Merge tag 'amd-drm-next-6.1-2022-09-08' of https://gitlab.freedesktop.org/agd5f/linux...
[linux-2.6-block.git] / drivers / gpu / drm / amd / display / dc / dml / dcn301 / dcn301_fpu.c
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 #include "resource.h"
26 #include "clk_mgr.h"
27 #include "dcn20/dcn20_resource.h"
28 #include "dcn301/dcn301_resource.h"
29 #include "clk_mgr/dcn301/vg_clk_mgr.h"
30
31 #include "dml/dcn20/dcn20_fpu.h"
32 #include "dcn301_fpu.h"
33
34 #define TO_DCN301_RES_POOL(pool)\
35         container_of(pool, struct dcn301_resource_pool, base)
36
37 /* Based on: //vidip/dc/dcn3/doc/architecture/DCN3x_Display_Mode.xlsm#83 */
38 struct _vcs_dpi_ip_params_st dcn3_01_ip = {
39         .odm_capable = 1,
40         .gpuvm_enable = 1,
41         .hostvm_enable = 1,
42         .gpuvm_max_page_table_levels = 1,
43         .hostvm_max_page_table_levels = 2,
44         .hostvm_cached_page_table_levels = 0,
45         .pte_group_size_bytes = 2048,
46         .num_dsc = 3,
47         .rob_buffer_size_kbytes = 184,
48         .det_buffer_size_kbytes = 184,
49         .dpte_buffer_size_in_pte_reqs_luma = 64,
50         .dpte_buffer_size_in_pte_reqs_chroma = 32,
51         .pde_proc_buffer_size_64k_reqs = 48,
52         .dpp_output_buffer_pixels = 2560,
53         .opp_output_buffer_lines = 1,
54         .pixel_chunk_size_kbytes = 8,
55         .meta_chunk_size_kbytes = 2,
56         .writeback_chunk_size_kbytes = 8,
57         .line_buffer_size_bits = 789504,
58         .is_line_buffer_bpp_fixed = 0,  // ?
59         .line_buffer_fixed_bpp = 48,     // ?
60         .dcc_supported = true,
61         .writeback_interface_buffer_size_kbytes = 90,
62         .writeback_line_buffer_buffer_size = 656640,
63         .max_line_buffer_lines = 12,
64         .writeback_luma_buffer_size_kbytes = 12,  // writeback_line_buffer_buffer_size = 656640
65         .writeback_chroma_buffer_size_kbytes = 8,
66         .writeback_chroma_line_buffer_width_pixels = 4,
67         .writeback_max_hscl_ratio = 1,
68         .writeback_max_vscl_ratio = 1,
69         .writeback_min_hscl_ratio = 1,
70         .writeback_min_vscl_ratio = 1,
71         .writeback_max_hscl_taps = 1,
72         .writeback_max_vscl_taps = 1,
73         .writeback_line_buffer_luma_buffer_size = 0,
74         .writeback_line_buffer_chroma_buffer_size = 14643,
75         .cursor_buffer_size = 8,
76         .cursor_chunk_size = 2,
77         .max_num_otg = 4,
78         .max_num_dpp = 4,
79         .max_num_wb = 1,
80         .max_dchub_pscl_bw_pix_per_clk = 4,
81         .max_pscl_lb_bw_pix_per_clk = 2,
82         .max_lb_vscl_bw_pix_per_clk = 4,
83         .max_vscl_hscl_bw_pix_per_clk = 4,
84         .max_hscl_ratio = 6,
85         .max_vscl_ratio = 6,
86         .hscl_mults = 4,
87         .vscl_mults = 4,
88         .max_hscl_taps = 8,
89         .max_vscl_taps = 8,
90         .dispclk_ramp_margin_percent = 1,
91         .underscan_factor = 1.11,
92         .min_vblank_lines = 32,
93         .dppclk_delay_subtotal = 46,
94         .dynamic_metadata_vm_enabled = true,
95         .dppclk_delay_scl_lb_only = 16,
96         .dppclk_delay_scl = 50,
97         .dppclk_delay_cnvc_formatter = 27,
98         .dppclk_delay_cnvc_cursor = 6,
99         .dispclk_delay_subtotal = 119,
100         .dcfclk_cstate_latency = 5.2, // SRExitTime
101         .max_inter_dcn_tile_repeaters = 8,
102         .max_num_hdmi_frl_outputs = 0,
103         .odm_combine_4to1_supported = true,
104
105         .xfc_supported = false,
106         .xfc_fill_bw_overhead_percent = 10.0,
107         .xfc_fill_constant_bytes = 0,
108         .gfx7_compat_tiling_supported = 0,
109         .number_of_cursors = 1,
110 };
111
112 struct _vcs_dpi_soc_bounding_box_st dcn3_01_soc = {
113         .clock_limits = {
114                 {
115                         .state = 0,
116                         .dram_speed_mts = 2400.0,
117                         .fabricclk_mhz = 600,
118                         .socclk_mhz = 278.0,
119                         .dcfclk_mhz = 400.0,
120                         .dscclk_mhz = 206.0,
121                         .dppclk_mhz = 1015.0,
122                         .dispclk_mhz = 1015.0,
123                         .phyclk_mhz = 600.0,
124                 },
125
126                 {
127                         .state = 1,
128                         .dram_speed_mts = 2400.0,
129                         .fabricclk_mhz = 688,
130                         .socclk_mhz = 278.0,
131                         .dcfclk_mhz = 400.0,
132                         .dscclk_mhz = 206.0,
133                         .dppclk_mhz = 1015.0,
134                         .dispclk_mhz = 1015.0,
135                         .phyclk_mhz = 600.0,
136                 },
137
138                 {
139                         .state = 2,
140                         .dram_speed_mts = 4267.0,
141                         .fabricclk_mhz = 1067,
142                         .socclk_mhz = 278.0,
143                         .dcfclk_mhz = 608.0,
144                         .dscclk_mhz = 296.0,
145                         .dppclk_mhz = 1015.0,
146                         .dispclk_mhz = 1015.0,
147                         .phyclk_mhz = 810.0,
148                 },
149
150                 {
151                         .state = 3,
152                         .dram_speed_mts = 4267.0,
153                         .fabricclk_mhz = 1067,
154                         .socclk_mhz = 715.0,
155                         .dcfclk_mhz = 676.0,
156                         .dscclk_mhz = 338.0,
157                         .dppclk_mhz = 1015.0,
158                         .dispclk_mhz = 1015.0,
159                         .phyclk_mhz = 810.0,
160                 },
161
162                 {
163                         .state = 4,
164                         .dram_speed_mts = 4267.0,
165                         .fabricclk_mhz = 1067,
166                         .socclk_mhz = 953.0,
167                         .dcfclk_mhz = 810.0,
168                         .dscclk_mhz = 338.0,
169                         .dppclk_mhz = 1015.0,
170                         .dispclk_mhz = 1015.0,
171                         .phyclk_mhz = 810.0,
172                 },
173         },
174
175         .sr_exit_time_us = 9.0,
176         .sr_enter_plus_exit_time_us = 11.0,
177         .urgent_latency_us = 4.0,
178         .urgent_latency_pixel_data_only_us = 4.0,
179         .urgent_latency_pixel_mixed_with_vm_data_us = 4.0,
180         .urgent_latency_vm_data_only_us = 4.0,
181         .urgent_out_of_order_return_per_channel_pixel_only_bytes = 4096,
182         .urgent_out_of_order_return_per_channel_pixel_and_vm_bytes = 4096,
183         .urgent_out_of_order_return_per_channel_vm_only_bytes = 4096,
184         .pct_ideal_dram_sdp_bw_after_urgent_pixel_only = 80.0,
185         .pct_ideal_dram_sdp_bw_after_urgent_pixel_and_vm = 75.0,
186         .pct_ideal_dram_sdp_bw_after_urgent_vm_only = 40.0,
187         .max_avg_sdp_bw_use_normal_percent = 60.0,
188         .max_avg_dram_bw_use_normal_percent = 60.0,
189         .writeback_latency_us = 12.0,
190         .max_request_size_bytes = 256,
191         .dram_channel_width_bytes = 4,
192         .fabric_datapath_to_dcn_data_return_bytes = 32,
193         .dcn_downspread_percent = 0.5,
194         .downspread_percent = 0.38,
195         .dram_page_open_time_ns = 50.0,
196         .dram_rw_turnaround_time_ns = 17.5,
197         .dram_return_buffer_per_channel_bytes = 8192,
198         .round_trip_ping_latency_dcfclk_cycles = 191,
199         .urgent_out_of_order_return_per_channel_bytes = 4096,
200         .channel_interleave_bytes = 256,
201         .num_banks = 8,
202         .num_chans = 4,
203         .gpuvm_min_page_size_bytes = 4096,
204         .hostvm_min_page_size_bytes = 4096,
205         .dram_clock_change_latency_us = 23.84,
206         .writeback_dram_clock_change_latency_us = 23.0,
207         .return_bus_width_bytes = 64,
208         .dispclk_dppclk_vco_speed_mhz = 3550,
209         .xfc_bus_transport_time_us = 20,      // ?
210         .xfc_xbuf_latency_tolerance_us = 4,  // ?
211         .use_urgent_burst_bw = 1,            // ?
212         .num_states = 5,
213         .do_urgent_latency_adjustment = false,
214         .urgent_latency_adjustment_fabric_clock_component_us = 0,
215         .urgent_latency_adjustment_fabric_clock_reference_mhz = 0,
216 };
217
218 struct wm_table ddr4_wm_table = {
219         .entries = {
220                 {
221                         .wm_inst = WM_A,
222                         .wm_type = WM_TYPE_PSTATE_CHG,
223                         .pstate_latency_us = 11.72,
224                         .sr_exit_time_us = 6.09,
225                         .sr_enter_plus_exit_time_us = 7.14,
226                         .valid = true,
227                 },
228                 {
229                         .wm_inst = WM_B,
230                         .wm_type = WM_TYPE_PSTATE_CHG,
231                         .pstate_latency_us = 11.72,
232                         .sr_exit_time_us = 10.12,
233                         .sr_enter_plus_exit_time_us = 11.48,
234                         .valid = true,
235                 },
236                 {
237                         .wm_inst = WM_C,
238                         .wm_type = WM_TYPE_PSTATE_CHG,
239                         .pstate_latency_us = 11.72,
240                         .sr_exit_time_us = 10.12,
241                         .sr_enter_plus_exit_time_us = 11.48,
242                         .valid = true,
243                 },
244                 {
245                         .wm_inst = WM_D,
246                         .wm_type = WM_TYPE_PSTATE_CHG,
247                         .pstate_latency_us = 11.72,
248                         .sr_exit_time_us = 10.12,
249                         .sr_enter_plus_exit_time_us = 11.48,
250                         .valid = true,
251                 },
252         }
253 };
254
255 struct wm_table lpddr5_wm_table = {
256         .entries = {
257                 {
258                         .wm_inst = WM_A,
259                         .wm_type = WM_TYPE_PSTATE_CHG,
260                         .pstate_latency_us = 11.65333,
261                         .sr_exit_time_us = 13.5,
262                         .sr_enter_plus_exit_time_us = 16.5,
263                         .valid = true,
264                 },
265                 {
266                         .wm_inst = WM_B,
267                         .wm_type = WM_TYPE_PSTATE_CHG,
268                         .pstate_latency_us = 11.65333,
269                         .sr_exit_time_us = 13.5,
270                         .sr_enter_plus_exit_time_us = 16.5,
271                         .valid = true,
272                 },
273                 {
274                         .wm_inst = WM_C,
275                         .wm_type = WM_TYPE_PSTATE_CHG,
276                         .pstate_latency_us = 11.65333,
277                         .sr_exit_time_us = 13.5,
278                         .sr_enter_plus_exit_time_us = 16.5,
279                         .valid = true,
280                 },
281                 {
282                         .wm_inst = WM_D,
283                         .wm_type = WM_TYPE_PSTATE_CHG,
284                         .pstate_latency_us = 11.65333,
285                         .sr_exit_time_us = 13.5,
286                         .sr_enter_plus_exit_time_us = 16.5,
287                         .valid = true,
288                 },
289         }
290 };
291
292 static void calculate_wm_set_for_vlevel(int vlevel,
293                 struct wm_range_table_entry *table_entry,
294                 struct dcn_watermarks *wm_set,
295                 struct display_mode_lib *dml,
296                 display_e2e_pipe_params_st *pipes,
297                 int pipe_cnt)
298 {
299         double dram_clock_change_latency_cached = dml->soc.dram_clock_change_latency_us;
300
301         ASSERT(vlevel < dml->soc.num_states);
302         /* only pipe 0 is read for voltage and dcf/soc clocks */
303         pipes[0].clks_cfg.voltage = vlevel;
304         pipes[0].clks_cfg.dcfclk_mhz = dml->soc.clock_limits[vlevel].dcfclk_mhz;
305         pipes[0].clks_cfg.socclk_mhz = dml->soc.clock_limits[vlevel].socclk_mhz;
306
307         dml->soc.dram_clock_change_latency_us = table_entry->pstate_latency_us;
308         dml->soc.sr_exit_time_us = table_entry->sr_exit_time_us;
309         dml->soc.sr_enter_plus_exit_time_us = table_entry->sr_enter_plus_exit_time_us;
310
311         wm_set->urgent_ns = get_wm_urgent(dml, pipes, pipe_cnt) * 1000;
312         wm_set->cstate_pstate.cstate_enter_plus_exit_ns = get_wm_stutter_enter_exit(dml, pipes, pipe_cnt) * 1000;
313         wm_set->cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(dml, pipes, pipe_cnt) * 1000;
314         wm_set->cstate_pstate.pstate_change_ns = get_wm_dram_clock_change(dml, pipes, pipe_cnt) * 1000;
315         wm_set->pte_meta_urgent_ns = get_wm_memory_trip(dml, pipes, pipe_cnt) * 1000;
316         wm_set->frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(dml, pipes, pipe_cnt) * 1000;
317         wm_set->frac_urg_bw_flip = get_fraction_of_urgent_bandwidth_imm_flip(dml, pipes, pipe_cnt) * 1000;
318         wm_set->urgent_latency_ns = get_urgent_latency(dml, pipes, pipe_cnt) * 1000;
319         dml->soc.dram_clock_change_latency_us = dram_clock_change_latency_cached;
320
321 }
322
323 void dcn301_update_bw_bounding_box(struct dc *dc, struct clk_bw_params *bw_params)
324 {
325         struct _vcs_dpi_voltage_scaling_st *s = dc->scratch.update_bw_bounding_box.clock_limits;
326         struct dcn301_resource_pool *pool = TO_DCN301_RES_POOL(dc->res_pool);
327         struct clk_limit_table *clk_table = &bw_params->clk_table;
328         unsigned int i, closest_clk_lvl;
329         int j;
330
331         dc_assert_fp_enabled();
332
333         memcpy(s, dcn3_01_soc.clock_limits, sizeof(dcn3_01_soc.clock_limits));
334
335         /* Default clock levels are used for diags, which may lead to overclocking. */
336         if (!IS_DIAG_DC(dc->ctx->dce_environment)) {
337                 dcn3_01_ip.max_num_otg = pool->base.res_cap->num_timing_generator;
338                 dcn3_01_ip.max_num_dpp = pool->base.pipe_count;
339                 dcn3_01_soc.num_chans = bw_params->num_channels;
340
341                 ASSERT(clk_table->num_entries);
342                 for (i = 0; i < clk_table->num_entries; i++) {
343                         /* loop backwards*/
344                         for (closest_clk_lvl = 0, j = dcn3_01_soc.num_states - 1; j >= 0; j--) {
345                                 if ((unsigned int) dcn3_01_soc.clock_limits[j].dcfclk_mhz <= clk_table->entries[i].dcfclk_mhz) {
346                                         closest_clk_lvl = j;
347                                         break;
348                                 }
349                         }
350
351                         s[i].state = i;
352                         s[i].dcfclk_mhz = clk_table->entries[i].dcfclk_mhz;
353                         s[i].fabricclk_mhz = clk_table->entries[i].fclk_mhz;
354                         s[i].socclk_mhz = clk_table->entries[i].socclk_mhz;
355                         s[i].dram_speed_mts = clk_table->entries[i].memclk_mhz * 2;
356
357                         s[i].dispclk_mhz = dcn3_01_soc.clock_limits[closest_clk_lvl].dispclk_mhz;
358                         s[i].dppclk_mhz = dcn3_01_soc.clock_limits[closest_clk_lvl].dppclk_mhz;
359                         s[i].dram_bw_per_chan_gbps =
360                                 dcn3_01_soc.clock_limits[closest_clk_lvl].dram_bw_per_chan_gbps;
361                         s[i].dscclk_mhz = dcn3_01_soc.clock_limits[closest_clk_lvl].dscclk_mhz;
362                         s[i].dtbclk_mhz = dcn3_01_soc.clock_limits[closest_clk_lvl].dtbclk_mhz;
363                         s[i].phyclk_d18_mhz =
364                                 dcn3_01_soc.clock_limits[closest_clk_lvl].phyclk_d18_mhz;
365                         s[i].phyclk_mhz = dcn3_01_soc.clock_limits[closest_clk_lvl].phyclk_mhz;
366                 }
367
368                 if (clk_table->num_entries) {
369                         dcn3_01_soc.num_states = clk_table->num_entries;
370                         /* duplicate last level */
371                         s[dcn3_01_soc.num_states] =
372                                 dcn3_01_soc.clock_limits[dcn3_01_soc.num_states - 1];
373                         s[dcn3_01_soc.num_states].state = dcn3_01_soc.num_states;
374                 }
375         }
376
377         memcpy(dcn3_01_soc.clock_limits, s, sizeof(dcn3_01_soc.clock_limits));
378
379         dcn3_01_soc.dispclk_dppclk_vco_speed_mhz = dc->clk_mgr->dentist_vco_freq_khz / 1000.0;
380         dc->dml.soc.dispclk_dppclk_vco_speed_mhz = dc->clk_mgr->dentist_vco_freq_khz / 1000.0;
381
382         dml_init_instance(&dc->dml, &dcn3_01_soc, &dcn3_01_ip, DML_PROJECT_DCN30);
383 }
384
385 void dcn301_fpu_set_wm_ranges(int i,
386         struct pp_smu_wm_range_sets *ranges,
387         struct _vcs_dpi_soc_bounding_box_st *loaded_bb)
388 {
389         dc_assert_fp_enabled();
390
391         ranges->reader_wm_sets[i].min_fill_clk_mhz = (i > 0) ? (loaded_bb->clock_limits[i - 1].dram_speed_mts / 16) + 1 : 0;
392         ranges->reader_wm_sets[i].max_fill_clk_mhz = loaded_bb->clock_limits[i].dram_speed_mts / 16;
393 }
394
395 void dcn301_fpu_init_soc_bounding_box(struct bp_soc_bb_info bb_info)
396 {
397         dc_assert_fp_enabled();
398
399         if (bb_info.dram_clock_change_latency_100ns > 0)
400                 dcn3_01_soc.dram_clock_change_latency_us = bb_info.dram_clock_change_latency_100ns * 10;
401
402         if (bb_info.dram_sr_enter_exit_latency_100ns > 0)
403                 dcn3_01_soc.sr_enter_plus_exit_time_us = bb_info.dram_sr_enter_exit_latency_100ns * 10;
404
405         if (bb_info.dram_sr_exit_latency_100ns > 0)
406                 dcn3_01_soc.sr_exit_time_us = bb_info.dram_sr_exit_latency_100ns * 10;
407 }
408
409 void dcn301_calculate_wm_and_dlg_fp(struct dc *dc,
410                 struct dc_state *context,
411                 display_e2e_pipe_params_st *pipes,
412                 int pipe_cnt,
413                 int vlevel_req)
414 {
415         int i, pipe_idx;
416         int vlevel, vlevel_max;
417         struct wm_range_table_entry *table_entry;
418         struct clk_bw_params *bw_params = dc->clk_mgr->bw_params;
419
420         ASSERT(bw_params);
421         dc_assert_fp_enabled();
422
423         vlevel_max = bw_params->clk_table.num_entries - 1;
424
425         /* WM Set D */
426         table_entry = &bw_params->wm_table.entries[WM_D];
427         if (table_entry->wm_type == WM_TYPE_RETRAINING)
428                 vlevel = 0;
429         else
430                 vlevel = vlevel_max;
431         calculate_wm_set_for_vlevel(vlevel, table_entry, &context->bw_ctx.bw.dcn.watermarks.d,
432                                                 &context->bw_ctx.dml, pipes, pipe_cnt);
433         /* WM Set C */
434         table_entry = &bw_params->wm_table.entries[WM_C];
435         vlevel = min(max(vlevel_req, 2), vlevel_max);
436         calculate_wm_set_for_vlevel(vlevel, table_entry, &context->bw_ctx.bw.dcn.watermarks.c,
437                                                 &context->bw_ctx.dml, pipes, pipe_cnt);
438         /* WM Set B */
439         table_entry = &bw_params->wm_table.entries[WM_B];
440         vlevel = min(max(vlevel_req, 1), vlevel_max);
441         calculate_wm_set_for_vlevel(vlevel, table_entry, &context->bw_ctx.bw.dcn.watermarks.b,
442                                                 &context->bw_ctx.dml, pipes, pipe_cnt);
443
444         /* WM Set A */
445         table_entry = &bw_params->wm_table.entries[WM_A];
446         vlevel = min(vlevel_req, vlevel_max);
447         calculate_wm_set_for_vlevel(vlevel, table_entry, &context->bw_ctx.bw.dcn.watermarks.a,
448                                                 &context->bw_ctx.dml, pipes, pipe_cnt);
449
450         for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
451                 if (!context->res_ctx.pipe_ctx[i].stream)
452                         continue;
453
454                 pipes[pipe_idx].clks_cfg.dispclk_mhz = get_dispclk_calculated(&context->bw_ctx.dml, pipes, pipe_cnt);
455                 pipes[pipe_idx].clks_cfg.dppclk_mhz = get_dppclk_calculated(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx);
456
457                 if (dc->config.forced_clocks) {
458                         pipes[pipe_idx].clks_cfg.dispclk_mhz = context->bw_ctx.dml.soc.clock_limits[0].dispclk_mhz;
459                         pipes[pipe_idx].clks_cfg.dppclk_mhz = context->bw_ctx.dml.soc.clock_limits[0].dppclk_mhz;
460                 }
461                 if (dc->debug.min_disp_clk_khz > pipes[pipe_idx].clks_cfg.dispclk_mhz * 1000)
462                         pipes[pipe_idx].clks_cfg.dispclk_mhz = dc->debug.min_disp_clk_khz / 1000.0;
463                 if (dc->debug.min_dpp_clk_khz > pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000)
464                         pipes[pipe_idx].clks_cfg.dppclk_mhz = dc->debug.min_dpp_clk_khz / 1000.0;
465                 pipe_idx++;
466         }
467
468         dcn20_calculate_dlg_params(dc, context, pipes, pipe_cnt, vlevel);
469 }