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 "dcn32/dcn32_resource.h"
28 #include "dcn20/dcn20_resource.h"
29 #include "display_mode_vba_util_32.h"
30 #include "dml/dcn32/display_mode_vba_32.h"
31 // We need this includes for WATERMARKS_* defines
32 #include "clk_mgr/dcn32/dcn32_smu13_driver_if.h"
33 #include "dcn30/dcn30_resource.h"
35 #include "dc_state_priv.h"
37 #define DC_LOGGER_INIT(logger)
39 static const struct subvp_high_refresh_list subvp_high_refresh_list = {
43 {.width = 3840, .height = 2160, },
44 {.width = 3440, .height = 1440, },
45 {.width = 2560, .height = 1440, },
46 {.width = 1920, .height = 1080, }},
49 static const struct subvp_active_margin_list subvp_active_margin_list = {
53 {.width = 2560, .height = 1440, },
54 {.width = 1920, .height = 1080, }},
57 struct _vcs_dpi_ip_params_st dcn3_2_ip = {
59 .gpuvm_max_page_table_levels = 4,
61 .rob_buffer_size_kbytes = 128,
62 .det_buffer_size_kbytes = DCN3_2_DEFAULT_DET_SIZE,
63 .config_return_buffer_size_in_kbytes = 1280,
64 .compressed_buffer_segment_size_in_kbytes = 64,
65 .meta_fifo_size_in_kentries = 22,
66 .zero_size_buffer_entries = 512,
67 .compbuf_reserved_space_64b = 256,
68 .compbuf_reserved_space_zs = 64,
69 .dpp_output_buffer_pixels = 2560,
70 .opp_output_buffer_lines = 1,
71 .pixel_chunk_size_kbytes = 8,
72 .alpha_pixel_chunk_size_kbytes = 4,
73 .min_pixel_chunk_size_bytes = 1024,
74 .dcc_meta_buffer_size_bytes = 6272,
75 .meta_chunk_size_kbytes = 2,
76 .min_meta_chunk_size_bytes = 256,
77 .writeback_chunk_size_kbytes = 8,
78 .ptoi_supported = false,
80 .maximum_dsc_bits_per_component = 12,
81 .maximum_pixels_per_line_per_dsc_unit = 6016,
82 .dsc422_native_support = true,
83 .is_line_buffer_bpp_fixed = true,
84 .line_buffer_fixed_bpp = 57,
85 .line_buffer_size_bits = 1171920,
86 .max_line_buffer_lines = 32,
87 .writeback_interface_buffer_size_kbytes = 90,
90 .max_num_hdmi_frl_outputs = 1,
92 .max_dchub_pscl_bw_pix_per_clk = 4,
93 .max_pscl_lb_bw_pix_per_clk = 2,
94 .max_lb_vscl_bw_pix_per_clk = 4,
95 .max_vscl_hscl_bw_pix_per_clk = 4,
100 .dpte_buffer_size_in_pte_reqs_luma = 64,
101 .dpte_buffer_size_in_pte_reqs_chroma = 34,
102 .dispclk_ramp_margin_percent = 1,
103 .max_inter_dcn_tile_repeaters = 8,
104 .cursor_buffer_size = 16,
105 .cursor_chunk_size = 2,
106 .writeback_line_buffer_buffer_size = 0,
107 .writeback_min_hscl_ratio = 1,
108 .writeback_min_vscl_ratio = 1,
109 .writeback_max_hscl_ratio = 1,
110 .writeback_max_vscl_ratio = 1,
111 .writeback_max_hscl_taps = 1,
112 .writeback_max_vscl_taps = 1,
113 .dppclk_delay_subtotal = 47,
114 .dppclk_delay_scl = 50,
115 .dppclk_delay_scl_lb_only = 16,
116 .dppclk_delay_cnvc_formatter = 28,
117 .dppclk_delay_cnvc_cursor = 6,
118 .dispclk_delay_subtotal = 125,
119 .dynamic_metadata_vm_enabled = false,
120 .odm_combine_4to1_supported = false,
121 .dcc_supported = true,
122 .max_num_dp2p0_outputs = 2,
123 .max_num_dp2p0_streams = 4,
126 struct _vcs_dpi_soc_bounding_box_st dcn3_2_soc = {
130 .dcfclk_mhz = 1564.0,
131 .fabricclk_mhz = 2500.0,
132 .dispclk_mhz = 2150.0,
133 .dppclk_mhz = 2150.0,
135 .phyclk_d18_mhz = 667.0,
136 .phyclk_d32_mhz = 625.0,
137 .socclk_mhz = 1200.0,
138 .dscclk_mhz = 716.667,
139 .dram_speed_mts = 18000.0,
140 .dtbclk_mhz = 1564.0,
144 .sr_exit_time_us = 42.97,
145 .sr_enter_plus_exit_time_us = 49.94,
146 .sr_exit_z8_time_us = 285.0,
147 .sr_enter_plus_exit_z8_time_us = 320,
148 .writeback_latency_us = 12.0,
149 .round_trip_ping_latency_dcfclk_cycles = 263,
150 .urgent_latency_pixel_data_only_us = 4.0,
151 .urgent_latency_pixel_mixed_with_vm_data_us = 4.0,
152 .urgent_latency_vm_data_only_us = 4.0,
153 .fclk_change_latency_us = 25,
154 .usr_retraining_latency_us = 2,
156 .mall_allocated_for_dcn_mbytes = 64,
157 .urgent_out_of_order_return_per_channel_pixel_only_bytes = 4096,
158 .urgent_out_of_order_return_per_channel_pixel_and_vm_bytes = 4096,
159 .urgent_out_of_order_return_per_channel_vm_only_bytes = 4096,
160 .pct_ideal_sdp_bw_after_urgent = 90.0,
161 .pct_ideal_fabric_bw_after_urgent = 67.0,
162 .pct_ideal_dram_sdp_bw_after_urgent_pixel_only = 20.0,
163 .pct_ideal_dram_sdp_bw_after_urgent_pixel_and_vm = 60.0, // N/A, for now keep as is until DML implemented
164 .pct_ideal_dram_sdp_bw_after_urgent_vm_only = 30.0, // N/A, for now keep as is until DML implemented
165 .pct_ideal_dram_bw_after_urgent_strobe = 67.0,
166 .max_avg_sdp_bw_use_normal_percent = 80.0,
167 .max_avg_fabric_bw_use_normal_percent = 60.0,
168 .max_avg_dram_bw_use_normal_strobe_percent = 50.0,
169 .max_avg_dram_bw_use_normal_percent = 15.0,
171 .dram_channel_width_bytes = 2,
172 .fabric_datapath_to_dcn_data_return_bytes = 64,
173 .return_bus_width_bytes = 64,
174 .downspread_percent = 0.38,
175 .dcn_downspread_percent = 0.5,
176 .dram_clock_change_latency_us = 400,
177 .dispclk_dppclk_vco_speed_mhz = 4300.0,
178 .do_urgent_latency_adjustment = true,
179 .urgent_latency_adjustment_fabric_clock_component_us = 1.0,
180 .urgent_latency_adjustment_fabric_clock_reference_mhz = 3000,
183 void dcn32_build_wm_range_table_fpu(struct clk_mgr_internal *clk_mgr)
186 double pstate_latency_us = clk_mgr->base.ctx->dc->dml.soc.dram_clock_change_latency_us;
187 double fclk_change_latency_us = clk_mgr->base.ctx->dc->dml.soc.fclk_change_latency_us;
188 double sr_exit_time_us = clk_mgr->base.ctx->dc->dml.soc.sr_exit_time_us;
189 double sr_enter_plus_exit_time_us = clk_mgr->base.ctx->dc->dml.soc.sr_enter_plus_exit_time_us;
190 /* For min clocks use as reported by PM FW and report those as min */
191 uint16_t min_uclk_mhz = clk_mgr->base.bw_params->clk_table.entries[0].memclk_mhz;
192 uint16_t min_dcfclk_mhz = clk_mgr->base.bw_params->clk_table.entries[0].dcfclk_mhz;
193 uint16_t setb_min_uclk_mhz = min_uclk_mhz;
194 uint16_t dcfclk_mhz_for_the_second_state = clk_mgr->base.ctx->dc->dml.soc.clock_limits[2].dcfclk_mhz;
196 dc_assert_fp_enabled();
198 /* For Set B ranges use min clocks state 2 when available, and report those to PM FW */
199 if (dcfclk_mhz_for_the_second_state)
200 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.min_dcfclk = dcfclk_mhz_for_the_second_state;
202 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;
204 if (clk_mgr->base.bw_params->clk_table.entries[2].memclk_mhz)
205 setb_min_uclk_mhz = clk_mgr->base.bw_params->clk_table.entries[2].memclk_mhz;
207 /* Set A - Normal - default values */
208 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].valid = true;
209 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us = pstate_latency_us;
210 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].dml_input.fclk_change_latency_us = fclk_change_latency_us;
211 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].dml_input.sr_exit_time_us = sr_exit_time_us;
212 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;
213 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.wm_type = WATERMARKS_CLOCK_RANGE;
214 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.min_dcfclk = min_dcfclk_mhz;
215 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.max_dcfclk = 0xFFFF;
216 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.min_uclk = min_uclk_mhz;
217 clk_mgr->base.bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.max_uclk = 0xFFFF;
219 /* Set B - Performance - higher clocks, using DPM[2] DCFCLK and UCLK */
220 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].valid = true;
221 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].dml_input.pstate_latency_us = pstate_latency_us;
222 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].dml_input.fclk_change_latency_us = fclk_change_latency_us;
223 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].dml_input.sr_exit_time_us = sr_exit_time_us;
224 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;
225 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.wm_type = WATERMARKS_CLOCK_RANGE;
226 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.max_dcfclk = 0xFFFF;
227 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.min_uclk = setb_min_uclk_mhz;
228 clk_mgr->base.bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.max_uclk = 0xFFFF;
230 /* Set C - Dummy P-State - P-State latency set to "dummy p-state" value */
231 /* 'DalDummyClockChangeLatencyNs' registry key option set to 0x7FFFFFFF can be used to disable Set C for dummy p-state */
232 if (clk_mgr->base.ctx->dc->bb_overrides.dummy_clock_change_latency_ns != 0x7FFFFFFF) {
233 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].valid = true;
234 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].dml_input.pstate_latency_us = 50;
235 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].dml_input.fclk_change_latency_us = fclk_change_latency_us;
236 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].dml_input.sr_exit_time_us = sr_exit_time_us;
237 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;
238 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.wm_type = WATERMARKS_DUMMY_PSTATE;
239 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.min_dcfclk = min_dcfclk_mhz;
240 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.max_dcfclk = 0xFFFF;
241 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.min_uclk = min_uclk_mhz;
242 clk_mgr->base.bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.max_uclk = 0xFFFF;
243 clk_mgr->base.bw_params->dummy_pstate_table[0].dram_speed_mts = clk_mgr->base.bw_params->clk_table.entries[0].memclk_mhz * 16;
244 clk_mgr->base.bw_params->dummy_pstate_table[0].dummy_pstate_latency_us = 50;
245 clk_mgr->base.bw_params->dummy_pstate_table[1].dram_speed_mts = clk_mgr->base.bw_params->clk_table.entries[1].memclk_mhz * 16;
246 clk_mgr->base.bw_params->dummy_pstate_table[1].dummy_pstate_latency_us = 9;
247 clk_mgr->base.bw_params->dummy_pstate_table[2].dram_speed_mts = clk_mgr->base.bw_params->clk_table.entries[2].memclk_mhz * 16;
248 clk_mgr->base.bw_params->dummy_pstate_table[2].dummy_pstate_latency_us = 8;
249 clk_mgr->base.bw_params->dummy_pstate_table[3].dram_speed_mts = clk_mgr->base.bw_params->clk_table.entries[3].memclk_mhz * 16;
250 clk_mgr->base.bw_params->dummy_pstate_table[3].dummy_pstate_latency_us = 5;
252 /* Set D - MALL - SR enter and exit time specific to MALL, TBD after bringup or later phase for now use DRAM values / 2 */
253 /* For MALL DRAM clock change latency is N/A, for watermak calculations use lowest value dummy P state latency */
254 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].valid = true;
255 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;
256 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].dml_input.fclk_change_latency_us = fclk_change_latency_us;
257 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].dml_input.sr_exit_time_us = sr_exit_time_us / 2; // TBD
258 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
259 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.wm_type = WATERMARKS_MALL;
260 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.min_dcfclk = min_dcfclk_mhz;
261 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.max_dcfclk = 0xFFFF;
262 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.min_uclk = min_uclk_mhz;
263 clk_mgr->base.bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.max_uclk = 0xFFFF;
267 * Finds dummy_latency_index when MCLK switching using firmware based
268 * vblank stretch is enabled. This function will iterate through the
269 * table of dummy pstate latencies until the lowest value that allows
270 * dm_allow_self_refresh_and_mclk_switch to happen is found
272 int dcn32_find_dummy_latency_index_for_fw_based_mclk_switch(struct dc *dc,
273 struct dc_state *context,
274 display_e2e_pipe_params_st *pipes,
278 const int max_latency_table_entries = 4;
279 struct vba_vars_st *vba = &context->bw_ctx.dml.vba;
280 int dummy_latency_index = 0;
281 enum clock_change_support temp_clock_change_support = vba->DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb];
283 dc_assert_fp_enabled();
285 while (dummy_latency_index < max_latency_table_entries) {
286 if (temp_clock_change_support != dm_dram_clock_change_unsupported)
287 vba->DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] = temp_clock_change_support;
288 context->bw_ctx.dml.soc.dram_clock_change_latency_us =
289 dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us;
290 dcn32_internal_validate_bw(dc, context, pipes, &pipe_cnt, &vlevel, false);
292 /* for subvp + DRR case, if subvp pipes are still present we support pstate */
293 if (vba->DRAMClockChangeSupport[vlevel][vba->maxMpcComb] == dm_dram_clock_change_unsupported &&
294 dcn32_subvp_in_use(dc, context))
295 vba->DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] = temp_clock_change_support;
297 if (vlevel < context->bw_ctx.dml.vba.soc.num_states &&
298 vba->DRAMClockChangeSupport[vlevel][vba->maxMpcComb] != dm_dram_clock_change_unsupported)
301 dummy_latency_index++;
304 if (dummy_latency_index == max_latency_table_entries) {
305 ASSERT(dummy_latency_index != max_latency_table_entries);
306 /* If the execution gets here, it means dummy p_states are
307 * not possible. This should never happen and would mean
308 * something is severely wrong.
309 * Here we reset dummy_latency_index to 3, because it is
310 * better to have underflows than system crashes.
312 dummy_latency_index = max_latency_table_entries - 1;
315 return dummy_latency_index;
319 * dcn32_helper_populate_phantom_dlg_params - Get DLG params for phantom pipes
320 * and populate pipe_ctx with those params.
321 * @dc: [in] current dc state
322 * @context: [in] new dc state
323 * @pipes: [in] DML pipe params array
324 * @pipe_cnt: [in] DML pipe count
326 * This function must be called AFTER the phantom pipes are added to context
327 * and run through DML (so that the DLG params for the phantom pipes can be
328 * populated), and BEFORE we program the timing for the phantom pipes.
330 void dcn32_helper_populate_phantom_dlg_params(struct dc *dc,
331 struct dc_state *context,
332 display_e2e_pipe_params_st *pipes,
335 uint32_t i, pipe_idx;
337 dc_assert_fp_enabled();
339 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
340 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
345 if (pipe->plane_state && dc_state_get_pipe_subvp_type(context, pipe) == SUBVP_PHANTOM) {
346 pipes[pipe_idx].pipe.dest.vstartup_start =
347 get_vstartup(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx);
348 pipes[pipe_idx].pipe.dest.vupdate_offset =
349 get_vupdate_offset(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx);
350 pipes[pipe_idx].pipe.dest.vupdate_width =
351 get_vupdate_width(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx);
352 pipes[pipe_idx].pipe.dest.vready_offset =
353 get_vready_offset(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx);
354 pipe->pipe_dlg_param = pipes[pipe_idx].pipe.dest;
360 static float calculate_net_bw_in_kbytes_sec(struct _vcs_dpi_voltage_scaling_st *entry)
362 float memory_bw_kbytes_sec;
363 float fabric_bw_kbytes_sec;
364 float sdp_bw_kbytes_sec;
365 float limiting_bw_kbytes_sec;
367 memory_bw_kbytes_sec = entry->dram_speed_mts *
368 dcn3_2_soc.num_chans *
369 dcn3_2_soc.dram_channel_width_bytes *
370 ((float)dcn3_2_soc.pct_ideal_dram_sdp_bw_after_urgent_pixel_only / 100);
372 fabric_bw_kbytes_sec = entry->fabricclk_mhz *
373 dcn3_2_soc.return_bus_width_bytes *
374 ((float)dcn3_2_soc.pct_ideal_fabric_bw_after_urgent / 100);
376 sdp_bw_kbytes_sec = entry->dcfclk_mhz *
377 dcn3_2_soc.return_bus_width_bytes *
378 ((float)dcn3_2_soc.pct_ideal_sdp_bw_after_urgent / 100);
380 limiting_bw_kbytes_sec = memory_bw_kbytes_sec;
382 if (fabric_bw_kbytes_sec < limiting_bw_kbytes_sec)
383 limiting_bw_kbytes_sec = fabric_bw_kbytes_sec;
385 if (sdp_bw_kbytes_sec < limiting_bw_kbytes_sec)
386 limiting_bw_kbytes_sec = sdp_bw_kbytes_sec;
388 return limiting_bw_kbytes_sec;
391 static void get_optimal_ntuple(struct _vcs_dpi_voltage_scaling_st *entry)
393 if (entry->dcfclk_mhz > 0) {
394 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);
396 entry->fabricclk_mhz = bw_on_sdp / (dcn3_2_soc.return_bus_width_bytes * ((float)dcn3_2_soc.pct_ideal_fabric_bw_after_urgent / 100));
397 entry->dram_speed_mts = bw_on_sdp / (dcn3_2_soc.num_chans *
398 dcn3_2_soc.dram_channel_width_bytes * ((float)dcn3_2_soc.pct_ideal_dram_sdp_bw_after_urgent_pixel_only / 100));
399 } else if (entry->fabricclk_mhz > 0) {
400 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);
402 entry->dcfclk_mhz = bw_on_fabric / (dcn3_2_soc.return_bus_width_bytes * ((float)dcn3_2_soc.pct_ideal_sdp_bw_after_urgent / 100));
403 entry->dram_speed_mts = bw_on_fabric / (dcn3_2_soc.num_chans *
404 dcn3_2_soc.dram_channel_width_bytes * ((float)dcn3_2_soc.pct_ideal_dram_sdp_bw_after_urgent_pixel_only / 100));
405 } else if (entry->dram_speed_mts > 0) {
406 float bw_on_dram = entry->dram_speed_mts * dcn3_2_soc.num_chans *
407 dcn3_2_soc.dram_channel_width_bytes * ((float)dcn3_2_soc.pct_ideal_dram_sdp_bw_after_urgent_pixel_only / 100);
409 entry->fabricclk_mhz = bw_on_dram / (dcn3_2_soc.return_bus_width_bytes * ((float)dcn3_2_soc.pct_ideal_fabric_bw_after_urgent / 100));
410 entry->dcfclk_mhz = bw_on_dram / (dcn3_2_soc.return_bus_width_bytes * ((float)dcn3_2_soc.pct_ideal_sdp_bw_after_urgent / 100));
414 static void insert_entry_into_table_sorted(struct _vcs_dpi_voltage_scaling_st *table,
415 unsigned int *num_entries,
416 struct _vcs_dpi_voltage_scaling_st *entry)
421 dc_assert_fp_enabled();
423 if (*num_entries == 0) {
427 while (entry->net_bw_in_kbytes_sec > table[index].net_bw_in_kbytes_sec) {
429 if (index >= *num_entries)
433 for (i = *num_entries; i > index; i--)
434 table[i] = table[i - 1];
436 table[index] = *entry;
442 * dcn32_set_phantom_stream_timing - Set timing params for the phantom stream
443 * @dc: current dc state
444 * @context: new dc state
445 * @ref_pipe: Main pipe for the phantom stream
446 * @phantom_stream: target phantom stream state
447 * @pipes: DML pipe params
448 * @pipe_cnt: number of DML pipes
449 * @dc_pipe_idx: DC pipe index for the main pipe (i.e. ref_pipe)
451 * Set timing params of the phantom stream based on calculated output from DML.
452 * This function first gets the DML pipe index using the DC pipe index, then
453 * calls into DML (get_subviewport_lines_needed_in_mall) to get the number of
454 * lines required for SubVP MCLK switching and assigns to the phantom stream
457 * - The number of SubVP lines calculated in DML does not take into account
458 * FW processing delays and required pstate allow width, so we must include
461 * - Set phantom backporch = vstartup of main pipe
463 void dcn32_set_phantom_stream_timing(struct dc *dc,
464 struct dc_state *context,
465 struct pipe_ctx *ref_pipe,
466 struct dc_stream_state *phantom_stream,
467 display_e2e_pipe_params_st *pipes,
468 unsigned int pipe_cnt,
469 unsigned int dc_pipe_idx)
471 unsigned int i, pipe_idx;
472 struct pipe_ctx *pipe;
473 uint32_t phantom_vactive, phantom_bp, pstate_width_fw_delay_lines;
474 unsigned int num_dpp;
475 unsigned int vlevel = context->bw_ctx.dml.vba.VoltageLevel;
476 unsigned int dcfclk = context->bw_ctx.dml.vba.DCFCLKState[vlevel][context->bw_ctx.dml.vba.maxMpcComb];
477 unsigned int socclk = context->bw_ctx.dml.vba.SOCCLKPerState[vlevel];
478 struct vba_vars_st *vba = &context->bw_ctx.dml.vba;
479 struct dc_stream_state *main_stream = ref_pipe->stream;
481 dc_assert_fp_enabled();
483 // Find DML pipe index (pipe_idx) using dc_pipe_idx
484 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
485 pipe = &context->res_ctx.pipe_ctx[i];
490 if (i == dc_pipe_idx)
496 // Calculate lines required for pstate allow width and FW processing delays
497 pstate_width_fw_delay_lines = ((double)(dc->caps.subvp_fw_processing_delay_us +
498 dc->caps.subvp_pstate_allow_width_us) / 1000000) *
499 (ref_pipe->stream->timing.pix_clk_100hz * 100) /
500 (double)ref_pipe->stream->timing.h_total;
502 // Update clks_cfg for calling into recalculate
503 pipes[0].clks_cfg.voltage = vlevel;
504 pipes[0].clks_cfg.dcfclk_mhz = dcfclk;
505 pipes[0].clks_cfg.socclk_mhz = socclk;
507 // DML calculation for MALL region doesn't take into account FW delay
508 // and required pstate allow width for multi-display cases
509 /* Add 16 lines margin to the MALL REGION because SUB_VP_START_LINE must be aligned
510 * to 2 swaths (i.e. 16 lines)
512 phantom_vactive = get_subviewport_lines_needed_in_mall(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx) +
513 pstate_width_fw_delay_lines + dc->caps.subvp_swath_height_margin_lines;
515 // W/A for DCC corruption with certain high resolution timings.
516 // Determing if pipesplit is used. If so, add meta_row_height to the phantom vactive.
517 num_dpp = vba->NoOfDPP[vba->VoltageLevel][vba->maxMpcComb][vba->pipe_plane[pipe_idx]];
518 phantom_vactive += num_dpp > 1 ? vba->meta_row_height[vba->pipe_plane[pipe_idx]] : 0;
520 /* dc->debug.subvp_extra_lines 0 by default*/
521 phantom_vactive += dc->debug.subvp_extra_lines;
523 // For backporch of phantom pipe, use vstartup of the main pipe
524 phantom_bp = get_vstartup(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx);
526 phantom_stream->dst.y = 0;
527 phantom_stream->dst.height = phantom_vactive;
528 /* When scaling, DML provides the end to end required number of lines for MALL.
529 * dst.height is always correct for this case, but src.height is not which causes a
530 * delta between main and phantom pipe scaling outputs. Need to adjust src.height on
531 * phantom for this case.
533 phantom_stream->src.y = 0;
534 phantom_stream->src.height = (double)phantom_vactive * (double)main_stream->src.height / (double)main_stream->dst.height;
536 phantom_stream->timing.v_addressable = phantom_vactive;
537 phantom_stream->timing.v_front_porch = 1;
538 phantom_stream->timing.v_total = phantom_stream->timing.v_addressable +
539 phantom_stream->timing.v_front_porch +
540 phantom_stream->timing.v_sync_width +
542 phantom_stream->timing.flags.DSC = 0; // Don't need DSC for phantom timing
546 * dcn32_get_num_free_pipes - Calculate number of free pipes
547 * @dc: current dc state
548 * @context: new dc state
550 * This function assumes that a "used" pipe is a pipe that has
551 * both a stream and a plane assigned to it.
553 * Return: Number of free pipes available in the context
555 static unsigned int dcn32_get_num_free_pipes(struct dc *dc, struct dc_state *context)
558 unsigned int free_pipes = 0;
559 unsigned int num_pipes = 0;
561 for (i = 0; i < dc->res_pool->pipe_count; i++) {
562 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
564 if (pipe->stream && !pipe->top_pipe) {
567 pipe = pipe->bottom_pipe;
572 free_pipes = dc->res_pool->pipe_count - num_pipes;
577 * dcn32_assign_subvp_pipe - Function to decide which pipe will use Sub-VP.
578 * @dc: current dc state
579 * @context: new dc state
580 * @index: [out] dc pipe index for the pipe chosen to have phantom pipes assigned
582 * We enter this function if we are Sub-VP capable (i.e. enough pipes available)
583 * and regular P-State switching (i.e. VACTIVE/VBLANK) is not supported, or if
584 * we are forcing SubVP P-State switching on the current config.
586 * The number of pipes used for the chosen surface must be less than or equal to the
587 * number of free pipes available.
589 * In general we choose surfaces with the longest frame time first (better for SubVP + VBLANK).
590 * For multi-display cases the ActiveDRAMClockChangeMargin doesn't provide enough info on its own
591 * for determining which should be the SubVP pipe (need a way to determine if a pipe / plane doesn't
592 * support MCLK switching naturally [i.e. ACTIVE or VBLANK]).
594 * Return: True if a valid pipe assignment was found for Sub-VP. Otherwise false.
596 static bool dcn32_assign_subvp_pipe(struct dc *dc,
597 struct dc_state *context,
600 unsigned int i, pipe_idx;
601 unsigned int max_frame_time = 0;
602 bool valid_assignment_found = false;
603 unsigned int free_pipes = dcn32_get_num_free_pipes(dc, context);
604 struct vba_vars_st *vba = &context->bw_ctx.dml.vba;
606 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
607 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
608 unsigned int num_pipes = 0;
609 unsigned int refresh_rate = 0;
615 refresh_rate = (pipe->stream->timing.pix_clk_100hz * 100 +
616 pipe->stream->timing.v_total * pipe->stream->timing.h_total - 1)
617 / (double)(pipe->stream->timing.v_total * pipe->stream->timing.h_total);
618 /* SubVP pipe candidate requirements:
619 * - Refresh rate < 120hz
620 * - Not able to switch in vactive naturally (switching in active means the
621 * DET provides enough buffer to hide the P-State switch latency -- trying
622 * to combine this with SubVP can cause issues with the scheduling).
625 if (pipe->plane_state && !pipe->top_pipe && !dcn32_is_center_timing(pipe) &&
626 (!dcn32_is_psr_capable(pipe) || (context->stream_count == 1 && dc->caps.dmub_caps.subvp_psr)) &&
627 dc_state_get_pipe_subvp_type(context, pipe) == SUBVP_NONE &&
628 (refresh_rate < 120 || dcn32_allow_subvp_high_refresh_rate(dc, context, pipe)) &&
629 !pipe->plane_state->address.tmz_surface &&
630 (vba->ActiveDRAMClockChangeLatencyMarginPerState[vba->VoltageLevel][vba->maxMpcComb][vba->pipe_plane[pipe_idx]] <= 0 ||
631 (vba->ActiveDRAMClockChangeLatencyMarginPerState[vba->VoltageLevel][vba->maxMpcComb][vba->pipe_plane[pipe_idx]] > 0 &&
632 dcn32_allow_subvp_with_active_margin(pipe)))) {
635 pipe = pipe->bottom_pipe;
638 pipe = &context->res_ctx.pipe_ctx[i];
639 if (num_pipes <= free_pipes) {
640 struct dc_stream_state *stream = pipe->stream;
641 unsigned int frame_us = (stream->timing.v_total * stream->timing.h_total /
642 (double)(stream->timing.pix_clk_100hz * 100)) * 1000000;
643 if (frame_us > max_frame_time) {
645 max_frame_time = frame_us;
646 valid_assignment_found = true;
652 return valid_assignment_found;
656 * dcn32_enough_pipes_for_subvp - Function to check if there are "enough" pipes for SubVP.
657 * @dc: current dc state
658 * @context: new dc state
660 * This function returns true if there are enough free pipes
661 * to create the required phantom pipes for any given stream
662 * (that does not already have phantom pipe assigned).
664 * e.g. For a 2 stream config where the first stream uses one
665 * pipe and the second stream uses 2 pipes (i.e. pipe split),
666 * this function will return true because there is 1 remaining
667 * pipe which can be used as the phantom pipe for the non pipe
671 * True if there are enough free pipes to assign phantom pipes to at least one
672 * stream that does not already have phantom pipes assigned. Otherwise false.
674 static bool dcn32_enough_pipes_for_subvp(struct dc *dc, struct dc_state *context)
676 unsigned int i, split_cnt, free_pipes;
677 unsigned int min_pipe_split = dc->res_pool->pipe_count + 1; // init as max number of pipes + 1
678 bool subvp_possible = false;
680 for (i = 0; i < dc->res_pool->pipe_count; i++) {
681 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
683 // Find the minimum pipe split count for non SubVP pipes
684 if (resource_is_pipe_type(pipe, OPP_HEAD) &&
685 dc_state_get_pipe_subvp_type(context, pipe) == SUBVP_NONE) {
689 pipe = pipe->bottom_pipe;
692 if (split_cnt < min_pipe_split)
693 min_pipe_split = split_cnt;
697 free_pipes = dcn32_get_num_free_pipes(dc, context);
699 // SubVP only possible if at least one pipe is being used (i.e. free_pipes
700 // should not equal to the pipe_count)
701 if (free_pipes >= min_pipe_split && free_pipes < dc->res_pool->pipe_count)
702 subvp_possible = true;
704 return subvp_possible;
708 * subvp_subvp_schedulable - Determine if SubVP + SubVP config is schedulable
709 * @dc: current dc state
710 * @context: new dc state
712 * High level algorithm:
713 * 1. Find longest microschedule length (in us) between the two SubVP pipes
714 * 2. Check if the worst case overlap (VBLANK in middle of ACTIVE) for both
715 * pipes still allows for the maximum microschedule to fit in the active
716 * region for both pipes.
718 * Return: True if the SubVP + SubVP config is schedulable, false otherwise
720 static bool subvp_subvp_schedulable(struct dc *dc, struct dc_state *context)
722 struct pipe_ctx *subvp_pipes[2];
723 struct dc_stream_state *phantom = NULL;
724 uint32_t microschedule_lines = 0;
727 uint32_t max_microschedule_us = 0;
728 int32_t vactive1_us, vactive2_us, vblank1_us, vblank2_us;
730 for (i = 0; i < dc->res_pool->pipe_count; i++) {
731 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
732 uint32_t time_us = 0;
734 /* Loop to calculate the maximum microschedule time between the two SubVP pipes,
735 * and also to store the two main SubVP pipe pointers in subvp_pipes[2].
737 if (pipe->stream && pipe->plane_state && !pipe->top_pipe &&
738 dc_state_get_pipe_subvp_type(context, pipe) == SUBVP_MAIN) {
739 phantom = dc_state_get_paired_subvp_stream(context, pipe->stream);
740 microschedule_lines = (phantom->timing.v_total - phantom->timing.v_front_porch) +
741 phantom->timing.v_addressable;
743 // Round up when calculating microschedule time (+ 1 at the end)
744 time_us = (microschedule_lines * phantom->timing.h_total) /
745 (double)(phantom->timing.pix_clk_100hz * 100) * 1000000 +
746 dc->caps.subvp_prefetch_end_to_mall_start_us +
747 dc->caps.subvp_fw_processing_delay_us + 1;
748 if (time_us > max_microschedule_us)
749 max_microschedule_us = time_us;
751 subvp_pipes[index] = pipe;
754 // Maximum 2 SubVP pipes
759 vactive1_us = ((subvp_pipes[0]->stream->timing.v_addressable * subvp_pipes[0]->stream->timing.h_total) /
760 (double)(subvp_pipes[0]->stream->timing.pix_clk_100hz * 100)) * 1000000;
761 vactive2_us = ((subvp_pipes[1]->stream->timing.v_addressable * subvp_pipes[1]->stream->timing.h_total) /
762 (double)(subvp_pipes[1]->stream->timing.pix_clk_100hz * 100)) * 1000000;
763 vblank1_us = (((subvp_pipes[0]->stream->timing.v_total - subvp_pipes[0]->stream->timing.v_addressable) *
764 subvp_pipes[0]->stream->timing.h_total) /
765 (double)(subvp_pipes[0]->stream->timing.pix_clk_100hz * 100)) * 1000000;
766 vblank2_us = (((subvp_pipes[1]->stream->timing.v_total - subvp_pipes[1]->stream->timing.v_addressable) *
767 subvp_pipes[1]->stream->timing.h_total) /
768 (double)(subvp_pipes[1]->stream->timing.pix_clk_100hz * 100)) * 1000000;
770 if ((vactive1_us - vblank2_us) / 2 > max_microschedule_us &&
771 (vactive2_us - vblank1_us) / 2 > max_microschedule_us)
778 * subvp_drr_schedulable() - Determine if SubVP + DRR config is schedulable
779 * @dc: current dc state
780 * @context: new dc state
782 * High level algorithm:
783 * 1. Get timing for SubVP pipe, phantom pipe, and DRR pipe
784 * 2. Determine the frame time for the DRR display when adding required margin for MCLK switching
785 * (the margin is equal to the MALL region + DRR margin (500us))
786 * 3.If (SubVP Active - Prefetch > Stretched DRR frame + max(MALL region, Stretched DRR frame))
787 * then report the configuration as supported
789 * Return: True if the SubVP + DRR config is schedulable, false otherwise
791 static bool subvp_drr_schedulable(struct dc *dc, struct dc_state *context)
793 bool schedulable = false;
795 struct pipe_ctx *pipe = NULL;
796 struct pipe_ctx *drr_pipe = NULL;
797 struct dc_crtc_timing *main_timing = NULL;
798 struct dc_crtc_timing *phantom_timing = NULL;
799 struct dc_crtc_timing *drr_timing = NULL;
800 int16_t prefetch_us = 0;
801 int16_t mall_region_us = 0;
802 int16_t drr_frame_us = 0; // nominal frame time
803 int16_t subvp_active_us = 0;
804 int16_t stretched_drr_us = 0;
805 int16_t drr_stretched_vblank_us = 0;
806 int16_t max_vblank_mallregion = 0;
807 struct dc_stream_state *phantom_stream;
808 bool subvp_found = false;
809 bool drr_found = false;
812 for (i = 0; i < dc->res_pool->pipe_count; i++) {
813 pipe = &context->res_ctx.pipe_ctx[i];
815 // We check for master pipe, but it shouldn't matter since we only need
816 // the pipe for timing info (stream should be same for any pipe splits)
817 if (!resource_is_pipe_type(pipe, OTG_MASTER) ||
818 !resource_is_pipe_type(pipe, DPP_PIPE))
821 // Find the SubVP pipe
822 if (dc_state_get_pipe_subvp_type(context, pipe) == SUBVP_MAIN) {
829 for (i = 0; i < dc->res_pool->pipe_count; i++) {
830 drr_pipe = &context->res_ctx.pipe_ctx[i];
832 // We check for master pipe only
833 if (!resource_is_pipe_type(drr_pipe, OTG_MASTER) ||
834 !resource_is_pipe_type(drr_pipe, DPP_PIPE))
837 if (dc_state_get_pipe_subvp_type(context, drr_pipe) == SUBVP_NONE && drr_pipe->stream->ignore_msa_timing_param &&
838 (drr_pipe->stream->allow_freesync || drr_pipe->stream->vrr_active_variable || drr_pipe->stream->vrr_active_fixed)) {
844 if (subvp_found && drr_found) {
845 phantom_stream = dc_state_get_paired_subvp_stream(context, pipe->stream);
846 main_timing = &pipe->stream->timing;
847 phantom_timing = &phantom_stream->timing;
848 drr_timing = &drr_pipe->stream->timing;
849 prefetch_us = (phantom_timing->v_total - phantom_timing->v_front_porch) * phantom_timing->h_total /
850 (double)(phantom_timing->pix_clk_100hz * 100) * 1000000 +
851 dc->caps.subvp_prefetch_end_to_mall_start_us;
852 subvp_active_us = main_timing->v_addressable * main_timing->h_total /
853 (double)(main_timing->pix_clk_100hz * 100) * 1000000;
854 drr_frame_us = drr_timing->v_total * drr_timing->h_total /
855 (double)(drr_timing->pix_clk_100hz * 100) * 1000000;
856 // P-State allow width and FW delays already included phantom_timing->v_addressable
857 mall_region_us = phantom_timing->v_addressable * phantom_timing->h_total /
858 (double)(phantom_timing->pix_clk_100hz * 100) * 1000000;
859 stretched_drr_us = drr_frame_us + mall_region_us + SUBVP_DRR_MARGIN_US;
860 drr_stretched_vblank_us = (drr_timing->v_total - drr_timing->v_addressable) * drr_timing->h_total /
861 (double)(drr_timing->pix_clk_100hz * 100) * 1000000 + (stretched_drr_us - drr_frame_us);
862 max_vblank_mallregion = drr_stretched_vblank_us > mall_region_us ? drr_stretched_vblank_us : mall_region_us;
865 /* We consider SubVP + DRR schedulable if the stretched frame duration of the DRR display (i.e. the
866 * highest refresh rate + margin that can support UCLK P-State switch) passes the static analysis
867 * for VBLANK: (VACTIVE region of the SubVP pipe can fit the MALL prefetch, VBLANK frame time,
868 * and the max of (VBLANK blanking time, MALL region)).
870 if (stretched_drr_us < (1 / (double)drr_timing->min_refresh_in_uhz) * 1000000 * 1000000 &&
871 subvp_active_us - prefetch_us - stretched_drr_us - max_vblank_mallregion > 0)
879 * subvp_vblank_schedulable - Determine if SubVP + VBLANK config is schedulable
880 * @dc: current dc state
881 * @context: new dc state
883 * High level algorithm:
884 * 1. Get timing for SubVP pipe, phantom pipe, and VBLANK pipe
885 * 2. If (SubVP Active - Prefetch > Vblank Frame Time + max(MALL region, Vblank blanking time))
886 * then report the configuration as supported
887 * 3. If the VBLANK display is DRR, then take the DRR static schedulability path
889 * Return: True if the SubVP + VBLANK/DRR config is schedulable, false otherwise
891 static bool subvp_vblank_schedulable(struct dc *dc, struct dc_state *context)
893 struct pipe_ctx *pipe = NULL;
894 struct pipe_ctx *subvp_pipe = NULL;
896 bool schedulable = false;
898 uint8_t vblank_index = 0;
899 uint16_t prefetch_us = 0;
900 uint16_t mall_region_us = 0;
901 uint16_t vblank_frame_us = 0;
902 uint16_t subvp_active_us = 0;
903 uint16_t vblank_blank_us = 0;
904 uint16_t max_vblank_mallregion = 0;
905 struct dc_crtc_timing *main_timing = NULL;
906 struct dc_crtc_timing *phantom_timing = NULL;
907 struct dc_crtc_timing *vblank_timing = NULL;
908 struct dc_stream_state *phantom_stream;
909 enum mall_stream_type pipe_mall_type;
911 /* For SubVP + VBLANK/DRR cases, we assume there can only be
912 * a single VBLANK/DRR display. If DML outputs SubVP + VBLANK
913 * is supported, it is either a single VBLANK case or two VBLANK
914 * displays which are synchronized (in which case they have identical
917 for (i = 0; i < dc->res_pool->pipe_count; i++) {
918 pipe = &context->res_ctx.pipe_ctx[i];
919 pipe_mall_type = dc_state_get_pipe_subvp_type(context, pipe);
921 // We check for master pipe, but it shouldn't matter since we only need
922 // the pipe for timing info (stream should be same for any pipe splits)
923 if (!resource_is_pipe_type(pipe, OTG_MASTER) ||
924 !resource_is_pipe_type(pipe, DPP_PIPE))
927 if (!found && pipe_mall_type == SUBVP_NONE) {
928 // Found pipe which is not SubVP or Phantom (i.e. the VBLANK pipe).
933 if (!subvp_pipe && pipe_mall_type == SUBVP_MAIN)
937 phantom_stream = dc_state_get_paired_subvp_stream(context, subvp_pipe->stream);
938 main_timing = &subvp_pipe->stream->timing;
939 phantom_timing = &phantom_stream->timing;
940 vblank_timing = &context->res_ctx.pipe_ctx[vblank_index].stream->timing;
941 // Prefetch time is equal to VACTIVE + BP + VSYNC of the phantom pipe
942 // Also include the prefetch end to mallstart delay time
943 prefetch_us = (phantom_timing->v_total - phantom_timing->v_front_porch) * phantom_timing->h_total /
944 (double)(phantom_timing->pix_clk_100hz * 100) * 1000000 +
945 dc->caps.subvp_prefetch_end_to_mall_start_us;
946 // P-State allow width and FW delays already included phantom_timing->v_addressable
947 mall_region_us = phantom_timing->v_addressable * phantom_timing->h_total /
948 (double)(phantom_timing->pix_clk_100hz * 100) * 1000000;
949 vblank_frame_us = vblank_timing->v_total * vblank_timing->h_total /
950 (double)(vblank_timing->pix_clk_100hz * 100) * 1000000;
951 vblank_blank_us = (vblank_timing->v_total - vblank_timing->v_addressable) * vblank_timing->h_total /
952 (double)(vblank_timing->pix_clk_100hz * 100) * 1000000;
953 subvp_active_us = main_timing->v_addressable * main_timing->h_total /
954 (double)(main_timing->pix_clk_100hz * 100) * 1000000;
955 max_vblank_mallregion = vblank_blank_us > mall_region_us ? vblank_blank_us : mall_region_us;
957 // Schedulable if VACTIVE region of the SubVP pipe can fit the MALL prefetch, VBLANK frame time,
958 // and the max of (VBLANK blanking time, MALL region)
959 // TODO: Possibly add some margin (i.e. the below conditions should be [...] > X instead of [...] > 0)
960 if (subvp_active_us - prefetch_us - vblank_frame_us - max_vblank_mallregion > 0)
967 * subvp_subvp_admissable() - Determine if subvp + subvp config is admissible
969 * @dc: Current DC state
970 * @context: New DC state to be programmed
972 * SubVP + SubVP is admissible under the following conditions:
973 * - All SubVP pipes are < 120Hz OR
974 * - All SubVP pipes are >= 120hz
976 * Return: True if admissible, false otherwise
978 static bool subvp_subvp_admissable(struct dc *dc,
979 struct dc_state *context)
983 uint8_t subvp_count = 0;
984 uint32_t min_refresh = subvp_high_refresh_list.min_refresh, max_refresh = 0;
985 uint64_t refresh_rate = 0;
987 for (i = 0; i < dc->res_pool->pipe_count; i++) {
988 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
993 if (pipe->plane_state && !pipe->top_pipe &&
994 dc_state_get_pipe_subvp_type(context, pipe) == SUBVP_MAIN) {
995 refresh_rate = (pipe->stream->timing.pix_clk_100hz * (uint64_t)100 +
996 pipe->stream->timing.v_total * pipe->stream->timing.h_total - (uint64_t)1);
997 refresh_rate = div_u64(refresh_rate, pipe->stream->timing.v_total);
998 refresh_rate = div_u64(refresh_rate, pipe->stream->timing.h_total);
1000 if ((uint32_t)refresh_rate < min_refresh)
1001 min_refresh = (uint32_t)refresh_rate;
1002 if ((uint32_t)refresh_rate > max_refresh)
1003 max_refresh = (uint32_t)refresh_rate;
1008 if (subvp_count == 2 && ((min_refresh < 120 && max_refresh < 120) ||
1009 (min_refresh >= subvp_high_refresh_list.min_refresh &&
1010 max_refresh <= subvp_high_refresh_list.max_refresh)))
1017 * subvp_validate_static_schedulability - Check which SubVP case is calculated
1018 * and handle static analysis based on the case.
1019 * @dc: current dc state
1020 * @context: new dc state
1021 * @vlevel: Voltage level calculated by DML
1025 * 2. SubVP + VBLANK (DRR checked internally)
1026 * 3. SubVP + VACTIVE (currently unsupported)
1028 * Return: True if statically schedulable, false otherwise
1030 static bool subvp_validate_static_schedulability(struct dc *dc,
1031 struct dc_state *context,
1034 bool schedulable = false;
1035 struct vba_vars_st *vba = &context->bw_ctx.dml.vba;
1036 uint32_t i, pipe_idx;
1037 uint8_t subvp_count = 0;
1038 uint8_t vactive_count = 0;
1039 uint8_t non_subvp_pipes = 0;
1041 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
1042 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
1043 enum mall_stream_type pipe_mall_type = dc_state_get_pipe_subvp_type(context, pipe);
1048 if (pipe->plane_state && !pipe->top_pipe) {
1049 if (pipe_mall_type == SUBVP_MAIN)
1051 if (pipe_mall_type == SUBVP_NONE)
1055 // Count how many planes that aren't SubVP/phantom are capable of VACTIVE
1056 // switching (SubVP + VACTIVE unsupported). In situations where we force
1057 // SubVP for a VACTIVE plane, we don't want to increment the vactive_count.
1058 if (vba->ActiveDRAMClockChangeLatencyMarginPerState[vlevel][vba->maxMpcComb][vba->pipe_plane[pipe_idx]] > 0 &&
1059 pipe_mall_type == SUBVP_NONE) {
1065 if (subvp_count == 2) {
1066 // Static schedulability check for SubVP + SubVP case
1067 schedulable = subvp_subvp_admissable(dc, context) && subvp_subvp_schedulable(dc, context);
1068 } else if (subvp_count == 1 && non_subvp_pipes == 0) {
1069 // Single SubVP configs will be supported by default as long as it's suppported by DML
1071 } else if (subvp_count == 1 && non_subvp_pipes == 1) {
1072 if (dcn32_subvp_drr_admissable(dc, context))
1073 schedulable = subvp_drr_schedulable(dc, context);
1074 else if (dcn32_subvp_vblank_admissable(dc, context, vlevel))
1075 schedulable = subvp_vblank_schedulable(dc, context);
1076 } else if (vba->DRAMClockChangeSupport[vlevel][vba->maxMpcComb] == dm_dram_clock_change_vactive_w_mall_sub_vp &&
1077 vactive_count > 0) {
1078 // For single display SubVP cases, DML will output dm_dram_clock_change_vactive_w_mall_sub_vp by default.
1079 // We tell the difference between SubVP vs. SubVP + VACTIVE by checking the vactive_count.
1080 // SubVP + VACTIVE currently unsupported
1081 schedulable = false;
1086 static void assign_subvp_index(struct dc *dc, struct dc_state *context)
1091 for (i = 0; i < dc->res_pool->pipe_count; i++) {
1092 struct pipe_ctx *pipe_ctx = &context->res_ctx.pipe_ctx[i];
1094 if (resource_is_pipe_type(pipe_ctx, OTG_MASTER) &&
1095 dc_state_get_pipe_subvp_type(context, pipe_ctx) == SUBVP_MAIN) {
1096 pipe_ctx->subvp_index = index++;
1098 pipe_ctx->subvp_index = 0;
1103 struct pipe_slice_table {
1105 struct dc_stream_state *stream;
1107 } odm_combines[MAX_STREAMS];
1108 int odm_combine_count;
1111 struct pipe_ctx *pri_pipe;
1112 struct dc_plane_state *plane;
1114 } mpc_combines[MAX_PLANES];
1115 int mpc_combine_count;
1119 static void update_slice_table_for_stream(struct pipe_slice_table *table,
1120 struct dc_stream_state *stream, int diff)
1124 for (i = 0; i < table->odm_combine_count; i++) {
1125 if (table->odm_combines[i].stream == stream) {
1126 table->odm_combines[i].slice_count += diff;
1131 if (i == table->odm_combine_count) {
1132 table->odm_combine_count++;
1133 table->odm_combines[i].stream = stream;
1134 table->odm_combines[i].slice_count = diff;
1138 static void update_slice_table_for_plane(struct pipe_slice_table *table,
1139 struct pipe_ctx *dpp_pipe, struct dc_plane_state *plane, int diff)
1142 struct pipe_ctx *pri_dpp_pipe = resource_get_primary_dpp_pipe(dpp_pipe);
1144 for (i = 0; i < table->mpc_combine_count; i++) {
1145 if (table->mpc_combines[i].plane == plane &&
1146 table->mpc_combines[i].pri_pipe == pri_dpp_pipe) {
1147 table->mpc_combines[i].slice_count += diff;
1152 if (i == table->mpc_combine_count) {
1153 table->mpc_combine_count++;
1154 table->mpc_combines[i].plane = plane;
1155 table->mpc_combines[i].pri_pipe = pri_dpp_pipe;
1156 table->mpc_combines[i].slice_count = diff;
1160 static void init_pipe_slice_table_from_context(
1161 struct pipe_slice_table *table,
1162 struct dc_state *context)
1165 struct pipe_ctx *otg_master;
1166 struct pipe_ctx *dpp_pipes[MAX_PIPES];
1167 struct dc_stream_state *stream;
1170 memset(table, 0, sizeof(*table));
1172 for (i = 0; i < context->stream_count; i++) {
1173 stream = context->streams[i];
1174 otg_master = resource_get_otg_master_for_stream(
1175 &context->res_ctx, stream);
1176 count = resource_get_odm_slice_count(otg_master);
1177 update_slice_table_for_stream(table, stream, count);
1179 count = resource_get_dpp_pipes_for_opp_head(otg_master,
1180 &context->res_ctx, dpp_pipes);
1181 for (j = 0; j < count; j++)
1182 if (dpp_pipes[j]->plane_state)
1183 update_slice_table_for_plane(table, dpp_pipes[j],
1184 dpp_pipes[j]->plane_state, 1);
1188 static bool update_pipe_slice_table_with_split_flags(
1189 struct pipe_slice_table *table,
1191 struct dc_state *context,
1192 struct vba_vars_st *vba,
1193 int split[MAX_PIPES],
1194 bool merge[MAX_PIPES])
1196 /* NOTE: we are deprecating the support for the concept of pipe splitting
1197 * or pipe merging. Instead we append slices to the end and remove
1198 * slices from the end. The following code converts a pipe split or
1199 * merge to an append or remove operation.
1202 * When split flags describe the following pipe connection transition
1205 * pipe 0 (split=2) -> pipe 1 (split=2)
1206 * to: (old behavior)
1207 * pipe 0 -> pipe 2 -> pipe 1 -> pipe 3
1209 * the code below actually does:
1210 * pipe 0 -> pipe 1 -> pipe 2 -> pipe 3
1212 * This is the new intended behavior and for future DCNs we will retire
1213 * the old concept completely.
1215 struct pipe_ctx *pipe;
1217 int dc_pipe_idx, dml_pipe_idx = 0;
1218 bool updated = false;
1220 for (dc_pipe_idx = 0;
1221 dc_pipe_idx < dc->res_pool->pipe_count; dc_pipe_idx++) {
1222 pipe = &context->res_ctx.pipe_ctx[dc_pipe_idx];
1223 if (resource_is_pipe_type(pipe, FREE_PIPE))
1226 if (merge[dc_pipe_idx]) {
1227 if (resource_is_pipe_type(pipe, OPP_HEAD))
1228 /* merging OPP head means reducing ODM slice
1231 update_slice_table_for_stream(table, pipe->stream, -1);
1232 else if (resource_is_pipe_type(pipe, DPP_PIPE) &&
1233 resource_get_odm_slice_index(resource_get_opp_head(pipe)) == 0)
1234 /* merging DPP pipe of the first ODM slice means
1235 * reducing MPC slice count by 1
1237 update_slice_table_for_plane(table, pipe, pipe->plane_state, -1);
1241 if (split[dc_pipe_idx]) {
1242 odm = vba->ODMCombineEnabled[vba->pipe_plane[dml_pipe_idx]] !=
1243 dm_odm_combine_mode_disabled;
1244 if (odm && resource_is_pipe_type(pipe, OPP_HEAD))
1245 update_slice_table_for_stream(
1246 table, pipe->stream, split[dc_pipe_idx] - 1);
1247 else if (!odm && resource_is_pipe_type(pipe, DPP_PIPE))
1248 update_slice_table_for_plane(table, pipe,
1249 pipe->plane_state, split[dc_pipe_idx] - 1);
1257 static void update_pipes_with_slice_table(struct dc *dc, struct dc_state *context,
1258 struct pipe_slice_table *table)
1262 for (i = 0; i < table->odm_combine_count; i++)
1263 resource_update_pipes_for_stream_with_slice_count(context,
1264 dc->current_state, dc->res_pool,
1265 table->odm_combines[i].stream,
1266 table->odm_combines[i].slice_count);
1268 for (i = 0; i < table->mpc_combine_count; i++)
1269 resource_update_pipes_for_plane_with_slice_count(context,
1270 dc->current_state, dc->res_pool,
1271 table->mpc_combines[i].plane,
1272 table->mpc_combines[i].slice_count);
1275 static bool update_pipes_with_split_flags(struct dc *dc, struct dc_state *context,
1276 struct vba_vars_st *vba, int split[MAX_PIPES],
1277 bool merge[MAX_PIPES])
1279 struct pipe_slice_table slice_table;
1282 init_pipe_slice_table_from_context(&slice_table, context);
1283 updated = update_pipe_slice_table_with_split_flags(
1284 &slice_table, dc, context, vba,
1286 update_pipes_with_slice_table(dc, context, &slice_table);
1290 static bool should_apply_odm_power_optimization(struct dc *dc,
1291 struct dc_state *context, struct vba_vars_st *v, int *split,
1294 struct dc_stream_state *stream = context->streams[0];
1295 struct pipe_slice_table slice_table;
1299 * this debug flag allows us to disable ODM power optimization feature
1300 * unconditionally. we force the feature off if this is set to false.
1302 if (!dc->debug.enable_single_display_2to1_odm_policy)
1305 /* current design and test coverage is only limited to allow ODM power
1306 * optimization for single stream. Supporting it for multiple streams
1307 * use case would require additional algorithm to decide how to
1308 * optimize power consumption when there are not enough free pipes to
1309 * allocate for all the streams. This level of optimization would
1310 * require multiple attempts of revalidation to make an optimized
1311 * decision. Unfortunately We do not support revalidation flow in
1312 * current version of DML.
1314 if (context->stream_count != 1)
1318 * Our hardware doesn't support ODM for HDMI TMDS
1320 if (dc_is_hdmi_signal(stream->signal))
1324 * ODM Combine 2:1 requires horizontal timing divisible by 2 so each
1325 * ODM segment has the same size.
1327 if (!is_h_timing_divisible_by_2(stream))
1331 * No power benefits if the timing's pixel clock is not high enough to
1332 * raise display clock from minimum power state.
1334 if (stream->timing.pix_clk_100hz * 100 <= DCN3_2_VMIN_DISPCLK_HZ)
1337 if (dc->config.enable_windowed_mpo_odm) {
1339 * ODM power optimization should only be allowed if the feature
1340 * can be seamlessly toggled off within an update. This would
1341 * require that the feature is applied on top of a minimal
1342 * state. A minimal state is defined as a state validated
1343 * without the need of pipe split. Therefore, when transition to
1344 * toggle the feature off, the same stream and plane
1345 * configuration can be supported by the pipe resource in the
1346 * first ODM slice alone without the need to acquire extra
1349 init_pipe_slice_table_from_context(&slice_table, context);
1350 update_pipe_slice_table_with_split_flags(
1351 &slice_table, dc, context, v,
1353 for (i = 0; i < slice_table.mpc_combine_count; i++)
1354 if (slice_table.mpc_combines[i].slice_count > 1)
1357 for (i = 0; i < slice_table.odm_combine_count; i++)
1358 if (slice_table.odm_combines[i].slice_count > 1)
1362 * the new ODM power optimization feature reduces software
1363 * design limitation and allows ODM power optimization to be
1364 * supported even with presence of overlay planes. The new
1365 * feature is enabled based on enable_windowed_mpo_odm flag. If
1366 * the flag is not set, we limit our feature scope due to
1367 * previous software design limitation
1369 if (context->stream_status[0].plane_count != 1)
1372 if (memcmp(&context->stream_status[0].plane_states[0]->clip_rect,
1373 &stream->src, sizeof(struct rect)) != 0)
1376 if (stream->src.width >= 5120 &&
1377 stream->src.width > stream->dst.width)
1383 static void try_odm_power_optimization_and_revalidate(
1385 struct dc_state *context,
1386 display_e2e_pipe_params_st *pipes,
1389 unsigned int *vlevel,
1393 unsigned int new_vlevel;
1394 unsigned int cur_policy[MAX_PIPES];
1396 for (i = 0; i < pipe_cnt; i++) {
1397 cur_policy[i] = pipes[i].pipe.dest.odm_combine_policy;
1398 pipes[i].pipe.dest.odm_combine_policy = dm_odm_combine_policy_2to1;
1401 new_vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, pipe_cnt);
1403 if (new_vlevel < context->bw_ctx.dml.soc.num_states) {
1404 memset(split, 0, MAX_PIPES * sizeof(int));
1405 memset(merge, 0, MAX_PIPES * sizeof(bool));
1406 *vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, new_vlevel, split, merge);
1407 context->bw_ctx.dml.vba.VoltageLevel = *vlevel;
1409 for (i = 0; i < pipe_cnt; i++)
1410 pipes[i].pipe.dest.odm_combine_policy = cur_policy[i];
1414 static bool is_test_pattern_enabled(
1415 struct dc_state *context)
1419 for (i = 0; i < context->stream_count; i++) {
1420 if (context->streams[i]->test_pattern.type != DP_TEST_PATTERN_VIDEO_MODE)
1427 static void dcn32_full_validate_bw_helper(struct dc *dc,
1428 struct dc_state *context,
1429 display_e2e_pipe_params_st *pipes,
1435 struct vba_vars_st *vba = &context->bw_ctx.dml.vba;
1436 unsigned int dc_pipe_idx = 0;
1438 bool found_supported_config = false;
1439 int vlevel_temp = 0;
1441 dc_assert_fp_enabled();
1444 * DML favors voltage over p-state, but we're more interested in
1445 * supporting p-state over voltage. We can't support p-state in
1446 * prefetch mode > 0 so try capping the prefetch mode to start.
1447 * Override present for testing.
1449 if (dc->debug.dml_disallow_alternate_prefetch_modes)
1450 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final =
1451 dm_prefetch_support_uclk_fclk_and_stutter;
1453 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final =
1454 dm_prefetch_support_uclk_fclk_and_stutter_if_possible;
1456 *vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, *pipe_cnt);
1457 /* This may adjust vlevel and maxMpcComb */
1458 if (*vlevel < context->bw_ctx.dml.soc.num_states) {
1459 *vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, *vlevel, split, merge);
1460 vba->VoltageLevel = *vlevel;
1463 /* Conditions for setting up phantom pipes for SubVP:
1464 * 1. Not force disable SubVP
1465 * 2. Full update (i.e. !fast_validate)
1466 * 3. Enough pipes are available to support SubVP (TODO: Which pipes will use VACTIVE / VBLANK / SUBVP?)
1467 * 4. Display configuration passes validation
1468 * 5. (Config doesn't support MCLK in VACTIVE/VBLANK || dc->debug.force_subvp_mclk_switch)
1470 if (!dc->debug.force_disable_subvp && !dc->caps.dmub_caps.gecc_enable && dcn32_all_pipes_have_stream_and_plane(dc, context) &&
1471 !dcn32_mpo_in_use(context) && !dcn32_any_surfaces_rotated(dc, context) && !is_test_pattern_enabled(context) &&
1472 (*vlevel == context->bw_ctx.dml.soc.num_states || (vba->DRAMSpeedPerState[*vlevel] != vba->DRAMSpeedPerState[0] &&
1473 vba->DRAMClockChangeSupport[*vlevel][vba->maxMpcComb] != dm_dram_clock_change_unsupported) ||
1474 vba->DRAMClockChangeSupport[*vlevel][vba->maxMpcComb] == dm_dram_clock_change_unsupported ||
1475 dc->debug.force_subvp_mclk_switch)) {
1477 dcn32_merge_pipes_for_subvp(dc, context);
1478 memset(merge, 0, MAX_PIPES * sizeof(bool));
1480 vlevel_temp = *vlevel;
1481 /* to re-initialize viewport after the pipe merge */
1482 for (i = 0; i < dc->res_pool->pipe_count; i++) {
1483 struct pipe_ctx *pipe_ctx = &context->res_ctx.pipe_ctx[i];
1485 if (!pipe_ctx->plane_state || !pipe_ctx->stream)
1488 resource_build_scaling_params(pipe_ctx);
1491 while (!found_supported_config && dcn32_enough_pipes_for_subvp(dc, context) &&
1492 dcn32_assign_subvp_pipe(dc, context, &dc_pipe_idx)) {
1493 /* For the case where *vlevel = num_states, bandwidth validation has failed for this config.
1494 * Adding phantom pipes won't change the validation result, so change the DML input param
1495 * for P-State support before adding phantom pipes and recalculating the DML result.
1496 * However, this case is only applicable for SubVP + DRR cases because the prefetch mode
1497 * will not allow for switch in VBLANK. The DRR display must have it's VBLANK stretched
1498 * enough to support MCLK switching.
1500 if (*vlevel == context->bw_ctx.dml.soc.num_states &&
1501 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final ==
1502 dm_prefetch_support_uclk_fclk_and_stutter) {
1503 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final =
1504 dm_prefetch_support_fclk_and_stutter;
1505 /* There are params (such as FabricClock) that need to be recalculated
1506 * after validation fails (otherwise it will be 0). Calculation for
1507 * phantom vactive requires call into DML, so we must ensure all the
1508 * vba params are valid otherwise we'll get incorrect phantom vactive.
1510 *vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, *pipe_cnt);
1513 dc->res_pool->funcs->add_phantom_pipes(dc, context, pipes, *pipe_cnt, dc_pipe_idx);
1515 *pipe_cnt = dc->res_pool->funcs->populate_dml_pipes(dc, context, pipes, false);
1516 // Populate dppclk to trigger a recalculate in dml_get_voltage_level
1517 // so the phantom pipe DLG params can be assigned correctly.
1518 pipes[0].clks_cfg.dppclk_mhz = get_dppclk_calculated(&context->bw_ctx.dml, pipes, *pipe_cnt, 0);
1519 *vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, *pipe_cnt);
1521 /* Check that vlevel requested supports pstate or not
1522 * if not, select the lowest vlevel that supports it
1524 for (i = *vlevel; i < context->bw_ctx.dml.soc.num_states; i++) {
1525 if (vba->DRAMClockChangeSupport[i][vba->maxMpcComb] != dm_dram_clock_change_unsupported) {
1531 if (*vlevel < context->bw_ctx.dml.soc.num_states
1532 && subvp_validate_static_schedulability(dc, context, *vlevel))
1533 found_supported_config = true;
1534 if (found_supported_config) {
1535 // For SubVP + DRR cases, we can force the lowest vlevel that supports the mode
1536 if (dcn32_subvp_drr_admissable(dc, context) && subvp_drr_schedulable(dc, context)) {
1537 /* find lowest vlevel that supports the config */
1538 for (i = *vlevel; i >= 0; i--) {
1539 if (vba->ModeSupport[i][vba->maxMpcComb]) {
1549 if (vba->DRAMSpeedPerState[*vlevel] >= vba->DRAMSpeedPerState[vlevel_temp])
1550 found_supported_config = false;
1552 // If SubVP pipe config is unsupported (or cannot be used for UCLK switching)
1553 // remove phantom pipes and repopulate dml pipes
1554 if (!found_supported_config) {
1555 dc_state_remove_phantom_streams_and_planes(dc, context);
1556 dc_state_release_phantom_streams_and_planes(dc, context);
1557 vba->DRAMClockChangeSupport[*vlevel][vba->maxMpcComb] = dm_dram_clock_change_unsupported;
1558 *pipe_cnt = dc->res_pool->funcs->populate_dml_pipes(dc, context, pipes, false);
1560 *vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, *pipe_cnt);
1561 /* This may adjust vlevel and maxMpcComb */
1562 if (*vlevel < context->bw_ctx.dml.soc.num_states) {
1563 *vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, *vlevel, split, merge);
1564 vba->VoltageLevel = *vlevel;
1567 // Most populate phantom DLG params before programming hardware / timing for phantom pipe
1568 dcn32_helper_populate_phantom_dlg_params(dc, context, pipes, *pipe_cnt);
1570 /* Call validate_apply_pipe_split flags after calling DML getters for
1571 * phantom dlg params, or some of the VBA params indicating pipe split
1572 * can be overwritten by the getters.
1574 * When setting up SubVP config, all pipes are merged before attempting to
1575 * add phantom pipes. If pipe split (ODM / MPC) is required, both the main
1576 * and phantom pipes will be split in the regular pipe splitting sequence.
1578 memset(split, 0, MAX_PIPES * sizeof(int));
1579 memset(merge, 0, MAX_PIPES * sizeof(bool));
1580 *vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, *vlevel, split, merge);
1581 vba->VoltageLevel = *vlevel;
1582 // Note: We can't apply the phantom pipes to hardware at this time. We have to wait
1583 // until driver has acquired the DMCUB lock to do it safely.
1584 assign_subvp_index(dc, context);
1588 if (should_apply_odm_power_optimization(dc, context, vba, split, merge))
1589 try_odm_power_optimization_and_revalidate(
1590 dc, context, pipes, split, merge, vlevel, *pipe_cnt);
1594 static bool is_dtbclk_required(struct dc *dc, struct dc_state *context)
1598 for (i = 0; i < dc->res_pool->pipe_count; i++) {
1599 if (!context->res_ctx.pipe_ctx[i].stream)
1601 if (dc->link_srv->dp_is_128b_132b_signal(&context->res_ctx.pipe_ctx[i]))
1607 static void dcn20_adjust_freesync_v_startup(const struct dc_crtc_timing *dc_crtc_timing, int *vstartup_start)
1609 struct dc_crtc_timing patched_crtc_timing;
1610 uint32_t asic_blank_end = 0;
1611 uint32_t asic_blank_start = 0;
1612 uint32_t newVstartup = 0;
1614 patched_crtc_timing = *dc_crtc_timing;
1616 if (patched_crtc_timing.flags.INTERLACE == 1) {
1617 if (patched_crtc_timing.v_front_porch < 2)
1618 patched_crtc_timing.v_front_porch = 2;
1620 if (patched_crtc_timing.v_front_porch < 1)
1621 patched_crtc_timing.v_front_porch = 1;
1624 /* blank_start = frame end - front porch */
1625 asic_blank_start = patched_crtc_timing.v_total -
1626 patched_crtc_timing.v_front_porch;
1628 /* blank_end = blank_start - active */
1629 asic_blank_end = asic_blank_start -
1630 patched_crtc_timing.v_border_bottom -
1631 patched_crtc_timing.v_addressable -
1632 patched_crtc_timing.v_border_top;
1634 newVstartup = asic_blank_end + (patched_crtc_timing.v_total - asic_blank_start);
1636 *vstartup_start = ((newVstartup > *vstartup_start) ? newVstartup : *vstartup_start);
1639 static void dcn32_calculate_dlg_params(struct dc *dc, struct dc_state *context,
1640 display_e2e_pipe_params_st *pipes,
1641 int pipe_cnt, int vlevel)
1643 int i, pipe_idx, active_hubp_count = 0;
1644 bool usr_retraining_support = false;
1645 bool unbounded_req_enabled = false;
1646 struct vba_vars_st *vba = &context->bw_ctx.dml.vba;
1648 dc_assert_fp_enabled();
1650 /* Writeback MCIF_WB arbitration parameters */
1651 dc->res_pool->funcs->set_mcif_arb_params(dc, context, pipes, pipe_cnt);
1653 context->bw_ctx.bw.dcn.clk.dispclk_khz = context->bw_ctx.dml.vba.DISPCLK * 1000;
1654 context->bw_ctx.bw.dcn.clk.dcfclk_khz = context->bw_ctx.dml.vba.DCFCLK * 1000;
1655 context->bw_ctx.bw.dcn.clk.socclk_khz = context->bw_ctx.dml.vba.SOCCLK * 1000;
1656 context->bw_ctx.bw.dcn.clk.dramclk_khz = context->bw_ctx.dml.vba.DRAMSpeed * 1000 / 16;
1657 context->bw_ctx.bw.dcn.clk.dcfclk_deep_sleep_khz = context->bw_ctx.dml.vba.DCFCLKDeepSleep * 1000;
1658 context->bw_ctx.bw.dcn.clk.fclk_khz = context->bw_ctx.dml.vba.FabricClock * 1000;
1659 context->bw_ctx.bw.dcn.clk.p_state_change_support =
1660 context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb]
1661 != dm_dram_clock_change_unsupported;
1663 /* Pstate change might not be supported by hardware, but it might be
1664 * possible with firmware driven vertical blank stretching.
1666 context->bw_ctx.bw.dcn.clk.p_state_change_support |= context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching;
1668 context->bw_ctx.bw.dcn.clk.dppclk_khz = 0;
1669 context->bw_ctx.bw.dcn.clk.dtbclk_en = is_dtbclk_required(dc, context);
1670 context->bw_ctx.bw.dcn.clk.ref_dtbclk_khz = context->bw_ctx.dml.vba.DTBCLKPerState[vlevel] * 1000;
1671 if (context->bw_ctx.dml.vba.FCLKChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] == dm_fclock_change_unsupported)
1672 context->bw_ctx.bw.dcn.clk.fclk_p_state_change_support = false;
1674 context->bw_ctx.bw.dcn.clk.fclk_p_state_change_support = true;
1676 usr_retraining_support = context->bw_ctx.dml.vba.USRRetrainingSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb];
1677 ASSERT(usr_retraining_support);
1679 if (context->bw_ctx.bw.dcn.clk.dispclk_khz < dc->debug.min_disp_clk_khz)
1680 context->bw_ctx.bw.dcn.clk.dispclk_khz = dc->debug.min_disp_clk_khz;
1682 unbounded_req_enabled = get_unbounded_request_enabled(&context->bw_ctx.dml, pipes, pipe_cnt);
1684 if (unbounded_req_enabled && pipe_cnt > 1) {
1685 // Unbounded requesting should not ever be used when more than 1 pipe is enabled.
1687 unbounded_req_enabled = false;
1690 context->bw_ctx.bw.dcn.mall_ss_size_bytes = 0;
1691 context->bw_ctx.bw.dcn.mall_ss_psr_active_size_bytes = 0;
1692 context->bw_ctx.bw.dcn.mall_subvp_size_bytes = 0;
1694 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
1695 if (!context->res_ctx.pipe_ctx[i].stream)
1697 if (context->res_ctx.pipe_ctx[i].plane_state)
1698 active_hubp_count++;
1699 pipes[pipe_idx].pipe.dest.vstartup_start = get_vstartup(&context->bw_ctx.dml, pipes, pipe_cnt,
1701 pipes[pipe_idx].pipe.dest.vupdate_offset = get_vupdate_offset(&context->bw_ctx.dml, pipes, pipe_cnt,
1703 pipes[pipe_idx].pipe.dest.vupdate_width = get_vupdate_width(&context->bw_ctx.dml, pipes, pipe_cnt,
1705 pipes[pipe_idx].pipe.dest.vready_offset = get_vready_offset(&context->bw_ctx.dml, pipes, pipe_cnt,
1708 if (dc_state_get_pipe_subvp_type(context, &context->res_ctx.pipe_ctx[i]) == SUBVP_PHANTOM) {
1709 // Phantom pipe requires that DET_SIZE = 0 and no unbounded requests
1710 context->res_ctx.pipe_ctx[i].det_buffer_size_kb = 0;
1711 context->res_ctx.pipe_ctx[i].unbounded_req = false;
1713 context->res_ctx.pipe_ctx[i].det_buffer_size_kb = get_det_buffer_size_kbytes(&context->bw_ctx.dml, pipes, pipe_cnt,
1715 context->res_ctx.pipe_ctx[i].unbounded_req = unbounded_req_enabled;
1718 if (context->bw_ctx.bw.dcn.clk.dppclk_khz < pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000)
1719 context->bw_ctx.bw.dcn.clk.dppclk_khz = pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000;
1720 if (context->res_ctx.pipe_ctx[i].plane_state)
1721 context->res_ctx.pipe_ctx[i].plane_res.bw.dppclk_khz = pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000;
1723 context->res_ctx.pipe_ctx[i].plane_res.bw.dppclk_khz = 0;
1724 context->res_ctx.pipe_ctx[i].pipe_dlg_param = pipes[pipe_idx].pipe.dest;
1726 context->res_ctx.pipe_ctx[i].surface_size_in_mall_bytes = get_surface_size_in_mall(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx);
1728 if (vba->ActiveDRAMClockChangeLatencyMarginPerState[vba->VoltageLevel][vba->maxMpcComb][vba->pipe_plane[pipe_idx]] > 0)
1729 context->res_ctx.pipe_ctx[i].has_vactive_margin = true;
1731 context->res_ctx.pipe_ctx[i].has_vactive_margin = false;
1733 /* MALL Allocation Sizes */
1734 /* count from active, top pipes per plane only */
1735 if (context->res_ctx.pipe_ctx[i].stream && context->res_ctx.pipe_ctx[i].plane_state &&
1736 (context->res_ctx.pipe_ctx[i].top_pipe == NULL ||
1737 context->res_ctx.pipe_ctx[i].plane_state != context->res_ctx.pipe_ctx[i].top_pipe->plane_state) &&
1738 context->res_ctx.pipe_ctx[i].prev_odm_pipe == NULL) {
1739 /* SS: all active surfaces stored in MALL */
1740 if (dc_state_get_pipe_subvp_type(context, &context->res_ctx.pipe_ctx[i]) != SUBVP_PHANTOM) {
1741 context->bw_ctx.bw.dcn.mall_ss_size_bytes += context->res_ctx.pipe_ctx[i].surface_size_in_mall_bytes;
1743 if (context->res_ctx.pipe_ctx[i].stream->link->psr_settings.psr_version == DC_PSR_VERSION_UNSUPPORTED) {
1744 /* SS PSR On: all active surfaces part of streams not supporting PSR stored in MALL */
1745 context->bw_ctx.bw.dcn.mall_ss_psr_active_size_bytes += context->res_ctx.pipe_ctx[i].surface_size_in_mall_bytes;
1748 /* SUBVP: phantom surfaces only stored in MALL */
1749 context->bw_ctx.bw.dcn.mall_subvp_size_bytes += context->res_ctx.pipe_ctx[i].surface_size_in_mall_bytes;
1753 if (context->res_ctx.pipe_ctx[i].stream->adaptive_sync_infopacket.valid)
1754 dcn20_adjust_freesync_v_startup(
1755 &context->res_ctx.pipe_ctx[i].stream->timing,
1756 &context->res_ctx.pipe_ctx[i].pipe_dlg_param.vstartup_start);
1760 /* If DCN isn't making memory requests we can allow pstate change and lower clocks */
1761 if (!active_hubp_count) {
1762 context->bw_ctx.bw.dcn.clk.socclk_khz = 0;
1763 context->bw_ctx.bw.dcn.clk.dppclk_khz = 0;
1764 context->bw_ctx.bw.dcn.clk.dcfclk_khz = 0;
1765 context->bw_ctx.bw.dcn.clk.dcfclk_deep_sleep_khz = 0;
1766 context->bw_ctx.bw.dcn.clk.dramclk_khz = 0;
1767 context->bw_ctx.bw.dcn.clk.fclk_khz = 0;
1768 context->bw_ctx.bw.dcn.clk.p_state_change_support = true;
1769 context->bw_ctx.bw.dcn.clk.fclk_p_state_change_support = true;
1771 /*save a original dppclock copy*/
1772 context->bw_ctx.bw.dcn.clk.bw_dppclk_khz = context->bw_ctx.bw.dcn.clk.dppclk_khz;
1773 context->bw_ctx.bw.dcn.clk.bw_dispclk_khz = context->bw_ctx.bw.dcn.clk.dispclk_khz;
1774 context->bw_ctx.bw.dcn.clk.max_supported_dppclk_khz = context->bw_ctx.dml.soc.clock_limits[vlevel].dppclk_mhz
1776 context->bw_ctx.bw.dcn.clk.max_supported_dispclk_khz = context->bw_ctx.dml.soc.clock_limits[vlevel].dispclk_mhz
1779 context->bw_ctx.bw.dcn.clk.num_ways = dcn32_helper_calculate_num_ways_for_subvp(dc, context);
1781 context->bw_ctx.bw.dcn.compbuf_size_kb = context->bw_ctx.dml.ip.config_return_buffer_size_in_kbytes;
1783 for (i = 0; i < dc->res_pool->pipe_count; i++) {
1784 if (context->res_ctx.pipe_ctx[i].stream)
1785 context->bw_ctx.bw.dcn.compbuf_size_kb -= context->res_ctx.pipe_ctx[i].det_buffer_size_kb;
1788 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
1790 if (!context->res_ctx.pipe_ctx[i].stream)
1793 context->bw_ctx.dml.funcs.rq_dlg_get_dlg_reg_v2(&context->bw_ctx.dml,
1794 &context->res_ctx.pipe_ctx[i].dlg_regs, &context->res_ctx.pipe_ctx[i].ttu_regs, pipes,
1795 pipe_cnt, pipe_idx);
1797 context->bw_ctx.dml.funcs.rq_dlg_get_rq_reg_v2(&context->res_ctx.pipe_ctx[i].rq_regs,
1798 &context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx);
1803 static struct pipe_ctx *dcn32_find_split_pipe(
1805 struct dc_state *context,
1808 struct pipe_ctx *pipe = NULL;
1811 if (old_index >= 0 && context->res_ctx.pipe_ctx[old_index].stream == NULL) {
1812 pipe = &context->res_ctx.pipe_ctx[old_index];
1813 pipe->pipe_idx = old_index;
1817 for (i = dc->res_pool->pipe_count - 1; i >= 0; i--) {
1818 if (dc->current_state->res_ctx.pipe_ctx[i].top_pipe == NULL
1819 && dc->current_state->res_ctx.pipe_ctx[i].prev_odm_pipe == NULL) {
1820 if (context->res_ctx.pipe_ctx[i].stream == NULL) {
1821 pipe = &context->res_ctx.pipe_ctx[i];
1829 * May need to fix pipes getting tossed from 1 opp to another on flip
1830 * Add for debugging transient underflow during topology updates:
1834 for (i = dc->res_pool->pipe_count - 1; i >= 0; i--) {
1835 if (context->res_ctx.pipe_ctx[i].stream == NULL) {
1836 pipe = &context->res_ctx.pipe_ctx[i];
1845 static bool dcn32_split_stream_for_mpc_or_odm(
1846 const struct dc *dc,
1847 struct resource_context *res_ctx,
1848 struct pipe_ctx *pri_pipe,
1849 struct pipe_ctx *sec_pipe,
1852 int pipe_idx = sec_pipe->pipe_idx;
1853 const struct resource_pool *pool = dc->res_pool;
1855 DC_LOGGER_INIT(dc->ctx->logger);
1857 if (odm && pri_pipe->plane_state) {
1858 /* ODM + window MPO, where MPO window is on left half only */
1859 if (pri_pipe->plane_state->clip_rect.x + pri_pipe->plane_state->clip_rect.width <=
1860 pri_pipe->stream->src.x + pri_pipe->stream->src.width/2) {
1862 DC_LOG_SCALER("%s - ODM + window MPO(left). pri_pipe:%d\n",
1864 pri_pipe->pipe_idx);
1868 /* ODM + window MPO, where MPO window is on right half only */
1869 if (pri_pipe->plane_state->clip_rect.x >= pri_pipe->stream->src.x + pri_pipe->stream->src.width/2) {
1871 DC_LOG_SCALER("%s - ODM + window MPO(right). pri_pipe:%d\n",
1873 pri_pipe->pipe_idx);
1878 *sec_pipe = *pri_pipe;
1880 sec_pipe->pipe_idx = pipe_idx;
1881 sec_pipe->plane_res.mi = pool->mis[pipe_idx];
1882 sec_pipe->plane_res.hubp = pool->hubps[pipe_idx];
1883 sec_pipe->plane_res.ipp = pool->ipps[pipe_idx];
1884 sec_pipe->plane_res.xfm = pool->transforms[pipe_idx];
1885 sec_pipe->plane_res.dpp = pool->dpps[pipe_idx];
1886 sec_pipe->plane_res.mpcc_inst = pool->dpps[pipe_idx]->inst;
1887 sec_pipe->stream_res.dsc = NULL;
1889 if (pri_pipe->next_odm_pipe) {
1890 ASSERT(pri_pipe->next_odm_pipe != sec_pipe);
1891 sec_pipe->next_odm_pipe = pri_pipe->next_odm_pipe;
1892 sec_pipe->next_odm_pipe->prev_odm_pipe = sec_pipe;
1894 if (pri_pipe->top_pipe && pri_pipe->top_pipe->next_odm_pipe) {
1895 pri_pipe->top_pipe->next_odm_pipe->bottom_pipe = sec_pipe;
1896 sec_pipe->top_pipe = pri_pipe->top_pipe->next_odm_pipe;
1898 if (pri_pipe->bottom_pipe && pri_pipe->bottom_pipe->next_odm_pipe) {
1899 pri_pipe->bottom_pipe->next_odm_pipe->top_pipe = sec_pipe;
1900 sec_pipe->bottom_pipe = pri_pipe->bottom_pipe->next_odm_pipe;
1902 pri_pipe->next_odm_pipe = sec_pipe;
1903 sec_pipe->prev_odm_pipe = pri_pipe;
1904 ASSERT(sec_pipe->top_pipe == NULL);
1906 if (!sec_pipe->top_pipe)
1907 sec_pipe->stream_res.opp = pool->opps[pipe_idx];
1909 sec_pipe->stream_res.opp = sec_pipe->top_pipe->stream_res.opp;
1910 if (sec_pipe->stream->timing.flags.DSC == 1) {
1911 dcn20_acquire_dsc(dc, res_ctx, &sec_pipe->stream_res.dsc, pipe_idx);
1912 ASSERT(sec_pipe->stream_res.dsc);
1913 if (sec_pipe->stream_res.dsc == NULL)
1917 if (pri_pipe->bottom_pipe) {
1918 ASSERT(pri_pipe->bottom_pipe != sec_pipe);
1919 sec_pipe->bottom_pipe = pri_pipe->bottom_pipe;
1920 sec_pipe->bottom_pipe->top_pipe = sec_pipe;
1922 pri_pipe->bottom_pipe = sec_pipe;
1923 sec_pipe->top_pipe = pri_pipe;
1925 ASSERT(pri_pipe->plane_state);
1931 bool dcn32_internal_validate_bw(struct dc *dc,
1932 struct dc_state *context,
1933 display_e2e_pipe_params_st *pipes,
1939 bool repopulate_pipes = false;
1940 int split[MAX_PIPES] = { 0 };
1941 bool merge[MAX_PIPES] = { false };
1942 bool newly_split[MAX_PIPES] = { false };
1943 int pipe_cnt, i, pipe_idx;
1944 int vlevel = context->bw_ctx.dml.soc.num_states;
1945 struct vba_vars_st *vba = &context->bw_ctx.dml.vba;
1947 dc_assert_fp_enabled();
1953 // For each full update, remove all existing phantom pipes first
1954 dc_state_remove_phantom_streams_and_planes(dc, context);
1955 dc_state_release_phantom_streams_and_planes(dc, context);
1957 dc->res_pool->funcs->update_soc_for_wm_a(dc, context);
1959 pipe_cnt = dc->res_pool->funcs->populate_dml_pipes(dc, context, pipes, fast_validate);
1966 dml_log_pipe_params(&context->bw_ctx.dml, pipes, pipe_cnt);
1967 context->bw_ctx.dml.soc.max_vratio_pre = dcn32_determine_max_vratio_prefetch(dc, context);
1970 dcn32_full_validate_bw_helper(dc, context, pipes, &vlevel, split, merge, &pipe_cnt);
1972 if (fast_validate ||
1973 (dc->debug.dml_disallow_alternate_prefetch_modes &&
1974 (vlevel == context->bw_ctx.dml.soc.num_states ||
1975 vba->DRAMClockChangeSupport[vlevel][vba->maxMpcComb] == dm_dram_clock_change_unsupported))) {
1977 * If dml_disallow_alternate_prefetch_modes is false, then we have already
1978 * tried alternate prefetch modes during full validation.
1980 * If mode is unsupported or there is no p-state support, then
1981 * fall back to favouring voltage.
1983 * If Prefetch mode 0 failed for this config, or passed with Max UCLK, then try
1984 * to support with Prefetch mode 1 (dm_prefetch_support_fclk_and_stutter == 2)
1986 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final =
1987 dm_prefetch_support_none;
1989 context->bw_ctx.dml.validate_max_state = fast_validate;
1990 vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, pipe_cnt);
1992 context->bw_ctx.dml.validate_max_state = false;
1994 if (vlevel < context->bw_ctx.dml.soc.num_states) {
1995 memset(split, 0, sizeof(split));
1996 memset(merge, 0, sizeof(merge));
1997 vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, vlevel, split, merge);
1998 // dcn20_validate_apply_pipe_split_flags can modify voltage level outside of DML
1999 vba->VoltageLevel = vlevel;
2003 dml_log_mode_support_params(&context->bw_ctx.dml);
2005 if (vlevel == context->bw_ctx.dml.soc.num_states)
2008 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
2009 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
2010 struct pipe_ctx *mpo_pipe = pipe->bottom_pipe;
2015 if (vba->ODMCombineEnabled[vba->pipe_plane[pipe_idx]] != dm_odm_combine_mode_disabled
2016 && !dc->config.enable_windowed_mpo_odm
2017 && pipe->plane_state && mpo_pipe
2018 && memcmp(&mpo_pipe->plane_state->clip_rect,
2020 sizeof(struct rect)) != 0) {
2021 ASSERT(mpo_pipe->plane_state != pipe->plane_state);
2027 if (dc->config.enable_windowed_mpo_odm) {
2028 repopulate_pipes = update_pipes_with_split_flags(
2029 dc, context, vba, split, merge);
2031 /* the code below will be removed once windowed mpo odm is fully
2034 /* merge pipes if necessary */
2035 for (i = 0; i < dc->res_pool->pipe_count; i++) {
2036 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
2038 /*skip pipes that don't need merging*/
2042 /* if ODM merge we ignore mpc tree, mpo pipes will have their own flags */
2043 if (pipe->prev_odm_pipe) {
2044 /*split off odm pipe*/
2045 pipe->prev_odm_pipe->next_odm_pipe = pipe->next_odm_pipe;
2046 if (pipe->next_odm_pipe)
2047 pipe->next_odm_pipe->prev_odm_pipe = pipe->prev_odm_pipe;
2049 /*2:1ODM+MPC Split MPO to Single Pipe + MPC Split MPO*/
2050 if (pipe->bottom_pipe) {
2051 if (pipe->bottom_pipe->prev_odm_pipe || pipe->bottom_pipe->next_odm_pipe) {
2052 /*MPC split rules will handle this case*/
2053 pipe->bottom_pipe->top_pipe = NULL;
2055 /* when merging an ODM pipes, the bottom MPC pipe must now point to
2056 * the previous ODM pipe and its associated stream assets
2058 if (pipe->prev_odm_pipe->bottom_pipe) {
2060 pipe->bottom_pipe->top_pipe = pipe->prev_odm_pipe->bottom_pipe;
2061 pipe->prev_odm_pipe->bottom_pipe->bottom_pipe = pipe->bottom_pipe;
2064 pipe->bottom_pipe->top_pipe = pipe->prev_odm_pipe;
2065 pipe->prev_odm_pipe->bottom_pipe = pipe->bottom_pipe;
2068 memcpy(&pipe->bottom_pipe->stream_res, &pipe->bottom_pipe->top_pipe->stream_res, sizeof(struct stream_resource));
2072 if (pipe->top_pipe) {
2073 pipe->top_pipe->bottom_pipe = NULL;
2076 pipe->bottom_pipe = NULL;
2077 pipe->next_odm_pipe = NULL;
2078 pipe->plane_state = NULL;
2079 pipe->stream = NULL;
2080 pipe->top_pipe = NULL;
2081 pipe->prev_odm_pipe = NULL;
2082 if (pipe->stream_res.dsc)
2083 dcn20_release_dsc(&context->res_ctx, dc->res_pool, &pipe->stream_res.dsc);
2084 memset(&pipe->plane_res, 0, sizeof(pipe->plane_res));
2085 memset(&pipe->stream_res, 0, sizeof(pipe->stream_res));
2086 memset(&pipe->link_res, 0, sizeof(pipe->link_res));
2087 repopulate_pipes = true;
2088 } else if (pipe->top_pipe && pipe->top_pipe->plane_state == pipe->plane_state) {
2089 struct pipe_ctx *top_pipe = pipe->top_pipe;
2090 struct pipe_ctx *bottom_pipe = pipe->bottom_pipe;
2092 top_pipe->bottom_pipe = bottom_pipe;
2094 bottom_pipe->top_pipe = top_pipe;
2096 pipe->top_pipe = NULL;
2097 pipe->bottom_pipe = NULL;
2098 pipe->plane_state = NULL;
2099 pipe->stream = NULL;
2100 memset(&pipe->plane_res, 0, sizeof(pipe->plane_res));
2101 memset(&pipe->stream_res, 0, sizeof(pipe->stream_res));
2102 memset(&pipe->link_res, 0, sizeof(pipe->link_res));
2103 repopulate_pipes = true;
2105 ASSERT(0); /* Should never try to merge master pipe */
2109 for (i = 0, pipe_idx = -1; i < dc->res_pool->pipe_count; i++) {
2110 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
2111 struct pipe_ctx *old_pipe = &dc->current_state->res_ctx.pipe_ctx[i];
2112 struct pipe_ctx *hsplit_pipe = NULL;
2116 if (!pipe->stream || newly_split[i])
2120 odm = vba->ODMCombineEnabled[vba->pipe_plane[pipe_idx]] != dm_odm_combine_mode_disabled;
2122 if (!pipe->plane_state && !odm)
2127 if (split[i] == 4 && old_pipe->next_odm_pipe && old_pipe->next_odm_pipe->next_odm_pipe)
2128 old_index = old_pipe->next_odm_pipe->next_odm_pipe->pipe_idx;
2129 else if (old_pipe->next_odm_pipe)
2130 old_index = old_pipe->next_odm_pipe->pipe_idx;
2132 if (split[i] == 4 && old_pipe->bottom_pipe && old_pipe->bottom_pipe->bottom_pipe &&
2133 old_pipe->bottom_pipe->bottom_pipe->plane_state == old_pipe->plane_state)
2134 old_index = old_pipe->bottom_pipe->bottom_pipe->pipe_idx;
2135 else if (old_pipe->bottom_pipe &&
2136 old_pipe->bottom_pipe->plane_state == old_pipe->plane_state)
2137 old_index = old_pipe->bottom_pipe->pipe_idx;
2139 hsplit_pipe = dcn32_find_split_pipe(dc, context, old_index);
2140 ASSERT(hsplit_pipe);
2144 if (!dcn32_split_stream_for_mpc_or_odm(
2145 dc, &context->res_ctx,
2146 pipe, hsplit_pipe, odm))
2149 newly_split[hsplit_pipe->pipe_idx] = true;
2150 repopulate_pipes = true;
2152 if (split[i] == 4) {
2153 struct pipe_ctx *pipe_4to1;
2155 if (odm && old_pipe->next_odm_pipe)
2156 old_index = old_pipe->next_odm_pipe->pipe_idx;
2157 else if (!odm && old_pipe->bottom_pipe &&
2158 old_pipe->bottom_pipe->plane_state == old_pipe->plane_state)
2159 old_index = old_pipe->bottom_pipe->pipe_idx;
2162 pipe_4to1 = dcn32_find_split_pipe(dc, context, old_index);
2166 if (!dcn32_split_stream_for_mpc_or_odm(
2167 dc, &context->res_ctx,
2168 pipe, pipe_4to1, odm))
2170 newly_split[pipe_4to1->pipe_idx] = true;
2172 if (odm && old_pipe->next_odm_pipe && old_pipe->next_odm_pipe->next_odm_pipe
2173 && old_pipe->next_odm_pipe->next_odm_pipe->next_odm_pipe)
2174 old_index = old_pipe->next_odm_pipe->next_odm_pipe->next_odm_pipe->pipe_idx;
2175 else if (!odm && old_pipe->bottom_pipe && old_pipe->bottom_pipe->bottom_pipe &&
2176 old_pipe->bottom_pipe->bottom_pipe->bottom_pipe &&
2177 old_pipe->bottom_pipe->bottom_pipe->bottom_pipe->plane_state == old_pipe->plane_state)
2178 old_index = old_pipe->bottom_pipe->bottom_pipe->bottom_pipe->pipe_idx;
2181 pipe_4to1 = dcn32_find_split_pipe(dc, context, old_index);
2185 if (!dcn32_split_stream_for_mpc_or_odm(
2186 dc, &context->res_ctx,
2187 hsplit_pipe, pipe_4to1, odm))
2189 newly_split[pipe_4to1->pipe_idx] = true;
2192 dcn20_build_mapped_resource(dc, context, pipe->stream);
2195 for (i = 0; i < dc->res_pool->pipe_count; i++) {
2196 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
2198 if (pipe->plane_state) {
2199 if (!resource_build_scaling_params(pipe))
2205 /* Actual dsc count per stream dsc validation*/
2206 if (!dcn20_validate_dsc(dc, context)) {
2207 vba->ValidationStatus[vba->soc.num_states] = DML_FAIL_DSC_VALIDATION_FAILURE;
2211 if (repopulate_pipes) {
2212 int flag_max_mpc_comb = vba->maxMpcComb;
2213 int flag_vlevel = vlevel;
2216 pipe_cnt = dc->res_pool->funcs->populate_dml_pipes(dc, context, pipes, fast_validate);
2217 if (!dc->config.enable_windowed_mpo_odm)
2218 dcn32_update_dml_pipes_odm_policy_based_on_context(dc, context, pipes);
2220 /* repopulate_pipes = 1 means the pipes were either split or merged. In this case
2221 * we have to re-calculate the DET allocation and run through DML once more to
2222 * ensure all the params are calculated correctly. We do not need to run the
2223 * pipe split check again after this call (pipes are already split / merged).
2225 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final =
2226 dm_prefetch_support_uclk_fclk_and_stutter_if_possible;
2228 vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, pipe_cnt);
2230 if (vlevel == context->bw_ctx.dml.soc.num_states) {
2231 /* failed after DET size changes */
2233 } else if (flag_max_mpc_comb == 0 &&
2234 flag_max_mpc_comb != context->bw_ctx.dml.vba.maxMpcComb) {
2235 /* check the context constructed with pipe split flags is still valid*/
2236 bool flags_valid = false;
2237 for (i = flag_vlevel; i < context->bw_ctx.dml.soc.num_states; i++) {
2238 if (vba->ModeSupport[i][flag_max_mpc_comb]) {
2239 vba->maxMpcComb = flag_max_mpc_comb;
2240 vba->VoltageLevel = i;
2247 /* this should never happen */
2252 *vlevel_out = vlevel;
2253 *pipe_cnt_out = pipe_cnt;
2266 void dcn32_calculate_wm_and_dlg_fpu(struct dc *dc, struct dc_state *context,
2267 display_e2e_pipe_params_st *pipes,
2271 int i, pipe_idx, vlevel_temp = 0;
2272 double dcfclk = dcn3_2_soc.clock_limits[0].dcfclk_mhz;
2273 double dcfclk_from_validation = context->bw_ctx.dml.vba.DCFCLKState[vlevel][context->bw_ctx.dml.vba.maxMpcComb];
2274 double dram_speed_from_validation = context->bw_ctx.dml.vba.DRAMSpeed;
2275 double dcfclk_from_fw_based_mclk_switching = dcfclk_from_validation;
2276 bool pstate_en = context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] !=
2277 dm_dram_clock_change_unsupported;
2278 unsigned int dummy_latency_index = 0;
2279 int maxMpcComb = context->bw_ctx.dml.vba.maxMpcComb;
2280 unsigned int min_dram_speed_mts = context->bw_ctx.dml.vba.DRAMSpeed;
2281 bool subvp_in_use = dcn32_subvp_in_use(dc, context);
2282 unsigned int min_dram_speed_mts_margin;
2283 bool need_fclk_lat_as_dummy = false;
2284 bool is_subvp_p_drr = false;
2285 struct dc_stream_state *fpo_candidate_stream = NULL;
2287 dc_assert_fp_enabled();
2289 /* need to find dummy latency index for subvp */
2291 /* Override DRAMClockChangeSupport for SubVP + DRR case where the DRR cannot switch without stretching it's VBLANK */
2293 context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][maxMpcComb] = dm_dram_clock_change_vblank_w_mall_sub_vp;
2294 context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final = dm_prefetch_support_fclk_and_stutter;
2296 is_subvp_p_drr = true;
2298 dummy_latency_index = dcn32_find_dummy_latency_index_for_fw_based_mclk_switch(dc,
2299 context, pipes, pipe_cnt, vlevel);
2301 /* For DCN32/321 need to validate with fclk pstate change latency equal to dummy so prefetch is
2302 * scheduled correctly to account for dummy pstate.
2304 if (context->bw_ctx.dml.soc.fclk_change_latency_us < dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us) {
2305 need_fclk_lat_as_dummy = true;
2306 context->bw_ctx.dml.soc.fclk_change_latency_us =
2307 dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us;
2309 context->bw_ctx.dml.soc.dram_clock_change_latency_us =
2310 dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us;
2311 dcn32_internal_validate_bw(dc, context, pipes, &pipe_cnt, &vlevel, false);
2312 maxMpcComb = context->bw_ctx.dml.vba.maxMpcComb;
2313 if (is_subvp_p_drr) {
2314 context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][maxMpcComb] = dm_dram_clock_change_vblank_w_mall_sub_vp;
2318 context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching = false;
2319 for (i = 0; i < context->stream_count; i++) {
2320 if (context->streams[i])
2321 context->streams[i]->fpo_in_use = false;
2324 if (!pstate_en || (!dc->debug.disable_fpo_optimizations &&
2325 pstate_en && vlevel != 0)) {
2326 /* only when the mclk switch can not be natural, is the fw based vblank stretch attempted */
2327 fpo_candidate_stream = dcn32_can_support_mclk_switch_using_fw_based_vblank_stretch(dc, context);
2328 if (fpo_candidate_stream) {
2329 fpo_candidate_stream->fpo_in_use = true;
2330 context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching = true;
2333 if (context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching) {
2334 dummy_latency_index = dcn32_find_dummy_latency_index_for_fw_based_mclk_switch(dc,
2335 context, pipes, pipe_cnt, vlevel);
2337 /* After calling dcn30_find_dummy_latency_index_for_fw_based_mclk_switch
2338 * we reinstate the original dram_clock_change_latency_us on the context
2339 * and all variables that may have changed up to this point, except the
2340 * newly found dummy_latency_index
2342 context->bw_ctx.dml.soc.dram_clock_change_latency_us =
2343 dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us;
2344 /* For DCN32/321 need to validate with fclk pstate change latency equal to dummy so
2345 * prefetch is scheduled correctly to account for dummy pstate.
2347 if (context->bw_ctx.dml.soc.fclk_change_latency_us < dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us) {
2348 need_fclk_lat_as_dummy = true;
2349 context->bw_ctx.dml.soc.fclk_change_latency_us =
2350 dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us;
2352 dcn32_internal_validate_bw(dc, context, pipes, &pipe_cnt, &vlevel_temp, false);
2353 if (vlevel_temp < vlevel) {
2354 vlevel = vlevel_temp;
2355 maxMpcComb = context->bw_ctx.dml.vba.maxMpcComb;
2356 dcfclk_from_fw_based_mclk_switching = context->bw_ctx.dml.vba.DCFCLKState[vlevel][context->bw_ctx.dml.vba.maxMpcComb];
2358 context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][maxMpcComb] = dm_dram_clock_change_vblank;
2360 /* Restore FCLK latency and re-run validation to go back to original validation
2361 * output if we find that enabling FPO does not give us any benefit (i.e. lower
2364 context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching = false;
2365 for (i = 0; i < context->stream_count; i++) {
2366 if (context->streams[i])
2367 context->streams[i]->fpo_in_use = false;
2369 context->bw_ctx.dml.soc.fclk_change_latency_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.fclk_change_latency_us;
2370 dcn32_internal_validate_bw(dc, context, pipes, &pipe_cnt, &vlevel, false);
2376 * For Set B calculations use clocks from clock_limits[2] when available i.e. when SMU is present,
2377 * otherwise use arbitrary low value from spreadsheet for DCFCLK as lower is safer for watermark
2378 * calculations to cover bootup clocks.
2379 * DCFCLK: soc.clock_limits[2] when available
2380 * UCLK: soc.clock_limits[2] when available
2382 if (dcn3_2_soc.num_states > 2) {
2384 dcfclk = dcn3_2_soc.clock_limits[2].dcfclk_mhz;
2386 dcfclk = 615; //DCFCLK Vmin_lv
2388 pipes[0].clks_cfg.voltage = vlevel_temp;
2389 pipes[0].clks_cfg.dcfclk_mhz = dcfclk;
2390 pipes[0].clks_cfg.socclk_mhz = context->bw_ctx.dml.soc.clock_limits[vlevel_temp].socclk_mhz;
2392 if (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_B].valid) {
2393 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;
2394 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;
2395 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;
2396 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;
2398 context->bw_ctx.bw.dcn.watermarks.b.urgent_ns = get_wm_urgent(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2399 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;
2400 context->bw_ctx.bw.dcn.watermarks.b.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2401 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;
2402 context->bw_ctx.bw.dcn.watermarks.b.pte_meta_urgent_ns = get_wm_memory_trip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2403 context->bw_ctx.bw.dcn.watermarks.b.frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2404 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;
2405 context->bw_ctx.bw.dcn.watermarks.b.urgent_latency_ns = get_urgent_latency(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2406 context->bw_ctx.bw.dcn.watermarks.b.cstate_pstate.fclk_pstate_change_ns = get_fclk_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2407 context->bw_ctx.bw.dcn.watermarks.b.usr_retraining_ns = get_usr_retraining_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2411 * DCFCLK: Min, as reported by PM FW when available
2412 * UCLK : Min, as reported by PM FW when available
2413 * sr_enter_exit/sr_exit should be lower than used for DRAM (TBD after bringup or later, use as decided in Clk Mgr)
2417 if (dcn3_2_soc.num_states > 2) {
2419 dcfclk = dc->clk_mgr->bw_params->clk_table.entries[0].dcfclk_mhz;
2421 dcfclk = 615; //DCFCLK Vmin_lv
2423 pipes[0].clks_cfg.voltage = vlevel_temp;
2424 pipes[0].clks_cfg.dcfclk_mhz = dcfclk;
2425 pipes[0].clks_cfg.socclk_mhz = context->bw_ctx.dml.soc.clock_limits[vlevel_temp].socclk_mhz;
2427 if (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_D].valid) {
2428 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;
2429 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;
2430 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;
2431 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;
2433 context->bw_ctx.bw.dcn.watermarks.d.urgent_ns = get_wm_urgent(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2434 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;
2435 context->bw_ctx.bw.dcn.watermarks.d.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2436 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;
2437 context->bw_ctx.bw.dcn.watermarks.d.pte_meta_urgent_ns = get_wm_memory_trip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2438 context->bw_ctx.bw.dcn.watermarks.d.frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2439 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;
2440 context->bw_ctx.bw.dcn.watermarks.d.urgent_latency_ns = get_urgent_latency(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2441 context->bw_ctx.bw.dcn.watermarks.d.cstate_pstate.fclk_pstate_change_ns = get_fclk_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2442 context->bw_ctx.bw.dcn.watermarks.d.usr_retraining_ns = get_usr_retraining_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2445 /* Set C, for Dummy P-State:
2447 * DCFCLK: Min, as reported by PM FW, when available
2448 * UCLK : Min, as reported by PM FW, when available
2449 * pstate latency as per UCLK state dummy pstate latency
2452 // For Set A and Set C use values from validation
2453 pipes[0].clks_cfg.voltage = vlevel;
2454 pipes[0].clks_cfg.dcfclk_mhz = dcfclk_from_validation;
2455 pipes[0].clks_cfg.socclk_mhz = context->bw_ctx.dml.soc.clock_limits[vlevel].socclk_mhz;
2457 if (context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching) {
2458 pipes[0].clks_cfg.dcfclk_mhz = dcfclk_from_fw_based_mclk_switching;
2461 if (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_C].valid) {
2462 min_dram_speed_mts = dram_speed_from_validation;
2463 min_dram_speed_mts_margin = 160;
2465 context->bw_ctx.dml.soc.dram_clock_change_latency_us =
2466 dc->clk_mgr->bw_params->dummy_pstate_table[0].dummy_pstate_latency_us;
2468 if (context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][maxMpcComb] ==
2469 dm_dram_clock_change_unsupported) {
2470 int min_dram_speed_mts_offset = dc->clk_mgr->bw_params->clk_table.num_entries_per_clk.num_memclk_levels - 1;
2472 min_dram_speed_mts =
2473 dc->clk_mgr->bw_params->clk_table.entries[min_dram_speed_mts_offset].memclk_mhz * 16;
2476 if (!context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching && !subvp_in_use) {
2477 /* find largest table entry that is lower than dram speed,
2478 * but lower than DPM0 still uses DPM0
2480 for (dummy_latency_index = 3; dummy_latency_index > 0; dummy_latency_index--)
2481 if (min_dram_speed_mts + min_dram_speed_mts_margin >
2482 dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dram_speed_mts)
2486 context->bw_ctx.dml.soc.dram_clock_change_latency_us =
2487 dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us;
2489 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;
2490 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;
2491 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;
2494 context->bw_ctx.bw.dcn.watermarks.c.urgent_ns = get_wm_urgent(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2495 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;
2496 context->bw_ctx.bw.dcn.watermarks.c.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2497 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;
2498 context->bw_ctx.bw.dcn.watermarks.c.pte_meta_urgent_ns = get_wm_memory_trip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2499 context->bw_ctx.bw.dcn.watermarks.c.frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2500 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;
2501 context->bw_ctx.bw.dcn.watermarks.c.urgent_latency_ns = get_urgent_latency(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2502 /* On DCN32/321, PMFW will set PSTATE_CHANGE_TYPE = 1 (FCLK) for UCLK dummy p-state.
2503 * In this case we must program FCLK WM Set C to use the UCLK dummy p-state WM
2506 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;
2507 context->bw_ctx.bw.dcn.watermarks.c.usr_retraining_ns = get_usr_retraining_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2509 if ((!pstate_en) && (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_C].valid)) {
2510 /* The only difference between A and C is p-state latency, if p-state is not supported
2511 * with full p-state latency we want to calculate DLG based on dummy p-state latency,
2512 * Set A p-state watermark set to 0 on DCN30, when p-state unsupported, for now keep as DCN30.
2514 context->bw_ctx.bw.dcn.watermarks.a = context->bw_ctx.bw.dcn.watermarks.c;
2515 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.pstate_change_ns = 0;
2516 /* Calculate FCLK p-state change watermark based on FCLK pstate change latency in case
2517 * UCLK p-state is not supported, to avoid underflow in case FCLK pstate is supported
2519 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.fclk_pstate_change_ns = get_fclk_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2523 * DCFCLK: Min, as reported by PM FW, when available
2524 * UCLK: Min, as reported by PM FW, when available
2527 /* For set A set the correct latency values (i.e. non-dummy values) unconditionally
2529 context->bw_ctx.dml.soc.dram_clock_change_latency_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us;
2530 context->bw_ctx.dml.soc.sr_enter_plus_exit_time_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.sr_enter_plus_exit_time_us;
2531 context->bw_ctx.dml.soc.sr_exit_time_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.sr_exit_time_us;
2533 context->bw_ctx.bw.dcn.watermarks.a.urgent_ns = get_wm_urgent(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2534 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;
2535 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2536 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;
2537 context->bw_ctx.bw.dcn.watermarks.a.pte_meta_urgent_ns = get_wm_memory_trip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2538 context->bw_ctx.bw.dcn.watermarks.a.frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2539 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;
2540 context->bw_ctx.bw.dcn.watermarks.a.urgent_latency_ns = get_urgent_latency(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2541 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.fclk_pstate_change_ns = get_fclk_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2542 context->bw_ctx.bw.dcn.watermarks.a.usr_retraining_ns = get_usr_retraining_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
2545 /* Make set D = set A since we do not optimized watermarks for MALL */
2546 context->bw_ctx.bw.dcn.watermarks.d = context->bw_ctx.bw.dcn.watermarks.a;
2548 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
2549 if (!context->res_ctx.pipe_ctx[i].stream)
2552 pipes[pipe_idx].clks_cfg.dispclk_mhz = get_dispclk_calculated(&context->bw_ctx.dml, pipes, pipe_cnt);
2553 pipes[pipe_idx].clks_cfg.dppclk_mhz = get_dppclk_calculated(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx);
2555 if (dc->config.forced_clocks) {
2556 pipes[pipe_idx].clks_cfg.dispclk_mhz = context->bw_ctx.dml.soc.clock_limits[0].dispclk_mhz;
2557 pipes[pipe_idx].clks_cfg.dppclk_mhz = context->bw_ctx.dml.soc.clock_limits[0].dppclk_mhz;
2559 if (dc->debug.min_disp_clk_khz > pipes[pipe_idx].clks_cfg.dispclk_mhz * 1000)
2560 pipes[pipe_idx].clks_cfg.dispclk_mhz = dc->debug.min_disp_clk_khz / 1000.0;
2561 if (dc->debug.min_dpp_clk_khz > pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000)
2562 pipes[pipe_idx].clks_cfg.dppclk_mhz = dc->debug.min_dpp_clk_khz / 1000.0;
2567 context->perf_params.stutter_period_us = context->bw_ctx.dml.vba.StutterPeriod;
2569 /* for proper prefetch calculations, if dummy lat > fclk lat, use fclk lat = dummy lat */
2570 if (need_fclk_lat_as_dummy)
2571 context->bw_ctx.dml.soc.fclk_change_latency_us =
2572 dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us;
2574 dcn32_calculate_dlg_params(dc, context, pipes, pipe_cnt, vlevel);
2577 /* Restore full p-state latency */
2578 context->bw_ctx.dml.soc.dram_clock_change_latency_us =
2579 dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us;
2581 /* revert fclk lat changes if required */
2582 if (need_fclk_lat_as_dummy)
2583 context->bw_ctx.dml.soc.fclk_change_latency_us =
2584 dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.fclk_change_latency_us;
2587 static void dcn32_get_optimal_dcfclk_fclk_for_uclk(unsigned int uclk_mts,
2588 unsigned int *optimal_dcfclk,
2589 unsigned int *optimal_fclk)
2591 double bw_from_dram, bw_from_dram1, bw_from_dram2;
2593 bw_from_dram1 = uclk_mts * dcn3_2_soc.num_chans *
2594 dcn3_2_soc.dram_channel_width_bytes * (dcn3_2_soc.max_avg_dram_bw_use_normal_percent / 100);
2595 bw_from_dram2 = uclk_mts * dcn3_2_soc.num_chans *
2596 dcn3_2_soc.dram_channel_width_bytes * (dcn3_2_soc.max_avg_sdp_bw_use_normal_percent / 100);
2598 bw_from_dram = (bw_from_dram1 < bw_from_dram2) ? bw_from_dram1 : bw_from_dram2;
2601 *optimal_fclk = bw_from_dram /
2602 (dcn3_2_soc.fabric_datapath_to_dcn_data_return_bytes * (dcn3_2_soc.max_avg_sdp_bw_use_normal_percent / 100));
2605 *optimal_dcfclk = bw_from_dram /
2606 (dcn3_2_soc.return_bus_width_bytes * (dcn3_2_soc.max_avg_sdp_bw_use_normal_percent / 100));
2609 static void remove_entry_from_table_at_index(struct _vcs_dpi_voltage_scaling_st *table, unsigned int *num_entries,
2614 if (*num_entries == 0)
2617 for (i = index; i < *num_entries - 1; i++) {
2618 table[i] = table[i + 1];
2620 memset(&table[--(*num_entries)], 0, sizeof(struct _vcs_dpi_voltage_scaling_st));
2623 void dcn32_patch_dpm_table(struct clk_bw_params *bw_params)
2626 unsigned int max_dcfclk_mhz = 0, max_dispclk_mhz = 0, max_dppclk_mhz = 0,
2627 max_phyclk_mhz = 0, max_dtbclk_mhz = 0, max_fclk_mhz = 0, max_uclk_mhz = 0;
2629 for (i = 0; i < MAX_NUM_DPM_LVL; i++) {
2630 if (bw_params->clk_table.entries[i].dcfclk_mhz > max_dcfclk_mhz)
2631 max_dcfclk_mhz = bw_params->clk_table.entries[i].dcfclk_mhz;
2632 if (bw_params->clk_table.entries[i].fclk_mhz > max_fclk_mhz)
2633 max_fclk_mhz = bw_params->clk_table.entries[i].fclk_mhz;
2634 if (bw_params->clk_table.entries[i].memclk_mhz > max_uclk_mhz)
2635 max_uclk_mhz = bw_params->clk_table.entries[i].memclk_mhz;
2636 if (bw_params->clk_table.entries[i].dispclk_mhz > max_dispclk_mhz)
2637 max_dispclk_mhz = bw_params->clk_table.entries[i].dispclk_mhz;
2638 if (bw_params->clk_table.entries[i].dppclk_mhz > max_dppclk_mhz)
2639 max_dppclk_mhz = bw_params->clk_table.entries[i].dppclk_mhz;
2640 if (bw_params->clk_table.entries[i].phyclk_mhz > max_phyclk_mhz)
2641 max_phyclk_mhz = bw_params->clk_table.entries[i].phyclk_mhz;
2642 if (bw_params->clk_table.entries[i].dtbclk_mhz > max_dtbclk_mhz)
2643 max_dtbclk_mhz = bw_params->clk_table.entries[i].dtbclk_mhz;
2646 /* Scan through clock values we currently have and if they are 0,
2647 * then populate it with dcn3_2_soc.clock_limits[] value.
2649 * Do it for DCFCLK, DISPCLK, DTBCLK and UCLK as any of those being
2650 * 0, will cause it to skip building the clock table.
2652 if (max_dcfclk_mhz == 0)
2653 bw_params->clk_table.entries[0].dcfclk_mhz = dcn3_2_soc.clock_limits[0].dcfclk_mhz;
2654 if (max_dispclk_mhz == 0)
2655 bw_params->clk_table.entries[0].dispclk_mhz = dcn3_2_soc.clock_limits[0].dispclk_mhz;
2656 if (max_dtbclk_mhz == 0)
2657 bw_params->clk_table.entries[0].dtbclk_mhz = dcn3_2_soc.clock_limits[0].dtbclk_mhz;
2658 if (max_uclk_mhz == 0)
2659 bw_params->clk_table.entries[0].memclk_mhz = dcn3_2_soc.clock_limits[0].dram_speed_mts / 16;
2662 static void swap_table_entries(struct _vcs_dpi_voltage_scaling_st *first_entry,
2663 struct _vcs_dpi_voltage_scaling_st *second_entry)
2665 struct _vcs_dpi_voltage_scaling_st temp_entry = *first_entry;
2666 *first_entry = *second_entry;
2667 *second_entry = temp_entry;
2671 * sort_entries_with_same_bw - Sort entries sharing the same bandwidth by DCFCLK
2673 static void sort_entries_with_same_bw(struct _vcs_dpi_voltage_scaling_st *table, unsigned int *num_entries)
2675 unsigned int start_index = 0;
2676 unsigned int end_index = 0;
2677 unsigned int current_bw = 0;
2679 for (int i = 0; i < (*num_entries - 1); i++) {
2680 if (table[i].net_bw_in_kbytes_sec == table[i+1].net_bw_in_kbytes_sec) {
2681 current_bw = table[i].net_bw_in_kbytes_sec;
2685 while ((i < (*num_entries - 1)) && (table[i+1].net_bw_in_kbytes_sec == current_bw))
2689 if (start_index != end_index) {
2690 for (int j = start_index; j < end_index; j++) {
2691 for (int k = start_index; k < end_index; k++) {
2692 if (table[k].dcfclk_mhz > table[k+1].dcfclk_mhz)
2693 swap_table_entries(&table[k], &table[k+1]);
2705 * remove_inconsistent_entries - Ensure entries with the same bandwidth have MEMCLK and FCLK monotonically increasing
2706 * and remove entries that do not
2708 static void remove_inconsistent_entries(struct _vcs_dpi_voltage_scaling_st *table, unsigned int *num_entries)
2710 for (int i = 0; i < (*num_entries - 1); i++) {
2711 if (table[i].net_bw_in_kbytes_sec == table[i+1].net_bw_in_kbytes_sec) {
2712 if ((table[i].dram_speed_mts > table[i+1].dram_speed_mts) ||
2713 (table[i].fabricclk_mhz > table[i+1].fabricclk_mhz))
2714 remove_entry_from_table_at_index(table, num_entries, i);
2720 * override_max_clk_values - Overwrite the max clock frequencies with the max DC mode timings
2722 * max_clk_limit - struct containing the desired clock timings
2724 * curr_clk_limit - struct containing the timings that need to be overwritten
2725 * Return: 0 upon success, non-zero for failure
2727 static int override_max_clk_values(struct clk_limit_table_entry *max_clk_limit,
2728 struct clk_limit_table_entry *curr_clk_limit)
2730 if (NULL == max_clk_limit || NULL == curr_clk_limit)
2731 return -1; //invalid parameters
2733 //only overwrite if desired max clock frequency is initialized
2734 if (max_clk_limit->dcfclk_mhz != 0)
2735 curr_clk_limit->dcfclk_mhz = max_clk_limit->dcfclk_mhz;
2737 if (max_clk_limit->fclk_mhz != 0)
2738 curr_clk_limit->fclk_mhz = max_clk_limit->fclk_mhz;
2740 if (max_clk_limit->memclk_mhz != 0)
2741 curr_clk_limit->memclk_mhz = max_clk_limit->memclk_mhz;
2743 if (max_clk_limit->socclk_mhz != 0)
2744 curr_clk_limit->socclk_mhz = max_clk_limit->socclk_mhz;
2746 if (max_clk_limit->dtbclk_mhz != 0)
2747 curr_clk_limit->dtbclk_mhz = max_clk_limit->dtbclk_mhz;
2749 if (max_clk_limit->dispclk_mhz != 0)
2750 curr_clk_limit->dispclk_mhz = max_clk_limit->dispclk_mhz;
2755 static int build_synthetic_soc_states(bool disable_dc_mode_overwrite, struct clk_bw_params *bw_params,
2756 struct _vcs_dpi_voltage_scaling_st *table, unsigned int *num_entries)
2759 struct _vcs_dpi_voltage_scaling_st entry = {0};
2760 struct clk_limit_table_entry max_clk_data = {0};
2762 unsigned int min_dcfclk_mhz = 199, min_fclk_mhz = 299;
2764 static const unsigned int num_dcfclk_stas = 5;
2765 unsigned int dcfclk_sta_targets[DC__VOLTAGE_STATES] = {199, 615, 906, 1324, 1564};
2767 unsigned int num_uclk_dpms = 0;
2768 unsigned int num_fclk_dpms = 0;
2769 unsigned int num_dcfclk_dpms = 0;
2771 unsigned int num_dc_uclk_dpms = 0;
2772 unsigned int num_dc_fclk_dpms = 0;
2773 unsigned int num_dc_dcfclk_dpms = 0;
2775 for (i = 0; i < MAX_NUM_DPM_LVL; i++) {
2776 if (bw_params->clk_table.entries[i].dcfclk_mhz > max_clk_data.dcfclk_mhz)
2777 max_clk_data.dcfclk_mhz = bw_params->clk_table.entries[i].dcfclk_mhz;
2778 if (bw_params->clk_table.entries[i].fclk_mhz > max_clk_data.fclk_mhz)
2779 max_clk_data.fclk_mhz = bw_params->clk_table.entries[i].fclk_mhz;
2780 if (bw_params->clk_table.entries[i].memclk_mhz > max_clk_data.memclk_mhz)
2781 max_clk_data.memclk_mhz = bw_params->clk_table.entries[i].memclk_mhz;
2782 if (bw_params->clk_table.entries[i].dispclk_mhz > max_clk_data.dispclk_mhz)
2783 max_clk_data.dispclk_mhz = bw_params->clk_table.entries[i].dispclk_mhz;
2784 if (bw_params->clk_table.entries[i].dppclk_mhz > max_clk_data.dppclk_mhz)
2785 max_clk_data.dppclk_mhz = bw_params->clk_table.entries[i].dppclk_mhz;
2786 if (bw_params->clk_table.entries[i].phyclk_mhz > max_clk_data.phyclk_mhz)
2787 max_clk_data.phyclk_mhz = bw_params->clk_table.entries[i].phyclk_mhz;
2788 if (bw_params->clk_table.entries[i].dtbclk_mhz > max_clk_data.dtbclk_mhz)
2789 max_clk_data.dtbclk_mhz = bw_params->clk_table.entries[i].dtbclk_mhz;
2791 if (bw_params->clk_table.entries[i].memclk_mhz > 0) {
2793 if (bw_params->clk_table.entries[i].memclk_mhz <= bw_params->dc_mode_limit.memclk_mhz)
2796 if (bw_params->clk_table.entries[i].fclk_mhz > 0) {
2798 if (bw_params->clk_table.entries[i].fclk_mhz <= bw_params->dc_mode_limit.fclk_mhz)
2801 if (bw_params->clk_table.entries[i].dcfclk_mhz > 0) {
2803 if (bw_params->clk_table.entries[i].dcfclk_mhz <= bw_params->dc_mode_limit.dcfclk_mhz)
2804 num_dc_dcfclk_dpms++;
2808 if (!disable_dc_mode_overwrite) {
2809 //Overwrite max frequencies with max DC mode frequencies for DC mode systems
2810 override_max_clk_values(&bw_params->dc_mode_limit, &max_clk_data);
2811 num_uclk_dpms = num_dc_uclk_dpms;
2812 num_fclk_dpms = num_dc_fclk_dpms;
2813 num_dcfclk_dpms = num_dc_dcfclk_dpms;
2814 bw_params->clk_table.num_entries_per_clk.num_memclk_levels = num_uclk_dpms;
2815 bw_params->clk_table.num_entries_per_clk.num_fclk_levels = num_fclk_dpms;
2818 if (num_dcfclk_dpms > 0 && bw_params->clk_table.entries[0].fclk_mhz > min_fclk_mhz)
2819 min_fclk_mhz = bw_params->clk_table.entries[0].fclk_mhz;
2821 if (!max_clk_data.dcfclk_mhz || !max_clk_data.dispclk_mhz || !max_clk_data.dtbclk_mhz)
2824 if (max_clk_data.dppclk_mhz == 0)
2825 max_clk_data.dppclk_mhz = max_clk_data.dispclk_mhz;
2827 if (max_clk_data.fclk_mhz == 0)
2828 max_clk_data.fclk_mhz = max_clk_data.dcfclk_mhz *
2829 dcn3_2_soc.pct_ideal_sdp_bw_after_urgent /
2830 dcn3_2_soc.pct_ideal_fabric_bw_after_urgent;
2832 if (max_clk_data.phyclk_mhz == 0)
2833 max_clk_data.phyclk_mhz = dcn3_2_soc.clock_limits[0].phyclk_mhz;
2836 entry.dispclk_mhz = max_clk_data.dispclk_mhz;
2837 entry.dscclk_mhz = max_clk_data.dispclk_mhz / 3;
2838 entry.dppclk_mhz = max_clk_data.dppclk_mhz;
2839 entry.dtbclk_mhz = max_clk_data.dtbclk_mhz;
2840 entry.phyclk_mhz = max_clk_data.phyclk_mhz;
2841 entry.phyclk_d18_mhz = dcn3_2_soc.clock_limits[0].phyclk_d18_mhz;
2842 entry.phyclk_d32_mhz = dcn3_2_soc.clock_limits[0].phyclk_d32_mhz;
2844 // Insert all the DCFCLK STAs
2845 for (i = 0; i < num_dcfclk_stas; i++) {
2846 entry.dcfclk_mhz = dcfclk_sta_targets[i];
2847 entry.fabricclk_mhz = 0;
2848 entry.dram_speed_mts = 0;
2850 get_optimal_ntuple(&entry);
2851 entry.net_bw_in_kbytes_sec = calculate_net_bw_in_kbytes_sec(&entry);
2852 insert_entry_into_table_sorted(table, num_entries, &entry);
2855 // Insert the max DCFCLK
2856 entry.dcfclk_mhz = max_clk_data.dcfclk_mhz;
2857 entry.fabricclk_mhz = 0;
2858 entry.dram_speed_mts = 0;
2860 get_optimal_ntuple(&entry);
2861 entry.net_bw_in_kbytes_sec = calculate_net_bw_in_kbytes_sec(&entry);
2862 insert_entry_into_table_sorted(table, num_entries, &entry);
2864 // Insert the UCLK DPMS
2865 for (i = 0; i < num_uclk_dpms; i++) {
2866 entry.dcfclk_mhz = 0;
2867 entry.fabricclk_mhz = 0;
2868 entry.dram_speed_mts = bw_params->clk_table.entries[i].memclk_mhz * 16;
2870 get_optimal_ntuple(&entry);
2871 entry.net_bw_in_kbytes_sec = calculate_net_bw_in_kbytes_sec(&entry);
2872 insert_entry_into_table_sorted(table, num_entries, &entry);
2875 // If FCLK is coarse grained, insert individual DPMs.
2876 if (num_fclk_dpms > 2) {
2877 for (i = 0; i < num_fclk_dpms; i++) {
2878 entry.dcfclk_mhz = 0;
2879 entry.fabricclk_mhz = bw_params->clk_table.entries[i].fclk_mhz;
2880 entry.dram_speed_mts = 0;
2882 get_optimal_ntuple(&entry);
2883 entry.net_bw_in_kbytes_sec = calculate_net_bw_in_kbytes_sec(&entry);
2884 insert_entry_into_table_sorted(table, num_entries, &entry);
2887 // If FCLK fine grained, only insert max
2889 entry.dcfclk_mhz = 0;
2890 entry.fabricclk_mhz = max_clk_data.fclk_mhz;
2891 entry.dram_speed_mts = 0;
2893 get_optimal_ntuple(&entry);
2894 entry.net_bw_in_kbytes_sec = calculate_net_bw_in_kbytes_sec(&entry);
2895 insert_entry_into_table_sorted(table, num_entries, &entry);
2898 // At this point, the table contains all "points of interest" based on
2899 // DPMs from PMFW, and STAs. Table is sorted by BW, and all clock
2900 // ratios (by derate, are exact).
2902 // Remove states that require higher clocks than are supported
2903 for (i = *num_entries - 1; i >= 0 ; i--) {
2904 if (table[i].dcfclk_mhz > max_clk_data.dcfclk_mhz ||
2905 table[i].fabricclk_mhz > max_clk_data.fclk_mhz ||
2906 table[i].dram_speed_mts > max_clk_data.memclk_mhz * 16)
2907 remove_entry_from_table_at_index(table, num_entries, i);
2910 // Insert entry with all max dc limits without bandwidth matching
2911 if (!disable_dc_mode_overwrite) {
2912 struct _vcs_dpi_voltage_scaling_st max_dc_limits_entry = entry;
2914 max_dc_limits_entry.dcfclk_mhz = max_clk_data.dcfclk_mhz;
2915 max_dc_limits_entry.fabricclk_mhz = max_clk_data.fclk_mhz;
2916 max_dc_limits_entry.dram_speed_mts = max_clk_data.memclk_mhz * 16;
2918 max_dc_limits_entry.net_bw_in_kbytes_sec = calculate_net_bw_in_kbytes_sec(&max_dc_limits_entry);
2919 insert_entry_into_table_sorted(table, num_entries, &max_dc_limits_entry);
2921 sort_entries_with_same_bw(table, num_entries);
2922 remove_inconsistent_entries(table, num_entries);
2925 // At this point, the table only contains supported points of interest
2926 // it could be used as is, but some states may be redundant due to
2927 // coarse grained nature of some clocks, so we want to round up to
2928 // coarse grained DPMs and remove duplicates.
2931 for (i = *num_entries - 1; i >= 0 ; i--) {
2932 for (j = 0; j < num_uclk_dpms; j++) {
2933 if (bw_params->clk_table.entries[j].memclk_mhz * 16 >= table[i].dram_speed_mts) {
2934 table[i].dram_speed_mts = bw_params->clk_table.entries[j].memclk_mhz * 16;
2940 // If FCLK is coarse grained, round up to next DPMs
2941 if (num_fclk_dpms > 2) {
2942 for (i = *num_entries - 1; i >= 0 ; i--) {
2943 for (j = 0; j < num_fclk_dpms; j++) {
2944 if (bw_params->clk_table.entries[j].fclk_mhz >= table[i].fabricclk_mhz) {
2945 table[i].fabricclk_mhz = bw_params->clk_table.entries[j].fclk_mhz;
2951 // Otherwise, round up to minimum.
2953 for (i = *num_entries - 1; i >= 0 ; i--) {
2954 if (table[i].fabricclk_mhz < min_fclk_mhz) {
2955 table[i].fabricclk_mhz = min_fclk_mhz;
2960 // Round DCFCLKs up to minimum
2961 for (i = *num_entries - 1; i >= 0 ; i--) {
2962 if (table[i].dcfclk_mhz < min_dcfclk_mhz) {
2963 table[i].dcfclk_mhz = min_dcfclk_mhz;
2967 // Remove duplicate states, note duplicate states are always neighbouring since table is sorted.
2969 while (i < *num_entries - 1) {
2970 if (table[i].dcfclk_mhz == table[i + 1].dcfclk_mhz &&
2971 table[i].fabricclk_mhz == table[i + 1].fabricclk_mhz &&
2972 table[i].dram_speed_mts == table[i + 1].dram_speed_mts)
2973 remove_entry_from_table_at_index(table, num_entries, i + 1);
2978 // Fix up the state indicies
2979 for (i = *num_entries - 1; i >= 0 ; i--) {
2987 * dcn32_update_bw_bounding_box
2989 * This would override some dcn3_2 ip_or_soc initial parameters hardcoded from
2990 * spreadsheet with actual values as per dGPU SKU:
2991 * - with passed few options from dc->config
2992 * - with dentist_vco_frequency from Clk Mgr (currently hardcoded, but might
2993 * need to get it from PM FW)
2994 * - with passed latency values (passed in ns units) in dc-> bb override for
2995 * debugging purposes
2996 * - with passed latencies from VBIOS (in 100_ns units) if available for
2998 * - with number of DRAM channels from VBIOS (which differ for certain dGPU SKU
3000 * - clocks levels with passed clk_table entries from Clk Mgr as reported by PM
3001 * FW for different clocks (which might differ for certain dGPU SKU of the
3004 void dcn32_update_bw_bounding_box_fpu(struct dc *dc, struct clk_bw_params *bw_params)
3006 dc_assert_fp_enabled();
3008 /* Overrides from dc->config options */
3009 dcn3_2_ip.clamp_min_dcfclk = dc->config.clamp_min_dcfclk;
3011 /* Override from passed dc->bb_overrides if available*/
3012 if ((int)(dcn3_2_soc.sr_exit_time_us * 1000) != dc->bb_overrides.sr_exit_time_ns
3013 && dc->bb_overrides.sr_exit_time_ns) {
3014 dc->dml2_options.bbox_overrides.sr_exit_latency_us =
3015 dcn3_2_soc.sr_exit_time_us = dc->bb_overrides.sr_exit_time_ns / 1000.0;
3018 if ((int)(dcn3_2_soc.sr_enter_plus_exit_time_us * 1000)
3019 != dc->bb_overrides.sr_enter_plus_exit_time_ns
3020 && dc->bb_overrides.sr_enter_plus_exit_time_ns) {
3021 dc->dml2_options.bbox_overrides.sr_enter_plus_exit_latency_us =
3022 dcn3_2_soc.sr_enter_plus_exit_time_us =
3023 dc->bb_overrides.sr_enter_plus_exit_time_ns / 1000.0;
3026 if ((int)(dcn3_2_soc.urgent_latency_us * 1000) != dc->bb_overrides.urgent_latency_ns
3027 && dc->bb_overrides.urgent_latency_ns) {
3028 dcn3_2_soc.urgent_latency_us = dc->bb_overrides.urgent_latency_ns / 1000.0;
3029 dc->dml2_options.bbox_overrides.urgent_latency_us =
3030 dcn3_2_soc.urgent_latency_pixel_data_only_us = dc->bb_overrides.urgent_latency_ns / 1000.0;
3033 if ((int)(dcn3_2_soc.dram_clock_change_latency_us * 1000)
3034 != dc->bb_overrides.dram_clock_change_latency_ns
3035 && dc->bb_overrides.dram_clock_change_latency_ns) {
3036 dc->dml2_options.bbox_overrides.dram_clock_change_latency_us =
3037 dcn3_2_soc.dram_clock_change_latency_us =
3038 dc->bb_overrides.dram_clock_change_latency_ns / 1000.0;
3041 if ((int)(dcn3_2_soc.fclk_change_latency_us * 1000)
3042 != dc->bb_overrides.fclk_clock_change_latency_ns
3043 && dc->bb_overrides.fclk_clock_change_latency_ns) {
3044 dc->dml2_options.bbox_overrides.fclk_change_latency_us =
3045 dcn3_2_soc.fclk_change_latency_us =
3046 dc->bb_overrides.fclk_clock_change_latency_ns / 1000;
3049 if ((int)(dcn3_2_soc.dummy_pstate_latency_us * 1000)
3050 != dc->bb_overrides.dummy_clock_change_latency_ns
3051 && dc->bb_overrides.dummy_clock_change_latency_ns) {
3052 dcn3_2_soc.dummy_pstate_latency_us =
3053 dc->bb_overrides.dummy_clock_change_latency_ns / 1000.0;
3056 /* Override from VBIOS if VBIOS bb_info available */
3057 if (dc->ctx->dc_bios->funcs->get_soc_bb_info) {
3058 struct bp_soc_bb_info bb_info = {0};
3060 if (dc->ctx->dc_bios->funcs->get_soc_bb_info(dc->ctx->dc_bios, &bb_info) == BP_RESULT_OK) {
3061 if (bb_info.dram_clock_change_latency_100ns > 0)
3062 dc->dml2_options.bbox_overrides.dram_clock_change_latency_us =
3063 dcn3_2_soc.dram_clock_change_latency_us =
3064 bb_info.dram_clock_change_latency_100ns * 10;
3066 if (bb_info.dram_sr_enter_exit_latency_100ns > 0)
3067 dc->dml2_options.bbox_overrides.sr_enter_plus_exit_latency_us =
3068 dcn3_2_soc.sr_enter_plus_exit_time_us =
3069 bb_info.dram_sr_enter_exit_latency_100ns * 10;
3071 if (bb_info.dram_sr_exit_latency_100ns > 0)
3072 dc->dml2_options.bbox_overrides.sr_exit_latency_us =
3073 dcn3_2_soc.sr_exit_time_us =
3074 bb_info.dram_sr_exit_latency_100ns * 10;
3078 /* Override from VBIOS for num_chan */
3079 if (dc->ctx->dc_bios->vram_info.num_chans) {
3080 dc->dml2_options.bbox_overrides.dram_num_chan =
3081 dcn3_2_soc.num_chans = dc->ctx->dc_bios->vram_info.num_chans;
3082 dcn3_2_soc.mall_allocated_for_dcn_mbytes = (double)(dcn32_calc_num_avail_chans_for_mall(dc,
3083 dc->ctx->dc_bios->vram_info.num_chans) * dc->caps.mall_size_per_mem_channel);
3086 if (dc->ctx->dc_bios->vram_info.dram_channel_width_bytes)
3087 dc->dml2_options.bbox_overrides.dram_chanel_width_bytes =
3088 dcn3_2_soc.dram_channel_width_bytes = dc->ctx->dc_bios->vram_info.dram_channel_width_bytes;
3090 /* DML DSC delay factor workaround */
3091 dcn3_2_ip.dsc_delay_factor_wa = dc->debug.dsc_delay_factor_wa_x1000 / 1000.0;
3093 dcn3_2_ip.min_prefetch_in_strobe_us = dc->debug.min_prefetch_in_strobe_ns / 1000.0;
3095 /* Override dispclk_dppclk_vco_speed_mhz from Clk Mgr */
3096 dcn3_2_soc.dispclk_dppclk_vco_speed_mhz = dc->clk_mgr->dentist_vco_freq_khz / 1000.0;
3097 dc->dml.soc.dispclk_dppclk_vco_speed_mhz = dc->clk_mgr->dentist_vco_freq_khz / 1000.0;
3098 dc->dml2_options.bbox_overrides.disp_pll_vco_speed_mhz = dc->clk_mgr->dentist_vco_freq_khz / 1000.0;
3099 dc->dml2_options.bbox_overrides.xtalclk_mhz = dc->ctx->dc_bios->fw_info.pll_info.crystal_frequency / 1000.0;
3100 dc->dml2_options.bbox_overrides.dchub_refclk_mhz = dc->res_pool->ref_clocks.dchub_ref_clock_inKhz / 1000.0;
3101 dc->dml2_options.bbox_overrides.dprefclk_mhz = dc->clk_mgr->dprefclk_khz / 1000.0;
3103 /* Overrides Clock levelsfrom CLK Mgr table entries as reported by PM FW */
3104 if (bw_params->clk_table.entries[0].memclk_mhz) {
3105 if (dc->debug.use_legacy_soc_bb_mechanism) {
3106 unsigned int i = 0, j = 0, num_states = 0;
3108 unsigned int dcfclk_mhz[DC__VOLTAGE_STATES] = {0};
3109 unsigned int dram_speed_mts[DC__VOLTAGE_STATES] = {0};
3110 unsigned int optimal_uclk_for_dcfclk_sta_targets[DC__VOLTAGE_STATES] = {0};
3111 unsigned int optimal_dcfclk_for_uclk[DC__VOLTAGE_STATES] = {0};
3112 unsigned int min_dcfclk = UINT_MAX;
3113 /* Set 199 as first value in STA target array to have a minimum DCFCLK value.
3114 * For DCN32 we set min to 199 so minimum FCLK DPM0 (300Mhz can be achieved) */
3115 unsigned int dcfclk_sta_targets[DC__VOLTAGE_STATES] = {199, 615, 906, 1324, 1564};
3116 unsigned int num_dcfclk_sta_targets = 4, num_uclk_states = 0;
3117 unsigned int max_dcfclk_mhz = 0, max_dispclk_mhz = 0, max_dppclk_mhz = 0, max_phyclk_mhz = 0;
3119 for (i = 0; i < MAX_NUM_DPM_LVL; i++) {
3120 if (bw_params->clk_table.entries[i].dcfclk_mhz > max_dcfclk_mhz)
3121 max_dcfclk_mhz = bw_params->clk_table.entries[i].dcfclk_mhz;
3122 if (bw_params->clk_table.entries[i].dcfclk_mhz != 0 &&
3123 bw_params->clk_table.entries[i].dcfclk_mhz < min_dcfclk)
3124 min_dcfclk = bw_params->clk_table.entries[i].dcfclk_mhz;
3125 if (bw_params->clk_table.entries[i].dispclk_mhz > max_dispclk_mhz)
3126 max_dispclk_mhz = bw_params->clk_table.entries[i].dispclk_mhz;
3127 if (bw_params->clk_table.entries[i].dppclk_mhz > max_dppclk_mhz)
3128 max_dppclk_mhz = bw_params->clk_table.entries[i].dppclk_mhz;
3129 if (bw_params->clk_table.entries[i].phyclk_mhz > max_phyclk_mhz)
3130 max_phyclk_mhz = bw_params->clk_table.entries[i].phyclk_mhz;
3132 if (min_dcfclk > dcfclk_sta_targets[0])
3133 dcfclk_sta_targets[0] = min_dcfclk;
3134 if (!max_dcfclk_mhz)
3135 max_dcfclk_mhz = dcn3_2_soc.clock_limits[0].dcfclk_mhz;
3136 if (!max_dispclk_mhz)
3137 max_dispclk_mhz = dcn3_2_soc.clock_limits[0].dispclk_mhz;
3138 if (!max_dppclk_mhz)
3139 max_dppclk_mhz = dcn3_2_soc.clock_limits[0].dppclk_mhz;
3140 if (!max_phyclk_mhz)
3141 max_phyclk_mhz = dcn3_2_soc.clock_limits[0].phyclk_mhz;
3143 if (max_dcfclk_mhz > dcfclk_sta_targets[num_dcfclk_sta_targets-1]) {
3144 // If max DCFCLK is greater than the max DCFCLK STA target, insert into the DCFCLK STA target array
3145 dcfclk_sta_targets[num_dcfclk_sta_targets] = max_dcfclk_mhz;
3146 num_dcfclk_sta_targets++;
3147 } else if (max_dcfclk_mhz < dcfclk_sta_targets[num_dcfclk_sta_targets-1]) {
3148 // If max DCFCLK is less than the max DCFCLK STA target, cap values and remove duplicates
3149 for (i = 0; i < num_dcfclk_sta_targets; i++) {
3150 if (dcfclk_sta_targets[i] > max_dcfclk_mhz) {
3151 dcfclk_sta_targets[i] = max_dcfclk_mhz;
3155 // Update size of array since we "removed" duplicates
3156 num_dcfclk_sta_targets = i + 1;
3159 num_uclk_states = bw_params->clk_table.num_entries;
3161 // Calculate optimal dcfclk for each uclk
3162 for (i = 0; i < num_uclk_states; i++) {
3163 dcn32_get_optimal_dcfclk_fclk_for_uclk(bw_params->clk_table.entries[i].memclk_mhz * 16,
3164 &optimal_dcfclk_for_uclk[i], NULL);
3165 if (optimal_dcfclk_for_uclk[i] < bw_params->clk_table.entries[0].dcfclk_mhz) {
3166 optimal_dcfclk_for_uclk[i] = bw_params->clk_table.entries[0].dcfclk_mhz;
3170 // Calculate optimal uclk for each dcfclk sta target
3171 for (i = 0; i < num_dcfclk_sta_targets; i++) {
3172 for (j = 0; j < num_uclk_states; j++) {
3173 if (dcfclk_sta_targets[i] < optimal_dcfclk_for_uclk[j]) {
3174 optimal_uclk_for_dcfclk_sta_targets[i] =
3175 bw_params->clk_table.entries[j].memclk_mhz * 16;
3183 // create the final dcfclk and uclk table
3184 while (i < num_dcfclk_sta_targets && j < num_uclk_states && num_states < DC__VOLTAGE_STATES) {
3185 if (dcfclk_sta_targets[i] < optimal_dcfclk_for_uclk[j] && i < num_dcfclk_sta_targets) {
3186 dcfclk_mhz[num_states] = dcfclk_sta_targets[i];
3187 dram_speed_mts[num_states++] = optimal_uclk_for_dcfclk_sta_targets[i++];
3189 if (j < num_uclk_states && optimal_dcfclk_for_uclk[j] <= max_dcfclk_mhz) {
3190 dcfclk_mhz[num_states] = optimal_dcfclk_for_uclk[j];
3191 dram_speed_mts[num_states++] = bw_params->clk_table.entries[j++].memclk_mhz * 16;
3193 j = num_uclk_states;
3198 while (i < num_dcfclk_sta_targets && num_states < DC__VOLTAGE_STATES) {
3199 dcfclk_mhz[num_states] = dcfclk_sta_targets[i];
3200 dram_speed_mts[num_states++] = optimal_uclk_for_dcfclk_sta_targets[i++];
3203 while (j < num_uclk_states && num_states < DC__VOLTAGE_STATES &&
3204 optimal_dcfclk_for_uclk[j] <= max_dcfclk_mhz) {
3205 dcfclk_mhz[num_states] = optimal_dcfclk_for_uclk[j];
3206 dram_speed_mts[num_states++] = bw_params->clk_table.entries[j++].memclk_mhz * 16;
3209 dcn3_2_soc.num_states = num_states;
3210 for (i = 0; i < dcn3_2_soc.num_states; i++) {
3211 dcn3_2_soc.clock_limits[i].state = i;
3212 dcn3_2_soc.clock_limits[i].dcfclk_mhz = dcfclk_mhz[i];
3213 dcn3_2_soc.clock_limits[i].fabricclk_mhz = dcfclk_mhz[i];
3215 /* Fill all states with max values of all these clocks */
3216 dcn3_2_soc.clock_limits[i].dispclk_mhz = max_dispclk_mhz;
3217 dcn3_2_soc.clock_limits[i].dppclk_mhz = max_dppclk_mhz;
3218 dcn3_2_soc.clock_limits[i].phyclk_mhz = max_phyclk_mhz;
3219 dcn3_2_soc.clock_limits[i].dscclk_mhz = max_dispclk_mhz / 3;
3221 /* Populate from bw_params for DTBCLK, SOCCLK */
3223 if (!bw_params->clk_table.entries[i].dtbclk_mhz) {
3224 dcn3_2_soc.clock_limits[i].dtbclk_mhz = dcn3_2_soc.clock_limits[i-1].dtbclk_mhz;
3226 dcn3_2_soc.clock_limits[i].dtbclk_mhz = bw_params->clk_table.entries[i].dtbclk_mhz;
3228 } else if (bw_params->clk_table.entries[i].dtbclk_mhz) {
3229 dcn3_2_soc.clock_limits[i].dtbclk_mhz = bw_params->clk_table.entries[i].dtbclk_mhz;
3232 if (!bw_params->clk_table.entries[i].socclk_mhz && i > 0)
3233 dcn3_2_soc.clock_limits[i].socclk_mhz = dcn3_2_soc.clock_limits[i-1].socclk_mhz;
3235 dcn3_2_soc.clock_limits[i].socclk_mhz = bw_params->clk_table.entries[i].socclk_mhz;
3237 if (!dram_speed_mts[i] && i > 0)
3238 dcn3_2_soc.clock_limits[i].dram_speed_mts = dcn3_2_soc.clock_limits[i-1].dram_speed_mts;
3240 dcn3_2_soc.clock_limits[i].dram_speed_mts = dram_speed_mts[i];
3242 /* These clocks cannot come from bw_params, always fill from dcn3_2_soc[0] */
3243 /* PHYCLK_D18, PHYCLK_D32 */
3244 dcn3_2_soc.clock_limits[i].phyclk_d18_mhz = dcn3_2_soc.clock_limits[0].phyclk_d18_mhz;
3245 dcn3_2_soc.clock_limits[i].phyclk_d32_mhz = dcn3_2_soc.clock_limits[0].phyclk_d32_mhz;
3248 build_synthetic_soc_states(dc->debug.disable_dc_mode_overwrite, bw_params,
3249 dcn3_2_soc.clock_limits, &dcn3_2_soc.num_states);
3252 /* Re-init DML with updated bb */
3253 dml_init_instance(&dc->dml, &dcn3_2_soc, &dcn3_2_ip, DML_PROJECT_DCN32);
3254 if (dc->current_state)
3255 dml_init_instance(&dc->current_state->bw_ctx.dml, &dcn3_2_soc, &dcn3_2_ip, DML_PROJECT_DCN32);
3258 if (dc->clk_mgr->bw_params->clk_table.num_entries > 1) {
3261 dc->dml2_options.bbox_overrides.clks_table.num_states = dc->clk_mgr->bw_params->clk_table.num_entries;
3263 dc->dml2_options.bbox_overrides.clks_table.num_entries_per_clk.num_dcfclk_levels =
3264 dc->clk_mgr->bw_params->clk_table.num_entries_per_clk.num_dcfclk_levels;
3266 dc->dml2_options.bbox_overrides.clks_table.num_entries_per_clk.num_fclk_levels =
3267 dc->clk_mgr->bw_params->clk_table.num_entries_per_clk.num_fclk_levels;
3269 dc->dml2_options.bbox_overrides.clks_table.num_entries_per_clk.num_memclk_levels =
3270 dc->clk_mgr->bw_params->clk_table.num_entries_per_clk.num_memclk_levels;
3272 dc->dml2_options.bbox_overrides.clks_table.num_entries_per_clk.num_socclk_levels =
3273 dc->clk_mgr->bw_params->clk_table.num_entries_per_clk.num_socclk_levels;
3275 dc->dml2_options.bbox_overrides.clks_table.num_entries_per_clk.num_dtbclk_levels =
3276 dc->clk_mgr->bw_params->clk_table.num_entries_per_clk.num_dtbclk_levels;
3278 dc->dml2_options.bbox_overrides.clks_table.num_entries_per_clk.num_dispclk_levels =
3279 dc->clk_mgr->bw_params->clk_table.num_entries_per_clk.num_dispclk_levels;
3281 dc->dml2_options.bbox_overrides.clks_table.num_entries_per_clk.num_dppclk_levels =
3282 dc->clk_mgr->bw_params->clk_table.num_entries_per_clk.num_dppclk_levels;
3284 for (i = 0; i < dc->clk_mgr->bw_params->clk_table.num_entries_per_clk.num_dcfclk_levels; i++) {
3285 if (dc->clk_mgr->bw_params->clk_table.entries[i].dcfclk_mhz)
3286 dc->dml2_options.bbox_overrides.clks_table.clk_entries[i].dcfclk_mhz =
3287 dc->clk_mgr->bw_params->clk_table.entries[i].dcfclk_mhz;
3290 for (i = 0; i < dc->clk_mgr->bw_params->clk_table.num_entries_per_clk.num_fclk_levels; i++) {
3291 if (dc->clk_mgr->bw_params->clk_table.entries[i].fclk_mhz)
3292 dc->dml2_options.bbox_overrides.clks_table.clk_entries[i].fclk_mhz =
3293 dc->clk_mgr->bw_params->clk_table.entries[i].fclk_mhz;
3296 for (i = 0; i < dc->clk_mgr->bw_params->clk_table.num_entries_per_clk.num_memclk_levels; i++) {
3297 if (dc->clk_mgr->bw_params->clk_table.entries[i].memclk_mhz)
3298 dc->dml2_options.bbox_overrides.clks_table.clk_entries[i].memclk_mhz =
3299 dc->clk_mgr->bw_params->clk_table.entries[i].memclk_mhz;
3302 for (i = 0; i < dc->clk_mgr->bw_params->clk_table.num_entries_per_clk.num_socclk_levels; i++) {
3303 if (dc->clk_mgr->bw_params->clk_table.entries[i].socclk_mhz)
3304 dc->dml2_options.bbox_overrides.clks_table.clk_entries[i].socclk_mhz =
3305 dc->clk_mgr->bw_params->clk_table.entries[i].socclk_mhz;
3308 for (i = 0; i < dc->clk_mgr->bw_params->clk_table.num_entries_per_clk.num_dtbclk_levels; i++) {
3309 if (dc->clk_mgr->bw_params->clk_table.entries[i].dtbclk_mhz)
3310 dc->dml2_options.bbox_overrides.clks_table.clk_entries[i].dtbclk_mhz =
3311 dc->clk_mgr->bw_params->clk_table.entries[i].dtbclk_mhz;
3314 for (i = 0; i < dc->clk_mgr->bw_params->clk_table.num_entries_per_clk.num_dispclk_levels; i++) {
3315 if (dc->clk_mgr->bw_params->clk_table.entries[i].dispclk_mhz) {
3316 dc->dml2_options.bbox_overrides.clks_table.clk_entries[i].dispclk_mhz =
3317 dc->clk_mgr->bw_params->clk_table.entries[i].dispclk_mhz;
3318 dc->dml2_options.bbox_overrides.clks_table.clk_entries[i].dppclk_mhz =
3319 dc->clk_mgr->bw_params->clk_table.entries[i].dispclk_mhz;
3325 void dcn32_zero_pipe_dcc_fraction(display_e2e_pipe_params_st *pipes,
3328 dc_assert_fp_enabled();
3330 pipes[pipe_cnt].pipe.src.dcc_fraction_of_zs_req_luma = 0;
3331 pipes[pipe_cnt].pipe.src.dcc_fraction_of_zs_req_chroma = 0;
3334 bool dcn32_allow_subvp_with_active_margin(struct pipe_ctx *pipe)
3337 uint32_t refresh_rate = 0;
3338 uint32_t min_refresh = subvp_active_margin_list.min_refresh;
3339 uint32_t max_refresh = subvp_active_margin_list.max_refresh;
3342 for (i = 0; i < SUBVP_ACTIVE_MARGIN_LIST_LEN; i++) {
3343 uint32_t width = subvp_active_margin_list.res[i].width;
3344 uint32_t height = subvp_active_margin_list.res[i].height;
3346 refresh_rate = (pipe->stream->timing.pix_clk_100hz * (uint64_t)100 +
3347 pipe->stream->timing.v_total * pipe->stream->timing.h_total - (uint64_t)1);
3348 refresh_rate = div_u64(refresh_rate, pipe->stream->timing.v_total);
3349 refresh_rate = div_u64(refresh_rate, pipe->stream->timing.h_total);
3351 if (refresh_rate >= min_refresh && refresh_rate <= max_refresh &&
3352 dcn32_check_native_scaling_for_res(pipe, width, height)) {
3361 * dcn32_allow_subvp_high_refresh_rate: Determine if the high refresh rate config will allow subvp
3363 * @dc: Current DC state
3364 * @context: New DC state to be programmed
3365 * @pipe: Pipe to be considered for use in subvp
3367 * On high refresh rate display configs, we will allow subvp under the following conditions:
3368 * 1. Resolution is 3840x2160, 3440x1440, or 2560x1440
3369 * 2. Refresh rate is between 120hz - 165hz
3371 * 4. Freesync is inactive
3372 * 5. For single display cases, freesync must be disabled
3374 * Return: True if pipe can be used for subvp, false otherwise
3376 bool dcn32_allow_subvp_high_refresh_rate(struct dc *dc, struct dc_state *context, struct pipe_ctx *pipe)
3379 uint32_t refresh_rate = 0;
3380 uint32_t subvp_min_refresh = subvp_high_refresh_list.min_refresh;
3381 uint32_t subvp_max_refresh = subvp_high_refresh_list.max_refresh;
3382 uint32_t min_refresh = subvp_max_refresh;
3385 /* Only allow SubVP on high refresh displays if all connected displays
3386 * are considered "high refresh" (i.e. >= 120hz). We do not want to
3387 * allow combinations such as 120hz (SubVP) + 60hz (SubVP).
3389 for (i = 0; i < dc->res_pool->pipe_count; i++) {
3390 struct pipe_ctx *pipe_ctx = &context->res_ctx.pipe_ctx[i];
3392 if (!pipe_ctx->stream)
3394 refresh_rate = (pipe_ctx->stream->timing.pix_clk_100hz * 100 +
3395 pipe_ctx->stream->timing.v_total * pipe_ctx->stream->timing.h_total - 1)
3396 / (double)(pipe_ctx->stream->timing.v_total * pipe_ctx->stream->timing.h_total);
3398 if (refresh_rate < min_refresh)
3399 min_refresh = refresh_rate;
3402 if (!dc->debug.disable_subvp_high_refresh && min_refresh >= subvp_min_refresh && pipe->stream &&
3403 pipe->plane_state && !(pipe->stream->vrr_active_variable || pipe->stream->vrr_active_fixed)) {
3404 refresh_rate = (pipe->stream->timing.pix_clk_100hz * 100 +
3405 pipe->stream->timing.v_total * pipe->stream->timing.h_total - 1)
3406 / (double)(pipe->stream->timing.v_total * pipe->stream->timing.h_total);
3407 if (refresh_rate >= subvp_min_refresh && refresh_rate <= subvp_max_refresh) {
3408 for (i = 0; i < SUBVP_HIGH_REFRESH_LIST_LEN; i++) {
3409 uint32_t width = subvp_high_refresh_list.res[i].width;
3410 uint32_t height = subvp_high_refresh_list.res[i].height;
3412 if (dcn32_check_native_scaling_for_res(pipe, width, height)) {
3413 if ((context->stream_count == 1 && !pipe->stream->allow_freesync) || context->stream_count > 1) {
3425 * dcn32_determine_max_vratio_prefetch: Determine max Vratio for prefetch by driver policy
3427 * @dc: Current DC state
3428 * @context: New DC state to be programmed
3430 * Return: Max vratio for prefetch
3432 double dcn32_determine_max_vratio_prefetch(struct dc *dc, struct dc_state *context)
3434 double max_vratio_pre = __DML_MAX_BW_RATIO_PRE__; // Default value is 4
3437 /* For single display MPO configs, allow the max vratio to be 8
3438 * if any plane is YUV420 format
3440 if (context->stream_count == 1 && context->stream_status[0].plane_count > 1) {
3441 for (i = 0; i < context->stream_status[0].plane_count; i++) {
3442 if (context->stream_status[0].plane_states[i]->format == SURFACE_PIXEL_FORMAT_VIDEO_420_YCbCr ||
3443 context->stream_status[0].plane_states[i]->format == SURFACE_PIXEL_FORMAT_VIDEO_420_YCrCb) {
3444 max_vratio_pre = __DML_MAX_VRATIO_PRE__;
3448 return max_vratio_pre;
3452 * dcn32_assign_fpo_vactive_candidate - Assign the FPO stream candidate for FPO + VActive case
3454 * This function chooses the FPO candidate stream for FPO + VActive cases (2 stream config).
3455 * For FPO + VAtive cases, the assumption is that one display has ActiveMargin > 0, and the
3456 * other display has ActiveMargin <= 0. This function will choose the pipe/stream that has
3457 * ActiveMargin <= 0 to be the FPO stream candidate if found.
3460 * @dc: current dc state
3461 * @context: new dc state
3462 * @fpo_candidate_stream: pointer to FPO stream candidate if one is found
3466 void dcn32_assign_fpo_vactive_candidate(struct dc *dc, const struct dc_state *context, struct dc_stream_state **fpo_candidate_stream)
3468 unsigned int i, pipe_idx;
3469 const struct vba_vars_st *vba = &context->bw_ctx.dml.vba;
3471 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
3472 const struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
3474 /* In DCN32/321, FPO uses per-pipe P-State force.
3475 * If there's no planes, HUBP is power gated and
3476 * therefore programming UCLK_PSTATE_FORCE does
3477 * nothing (P-State will always be asserted naturally
3478 * on a pipe that has HUBP power gated. Therefore we
3479 * only want to enable FPO if the FPO pipe has both
3480 * a stream and a plane.
3482 if (!pipe->stream || !pipe->plane_state)
3485 if (vba->ActiveDRAMClockChangeLatencyMarginPerState[vba->VoltageLevel][vba->maxMpcComb][vba->pipe_plane[pipe_idx]] <= 0) {
3486 *fpo_candidate_stream = pipe->stream;
3494 * dcn32_find_vactive_pipe - Determines if the config has a pipe that can switch in VACTIVE
3496 * @dc: current dc state
3497 * @context: new dc state
3498 * @vactive_margin_req_us: The vactive marign required for a vactive pipe to be considered "found"
3500 * Return: True if VACTIVE display is found, false otherwise
3502 bool dcn32_find_vactive_pipe(struct dc *dc, const struct dc_state *context, uint32_t vactive_margin_req_us)
3504 unsigned int i, pipe_idx;
3505 const struct vba_vars_st *vba = &context->bw_ctx.dml.vba;
3506 bool vactive_found = false;
3507 unsigned int blank_us = 0;
3509 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
3510 const struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
3515 blank_us = ((pipe->stream->timing.v_total - pipe->stream->timing.v_addressable) * pipe->stream->timing.h_total /
3516 (double)(pipe->stream->timing.pix_clk_100hz * 100)) * 1000000;
3517 if (vba->ActiveDRAMClockChangeLatencyMarginPerState[vba->VoltageLevel][vba->maxMpcComb][vba->pipe_plane[pipe_idx]] >= vactive_margin_req_us &&
3518 !(pipe->stream->vrr_active_variable || pipe->stream->vrr_active_fixed) && blank_us < dc->debug.fpo_vactive_max_blank_us) {
3519 vactive_found = true;
3524 return vactive_found;
3527 void dcn32_set_clock_limits(const struct _vcs_dpi_soc_bounding_box_st *soc_bb)
3529 dc_assert_fp_enabled();
3530 dcn3_2_soc.clock_limits[0].dcfclk_mhz = 1200.0;
3533 void dcn32_override_min_req_memclk(struct dc *dc, struct dc_state *context)
3535 // WA: restrict FPO and SubVP to use first non-strobe mode (DCN32 BW issue)
3536 if ((context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching || dcn32_subvp_in_use(dc, context)) &&
3537 dc->dml.soc.num_chans <= 8) {
3538 int num_mclk_levels = dc->clk_mgr->bw_params->clk_table.num_entries_per_clk.num_memclk_levels;
3540 if (context->bw_ctx.dml.vba.DRAMSpeed <= dc->clk_mgr->bw_params->clk_table.entries[0].memclk_mhz * 16 &&
3541 num_mclk_levels > 1) {
3542 context->bw_ctx.dml.vba.DRAMSpeed = dc->clk_mgr->bw_params->clk_table.entries[1].memclk_mhz * 16;
3543 context->bw_ctx.bw.dcn.clk.dramclk_khz = context->bw_ctx.dml.vba.DRAMSpeed * 1000 / 16;