_pipes(dc, context, pipes, fast_validate);
1495 dml_log_pipe_params(&context->bw_ctx.dml, pipes, pipe_cnt);
1497 if (!fast_validate) {
1499 dcn32_full_validate_bw_helper(dc, context, pipes, &vlevel, split, merge, &pipe_cnt);
1503 if (fast_validate ||
1504 (dc->debug.dml_disallow_alternate_prefetch_modes &&
1505 (vlevel == context->bw_ctx.dml.soc.num_states ||
1506 vba->DRAMClockChangeSupport[vlevel][vba->maxMpcComb] == dm_dram_clock_change_unsupported))) {
1508 * If dml_disallow_alternate_prefetch_modes is false, then we have already
1509 * tried alternate prefetch modes during full validation.
1511 * If mode is unsupported or there is no p-state support, then
1512 * fall back to favouring voltage.
1514 * If Prefetch mode 0 failed for this config, or passed with Max UCLK, then try
1515 * to support with Prefetch mode 1 (dm_prefetch_support_fclk_and_stutter == 2)
1517 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final =
1518 dm_prefetch_support_fclk_and_stutter;
1520 vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, pipe_cnt);
1522 /* Last attempt with Prefetch mode 2 (dm_prefetch_support_stutter == 3) */
1523 if (vlevel == context->bw_ctx.dml.soc.num_states) {
1524 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final =
1525 dm_prefetch_support_stutter;
1526 vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, pipe_cnt);
1529 if (vlevel < context->bw_ctx.dml.soc.num_states) {
1530 memset(split, 0, sizeof(split));
1531 memset(merge, 0, sizeof(merge));
1532 vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, vlevel, split, merge);
1533 // dcn20_validate_apply_pipe_split_flags can modify voltage level outside of DML
1534 vba->VoltageLevel = vlevel;
1538 dml_log_mode_support_params(&context->bw_ctx.dml);
1540 if (vlevel == context->bw_ctx.dml.soc.num_states)
1543 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
1544 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
1545 struct pipe_ctx *mpo_pipe = pipe->bottom_pipe;
1550 if (vba->ODMCombineEnabled[vba->pipe_plane[pipe_idx]] != dm_odm_combine_mode_disabled
1551 && !dc->config.enable_windowed_mpo_odm
1552 && pipe->plane_state && mpo_pipe
1553 && memcmp(&mpo_pipe->plane_res.scl_data.recout,
1554 &pipe->plane_res.scl_data.recout,
1555 sizeof(struct rect)) != 0) {
1556 ASSERT(mpo_pipe->plane_state != pipe->plane_state);
1562 /* merge pipes if necessary */
1563 for (i = 0; i < dc->res_pool->pipe_count; i++) {
1564 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
1566 /*skip pipes that don't need merging*/
1570 /* if ODM merge we ignore mpc tree, mpo pipes will have their own flags */
1571 if (pipe->prev_odm_pipe) {
1572 /*split off odm pipe*/
1573 pipe->prev_odm_pipe->next_odm_pipe = pipe->next_odm_pipe;
1574 if (pipe->next_odm_pipe)
1575 pipe->next_odm_pipe->prev_odm_pipe = pipe->prev_odm_pipe;
1577 /*2:1ODM+MPC Split MPO to Single Pipe + MPC Split MPO*/
1578 if (pipe->bottom_pipe) {
1579 if (pipe->bottom_pipe->prev_odm_pipe || pipe->bottom_pipe->next_odm_pipe) {
1580 /*MPC split rules will handle this case*/
1581 pipe->bottom_pipe->top_pipe = NULL;
1583 /* when merging an ODM pipes, the bottom MPC pipe must now point to
1584 * the previous ODM pipe and its associated stream assets
1586 if (pipe->prev_odm_pipe->bottom_pipe) {
1588 pipe->bottom_pipe->top_pipe = pipe->prev_odm_pipe->bottom_pipe;
1589 pipe->prev_odm_pipe->bottom_pipe->bottom_pipe = pipe->bottom_pipe;
1592 pipe->bottom_pipe->top_pipe = pipe->prev_odm_pipe;
1593 pipe->prev_odm_pipe->bottom_pipe = pipe->bottom_pipe;
1596 memcpy(&pipe->bottom_pipe->stream_res, &pipe->bottom_pipe->top_pipe->stream_res, sizeof(struct stream_resource));
1600 if (pipe->top_pipe) {
1601 pipe->top_pipe->bottom_pipe = NULL;
1604 pipe->bottom_pipe = NULL;
1605 pipe->next_odm_pipe = NULL;
1606 pipe->plane_state = NULL;
1607 pipe->stream = NULL;
1608 pipe->top_pipe = NULL;
1609 pipe->prev_odm_pipe = NULL;
1610 if (pipe->stream_res.dsc)
1611 dcn20_release_dsc(&context->res_ctx, dc->res_pool, &pipe->stream_res.dsc);
1612 memset(&pipe->plane_res, 0, sizeof(pipe->plane_res));
1613 memset(&pipe->stream_res, 0, sizeof(pipe->stream_res));
1614 repopulate_pipes = true;
1615 } else if (pipe->top_pipe && pipe->top_pipe->plane_state == pipe->plane_state) {
1616 struct pipe_ctx *top_pipe = pipe->top_pipe;
1617 struct pipe_ctx *bottom_pipe = pipe->bottom_pipe;
1619 top_pipe->bottom_pipe = bottom_pipe;
1621 bottom_pipe->top_pipe = top_pipe;
1623 pipe->top_pipe = NULL;
1624 pipe->bottom_pipe = NULL;
1625 pipe->plane_state = NULL;
1626 pipe->stream = NULL;
1627 memset(&pipe->plane_res, 0, sizeof(pipe->plane_res));
1628 memset(&pipe->stream_res, 0, sizeof(pipe->stream_res));
1629 repopulate_pipes = true;
1631 ASSERT(0); /* Should never try to merge master pipe */
1635 for (i = 0, pipe_idx = -1; i < dc->res_pool->pipe_count; i++) {
1636 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
1637 struct pipe_ctx *old_pipe = &dc->current_state->res_ctx.pipe_ctx[i];
1638 struct pipe_ctx *hsplit_pipe = NULL;
1642 if (!pipe->stream || newly_split[i])
1646 odm = vba->ODMCombineEnabled[vba->pipe_plane[pipe_idx]] != dm_odm_combine_mode_disabled;
1648 if (!pipe->plane_state && !odm)
1653 if (split[i] == 4 && old_pipe->next_odm_pipe && old_pipe->next_odm_pipe->next_odm_pipe)
1654 old_index = old_pipe->next_odm_pipe->next_odm_pipe->pipe_idx;
1655 else if (old_pipe->next_odm_pipe)
1656 old_index = old_pipe->next_odm_pipe->pipe_idx;
1658 if (split[i] == 4 && old_pipe->bottom_pipe && old_pipe->bottom_pipe->bottom_pipe &&
1659 old_pipe->bottom_pipe->bottom_pipe->plane_state == old_pipe->plane_state)
1660 old_index = old_pipe->bottom_pipe->bottom_pipe->pipe_idx;
1661 else if (old_pipe->bottom_pipe &&
1662 old_pipe->bottom_pipe->plane_state == old_pipe->plane_state)
1663 old_index = old_pipe->bottom_pipe->pipe_idx;
1665 hsplit_pipe = dcn32_find_split_pipe(dc, context, old_index);
1666 ASSERT(hsplit_pipe);
1670 if (!dcn32_split_stream_for_mpc_or_odm(
1671 dc, &context->res_ctx,
1672 pipe, hsplit_pipe, odm))
1675 newly_split[hsplit_pipe->pipe_idx] = true;
1676 repopulate_pipes = true;
1678 if (split[i] == 4) {
1679 struct pipe_ctx *pipe_4to1;
1681 if (odm && old_pipe->next_odm_pipe)
1682 old_index = old_pipe->next_odm_pipe->pipe_idx;
1683 else if (!odm && old_pipe->bottom_pipe &&
1684 old_pipe->bottom_pipe->plane_state == old_pipe->plane_state)
1685 old_index = old_pipe->bottom_pipe->pipe_idx;
1688 pipe_4to1 = dcn32_find_split_pipe(dc, context, old_index);
1692 if (!dcn32_split_stream_for_mpc_or_odm(
1693 dc, &context->res_ctx,
1694 pipe, pipe_4to1, odm))
1696 newly_split[pipe_4to1->pipe_idx] = true;
1698 if (odm && old_pipe->next_odm_pipe && old_pipe->next_odm_pipe->next_odm_pipe
1699 && old_pipe->next_odm_pipe->next_odm_pipe->next_odm_pipe)
1700 old_index = old_pipe->next_odm_pipe->next_odm_pipe->next_odm_pipe->pipe_idx;
1701 else if (!odm && old_pipe->bottom_pipe && old_pipe->bottom_pipe->bottom_pipe &&
1702 old_pipe->bottom_pipe->bottom_pipe->bottom_pipe &&
1703 old_pipe->bottom_pipe->bottom_pipe->bottom_pipe->plane_state == old_pipe->plane_state)
1704 old_index = old_pipe->bottom_pipe->bottom_pipe->bottom_pipe->pipe_idx;
1707 pipe_4to1 = dcn32_find_split_pipe(dc, context, old_index);
1711 if (!dcn32_split_stream_for_mpc_or_odm(
1712 dc, &context->res_ctx,
1713 hsplit_pipe, pipe_4to1, odm))
1715 newly_split[pipe_4to1->pipe_idx] = true;
1718 dcn20_build_mapped_resource(dc, context, pipe->stream);
1721 for (i = 0; i < dc->res_pool->pipe_count; i++) {
1722 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
1724 if (pipe->plane_state) {
1725 if (!resource_build_scaling_params(pipe))
1730 /* Actual dsc count per stream dsc validation*/
1731 if (!dcn20_validate_dsc(dc, context)) {
1732 vba->ValidationStatus[vba->soc.num_states] = DML_FAIL_DSC_VALIDATION_FAILURE;
1736 if (repopulate_pipes) {
1737 pipe_cnt = dc->res_pool->funcs->populate_dml_pipes(dc, context, pipes, fast_validate);
1739 /* repopulate_pipes = 1 means the pipes were either split or merged. In this case
1740 * we have to re-calculate the DET allocation and run through DML once more to
1741 * ensure all the params are calculated correctly. We do not need to run the
1742 * pipe split check again after this call (pipes are already split / merged).
1744 if (!fast_validate) {
1745 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final =
1746 dm_prefetch_support_uclk_fclk_and_stutter_if_possible;
1747 vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, pipe_cnt);
1750 *vlevel_out = vlevel;
1751 *pipe_cnt_out = pipe_cnt;
1764 void dcn32_calculate_wm_and_dlg_fpu(struct dc *dc, struct dc_state *context,
1765 display_e2e_pipe_params_st *pipes,
1769 int i, pipe_idx, vlevel_temp = 0;
1770 double dcfclk = dcn3_2_soc.clock_limits[0].dcfclk_mhz;
1771 double dcfclk_from_validation = context->bw_ctx.dml.vba.DCFCLKState[vlevel][context->bw_ctx.dml.vba.maxMpcComb];
1772 double dcfclk_from_fw_based_mclk_switching = dcfclk_from_validation;
1773 bool pstate_en = context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] !=
1774 dm_dram_clock_change_unsupported;
1775 unsigned int dummy_latency_index = 0;
1776 int maxMpcComb = context->bw_ctx.dml.vba.maxMpcComb;
1777 unsigned int min_dram_speed_mts = context->bw_ctx.dml.vba.DRAMSpeed;
1778 unsigned int min_dram_speed_mts_margin;
1780 dc_assert_fp_enabled();
1782 // Override DRAMClockChangeSupport for SubVP + DRR case where the DRR cannot switch without stretching it's VBLANK
1783 if (!pstate_en && dcn32_subvp_in_use(dc, context)) {
1784 context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] = dm_dram_clock_change_vblank_w_mall_sub_vp;
1788 context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching = false;
1791 /* only when the mclk switch can not be natural, is the fw based vblank stretch attempted */
1792 context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching =
1793 dcn30_can_support_mclk_switch_using_fw_based_vblank_stretch(dc, context);
1795 if (context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching) {
1796 dummy_latency_index = dcn32_find_dummy_latency_index_for_fw_based_mclk_switch(dc,
1797 context, pipes, pipe_cnt, vlevel);
1799 /* After calling dcn30_find_dummy_latency_index_for_fw_based_mclk_switch
1800 * we reinstate the original dram_clock_change_latency_us on the context
1801 * and all variables that may have changed up to this point, except the
1802 * newly found dummy_latency_index
1804 context->bw_ctx.dml.soc.dram_clock_change_latency_us =
1805 dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us;
1806 /* For DCN32/321 need to validate with fclk pstate change latency equal to dummy so
1807 * prefetch is scheduled correctly to account for dummy pstate.
1809 if (dummy_latency_index == 0)
1810 context->bw_ctx.dml.soc.fclk_change_latency_us =
1811 dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us;
1812 dcn32_internal_validate_bw(dc, context, pipes, &pipe_cnt, &vlevel, false);
1813 maxMpcComb = context->bw_ctx.dml.vba.maxMpcComb;
1814 dcfclk_from_fw_based_mclk_switching = context->bw_ctx.dml.vba.DCFCLKState[vlevel][context->bw_ctx.dml.vba.maxMpcComb];
1815 pstate_en = context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][maxMpcComb] !=
1816 dm_dram_clock_change_unsupported;
1821 * For Set B calculations use clocks from clock_limits[2] when available i.e. when SMU is present,
1822 * otherwise use arbitrary low value from spreadsheet for DCFCLK as lower is safer for watermark
1823 * calculations to cover bootup clocks.
1824 * DCFCLK: soc.clock_limits[2] when available
1825 * UCLK: soc.clock_limits[2] when available
1827 if (dcn3_2_soc.num_states > 2) {
1829 dcfclk = dcn3_2_soc.clock_limits[2].dcfclk_mhz;
1831 dcfclk = 615; //DCFCLK Vmin_lv
1833 pipes[0].clks_cfg.voltage = vlevel_temp;
1834 pipes[0].clks_cfg.dcfclk_mhz = dcfclk;
1835 pipes[0].clks_cfg.socclk_mhz = context->bw_ctx.dml.soc.clock_limits[vlevel_temp].socclk_mhz;
1837 if (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_B].valid) {
1838 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;
1839 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;
1840 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;
1841 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;
1843 context->bw_ctx.bw.dcn.watermarks.b.urgent_ns = get_wm_urgent(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1844 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;
1845 context->bw_ctx.bw.dcn.watermarks.b.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1846 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;
1847 context->bw_ctx.bw.dcn.watermarks.b.pte_meta_urgent_ns = get_wm_memory_trip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1848 context->bw_ctx.bw.dcn.watermarks.b.frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1849 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;
1850 context->bw_ctx.bw.dcn.watermarks.b.urgent_latency_ns = get_urgent_latency(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1851 context->bw_ctx.bw.dcn.watermarks.b.cstate_pstate.fclk_pstate_change_ns = get_fclk_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1852 context->bw_ctx.bw.dcn.watermarks.b.usr_retraining_ns = get_usr_retraining_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1856 * DCFCLK: Min, as reported by PM FW when available
1857 * UCLK : Min, as reported by PM FW when available
1858 * sr_enter_exit/sr_exit should be lower than used for DRAM (TBD after bringup or later, use as decided in Clk Mgr)
1861 if (dcn3_2_soc.num_states > 2) {
1863 dcfclk = dc->clk_mgr->bw_params->clk_table.entries[0].dcfclk_mhz;
1865 dcfclk = 615; //DCFCLK Vmin_lv
1867 pipes[0].clks_cfg.voltage = vlevel_temp;
1868 pipes[0].clks_cfg.dcfclk_mhz = dcfclk;
1869 pipes[0].clks_cfg.socclk_mhz = context->bw_ctx.dml.soc.clock_limits[vlevel_temp].socclk_mhz;
1871 if (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_D].valid) {
1872 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;
1873 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;
1874 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;
1875 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;
1877 context->bw_ctx.bw.dcn.watermarks.d.urgent_ns = get_wm_urgent(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1878 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;
1879 context->bw_ctx.bw.dcn.watermarks.d.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1880 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;
1881 context->bw_ctx.bw.dcn.watermarks.d.pte_meta_urgent_ns = get_wm_memory_trip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1882 context->bw_ctx.bw.dcn.watermarks.d.frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1883 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;
1884 context->bw_ctx.bw.dcn.watermarks.d.urgent_latency_ns = get_urgent_latency(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1885 context->bw_ctx.bw.dcn.watermarks.d.cstate_pstate.fclk_pstate_change_ns = get_fclk_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1886 context->bw_ctx.bw.dcn.watermarks.d.usr_retraining_ns = get_usr_retraining_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1888 /* Set C, for Dummy P-State:
1890 * DCFCLK: Min, as reported by PM FW, when available
1891 * UCLK : Min, as reported by PM FW, when available
1892 * pstate latency as per UCLK state dummy pstate latency
1895 // For Set A and Set C use values from validation
1896 pipes[0].clks_cfg.voltage = vlevel;
1897 pipes[0].clks_cfg.dcfclk_mhz = dcfclk_from_validation;
1898 pipes[0].clks_cfg.socclk_mhz = context->bw_ctx.dml.soc.clock_limits[vlevel].socclk_mhz;
1900 if (context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching) {
1901 pipes[0].clks_cfg.dcfclk_mhz = dcfclk_from_fw_based_mclk_switching;
1904 if (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_C].valid) {
1905 min_dram_speed_mts = context->bw_ctx.dml.vba.DRAMSpeed;
1906 min_dram_speed_mts_margin = 160;
1908 context->bw_ctx.dml.soc.dram_clock_change_latency_us =
1909 dc->clk_mgr->bw_params->dummy_pstate_table[0].dummy_pstate_latency_us;
1911 if (context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][maxMpcComb] ==
1912 dm_dram_clock_change_unsupported) {
1913 int min_dram_speed_mts_offset = dc->clk_mgr->bw_params->clk_table.num_entries - 1;
1915 min_dram_speed_mts =
1916 dc->clk_mgr->bw_params->clk_table.entries[min_dram_speed_mts_offset].memclk_mhz * 16;
1919 if (!context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching) {
1920 /* find largest table entry that is lower than dram speed,
1921 * but lower than DPM0 still uses DPM0
1923 for (dummy_latency_index = 3; dummy_latency_index > 0; dummy_latency_index--)
1924 if (min_dram_speed_mts + min_dram_speed_mts_margin >
1925 dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dram_speed_mts)
1929 context->bw_ctx.dml.soc.dram_clock_change_latency_us =
1930 dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us;
1932 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;
1933 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;
1934 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;
1937 context->bw_ctx.bw.dcn.watermarks.c.urgent_ns = get_wm_urgent(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1938 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;
1939 context->bw_ctx.bw.dcn.watermarks.c.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1940 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;
1941 context->bw_ctx.bw.dcn.watermarks.c.pte_meta_urgent_ns = get_wm_memory_trip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1942 context->bw_ctx.bw.dcn.watermarks.c.frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1943 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;
1944 context->bw_ctx.bw.dcn.watermarks.c.urgent_latency_ns = get_urgent_latency(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1945 /* On DCN32/321, PMFW will set PSTATE_CHANGE_TYPE = 1 (FCLK) for UCLK dummy p-state.
1946 * In this case we must program FCLK WM Set C to use the UCLK dummy p-state WM
1949 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;
1950 context->bw_ctx.bw.dcn.watermarks.c.usr_retraining_ns = get_usr_retraining_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1952 if ((!pstate_en) && (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_C].valid)) {
1953 /* The only difference between A and C is p-state latency, if p-state is not supported
1954 * with full p-state latency we want to calculate DLG based on dummy p-state latency,
1955 * Set A p-state watermark set to 0 on DCN30, when p-state unsupported, for now keep as DCN30.
1957 context->bw_ctx.bw.dcn.watermarks.a = context->bw_ctx.bw.dcn.watermarks.c;
1958 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.pstate_change_ns = 0;
1962 * DCFCLK: Min, as reported by PM FW, when available
1963 * UCLK: Min, as reported by PM FW, when available
1965 dc->res_pool->funcs->update_soc_for_wm_a(dc, context);
1966 context->bw_ctx.bw.dcn.watermarks.a.urgent_ns = get_wm_urgent(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1967 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;
1968 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1969 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;
1970 context->bw_ctx.bw.dcn.watermarks.a.pte_meta_urgent_ns = get_wm_memory_trip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1971 context->bw_ctx.bw.dcn.watermarks.a.frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1972 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;
1973 context->bw_ctx.bw.dcn.watermarks.a.urgent_latency_ns = get_urgent_latency(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1974 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.fclk_pstate_change_ns = get_fclk_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1975 context->bw_ctx.bw.dcn.watermarks.a.usr_retraining_ns = get_usr_retraining_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
1978 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
1979 if (!context->res_ctx.pipe_ctx[i].stream)
1982 pipes[pipe_idx].clks_cfg.dispclk_mhz = get_dispclk_calculated(&context->bw_ctx.dml, pipes, pipe_cnt);
1983 pipes[pipe_idx].clks_cfg.dppclk_mhz = get_dppclk_calculated(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx);
1985 if (dc->config.forced_clocks) {
1986 pipes[pipe_idx].clks_cfg.dispclk_mhz = context->bw_ctx.dml.soc.clock_limits[0].dispclk_mhz;
1987 pipes[pipe_idx].clks_cfg.dppclk_mhz = context->bw_ctx.dml.soc.clock_limits[0].dppclk_mhz;
1989 if (dc->debug.min_disp_clk_khz > pipes[pipe_idx].clks_cfg.dispclk_mhz * 1000)
1990 pipes[pipe_idx].clks_cfg.dispclk_mhz = dc->debug.min_disp_clk_khz / 1000.0;
1991 if (dc->debug.min_dpp_clk_khz > pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000)
1992 pipes[pipe_idx].clks_cfg.dppclk_mhz = dc->debug.min_dpp_clk_khz / 1000.0;
1997 context->perf_params.stutter_period_us = context->bw_ctx.dml.vba.StutterPeriod;
1999 if (context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching && dummy_latency_index == 0)
2000 context->bw_ctx.dml.soc.fclk_change_latency_us =
2001 dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us;
2003 dcn32_calculate_dlg_params(dc, context, pipes, pipe_cnt, vlevel);
2006 /* Restore full p-state latency */
2007 context->bw_ctx.dml.soc.dram_clock_change_latency_us =
2008 dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us;
2010 if (context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching) {
2011 dcn30_setup_mclk_switch_using_fw_based_vblank_stretch(dc, context);
2012 if (dummy_latency_index == 0)
2013 context->bw_ctx.dml.soc.fclk_change_latency_us =
2014 dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.fclk_change_latency_us;
2018 static void dcn32_get_optimal_dcfclk_fclk_for_uclk(unsigned int uclk_mts,
2019 unsigned int *optimal_dcfclk,
2020 unsigned int *optimal_fclk)
2022 double bw_from_dram, bw_from_dram1, bw_from_dram2;
2024 bw_from_dram1 = uclk_mts * dcn3_2_soc.num_chans *
2025 dcn3_2_soc.dram_channel_width_bytes * (dcn3_2_soc.max_avg_dram_bw_use_normal_percent / 100);
2026 bw_from_dram2 = uclk_mts * dcn3_2_soc.num_chans *
2027 dcn3_2_soc.dram_channel_width_bytes * (dcn3_2_soc.max_avg_sdp_bw_use_normal_percent / 100);
2029 bw_from_dram = (bw_from_dram1 < bw_from_dram2) ? bw_from_dram1 : bw_from_dram2;
2032 *optimal_fclk = bw_from_dram /
2033 (dcn3_2_soc.fabric_datapath_to_dcn_data_return_bytes * (dcn3_2_soc.max_avg_sdp_bw_use_normal_percent / 100));
2036 *optimal_dcfclk = bw_from_dram /
2037 (dcn3_2_soc.return_bus_width_bytes * (dcn3_2_soc.max_avg_sdp_bw_use_normal_percent / 100));
2040 static void remove_entry_from_table_at_index(struct _vcs_dpi_voltage_scaling_st *table, unsigned int *num_entries,
2045 if (*num_entries == 0)
2048 for (i = index; i < *num_entries - 1; i++) {
2049 table[i] = table[i + 1];
2051 memset(&table[--(*num_entries)], 0, sizeof(struct _vcs_dpi_voltage_scaling_st));
2054 void dcn32_patch_dpm_table(struct clk_bw_params *bw_params)
2057 unsigned int max_dcfclk_mhz = 0, max_dispclk_mhz = 0, max_dppclk_mhz = 0,
2058 max_phyclk_mhz = 0, max_dtbclk_mhz = 0, max_fclk_mhz = 0, max_uclk_mhz = 0;
2060 for (i = 0; i < MAX_NUM_DPM_LVL; i++) {
2061 if (bw_params->clk_table.entries[i].dcfclk_mhz > max_dcfclk_mhz)
2062 max_dcfclk_mhz = bw_params->clk_table.entries[i].dcfclk_mhz;
2063 if (bw_params->clk_table.entries[i].fclk_mhz > max_fclk_mhz)
2064 max_fclk_mhz = bw_params->clk_table.entries[i].fclk_mhz;
2065 if (bw_params->clk_table.entries[i].memclk_mhz > max_uclk_mhz)
2066 max_uclk_mhz = bw_params->clk_table.entries[i].memclk_mhz;
2067 if (bw_params->clk_table.entries[i].dispclk_mhz > max_dispclk_mhz)
2068 max_dispclk_mhz = bw_params->clk_table.entries[i].dispclk_mhz;
2069 if (bw_params->clk_table.entries[i].dppclk_mhz > max_dppclk_mhz)
2070 max_dppclk_mhz = bw_params->clk_table.entries[i].dppclk_mhz;
2071 if (bw_params->clk_table.entries[i].phyclk_mhz > max_phyclk_mhz)
2072 max_phyclk_mhz = bw_params->clk_table.entries[i].phyclk_mhz;
2073 if (bw_params->clk_table.entries[i].dtbclk_mhz > max_dtbclk_mhz)
2074 max_dtbclk_mhz = bw_params->clk_table.entries[i].dtbclk_mhz;
2077 /* Scan through clock values we currently have and if they are 0,
2078 * then populate it with dcn3_2_soc.clock_limits[] value.
2080 * Do it for DCFCLK, DISPCLK, DTBCLK and UCLK as any of those being
2081 * 0, will cause it to skip building the clock table.
2083 if (max_dcfclk_mhz == 0)
2084 bw_params->clk_table.entries[0].dcfclk_mhz = dcn3_2_soc.clock_limits[0].dcfclk_mhz;
2085 if (max_dispclk_mhz == 0)
2086 bw_params->clk_table.entries[0].dispclk_mhz = dcn3_2_soc.clock_limits[0].dispclk_mhz;
2087 if (max_dtbclk_mhz == 0)
2088 bw_params->clk_table.entries[0].dtbclk_mhz = dcn3_2_soc.clock_limits[0].dtbclk_mhz;
2089 if (max_uclk_mhz == 0)
2090 bw_params->clk_table.entries[0].memclk_mhz = dcn3_2_soc.clock_limits[0].dram_speed_mts / 16;
2093 static int build_synthetic_soc_states(struct clk_bw_params *bw_params,
2094 struct _vcs_dpi_voltage_scaling_st *table, unsigned int *num_entries)
2097 struct _vcs_dpi_voltage_scaling_st entry = {0};
2099 unsigned int max_dcfclk_mhz = 0, max_dispclk_mhz = 0, max_dppclk_mhz = 0,
2100 max_phyclk_mhz = 0, max_dtbclk_mhz = 0, max_fclk_mhz = 0, max_uclk_mhz = 0;
2102 unsigned int min_dcfclk_mhz = 199, min_fclk_mhz = 299;
2104 static const unsigned int num_dcfclk_stas = 5;
2105 unsigned int dcfclk_sta_targets[DC__VOLTAGE_STATES] = {199, 615, 906, 1324, 1564};
2107 unsigned int num_uclk_dpms = 0;
2108 unsigned int num_fclk_dpms = 0;
2109 unsigned int num_dcfclk_dpms = 0;
2111 for (i = 0; i < MAX_NUM_DPM_LVL; i++) {
2112 if (bw_params->clk_table.entries[i].dcfclk_mhz > max_dcfclk_mhz)
2113 max_dcfclk_mhz = bw_params->clk_table.entries[i].dcfclk_mhz;
2114 if (bw_params->clk_table.entries[i].fclk_mhz > max_fclk_mhz)
2115 max_fclk_mhz = bw_params->clk_table.entries[i].fclk_mhz;
2116 if (bw_params->clk_table.entries[i].memclk_mhz > max_uclk_mhz)
2117 max_uclk_mhz = bw_params->clk_table.entries[i].memclk_mhz;
2118 if (bw_params->clk_table.entries[i].dispclk_mhz > max_dispclk_mhz)
2119 max_dispclk_mhz = bw_params->clk_table.entries[i].dispclk_mhz;
2120 if (bw_params->clk_table.entries[i].dppclk_mhz > max_dppclk_mhz)
2121 max_dppclk_mhz = bw_params->clk_table.entries[i].dppclk_mhz;
2122 if (bw_params->clk_table.entries[i].phyclk_mhz > max_phyclk_mhz)
2123 max_phyclk_mhz = bw_params->clk_table.entries[i].phyclk_mhz;
2124 if (bw_params->clk_table.entries[i].dtbclk_mhz > max_dtbclk_mhz)
2125 max_dtbclk_mhz = bw_params->clk_table.entries[i].dtbclk_mhz;
2127 if (bw_params->clk_table.entries[i].memclk_mhz > 0)
2129 if (bw_params->clk_table.entries[i].fclk_mhz > 0)
2131 if (bw_params->clk_table.entries[i].dcfclk_mhz > 0)
2135 if (!max_dcfclk_mhz || !max_dispclk_mhz || !max_dtbclk_mhz)
2138 if (max_dppclk_mhz == 0)
2139 max_dppclk_mhz = max_dispclk_mhz;
2141 if (max_fclk_mhz == 0)
2142 max_fclk_mhz = max_dcfclk_mhz * dcn3_2_soc.pct_ideal_sdp_bw_after_urgent / dcn3_2_soc.pct_ideal_fabric_bw_after_urgent;
2144 if (max_phyclk_mhz == 0)
2145 max_phyclk_mhz = dcn3_2_soc.clock_limits[0].phyclk_mhz;
2148 entry.dispclk_mhz = max_dispclk_mhz;
2149 entry.dscclk_mhz = max_dispclk_mhz / 3;
2150 entry.dppclk_mhz = max_dppclk_mhz;
2151 entry.dtbclk_mhz = max_dtbclk_mhz;
2152 entry.phyclk_mhz = max_phyclk_mhz;
2153 entry.phyclk_d18_mhz = dcn3_2_soc.clock_limits[0].phyclk_d18_mhz;
2154 entry.phyclk_d32_mhz = dcn3_2_soc.clock_limits[0].phyclk_d32_mhz;
2156 // Insert all the DCFCLK STAs
2157 for (i = 0; i < num_dcfclk_stas; i++) {
2158 entry.dcfclk_mhz = dcfclk_sta_targets[i];
2159 entry.fabricclk_mhz = 0;
2160 entry.dram_speed_mts = 0;
2163 insert_entry_into_table_sorted(table, num_entries, &entry);
2167 // Insert the max DCFCLK
2168 entry.dcfclk_mhz = max_dcfclk_mhz;
2169 entry.fabricclk_mhz = 0;
2170 entry.dram_speed_mts = 0;
2173 insert_entry_into_table_sorted(table, num_entries, &entry);
2176 // Insert the UCLK DPMS
2177 for (i = 0; i < num_uclk_dpms; i++) {
2178 entry.dcfclk_mhz = 0;
2179 entry.fabricclk_mhz = 0;
2180 entry.dram_speed_mts = bw_params->clk_table.entries[i].memclk_mhz * 16;
2183 insert_entry_into_table_sorted(table, num_entries, &entry);
2187 // If FCLK is coarse grained, insert individual DPMs.
2188 if (num_fclk_dpms > 2) {
2189 for (i = 0; i < num_fclk_dpms; i++) {
2190 entry.dcfclk_mhz = 0;
2191 entry.fabricclk_mhz = bw_params->clk_table.entries[i].fclk_mhz;
2192 entry.dram_speed_mts = 0;
2195 insert_entry_into_table_sorted(table, num_entries, &entry);
2199 // If FCLK fine grained, only insert max
2201 entry.dcfclk_mhz = 0;
2202 entry.fabricclk_mhz = max_fclk_mhz;
2203 entry.dram_speed_mts = 0;
2206 insert_entry_into_table_sorted(table, num_entries, &entry);
2210 // At this point, the table contains all "points of interest" based on
2211 // DPMs from PMFW, and STAs. Table is sorted by BW, and all clock
2212 // ratios (by derate, are exact).
2214 // Remove states that require higher clocks than are supported
2215 for (i = *num_entries - 1; i >= 0 ; i--) {
2216 if (table[i].dcfclk_mhz > max_dcfclk_mhz ||
2217 table[i].fabricclk_mhz > max_fclk_mhz ||
2218 table[i].dram_speed_mts > max_uclk_mhz * 16)
2219 remove_entry_from_table_at_index(table, num_entries, i);
2222 // At this point, the table only contains supported points of interest
2223 // it could be used as is, but some states may be redundant due to
2224 // coarse grained nature of some clocks, so we want to round up to
2225 // coarse grained DPMs and remove duplicates.
2228 for (i = *num_entries - 1; i >= 0 ; i--) {
2229 for (j = 0; j < num_uclk_dpms; j++) {
2230 if (bw_params->clk_table.entries[j].memclk_mhz * 16 >= table[i].dram_speed_mts) {
2231 table[i].dram_speed_mts = bw_params->clk_table.entries[j].memclk_mhz * 16;
2237 // If FCLK is coarse grained, round up to next DPMs
2238 if (num_fclk_dpms > 2) {
2239 for (i = *num_entries - 1; i >= 0 ; i--) {
2240 for (j = 0; j < num_fclk_dpms; j++) {
2241 if (bw_params->clk_table.entries[j].fclk_mhz >= table[i].fabricclk_mhz) {
2242 table[i].fabricclk_mhz = bw_params->clk_table.entries[j].fclk_mhz;
2248 // Otherwise, round up to minimum.
2250 for (i = *num_entries - 1; i >= 0 ; i--) {
2251 if (table[i].fabricclk_mhz < min_fclk_mhz) {
2252 table[i].fabricclk_mhz = min_fclk_mhz;
2258 // Round DCFCLKs up to minimum
2259 for (i = *num_entries - 1; i >= 0 ; i--) {
2260 if (table[i].dcfclk_mhz < min_dcfclk_mhz) {
2261 table[i].dcfclk_mhz = min_dcfclk_mhz;
2266 // Remove duplicate states, note duplicate states are always neighbouring since table is sorted.
2268 while (i < *num_entries - 1) {
2269 if (table[i].dcfclk_mhz == table[i + 1].dcfclk_mhz &&
2270 table[i].fabricclk_mhz == table[i + 1].fabricclk_mhz &&
2271 table[i].dram_speed_mts == table[i + 1].dram_speed_mts)
2272 remove_entry_from_table_at_index(table, num_entries, i + 1);
2277 // Fix up the state indicies
2278 for (i = *num_entries - 1; i >= 0 ; i--) {
2286 * dcn32_update_bw_bounding_box
2288 * This would override some dcn3_2 ip_or_soc initial parameters hardcoded from
2289 * spreadsheet with actual values as per dGPU SKU:
2290 * - with passed few options from dc->config
2291 * - with dentist_vco_frequency from Clk Mgr (currently hardcoded, but might
2292 * need to get it from PM FW)
2293 * - with passed latency values (passed in ns units) in dc-> bb override for
2294 * debugging purposes
2295 * - with passed latencies from VBIOS (in 100_ns units) if available for
2297 * - with number of DRAM channels from VBIOS (which differ for certain dGPU SKU
2299 * - clocks levels with passed clk_table entries from Clk Mgr as reported by PM
2300 * FW for different clocks (which might differ for certain dGPU SKU of the
2303 void dcn32_update_bw_bounding_box_fpu(struct dc *dc, struct clk_bw_params *bw_params)
2305 dc_assert_fp_enabled();
2307 if (!IS_FPGA_MAXIMUS_DC(dc->ctx->dce_environment)) {
2308 /* Overrides from dc->config options */
2309 dcn3_2_ip.clamp_min_dcfclk = dc->config.clamp_min_dcfclk;
2311 /* Override from passed dc->bb_overrides if available*/
2312 if ((int)(dcn3_2_soc.sr_exit_time_us * 1000) != dc->bb_overrides.sr_exit_time_ns
2313 && dc->bb_overrides.sr_exit_time_ns) {
2314 dcn3_2_soc.sr_exit_time_us = dc->bb_overrides.sr_exit_time_ns / 1000.0;
2317 if ((int)(dcn3_2_soc.sr_enter_plus_exit_time_us * 1000)
2318 != dc->bb_overrides.sr_enter_plus_exit_time_ns
2319 && dc->bb_overrides.sr_enter_plus_exit_time_ns) {
2320 dcn3_2_soc.sr_enter_plus_exit_time_us =
2321 dc->bb_overrides.sr_enter_plus_exit_time_ns / 1000.0;
2324 if ((int)(dcn3_2_soc.urgent_latency_us * 1000) != dc->bb_overrides.urgent_latency_ns
2325 && dc->bb_overrides.urgent_latency_ns) {
2326 dcn3_2_soc.urgent_latency_us = dc->bb_overrides.urgent_latency_ns / 1000.0;
2327 dcn3_2_soc.urgent_latency_pixel_data_only_us = dc->bb_overrides.urgent_latency_ns / 1000.0;
2330 if ((int)(dcn3_2_soc.dram_clock_change_latency_us * 1000)
2331 != dc->bb_overrides.dram_clock_change_latency_ns
2332 && dc->bb_overrides.dram_clock_change_latency_ns) {
2333 dcn3_2_soc.dram_clock_change_latency_us =
2334 dc->bb_overrides.dram_clock_change_latency_ns / 1000.0;
2337 if ((int)(dcn3_2_soc.fclk_change_latency_us * 1000)
2338 != dc->bb_overrides.fclk_clock_change_latency_ns
2339 && dc->bb_overrides.fclk_clock_change_latency_ns) {
2340 dcn3_2_soc.fclk_change_latency_us =
2341 dc->bb_overrides.fclk_clock_change_latency_ns / 1000;
2344 if ((int)(dcn3_2_soc.dummy_pstate_latency_us * 1000)
2345 != dc->bb_overrides.dummy_clock_change_latency_ns
2346 && dc->bb_overrides.dummy_clock_change_latency_ns) {
2347 dcn3_2_soc.dummy_pstate_latency_us =
2348 dc->bb_overrides.dummy_clock_change_latency_ns / 1000.0;
2351 /* Override from VBIOS if VBIOS bb_info available */
2352 if (dc->ctx->dc_bios->funcs->get_soc_bb_info) {
2353 struct bp_soc_bb_info bb_info = {0};
2355 if (dc->ctx->dc_bios->funcs->get_soc_bb_info(dc->ctx->dc_bios, &bb_info) == BP_RESULT_OK) {
2356 if (bb_info.dram_clock_change_latency_100ns > 0)
2357 dcn3_2_soc.dram_clock_change_latency_us =
2358 bb_info.dram_clock_change_latency_100ns * 10;
2360 if (bb_info.dram_sr_enter_exit_latency_100ns > 0)
2361 dcn3_2_soc.sr_enter_plus_exit_time_us =
2362 bb_info.dram_sr_enter_exit_latency_100ns * 10;
2364 if (bb_info.dram_sr_exit_latency_100ns > 0)
2365 dcn3_2_soc.sr_exit_time_us =
2366 bb_info.dram_sr_exit_latency_100ns * 10;
2370 /* Override from VBIOS for num_chan */
2371 if (dc->ctx->dc_bios->vram_info.num_chans)
2372 dcn3_2_soc.num_chans = dc->ctx->dc_bios->vram_info.num_chans;
2374 if (dc->ctx->dc_bios->vram_info.dram_channel_width_bytes)
2375 dcn3_2_soc.dram_channel_width_bytes = dc->ctx->dc_bios->vram_info.dram_channel_width_bytes;
2378 /* DML DSC delay factor workaround */
2379 dcn3_2_ip.dsc_delay_factor_wa = dc->debug.dsc_delay_factor_wa_x1000 / 1000.0;
2381 dcn3_2_ip.min_prefetch_in_strobe_us = dc->debug.min_prefetch_in_strobe_ns / 1000.0;
2383 /* Override dispclk_dppclk_vco_speed_mhz from Clk Mgr */
2384 dcn3_2_soc.dispclk_dppclk_vco_speed_mhz = dc->clk_mgr->dentist_vco_freq_khz / 1000.0;
2385 dc->dml.soc.dispclk_dppclk_vco_speed_mhz = dc->clk_mgr->dentist_vco_freq_khz / 1000.0;
2387 /* Overrides Clock levelsfrom CLK Mgr table entries as reported by PM FW */
2388 if ((!IS_FPGA_MAXIMUS_DC(dc->ctx->dce_environment)) && (bw_params->clk_table.entries[0].memclk_mhz)) {
2389 if (dc->debug.use_legacy_soc_bb_mechanism) {
2390 unsigned int i = 0, j = 0, num_states = 0;
2392 unsigned int dcfclk_mhz[DC__VOLTAGE_STATES] = {0};
2393 unsigned int dram_speed_mts[DC__VOLTAGE_STATES] = {0};
2394 unsigned int optimal_uclk_for_dcfclk_sta_targets[DC__VOLTAGE_STATES] = {0};
2395 unsigned int optimal_dcfclk_for_uclk[DC__VOLTAGE_STATES] = {0};
2396 unsigned int min_dcfclk = UINT_MAX;
2397 /* Set 199 as first value in STA target array to have a minimum DCFCLK value.
2398 * For DCN32 we set min to 199 so minimum FCLK DPM0 (300Mhz can be achieved) */
2399 unsigned int dcfclk_sta_targets[DC__VOLTAGE_STATES] = {199, 615, 906, 1324, 1564};
2400 unsigned int num_dcfclk_sta_targets = 4, num_uclk_states = 0;
2401 unsigned int max_dcfclk_mhz = 0, max_dispclk_mhz = 0, max_dppclk_mhz = 0, max_phyclk_mhz = 0;
2403 for (i = 0; i < MAX_NUM_DPM_LVL; i++) {
2404 if (bw_params->clk_table.entries[i].dcfclk_mhz > max_dcfclk_mhz)
2405 max_dcfclk_mhz = bw_params->clk_table.entries[i].dcfclk_mhz;
2406 if (bw_params->clk_table.entries[i].dcfclk_mhz != 0 &&
2407 bw_params->clk_table.entries[i].dcfclk_mhz < min_dcfclk)
2408 min_dcfclk = bw_params->clk_table.entries[i].dcfclk_mhz;
2409 if (bw_params->clk_table.entries[i].dispclk_mhz > max_dispclk_mhz)
2410 max_dispclk_mhz = bw_params->clk_table.entries[i].dispclk_mhz;
2411 if (bw_params->clk_table.entries[i].dppclk_mhz > max_dppclk_mhz)
2412 max_dppclk_mhz = bw_params->clk_table.entries[i].dppclk_mhz;
2413 if (bw_params->clk_table.entries[i].phyclk_mhz > max_phyclk_mhz)
2414 max_phyclk_mhz = bw_params->clk_table.entries[i].phyclk_mhz;
2416 if (min_dcfclk > dcfclk_sta_targets[0])
2417 dcfclk_sta_targets[0] = min_dcfclk;
2418 if (!max_dcfclk_mhz)
2419 max_dcfclk_mhz = dcn3_2_soc.clock_limits[0].dcfclk_mhz;
2420 if (!max_dispclk_mhz)
2421 max_dispclk_mhz = dcn3_2_soc.clock_limits[0].dispclk_mhz;
2422 if (!max_dppclk_mhz)
2423 max_dppclk_mhz = dcn3_2_soc.clock_limits[0].dppclk_mhz;
2424 if (!max_phyclk_mhz)
2425 max_phyclk_mhz = dcn3_2_soc.clock_limits[0].phyclk_mhz;
2427 if (max_dcfclk_mhz > dcfclk_sta_targets[num_dcfclk_sta_targets-1]) {
2428 // If max DCFCLK is greater than the max DCFCLK STA target, insert into the DCFCLK STA target array
2429 dcfclk_sta_targets[num_dcfclk_sta_targets] = max_dcfclk_mhz;
2430 num_dcfclk_sta_targets++;
2431 } else if (max_dcfclk_mhz < dcfclk_sta_targets[num_dcfclk_sta_targets-1]) {
2432 // If max DCFCLK is less than the max DCFCLK STA target, cap values and remove duplicates
2433 for (i = 0; i < num_dcfclk_sta_targets; i++) {
2434 if (dcfclk_sta_targets[i] > max_dcfclk_mhz) {
2435 dcfclk_sta_targets[i] = max_dcfclk_mhz;
2439 // Update size of array since we "removed" duplicates
2440 num_dcfclk_sta_targets = i + 1;
2443 num_uclk_states = bw_params->clk_table.num_entries;
2445 // Calculate optimal dcfclk for each uclk
2446 for (i = 0; i < num_uclk_states; i++) {
2447 dcn32_get_optimal_dcfclk_fclk_for_uclk(bw_params->clk_table.entries[i].memclk_mhz * 16,
2448 &optimal_dcfclk_for_uclk[i], NULL);
2449 if (optimal_dcfclk_for_uclk[i] < bw_params->clk_table.entries[0].dcfclk_mhz) {
2450 optimal_dcfclk_for_uclk[i] = bw_params->clk_table.entries[0].dcfclk_mhz;
2454 // Calculate optimal uclk for each dcfclk sta target
2455 for (i = 0; i < num_dcfclk_sta_targets; i++) {
2456 for (j = 0; j < num_uclk_states; j++) {
2457 if (dcfclk_sta_targets[i] < optimal_dcfclk_for_uclk[j]) {
2458 optimal_uclk_for_dcfclk_sta_targets[i] =
2459 bw_params->clk_table.entries[j].memclk_mhz * 16;
2467 // create the final dcfclk and uclk table
2468 while (i < num_dcfclk_sta_targets && j < num_uclk_states && num_states < DC__VOLTAGE_STATES) {
2469 if (dcfclk_sta_targets[i] < optimal_dcfclk_for_uclk[j] && i < num_dcfclk_sta_targets) {
2470 dcfclk_mhz[num_states] = dcfclk_sta_targets[i];
2471 dram_speed_mts[num_states++] = optimal_uclk_for_dcfclk_sta_targets[i++];
2473 if (j < num_uclk_states && optimal_dcfclk_for_uclk[j] <= max_dcfclk_mhz) {
2474 dcfclk_mhz[num_states] = optimal_dcfclk_for_uclk[j];
2475 dram_speed_mts[num_states++] = bw_params->clk_table.entries[j++].memclk_mhz * 16;
2477 j = num_uclk_states;
2482 while (i < num_dcfclk_sta_targets && num_states < DC__VOLTAGE_STATES) {
2483 dcfclk_mhz[num_states] = dcfclk_sta_targets[i];
2484 dram_speed_mts[num_states++] = optimal_uclk_for_dcfclk_sta_targets[i++];
2487 while (j < num_uclk_states && num_states < DC__VOLTAGE_STATES &&
2488 optimal_dcfclk_for_uclk[j] <= max_dcfclk_mhz) {
2489 dcfclk_mhz[num_states] = optimal_dcfclk_for_uclk[j];
2490 dram_speed_mts[num_states++] = bw_params->clk_table.entries[j++].memclk_mhz * 16;
2493 dcn3_2_soc.num_states = num_states;
2494 for (i = 0; i < dcn3_2_soc.num_states; i++) {
2495 dcn3_2_soc.clock_limits[i].state = i;
2496 dcn3_2_soc.clock_limits[i].dcfclk_mhz = dcfclk_mhz[i];
2497 dcn3_2_soc.clock_limits[i].fabricclk_mhz = dcfclk_mhz[i];
2499 /* Fill all states with max values of all these clocks */
2500 dcn3_2_soc.clock_limits[i].dispclk_mhz = max_dispclk_mhz;
2501 dcn3_2_soc.clock_limits[i].dppclk_mhz = max_dppclk_mhz;
2502 dcn3_2_soc.clock_limits[i].phyclk_mhz = max_phyclk_mhz;
2503 dcn3_2_soc.clock_limits[i].dscclk_mhz = max_dispclk_mhz / 3;
2505 /* Populate from bw_params for DTBCLK, SOCCLK */
2507 if (!bw_params->clk_table.entries[i].dtbclk_mhz) {
2508 dcn3_2_soc.clock_limits[i].dtbclk_mhz = dcn3_2_soc.clock_limits[i-1].dtbclk_mhz;
2510 dcn3_2_soc.clock_limits[i].dtbclk_mhz = bw_params->clk_table.entries[i].dtbclk_mhz;
2512 } else if (bw_params->clk_table.entries[i].dtbclk_mhz) {
2513 dcn3_2_soc.clock_limits[i].dtbclk_mhz = bw_params->clk_table.entries[i].dtbclk_mhz;
2516 if (!bw_params->clk_table.entries[i].socclk_mhz && i > 0)
2517 dcn3_2_soc.clock_limits[i].socclk_mhz = dcn3_2_soc.clock_limits[i-1].socclk_mhz;
2519 dcn3_2_soc.clock_limits[i].socclk_mhz = bw_params->clk_table.entries[i].socclk_mhz;
2521 if (!dram_speed_mts[i] && i > 0)
2522 dcn3_2_soc.clock_limits[i].dram_speed_mts = dcn3_2_soc.clock_limits[i-1].dram_speed_mts;
2524 dcn3_2_soc.clock_limits[i].dram_speed_mts = dram_speed_mts[i];
2526 /* These clocks cannot come from bw_params, always fill from dcn3_2_soc[0] */
2527 /* PHYCLK_D18, PHYCLK_D32 */
2528 dcn3_2_soc.clock_limits[i].phyclk_d18_mhz = dcn3_2_soc.clock_limits[0].phyclk_d18_mhz;
2529 dcn3_2_soc.clock_limits[i].phyclk_d32_mhz = dcn3_2_soc.clock_limits[0].phyclk_d32_mhz;
2532 build_synthetic_soc_states(bw_params, dcn3_2_soc.clock_limits, &dcn3_2_soc.num_states);
2535 /* Re-init DML with updated bb */
2536 dml_init_instance(&dc->dml, &dcn3_2_soc, &dcn3_2_ip, DML_PROJECT_DCN32);
2537 if (dc->current_state)
2538 dml_init_instance(&dc->current_state->bw_ctx.dml, &dcn3_2_soc, &dcn3_2_ip, DML_PROJECT_DCN32);