1/* 2 * Copyright (C) 2017-2019 Alyssa Rosenzweig 3 * Copyright (C) 2017-2019 Connor Abbott 4 * 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: 11 * 12 * The above copyright notice and this permission notice (including the next 13 * paragraph) shall be included in all copies or substantial portions of the 14 * Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 22 * SOFTWARE. 23 */ 24 25#include <panfrost-job.h> 26#include <stdio.h> 27#include <stdlib.h> 28#include <memory.h> 29#include <stdbool.h> 30#include <stdarg.h> 31#include "mmap.h" 32 33#include "../pan_pretty_print.h" 34#include "../midgard/disassemble.h" 35int pandecode_replay_jc(mali_ptr jc_gpu_va, bool bifrost); 36 37#define MEMORY_PROP(obj, p) {\ 38 char *a = pointer_as_memory_reference(obj->p); \ 39 pandecode_prop("%s = %s", #p, a); \ 40 free(a); \ 41} 42 43#define MEMORY_COMMENT(obj, p) {\ 44 char *a = pointer_as_memory_reference(obj->p); \ 45 pandecode_msg("%s = %s\n", #p, a); \ 46 free(a); \ 47} 48 49#define DYN_MEMORY_PROP(obj, no, p) { \ 50 if (obj->p) \ 51 pandecode_prop("%s = %s_%d_p", #p, #p, no); \ 52} 53 54/* Semantic logging type. 55 * 56 * Raw: for raw messages to be printed as is. 57 * Message: for helpful information to be commented out in replays. 58 * Property: for properties of a struct 59 * 60 * Use one of pandecode_log, pandecode_msg, or pandecode_prop as syntax sugar. 61 */ 62 63enum pandecode_log_type { 64 PANDECODE_RAW, 65 PANDECODE_MESSAGE, 66 PANDECODE_PROPERTY 67}; 68 69#define pandecode_log(...) pandecode_log_typed(PANDECODE_RAW, __VA_ARGS__) 70#define pandecode_msg(...) pandecode_log_typed(PANDECODE_MESSAGE, __VA_ARGS__) 71#define pandecode_prop(...) pandecode_log_typed(PANDECODE_PROPERTY, __VA_ARGS__) 72 73unsigned pandecode_indent = 0; 74 75static void 76pandecode_make_indent(void) 77{ 78 for (unsigned i = 0; i < pandecode_indent; ++i) 79 printf(" "); 80} 81 82static void 83pandecode_log_typed(enum pandecode_log_type type, const char *format, ...) 84{ 85 va_list ap; 86 87 pandecode_make_indent(); 88 89 if (type == PANDECODE_MESSAGE) 90 printf("// "); 91 else if (type == PANDECODE_PROPERTY) 92 printf("."); 93 94 va_start(ap, format); 95 vprintf(format, ap); 96 va_end(ap); 97 98 if (type == PANDECODE_PROPERTY) 99 printf(",\n"); 100} 101 102static void 103pandecode_log_cont(const char *format, ...) 104{ 105 va_list ap; 106 107 va_start(ap, format); 108 vprintf(format, ap); 109 va_end(ap); 110} 111 112struct pandecode_flag_info { 113 u64 flag; 114 const char *name; 115}; 116 117static void 118pandecode_log_decoded_flags(const struct pandecode_flag_info *flag_info, 119 u64 flags) 120{ 121 bool decodable_flags_found = false; 122 123 for (int i = 0; flag_info[i].name; i++) { 124 if ((flags & flag_info[i].flag) != flag_info[i].flag) 125 continue; 126 127 if (!decodable_flags_found) { 128 decodable_flags_found = true; 129 } else { 130 pandecode_log_cont(" | "); 131 } 132 133 pandecode_log_cont("%s", flag_info[i].name); 134 135 flags &= ~flag_info[i].flag; 136 } 137 138 if (decodable_flags_found) { 139 if (flags) 140 pandecode_log_cont(" | 0x%" PRIx64, flags); 141 } else { 142 pandecode_log_cont("0x%" PRIx64, flags); 143 } 144} 145 146#define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag } 147static const struct pandecode_flag_info gl_enable_flag_info[] = { 148 FLAG_INFO(CULL_FACE_FRONT), 149 FLAG_INFO(CULL_FACE_BACK), 150 FLAG_INFO(OCCLUSION_QUERY), 151 FLAG_INFO(OCCLUSION_PRECISE), 152 {} 153}; 154#undef FLAG_INFO 155 156#define FLAG_INFO(flag) { MALI_CLEAR_##flag, "MALI_CLEAR_" #flag } 157static const struct pandecode_flag_info clear_flag_info[] = { 158 FLAG_INFO(FAST), 159 FLAG_INFO(SLOW), 160 FLAG_INFO(SLOW_STENCIL), 161 {} 162}; 163#undef FLAG_INFO 164 165#define FLAG_INFO(flag) { MALI_MASK_##flag, "MALI_MASK_" #flag } 166static const struct pandecode_flag_info mask_flag_info[] = { 167 FLAG_INFO(R), 168 FLAG_INFO(G), 169 FLAG_INFO(B), 170 FLAG_INFO(A), 171 {} 172}; 173#undef FLAG_INFO 174 175#define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag } 176static const struct pandecode_flag_info u3_flag_info[] = { 177 FLAG_INFO(HAS_MSAA), 178 FLAG_INFO(CAN_DISCARD), 179 FLAG_INFO(HAS_BLEND_SHADER), 180 FLAG_INFO(DEPTH_TEST), 181 {} 182}; 183 184static const struct pandecode_flag_info u4_flag_info[] = { 185 FLAG_INFO(NO_MSAA), 186 FLAG_INFO(NO_DITHER), 187 FLAG_INFO(DEPTH_RANGE_A), 188 FLAG_INFO(DEPTH_RANGE_B), 189 FLAG_INFO(STENCIL_TEST), 190 FLAG_INFO(SAMPLE_ALPHA_TO_COVERAGE_NO_BLEND_SHADER), 191 {} 192}; 193#undef FLAG_INFO 194 195#define FLAG_INFO(flag) { MALI_FRAMEBUFFER_##flag, "MALI_FRAMEBUFFER_" #flag } 196static const struct pandecode_flag_info fb_fmt_flag_info[] = { 197 FLAG_INFO(MSAA_A), 198 FLAG_INFO(MSAA_B), 199 FLAG_INFO(MSAA_8), 200 {} 201}; 202#undef FLAG_INFO 203 204#define FLAG_INFO(flag) { MALI_MFBD_FORMAT_##flag, "MALI_MFBD_FORMAT_" #flag } 205static const struct pandecode_flag_info mfbd_fmt_flag_info[] = { 206 FLAG_INFO(AFBC), 207 FLAG_INFO(MSAA), 208 {} 209}; 210#undef FLAG_INFO 211 212#define FLAG_INFO(flag) { MALI_EXTRA_##flag, "MALI_EXTRA_" #flag } 213static const struct pandecode_flag_info mfbd_extra_flag_info[] = { 214 FLAG_INFO(PRESENT), 215 FLAG_INFO(AFBC), 216 FLAG_INFO(ZS), 217 {} 218}; 219#undef FLAG_INFO 220 221#define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag } 222static const struct pandecode_flag_info shader_unknown1_flag_info [] = { 223 FLAG_INFO(NO_ALPHA_TO_COVERAGE), 224 FLAG_INFO(READS_TILEBUFFER), 225 FLAG_INFO(READS_ZS), 226 {} 227}; 228#undef FLAG_INFO 229 230extern char *replace_fragment; 231extern char *replace_vertex; 232 233static char * 234pandecode_job_type_name(enum mali_job_type type) 235{ 236#define DEFINE_CASE(name) case JOB_TYPE_ ## name: return "JOB_TYPE_" #name 237 238 switch (type) { 239 DEFINE_CASE(NULL); 240 DEFINE_CASE(SET_VALUE); 241 DEFINE_CASE(CACHE_FLUSH); 242 DEFINE_CASE(COMPUTE); 243 DEFINE_CASE(VERTEX); 244 DEFINE_CASE(TILER); 245 DEFINE_CASE(FUSED); 246 DEFINE_CASE(FRAGMENT); 247 248 case JOB_NOT_STARTED: 249 return "NOT_STARTED"; 250 251 default: 252 pandecode_log("Warning! Unknown job type %x\n", type); 253 return "!?!?!?"; 254 } 255 256#undef DEFINE_CASE 257} 258 259static char * 260pandecode_draw_mode_name(enum mali_draw_mode mode) 261{ 262#define DEFINE_CASE(name) case MALI_ ## name: return "MALI_" #name 263 264 switch (mode) { 265 DEFINE_CASE(DRAW_NONE); 266 DEFINE_CASE(POINTS); 267 DEFINE_CASE(LINES); 268 DEFINE_CASE(TRIANGLES); 269 DEFINE_CASE(TRIANGLE_STRIP); 270 DEFINE_CASE(TRIANGLE_FAN); 271 DEFINE_CASE(LINE_STRIP); 272 DEFINE_CASE(LINE_LOOP); 273 DEFINE_CASE(POLYGON); 274 DEFINE_CASE(QUADS); 275 DEFINE_CASE(QUAD_STRIP); 276 277 default: 278 return "MALI_TRIANGLES /* XXX: Unknown GL mode, check dump */"; 279 } 280 281#undef DEFINE_CASE 282} 283 284#define DEFINE_CASE(name) case MALI_FUNC_ ## name: return "MALI_FUNC_" #name 285static char * 286pandecode_func_name(enum mali_func mode) 287{ 288 switch (mode) { 289 DEFINE_CASE(NEVER); 290 DEFINE_CASE(LESS); 291 DEFINE_CASE(EQUAL); 292 DEFINE_CASE(LEQUAL); 293 DEFINE_CASE(GREATER); 294 DEFINE_CASE(NOTEQUAL); 295 DEFINE_CASE(GEQUAL); 296 DEFINE_CASE(ALWAYS); 297 298 default: 299 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */"; 300 } 301} 302#undef DEFINE_CASE 303 304/* Why is this duplicated? Who knows... */ 305#define DEFINE_CASE(name) case MALI_ALT_FUNC_ ## name: return "MALI_ALT_FUNC_" #name 306static char * 307pandecode_alt_func_name(enum mali_alt_func mode) 308{ 309 switch (mode) { 310 DEFINE_CASE(NEVER); 311 DEFINE_CASE(LESS); 312 DEFINE_CASE(EQUAL); 313 DEFINE_CASE(LEQUAL); 314 DEFINE_CASE(GREATER); 315 DEFINE_CASE(NOTEQUAL); 316 DEFINE_CASE(GEQUAL); 317 DEFINE_CASE(ALWAYS); 318 319 default: 320 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */"; 321 } 322} 323#undef DEFINE_CASE 324 325#define DEFINE_CASE(name) case MALI_STENCIL_ ## name: return "MALI_STENCIL_" #name 326static char * 327pandecode_stencil_op_name(enum mali_stencil_op op) 328{ 329 switch (op) { 330 DEFINE_CASE(KEEP); 331 DEFINE_CASE(REPLACE); 332 DEFINE_CASE(ZERO); 333 DEFINE_CASE(INVERT); 334 DEFINE_CASE(INCR_WRAP); 335 DEFINE_CASE(DECR_WRAP); 336 DEFINE_CASE(INCR); 337 DEFINE_CASE(DECR); 338 339 default: 340 return "MALI_STENCIL_KEEP /* XXX: Unknown stencil op, check dump */"; 341 } 342} 343 344#undef DEFINE_CASE 345 346#define DEFINE_CASE(name) case MALI_ATTR_ ## name: return "MALI_ATTR_" #name 347static char *pandecode_attr_mode_name(enum mali_attr_mode mode) 348{ 349 switch(mode) { 350 DEFINE_CASE(UNUSED); 351 DEFINE_CASE(LINEAR); 352 DEFINE_CASE(POT_DIVIDE); 353 DEFINE_CASE(MODULO); 354 DEFINE_CASE(NPOT_DIVIDE); 355 default: return "MALI_ATTR_UNUSED /* XXX: Unknown stencil op, check dump */"; 356 } 357} 358 359#undef DEFINE_CASE 360 361#define DEFINE_CASE(name) case MALI_CHANNEL_## name: return "MALI_CHANNEL_" #name 362static char * 363pandecode_channel_name(enum mali_channel channel) 364{ 365 switch (channel) { 366 DEFINE_CASE(RED); 367 DEFINE_CASE(GREEN); 368 DEFINE_CASE(BLUE); 369 DEFINE_CASE(ALPHA); 370 DEFINE_CASE(ZERO); 371 DEFINE_CASE(ONE); 372 DEFINE_CASE(RESERVED_0); 373 DEFINE_CASE(RESERVED_1); 374 375 default: 376 return "MALI_CHANNEL_ZERO /* XXX: Unknown channel, check dump */"; 377 } 378} 379#undef DEFINE_CASE 380 381#define DEFINE_CASE(name) case MALI_WRAP_## name: return "MALI_WRAP_" #name 382static char * 383pandecode_wrap_mode_name(enum mali_wrap_mode op) 384{ 385 switch (op) { 386 DEFINE_CASE(REPEAT); 387 DEFINE_CASE(CLAMP_TO_EDGE); 388 DEFINE_CASE(CLAMP_TO_BORDER); 389 DEFINE_CASE(MIRRORED_REPEAT); 390 391 default: 392 return "MALI_WRAP_REPEAT /* XXX: Unknown wrap mode, check dump */"; 393 } 394} 395#undef DEFINE_CASE 396 397static inline char * 398pandecode_decode_fbd_type(enum mali_fbd_type type) 399{ 400 if (type == MALI_SFBD) return "SFBD"; 401 else if (type == MALI_MFBD) return "MFBD"; 402 else return "WATFBD /* XXX */"; 403} 404 405static void 406pandecode_replay_sfbd(uint64_t gpu_va, int job_no) 407{ 408 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va); 409 const struct mali_single_framebuffer *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va); 410 411 pandecode_log("struct mali_single_framebuffer framebuffer_%d = {\n", job_no); 412 pandecode_indent++; 413 414 pandecode_prop("unknown1 = 0x%" PRIx32, s->unknown1); 415 pandecode_prop("unknown2 = 0x%" PRIx32, s->unknown2); 416 417 pandecode_log(".format = "); 418 pandecode_log_decoded_flags(fb_fmt_flag_info, s->format); 419 pandecode_log_cont(",\n"); 420 421 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", s->width + 1); 422 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", s->height + 1); 423 424 MEMORY_PROP(s, framebuffer); 425 pandecode_prop("stride = %d", s->stride); 426 427 /* Earlier in the actual commandstream -- right before width -- but we 428 * delay to flow nicer */ 429 430 pandecode_log(".clear_flags = "); 431 pandecode_log_decoded_flags(clear_flag_info, s->clear_flags); 432 pandecode_log_cont(",\n"); 433 434 if (s->depth_buffer | s->depth_buffer_enable) { 435 MEMORY_PROP(s, depth_buffer); 436 pandecode_prop("depth_buffer_enable = %s", DS_ENABLE(s->depth_buffer_enable)); 437 } 438 439 if (s->stencil_buffer | s->stencil_buffer_enable) { 440 MEMORY_PROP(s, stencil_buffer); 441 pandecode_prop("stencil_buffer_enable = %s", DS_ENABLE(s->stencil_buffer_enable)); 442 } 443 444 if (s->clear_color_1 | s->clear_color_2 | s->clear_color_3 | s->clear_color_4) { 445 pandecode_prop("clear_color_1 = 0x%" PRIx32, s->clear_color_1); 446 pandecode_prop("clear_color_2 = 0x%" PRIx32, s->clear_color_2); 447 pandecode_prop("clear_color_3 = 0x%" PRIx32, s->clear_color_3); 448 pandecode_prop("clear_color_4 = 0x%" PRIx32, s->clear_color_4); 449 } 450 451 if (s->clear_depth_1 != 0 || s->clear_depth_2 != 0 || s->clear_depth_3 != 0 || s->clear_depth_4 != 0) { 452 pandecode_prop("clear_depth_1 = %f", s->clear_depth_1); 453 pandecode_prop("clear_depth_2 = %f", s->clear_depth_2); 454 pandecode_prop("clear_depth_3 = %f", s->clear_depth_3); 455 pandecode_prop("clear_depth_4 = %f", s->clear_depth_4); 456 } 457 458 if (s->clear_stencil) { 459 pandecode_prop("clear_stencil = 0x%x", s->clear_stencil); 460 } 461 462 MEMORY_PROP(s, unknown_address_0); 463 MEMORY_PROP(s, unknown_address_1); 464 MEMORY_PROP(s, unknown_address_2); 465 466 pandecode_prop("resolution_check = 0x%" PRIx32, s->resolution_check); 467 pandecode_prop("tiler_flags = 0x%" PRIx32, s->tiler_flags); 468 469 MEMORY_PROP(s, tiler_heap_free); 470 MEMORY_PROP(s, tiler_heap_end); 471 472 pandecode_indent--; 473 pandecode_log("};\n"); 474 475 pandecode_prop("zero0 = 0x%" PRIx64, s->zero0); 476 pandecode_prop("zero1 = 0x%" PRIx64, s->zero1); 477 pandecode_prop("zero2 = 0x%" PRIx32, s->zero2); 478 pandecode_prop("zero4 = 0x%" PRIx32, s->zero4); 479 480 printf(".zero3 = {"); 481 482 for (int i = 0; i < sizeof(s->zero3) / sizeof(s->zero3[0]); ++i) 483 printf("%X, ", s->zero3[i]); 484 485 printf("},\n"); 486 487 printf(".zero6 = {"); 488 489 for (int i = 0; i < sizeof(s->zero6) / sizeof(s->zero6[0]); ++i) 490 printf("%X, ", s->zero6[i]); 491 492 printf("},\n"); 493} 494 495static void 496pandecode_replay_swizzle(unsigned swizzle) 497{ 498 pandecode_prop("swizzle = %s | (%s << 3) | (%s << 6) | (%s << 9)", 499 pandecode_channel_name((swizzle >> 0) & 0x7), 500 pandecode_channel_name((swizzle >> 3) & 0x7), 501 pandecode_channel_name((swizzle >> 6) & 0x7), 502 pandecode_channel_name((swizzle >> 9) & 0x7)); 503} 504 505static void 506pandecode_rt_format(struct mali_rt_format format) 507{ 508 pandecode_log(".format = {\n"); 509 pandecode_indent++; 510 511 pandecode_prop("unk1 = 0x%" PRIx32, format.unk1); 512 pandecode_prop("unk2 = 0x%" PRIx32, format.unk2); 513 514 pandecode_prop("nr_channels = MALI_POSITIVE(%d)", 515 MALI_NEGATIVE(format.nr_channels)); 516 517 pandecode_log(".flags = "); 518 pandecode_log_decoded_flags(mfbd_fmt_flag_info, format.flags); 519 pandecode_log_cont(",\n"); 520 521 pandecode_replay_swizzle(format.swizzle); 522 523 pandecode_prop("unk4 = 0x%" PRIx32, format.unk4); 524 525 pandecode_indent--; 526 pandecode_log("},\n"); 527} 528 529static void 530pandecode_replay_mfbd_bfr(uint64_t gpu_va, int job_no) 531{ 532 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va); 533 const struct bifrost_framebuffer *PANDECODE_PTR_VAR(fb, mem, (mali_ptr) gpu_va); 534 535 if (fb->sample_locations) { 536 /* The blob stores all possible sample locations in a single buffer 537 * allocated on startup, and just switches the pointer when switching 538 * MSAA state. For now, we just put the data into the cmdstream, but we 539 * should do something like what the blob does with a real driver. 540 * 541 * There seem to be 32 slots for sample locations, followed by another 542 * 16. The second 16 is just the center location followed by 15 zeros 543 * in all the cases I've identified (maybe shader vs. depth/color 544 * samples?). 545 */ 546 547 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(fb->sample_locations); 548 549 const u16 *PANDECODE_PTR_VAR(samples, smem, fb->sample_locations); 550 551 pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no); 552 pandecode_indent++; 553 554 for (int i = 0; i < 32 + 16; i++) { 555 pandecode_log("%d, %d,\n", samples[2 * i], samples[2 * i + 1]); 556 } 557 558 pandecode_indent--; 559 pandecode_log("};\n"); 560 } 561 562 pandecode_log("struct bifrost_framebuffer framebuffer_%d = {\n", job_no); 563 pandecode_indent++; 564 565 pandecode_prop("unk0 = 0x%x", fb->unk0); 566 567 if (fb->sample_locations) 568 pandecode_prop("sample_locations = sample_locations_%d", job_no); 569 570 /* Assume that unknown1 and tiler_meta were emitted in the last job for 571 * now */ 572 /*pandecode_prop("unknown1 = unknown1_%d_p", job_no - 1); 573 pandecode_prop("tiler_meta = tiler_meta_%d_p", job_no - 1);*/ 574 MEMORY_PROP(fb, unknown1); 575 MEMORY_PROP(fb, tiler_meta); 576 577 pandecode_prop("width1 = MALI_POSITIVE(%d)", fb->width1 + 1); 578 pandecode_prop("height1 = MALI_POSITIVE(%d)", fb->height1 + 1); 579 pandecode_prop("width2 = MALI_POSITIVE(%d)", fb->width2 + 1); 580 pandecode_prop("height2 = MALI_POSITIVE(%d)", fb->height2 + 1); 581 582 pandecode_prop("unk1 = 0x%x", fb->unk1); 583 pandecode_prop("unk2 = 0x%x", fb->unk2); 584 pandecode_prop("rt_count_1 = MALI_POSITIVE(%d)", fb->rt_count_1 + 1); 585 pandecode_prop("rt_count_2 = %d", fb->rt_count_2); 586 587 pandecode_prop("unk3 = 0x%x", fb->unk3); 588 pandecode_prop("clear_stencil = 0x%x", fb->clear_stencil); 589 pandecode_prop("clear_depth = %f", fb->clear_depth); 590 591 pandecode_prop("unknown2 = 0x%x", fb->unknown2); 592 MEMORY_PROP(fb, scratchpad); 593 MEMORY_PROP(fb, tiler_scratch_start); 594 MEMORY_PROP(fb, tiler_scratch_middle); 595 MEMORY_PROP(fb, tiler_heap_start); 596 MEMORY_PROP(fb, tiler_heap_end); 597 598 if (fb->zero3 || fb->zero4 || fb->zero9 || fb->zero10 || fb->zero11 || fb->zero12) { 599 pandecode_msg("framebuffer zeros tripped\n"); 600 pandecode_prop("zero3 = 0x%" PRIx32, fb->zero3); 601 pandecode_prop("zero4 = 0x%" PRIx32, fb->zero4); 602 pandecode_prop("zero9 = 0x%" PRIx64, fb->zero9); 603 pandecode_prop("zero10 = 0x%" PRIx64, fb->zero10); 604 pandecode_prop("zero11 = 0x%" PRIx64, fb->zero11); 605 pandecode_prop("zero12 = 0x%" PRIx64, fb->zero12); 606 } 607 608 pandecode_indent--; 609 pandecode_log("};\n"); 610 611 gpu_va += sizeof(struct bifrost_framebuffer); 612 613 if (fb->unk3 & MALI_MFBD_EXTRA) { 614 mem = pandecode_find_mapped_gpu_mem_containing(gpu_va); 615 const struct bifrost_fb_extra *PANDECODE_PTR_VAR(fbx, mem, (mali_ptr) gpu_va); 616 617 pandecode_log("struct bifrost_fb_extra fb_extra_%d = {\n", job_no); 618 pandecode_indent++; 619 620 MEMORY_PROP(fbx, checksum); 621 622 if (fbx->checksum_stride) 623 pandecode_prop("checksum_stride = %d", fbx->checksum_stride); 624 625 pandecode_log(".flags = "); 626 pandecode_log_decoded_flags(mfbd_extra_flag_info, fbx->flags); 627 pandecode_log_cont(",\n"); 628 629 if (fbx->flags & MALI_EXTRA_AFBC_ZS) { 630 pandecode_log(".ds_afbc = {\n"); 631 pandecode_indent++; 632 633 MEMORY_PROP((&fbx->ds_afbc), depth_stencil_afbc_metadata); 634 pandecode_prop("depth_stencil_afbc_stride = %d", 635 fbx->ds_afbc.depth_stencil_afbc_stride); 636 MEMORY_PROP((&fbx->ds_afbc), depth_stencil); 637 638 if (fbx->ds_afbc.zero1 || fbx->ds_afbc.padding) { 639 pandecode_msg("Depth/stencil AFBC zeros tripped\n"); 640 pandecode_prop("zero1 = 0x%" PRIx32, 641 fbx->ds_afbc.zero1); 642 pandecode_prop("padding = 0x%" PRIx64, 643 fbx->ds_afbc.padding); 644 } 645 646 pandecode_indent--; 647 pandecode_log("},\n"); 648 } else { 649 pandecode_log(".ds_linear = {\n"); 650 pandecode_indent++; 651 652 if (fbx->ds_linear.depth) { 653 MEMORY_PROP((&fbx->ds_linear), depth); 654 pandecode_prop("depth_stride = %d", 655 fbx->ds_linear.depth_stride); 656 } 657 658 if (fbx->ds_linear.stencil) { 659 MEMORY_PROP((&fbx->ds_linear), stencil); 660 pandecode_prop("stencil_stride = %d", 661 fbx->ds_linear.stencil_stride); 662 } 663 664 if (fbx->ds_linear.depth_stride_zero || 665 fbx->ds_linear.stencil_stride_zero || 666 fbx->ds_linear.zero1 || fbx->ds_linear.zero2) { 667 pandecode_msg("Depth/stencil zeros tripped\n"); 668 pandecode_prop("depth_stride_zero = 0x%x", 669 fbx->ds_linear.depth_stride_zero); 670 pandecode_prop("stencil_stride_zero = 0x%x", 671 fbx->ds_linear.stencil_stride_zero); 672 pandecode_prop("zero1 = 0x%" PRIx32, 673 fbx->ds_linear.zero1); 674 pandecode_prop("zero2 = 0x%" PRIx32, 675 fbx->ds_linear.zero2); 676 } 677 678 pandecode_indent--; 679 pandecode_log("},\n"); 680 } 681 682 if (fbx->zero3 || fbx->zero4) { 683 pandecode_msg("fb_extra zeros tripped\n"); 684 pandecode_prop("zero3 = 0x%" PRIx64, fbx->zero3); 685 pandecode_prop("zero4 = 0x%" PRIx64, fbx->zero4); 686 } 687 688 pandecode_indent--; 689 pandecode_log("};\n"); 690 691 gpu_va += sizeof(struct bifrost_fb_extra); 692 } 693 694 pandecode_log("struct bifrost_render_target rts_list_%d[] = {\n", job_no); 695 pandecode_indent++; 696 697 for (int i = 0; i < MALI_NEGATIVE(fb->rt_count_1); i++) { 698 mali_ptr rt_va = gpu_va + i * sizeof(struct bifrost_render_target); 699 mem = pandecode_find_mapped_gpu_mem_containing(rt_va); 700 const struct bifrost_render_target *PANDECODE_PTR_VAR(rt, mem, (mali_ptr) rt_va); 701 702 pandecode_log("{\n"); 703 pandecode_indent++; 704 705 pandecode_rt_format(rt->format); 706 707 /* TODO: How the actual heck does AFBC enabling work here? */ 708 if (0) { 709 pandecode_log(".afbc = {\n"); 710 pandecode_indent++; 711 712 char *a = pointer_as_memory_reference(rt->afbc.metadata); 713 pandecode_prop("metadata = %s", a); 714 free(a); 715 716 pandecode_prop("stride = %d", rt->afbc.stride); 717 pandecode_prop("unk = 0x%" PRIx32, rt->afbc.unk); 718 719 pandecode_indent--; 720 pandecode_log("},\n"); 721 } else { 722 pandecode_log(".chunknown = {\n"); 723 pandecode_indent++; 724 725 pandecode_prop("unk = 0x%" PRIx64, rt->chunknown.unk); 726 727 char *a = pointer_as_memory_reference(rt->chunknown.pointer); 728 pandecode_prop("pointer = %s", a); 729 free(a); 730 731 pandecode_indent--; 732 pandecode_log("},\n"); 733 } 734 735 MEMORY_PROP(rt, framebuffer); 736 pandecode_prop("framebuffer_stride = %d", rt->framebuffer_stride); 737 738 if (rt->clear_color_1 | rt->clear_color_2 | rt->clear_color_3 | rt->clear_color_4) { 739 pandecode_prop("clear_color_1 = 0x%" PRIx32, rt->clear_color_1); 740 pandecode_prop("clear_color_2 = 0x%" PRIx32, rt->clear_color_2); 741 pandecode_prop("clear_color_3 = 0x%" PRIx32, rt->clear_color_3); 742 pandecode_prop("clear_color_4 = 0x%" PRIx32, rt->clear_color_4); 743 } 744 745 if (rt->zero1 || rt->zero2 || rt->zero3) { 746 pandecode_msg("render target zeros tripped\n"); 747 pandecode_prop("zero1 = 0x%" PRIx64, rt->zero1); 748 pandecode_prop("zero2 = 0x%" PRIx32, rt->zero2); 749 pandecode_prop("zero3 = 0x%" PRIx32, rt->zero3); 750 } 751 752 pandecode_indent--; 753 pandecode_log("},\n"); 754 } 755 756 pandecode_indent--; 757 pandecode_log("};\n"); 758} 759 760static void 761pandecode_replay_attributes(const struct pandecode_mapped_memory *mem, 762 mali_ptr addr, int job_no, char *suffix, 763 int count, bool varying) 764{ 765 char *prefix = varying ? "varyings" : "attributes"; 766 767 union mali_attr *attr = pandecode_fetch_gpu_mem(mem, addr, sizeof(union mali_attr) * count); 768 769 char base[128]; 770 snprintf(base, sizeof(base), "%s_data_%d%s", prefix, job_no, suffix); 771 772 for (int i = 0; i < count; ++i) { 773 enum mali_attr_mode mode = attr[i].elements & 7; 774 775 if (mode == MALI_ATTR_UNUSED) 776 continue; 777 778 mali_ptr raw_elements = attr[i].elements & ~7; 779 780 /* TODO: Do we maybe want to dump the attribute values 781 * themselves given the specified format? Or is that too hard? 782 * */ 783 784 char *a = pointer_as_memory_reference(raw_elements); 785 pandecode_log("mali_ptr %s_%d_p = %s;\n", base, i, a); 786 free(a); 787 } 788 789 pandecode_log("union mali_attr %s_%d[] = {\n", prefix, job_no); 790 pandecode_indent++; 791 792 for (int i = 0; i < count; ++i) { 793 pandecode_log("{\n"); 794 pandecode_indent++; 795 796 pandecode_prop("elements = (%s_%d_p) | %s", base, i, pandecode_attr_mode_name(attr[i].elements & 7)); 797 pandecode_prop("shift = %d", attr[i].shift); 798 pandecode_prop("extra_flags = %d", attr[i].extra_flags); 799 pandecode_prop("stride = 0x%" PRIx32, attr[i].stride); 800 pandecode_prop("size = 0x%" PRIx32, attr[i].size); 801 pandecode_indent--; 802 pandecode_log("}, \n"); 803 804 if ((attr[i].elements & 7) == MALI_ATTR_NPOT_DIVIDE) { 805 i++; 806 pandecode_log("{\n"); 807 pandecode_indent++; 808 pandecode_prop("unk = 0x%x", attr[i].unk); 809 pandecode_prop("magic_divisor = 0x%08x", attr[i].magic_divisor); 810 if (attr[i].zero != 0) 811 pandecode_prop("zero = 0x%x /* XXX zero tripped */", attr[i].zero); 812 pandecode_prop("divisor = %d", attr[i].divisor); 813 pandecode_indent--; 814 pandecode_log("}, \n"); 815 } 816 817 } 818 819 pandecode_indent--; 820 pandecode_log("};\n"); 821} 822 823static mali_ptr 824pandecode_replay_shader_address(const char *name, mali_ptr ptr) 825{ 826 /* TODO: Decode flags */ 827 mali_ptr shader_ptr = ptr & ~15; 828 829 char *a = pointer_as_memory_reference(shader_ptr); 830 pandecode_prop("%s = (%s) | %d", name, a, (int) (ptr & 15)); 831 free(a); 832 833 return shader_ptr; 834} 835 836static void 837pandecode_replay_stencil(const char *name, const struct mali_stencil_test *stencil) 838{ 839 const char *func = pandecode_func_name(stencil->func); 840 const char *sfail = pandecode_stencil_op_name(stencil->sfail); 841 const char *dpfail = pandecode_stencil_op_name(stencil->dpfail); 842 const char *dppass = pandecode_stencil_op_name(stencil->dppass); 843 844 if (stencil->zero) 845 pandecode_msg("Stencil zero tripped: %X\n", stencil->zero); 846 847 pandecode_log(".stencil_%s = {\n", name); 848 pandecode_indent++; 849 pandecode_prop("ref = %d", stencil->ref); 850 pandecode_prop("mask = 0x%02X", stencil->mask); 851 pandecode_prop("func = %s", func); 852 pandecode_prop("sfail = %s", sfail); 853 pandecode_prop("dpfail = %s", dpfail); 854 pandecode_prop("dppass = %s", dppass); 855 pandecode_indent--; 856 pandecode_log("},\n"); 857} 858 859static void 860pandecode_replay_blend_equation(const struct mali_blend_equation *blend, const char *suffix) 861{ 862 if (blend->zero1) 863 pandecode_msg("Blend zero tripped: %X\n", blend->zero1); 864 865 pandecode_log(".blend_equation%s = {\n", suffix); 866 pandecode_indent++; 867 868 pandecode_prop("rgb_mode = 0x%X", blend->rgb_mode); 869 pandecode_prop("alpha_mode = 0x%X", blend->alpha_mode); 870 871 pandecode_log(".color_mask = "); 872 pandecode_log_decoded_flags(mask_flag_info, blend->color_mask); 873 pandecode_log_cont(",\n"); 874 875 pandecode_indent--; 876 pandecode_log("},\n"); 877} 878 879static int 880pandecode_replay_attribute_meta(int job_no, int count, const struct mali_vertex_tiler_postfix *v, bool varying, char *suffix) 881{ 882 char base[128]; 883 char *prefix = varying ? "varying" : "attribute"; 884 unsigned max_index = 0; 885 snprintf(base, sizeof(base), "%s_meta", prefix); 886 887 pandecode_log("struct mali_attr_meta %s_%d%s[] = {\n", base, job_no, suffix); 888 pandecode_indent++; 889 890 struct mali_attr_meta *attr_meta; 891 mali_ptr p = varying ? (v->varying_meta & ~0xF) : v->attribute_meta; 892 893 struct pandecode_mapped_memory *attr_mem = pandecode_find_mapped_gpu_mem_containing(p); 894 895 for (int i = 0; i < count; ++i, p += sizeof(struct mali_attr_meta)) { 896 attr_meta = pandecode_fetch_gpu_mem(attr_mem, p, 897 sizeof(*attr_mem)); 898 899 pandecode_log("{\n"); 900 pandecode_indent++; 901 pandecode_prop("index = %d", attr_meta->index); 902 903 if (attr_meta->index > max_index) 904 max_index = attr_meta->index; 905 pandecode_replay_swizzle(attr_meta->swizzle); 906 pandecode_prop("format = %s", pandecode_format_name(attr_meta->format)); 907 908 pandecode_prop("unknown1 = 0x%" PRIx64, (u64) attr_meta->unknown1); 909 pandecode_prop("unknown3 = 0x%" PRIx64, (u64) attr_meta->unknown3); 910 pandecode_prop("src_offset = 0x%" PRIx64, (u64) attr_meta->src_offset); 911 pandecode_indent--; 912 pandecode_log("},\n"); 913 914 } 915 916 pandecode_indent--; 917 pandecode_log("};\n"); 918 919 return max_index; 920} 921 922static void 923pandecode_replay_indices(uintptr_t pindices, uint32_t index_count, int job_no) 924{ 925 struct pandecode_mapped_memory *imem = pandecode_find_mapped_gpu_mem_containing(pindices); 926 927 if (imem) { 928 /* Indices are literally just a u32 array :) */ 929 930 uint32_t *PANDECODE_PTR_VAR(indices, imem, pindices); 931 932 pandecode_log("uint32_t indices_%d[] = {\n", job_no); 933 pandecode_indent++; 934 935 for (unsigned i = 0; i < (index_count + 1); i += 3) 936 pandecode_log("%d, %d, %d,\n", 937 indices[i], 938 indices[i + 1], 939 indices[i + 2]); 940 941 pandecode_indent--; 942 pandecode_log("};\n"); 943 } 944} 945 946/* return bits [lo, hi) of word */ 947static u32 948bits(u32 word, u32 lo, u32 hi) 949{ 950 if (hi - lo >= 32) 951 return word; // avoid undefined behavior with the shift 952 953 return (word >> lo) & ((1 << (hi - lo)) - 1); 954} 955 956static void 957pandecode_replay_vertex_tiler_prefix(struct mali_vertex_tiler_prefix *p, int job_no) 958{ 959 pandecode_log_cont("{\n"); 960 pandecode_indent++; 961 962 pandecode_prop("invocation_count = %" PRIx32, p->invocation_count); 963 pandecode_prop("size_y_shift = %d", p->size_y_shift); 964 pandecode_prop("size_z_shift = %d", p->size_z_shift); 965 pandecode_prop("workgroups_x_shift = %d", p->workgroups_x_shift); 966 pandecode_prop("workgroups_y_shift = %d", p->workgroups_y_shift); 967 pandecode_prop("workgroups_z_shift = %d", p->workgroups_z_shift); 968 pandecode_prop("workgroups_x_shift_2 = 0x%" PRIx32, p->workgroups_x_shift_2); 969 970 /* Decode invocation_count. See the comment before the definition of 971 * invocation_count for an explanation. 972 */ 973 pandecode_msg("size: (%d, %d, %d)\n", 974 bits(p->invocation_count, 0, p->size_y_shift) + 1, 975 bits(p->invocation_count, p->size_y_shift, p->size_z_shift) + 1, 976 bits(p->invocation_count, p->size_z_shift, 977 p->workgroups_x_shift) + 1); 978 pandecode_msg("workgroups: (%d, %d, %d)\n", 979 bits(p->invocation_count, p->workgroups_x_shift, 980 p->workgroups_y_shift) + 1, 981 bits(p->invocation_count, p->workgroups_y_shift, 982 p->workgroups_z_shift) + 1, 983 bits(p->invocation_count, p->workgroups_z_shift, 984 32) + 1); 985 986 /* TODO: Decode */ 987 pandecode_prop("unknown_draw = 0x%" PRIx32, p->unknown_draw); 988 pandecode_prop("workgroups_x_shift_3 = 0x%" PRIx32, p->workgroups_x_shift_3); 989 990 pandecode_prop("draw_mode = %s", pandecode_draw_mode_name(p->draw_mode)); 991 992 /* Index count only exists for tiler jobs anyway */ 993 994 if (p->index_count) 995 pandecode_prop("index_count = MALI_POSITIVE(%" PRId32 ")", p->index_count + 1); 996 997 pandecode_prop("negative_start = %d", p->negative_start); 998 999 DYN_MEMORY_PROP(p, job_no, indices); 1000 1001 if (p->zero1) { 1002 pandecode_msg("Zero tripped\n"); 1003 pandecode_prop("zero1 = 0x%" PRIx32, p->zero1); 1004 } 1005 1006 pandecode_indent--; 1007 pandecode_log("},\n"); 1008} 1009 1010static void 1011pandecode_replay_uniform_buffers(mali_ptr pubufs, int ubufs_count, int job_no) 1012{ 1013 struct pandecode_mapped_memory *umem = pandecode_find_mapped_gpu_mem_containing(pubufs); 1014 1015 struct mali_uniform_buffer_meta *PANDECODE_PTR_VAR(ubufs, umem, pubufs); 1016 1017 for (int i = 0; i < ubufs_count; i++) { 1018 mali_ptr ptr = ubufs[i].ptr << 2; 1019 struct pandecode_mapped_memory *umem2 = pandecode_find_mapped_gpu_mem_containing(ptr); 1020 uint32_t *PANDECODE_PTR_VAR(ubuf, umem2, ptr); 1021 char name[50]; 1022 snprintf(name, sizeof(name), "ubuf_%d", i); 1023 /* The blob uses ubuf 0 to upload internal stuff and 1024 * uniforms that won't fit/are accessed indirectly, so 1025 * it puts it in the batchbuffer. 1026 */ 1027 pandecode_log("uint32_t %s_%d[] = {\n", name, job_no); 1028 pandecode_indent++; 1029 1030 for (int j = 0; j <= ubufs[i].size; j++) { 1031 for (int k = 0; k < 4; k++) { 1032 if (k == 0) 1033 pandecode_log("0x%"PRIx32", ", ubuf[4 * j + k]); 1034 else 1035 pandecode_log_cont("0x%"PRIx32", ", ubuf[4 * j + k]); 1036 1037 } 1038 1039 pandecode_log_cont("\n"); 1040 } 1041 1042 pandecode_indent--; 1043 pandecode_log("};\n"); 1044 } 1045 1046 pandecode_log("struct mali_uniform_buffer_meta uniform_buffers_%d[] = {\n", 1047 job_no); 1048 pandecode_indent++; 1049 1050 for (int i = 0; i < ubufs_count; i++) { 1051 pandecode_log("{\n"); 1052 pandecode_indent++; 1053 pandecode_prop("size = MALI_POSITIVE(%d)", ubufs[i].size + 1); 1054 pandecode_prop("ptr = ubuf_%d_%d_p >> 2", i, job_no); 1055 pandecode_indent--; 1056 pandecode_log("},\n"); 1057 } 1058 1059 pandecode_indent--; 1060 pandecode_log("};\n"); 1061} 1062 1063static void 1064pandecode_replay_scratchpad(uintptr_t pscratchpad, int job_no, char *suffix) 1065{ 1066 1067 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(pscratchpad); 1068 1069 struct bifrost_scratchpad *PANDECODE_PTR_VAR(scratchpad, mem, pscratchpad); 1070 1071 if (scratchpad->zero) 1072 pandecode_msg("XXX scratchpad zero tripped"); 1073 1074 pandecode_log("struct bifrost_scratchpad scratchpad_%d%s = {\n", job_no, suffix); 1075 pandecode_indent++; 1076 1077 pandecode_prop("flags = 0x%x", scratchpad->flags); 1078 MEMORY_PROP(scratchpad, gpu_scratchpad); 1079 1080 pandecode_indent--; 1081 pandecode_log("};\n"); 1082} 1083 1084static void 1085pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type, 1086 bool is_bifrost) 1087{ 1088 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(shader_ptr); 1089 uint8_t *PANDECODE_PTR_VAR(code, mem, shader_ptr); 1090 1091 /* Compute maximum possible size */ 1092 size_t sz = mem->length - (shader_ptr - mem->gpu_va); 1093 1094 /* TODO: When Bifrost is upstreamed, disassemble that too */ 1095 if (is_bifrost) { 1096 pandecode_msg("Bifrost disassembler not yet upstreamed"); 1097 return; 1098 } 1099 1100 /* Print some boilerplate to clearly denote the assembly (which doesn't 1101 * obey indentation rules), and actually do the disassembly! */ 1102 1103 printf("\n\n"); 1104 disassemble_midgard(code, sz); 1105 printf("\n\n"); 1106} 1107 1108static void 1109pandecode_replay_vertex_tiler_postfix_pre(const struct mali_vertex_tiler_postfix *p, 1110 int job_no, enum mali_job_type job_type, 1111 char *suffix, bool is_bifrost) 1112{ 1113 mali_ptr shader_meta_ptr = (u64) (uintptr_t) (p->_shader_upper << 4); 1114 struct pandecode_mapped_memory *attr_mem; 1115 1116 /* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad 1117 * are the only things actually needed from the FBD, vertex/tiler jobs 1118 * no longer reference the FBD -- instead, this field points to some 1119 * info about the scratchpad. 1120 */ 1121 if (is_bifrost) 1122 pandecode_replay_scratchpad(p->framebuffer & ~FBD_TYPE, job_no, suffix); 1123 else if (p->framebuffer & MALI_MFBD) 1124 pandecode_replay_mfbd_bfr((u64) ((uintptr_t) p->framebuffer) & FBD_MASK, job_no); 1125 else 1126 pandecode_replay_sfbd((u64) (uintptr_t) p->framebuffer, job_no); 1127 1128 int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0; 1129 int texture_count = 0, sampler_count = 0; 1130 1131 if (shader_meta_ptr) { 1132 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(shader_meta_ptr); 1133 struct mali_shader_meta *PANDECODE_PTR_VAR(s, smem, shader_meta_ptr); 1134 1135 pandecode_log("struct mali_shader_meta shader_meta_%d%s = {\n", job_no, suffix); 1136 pandecode_indent++; 1137 1138 /* Save for dumps */ 1139 attribute_count = s->attribute_count; 1140 varying_count = s->varying_count; 1141 texture_count = s->texture_count; 1142 sampler_count = s->sampler_count; 1143 1144 if (is_bifrost) { 1145 uniform_count = s->bifrost2.uniform_count; 1146 uniform_buffer_count = s->bifrost1.uniform_buffer_count; 1147 } else { 1148 uniform_count = s->midgard1.uniform_count; 1149 /* TODO figure this out */ 1150 uniform_buffer_count = 1; 1151 } 1152 1153 mali_ptr shader_ptr = pandecode_replay_shader_address("shader", s->shader); 1154 1155 pandecode_prop("texture_count = %" PRId16, s->texture_count); 1156 pandecode_prop("sampler_count = %" PRId16, s->sampler_count); 1157 pandecode_prop("attribute_count = %" PRId16, s->attribute_count); 1158 pandecode_prop("varying_count = %" PRId16, s->varying_count); 1159 1160 if (is_bifrost) { 1161 pandecode_log(".bifrost1 = {\n"); 1162 pandecode_indent++; 1163 1164 pandecode_prop("uniform_buffer_count = %" PRId32, s->bifrost1.uniform_buffer_count); 1165 pandecode_prop("unk1 = 0x%" PRIx32, s->bifrost1.unk1); 1166 1167 pandecode_indent--; 1168 pandecode_log("},\n"); 1169 } else { 1170 pandecode_log(".midgard1 = {\n"); 1171 pandecode_indent++; 1172 1173 pandecode_prop("uniform_count = %" PRId16, s->midgard1.uniform_count); 1174 pandecode_prop("work_count = %" PRId16, s->midgard1.work_count); 1175 1176 pandecode_log(".unknown1 = "); 1177 pandecode_log_decoded_flags(shader_unknown1_flag_info, s->midgard1.unknown1); 1178 pandecode_log_cont(",\n"); 1179 1180 pandecode_prop("unknown2 = 0x%" PRIx32, s->midgard1.unknown2); 1181 1182 pandecode_indent--; 1183 pandecode_log("},\n"); 1184 } 1185 1186 if (s->depth_units || s->depth_factor) { 1187 if (is_bifrost) 1188 pandecode_prop("depth_units = %f", s->depth_units); 1189 else 1190 pandecode_prop("depth_units = MALI_NEGATIVE(%f)", s->depth_units - 1.0f); 1191 1192 pandecode_prop("depth_factor = %f", s->depth_factor); 1193 } 1194 1195 bool invert_alpha_coverage = s->alpha_coverage & 0xFFF0; 1196 uint16_t inverted_coverage = invert_alpha_coverage ? ~s->alpha_coverage : s->alpha_coverage; 1197 1198 pandecode_prop("alpha_coverage = %sMALI_ALPHA_COVERAGE(%f)", 1199 invert_alpha_coverage ? "~" : "", 1200 MALI_GET_ALPHA_COVERAGE(inverted_coverage)); 1201 1202 pandecode_log(".unknown2_3 = "); 1203 1204 int unknown2_3 = s->unknown2_3; 1205 int unknown2_4 = s->unknown2_4; 1206 1207 /* We're not quite sure what these flags mean without the depth test, if anything */ 1208 1209 if (unknown2_3 & (MALI_DEPTH_TEST | MALI_DEPTH_FUNC_MASK)) { 1210 const char *func = pandecode_func_name(MALI_GET_DEPTH_FUNC(unknown2_3)); 1211 unknown2_3 &= ~MALI_DEPTH_FUNC_MASK; 1212 1213 pandecode_log_cont("MALI_DEPTH_FUNC(%s) | ", func); 1214 } 1215 1216 pandecode_log_decoded_flags(u3_flag_info, unknown2_3); 1217 pandecode_log_cont(",\n"); 1218 1219 pandecode_prop("stencil_mask_front = 0x%02X", s->stencil_mask_front); 1220 pandecode_prop("stencil_mask_back = 0x%02X", s->stencil_mask_back); 1221 1222 pandecode_log(".unknown2_4 = "); 1223 pandecode_log_decoded_flags(u4_flag_info, unknown2_4); 1224 pandecode_log_cont(",\n"); 1225 1226 pandecode_replay_stencil("front", &s->stencil_front); 1227 pandecode_replay_stencil("back", &s->stencil_back); 1228 1229 if (is_bifrost) { 1230 pandecode_log(".bifrost2 = {\n"); 1231 pandecode_indent++; 1232 1233 pandecode_prop("unk3 = 0x%" PRIx32, s->bifrost2.unk3); 1234 pandecode_prop("preload_regs = 0x%" PRIx32, s->bifrost2.preload_regs); 1235 pandecode_prop("uniform_count = %" PRId32, s->bifrost2.uniform_count); 1236 pandecode_prop("unk4 = 0x%" PRIx32, s->bifrost2.unk4); 1237 1238 pandecode_indent--; 1239 pandecode_log("},\n"); 1240 } else { 1241 pandecode_log(".midgard2 = {\n"); 1242 pandecode_indent++; 1243 1244 pandecode_prop("unknown2_7 = 0x%" PRIx32, s->midgard2.unknown2_7); 1245 pandecode_indent--; 1246 pandecode_log("},\n"); 1247 } 1248 1249 pandecode_prop("unknown2_8 = 0x%" PRIx32, s->unknown2_8); 1250 1251 bool blend_shader = false; 1252 1253 if (!is_bifrost) { 1254 if (s->unknown2_3 & MALI_HAS_BLEND_SHADER) { 1255 blend_shader = true; 1256 pandecode_replay_shader_address("blend_shader", s->blend_shader); 1257 } else { 1258 pandecode_replay_blend_equation(&s->blend_equation, ""); 1259 } 1260 } 1261 1262 pandecode_indent--; 1263 pandecode_log("};\n"); 1264 1265 /* MRT blend fields are used whenever MFBD is used */ 1266 1267 if (job_type == JOB_TYPE_TILER) { 1268 pandecode_log("struct mali_blend_meta blend_meta_%d[] = {\n", 1269 job_no); 1270 pandecode_indent++; 1271 1272 int i; 1273 1274 for (i = 0; i < 4; i++) { 1275 const struct mali_blend_meta *b = &s->blend_meta[i]; 1276 pandecode_log("{\n"); 1277 pandecode_indent++; 1278 1279#ifndef BIFROST 1280 pandecode_prop("unk1 = 0x%" PRIx64, b->unk1); 1281 1282 /* Depending on unk1, we determine if there's a 1283 * blend shader */ 1284 1285 if ((b->unk1 & 0xF) >= 0x2) { 1286 blend_shader = true; 1287 pandecode_replay_shader_address("blend_shader", b->blend_shader); 1288 } else { 1289 pandecode_replay_blend_equation(&b->blend_equation_1, "_1"); 1290 } 1291 1292 /* This is always an equation, I think. If 1293 * there's a shader, it just defaults to 1294 * REPLACE (0x122) */ 1295 pandecode_replay_blend_equation(&b->blend_equation_2, "_2"); 1296 1297 if (b->zero2) { 1298 pandecode_msg("blend zero tripped\n"); 1299 pandecode_prop("zero2 = 0x%x", b->zero2); 1300 } 1301 1302#else 1303 1304 pandecode_prop("unk1 = 0x%" PRIx32, b->unk1); 1305 /* TODO figure out blend shader enable bit */ 1306 pandecode_replay_blend_equation(&b->blend_equation); 1307 pandecode_prop("unk2 = 0x%" PRIx16, b->unk2); 1308 pandecode_prop("index = 0x%" PRIx16, b->index); 1309 pandecode_prop("unk3 = 0x%" PRIx32, b->unk3); 1310#endif 1311 1312 pandecode_indent--; 1313 pandecode_log("},\n"); 1314 1315#ifdef BIFROST 1316 if (b->unk2 == 3) 1317 break; 1318#else 1319 /* TODO: What's this supposed to be? */ 1320 if (b->unk1 & 0x200) 1321 break; 1322#endif 1323 1324 } 1325 1326 pandecode_indent--; 1327 pandecode_log("};\n"); 1328 1329 /* This needs to be uploaded right after the 1330 * shader_meta since it's technically part of the same 1331 * (variable-size) structure. 1332 */ 1333 } 1334 1335 pandecode_shader_disassemble(shader_ptr, job_no, job_type, is_bifrost); 1336 1337 if (!is_bifrost && blend_shader) 1338 pandecode_shader_disassemble(s->blend_shader & ~0xF, job_no, job_type, false); 1339 1340 } else 1341 pandecode_msg("<no shader>\n"); 1342 1343 if (p->viewport) { 1344 struct pandecode_mapped_memory *fmem = pandecode_find_mapped_gpu_mem_containing(p->viewport); 1345 struct mali_viewport *PANDECODE_PTR_VAR(f, fmem, p->viewport); 1346 1347 pandecode_log("struct mali_viewport viewport_%d%s = {\n", job_no, suffix); 1348 pandecode_indent++; 1349 1350 pandecode_prop("clip_minx = %f", f->clip_minx); 1351 pandecode_prop("clip_miny = %f", f->clip_miny); 1352 pandecode_prop("clip_minz = %f", f->clip_minz); 1353 pandecode_prop("clip_maxx = %f", f->clip_maxx); 1354 pandecode_prop("clip_maxy = %f", f->clip_maxy); 1355 pandecode_prop("clip_maxz = %f", f->clip_maxz); 1356 1357 /* Only the higher coordinates are MALI_POSITIVE scaled */ 1358 1359 pandecode_prop("viewport0 = { %d, %d }", 1360 f->viewport0[0], f->viewport0[1]); 1361 1362 pandecode_prop("viewport1 = { MALI_POSITIVE(%d), MALI_POSITIVE(%d) }", 1363 f->viewport1[0] + 1, f->viewport1[1] + 1); 1364 1365 pandecode_indent--; 1366 pandecode_log("};\n"); 1367 } 1368 1369 if (p->attribute_meta) { 1370 unsigned max_attr_index = pandecode_replay_attribute_meta(job_no, attribute_count, p, false, suffix); 1371 1372 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->attributes); 1373 pandecode_replay_attributes(attr_mem, p->attributes, job_no, suffix, max_attr_index + 1, false); 1374 } 1375 1376 /* Varyings are encoded like attributes but not actually sent; we just 1377 * pass a zero buffer with the right stride/size set, (or whatever) 1378 * since the GPU will write to it itself */ 1379 1380 if (p->varyings) { 1381 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->varyings); 1382 1383 /* Number of descriptors depends on whether there are 1384 * non-internal varyings */ 1385 1386 pandecode_replay_attributes(attr_mem, p->varyings, job_no, suffix, varying_count > 1 ? 4 : 1, true); 1387 } 1388 1389 if (p->varying_meta) { 1390 pandecode_replay_attribute_meta(job_no, varying_count, p, true, suffix); 1391 } 1392 1393 if (p->uniforms) { 1394 int rows = uniform_count, width = 4; 1395 size_t sz = rows * width * sizeof(float); 1396 1397 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms); 1398 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz); 1399 float *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms); 1400 1401 pandecode_log("float uniforms_%d%s[] = {\n", job_no, suffix); 1402 1403 pandecode_indent++; 1404 1405 for (int row = 0; row < rows; row++) { 1406 for (int i = 0; i < width; i++) 1407 pandecode_log_cont("%ff, ", uniforms[i]); 1408 1409 pandecode_log_cont("\n"); 1410 1411 uniforms += width; 1412 } 1413 1414 pandecode_indent--; 1415 pandecode_log("};\n"); 1416 } 1417 1418 if (p->uniform_buffers) { 1419 pandecode_replay_uniform_buffers(p->uniform_buffers, uniform_buffer_count, job_no); 1420 } 1421 1422 if (p->texture_trampoline) { 1423 struct pandecode_mapped_memory *mmem = pandecode_find_mapped_gpu_mem_containing(p->texture_trampoline); 1424 1425 if (mmem) { 1426 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline); 1427 1428 pandecode_log("uint64_t texture_trampoline_%d[] = {\n", job_no); 1429 pandecode_indent++; 1430 1431 for (int tex = 0; tex < texture_count; ++tex) { 1432 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr)); 1433 char *a = pointer_as_memory_reference(*u); 1434 pandecode_log("%s,\n", a); 1435 free(a); 1436 } 1437 1438 pandecode_indent--; 1439 pandecode_log("};\n"); 1440 1441 /* Now, finally, descend down into the texture descriptor */ 1442 for (int tex = 0; tex < texture_count; ++tex) { 1443 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr)); 1444 struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(*u); 1445 1446 if (tmem) { 1447 struct mali_texture_descriptor *PANDECODE_PTR_VAR(t, tmem, *u); 1448 1449 pandecode_log("struct mali_texture_descriptor texture_descriptor_%d_%d = {\n", job_no, tex); 1450 pandecode_indent++; 1451 1452 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", t->width + 1); 1453 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", t->height + 1); 1454 pandecode_prop("depth = MALI_POSITIVE(%" PRId16 ")", t->depth + 1); 1455 1456 pandecode_prop("unknown3 = %" PRId16, t->unknown3); 1457 pandecode_prop("unknown3A = %" PRId8, t->unknown3A); 1458 pandecode_prop("nr_mipmap_levels = %" PRId8, t->nr_mipmap_levels); 1459 1460 struct mali_texture_format f = t->format; 1461 1462 pandecode_log(".format = {\n"); 1463 pandecode_indent++; 1464 1465 pandecode_replay_swizzle(f.swizzle); 1466 pandecode_prop("format = %s", pandecode_format_name(f.format)); 1467 1468 pandecode_prop("usage1 = 0x%" PRIx32, f.usage1); 1469 pandecode_prop("is_not_cubemap = %" PRId32, f.is_not_cubemap); 1470 pandecode_prop("usage2 = 0x%" PRIx32, f.usage2); 1471 1472 pandecode_indent--; 1473 pandecode_log("},\n"); 1474 1475 pandecode_replay_swizzle(t->swizzle); 1476 1477 if (t->swizzle_zero) { 1478 /* Shouldn't happen */ 1479 pandecode_msg("Swizzle zero tripped but replay will be fine anyway"); 1480 pandecode_prop("swizzle_zero = %d", t->swizzle_zero); 1481 } 1482 1483 pandecode_prop("unknown3 = 0x%" PRIx32, t->unknown3); 1484 1485 pandecode_prop("unknown5 = 0x%" PRIx32, t->unknown5); 1486 pandecode_prop("unknown6 = 0x%" PRIx32, t->unknown6); 1487 pandecode_prop("unknown7 = 0x%" PRIx32, t->unknown7); 1488 1489 pandecode_log(".swizzled_bitmaps = {\n"); 1490 pandecode_indent++; 1491 1492 int bitmap_count = MALI_NEGATIVE(t->nr_mipmap_levels); 1493 1494 if (!f.is_not_cubemap) { 1495 /* Miptree for each face */ 1496 bitmap_count *= 6; 1497 } 1498 1499 int max_count = sizeof(t->swizzled_bitmaps) / sizeof(t->swizzled_bitmaps[0]); 1500 1501 if (bitmap_count > max_count) { 1502 pandecode_msg("XXX: bitmap count tripped"); 1503 bitmap_count = max_count; 1504 } 1505 1506 for (int i = 0; i < bitmap_count; ++i) { 1507 char *a = pointer_as_memory_reference(t->swizzled_bitmaps[i]); 1508 pandecode_log("%s, \n", a); 1509 free(a); 1510 } 1511 1512 pandecode_indent--; 1513 pandecode_log("},\n"); 1514 1515 pandecode_indent--; 1516 pandecode_log("};\n"); 1517 } 1518 } 1519 } 1520 } 1521 1522 if (p->sampler_descriptor) { 1523 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(p->sampler_descriptor); 1524 1525 if (smem) { 1526 struct mali_sampler_descriptor *s; 1527 1528 mali_ptr d = p->sampler_descriptor; 1529 1530 for (int i = 0; i < sampler_count; ++i) { 1531 s = pandecode_fetch_gpu_mem(smem, d + sizeof(*s) * i, sizeof(*s)); 1532 1533 pandecode_log("struct mali_sampler_descriptor sampler_descriptor_%d_%d = {\n", job_no, i); 1534 pandecode_indent++; 1535 1536 /* Only the lower two bits are understood right now; the rest we display as hex */ 1537 pandecode_log(".filter_mode = MALI_TEX_MIN(%s) | MALI_TEX_MAG(%s) | 0x%" PRIx32",\n", 1538 MALI_FILTER_NAME(s->filter_mode & MALI_TEX_MIN_MASK), 1539 MALI_FILTER_NAME(s->filter_mode & MALI_TEX_MAG_MASK), 1540 s->filter_mode & ~3); 1541 1542 pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s->min_lod)); 1543 pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s->max_lod)); 1544 1545 pandecode_prop("wrap_s = %s", pandecode_wrap_mode_name(s->wrap_s)); 1546 pandecode_prop("wrap_t = %s", pandecode_wrap_mode_name(s->wrap_t)); 1547 pandecode_prop("wrap_r = %s", pandecode_wrap_mode_name(s->wrap_r)); 1548 1549 pandecode_prop("compare_func = %s", pandecode_alt_func_name(s->compare_func)); 1550 1551 if (s->zero || s->zero2) { 1552 pandecode_msg("Zero tripped\n"); 1553 pandecode_prop("zero = 0x%X, 0x%X\n", s->zero, s->zero2); 1554 } 1555 1556 pandecode_prop("unknown2 = %d", s->unknown2); 1557 1558 pandecode_prop("border_color = { %f, %f, %f, %f }", 1559 s->border_color[0], 1560 s->border_color[1], 1561 s->border_color[2], 1562 s->border_color[3]); 1563 1564 pandecode_indent--; 1565 pandecode_log("};\n"); 1566 } 1567 } 1568 } 1569} 1570 1571static void 1572pandecode_replay_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix *p, int job_no, bool is_bifrost) 1573{ 1574 pandecode_log_cont("{\n"); 1575 pandecode_indent++; 1576 1577 MEMORY_PROP(p, position_varying); 1578 MEMORY_COMMENT(p, position_varying); 1579 DYN_MEMORY_PROP(p, job_no, uniform_buffers); 1580 MEMORY_COMMENT(p, uniform_buffers); 1581 DYN_MEMORY_PROP(p, job_no, texture_trampoline); 1582 MEMORY_COMMENT(p, texture_trampoline); 1583 DYN_MEMORY_PROP(p, job_no, sampler_descriptor); 1584 MEMORY_COMMENT(p, sampler_descriptor); 1585 DYN_MEMORY_PROP(p, job_no, uniforms); 1586 MEMORY_COMMENT(p, uniforms); 1587 DYN_MEMORY_PROP(p, job_no, attributes); 1588 MEMORY_COMMENT(p, attributes); 1589 DYN_MEMORY_PROP(p, job_no, attribute_meta); 1590 MEMORY_COMMENT(p, attribute_meta); 1591 DYN_MEMORY_PROP(p, job_no, varyings); 1592 MEMORY_COMMENT(p, varyings); 1593 DYN_MEMORY_PROP(p, job_no, varying_meta); 1594 MEMORY_COMMENT(p, varying_meta); 1595 DYN_MEMORY_PROP(p, job_no, viewport); 1596 MEMORY_COMMENT(p, viewport); 1597 DYN_MEMORY_PROP(p, job_no, occlusion_counter); 1598 MEMORY_COMMENT(p, occlusion_counter); 1599 MEMORY_COMMENT(p, framebuffer & ~1); 1600 pandecode_msg("%" PRIx64 "\n", p->viewport); 1601 pandecode_msg("%" PRIx64 "\n", p->framebuffer); 1602 1603 if (is_bifrost) 1604 pandecode_prop("framebuffer = scratchpad_%d_p", job_no); 1605 else 1606 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, p->framebuffer & MALI_MFBD ? "MALI_MFBD" : "0"); 1607 1608 pandecode_prop("_shader_upper = (shader_meta_%d_p) >> 4", job_no); 1609 pandecode_prop("flags = %d", p->flags); 1610 1611 pandecode_indent--; 1612 pandecode_log("},\n"); 1613} 1614 1615static void 1616pandecode_replay_vertex_only_bfr(struct bifrost_vertex_only *v) 1617{ 1618 pandecode_log_cont("{\n"); 1619 pandecode_indent++; 1620 1621 pandecode_prop("unk2 = 0x%x", v->unk2); 1622 1623 if (v->zero0 || v->zero1) { 1624 pandecode_msg("vertex only zero tripped"); 1625 pandecode_prop("zero0 = 0x%" PRIx32, v->zero0); 1626 pandecode_prop("zero1 = 0x%" PRIx64, v->zero1); 1627 } 1628 1629 pandecode_indent--; 1630 pandecode_log("}\n"); 1631} 1632 1633static void 1634pandecode_replay_tiler_heap_meta(mali_ptr gpu_va, int job_no) 1635{ 1636 1637 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va); 1638 const struct bifrost_tiler_heap_meta *PANDECODE_PTR_VAR(h, mem, gpu_va); 1639 1640 pandecode_log("struct mali_tiler_heap_meta tiler_heap_meta_%d = {\n", job_no); 1641 pandecode_indent++; 1642 1643 if (h->zero) { 1644 pandecode_msg("tiler heap zero tripped\n"); 1645 pandecode_prop("zero = 0x%x", h->zero); 1646 } 1647 1648 for (int i = 0; i < 12; i++) { 1649 if (h->zeros[i] != 0) { 1650 pandecode_msg("tiler heap zero %d tripped, value %x\n", 1651 i, h->zeros[i]); 1652 } 1653 } 1654 1655 pandecode_prop("heap_size = 0x%x", h->heap_size); 1656 MEMORY_PROP(h, tiler_heap_start); 1657 MEMORY_PROP(h, tiler_heap_free); 1658 1659 /* this might point to the beginning of another buffer, when it's 1660 * really the end of the tiler heap buffer, so we have to be careful 1661 * here. 1662 */ 1663 char *a = pointer_as_memory_reference(h->tiler_heap_end - 1); 1664 pandecode_prop("tiler_heap_end = %s + 1", a); 1665 free(a); 1666 1667 pandecode_indent--; 1668 pandecode_log("};\n"); 1669} 1670 1671static void 1672pandecode_replay_tiler_meta(mali_ptr gpu_va, int job_no) 1673{ 1674 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va); 1675 const struct bifrost_tiler_meta *PANDECODE_PTR_VAR(t, mem, gpu_va); 1676 1677 pandecode_replay_tiler_heap_meta(t->tiler_heap_meta, job_no); 1678 1679 pandecode_log("struct mali_tiler_meta tiler_meta_%d = {\n", job_no); 1680 pandecode_indent++; 1681 1682 if (t->zero0 || t->zero1) { 1683 pandecode_msg("tiler meta zero tripped"); 1684 pandecode_prop("zero0 = 0x%" PRIx64, t->zero0); 1685 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1); 1686 } 1687 1688 pandecode_prop("unk = 0x%x", t->unk); 1689 pandecode_prop("width = MALI_POSITIVE(%d)", t->width + 1); 1690 pandecode_prop("height = MALI_POSITIVE(%d)", t->height + 1); 1691 DYN_MEMORY_PROP(t, job_no, tiler_heap_meta); 1692 1693 for (int i = 0; i < 12; i++) { 1694 if (t->zeros[i] != 0) { 1695 pandecode_msg("tiler heap zero %d tripped, value %" PRIx64 "\n", 1696 i, t->zeros[i]); 1697 } 1698 } 1699 1700 pandecode_indent--; 1701 pandecode_log("};\n"); 1702} 1703 1704static void 1705pandecode_replay_gl_enables(uint32_t gl_enables, int job_type) 1706{ 1707 pandecode_log(".gl_enables = "); 1708 1709 if (job_type == JOB_TYPE_TILER) { 1710 pandecode_log_cont("MALI_FRONT_FACE(MALI_%s) | ", 1711 gl_enables & MALI_FRONT_FACE(MALI_CW) ? "CW" : "CCW"); 1712 1713 gl_enables &= ~(MALI_FRONT_FACE(1)); 1714 } 1715 1716 pandecode_log_decoded_flags(gl_enable_flag_info, gl_enables); 1717 1718 pandecode_log_cont(",\n"); 1719} 1720 1721static void 1722pandecode_replay_primitive_size(union midgard_primitive_size u, bool constant) 1723{ 1724 pandecode_log(".primitive_size = {\n"); 1725 pandecode_indent++; 1726 1727 if (constant) { 1728 pandecode_prop("constant = %f", u.constant); 1729 } else { 1730 MEMORY_PROP((&u), pointer); 1731 } 1732 1733 pandecode_indent--; 1734 pandecode_log("},\n"); 1735} 1736 1737static void 1738pandecode_replay_tiler_only_bfr(const struct bifrost_tiler_only *t, int job_no) 1739{ 1740 pandecode_log_cont("{\n"); 1741 pandecode_indent++; 1742 1743 /* TODO: gl_PointSize on Bifrost */ 1744 pandecode_replay_primitive_size(t->primitive_size, true); 1745 1746 DYN_MEMORY_PROP(t, job_no, tiler_meta); 1747 pandecode_replay_gl_enables(t->gl_enables, JOB_TYPE_TILER); 1748 1749 if (t->zero1 || t->zero2 || t->zero3 || t->zero4 || t->zero5 1750 || t->zero6 || t->zero7 || t->zero8) { 1751 pandecode_msg("tiler only zero tripped"); 1752 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1); 1753 pandecode_prop("zero2 = 0x%" PRIx64, t->zero2); 1754 pandecode_prop("zero3 = 0x%" PRIx64, t->zero3); 1755 pandecode_prop("zero4 = 0x%" PRIx64, t->zero4); 1756 pandecode_prop("zero5 = 0x%" PRIx64, t->zero5); 1757 pandecode_prop("zero6 = 0x%" PRIx64, t->zero6); 1758 pandecode_prop("zero7 = 0x%" PRIx32, t->zero7); 1759 pandecode_prop("zero8 = 0x%" PRIx64, t->zero8); 1760 } 1761 1762 pandecode_indent--; 1763 pandecode_log("},\n"); 1764} 1765 1766static int 1767pandecode_replay_vertex_job_bfr(const struct mali_job_descriptor_header *h, 1768 const struct pandecode_mapped_memory *mem, 1769 mali_ptr payload, int job_no) 1770{ 1771 struct bifrost_payload_vertex *PANDECODE_PTR_VAR(v, mem, payload); 1772 1773 pandecode_replay_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", true); 1774 1775 pandecode_log("struct bifrost_payload_vertex payload_%d = {\n", job_no); 1776 pandecode_indent++; 1777 1778 pandecode_log(".prefix = "); 1779 pandecode_replay_vertex_tiler_prefix(&v->prefix, job_no); 1780 1781 pandecode_log(".vertex = "); 1782 pandecode_replay_vertex_only_bfr(&v->vertex); 1783 1784 pandecode_log(".postfix = "); 1785 pandecode_replay_vertex_tiler_postfix(&v->postfix, job_no, true); 1786 1787 pandecode_indent--; 1788 pandecode_log("};\n"); 1789 1790 return sizeof(*v); 1791} 1792 1793static int 1794pandecode_replay_tiler_job_bfr(const struct mali_job_descriptor_header *h, 1795 const struct pandecode_mapped_memory *mem, 1796 mali_ptr payload, int job_no) 1797{ 1798 struct bifrost_payload_tiler *PANDECODE_PTR_VAR(t, mem, payload); 1799 1800 pandecode_replay_vertex_tiler_postfix_pre(&t->postfix, job_no, h->job_type, "", true); 1801 1802 pandecode_replay_indices(t->prefix.indices, t->prefix.index_count, job_no); 1803 pandecode_replay_tiler_meta(t->tiler.tiler_meta, job_no); 1804 1805 pandecode_log("struct bifrost_payload_tiler payload_%d = {\n", job_no); 1806 pandecode_indent++; 1807 1808 pandecode_log(".prefix = "); 1809 pandecode_replay_vertex_tiler_prefix(&t->prefix, job_no); 1810 1811 pandecode_log(".tiler = "); 1812 pandecode_replay_tiler_only_bfr(&t->tiler, job_no); 1813 1814 pandecode_log(".postfix = "); 1815 pandecode_replay_vertex_tiler_postfix(&t->postfix, job_no, true); 1816 1817 pandecode_indent--; 1818 pandecode_log("};\n"); 1819 1820 return sizeof(*t); 1821} 1822 1823static int 1824pandecode_replay_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header *h, 1825 const struct pandecode_mapped_memory *mem, 1826 mali_ptr payload, int job_no) 1827{ 1828 struct midgard_payload_vertex_tiler *PANDECODE_PTR_VAR(v, mem, payload); 1829 1830 char *a = pointer_as_memory_reference(payload); 1831 pandecode_msg("vt payload: %s\n", a); 1832 free(a); 1833 1834 pandecode_replay_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", false); 1835 1836 pandecode_replay_indices(v->prefix.indices, v->prefix.index_count, job_no); 1837 1838 pandecode_log("struct midgard_payload_vertex_tiler payload_%d = {\n", job_no); 1839 pandecode_indent++; 1840 1841 bool has_primitive_pointer = v->prefix.unknown_draw & MALI_DRAW_VARYING_SIZE; 1842 pandecode_replay_primitive_size(v->primitive_size, !has_primitive_pointer); 1843 1844 pandecode_log(".prefix = "); 1845 pandecode_replay_vertex_tiler_prefix(&v->prefix, job_no); 1846 1847 pandecode_replay_gl_enables(v->gl_enables, h->job_type); 1848 pandecode_prop("draw_start = %d", v->draw_start); 1849 1850#ifndef __LP64__ 1851 1852 if (v->zero3) { 1853 pandecode_msg("Zero tripped\n"); 1854 pandecode_prop("zero3 = 0x%" PRIx32, v->zero3); 1855 } 1856 1857#endif 1858 1859 if (v->zero5) { 1860 pandecode_msg("Zero tripped\n"); 1861 pandecode_prop("zero5 = 0x%" PRIx64, v->zero5); 1862 } 1863 1864 pandecode_log(".postfix = "); 1865 pandecode_replay_vertex_tiler_postfix(&v->postfix, job_no, false); 1866 1867 pandecode_indent--; 1868 pandecode_log("};\n"); 1869 1870 return sizeof(*v); 1871} 1872 1873static int 1874pandecode_replay_fragment_job(const struct pandecode_mapped_memory *mem, 1875 mali_ptr payload, int job_no, 1876 bool is_bifrost) 1877{ 1878 const struct mali_payload_fragment *PANDECODE_PTR_VAR(s, mem, payload); 1879 1880 bool fbd_dumped = false; 1881 1882 if (!is_bifrost && (s->framebuffer & FBD_TYPE) == MALI_SFBD) { 1883 /* Only SFBDs are understood, not MFBDs. We're speculating, 1884 * based on the versioning, kernel code, etc, that the 1885 * difference is between Single FrameBuffer Descriptor and 1886 * Multiple FrmaeBuffer Descriptor; the change apparently lines 1887 * up with multi-framebuffer support being added (T7xx onwards, 1888 * including Gxx). In any event, there's some field shuffling 1889 * that we haven't looked into yet. */ 1890 1891 pandecode_replay_sfbd(s->framebuffer & FBD_MASK, job_no); 1892 fbd_dumped = true; 1893 } else if ((s->framebuffer & FBD_TYPE) == MALI_MFBD) { 1894 /* We don't know if Bifrost supports SFBD's at all, since the 1895 * driver never uses them. And the format is different from 1896 * Midgard anyways, due to the tiler heap and scratchpad being 1897 * moved out into separate structures, so it's not clear what a 1898 * Bifrost SFBD would even look like without getting an actual 1899 * trace, which appears impossible. 1900 */ 1901 1902 pandecode_replay_mfbd_bfr(s->framebuffer & FBD_MASK, job_no); 1903 fbd_dumped = true; 1904 } 1905 1906 uintptr_t p = (uintptr_t) s->framebuffer & FBD_MASK; 1907 1908 pandecode_log("struct mali_payload_fragment payload_%d = {\n", job_no); 1909 pandecode_indent++; 1910 1911 /* See the comments by the macro definitions for mathematical context 1912 * on why this is so weird */ 1913 1914 if (MALI_TILE_COORD_FLAGS(s->max_tile_coord) || MALI_TILE_COORD_FLAGS(s->min_tile_coord)) 1915 pandecode_msg("Tile coordinate flag missed, replay wrong\n"); 1916 1917 pandecode_prop("min_tile_coord = MALI_COORDINATE_TO_TILE_MIN(%d, %d)", 1918 MALI_TILE_COORD_X(s->min_tile_coord) << MALI_TILE_SHIFT, 1919 MALI_TILE_COORD_Y(s->min_tile_coord) << MALI_TILE_SHIFT); 1920 1921 pandecode_prop("max_tile_coord = MALI_COORDINATE_TO_TILE_MAX(%d, %d)", 1922 (MALI_TILE_COORD_X(s->max_tile_coord) + 1) << MALI_TILE_SHIFT, 1923 (MALI_TILE_COORD_Y(s->max_tile_coord) + 1) << MALI_TILE_SHIFT); 1924 1925 /* If the FBD was just decoded, we can refer to it by pointer. If not, 1926 * we have to fallback on offsets. */ 1927 1928 const char *fbd_type = s->framebuffer & MALI_MFBD ? "MALI_MFBD" : "MALI_SFBD"; 1929 1930 if (fbd_dumped) 1931 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, fbd_type); 1932 else 1933 pandecode_prop("framebuffer = %s | %s", pointer_as_memory_reference(p), fbd_type); 1934 1935 pandecode_indent--; 1936 pandecode_log("};\n"); 1937 1938 return sizeof(*s); 1939} 1940 1941static int job_descriptor_number = 0; 1942 1943int 1944pandecode_replay_jc(mali_ptr jc_gpu_va, bool bifrost) 1945{ 1946 struct mali_job_descriptor_header *h; 1947 1948 int start_number = 0; 1949 1950 bool first = true; 1951 bool last_size; 1952 1953 do { 1954 struct pandecode_mapped_memory *mem = 1955 pandecode_find_mapped_gpu_mem_containing(jc_gpu_va); 1956 1957 void *payload; 1958 1959 h = PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_descriptor_header); 1960 1961 /* On Midgard, for 32-bit jobs except for fragment jobs, the 1962 * high 32-bits of the 64-bit pointer are reused to store 1963 * something else. 1964 */ 1965 int offset = h->job_descriptor_size == MALI_JOB_32 && 1966 h->job_type != JOB_TYPE_FRAGMENT ? 4 : 0; 1967 mali_ptr payload_ptr = jc_gpu_va + sizeof(*h) - offset; 1968 1969 payload = pandecode_fetch_gpu_mem(mem, payload_ptr, 1970 MALI_PAYLOAD_SIZE); 1971 1972 int job_no = job_descriptor_number++; 1973 1974 if (first) 1975 start_number = job_no; 1976 1977 pandecode_log("struct mali_job_descriptor_header job_%d = {\n", job_no); 1978 pandecode_indent++; 1979 1980 pandecode_prop("job_type = %s", pandecode_job_type_name(h->job_type)); 1981 1982 /* Save for next job fixing */ 1983 last_size = h->job_descriptor_size; 1984 1985 if (h->job_descriptor_size) 1986 pandecode_prop("job_descriptor_size = %d", h->job_descriptor_size); 1987 1988 if (h->exception_status) 1989 pandecode_prop("exception_status = %d", h->exception_status); 1990 1991 if (h->first_incomplete_task) 1992 pandecode_prop("first_incomplete_task = %d", h->first_incomplete_task); 1993 1994 if (h->fault_pointer) 1995 pandecode_prop("fault_pointer = 0x%" PRIx64, h->fault_pointer); 1996 1997 if (h->job_barrier) 1998 pandecode_prop("job_barrier = %d", h->job_barrier); 1999 2000 pandecode_prop("job_index = %d", h->job_index); 2001 2002 if (h->unknown_flags) 2003 pandecode_prop("unknown_flags = %d", h->unknown_flags); 2004 2005 if (h->job_dependency_index_1) 2006 pandecode_prop("job_dependency_index_1 = %d", h->job_dependency_index_1); 2007 2008 if (h->job_dependency_index_2) 2009 pandecode_prop("job_dependency_index_2 = %d", h->job_dependency_index_2); 2010 2011 pandecode_indent--; 2012 pandecode_log("};\n"); 2013 2014 /* Do not touch the field yet -- decode the payload first, and 2015 * don't touch that either. This is essential for the uploads 2016 * to occur in sequence and therefore be dynamically allocated 2017 * correctly. Do note the size, however, for that related 2018 * reason. */ 2019 2020 switch (h->job_type) { 2021 case JOB_TYPE_SET_VALUE: { 2022 struct mali_payload_set_value *s = payload; 2023 2024 pandecode_log("struct mali_payload_set_value payload_%d = {\n", job_no); 2025 pandecode_indent++; 2026 MEMORY_PROP(s, out); 2027 pandecode_prop("unknown = 0x%" PRIX64, s->unknown); 2028 pandecode_indent--; 2029 pandecode_log("};\n"); 2030 2031 break; 2032 } 2033 2034 case JOB_TYPE_TILER: 2035 case JOB_TYPE_VERTEX: 2036 case JOB_TYPE_COMPUTE: 2037 if (bifrost) { 2038 if (h->job_type == JOB_TYPE_TILER) 2039 pandecode_replay_tiler_job_bfr(h, mem, payload_ptr, job_no); 2040 else 2041 pandecode_replay_vertex_job_bfr(h, mem, payload_ptr, job_no); 2042 } else 2043 pandecode_replay_vertex_or_tiler_job_mdg(h, mem, payload_ptr, job_no); 2044 2045 break; 2046 2047 case JOB_TYPE_FRAGMENT: 2048 pandecode_replay_fragment_job(mem, payload_ptr, job_no, bifrost); 2049 break; 2050 2051 default: 2052 break; 2053 } 2054 2055 /* Handle linkage */ 2056 2057 if (!first) { 2058 pandecode_log("((struct mali_job_descriptor_header *) (uintptr_t) job_%d_p)->", job_no - 1); 2059 2060 if (last_size) 2061 pandecode_log_cont("next_job_64 = job_%d_p;\n\n", job_no); 2062 else 2063 pandecode_log_cont("next_job_32 = (u32) (uintptr_t) job_%d_p;\n\n", job_no); 2064 } 2065 2066 first = false; 2067 2068 } while ((jc_gpu_va = h->job_descriptor_size ? h->next_job_64 : h->next_job_32)); 2069 2070 return start_number; 2071} 2072