src/gallium/drivers/radeonsi/si_debug.c

   1 /*
   2  * Copyright 2015 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  * on the rights to use, copy, modify, merge, publish, distribute, sub
   8  * license, and/or sell copies of the Software, and to permit persons to whom
   9  * the Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  18  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  19  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  20  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  21  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  22  *
  23  * Authors:
  24  *      Marek Olšák <maraeo@gmail.com>
  25  */
  26
  27 #include "si_pipe.h"
  28 #include "sid.h"
  29 #include "sid_tables.h"
  30 #include "radeon/radeon_elf_util.h"
  31 #include "ddebug/dd_util.h"
  32 #include "util/u_memory.h"
  33
  34 DEBUG_GET_ONCE_OPTION(replace_shaders, "RADEON_REPLACE_SHADERS", NULL)
  35
  36 static void si_dump_shader(struct si_screen *sscreen,
  37                            struct si_shader_ctx_state *state, FILE *f)
  38 {
  39         struct si_shader *current = state->current;
  40
  41         if (!state->cso || !current)
  42                 return;
  43
  44         if (current->shader_log)
  45                 fwrite(current->shader_log, current->shader_log_size, 1, f);
  46         else
  47                 si_shader_dump(sscreen, state->current, NULL,
  48                                state->cso->info.processor, f);
  49 }
  50
  51 /**
  52  * Shader compiles can be overridden with arbitrary ELF objects by setting
  53  * the environment variable RADEON_REPLACE_SHADERS=num1:filename1[;num2:filename2]
  54  */
  55 bool si_replace_shader(unsigned num, struct radeon_shader_binary *binary)
  56 {
  57         const char *p = debug_get_option_replace_shaders();
  58         const char *semicolon;
  59         char *copy = NULL;
  60         FILE *f;
  61         long filesize, nread;
  62         char *buf = NULL;
  63         bool replaced = false;
  64
  65         if (!p)
  66                 return false;
  67
  68         while (*p) {
  69                 unsigned long i;
  70                 char *endp;
  71                 i = strtoul(p, &endp, 0);
  72
  73                 p = endp;
  74                 if (*p != ':') {
  75                         fprintf(stderr, "RADEON_REPLACE_SHADERS formatted badly.\n");
  76                         exit(1);
  77                 }
  78                 ++p;
  79
  80                 if (i == num)
  81                         break;
  82
  83                 p = strchr(p, ';');
  84                 if (!p)
  85                         return false;
  86                 ++p;
  87         }
  88         if (!*p)
  89                 return false;
  90
  91         semicolon = strchr(p, ';');
  92         if (semicolon) {
  93                 p = copy = strndup(p, semicolon - p);
  94                 if (!copy) {
  95                         fprintf(stderr, "out of memory\n");
  96                         return false;
  97                 }
  98         }
  99
 100         fprintf(stderr, "radeonsi: replace shader %u by %s\n", num, p);
 101
 102         f = fopen(p, "r");
 103         if (!f) {
 104                 perror("radeonsi: failed to open file");
 105                 goto out_free;
 106         }
 107
 108         if (fseek(f, 0, SEEK_END) != 0)
 109                 goto file_error;
 110
 111         filesize = ftell(f);
 112         if (filesize < 0)
 113                 goto file_error;
 114
 115         if (fseek(f, 0, SEEK_SET) != 0)
 116                 goto file_error;
 117
 118         buf = MALLOC(filesize);
 119         if (!buf) {
 120                 fprintf(stderr, "out of memory\n");
 121                 goto out_close;
 122         }
 123
 124         nread = fread(buf, 1, filesize, f);
 125         if (nread != filesize)
 126                 goto file_error;
 127
 128         radeon_elf_read(buf, filesize, binary);
 129         replaced = true;
 130
 131 out_close:
 132         fclose(f);
 133 out_free:
 134         FREE(buf);
 135         free(copy);
 136         return replaced;
 137
 138 file_error:
 139         perror("radeonsi: reading shader");
 140         goto out_close;
 141 }
 142
 143 /* Parsed IBs are difficult to read without colors. Use "less -R file" to
 144  * read them, or use "aha -b -f file" to convert them to html.
 145  */
 146 #define COLOR_RESET     "\033[0m"
 147 #define COLOR_RED       "\033[31m"
 148 #define COLOR_GREEN     "\033[1;32m"
 149 #define COLOR_YELLOW    "\033[1;33m"
 150 #define COLOR_CYAN      "\033[1;36m"
 151
 152 #define INDENT_PKT 8
 153
 154 static void print_spaces(FILE *f, unsigned num)
 155 {
 156         fprintf(f, "%*s", num, "");
 157 }
 158
 159 static void print_value(FILE *file, uint32_t value, int bits)
 160 {
 161         /* Guess if it's int or float */
 162         if (value <= (1 << 15)) {
 163                 if (value <= 9)
 164                         fprintf(file, "%u\n", value);
 165                 else
 166                         fprintf(file, "%u (0x%0*x)\n", value, bits / 4, value);
 167         } else {
 168                 float f = uif(value);
 169
 170                 if (fabs(f) < 100000 && f*10 == floor(f*10))
 171                         fprintf(file, "%.1ff (0x%0*x)\n", f, bits / 4, value);
 172                 else
 173                         /* Don't print more leading zeros than there are bits. */
 174                         fprintf(file, "0x%0*x\n", bits / 4, value);
 175         }
 176 }
 177
 178 static void print_named_value(FILE *file, const char *name, uint32_t value,
 179                               int bits)
 180 {
 181         print_spaces(file, INDENT_PKT);
 182         fprintf(file, COLOR_YELLOW "%s" COLOR_RESET " <- ", name);
 183         print_value(file, value, bits);
 184 }
 185
 186 static void si_dump_reg(FILE *file, unsigned offset, uint32_t value,
 187                         uint32_t field_mask)
 188 {
 189         int r, f;
 190
 191         for (r = 0; r < ARRAY_SIZE(sid_reg_table); r++) {
 192                 const struct si_reg *reg = &sid_reg_table[r];
 193                 const char *reg_name = sid_strings + reg->name_offset;
 194
 195                 if (reg->offset == offset) {
 196                         bool first_field = true;
 197
 198                         print_spaces(file, INDENT_PKT);
 199                         fprintf(file, COLOR_YELLOW "%s" COLOR_RESET " <- ",
 200                                 reg_name);
 201
 202                         if (!reg->num_fields) {
 203                                 print_value(file, value, 32);
 204                                 return;
 205                         }
 206
 207                         for (f = 0; f < reg->num_fields; f++) {
 208                                 const struct si_field *field = sid_fields_table + reg->fields_offset + f;
 209                                 const int *values_offsets = sid_strings_offsets + field->values_offset;
 210                                 uint32_t val = (value & field->mask) >>
 211                                                (ffs(field->mask) - 1);
 212
 213                                 if (!(field->mask & field_mask))
 214                                         continue;
 215
 216                                 /* Indent the field. */
 217                                 if (!first_field)
 218                                         print_spaces(file,
 219                                                      INDENT_PKT + strlen(reg_name) + 4);
 220
 221                                 /* Print the field. */
 222                                 fprintf(file, "%s = ", sid_strings + field->name_offset);
 223
 224                                 if (val < field->num_values && values_offsets[val] >= 0)
 225                                         fprintf(file, "%s\n", sid_strings + values_offsets[val]);
 226                                 else
 227                                         print_value(file, val,
 228                                                     util_bitcount(field->mask));
 229
 230                                 first_field = false;
 231                         }
 232                         return;
 233                 }
 234         }
 235
 236         fprintf(file, COLOR_YELLOW "0x%05x" COLOR_RESET " = 0x%08x", offset, value);
 237 }
 238
 239 static void si_parse_set_reg_packet(FILE *f, uint32_t *ib, unsigned count,
 240                                     unsigned reg_offset)
 241 {
 242         unsigned reg = (ib[1] << 2) + reg_offset;
 243         int i;
 244
 245         for (i = 0; i < count; i++)
 246                 si_dump_reg(f, reg + i*4, ib[2+i], ~0);
 247 }
 248
 249 static uint32_t *si_parse_packet3(FILE *f, uint32_t *ib, int *num_dw,
 250                                   int trace_id, enum chip_class chip_class)
 251 {
 252         unsigned count = PKT_COUNT_G(ib[0]);
 253         unsigned op = PKT3_IT_OPCODE_G(ib[0]);
 254         const char *predicate = PKT3_PREDICATE(ib[0]) ? "(predicate)" : "";
 255         int i;
 256
 257         /* Print the name first. */
 258         for (i = 0; i < ARRAY_SIZE(packet3_table); i++)
 259                 if (packet3_table[i].op == op)
 260                         break;
 261
 262         if (i < ARRAY_SIZE(packet3_table)) {
 263                 const char *name = sid_strings + packet3_table[i].name_offset;
 264
 265                 if (op == PKT3_SET_CONTEXT_REG ||
 266                     op == PKT3_SET_CONFIG_REG ||
 267                     op == PKT3_SET_UCONFIG_REG ||
 268                     op == PKT3_SET_SH_REG)
 269                         fprintf(f, COLOR_CYAN "%s%s" COLOR_CYAN ":\n",
 270                                 name, predicate);
 271                 else
 272                         fprintf(f, COLOR_GREEN "%s%s" COLOR_RESET ":\n",
 273                                 name, predicate);
 274         } else
 275                 fprintf(f, COLOR_RED "PKT3_UNKNOWN 0x%x%s" COLOR_RESET ":\n",
 276                         op, predicate);
 277
 278         /* Print the contents. */
 279         switch (op) {
 280         case PKT3_SET_CONTEXT_REG:
 281                 si_parse_set_reg_packet(f, ib, count, SI_CONTEXT_REG_OFFSET);
 282                 break;
 283         case PKT3_SET_CONFIG_REG:
 284                 si_parse_set_reg_packet(f, ib, count, SI_CONFIG_REG_OFFSET);
 285                 break;
 286         case PKT3_SET_UCONFIG_REG:
 287                 si_parse_set_reg_packet(f, ib, count, CIK_UCONFIG_REG_OFFSET);
 288                 break;
 289         case PKT3_SET_SH_REG:
 290                 si_parse_set_reg_packet(f, ib, count, SI_SH_REG_OFFSET);
 291                 break;
 292         case PKT3_ACQUIRE_MEM:
 293                 si_dump_reg(f, R_0301F0_CP_COHER_CNTL, ib[1], ~0);
 294                 si_dump_reg(f, R_0301F4_CP_COHER_SIZE, ib[2], ~0);
 295                 si_dump_reg(f, R_030230_CP_COHER_SIZE_HI, ib[3], ~0);
 296                 si_dump_reg(f, R_0301F8_CP_COHER_BASE, ib[4], ~0);
 297                 si_dump_reg(f, R_0301E4_CP_COHER_BASE_HI, ib[5], ~0);
 298                 print_named_value(f, "POLL_INTERVAL", ib[6], 16);
 299                 break;
 300         case PKT3_SURFACE_SYNC:
 301                 if (chip_class >= CIK) {
 302                         si_dump_reg(f, R_0301F0_CP_COHER_CNTL, ib[1], ~0);
 303                         si_dump_reg(f, R_0301F4_CP_COHER_SIZE, ib[2], ~0);
 304                         si_dump_reg(f, R_0301F8_CP_COHER_BASE, ib[3], ~0);
 305                 } else {
 306                         si_dump_reg(f, R_0085F0_CP_COHER_CNTL, ib[1], ~0);
 307                         si_dump_reg(f, R_0085F4_CP_COHER_SIZE, ib[2], ~0);
 308                         si_dump_reg(f, R_0085F8_CP_COHER_BASE, ib[3], ~0);
 309                 }
 310                 print_named_value(f, "POLL_INTERVAL", ib[4], 16);
 311                 break;
 312         case PKT3_EVENT_WRITE:
 313                 si_dump_reg(f, R_028A90_VGT_EVENT_INITIATOR, ib[1],
 314                             S_028A90_EVENT_TYPE(~0));
 315                 print_named_value(f, "EVENT_INDEX", (ib[1] >> 8) & 0xf, 4);
 316                 print_named_value(f, "INV_L2", (ib[1] >> 20) & 0x1, 1);
 317                 if (count > 0) {
 318                         print_named_value(f, "ADDRESS_LO", ib[2], 32);
 319                         print_named_value(f, "ADDRESS_HI", ib[3], 16);
 320                 }
 321                 break;
 322         case PKT3_DRAW_INDEX_AUTO:
 323                 si_dump_reg(f, R_030930_VGT_NUM_INDICES, ib[1], ~0);
 324                 si_dump_reg(f, R_0287F0_VGT_DRAW_INITIATOR, ib[2], ~0);
 325                 break;
 326         case PKT3_DRAW_INDEX_2:
 327                 si_dump_reg(f, R_028A78_VGT_DMA_MAX_SIZE, ib[1], ~0);
 328                 si_dump_reg(f, R_0287E8_VGT_DMA_BASE, ib[2], ~0);
 329                 si_dump_reg(f, R_0287E4_VGT_DMA_BASE_HI, ib[3], ~0);
 330                 si_dump_reg(f, R_030930_VGT_NUM_INDICES, ib[4], ~0);
 331                 si_dump_reg(f, R_0287F0_VGT_DRAW_INITIATOR, ib[5], ~0);
 332                 break;
 333         case PKT3_INDEX_TYPE:
 334                 si_dump_reg(f, R_028A7C_VGT_DMA_INDEX_TYPE, ib[1], ~0);
 335                 break;
 336         case PKT3_NUM_INSTANCES:
 337                 si_dump_reg(f, R_030934_VGT_NUM_INSTANCES, ib[1], ~0);
 338                 break;
 339         case PKT3_WRITE_DATA:
 340                 si_dump_reg(f, R_370_CONTROL, ib[1], ~0);
 341                 si_dump_reg(f, R_371_DST_ADDR_LO, ib[2], ~0);
 342                 si_dump_reg(f, R_372_DST_ADDR_HI, ib[3], ~0);
 343                 for (i = 2; i < count; i++) {
 344                         print_spaces(f, INDENT_PKT);
 345                         fprintf(f, "0x%08x\n", ib[2+i]);
 346                 }
 347                 break;
 348         case PKT3_CP_DMA:
 349                 si_dump_reg(f, R_410_CP_DMA_WORD0, ib[1], ~0);
 350                 si_dump_reg(f, R_411_CP_DMA_WORD1, ib[2], ~0);
 351                 si_dump_reg(f, R_412_CP_DMA_WORD2, ib[3], ~0);
 352                 si_dump_reg(f, R_413_CP_DMA_WORD3, ib[4], ~0);
 353                 si_dump_reg(f, R_414_COMMAND, ib[5], ~0);
 354                 break;
 355         case PKT3_DMA_DATA:
 356                 si_dump_reg(f, R_500_DMA_DATA_WORD0, ib[1], ~0);
 357                 si_dump_reg(f, R_501_SRC_ADDR_LO, ib[2], ~0);
 358                 si_dump_reg(f, R_502_SRC_ADDR_HI, ib[3], ~0);
 359                 si_dump_reg(f, R_503_DST_ADDR_LO, ib[4], ~0);
 360                 si_dump_reg(f, R_504_DST_ADDR_HI, ib[5], ~0);
 361                 si_dump_reg(f, R_414_COMMAND, ib[6], ~0);
 362                 break;
 363         case PKT3_INDIRECT_BUFFER_SI:
 364         case PKT3_INDIRECT_BUFFER_CONST:
 365         case PKT3_INDIRECT_BUFFER_CIK:
 366                 si_dump_reg(f, R_3F0_IB_BASE_LO, ib[1], ~0);
 367                 si_dump_reg(f, R_3F1_IB_BASE_HI, ib[2], ~0);
 368                 si_dump_reg(f, R_3F2_CONTROL, ib[3], ~0);
 369                 break;
 370         case PKT3_NOP:
 371                 if (ib[0] == 0xffff1000) {
 372                         count = -1; /* One dword NOP. */
 373                         break;
 374                 } else if (count == 0 && SI_IS_TRACE_POINT(ib[1])) {
 375                         unsigned packet_id = SI_GET_TRACE_POINT_ID(ib[1]);
 376
 377                         print_spaces(f, INDENT_PKT);
 378                         fprintf(f, COLOR_RED "Trace point ID: %u\n", packet_id);
 379
 380                         if (trace_id == -1)
 381                                 break; /* tracing was disabled */
 382
 383                         print_spaces(f, INDENT_PKT);
 384                         if (packet_id < trace_id)
 385                                 fprintf(f, COLOR_RED
 386                                         "This trace point was reached by the CP."
 387                                         COLOR_RESET "\n");
 388                         else if (packet_id == trace_id)
 389                                 fprintf(f, COLOR_RED
 390                                         "!!!!! This is the last trace point that "
 391                                         "was reached by the CP !!!!!"
 392                                         COLOR_RESET "\n");
 393                         else if (packet_id+1 == trace_id)
 394                                 fprintf(f, COLOR_RED
 395                                         "!!!!! This is the first trace point that "
 396                                         "was NOT been reached by the CP !!!!!"
 397                                         COLOR_RESET "\n");
 398                         else
 399                                 fprintf(f, COLOR_RED
 400                                         "!!!!! This trace point was NOT reached "
 401                                         "by the CP !!!!!"
 402                                         COLOR_RESET "\n");
 403                         break;
 404                 }
 405                 /* fall through, print all dwords */
 406         default:
 407                 for (i = 0; i < count+1; i++) {
 408                         print_spaces(f, INDENT_PKT);
 409                         fprintf(f, "0x%08x\n", ib[1+i]);
 410                 }
 411         }
 412
 413         ib += count + 2;
 414         *num_dw -= count + 2;
 415         return ib;
 416 }
 417
 418 /**
 419  * Parse and print an IB into a file.
 420  *
 421  * \param f             file
 422  * \param ib            IB
 423  * \param num_dw        size of the IB
 424  * \param chip_class    chip class
 425  * \param trace_id      the last trace ID that is known to have been reached
 426  *                      and executed by the CP, typically read from a buffer
 427  */
 428 static void si_parse_ib(FILE *f, uint32_t *ib, int num_dw, int trace_id,
 429                         const char *name, enum chip_class chip_class)
 430 {
 431         fprintf(f, "------------------ %s begin ------------------\n", name);
 432
 433         while (num_dw > 0) {
 434                 unsigned type = PKT_TYPE_G(ib[0]);
 435
 436                 switch (type) {
 437                 case 3:
 438                         ib = si_parse_packet3(f, ib, &num_dw, trace_id,
 439                                               chip_class);
 440                         break;
 441                 case 2:
 442                         /* type-2 nop */
 443                         if (ib[0] == 0x80000000) {
 444                                 fprintf(f, COLOR_GREEN "NOP (type 2)" COLOR_RESET "\n");
 445                                 ib++;
 446                                 break;
 447                         }
 448                         /* fall through */
 449                 default:
 450                         fprintf(f, "Unknown packet type %i\n", type);
 451                         return;
 452                 }
 453         }
 454
 455         fprintf(f, "------------------- %s end -------------------\n", name);
 456         if (num_dw < 0) {
 457                 printf("Packet ends after the end of IB.\n");
 458                 exit(0);
 459         }
 460         fprintf(f, "\n");
 461 }
 462
 463 static void si_dump_mmapped_reg(struct si_context *sctx, FILE *f,
 464                                 unsigned offset)
 465 {
 466         struct radeon_winsys *ws = sctx->b.ws;
 467         uint32_t value;
 468
 469         if (ws->read_registers(ws, offset, 1, &value))
 470                 si_dump_reg(f, offset, value, ~0);
 471 }
 472
 473 static void si_dump_debug_registers(struct si_context *sctx, FILE *f)
 474 {
 475         if (sctx->screen->b.info.drm_major == 2 &&
 476             sctx->screen->b.info.drm_minor < 42)
 477                 return; /* no radeon support */
 478
 479         fprintf(f, "Memory-mapped registers:\n");
 480         si_dump_mmapped_reg(sctx, f, R_008010_GRBM_STATUS);
 481
 482         /* No other registers can be read on DRM < 3.1.0. */
 483         if (sctx->screen->b.info.drm_major < 3 ||
 484             sctx->screen->b.info.drm_minor < 1) {
 485                 fprintf(f, "\n");
 486                 return;
 487         }
 488
 489         si_dump_mmapped_reg(sctx, f, R_008008_GRBM_STATUS2);
 490         si_dump_mmapped_reg(sctx, f, R_008014_GRBM_STATUS_SE0);
 491         si_dump_mmapped_reg(sctx, f, R_008018_GRBM_STATUS_SE1);
 492         si_dump_mmapped_reg(sctx, f, R_008038_GRBM_STATUS_SE2);
 493         si_dump_mmapped_reg(sctx, f, R_00803C_GRBM_STATUS_SE3);
 494         si_dump_mmapped_reg(sctx, f, R_00D034_SDMA0_STATUS_REG);
 495         si_dump_mmapped_reg(sctx, f, R_00D834_SDMA1_STATUS_REG);
 496         si_dump_mmapped_reg(sctx, f, R_000E50_SRBM_STATUS);
 497         si_dump_mmapped_reg(sctx, f, R_000E4C_SRBM_STATUS2);
 498         si_dump_mmapped_reg(sctx, f, R_000E54_SRBM_STATUS3);
 499         si_dump_mmapped_reg(sctx, f, R_008680_CP_STAT);
 500         si_dump_mmapped_reg(sctx, f, R_008674_CP_STALLED_STAT1);
 501         si_dump_mmapped_reg(sctx, f, R_008678_CP_STALLED_STAT2);
 502         si_dump_mmapped_reg(sctx, f, R_008670_CP_STALLED_STAT3);
 503         si_dump_mmapped_reg(sctx, f, R_008210_CP_CPC_STATUS);
 504         si_dump_mmapped_reg(sctx, f, R_008214_CP_CPC_BUSY_STAT);
 505         si_dump_mmapped_reg(sctx, f, R_008218_CP_CPC_STALLED_STAT1);
 506         si_dump_mmapped_reg(sctx, f, R_00821C_CP_CPF_STATUS);
 507         si_dump_mmapped_reg(sctx, f, R_008220_CP_CPF_BUSY_STAT);
 508         si_dump_mmapped_reg(sctx, f, R_008224_CP_CPF_STALLED_STAT1);
 509         fprintf(f, "\n");
 510 }
 511
 512 static void si_dump_last_ib(struct si_context *sctx, FILE *f)
 513 {
 514         int last_trace_id = -1;
 515
 516         if (!sctx->last_gfx.ib)
 517                 return;
 518
 519         if (sctx->last_trace_buf) {
 520                 /* We are expecting that the ddebug pipe has already
 521                  * waited for the context, so this buffer should be idle.
 522                  * If the GPU is hung, there is no point in waiting for it.
 523                  */
 524                 uint32_t *map = sctx->b.ws->buffer_map(sctx->last_trace_buf->buf,
 525                                                        NULL,
 526                                                        PIPE_TRANSFER_UNSYNCHRONIZED |
 527                                                        PIPE_TRANSFER_READ);
 528                 if (map)
 529                         last_trace_id = *map;
 530         }
 531
 532         if (sctx->init_config)
 533                 si_parse_ib(f, sctx->init_config->pm4, sctx->init_config->ndw,
 534                             -1, "IB2: Init config", sctx->b.chip_class);
 535
 536         if (sctx->init_config_gs_rings)
 537                 si_parse_ib(f, sctx->init_config_gs_rings->pm4,
 538                             sctx->init_config_gs_rings->ndw,
 539                             -1, "IB2: Init GS rings", sctx->b.chip_class);
 540
 541         si_parse_ib(f, sctx->last_gfx.ib, sctx->last_gfx.num_dw,
 542                     last_trace_id, "IB", sctx->b.chip_class);
 543 }
 544
 545 static const char *priority_to_string(enum radeon_bo_priority priority)
 546 {
 547 #define ITEM(x) [RADEON_PRIO_##x] = #x
 548         static const char *table[64] = {
 549                 ITEM(FENCE),
 550                 ITEM(TRACE),
 551                 ITEM(SO_FILLED_SIZE),
 552                 ITEM(QUERY),
 553                 ITEM(IB1),
 554                 ITEM(IB2),
 555                 ITEM(DRAW_INDIRECT),
 556                 ITEM(INDEX_BUFFER),
 557                 ITEM(VCE),
 558                 ITEM(UVD),
 559                 ITEM(SDMA_BUFFER),
 560                 ITEM(SDMA_TEXTURE),
 561                 ITEM(CP_DMA),
 562                 ITEM(CONST_BUFFER),
 563                 ITEM(DESCRIPTORS),
 564                 ITEM(BORDER_COLORS),
 565                 ITEM(SAMPLER_BUFFER),
 566                 ITEM(VERTEX_BUFFER),
 567                 ITEM(SHADER_RW_BUFFER),
 568                 ITEM(COMPUTE_GLOBAL),
 569                 ITEM(SAMPLER_TEXTURE),
 570                 ITEM(SHADER_RW_IMAGE),
 571                 ITEM(SAMPLER_TEXTURE_MSAA),
 572                 ITEM(COLOR_BUFFER),
 573                 ITEM(DEPTH_BUFFER),
 574                 ITEM(COLOR_BUFFER_MSAA),
 575                 ITEM(DEPTH_BUFFER_MSAA),
 576                 ITEM(CMASK),
 577                 ITEM(DCC),
 578                 ITEM(HTILE),
 579                 ITEM(SHADER_BINARY),
 580                 ITEM(SHADER_RINGS),
 581                 ITEM(SCRATCH_BUFFER),
 582         };
 583 #undef ITEM
 584
 585         assert(priority < ARRAY_SIZE(table));
 586         return table[priority];
 587 }
 588
 589 static int bo_list_compare_va(const struct radeon_bo_list_item *a,
 590                                    const struct radeon_bo_list_item *b)
 591 {
 592         return a->vm_address < b->vm_address ? -1 :
 593                a->vm_address > b->vm_address ? 1 : 0;
 594 }
 595
 596 static void si_dump_bo_list(struct si_context *sctx,
 597                             const struct radeon_saved_cs *saved, FILE *f)
 598 {
 599         unsigned i,j;
 600
 601         if (!saved->bo_list)
 602                 return;
 603
 604         /* Sort the list according to VM adddresses first. */
 605         qsort(saved->bo_list, saved->bo_count,
 606               sizeof(saved->bo_list[0]), (void*)bo_list_compare_va);
 607
 608         fprintf(f, "Buffer list (in units of pages = 4kB):\n"
 609                 COLOR_YELLOW "        Size    VM start page         "
 610                 "VM end page           Usage" COLOR_RESET "\n");
 611
 612         for (i = 0; i < saved->bo_count; i++) {
 613                 /* Note: Buffer sizes are expected to be aligned to 4k by the winsys. */
 614                 const unsigned page_size = sctx->b.screen->info.gart_page_size;
 615                 uint64_t va = saved->bo_list[i].vm_address;
 616                 uint64_t size = saved->bo_list[i].bo_size;
 617                 bool hit = false;
 618
 619                 /* If there's unused virtual memory between 2 buffers, print it. */
 620                 if (i) {
 621                         uint64_t previous_va_end = saved->bo_list[i-1].vm_address +
 622                                                    saved->bo_list[i-1].bo_size;
 623
 624                         if (va > previous_va_end) {
 625                                 fprintf(f, "  %10"PRIu64"    -- hole --\n",
 626                                         (va - previous_va_end) / page_size);
 627                         }
 628                 }
 629
 630                 /* Print the buffer. */
 631                 fprintf(f, "  %10"PRIu64"    0x%013"PRIx64"       0x%013"PRIx64"       ",
 632                         size / page_size, va / page_size, (va + size) / page_size);
 633
 634                 /* Print the usage. */
 635                 for (j = 0; j < 64; j++) {
 636                         if (!(saved->bo_list[i].priority_usage & (1llu << j)))
 637                                 continue;
 638
 639                         fprintf(f, "%s%s", !hit ? "" : ", ", priority_to_string(j));
 640                         hit = true;
 641                 }
 642                 fprintf(f, "\n");
 643         }
 644         fprintf(f, "\nNote: The holes represent memory not used by the IB.\n"
 645                    "      Other buffers can still be allocated there.\n\n");
 646 }
 647
 648 static void si_dump_framebuffer(struct si_context *sctx, FILE *f)
 649 {
 650         struct pipe_framebuffer_state *state = &sctx->framebuffer.state;
 651         struct r600_texture *rtex;
 652         int i;
 653
 654         for (i = 0; i < state->nr_cbufs; i++) {
 655                 if (!state->cbufs[i])
 656                         continue;
 657
 658                 rtex = (struct r600_texture*)state->cbufs[i]->texture;
 659                 fprintf(f, COLOR_YELLOW "Color buffer %i:" COLOR_RESET "\n", i);
 660                 r600_print_texture_info(rtex, f);
 661                 fprintf(f, "\n");
 662         }
 663
 664         if (state->zsbuf) {
 665                 rtex = (struct r600_texture*)state->zsbuf->texture;
 666                 fprintf(f, COLOR_YELLOW "Depth-stencil buffer:" COLOR_RESET "\n");
 667                 r600_print_texture_info(rtex, f);
 668                 fprintf(f, "\n");
 669         }
 670 }
 671
 672 static void si_dump_debug_state(struct pipe_context *ctx, FILE *f,
 673                                 unsigned flags)
 674 {
 675         struct si_context *sctx = (struct si_context*)ctx;
 676
 677         if (flags & PIPE_DUMP_DEVICE_STATUS_REGISTERS)
 678                 si_dump_debug_registers(sctx, f);
 679
 680         if (flags & PIPE_DUMP_CURRENT_STATES)
 681                 si_dump_framebuffer(sctx, f);
 682
 683         if (flags & PIPE_DUMP_CURRENT_SHADERS) {
 684                 si_dump_shader(sctx->screen, &sctx->vs_shader, f);
 685                 si_dump_shader(sctx->screen, &sctx->tcs_shader, f);
 686                 si_dump_shader(sctx->screen, &sctx->tes_shader, f);
 687                 si_dump_shader(sctx->screen, &sctx->gs_shader, f);
 688                 si_dump_shader(sctx->screen, &sctx->ps_shader, f);
 689         }
 690
 691         if (flags & PIPE_DUMP_LAST_COMMAND_BUFFER) {
 692                 si_dump_bo_list(sctx, &sctx->last_gfx, f);
 693                 si_dump_last_ib(sctx, f);
 694
 695                 fprintf(f, "Done.\n");
 696
 697                 /* dump only once */
 698                 radeon_clear_saved_cs(&sctx->last_gfx);
 699                 r600_resource_reference(&sctx->last_trace_buf, NULL);
 700         }
 701 }
 702
 703 static void si_dump_dma(struct si_context *sctx,
 704                         struct radeon_saved_cs *saved, FILE *f)
 705 {
 706         static const char ib_name[] = "sDMA IB";
 707         unsigned i;
 708
 709         si_dump_bo_list(sctx, saved, f);
 710
 711         fprintf(f, "------------------ %s begin ------------------\n", ib_name);
 712
 713         for (i = 0; i < saved->num_dw; ++i) {
 714                 fprintf(f, " %08x\n", saved->ib[i]);
 715         }
 716
 717         fprintf(f, "------------------- %s end -------------------\n", ib_name);
 718         fprintf(f, "\n");
 719
 720         fprintf(f, "SDMA Dump Done.\n");
 721 }
 722
 723 static bool si_vm_fault_occured(struct si_context *sctx, uint32_t *out_addr)
 724 {
 725         char line[2000];
 726         unsigned sec, usec;
 727         int progress = 0;
 728         uint64_t timestamp = 0;
 729         bool fault = false;
 730
 731         FILE *p = popen("dmesg", "r");
 732         if (!p)
 733                 return false;
 734
 735         while (fgets(line, sizeof(line), p)) {
 736                 char *msg, len;
 737
 738                 if (!line[0] || line[0] == '\n')
 739                         continue;
 740
 741                 /* Get the timestamp. */
 742                 if (sscanf(line, "[%u.%u]", &sec, &usec) != 2) {
 743                         static bool hit = false;
 744                         if (!hit) {
 745                                 fprintf(stderr, "%s: failed to parse line '%s'\n",
 746                                         __func__, line);
 747                                 hit = true;
 748                         }
 749                         continue;
 750                 }
 751                 timestamp = sec * 1000000llu + usec;
 752
 753                 /* If just updating the timestamp. */
 754                 if (!out_addr)
 755                         continue;
 756
 757                 /* Process messages only if the timestamp is newer. */
 758                 if (timestamp <= sctx->dmesg_timestamp)
 759                         continue;
 760
 761                 /* Only process the first VM fault. */
 762                 if (fault)
 763                         continue;
 764
 765                 /* Remove trailing \n */
 766                 len = strlen(line);
 767                 if (len && line[len-1] == '\n')
 768                         line[len-1] = 0;
 769
 770                 /* Get the message part. */
 771                 msg = strchr(line, ']');
 772                 if (!msg) {
 773                         assert(0);
 774                         continue;
 775                 }
 776                 msg++;
 777
 778                 switch (progress) {
 779                 case 0:
 780                         if (strstr(msg, "GPU fault detected:"))
 781                                 progress = 1;
 782                         break;
 783                 case 1:
 784                         msg = strstr(msg, "VM_CONTEXT1_PROTECTION_FAULT_ADDR");
 785                         if (msg) {
 786                                 msg = strstr(msg, "0x");
 787                                 if (msg) {
 788                                         msg += 2;
 789                                         if (sscanf(msg, "%X", out_addr) == 1)
 790                                                 fault = true;
 791                                 }
 792                         }
 793                         progress = 0;
 794                         break;
 795                 default:
 796                         progress = 0;
 797                 }
 798         }
 799         pclose(p);
 800
 801         if (timestamp > sctx->dmesg_timestamp)
 802                 sctx->dmesg_timestamp = timestamp;
 803         return fault;
 804 }
 805
 806 void si_check_vm_faults(struct r600_common_context *ctx,
 807                         struct radeon_saved_cs *saved, enum ring_type ring)
 808 {
 809         struct si_context *sctx = (struct si_context *)ctx;
 810         struct pipe_screen *screen = sctx->b.b.screen;
 811         FILE *f;
 812         uint32_t addr;
 813         char cmd_line[4096];
 814
 815         if (!si_vm_fault_occured(sctx, &addr))
 816                 return;
 817
 818         f = dd_get_debug_file(false);
 819         if (!f)
 820                 return;
 821
 822         fprintf(f, "VM fault report.\n\n");
 823         if (os_get_command_line(cmd_line, sizeof(cmd_line)))
 824                 fprintf(f, "Command: %s\n", cmd_line);
 825         fprintf(f, "Driver vendor: %s\n", screen->get_vendor(screen));
 826         fprintf(f, "Device vendor: %s\n", screen->get_device_vendor(screen));
 827         fprintf(f, "Device name: %s\n\n", screen->get_name(screen));
 828         fprintf(f, "Failing VM page: 0x%08x\n\n", addr);
 829
 830         if (sctx->apitrace_call_number)
 831                 fprintf(f, "Last apitrace call: %u\n\n",
 832                         sctx->apitrace_call_number);
 833
 834         switch (ring) {
 835         case RING_GFX:
 836                 si_dump_debug_state(&sctx->b.b, f,
 837                                     PIPE_DUMP_CURRENT_STATES |
 838                                     PIPE_DUMP_CURRENT_SHADERS |
 839                                     PIPE_DUMP_LAST_COMMAND_BUFFER);
 840                 break;
 841
 842         case RING_DMA:
 843                 si_dump_dma(sctx, saved, f);
 844                 break;
 845
 846         default:
 847                 break;
 848         }
 849
 850         fclose(f);
 851
 852         fprintf(stderr, "Detected a VM fault, exiting...\n");
 853         exit(0);
 854 }
 855
 856 void si_init_debug_functions(struct si_context *sctx)
 857 {
 858         sctx->b.b.dump_debug_state = si_dump_debug_state;
 859         sctx->b.check_vm_faults = si_check_vm_faults;
 860
 861         /* Set the initial dmesg timestamp for this context, so that
 862          * only new messages will be checked for VM faults.
 863          */
 864         if (sctx->screen->b.debug_flags & DBG_CHECK_VM)
 865                 si_vm_fault_occured(sctx, NULL);
 866 }