1/*
2 * Copyright 2019-2021 Advanced Micro Devices, Inc.
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
21 *
22 * Authors: AMD
23 *
24 */
25
26#include "resource.h"
27#include "clk_mgr.h"
28#include "dcn31/dcn31_resource.h"
29#include "dcn315/dcn315_resource.h"
30#include "dcn316/dcn316_resource.h"
31
32#include "dml/dcn20/dcn20_fpu.h"
33#include "dcn31_fpu.h"
34
35/**
36 * DOC: DCN31x FPU manipulation Overview
37 *
38 * The DCN architecture relies on FPU operations, which require special
39 * compilation flags and the use of kernel_fpu_begin/end functions; ideally, we
40 * want to avoid spreading FPU access across multiple files. With this idea in
41 * mind, this file aims to centralize all DCN3.1.x functions that require FPU
42 * access in a single place. Code in this file follows the following code
43 * pattern:
44 *
45 * 1. Functions that use FPU operations should be isolated in static functions.
46 * 2. The FPU functions should have the noinline attribute to ensure anything
47 * that deals with FP register is contained within this call.
48 * 3. All function that needs to be accessed outside this file requires a
49 * public interface that not uses any FPU reference.
50 * 4. Developers **must not** use DC_FP_START/END in this file, but they need
51 * to ensure that the caller invokes it before access any function available
52 * in this file. For this reason, public functions in this file must invoke
53 * dc_assert_fp_enabled();
54 */
55
56struct _vcs_dpi_ip_params_st dcn3_1_ip = {
57 .gpuvm_enable = 1,
58 .gpuvm_max_page_table_levels = 1,
59 .hostvm_enable = 1,
60 .hostvm_max_page_table_levels = 2,
61 .rob_buffer_size_kbytes = 64,
62 .det_buffer_size_kbytes = DCN3_1_DEFAULT_DET_SIZE,
63 .config_return_buffer_size_in_kbytes = 1792,
64 .compressed_buffer_segment_size_in_kbytes = 64,
65 .meta_fifo_size_in_kentries = 32,
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 .meta_chunk_size_kbytes = 2,
73 .min_meta_chunk_size_bytes = 256,
74 .writeback_chunk_size_kbytes = 8,
75 .ptoi_supported = false,
76 .num_dsc = 3,
77 .maximum_dsc_bits_per_component = 10,
78 .dsc422_native_support = false,
79 .is_line_buffer_bpp_fixed = true,
80 .line_buffer_fixed_bpp = 48,
81 .line_buffer_size_bits = 789504,
82 .max_line_buffer_lines = 12,
83 .writeback_interface_buffer_size_kbytes = 90,
84 .max_num_dpp = 4,
85 .max_num_otg = 4,
86 .max_num_hdmi_frl_outputs = 1,
87 .max_num_wb = 1,
88 .max_dchub_pscl_bw_pix_per_clk = 4,
89 .max_pscl_lb_bw_pix_per_clk = 2,
90 .max_lb_vscl_bw_pix_per_clk = 4,
91 .max_vscl_hscl_bw_pix_per_clk = 4,
92 .max_hscl_ratio = 6,
93 .max_vscl_ratio = 6,
94 .max_hscl_taps = 8,
95 .max_vscl_taps = 8,
96 .dpte_buffer_size_in_pte_reqs_luma = 64,
97 .dpte_buffer_size_in_pte_reqs_chroma = 34,
98 .dispclk_ramp_margin_percent = 1,
99 .max_inter_dcn_tile_repeaters = 8,
100 .cursor_buffer_size = 16,
101 .cursor_chunk_size = 2,
102 .writeback_line_buffer_buffer_size = 0,
103 .writeback_min_hscl_ratio = 1,
104 .writeback_min_vscl_ratio = 1,
105 .writeback_max_hscl_ratio = 1,
106 .writeback_max_vscl_ratio = 1,
107 .writeback_max_hscl_taps = 1,
108 .writeback_max_vscl_taps = 1,
109 .dppclk_delay_subtotal = 46,
110 .dppclk_delay_scl = 50,
111 .dppclk_delay_scl_lb_only = 16,
112 .dppclk_delay_cnvc_formatter = 27,
113 .dppclk_delay_cnvc_cursor = 6,
114 .dispclk_delay_subtotal = 119,
115 .dynamic_metadata_vm_enabled = false,
116 .odm_combine_4to1_supported = false,
117 .dcc_supported = true,
118};
119
120static struct _vcs_dpi_soc_bounding_box_st dcn3_1_soc = {
121 /*TODO: correct dispclk/dppclk voltage level determination*/
122 .clock_limits = {
123 {
124 .state = 0,
125 .dispclk_mhz = 1200.0,
126 .dppclk_mhz = 1200.0,
127 .phyclk_mhz = 600.0,
128 .phyclk_d18_mhz = 667.0,
129 .dscclk_mhz = 186.0,
130 .dtbclk_mhz = 625.0,
131 },
132 {
133 .state = 1,
134 .dispclk_mhz = 1200.0,
135 .dppclk_mhz = 1200.0,
136 .phyclk_mhz = 810.0,
137 .phyclk_d18_mhz = 667.0,
138 .dscclk_mhz = 209.0,
139 .dtbclk_mhz = 625.0,
140 },
141 {
142 .state = 2,
143 .dispclk_mhz = 1200.0,
144 .dppclk_mhz = 1200.0,
145 .phyclk_mhz = 810.0,
146 .phyclk_d18_mhz = 667.0,
147 .dscclk_mhz = 209.0,
148 .dtbclk_mhz = 625.0,
149 },
150 {
151 .state = 3,
152 .dispclk_mhz = 1200.0,
153 .dppclk_mhz = 1200.0,
154 .phyclk_mhz = 810.0,
155 .phyclk_d18_mhz = 667.0,
156 .dscclk_mhz = 371.0,
157 .dtbclk_mhz = 625.0,
158 },
159 {
160 .state = 4,
161 .dispclk_mhz = 1200.0,
162 .dppclk_mhz = 1200.0,
163 .phyclk_mhz = 810.0,
164 .phyclk_d18_mhz = 667.0,
165 .dscclk_mhz = 417.0,
166 .dtbclk_mhz = 625.0,
167 },
168 },
169 .num_states = 5,
170 .sr_exit_time_us = 9.0,
171 .sr_enter_plus_exit_time_us = 11.0,
172 .sr_exit_z8_time_us = 442.0,
173 .sr_enter_plus_exit_z8_time_us = 560.0,
174 .writeback_latency_us = 12.0,
175 .dram_channel_width_bytes = 4,
176 .round_trip_ping_latency_dcfclk_cycles = 106,
177 .urgent_latency_pixel_data_only_us = 4.0,
178 .urgent_latency_pixel_mixed_with_vm_data_us = 4.0,
179 .urgent_latency_vm_data_only_us = 4.0,
180 .urgent_out_of_order_return_per_channel_pixel_only_bytes = 4096,
181 .urgent_out_of_order_return_per_channel_pixel_and_vm_bytes = 4096,
182 .urgent_out_of_order_return_per_channel_vm_only_bytes = 4096,
183 .pct_ideal_sdp_bw_after_urgent = 80.0,
184 .pct_ideal_dram_sdp_bw_after_urgent_pixel_only = 65.0,
185 .pct_ideal_dram_sdp_bw_after_urgent_pixel_and_vm = 60.0,
186 .pct_ideal_dram_sdp_bw_after_urgent_vm_only = 30.0,
187 .max_avg_sdp_bw_use_normal_percent = 60.0,
188 .max_avg_dram_bw_use_normal_percent = 60.0,
189 .fabric_datapath_to_dcn_data_return_bytes = 32,
190 .return_bus_width_bytes = 64,
191 .downspread_percent = 0.38,
192 .dcn_downspread_percent = 0.5,
193 .gpuvm_min_page_size_bytes = 4096,
194 .hostvm_min_page_size_bytes = 4096,
195 .do_urgent_latency_adjustment = false,
196 .urgent_latency_adjustment_fabric_clock_component_us = 0,
197 .urgent_latency_adjustment_fabric_clock_reference_mhz = 0,
198};
199
200struct _vcs_dpi_ip_params_st dcn3_15_ip = {
201 .gpuvm_enable = 1,
202 .gpuvm_max_page_table_levels = 1,
203 .hostvm_enable = 1,
204 .hostvm_max_page_table_levels = 2,
205 .rob_buffer_size_kbytes = 64,
206 .det_buffer_size_kbytes = DCN3_15_DEFAULT_DET_SIZE,
207 .min_comp_buffer_size_kbytes = 64,
208 .config_return_buffer_size_in_kbytes = 1024,
209 .compressed_buffer_segment_size_in_kbytes = 64,
210 .meta_fifo_size_in_kentries = 32,
211 .zero_size_buffer_entries = 512,
212 .compbuf_reserved_space_64b = 256,
213 .compbuf_reserved_space_zs = 64,
214 .dpp_output_buffer_pixels = 2560,
215 .opp_output_buffer_lines = 1,
216 .pixel_chunk_size_kbytes = 8,
217 .meta_chunk_size_kbytes = 2,
218 .min_meta_chunk_size_bytes = 256,
219 .writeback_chunk_size_kbytes = 8,
220 .ptoi_supported = false,
221 .num_dsc = 3,
222 .maximum_dsc_bits_per_component = 10,
223 .dsc422_native_support = false,
224 .is_line_buffer_bpp_fixed = true,
225 .line_buffer_fixed_bpp = 48,
226 .line_buffer_size_bits = 789504,
227 .max_line_buffer_lines = 12,
228 .writeback_interface_buffer_size_kbytes = 90,
229 .max_num_dpp = 4,
230 .max_num_otg = 4,
231 .max_num_hdmi_frl_outputs = 1,
232 .max_num_wb = 1,
233 .max_dchub_pscl_bw_pix_per_clk = 4,
234 .max_pscl_lb_bw_pix_per_clk = 2,
235 .max_lb_vscl_bw_pix_per_clk = 4,
236 .max_vscl_hscl_bw_pix_per_clk = 4,
237 .max_hscl_ratio = 6,
238 .max_vscl_ratio = 6,
239 .max_hscl_taps = 8,
240 .max_vscl_taps = 8,
241 .dpte_buffer_size_in_pte_reqs_luma = 64,
242 .dpte_buffer_size_in_pte_reqs_chroma = 34,
243 .dispclk_ramp_margin_percent = 1,
244 .max_inter_dcn_tile_repeaters = 9,
245 .cursor_buffer_size = 16,
246 .cursor_chunk_size = 2,
247 .writeback_line_buffer_buffer_size = 0,
248 .writeback_min_hscl_ratio = 1,
249 .writeback_min_vscl_ratio = 1,
250 .writeback_max_hscl_ratio = 1,
251 .writeback_max_vscl_ratio = 1,
252 .writeback_max_hscl_taps = 1,
253 .writeback_max_vscl_taps = 1,
254 .dppclk_delay_subtotal = 46,
255 .dppclk_delay_scl = 50,
256 .dppclk_delay_scl_lb_only = 16,
257 .dppclk_delay_cnvc_formatter = 27,
258 .dppclk_delay_cnvc_cursor = 6,
259 .dispclk_delay_subtotal = 119,
260 .dynamic_metadata_vm_enabled = false,
261 .odm_combine_4to1_supported = false,
262 .dcc_supported = true,
263};
264
265static struct _vcs_dpi_soc_bounding_box_st dcn3_15_soc = {
266 .sr_exit_time_us = 9.0,
267 .sr_enter_plus_exit_time_us = 11.0,
268 .sr_exit_z8_time_us = 50.0,
269 .sr_enter_plus_exit_z8_time_us = 50.0,
270 .writeback_latency_us = 12.0,
271 .dram_channel_width_bytes = 4,
272 .round_trip_ping_latency_dcfclk_cycles = 106,
273 .urgent_latency_pixel_data_only_us = 4.0,
274 .urgent_latency_pixel_mixed_with_vm_data_us = 4.0,
275 .urgent_latency_vm_data_only_us = 4.0,
276 .urgent_out_of_order_return_per_channel_pixel_only_bytes = 4096,
277 .urgent_out_of_order_return_per_channel_pixel_and_vm_bytes = 4096,
278 .urgent_out_of_order_return_per_channel_vm_only_bytes = 4096,
279 .pct_ideal_sdp_bw_after_urgent = 80.0,
280 .pct_ideal_dram_sdp_bw_after_urgent_pixel_only = 65.0,
281 .pct_ideal_dram_sdp_bw_after_urgent_pixel_and_vm = 60.0,
282 .pct_ideal_dram_sdp_bw_after_urgent_vm_only = 30.0,
283 .max_avg_sdp_bw_use_normal_percent = 60.0,
284 .max_avg_dram_bw_use_normal_percent = 60.0,
285 .fabric_datapath_to_dcn_data_return_bytes = 32,
286 .return_bus_width_bytes = 64,
287 .downspread_percent = 0.38,
288 .dcn_downspread_percent = 0.38,
289 .gpuvm_min_page_size_bytes = 4096,
290 .hostvm_min_page_size_bytes = 4096,
291 .do_urgent_latency_adjustment = false,
292 .urgent_latency_adjustment_fabric_clock_component_us = 0,
293 .urgent_latency_adjustment_fabric_clock_reference_mhz = 0,
294 .num_chans = 4,
295 .dummy_pstate_latency_us = 10.0
296};
297
298struct _vcs_dpi_ip_params_st dcn3_16_ip = {
299 .gpuvm_enable = 1,
300 .gpuvm_max_page_table_levels = 1,
301 .hostvm_enable = 1,
302 .hostvm_max_page_table_levels = 2,
303 .rob_buffer_size_kbytes = 64,
304 .det_buffer_size_kbytes = DCN3_16_DEFAULT_DET_SIZE,
305 .min_comp_buffer_size_kbytes = 64,
306 .config_return_buffer_size_in_kbytes = 1024,
307 .compressed_buffer_segment_size_in_kbytes = 64,
308 .meta_fifo_size_in_kentries = 32,
309 .zero_size_buffer_entries = 512,
310 .compbuf_reserved_space_64b = 256,
311 .compbuf_reserved_space_zs = 64,
312 .dpp_output_buffer_pixels = 2560,
313 .opp_output_buffer_lines = 1,
314 .pixel_chunk_size_kbytes = 8,
315 .meta_chunk_size_kbytes = 2,
316 .min_meta_chunk_size_bytes = 256,
317 .writeback_chunk_size_kbytes = 8,
318 .ptoi_supported = false,
319 .num_dsc = 3,
320 .maximum_dsc_bits_per_component = 10,
321 .dsc422_native_support = false,
322 .is_line_buffer_bpp_fixed = true,
323 .line_buffer_fixed_bpp = 48,
324 .line_buffer_size_bits = 789504,
325 .max_line_buffer_lines = 12,
326 .writeback_interface_buffer_size_kbytes = 90,
327 .max_num_dpp = 4,
328 .max_num_otg = 4,
329 .max_num_hdmi_frl_outputs = 1,
330 .max_num_wb = 1,
331 .max_dchub_pscl_bw_pix_per_clk = 4,
332 .max_pscl_lb_bw_pix_per_clk = 2,
333 .max_lb_vscl_bw_pix_per_clk = 4,
334 .max_vscl_hscl_bw_pix_per_clk = 4,
335 .max_hscl_ratio = 6,
336 .max_vscl_ratio = 6,
337 .max_hscl_taps = 8,
338 .max_vscl_taps = 8,
339 .dpte_buffer_size_in_pte_reqs_luma = 64,
340 .dpte_buffer_size_in_pte_reqs_chroma = 34,
341 .dispclk_ramp_margin_percent = 1,
342 .max_inter_dcn_tile_repeaters = 8,
343 .cursor_buffer_size = 16,
344 .cursor_chunk_size = 2,
345 .writeback_line_buffer_buffer_size = 0,
346 .writeback_min_hscl_ratio = 1,
347 .writeback_min_vscl_ratio = 1,
348 .writeback_max_hscl_ratio = 1,
349 .writeback_max_vscl_ratio = 1,
350 .writeback_max_hscl_taps = 1,
351 .writeback_max_vscl_taps = 1,
352 .dppclk_delay_subtotal = 46,
353 .dppclk_delay_scl = 50,
354 .dppclk_delay_scl_lb_only = 16,
355 .dppclk_delay_cnvc_formatter = 27,
356 .dppclk_delay_cnvc_cursor = 6,
357 .dispclk_delay_subtotal = 119,
358 .dynamic_metadata_vm_enabled = false,
359 .odm_combine_4to1_supported = false,
360 .dcc_supported = true,
361};
362
363static struct _vcs_dpi_soc_bounding_box_st dcn3_16_soc = {
364 /*TODO: correct dispclk/dppclk voltage level determination*/
365 .clock_limits = {
366 {
367 .state = 0,
368 .dispclk_mhz = 556.0,
369 .dppclk_mhz = 556.0,
370 .phyclk_mhz = 600.0,
371 .phyclk_d18_mhz = 445.0,
372 .dscclk_mhz = 186.0,
373 .dtbclk_mhz = 625.0,
374 },
375 {
376 .state = 1,
377 .dispclk_mhz = 625.0,
378 .dppclk_mhz = 625.0,
379 .phyclk_mhz = 810.0,
380 .phyclk_d18_mhz = 667.0,
381 .dscclk_mhz = 209.0,
382 .dtbclk_mhz = 625.0,
383 },
384 {
385 .state = 2,
386 .dispclk_mhz = 625.0,
387 .dppclk_mhz = 625.0,
388 .phyclk_mhz = 810.0,
389 .phyclk_d18_mhz = 667.0,
390 .dscclk_mhz = 209.0,
391 .dtbclk_mhz = 625.0,
392 },
393 {
394 .state = 3,
395 .dispclk_mhz = 1112.0,
396 .dppclk_mhz = 1112.0,
397 .phyclk_mhz = 810.0,
398 .phyclk_d18_mhz = 667.0,
399 .dscclk_mhz = 371.0,
400 .dtbclk_mhz = 625.0,
401 },
402 {
403 .state = 4,
404 .dispclk_mhz = 1250.0,
405 .dppclk_mhz = 1250.0,
406 .phyclk_mhz = 810.0,
407 .phyclk_d18_mhz = 667.0,
408 .dscclk_mhz = 417.0,
409 .dtbclk_mhz = 625.0,
410 },
411 },
412 .num_states = 5,
413 .sr_exit_time_us = 9.0,
414 .sr_enter_plus_exit_time_us = 11.0,
415 .sr_exit_z8_time_us = 442.0,
416 .sr_enter_plus_exit_z8_time_us = 560.0,
417 .writeback_latency_us = 12.0,
418 .dram_channel_width_bytes = 4,
419 .round_trip_ping_latency_dcfclk_cycles = 106,
420 .urgent_latency_pixel_data_only_us = 4.0,
421 .urgent_latency_pixel_mixed_with_vm_data_us = 4.0,
422 .urgent_latency_vm_data_only_us = 4.0,
423 .urgent_out_of_order_return_per_channel_pixel_only_bytes = 4096,
424 .urgent_out_of_order_return_per_channel_pixel_and_vm_bytes = 4096,
425 .urgent_out_of_order_return_per_channel_vm_only_bytes = 4096,
426 .pct_ideal_sdp_bw_after_urgent = 80.0,
427 .pct_ideal_dram_sdp_bw_after_urgent_pixel_only = 65.0,
428 .pct_ideal_dram_sdp_bw_after_urgent_pixel_and_vm = 60.0,
429 .pct_ideal_dram_sdp_bw_after_urgent_vm_only = 30.0,
430 .max_avg_sdp_bw_use_normal_percent = 60.0,
431 .max_avg_dram_bw_use_normal_percent = 60.0,
432 .fabric_datapath_to_dcn_data_return_bytes = 32,
433 .return_bus_width_bytes = 64,
434 .downspread_percent = 0.38,
435 .dcn_downspread_percent = 0.5,
436 .gpuvm_min_page_size_bytes = 4096,
437 .hostvm_min_page_size_bytes = 4096,
438 .do_urgent_latency_adjustment = false,
439 .urgent_latency_adjustment_fabric_clock_component_us = 0,
440 .urgent_latency_adjustment_fabric_clock_reference_mhz = 0,
441};
442
443void dcn31_zero_pipe_dcc_fraction(display_e2e_pipe_params_st *pipes,
444 int pipe_cnt)
445{
446 dc_assert_fp_enabled();
447
448 pipes[pipe_cnt].pipe.src.dcc_fraction_of_zs_req_luma = 0;
449 pipes[pipe_cnt].pipe.src.dcc_fraction_of_zs_req_chroma = 0;
450}
451
452void dcn31_update_soc_for_wm_a(struct dc *dc, struct dc_state *context)
453{
454 dc_assert_fp_enabled();
455
456 if (dc->clk_mgr->bw_params->wm_table.entries[WM_A].valid) {
457 context->bw_ctx.dml.soc.dram_clock_change_latency_us = dc->clk_mgr->bw_params->wm_table.entries[WM_A].pstate_latency_us;
458 context->bw_ctx.dml.soc.sr_enter_plus_exit_time_us = dc->clk_mgr->bw_params->wm_table.entries[WM_A].sr_enter_plus_exit_time_us;
459 context->bw_ctx.dml.soc.sr_exit_time_us = dc->clk_mgr->bw_params->wm_table.entries[WM_A].sr_exit_time_us;
460 }
461}
462
463void dcn315_update_soc_for_wm_a(struct dc *dc, struct dc_state *context)
464{
465 dc_assert_fp_enabled();
466
467 if (dc->clk_mgr->bw_params->wm_table.entries[WM_A].valid) {
468 /* For 315 pstate change is only supported if possible in vactive */
469 if (context->bw_ctx.dml.vba.DRAMClockChangeSupport[context->bw_ctx.dml.vba.VoltageLevel][context->bw_ctx.dml.vba.maxMpcComb] != dm_dram_clock_change_vactive)
470 context->bw_ctx.dml.soc.dram_clock_change_latency_us = context->bw_ctx.dml.soc.dummy_pstate_latency_us;
471 else
472 context->bw_ctx.dml.soc.dram_clock_change_latency_us = dc->clk_mgr->bw_params->wm_table.entries[WM_A].pstate_latency_us;
473 context->bw_ctx.dml.soc.sr_enter_plus_exit_time_us =
474 dc->clk_mgr->bw_params->wm_table.entries[WM_A].sr_enter_plus_exit_time_us;
475 context->bw_ctx.dml.soc.sr_exit_time_us =
476 dc->clk_mgr->bw_params->wm_table.entries[WM_A].sr_exit_time_us;
477 }
478}
479
480void dcn31_calculate_wm_and_dlg_fp(
481 struct dc *dc, struct dc_state *context,
482 display_e2e_pipe_params_st *pipes,
483 int pipe_cnt,
484 int vlevel)
485{
486 int i, pipe_idx, total_det = 0, active_hubp_count = 0;
487 double dcfclk = context->bw_ctx.dml.vba.DCFCLKState[vlevel][context->bw_ctx.dml.vba.maxMpcComb];
488
489 dc_assert_fp_enabled();
490
491 if (context->bw_ctx.dml.soc.min_dcfclk > dcfclk)
492 dcfclk = context->bw_ctx.dml.soc.min_dcfclk;
493
494 /* We don't recalculate clocks for 0 pipe configs, which can block
495 * S0i3 as high clocks will block low power states
496 * Override any clocks that can block S0i3 to min here
497 */
498 if (pipe_cnt == 0) {
499 context->bw_ctx.bw.dcn.clk.dcfclk_khz = dcfclk; // always should be vlevel 0
500 return;
501 }
502
503 pipes[0].clks_cfg.voltage = vlevel;
504 pipes[0].clks_cfg.dcfclk_mhz = dcfclk;
505 pipes[0].clks_cfg.socclk_mhz = context->bw_ctx.dml.soc.clock_limits[vlevel].socclk_mhz;
506
507 /* Set A:
508 * All clocks min required
509 *
510 * Set A calculated last so that following calculations are based on Set A
511 */
512 dc->res_pool->funcs->update_soc_for_wm_a(dc, context);
513 context->bw_ctx.bw.dcn.watermarks.a.urgent_ns = get_wm_urgent(mode_lib: &context->bw_ctx.dml, pipes, num_pipes: pipe_cnt) * 1000;
514 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.cstate_enter_plus_exit_ns = get_wm_stutter_enter_exit(mode_lib: &context->bw_ctx.dml, pipes, num_pipes: pipe_cnt) * 1000;
515 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(mode_lib: &context->bw_ctx.dml, pipes, num_pipes: pipe_cnt) * 1000;
516 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.pstate_change_ns = get_wm_dram_clock_change(mode_lib: &context->bw_ctx.dml, pipes, num_pipes: pipe_cnt) * 1000;
517 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.cstate_enter_plus_exit_z8_ns = get_wm_z8_stutter_enter_exit(mode_lib: &context->bw_ctx.dml, pipes, num_pipes: pipe_cnt) * 1000;
518 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.cstate_exit_z8_ns = get_wm_z8_stutter_exit(mode_lib: &context->bw_ctx.dml, pipes, num_pipes: pipe_cnt) * 1000;
519 context->bw_ctx.bw.dcn.watermarks.a.pte_meta_urgent_ns = get_wm_memory_trip(mode_lib: &context->bw_ctx.dml, pipes, num_pipes: pipe_cnt) * 1000;
520 context->bw_ctx.bw.dcn.watermarks.a.frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(mode_lib: &context->bw_ctx.dml, pipes, num_pipes: pipe_cnt) * 1000;
521 context->bw_ctx.bw.dcn.watermarks.a.frac_urg_bw_flip = get_fraction_of_urgent_bandwidth_imm_flip(mode_lib: &context->bw_ctx.dml, pipes, num_pipes: pipe_cnt) * 1000;
522 context->bw_ctx.bw.dcn.watermarks.a.urgent_latency_ns = get_urgent_latency(mode_lib: &context->bw_ctx.dml, pipes, num_pipes: pipe_cnt) * 1000;
523 context->bw_ctx.bw.dcn.watermarks.b = context->bw_ctx.bw.dcn.watermarks.a;
524 context->bw_ctx.bw.dcn.watermarks.c = context->bw_ctx.bw.dcn.watermarks.a;
525 context->bw_ctx.bw.dcn.watermarks.d = context->bw_ctx.bw.dcn.watermarks.a;
526
527 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
528 if (!context->res_ctx.pipe_ctx[i].stream)
529 continue;
530
531 if (context->res_ctx.pipe_ctx[i].plane_state)
532 active_hubp_count++;
533
534 pipes[pipe_idx].clks_cfg.dispclk_mhz = get_dispclk_calculated(mode_lib: &context->bw_ctx.dml, pipes, num_pipes: pipe_cnt);
535 pipes[pipe_idx].clks_cfg.dppclk_mhz = get_dppclk_calculated(mode_lib: &context->bw_ctx.dml, pipes, num_pipes: pipe_cnt, which_pipe: pipe_idx);
536
537 if (dc->config.forced_clocks || dc->debug.max_disp_clk) {
538 pipes[pipe_idx].clks_cfg.dispclk_mhz = context->bw_ctx.dml.soc.clock_limits[0].dispclk_mhz;
539 pipes[pipe_idx].clks_cfg.dppclk_mhz = context->bw_ctx.dml.soc.clock_limits[0].dppclk_mhz;
540 }
541 if (dc->debug.min_disp_clk_khz > pipes[pipe_idx].clks_cfg.dispclk_mhz * 1000)
542 pipes[pipe_idx].clks_cfg.dispclk_mhz = dc->debug.min_disp_clk_khz / 1000.0;
543 if (dc->debug.min_dpp_clk_khz > pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000)
544 pipes[pipe_idx].clks_cfg.dppclk_mhz = dc->debug.min_dpp_clk_khz / 1000.0;
545
546 pipe_idx++;
547 }
548
549 dcn20_calculate_dlg_params(dc, context, pipes, pipe_cnt, vlevel);
550 /* For 31x apu pstate change is only supported if possible in vactive*/
551 context->bw_ctx.bw.dcn.clk.p_state_change_support =
552 context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] == dm_dram_clock_change_vactive;
553 /* If DCN isn't making memory requests we can allow pstate change and lower clocks */
554 if (!active_hubp_count) {
555 context->bw_ctx.bw.dcn.clk.socclk_khz = 0;
556 context->bw_ctx.bw.dcn.clk.dppclk_khz = 0;
557 context->bw_ctx.bw.dcn.clk.dcfclk_khz = 0;
558 context->bw_ctx.bw.dcn.clk.dcfclk_deep_sleep_khz = 0;
559 context->bw_ctx.bw.dcn.clk.dramclk_khz = 0;
560 context->bw_ctx.bw.dcn.clk.fclk_khz = 0;
561 context->bw_ctx.bw.dcn.clk.p_state_change_support = true;
562 for (i = 0; i < dc->res_pool->pipe_count; i++)
563 if (context->res_ctx.pipe_ctx[i].stream)
564 context->res_ctx.pipe_ctx[i].plane_res.bw.dppclk_khz = 0;
565 }
566 for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
567 if (!context->res_ctx.pipe_ctx[i].stream)
568 continue;
569
570 context->res_ctx.pipe_ctx[i].det_buffer_size_kb =
571 get_det_buffer_size_kbytes(mode_lib: &context->bw_ctx.dml, pipes, num_pipes: pipe_cnt, which_pipe: pipe_idx);
572 if (context->res_ctx.pipe_ctx[i].det_buffer_size_kb > 384)
573 context->res_ctx.pipe_ctx[i].det_buffer_size_kb /= 2;
574 total_det += context->res_ctx.pipe_ctx[i].det_buffer_size_kb;
575 pipe_idx++;
576 }
577 context->bw_ctx.bw.dcn.compbuf_size_kb = context->bw_ctx.dml.ip.config_return_buffer_size_in_kbytes - total_det;
578}
579
580void dcn31_update_bw_bounding_box(struct dc *dc, struct clk_bw_params *bw_params)
581{
582 struct _vcs_dpi_voltage_scaling_st *s = dc->scratch.update_bw_bounding_box.clock_limits;
583 struct clk_limit_table *clk_table = &bw_params->clk_table;
584 unsigned int i, closest_clk_lvl;
585 int max_dispclk_mhz = 0, max_dppclk_mhz = 0;
586 int j;
587
588 dc_assert_fp_enabled();
589
590 memcpy(s, dcn3_1_soc.clock_limits, sizeof(dcn3_1_soc.clock_limits));
591
592 // Default clock levels are used for diags, which may lead to overclocking.
593 dcn3_1_ip.max_num_otg = dc->res_pool->res_cap->num_timing_generator;
594 dcn3_1_ip.max_num_dpp = dc->res_pool->pipe_count;
595 dcn3_1_soc.num_chans = bw_params->num_channels;
596
597 ASSERT(clk_table->num_entries);
598
599 /* Prepass to find max clocks independent of voltage level. */
600 for (i = 0; i < clk_table->num_entries; ++i) {
601 if (clk_table->entries[i].dispclk_mhz > max_dispclk_mhz)
602 max_dispclk_mhz = clk_table->entries[i].dispclk_mhz;
603 if (clk_table->entries[i].dppclk_mhz > max_dppclk_mhz)
604 max_dppclk_mhz = clk_table->entries[i].dppclk_mhz;
605 }
606
607 for (i = 0; i < clk_table->num_entries; i++) {
608 /* loop backwards*/
609 for (closest_clk_lvl = 0, j = dcn3_1_soc.num_states - 1; j >= 0; j--) {
610 if ((unsigned int) dcn3_1_soc.clock_limits[j].dcfclk_mhz <= clk_table->entries[i].dcfclk_mhz) {
611 closest_clk_lvl = j;
612 break;
613 }
614 }
615
616 s[i].state = i;
617
618 /* Clocks dependent on voltage level. */
619 s[i].dcfclk_mhz = clk_table->entries[i].dcfclk_mhz;
620 s[i].fabricclk_mhz = clk_table->entries[i].fclk_mhz;
621 s[i].socclk_mhz = clk_table->entries[i].socclk_mhz;
622 s[i].dram_speed_mts = clk_table->entries[i].memclk_mhz *
623 2 * clk_table->entries[i].wck_ratio;
624
625 /* Clocks independent of voltage level. */
626 s[i].dispclk_mhz = max_dispclk_mhz ? max_dispclk_mhz :
627 dcn3_1_soc.clock_limits[closest_clk_lvl].dispclk_mhz;
628
629 s[i].dppclk_mhz = max_dppclk_mhz ? max_dppclk_mhz :
630 dcn3_1_soc.clock_limits[closest_clk_lvl].dppclk_mhz;
631
632 s[i].dram_bw_per_chan_gbps =
633 dcn3_1_soc.clock_limits[closest_clk_lvl].dram_bw_per_chan_gbps;
634 s[i].dscclk_mhz = dcn3_1_soc.clock_limits[closest_clk_lvl].dscclk_mhz;
635 s[i].dtbclk_mhz = dcn3_1_soc.clock_limits[closest_clk_lvl].dtbclk_mhz;
636 s[i].phyclk_d18_mhz =
637 dcn3_1_soc.clock_limits[closest_clk_lvl].phyclk_d18_mhz;
638 s[i].phyclk_mhz = dcn3_1_soc.clock_limits[closest_clk_lvl].phyclk_mhz;
639 }
640 if (clk_table->num_entries) {
641 dcn3_1_soc.num_states = clk_table->num_entries;
642 }
643
644 memcpy(dcn3_1_soc.clock_limits, s, sizeof(dcn3_1_soc.clock_limits));
645
646 dcn3_1_soc.dispclk_dppclk_vco_speed_mhz = dc->clk_mgr->dentist_vco_freq_khz / 1000.0;
647 dc->dml.soc.dispclk_dppclk_vco_speed_mhz = dc->clk_mgr->dentist_vco_freq_khz / 1000.0;
648
649 if ((int)(dcn3_1_soc.dram_clock_change_latency_us * 1000)
650 != dc->debug.dram_clock_change_latency_ns
651 && dc->debug.dram_clock_change_latency_ns) {
652 dcn3_1_soc.dram_clock_change_latency_us = dc->debug.dram_clock_change_latency_ns / 1000;
653 }
654
655 dml_init_instance(lib: &dc->dml, soc_bb: &dcn3_1_soc, ip_params: &dcn3_1_ip, project: DML_PROJECT_DCN31);
656}
657
658void dcn315_update_bw_bounding_box(struct dc *dc, struct clk_bw_params *bw_params)
659{
660 struct clk_limit_table *clk_table = &bw_params->clk_table;
661 int i, max_dispclk_mhz = 0, max_dppclk_mhz = 0;
662
663 dc_assert_fp_enabled();
664
665 dcn3_15_ip.max_num_otg = dc->res_pool->res_cap->num_timing_generator;
666 dcn3_15_ip.max_num_dpp = dc->res_pool->pipe_count;
667
668 if (bw_params->num_channels > 0)
669 dcn3_15_soc.num_chans = bw_params->num_channels;
670 if (bw_params->dram_channel_width_bytes > 0)
671 dcn3_15_soc.dram_channel_width_bytes = bw_params->dram_channel_width_bytes;
672
673 ASSERT(clk_table->num_entries);
674
675 /* Setup soc to always use max dispclk/dppclk to avoid odm-to-lower-voltage */
676 for (i = 0; i < clk_table->num_entries; ++i) {
677 if (clk_table->entries[i].dispclk_mhz > max_dispclk_mhz)
678 max_dispclk_mhz = clk_table->entries[i].dispclk_mhz;
679 if (clk_table->entries[i].dppclk_mhz > max_dppclk_mhz)
680 max_dppclk_mhz = clk_table->entries[i].dppclk_mhz;
681 }
682
683 for (i = 0; i < clk_table->num_entries; i++) {
684 dcn3_15_soc.clock_limits[i].state = i;
685
686 /* Clocks dependent on voltage level. */
687 dcn3_15_soc.clock_limits[i].dcfclk_mhz = clk_table->entries[i].dcfclk_mhz;
688 dcn3_15_soc.clock_limits[i].fabricclk_mhz = clk_table->entries[i].fclk_mhz;
689 dcn3_15_soc.clock_limits[i].socclk_mhz = clk_table->entries[i].socclk_mhz;
690 dcn3_15_soc.clock_limits[i].dram_speed_mts = clk_table->entries[i].memclk_mhz * 2 * clk_table->entries[i].wck_ratio;
691
692 /* These aren't actually read from smu, but rather set in clk_mgr defaults */
693 dcn3_15_soc.clock_limits[i].dtbclk_mhz = clk_table->entries[i].dtbclk_mhz;
694 dcn3_15_soc.clock_limits[i].phyclk_d18_mhz = clk_table->entries[i].phyclk_d18_mhz;
695 dcn3_15_soc.clock_limits[i].phyclk_mhz = clk_table->entries[i].phyclk_mhz;
696
697 /* Clocks independent of voltage level. */
698 dcn3_15_soc.clock_limits[i].dispclk_mhz = max_dispclk_mhz;
699 dcn3_15_soc.clock_limits[i].dppclk_mhz = max_dppclk_mhz;
700 dcn3_15_soc.clock_limits[i].dscclk_mhz = max_dispclk_mhz / 3.0;
701 }
702 dcn3_15_soc.num_states = clk_table->num_entries;
703
704
705 /* Set vco to max_dispclk * 2 to make sure the highest dispclk is always available for dml calcs,
706 * no impact outside of dml validation
707 */
708 dcn3_15_soc.dispclk_dppclk_vco_speed_mhz = max_dispclk_mhz * 2;
709
710 if ((int)(dcn3_15_soc.dram_clock_change_latency_us * 1000)
711 != dc->debug.dram_clock_change_latency_ns
712 && dc->debug.dram_clock_change_latency_ns) {
713 dcn3_15_soc.dram_clock_change_latency_us = dc->debug.dram_clock_change_latency_ns / 1000;
714 }
715
716 dml_init_instance(lib: &dc->dml, soc_bb: &dcn3_15_soc, ip_params: &dcn3_15_ip, project: DML_PROJECT_DCN315);
717}
718
719void dcn316_update_bw_bounding_box(struct dc *dc, struct clk_bw_params *bw_params)
720{
721 struct _vcs_dpi_voltage_scaling_st *s = dc->scratch.update_bw_bounding_box.clock_limits;
722 struct clk_limit_table *clk_table = &bw_params->clk_table;
723 unsigned int i, closest_clk_lvl;
724 int max_dispclk_mhz = 0, max_dppclk_mhz = 0;
725 int j;
726
727 dc_assert_fp_enabled();
728
729 memcpy(s, dcn3_16_soc.clock_limits, sizeof(dcn3_16_soc.clock_limits));
730
731 // Default clock levels are used for diags, which may lead to overclocking.
732 dcn3_16_ip.max_num_otg = dc->res_pool->res_cap->num_timing_generator;
733 dcn3_16_ip.max_num_dpp = dc->res_pool->pipe_count;
734 dcn3_16_soc.num_chans = bw_params->num_channels;
735
736 ASSERT(clk_table->num_entries);
737
738 /* Prepass to find max clocks independent of voltage level. */
739 for (i = 0; i < clk_table->num_entries; ++i) {
740 if (clk_table->entries[i].dispclk_mhz > max_dispclk_mhz)
741 max_dispclk_mhz = clk_table->entries[i].dispclk_mhz;
742 if (clk_table->entries[i].dppclk_mhz > max_dppclk_mhz)
743 max_dppclk_mhz = clk_table->entries[i].dppclk_mhz;
744 }
745
746 for (i = 0; i < clk_table->num_entries; i++) {
747 /* loop backwards*/
748 for (closest_clk_lvl = 0, j = dcn3_16_soc.num_states - 1; j >= 0; j--) {
749 if ((unsigned int) dcn3_16_soc.clock_limits[j].dcfclk_mhz <=
750 clk_table->entries[i].dcfclk_mhz) {
751 closest_clk_lvl = j;
752 break;
753 }
754 }
755 // Ported from DCN315
756 if (clk_table->num_entries == 1) {
757 /*smu gives one DPM level, let's take the highest one*/
758 closest_clk_lvl = dcn3_16_soc.num_states - 1;
759 }
760
761 s[i].state = i;
762
763 /* Clocks dependent on voltage level. */
764 s[i].dcfclk_mhz = clk_table->entries[i].dcfclk_mhz;
765 if (clk_table->num_entries == 1 &&
766 s[i].dcfclk_mhz <
767 dcn3_16_soc.clock_limits[closest_clk_lvl].dcfclk_mhz) {
768 /*SMU fix not released yet*/
769 s[i].dcfclk_mhz =
770 dcn3_16_soc.clock_limits[closest_clk_lvl].dcfclk_mhz;
771 }
772 s[i].fabricclk_mhz = clk_table->entries[i].fclk_mhz;
773 s[i].socclk_mhz = clk_table->entries[i].socclk_mhz;
774 s[i].dram_speed_mts = clk_table->entries[i].memclk_mhz *
775 2 * clk_table->entries[i].wck_ratio;
776
777 /* Clocks independent of voltage level. */
778 s[i].dispclk_mhz = max_dispclk_mhz ? max_dispclk_mhz :
779 dcn3_16_soc.clock_limits[closest_clk_lvl].dispclk_mhz;
780
781 s[i].dppclk_mhz = max_dppclk_mhz ? max_dppclk_mhz :
782 dcn3_16_soc.clock_limits[closest_clk_lvl].dppclk_mhz;
783
784 s[i].dram_bw_per_chan_gbps =
785 dcn3_16_soc.clock_limits[closest_clk_lvl].dram_bw_per_chan_gbps;
786 s[i].dscclk_mhz = dcn3_16_soc.clock_limits[closest_clk_lvl].dscclk_mhz;
787 s[i].dtbclk_mhz = dcn3_16_soc.clock_limits[closest_clk_lvl].dtbclk_mhz;
788 s[i].phyclk_d18_mhz =
789 dcn3_16_soc.clock_limits[closest_clk_lvl].phyclk_d18_mhz;
790 s[i].phyclk_mhz = dcn3_16_soc.clock_limits[closest_clk_lvl].phyclk_mhz;
791 }
792 if (clk_table->num_entries) {
793 dcn3_16_soc.num_states = clk_table->num_entries;
794 }
795
796 memcpy(dcn3_16_soc.clock_limits, s, sizeof(dcn3_16_soc.clock_limits));
797
798 if (max_dispclk_mhz) {
799 dcn3_16_soc.dispclk_dppclk_vco_speed_mhz = max_dispclk_mhz * 2;
800 dc->dml.soc.dispclk_dppclk_vco_speed_mhz = max_dispclk_mhz * 2;
801 }
802 if ((int)(dcn3_16_soc.dram_clock_change_latency_us * 1000)
803 != dc->debug.dram_clock_change_latency_ns
804 && dc->debug.dram_clock_change_latency_ns) {
805 dcn3_16_soc.dram_clock_change_latency_us = dc->debug.dram_clock_change_latency_ns / 1000;
806 }
807
808 dml_init_instance(lib: &dc->dml, soc_bb: &dcn3_16_soc, ip_params: &dcn3_16_ip, project: DML_PROJECT_DCN31);
809}
810
811int dcn_get_max_non_odm_pix_rate_100hz(struct _vcs_dpi_soc_bounding_box_st *soc)
812{
813 return soc->clock_limits[0].dispclk_mhz * 10000.0 / (1.0 + soc->dcn_downspread_percent / 100.0);
814}
815
816int dcn_get_approx_det_segs_required_for_pstate(
817 struct _vcs_dpi_soc_bounding_box_st *soc,
818 int pix_clk_100hz, int bpp, int seg_size_kb)
819{
820 /* Roughly calculate required crb to hide latency. In practice there is slightly
821 * more buffer available for latency hiding
822 */
823 return (int)(soc->dram_clock_change_latency_us * pix_clk_100hz * bpp
824 / 10240000 + seg_size_kb - 1) / seg_size_kb;
825}
826

source code of linux/drivers/gpu/drm/amd/display/dc/dml/dcn31/dcn31_fpu.c