2 * Copyright 2020-2021 Advanced Micro Devices, Inc.
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
27 #include "reg_helper.h"
28 #include "dcn_calc_math.h"
29 #include "dcn20/dcn20_resource.h"
30 #include "dcn30/dcn30_resource.h"
32 #include "clk_mgr/dcn30/dcn30_smu11_driver_if.h"
33 #include "display_mode_vba_30.h"
34 #include "dcn30_fpu.h"
43 #define FN(reg_name, field_name) \
44 optc1->tg_shift->field_name, optc1->tg_mask->field_name
47 struct _vcs_dpi_ip_params_st dcn3_0_ip = {
49 .clamp_min_dcfclk = 0,
53 .gpuvm_max_page_table_levels = 4,
54 .hostvm_max_page_table_levels = 4,
55 .hostvm_cached_page_table_levels = 0,
56 .pte_group_size_bytes = 2048,
58 .rob_buffer_size_kbytes = 184,
59 .det_buffer_size_kbytes = 184,
60 .dpte_buffer_size_in_pte_reqs_luma = 84,
61 .pde_proc_buffer_size_64k_reqs = 48,
62 .dpp_output_buffer_pixels = 2560,
63 .opp_output_buffer_lines = 1,
64 .pixel_chunk_size_kbytes = 8,
66 .max_page_table_levels = 2,
67 .pte_chunk_size_kbytes = 2, // ?
68 .meta_chunk_size_kbytes = 2,
69 .writeback_chunk_size_kbytes = 8,
70 .line_buffer_size_bits = 789504,
71 .is_line_buffer_bpp_fixed = 0, // ?
72 .line_buffer_fixed_bpp = 0, // ?
73 .dcc_supported = true,
74 .writeback_interface_buffer_size_kbytes = 90,
75 .writeback_line_buffer_buffer_size = 0,
76 .max_line_buffer_lines = 12,
77 .writeback_luma_buffer_size_kbytes = 12, // writeback_line_buffer_buffer_size = 656640
78 .writeback_chroma_buffer_size_kbytes = 8,
79 .writeback_chroma_line_buffer_width_pixels = 4,
80 .writeback_max_hscl_ratio = 1,
81 .writeback_max_vscl_ratio = 1,
82 .writeback_min_hscl_ratio = 1,
83 .writeback_min_vscl_ratio = 1,
84 .writeback_max_hscl_taps = 1,
85 .writeback_max_vscl_taps = 1,
86 .writeback_line_buffer_luma_buffer_size = 0,
87 .writeback_line_buffer_chroma_buffer_size = 14643,
88 .cursor_buffer_size = 8,
89 .cursor_chunk_size = 2,
93 .max_dchub_pscl_bw_pix_per_clk = 4,
94 .max_pscl_lb_bw_pix_per_clk = 2,
95 .max_lb_vscl_bw_pix_per_clk = 4,
96 .max_vscl_hscl_bw_pix_per_clk = 4,
103 .dispclk_ramp_margin_percent = 1,
104 .underscan_factor = 1.11,
105 .min_vblank_lines = 32,
106 .dppclk_delay_subtotal = 46,
107 .dynamic_metadata_vm_enabled = true,
108 .dppclk_delay_scl_lb_only = 16,
109 .dppclk_delay_scl = 50,
110 .dppclk_delay_cnvc_formatter = 27,
111 .dppclk_delay_cnvc_cursor = 6,
112 .dispclk_delay_subtotal = 119,
113 .dcfclk_cstate_latency = 5.2, // SRExitTime
114 .max_inter_dcn_tile_repeaters = 8,
115 .max_num_hdmi_frl_outputs = 1,
116 .odm_combine_4to1_supported = true,
118 .xfc_supported = false,
119 .xfc_fill_bw_overhead_percent = 10.0,
120 .xfc_fill_constant_bytes = 0,
121 .gfx7_compat_tiling_supported = 0,
122 .number_of_cursors = 1,
125 struct _vcs_dpi_soc_bounding_box_st dcn3_0_soc = {
129 .dispclk_mhz = 562.0,
132 .phyclk_d18_mhz = 667.0,
137 .min_dcfclk = 500.0, /* TODO: set this to actual min DCFCLK */
139 .sr_exit_time_us = 15.5,
140 .sr_enter_plus_exit_time_us = 20,
141 .urgent_latency_us = 4.0,
142 .urgent_latency_pixel_data_only_us = 4.0,
143 .urgent_latency_pixel_mixed_with_vm_data_us = 4.0,
144 .urgent_latency_vm_data_only_us = 4.0,
145 .urgent_out_of_order_return_per_channel_pixel_only_bytes = 4096,
146 .urgent_out_of_order_return_per_channel_pixel_and_vm_bytes = 4096,
147 .urgent_out_of_order_return_per_channel_vm_only_bytes = 4096,
148 .pct_ideal_dram_sdp_bw_after_urgent_pixel_only = 80.0,
149 .pct_ideal_dram_sdp_bw_after_urgent_pixel_and_vm = 60.0,
150 .pct_ideal_dram_sdp_bw_after_urgent_vm_only = 40.0,
151 .max_avg_sdp_bw_use_normal_percent = 60.0,
152 .max_avg_dram_bw_use_normal_percent = 40.0,
153 .writeback_latency_us = 12.0,
154 .max_request_size_bytes = 256,
155 .fabric_datapath_to_dcn_data_return_bytes = 64,
156 .dcn_downspread_percent = 0.5,
157 .downspread_percent = 0.38,
158 .dram_page_open_time_ns = 50.0,
159 .dram_rw_turnaround_time_ns = 17.5,
160 .dram_return_buffer_per_channel_bytes = 8192,
161 .round_trip_ping_latency_dcfclk_cycles = 191,
162 .urgent_out_of_order_return_per_channel_bytes = 4096,
163 .channel_interleave_bytes = 256,
165 .gpuvm_min_page_size_bytes = 4096,
166 .hostvm_min_page_size_bytes = 4096,
167 .dram_clock_change_latency_us = 404,
168 .dummy_pstate_latency_us = 5,
169 .writeback_dram_clock_change_latency_us = 23.0,
170 .return_bus_width_bytes = 64,
171 .dispclk_dppclk_vco_speed_mhz = 3650,
172 .xfc_bus_transport_time_us = 20, // ?
173 .xfc_xbuf_latency_tolerance_us = 4, // ?
174 .use_urgent_burst_bw = 1, // ?
175 .do_urgent_latency_adjustment = true,
176 .urgent_latency_adjustment_fabric_clock_component_us = 1.0,
177 .urgent_latency_adjustment_fabric_clock_reference_mhz = 1000,
181 void optc3_fpu_set_vrr_m_const(struct timing_generator *optc,
184 struct optc *optc1 = DCN10TG_FROM_TG(optc);
185 double vtotal_min, vtotal_max;
186 double ratio, modulo, phase;
187 uint32_t vblank_start;
188 uint32_t v_total_mask_value = 0;
190 dc_assert_fp_enabled();
192 /* Compute VTOTAL_MIN and VTOTAL_MAX, so that
193 * VOTAL_MAX - VTOTAL_MIN = 1
195 v_total_mask_value = 16;
196 vtotal_min = dcn_bw_floor(vtotal_avg);
197 vtotal_max = dcn_bw_ceil(vtotal_avg);
199 /* Check that bottom VBLANK is at least 2 lines tall when running with
200 * VTOTAL_MIN. Note that VTOTAL registers are defined as 'total number
201 * of lines in a frame - 1'.
203 REG_GET(OTG_V_BLANK_START_END, OTG_V_BLANK_START,
205 ASSERT(vtotal_min >= vblank_start + 1);
207 /* Special case where the average frame rate can be achieved
208 * without using the DTO
210 if (vtotal_min == vtotal_max) {
211 REG_SET(OTG_V_TOTAL, 0, OTG_V_TOTAL, (uint32_t)vtotal_min);
213 optc->funcs->set_vtotal_min_max(optc, 0, 0);
214 REG_SET(OTG_M_CONST_DTO0, 0, OTG_M_CONST_DTO_PHASE, 0);
215 REG_SET(OTG_M_CONST_DTO1, 0, OTG_M_CONST_DTO_MODULO, 0);
216 REG_UPDATE_3(OTG_V_TOTAL_CONTROL,
217 OTG_V_TOTAL_MIN_SEL, 0,
218 OTG_V_TOTAL_MAX_SEL, 0,
219 OTG_SET_V_TOTAL_MIN_MASK_EN, 0);
223 ratio = vtotal_max - vtotal_avg;
224 modulo = 65536.0 * 65536.0 - 1.0; /* 2^32 - 1 */
225 phase = ratio * modulo;
227 /* Special cases where the DTO phase gets rounded to 0 or
230 if (phase <= 0 || phase >= modulo) {
231 REG_SET(OTG_V_TOTAL, 0, OTG_V_TOTAL,
233 (uint32_t)vtotal_max : (uint32_t)vtotal_min);
234 REG_SET(OTG_V_TOTAL_MIN, 0, OTG_V_TOTAL_MIN, 0);
235 REG_SET(OTG_V_TOTAL_MAX, 0, OTG_V_TOTAL_MAX, 0);
236 REG_SET(OTG_M_CONST_DTO0, 0, OTG_M_CONST_DTO_PHASE, 0);
237 REG_SET(OTG_M_CONST_DTO1, 0, OTG_M_CONST_DTO_MODULO, 0);
238 REG_UPDATE_3(OTG_V_TOTAL_CONTROL,
239 OTG_V_TOTAL_MIN_SEL, 0,
240 OTG_V_TOTAL_MAX_SEL, 0,
241 OTG_SET_V_TOTAL_MIN_MASK_EN, 0);
244 REG_UPDATE_6(OTG_V_TOTAL_CONTROL,
245 OTG_V_TOTAL_MIN_SEL, 1,
246 OTG_V_TOTAL_MAX_SEL, 1,
247 OTG_SET_V_TOTAL_MIN_MASK_EN, 1,
248 OTG_SET_V_TOTAL_MIN_MASK, v_total_mask_value,
249 OTG_VTOTAL_MID_REPLACING_MIN_EN, 0,
250 OTG_VTOTAL_MID_REPLACING_MAX_EN, 0);
251 REG_SET(OTG_V_TOTAL, 0, OTG_V_TOTAL, (uint32_t)vtotal_min);
252 optc->funcs->set_vtotal_min_max(optc, vtotal_min, vtotal_max);
253 REG_SET(OTG_M_CONST_DTO0, 0, OTG_M_CONST_DTO_PHASE, (uint32_t)phase);
254 REG_SET(OTG_M_CONST_DTO1, 0, OTG_M_CONST_DTO_MODULO, (uint32_t)modulo);
257 void dcn30_fpu_populate_dml_writeback_from_context(
258 struct dc *dc, struct resource_context *res_ctx, display_e2e_pipe_params_st *pipes)
261 double max_calc_writeback_dispclk;
262 double writeback_dispclk;
263 struct writeback_st dout_wb;
265 dc_assert_fp_enabled();
267 for (i = 0, pipe_cnt = 0; i < dc->res_pool->pipe_count; i++) {
268 struct dc_stream_state *stream = res_ctx->pipe_ctx[i].stream;
272 max_calc_writeback_dispclk = 0;
274 /* Set writeback information */
275 pipes[pipe_cnt].dout.wb_enable = 0;
276 pipes[pipe_cnt].dout.num_active_wb = 0;
277 for (j = 0; j < stream->num_wb_info; j++) {
278 struct dc_writeback_info *wb_info = &stream->writeback_info[j];
280 if (wb_info->wb_enabled && wb_info->writeback_source_plane &&
281 (wb_info->writeback_source_plane == res_ctx->pipe_ctx[i].plane_state)) {
282 pipes[pipe_cnt].dout.wb_enable = 1;
283 pipes[pipe_cnt].dout.num_active_wb++;
284 dout_wb.wb_src_height = wb_info->dwb_params.cnv_params.crop_en ?
285 wb_info->dwb_params.cnv_params.crop_height :
286 wb_info->dwb_params.cnv_params.src_height;
287 dout_wb.wb_src_width = wb_info->dwb_params.cnv_params.crop_en ?
288 wb_info->dwb_params.cnv_params.crop_width :
289 wb_info->dwb_params.cnv_params.src_width;
290 dout_wb.wb_dst_width = wb_info->dwb_params.dest_width;
291 dout_wb.wb_dst_height = wb_info->dwb_params.dest_height;
293 /* For IP that doesn't support WB scaling, set h/v taps to 1 to avoid DML validation failure */
294 if (dc->dml.ip.writeback_max_hscl_taps > 1) {
295 dout_wb.wb_htaps_luma = wb_info->dwb_params.scaler_taps.h_taps;
296 dout_wb.wb_vtaps_luma = wb_info->dwb_params.scaler_taps.v_taps;
298 dout_wb.wb_htaps_luma = 1;
299 dout_wb.wb_vtaps_luma = 1;
301 dout_wb.wb_htaps_chroma = 0;
302 dout_wb.wb_vtaps_chroma = 0;
303 dout_wb.wb_hratio = wb_info->dwb_params.cnv_params.crop_en ?
304 (double)wb_info->dwb_params.cnv_params.crop_width /
305 (double)wb_info->dwb_params.dest_width :
306 (double)wb_info->dwb_params.cnv_params.src_width /
307 (double)wb_info->dwb_params.dest_width;
308 dout_wb.wb_vratio = wb_info->dwb_params.cnv_params.crop_en ?
309 (double)wb_info->dwb_params.cnv_params.crop_height /
310 (double)wb_info->dwb_params.dest_height :
311 (double)wb_info->dwb_params.cnv_params.src_height /
312 (double)wb_info->dwb_params.dest_height;
313 if (wb_info->dwb_params.cnv_params.fc_out_format == DWB_OUT_FORMAT_64BPP_ARGB ||
314 wb_info->dwb_params.cnv_params.fc_out_format == DWB_OUT_FORMAT_64BPP_RGBA)
315 dout_wb.wb_pixel_format = dm_444_64;
317 dout_wb.wb_pixel_format = dm_444_32;
319 /* Workaround for cases where multiple writebacks are connected to same plane
320 * In which case, need to compute worst case and set the associated writeback parameters
321 * This workaround is necessary due to DML computation assuming only 1 set of writeback
322 * parameters per pipe
324 writeback_dispclk = dml30_CalculateWriteBackDISPCLK(
325 dout_wb.wb_pixel_format,
326 pipes[pipe_cnt].pipe.dest.pixel_rate_mhz,
329 dout_wb.wb_htaps_luma,
330 dout_wb.wb_vtaps_luma,
331 dout_wb.wb_src_width,
332 dout_wb.wb_dst_width,
333 pipes[pipe_cnt].pipe.dest.htotal,
334 dc->current_state->bw_ctx.dml.ip.writeback_line_buffer_buffer_size);
336 if (writeback_dispclk > max_calc_writeback_dispclk) {
337 max_calc_writeback_dispclk = writeback_dispclk;
338 pipes[pipe_cnt].dout.wb = dout_wb;
347 void dcn30_fpu_set_mcif_arb_params(struct mcif_arb_params *wb_arb_params,
348 struct display_mode_lib *dml,
349 display_e2e_pipe_params_st *pipes,
355 dc_assert_fp_enabled();
357 for (i = 0; i < sizeof(wb_arb_params->cli_watermark)/sizeof(wb_arb_params->cli_watermark[0]); i++) {
358 wb_arb_params->cli_watermark[i] = get_wm_writeback_urgent(dml, pipes, pipe_cnt) * 1000;
359 wb_arb_params->pstate_watermark[i] = get_wm_writeback_dram_clock_change(dml, pipes, pipe_cnt) * 1000;
362 wb_arb_params->dram_speed_change_duration = dml->vba.WritebackAllowDRAMClockChangeEndPosition[cur_pipe] * pipes[0].clks_cfg.refclk_mhz; /* num_clock_cycles = us * MHz */
365 void dcn30_fpu_update_soc_for_wm_a(struct dc *dc, struct dc_state *context)
368 dc_assert_fp_enabled();
370 if (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].valid) {
371 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;
372 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;
373 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;
377 void dcn30_fpu_calculate_wm_and_dlg(
378 struct dc *dc, struct dc_state *context,
379 display_e2e_pipe_params_st *pipes,
383 int maxMpcComb = context->bw_ctx.dml.vba.maxMpcComb;
385 double dcfclk = context->bw_ctx.dml.vba.DCFCLKState[vlevel][maxMpcComb];
386 bool pstate_en = context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][maxMpcComb] != dm_dram_clock_change_unsupported;
388 dc_assert_fp_enabled();
390 if (context->bw_ctx.dml.soc.min_dcfclk > dcfclk)
391 dcfclk = context->bw_ctx.dml.soc.min_dcfclk;
393 pipes[0].clks_cfg.voltage = vlevel;
394 pipes[0].clks_cfg.dcfclk_mhz = dcfclk;
395 pipes[0].clks_cfg.socclk_mhz = context->bw_ctx.dml.soc.clock_limits[vlevel].socclk_mhz;
398 * DCFCLK: 1GHz or min required above 1GHz
401 if (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_B].valid) {
403 pipes[0].clks_cfg.voltage = 1;
404 pipes[0].clks_cfg.dcfclk_mhz = context->bw_ctx.dml.soc.clock_limits[0].dcfclk_mhz;
406 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;
407 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;
408 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;
410 context->bw_ctx.bw.dcn.watermarks.b.urgent_ns = get_wm_urgent(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
411 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;
412 context->bw_ctx.bw.dcn.watermarks.b.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
413 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;
414 context->bw_ctx.bw.dcn.watermarks.b.pte_meta_urgent_ns = get_wm_memory_trip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
415 context->bw_ctx.bw.dcn.watermarks.b.frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
416 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;
417 context->bw_ctx.bw.dcn.watermarks.b.urgent_latency_ns = get_urgent_latency(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
419 pipes[0].clks_cfg.voltage = vlevel;
420 pipes[0].clks_cfg.dcfclk_mhz = dcfclk;
423 * DCFCLK: Min Required
424 * FCLK(proportional to UCLK): 1GHz or Max
425 * MALL stutter, sr_enter_exit = 4, sr_exit = 2us
428 if (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_D].valid) {
429 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;
430 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;
431 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;
433 context->bw_ctx.bw.dcn.watermarks.d.urgent_ns = get_wm_urgent(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
434 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;
435 context->bw_ctx.bw.dcn.watermarks.d.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
436 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;
437 context->bw_ctx.bw.dcn.watermarks.d.pte_meta_urgent_ns = get_wm_memory_trip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
438 context->bw_ctx.bw.dcn.watermarks.d.frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
439 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;
440 context->bw_ctx.bw.dcn.watermarks.d.urgent_latency_ns = get_urgent_latency(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
444 * DCFCLK: Min Required
445 * FCLK(proportional to UCLK): 1GHz or Max
446 * pstate latency overridden to 5us
448 if (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_C].valid) {
449 unsigned int min_dram_speed_mts = context->bw_ctx.dml.vba.DRAMSpeed;
450 unsigned int min_dram_speed_mts_margin = 160;
452 if (context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] == dm_dram_clock_change_unsupported)
453 min_dram_speed_mts = dc->clk_mgr->bw_params->clk_table.entries[dc->clk_mgr->bw_params->clk_table.num_entries - 1].memclk_mhz * 16;
455 /* find largest table entry that is lower than dram speed, but lower than DPM0 still uses DPM0 */
456 for (i = 3; i > 0; i--)
457 if (min_dram_speed_mts + min_dram_speed_mts_margin > dc->clk_mgr->bw_params->dummy_pstate_table[i].dram_speed_mts)
460 context->bw_ctx.dml.soc.dram_clock_change_latency_us = dc->clk_mgr->bw_params->dummy_pstate_table[i].dummy_pstate_latency_us;
462 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;
463 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;
466 context->bw_ctx.bw.dcn.watermarks.c.urgent_ns = get_wm_urgent(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
467 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;
468 context->bw_ctx.bw.dcn.watermarks.c.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
469 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;
470 context->bw_ctx.bw.dcn.watermarks.c.pte_meta_urgent_ns = get_wm_memory_trip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
471 context->bw_ctx.bw.dcn.watermarks.c.frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
472 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;
473 context->bw_ctx.bw.dcn.watermarks.c.urgent_latency_ns = get_urgent_latency(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
476 /* The only difference between A and C is p-state latency, if p-state is not supported we want to
477 * calculate DLG based on dummy p-state latency, and max out the set A p-state watermark
479 context->bw_ctx.bw.dcn.watermarks.a = context->bw_ctx.bw.dcn.watermarks.c;
480 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.pstate_change_ns = 0;
483 * DCFCLK: Min Required
484 * FCLK(proportional to UCLK): 1GHz or Max
486 * Set A calculated last so that following calculations are based on Set A
488 dc->res_pool->funcs->update_soc_for_wm_a(dc, context);
489 context->bw_ctx.bw.dcn.watermarks.a.urgent_ns = get_wm_urgent(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
490 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;
491 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
492 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;
493 context->bw_ctx.bw.dcn.watermarks.a.pte_meta_urgent_ns = get_wm_memory_trip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
494 context->bw_ctx.bw.dcn.watermarks.a.frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
495 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;
496 context->bw_ctx.bw.dcn.watermarks.a.urgent_latency_ns = get_urgent_latency(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
499 context->perf_params.stutter_period_us = context->bw_ctx.dml.vba.StutterPeriod;
501 /* Make set D = set A until set D is enabled */
502 context->bw_ctx.bw.dcn.watermarks.d = context->bw_ctx.bw.dcn.watermarks.a;
504 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
505 if (!context->res_ctx.pipe_ctx[i].stream)
508 pipes[pipe_idx].clks_cfg.dispclk_mhz = get_dispclk_calculated(&context->bw_ctx.dml, pipes, pipe_cnt);
509 pipes[pipe_idx].clks_cfg.dppclk_mhz = get_dppclk_calculated(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx);
511 if (dc->config.forced_clocks) {
512 pipes[pipe_idx].clks_cfg.dispclk_mhz = context->bw_ctx.dml.soc.clock_limits[0].dispclk_mhz;
513 pipes[pipe_idx].clks_cfg.dppclk_mhz = context->bw_ctx.dml.soc.clock_limits[0].dppclk_mhz;
515 if (dc->debug.min_disp_clk_khz > pipes[pipe_idx].clks_cfg.dispclk_mhz * 1000)
516 pipes[pipe_idx].clks_cfg.dispclk_mhz = dc->debug.min_disp_clk_khz / 1000.0;
517 if (dc->debug.min_dpp_clk_khz > pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000)
518 pipes[pipe_idx].clks_cfg.dppclk_mhz = dc->debug.min_dpp_clk_khz / 1000.0;
524 dcn20_calculate_dlg_params(dc, context, pipes, pipe_cnt, vlevel);
528 /* Restore full p-state latency */
529 context->bw_ctx.dml.soc.dram_clock_change_latency_us =
530 dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us;
532 if (context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching)
533 dcn30_setup_mclk_switch_using_fw_based_vblank_stretch(dc, context);
536 void dcn30_fpu_update_dram_channel_width_bytes(struct dc *dc)
538 dc_assert_fp_enabled();
540 if (dc->ctx->dc_bios->vram_info.dram_channel_width_bytes)
541 dcn3_0_soc.dram_channel_width_bytes = dc->ctx->dc_bios->vram_info.dram_channel_width_bytes;
544 void dcn30_fpu_update_max_clk(struct dc_bounding_box_max_clk *dcn30_bb_max_clk)
546 dc_assert_fp_enabled();
548 if (!dcn30_bb_max_clk->max_dcfclk_mhz)
549 dcn30_bb_max_clk->max_dcfclk_mhz = dcn3_0_soc.clock_limits[0].dcfclk_mhz;
550 if (!dcn30_bb_max_clk->max_dispclk_mhz)
551 dcn30_bb_max_clk->max_dispclk_mhz = dcn3_0_soc.clock_limits[0].dispclk_mhz;
552 if (!dcn30_bb_max_clk->max_dppclk_mhz)
553 dcn30_bb_max_clk->max_dppclk_mhz = dcn3_0_soc.clock_limits[0].dppclk_mhz;
554 if (!dcn30_bb_max_clk->max_phyclk_mhz)
555 dcn30_bb_max_clk->max_phyclk_mhz = dcn3_0_soc.clock_limits[0].phyclk_mhz;
558 void dcn30_fpu_get_optimal_dcfclk_fclk_for_uclk(unsigned int uclk_mts,
559 unsigned int *optimal_dcfclk,
560 unsigned int *optimal_fclk)
562 double bw_from_dram, bw_from_dram1, bw_from_dram2;
564 dc_assert_fp_enabled();
566 bw_from_dram1 = uclk_mts * dcn3_0_soc.num_chans *
567 dcn3_0_soc.dram_channel_width_bytes * (dcn3_0_soc.max_avg_dram_bw_use_normal_percent / 100);
568 bw_from_dram2 = uclk_mts * dcn3_0_soc.num_chans *
569 dcn3_0_soc.dram_channel_width_bytes * (dcn3_0_soc.max_avg_sdp_bw_use_normal_percent / 100);
571 bw_from_dram = (bw_from_dram1 < bw_from_dram2) ? bw_from_dram1 : bw_from_dram2;
574 *optimal_fclk = bw_from_dram /
575 (dcn3_0_soc.fabric_datapath_to_dcn_data_return_bytes * (dcn3_0_soc.max_avg_sdp_bw_use_normal_percent / 100));
578 *optimal_dcfclk = bw_from_dram /
579 (dcn3_0_soc.return_bus_width_bytes * (dcn3_0_soc.max_avg_sdp_bw_use_normal_percent / 100));
582 void dcn30_fpu_update_bw_bounding_box(struct dc *dc,
583 struct clk_bw_params *bw_params,
584 struct dc_bounding_box_max_clk *dcn30_bb_max_clk,
585 unsigned int *dcfclk_mhz,
586 unsigned int *dram_speed_mts)
590 dc_assert_fp_enabled();
592 dcn3_0_soc.dispclk_dppclk_vco_speed_mhz = dc->clk_mgr->dentist_vco_freq_khz / 1000.0;
593 dc->dml.soc.dispclk_dppclk_vco_speed_mhz = dc->clk_mgr->dentist_vco_freq_khz / 1000.0;
595 for (i = 0; i < dcn3_0_soc.num_states; i++) {
596 dcn3_0_soc.clock_limits[i].state = i;
597 dcn3_0_soc.clock_limits[i].dcfclk_mhz = dcfclk_mhz[i];
598 dcn3_0_soc.clock_limits[i].fabricclk_mhz = dcfclk_mhz[i];
599 dcn3_0_soc.clock_limits[i].dram_speed_mts = dram_speed_mts[i];
601 /* Fill all states with max values of all other clocks */
602 dcn3_0_soc.clock_limits[i].dispclk_mhz = dcn30_bb_max_clk->max_dispclk_mhz;
603 dcn3_0_soc.clock_limits[i].dppclk_mhz = dcn30_bb_max_clk->max_dppclk_mhz;
604 dcn3_0_soc.clock_limits[i].phyclk_mhz = dcn30_bb_max_clk->max_phyclk_mhz;
605 dcn3_0_soc.clock_limits[i].dtbclk_mhz = dcn3_0_soc.clock_limits[0].dtbclk_mhz;
606 /* These clocks cannot come from bw_params, always fill from dcn3_0_soc[1] */
607 /* FCLK, PHYCLK_D18, SOCCLK, DSCCLK */
608 dcn3_0_soc.clock_limits[i].phyclk_d18_mhz = dcn3_0_soc.clock_limits[0].phyclk_d18_mhz;
609 dcn3_0_soc.clock_limits[i].socclk_mhz = dcn3_0_soc.clock_limits[0].socclk_mhz;
610 dcn3_0_soc.clock_limits[i].dscclk_mhz = dcn3_0_soc.clock_limits[0].dscclk_mhz;
612 /* re-init DML with updated bb */
613 dml_init_instance(&dc->dml, &dcn3_0_soc, &dcn3_0_ip, DML_PROJECT_DCN30);
614 if (dc->current_state)
615 dml_init_instance(&dc->current_state->bw_ctx.dml, &dcn3_0_soc, &dcn3_0_ip, DML_PROJECT_DCN30);
619 void dcn3_fpu_build_wm_range_table(struct clk_mgr *base)
622 double pstate_latency_us = base->ctx->dc->dml.soc.dram_clock_change_latency_us;
623 double sr_exit_time_us = base->ctx->dc->dml.soc.sr_exit_time_us;
624 double sr_enter_plus_exit_time_us = base->ctx->dc->dml.soc.sr_enter_plus_exit_time_us;
625 uint16_t min_uclk_mhz = base->bw_params->clk_table.entries[0].memclk_mhz;
627 dc_assert_fp_enabled();
629 /* Set A - Normal - default values*/
630 base->bw_params->wm_table.nv_entries[WM_A].valid = true;
631 base->bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us = pstate_latency_us;
632 base->bw_params->wm_table.nv_entries[WM_A].dml_input.sr_exit_time_us = sr_exit_time_us;
633 base->bw_params->wm_table.nv_entries[WM_A].dml_input.sr_enter_plus_exit_time_us = sr_enter_plus_exit_time_us;
634 base->bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.wm_type = WATERMARKS_CLOCK_RANGE;
635 base->bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.min_dcfclk = 0;
636 base->bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.max_dcfclk = 0xFFFF;
637 base->bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.min_uclk = min_uclk_mhz;
638 base->bw_params->wm_table.nv_entries[WM_A].pmfw_breakdown.max_uclk = 0xFFFF;
640 /* Set B - Performance - higher minimum clocks */
641 // base->bw_params->wm_table.nv_entries[WM_B].valid = true;
642 // base->bw_params->wm_table.nv_entries[WM_B].dml_input.pstate_latency_us = pstate_latency_us;
643 // base->bw_params->wm_table.nv_entries[WM_B].dml_input.sr_exit_time_us = sr_exit_time_us;
644 // base->bw_params->wm_table.nv_entries[WM_B].dml_input.sr_enter_plus_exit_time_us = sr_enter_plus_exit_time_us;
645 // base->bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.wm_type = WATERMARKS_CLOCK_RANGE;
646 // base->bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.min_dcfclk = TUNED VALUE;
647 // base->bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.max_dcfclk = 0xFFFF;
648 // base->bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.min_uclk = TUNED VALUE;
649 // base->bw_params->wm_table.nv_entries[WM_B].pmfw_breakdown.max_uclk = 0xFFFF;
651 /* Set C - Dummy P-State - P-State latency set to "dummy p-state" value */
652 base->bw_params->wm_table.nv_entries[WM_C].valid = true;
653 base->bw_params->wm_table.nv_entries[WM_C].dml_input.pstate_latency_us = 0;
654 base->bw_params->wm_table.nv_entries[WM_C].dml_input.sr_exit_time_us = sr_exit_time_us;
655 base->bw_params->wm_table.nv_entries[WM_C].dml_input.sr_enter_plus_exit_time_us = sr_enter_plus_exit_time_us;
656 base->bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.wm_type = WATERMARKS_DUMMY_PSTATE;
657 base->bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.min_dcfclk = 0;
658 base->bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.max_dcfclk = 0xFFFF;
659 base->bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.min_uclk = min_uclk_mhz;
660 base->bw_params->wm_table.nv_entries[WM_C].pmfw_breakdown.max_uclk = 0xFFFF;
661 base->bw_params->dummy_pstate_table[0].dram_speed_mts = 1600;
662 base->bw_params->dummy_pstate_table[0].dummy_pstate_latency_us = 38;
663 base->bw_params->dummy_pstate_table[1].dram_speed_mts = 8000;
664 base->bw_params->dummy_pstate_table[1].dummy_pstate_latency_us = 9;
665 base->bw_params->dummy_pstate_table[2].dram_speed_mts = 10000;
666 base->bw_params->dummy_pstate_table[2].dummy_pstate_latency_us = 8;
667 base->bw_params->dummy_pstate_table[3].dram_speed_mts = 16000;
668 base->bw_params->dummy_pstate_table[3].dummy_pstate_latency_us = 5;
670 /* Set D - MALL - SR enter and exit times adjusted for MALL */
671 base->bw_params->wm_table.nv_entries[WM_D].valid = true;
672 base->bw_params->wm_table.nv_entries[WM_D].dml_input.pstate_latency_us = pstate_latency_us;
673 base->bw_params->wm_table.nv_entries[WM_D].dml_input.sr_exit_time_us = 2;
674 base->bw_params->wm_table.nv_entries[WM_D].dml_input.sr_enter_plus_exit_time_us = 4;
675 base->bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.wm_type = WATERMARKS_MALL;
676 base->bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.min_dcfclk = 0;
677 base->bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.max_dcfclk = 0xFFFF;
678 base->bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.min_uclk = min_uclk_mhz;
679 base->bw_params->wm_table.nv_entries[WM_D].pmfw_breakdown.max_uclk = 0xFFFF;