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 "si_shader.h"
  29 #include "sid.h"
  30 #include "sid_tables.h"
  31 #include "radeon/radeon_elf_util.h"
  32 #include "ddebug/dd_util.h"
  33 #include "util/u_memory.h"
  34
  35 DEBUG_GET_ONCE_OPTION(replace_shaders, "RADEON_REPLACE_SHADERS", NULL)
  36
  37 static void si_dump_shader(struct si_screen *sscreen,
  38                            struct si_shader_ctx_state *state, FILE *f)
  39 {
  40         struct si_shader *current = state->current;
  41
  42         if (!state->cso || !current)
  43                 return;
  44
  45         if (current->shader_log)
  46                 fwrite(current->shader_log, current->shader_log_size, 1, f);
  47         else
  48                 si_shader_dump(sscreen, state->current, NULL,
  49                                state->cso->info.processor, f);
  50 }
  51
  52 /**
  53  * Shader compiles can be overridden with arbitrary ELF objects by setting
  54  * the environment variable RADEON_REPLACE_SHADERS=num1:filename1[;num2:filename2]
  55  */
  56 bool si_replace_shader(unsigned num, struct radeon_shader_binary *binary)
  57 {
  58         const char *p = debug_get_option_replace_shaders();
  59         const char *semicolon;
  60         char *copy = NULL;
  61         FILE *f;
  62         long filesize, nread;
  63         char *buf = NULL;
  64         bool replaced = false;
  65
  66         if (!p)
  67                 return false;
  68
  69         while (*p) {
  70                 unsigned long i;
  71                 char *endp;
  72                 i = strtoul(p, &endp, 0);
  73
  74                 p = endp;
  75                 if (*p != ':') {
  76                         fprintf(stderr, "RADEON_REPLACE_SHADERS formatted badly.\n");
  77                         exit(1);
  78                 }
  79                 ++p;
  80
  81                 if (i == num)
  82                         break;
  83
  84                 p = strchr(p, ';');
  85                 if (!p)
  86                         return false;
  87                 ++p;
  88         }
  89         if (!*p)
  90                 return false;
  91
  92         semicolon = strchr(p, ';');
  93         if (semicolon) {
  94                 p = copy = strndup(p, semicolon - p);
  95                 if (!copy) {
  96                         fprintf(stderr, "out of memory\n");
  97                         return false;
  98                 }
  99         }
 100
 101         fprintf(stderr, "radeonsi: replace shader %u by %s\n", num, p);
 102
 103         f = fopen(p, "r");
 104         if (!f) {
 105                 perror("radeonsi: failed to open file");
 106                 goto out_free;
 107         }
 108
 109         if (fseek(f, 0, SEEK_END) != 0)
 110                 goto file_error;
 111
 112         filesize = ftell(f);
 113         if (filesize < 0)
 114                 goto file_error;
 115
 116         if (fseek(f, 0, SEEK_SET) != 0)
 117                 goto file_error;
 118
 119         buf = MALLOC(filesize);
 120         if (!buf) {
 121                 fprintf(stderr, "out of memory\n");
 122                 goto out_close;
 123         }
 124
 125         nread = fread(buf, 1, filesize, f);
 126         if (nread != filesize)
 127                 goto file_error;
 128
 129         radeon_elf_read(buf, filesize, binary);
 130         replaced = true;
 131
 132 out_close:
 133         fclose(f);
 134 out_free:
 135         FREE(buf);
 136         free(copy);
 137         return replaced;
 138
 139 file_error:
 140         perror("radeonsi: reading shader");
 141         goto out_close;
 142 }
 143
 144 /* Parsed IBs are difficult to read without colors. Use "less -R file" to
 145  * read them, or use "aha -b -f file" to convert them to html.
 146  */
 147 #define COLOR_RESET     "\033[0m"
 148 #define COLOR_RED       "\033[31m"
 149 #define COLOR_GREEN     "\033[1;32m"
 150 #define COLOR_YELLOW    "\033[1;33m"
 151 #define COLOR_CYAN      "\033[1;36m"
 152
 153 #define INDENT_PKT 8
 154
 155 static void print_spaces(FILE *f, unsigned num)
 156 {
 157         fprintf(f, "%*s", num, "");
 158 }
 159
 160 static void print_value(FILE *file, uint32_t value, int bits)
 161 {
 162         /* Guess if it's int or float */
 163         if (value <= (1 << 15)) {
 164                 if (value <= 9)
 165                         fprintf(file, "%u\n", value);
 166                 else
 167                         fprintf(file, "%u (0x%0*x)\n", value, bits / 4, value);
 168         } else {
 169                 float f = uif(value);
 170
 171                 if (fabs(f) < 100000 && f*10 == floor(f*10))
 172                         fprintf(file, "%.1ff (0x%0*x)\n", f, bits / 4, value);
 173                 else
 174                         /* Don't print more leading zeros than there are bits. */
 175                         fprintf(file, "0x%0*x\n", bits / 4, value);
 176         }
 177 }
 178
 179 static void print_named_value(FILE *file, const char *name, uint32_t value,
 180                               int bits)
 181 {
 182         print_spaces(file, INDENT_PKT);
 183         fprintf(file, COLOR_YELLOW "%s" COLOR_RESET " <- ", name);
 184         print_value(file, value, bits);
 185 }
 186
 187 static void si_dump_reg(FILE *file, unsigned offset, uint32_t value,
 188                         uint32_t field_mask)
 189 {
 190         int r, f;
 191
 192         for (r = 0; r < ARRAY_SIZE(sid_reg_table); r++) {
 193                 const struct si_reg *reg = &sid_reg_table[r];
 194                 const char *reg_name = sid_strings + reg->name_offset;
 195
 196                 if (reg->offset == offset) {
 197                         bool first_field = true;
 198
 199                         print_spaces(file, INDENT_PKT);
 200                         fprintf(file, COLOR_YELLOW "%s" COLOR_RESET " <- ",
 201                                 reg_name);
 202
 203                         if (!reg->num_fields) {
 204                                 print_value(file, value, 32);
 205                                 return;
 206                         }
 207
 208                         for (f = 0; f < reg->num_fields; f++) {
 209                                 const struct si_field *field = sid_fields_table + reg->fields_offset + f;
 210                                 const int *values_offsets = sid_strings_offsets + field->values_offset;
 211                                 uint32_t val = (value & field->mask) >>
 212                                                (ffs(field->mask) - 1);
 213
 214                                 if (!(field->mask & field_mask))
 215                                         continue;
 216
 217                                 /* Indent the field. */
 218                                 if (!first_field)
 219                                         print_spaces(file,
 220                                                      INDENT_PKT + strlen(reg_name) + 4);
 221
 222                                 /* Print the field. */
 223                                 fprintf(file, "%s = ", sid_strings + field->name_offset);
 224
 225                                 if (val < field->num_values && values_offsets[val] >= 0)
 226                                         fprintf(file, "%s\n", sid_strings + values_offsets[val]);
 227                                 else
 228                                         print_value(file, val,
 229                                                     util_bitcount(field->mask));
 230
 231                                 first_field = false;
 232                         }
 233                         return;
 234                 }
 235         }
 236
 237         fprintf(file, COLOR_YELLOW "0x%05x" COLOR_RESET " = 0x%08x", offset, value);
 238 }
 239
 240 static void si_parse_set_reg_packet(FILE *f, uint32_t *ib, unsigned count,
 241                                     unsigned reg_offset)
 242 {
 243         unsigned reg = (ib[1] << 2) + reg_offset;
 244         int i;
 245
 246         for (i = 0; i < count; i++)
 247                 si_dump_reg(f, reg + i*4, ib[2+i], ~0);
 248 }
 249
 250 static uint32_t *si_parse_packet3(FILE *f, uint32_t *ib, int *num_dw,
 251                                   int trace_id, enum chip_class chip_class)
 252 {
 253         unsigned count = PKT_COUNT_G(ib[0]);
 254         unsigned op = PKT3_IT_OPCODE_G(ib[0]);
 255         const char *predicate = PKT3_PREDICATE(ib[0]) ? "(predicate)" : "";
 256         int i;
 257
 258         /* Print the name first. */
 259         for (i = 0; i < ARRAY_SIZE(packet3_table); i++)
 260                 if (packet3_table[i].op == op)
 261                         break;
 262
 263         if (i < ARRAY_SIZE(packet3_table)) {
 264                 const char *name = sid_strings + packet3_table[i].name_offset;
 265
 266                 if (op == PKT3_SET_CONTEXT_REG ||
 267                     op == PKT3_SET_CONFIG_REG ||
 268                     op == PKT3_SET_UCONFIG_REG ||
 269                     op == PKT3_SET_SH_REG)
 270                         fprintf(f, COLOR_CYAN "%s%s" COLOR_CYAN ":\n",
 271                                 name, predicate);
 272                 else
 273                         fprintf(f, COLOR_GREEN "%s%s" COLOR_RESET ":\n",
 274                                 name, predicate);
 275         } else
 276                 fprintf(f, COLOR_RED "PKT3_UNKNOWN 0x%x%s" COLOR_RESET ":\n",
 277                         op, predicate);
 278
 279         /* Print the contents. */
 280         switch (op) {
 281         case PKT3_SET_CONTEXT_REG:
 282                 si_parse_set_reg_packet(f, ib, count, SI_CONTEXT_REG_OFFSET);
 283                 break;
 284         case PKT3_SET_CONFIG_REG:
 285                 si_parse_set_reg_packet(f, ib, count, SI_CONFIG_REG_OFFSET);
 286                 break;
 287         case PKT3_SET_UCONFIG_REG:
 288                 si_parse_set_reg_packet(f, ib, count, CIK_UCONFIG_REG_OFFSET);
 289                 break;
 290         case PKT3_SET_SH_REG:
 291                 si_parse_set_reg_packet(f, ib, count, SI_SH_REG_OFFSET);
 292                 break;
 293         case PKT3_ACQUIRE_MEM:
 294                 si_dump_reg(f, R_0301F0_CP_COHER_CNTL, ib[1], ~0);
 295                 si_dump_reg(f, R_0301F4_CP_COHER_SIZE, ib[2], ~0);
 296                 si_dump_reg(f, R_030230_CP_COHER_SIZE_HI, ib[3], ~0);
 297                 si_dump_reg(f, R_0301F8_CP_COHER_BASE, ib[4], ~0);
 298                 si_dump_reg(f, R_0301E4_CP_COHER_BASE_HI, ib[5], ~0);
 299                 print_named_value(f, "POLL_INTERVAL", ib[6], 16);
 300                 break;
 301         case PKT3_SURFACE_SYNC:
 302                 if (chip_class >= CIK) {
 303                         si_dump_reg(f, R_0301F0_CP_COHER_CNTL, ib[1], ~0);
 304                         si_dump_reg(f, R_0301F4_CP_COHER_SIZE, ib[2], ~0);
 305                         si_dump_reg(f, R_0301F8_CP_COHER_BASE, ib[3], ~0);
 306                 } else {
 307                         si_dump_reg(f, R_0085F0_CP_COHER_CNTL, ib[1], ~0);
 308                         si_dump_reg(f, R_0085F4_CP_COHER_SIZE, ib[2], ~0);
 309                         si_dump_reg(f, R_0085F8_CP_COHER_BASE, ib[3], ~0);
 310                 }
 311                 print_named_value(f, "POLL_INTERVAL", ib[4], 16);
 312                 break;
 313         case PKT3_EVENT_WRITE:
 314                 si_dump_reg(f, R_028A90_VGT_EVENT_INITIATOR, ib[1],
 315                             S_028A90_EVENT_TYPE(~0));
 316                 print_named_value(f, "EVENT_INDEX", (ib[1] >> 8) & 0xf, 4);
 317                 print_named_value(f, "INV_L2", (ib[1] >> 20) & 0x1, 1);
 318                 if (count > 0) {
 319                         print_named_value(f, "ADDRESS_LO", ib[2], 32);
 320                         print_named_value(f, "ADDRESS_HI", ib[3], 16);
 321                 }
 322                 break;
 323         case PKT3_DRAW_INDEX_AUTO:
 324                 si_dump_reg(f, R_030930_VGT_NUM_INDICES, ib[1], ~0);
 325                 si_dump_reg(f, R_0287F0_VGT_DRAW_INITIATOR, ib[2], ~0);
 326                 break;
 327         case PKT3_DRAW_INDEX_2:
 328                 si_dump_reg(f, R_028A78_VGT_DMA_MAX_SIZE, ib[1], ~0);
 329                 si_dump_reg(f, R_0287E8_VGT_DMA_BASE, ib[2], ~0);
 330                 si_dump_reg(f, R_0287E4_VGT_DMA_BASE_HI, ib[3], ~0);
 331                 si_dump_reg(f, R_030930_VGT_NUM_INDICES, ib[4], ~0);
 332                 si_dump_reg(f, R_0287F0_VGT_DRAW_INITIATOR, ib[5], ~0);
 333                 break;
 334         case PKT3_INDEX_TYPE:
 335                 si_dump_reg(f, R_028A7C_VGT_DMA_INDEX_TYPE, ib[1], ~0);
 336                 break;
 337         case PKT3_NUM_INSTANCES:
 338                 si_dump_reg(f, R_030934_VGT_NUM_INSTANCES, ib[1], ~0);
 339                 break;
 340         case PKT3_WRITE_DATA:
 341                 si_dump_reg(f, R_370_CONTROL, ib[1], ~0);
 342                 si_dump_reg(f, R_371_DST_ADDR_LO, ib[2], ~0);
 343                 si_dump_reg(f, R_372_DST_ADDR_HI, ib[3], ~0);
 344                 for (i = 2; i < count; i++) {
 345                         print_spaces(f, INDENT_PKT);
 346                         fprintf(f, "0x%08x\n", ib[2+i]);
 347                 }
 348                 break;
 349         case PKT3_CP_DMA:
 350                 si_dump_reg(f, R_410_CP_DMA_WORD0, ib[1], ~0);
 351                 si_dump_reg(f, R_411_CP_DMA_WORD1, ib[2], ~0);
 352                 si_dump_reg(f, R_412_CP_DMA_WORD2, ib[3], ~0);
 353                 si_dump_reg(f, R_413_CP_DMA_WORD3, ib[4], ~0);
 354                 si_dump_reg(f, R_414_COMMAND, ib[5], ~0);
 355                 break;
 356         case PKT3_DMA_DATA:
 357                 si_dump_reg(f, R_500_DMA_DATA_WORD0, ib[1], ~0);
 358                 si_dump_reg(f, R_501_SRC_ADDR_LO, ib[2], ~0);
 359                 si_dump_reg(f, R_502_SRC_ADDR_HI, ib[3], ~0);
 360                 si_dump_reg(f, R_503_DST_ADDR_LO, ib[4], ~0);
 361                 si_dump_reg(f, R_504_DST_ADDR_HI, ib[5], ~0);
 362                 si_dump_reg(f, R_414_COMMAND, ib[6], ~0);
 363                 break;
 364         case PKT3_INDIRECT_BUFFER_SI:
 365         case PKT3_INDIRECT_BUFFER_CONST:
 366         case PKT3_INDIRECT_BUFFER_CIK:
 367                 si_dump_reg(f, R_3F0_IB_BASE_LO, ib[1], ~0);
 368                 si_dump_reg(f, R_3F1_IB_BASE_HI, ib[2], ~0);
 369                 si_dump_reg(f, R_3F2_CONTROL, ib[3], ~0);
 370                 break;
 371         case PKT3_NOP:
 372                 if (ib[0] == 0xffff1000) {
 373                         count = -1; /* One dword NOP. */
 374                         break;
 375                 } else if (count == 0 && SI_IS_TRACE_POINT(ib[1])) {
 376                         unsigned packet_id = SI_GET_TRACE_POINT_ID(ib[1]);
 377
 378                         print_spaces(f, INDENT_PKT);
 379                         fprintf(f, COLOR_RED "Trace point ID: %u\n", packet_id);
 380
 381                         if (trace_id == -1)
 382                                 break; /* tracing was disabled */
 383
 384                         print_spaces(f, INDENT_PKT);
 385                         if (packet_id < trace_id)
 386                                 fprintf(f, COLOR_RED
 387                                         "This trace point was reached by the CP."
 388                                         COLOR_RESET "\n");
 389                         else if (packet_id == trace_id)
 390                                 fprintf(f, COLOR_RED
 391                                         "!!!!! This is the last trace point that "
 392                                         "was reached by the CP !!!!!"
 393                                         COLOR_RESET "\n");
 394                         else if (packet_id+1 == trace_id)
 395                                 fprintf(f, COLOR_RED
 396                                         "!!!!! This is the first trace point that "
 397                                         "was NOT been reached by the CP !!!!!"
 398                                         COLOR_RESET "\n");
 399                         else
 400                                 fprintf(f, COLOR_RED
 401                                         "!!!!! This trace point was NOT reached "
 402                                         "by the CP !!!!!"
 403                                         COLOR_RESET "\n");
 404                         break;
 405                 }
 406                 /* fall through, print all dwords */
 407         default:
 408                 for (i = 0; i < count+1; i++) {
 409                         print_spaces(f, INDENT_PKT);
 410                         fprintf(f, "0x%08x\n", ib[1+i]);
 411                 }
 412         }
 413
 414         ib += count + 2;
 415         *num_dw -= count + 2;
 416         return ib;
 417 }
 418
 419 /**
 420  * Parse and print an IB into a file.
 421  *
 422  * \param f             file
 423  * \param ib            IB
 424  * \param num_dw        size of the IB
 425  * \param chip_class    chip class
 426  * \param trace_id      the last trace ID that is known to have been reached
 427  *                      and executed by the CP, typically read from a buffer
 428  */
 429 static void si_parse_ib(FILE *f, uint32_t *ib, int num_dw, int trace_id,
 430                         const char *name, enum chip_class chip_class)
 431 {
 432         fprintf(f, "------------------ %s begin ------------------\n", name);
 433
 434         while (num_dw > 0) {
 435                 unsigned type = PKT_TYPE_G(ib[0]);
 436
 437                 switch (type) {
 438                 case 3:
 439                         ib = si_parse_packet3(f, ib, &num_dw, trace_id,
 440                                               chip_class);
 441                         break;
 442                 case 2:
 443                         /* type-2 nop */
 444                         if (ib[0] == 0x80000000) {
 445                                 fprintf(f, COLOR_GREEN "NOP (type 2)" COLOR_RESET "\n");
 446                                 ib++;
 447                                 break;
 448                         }
 449                         /* fall through */
 450                 default:
 451                         fprintf(f, "Unknown packet type %i\n", type);
 452                         return;
 453                 }
 454         }
 455
 456         fprintf(f, "------------------- %s end -------------------\n", name);
 457         if (num_dw < 0) {
 458                 printf("Packet ends after the end of IB.\n");
 459                 exit(0);
 460         }
 461         fprintf(f, "\n");
 462 }
 463
 464 static void si_dump_mmapped_reg(struct si_context *sctx, FILE *f,
 465                                 unsigned offset)
 466 {
 467         struct radeon_winsys *ws = sctx->b.ws;
 468         uint32_t value;
 469
 470         if (ws->read_registers(ws, offset, 1, &value))
 471                 si_dump_reg(f, offset, value, ~0);
 472 }
 473
 474 static void si_dump_debug_registers(struct si_context *sctx, FILE *f)
 475 {
 476         if (sctx->screen->b.info.drm_major == 2 &&
 477             sctx->screen->b.info.drm_minor < 42)
 478                 return; /* no radeon support */
 479
 480         fprintf(f, "Memory-mapped registers:\n");
 481         si_dump_mmapped_reg(sctx, f, R_008010_GRBM_STATUS);
 482
 483         /* No other registers can be read on DRM < 3.1.0. */
 484         if (sctx->screen->b.info.drm_major < 3 ||
 485             sctx->screen->b.info.drm_minor < 1) {
 486                 fprintf(f, "\n");
 487                 return;
 488         }
 489
 490         si_dump_mmapped_reg(sctx, f, R_008008_GRBM_STATUS2);
 491         si_dump_mmapped_reg(sctx, f, R_008014_GRBM_STATUS_SE0);
 492         si_dump_mmapped_reg(sctx, f, R_008018_GRBM_STATUS_SE1);
 493         si_dump_mmapped_reg(sctx, f, R_008038_GRBM_STATUS_SE2);
 494         si_dump_mmapped_reg(sctx, f, R_00803C_GRBM_STATUS_SE3);
 495         si_dump_mmapped_reg(sctx, f, R_00D034_SDMA0_STATUS_REG);
 496         si_dump_mmapped_reg(sctx, f, R_00D834_SDMA1_STATUS_REG);
 497         si_dump_mmapped_reg(sctx, f, R_000E50_SRBM_STATUS);
 498         si_dump_mmapped_reg(sctx, f, R_000E4C_SRBM_STATUS2);
 499         si_dump_mmapped_reg(sctx, f, R_000E54_SRBM_STATUS3);
 500         si_dump_mmapped_reg(sctx, f, R_008680_CP_STAT);
 501         si_dump_mmapped_reg(sctx, f, R_008674_CP_STALLED_STAT1);
 502         si_dump_mmapped_reg(sctx, f, R_008678_CP_STALLED_STAT2);
 503         si_dump_mmapped_reg(sctx, f, R_008670_CP_STALLED_STAT3);
 504         si_dump_mmapped_reg(sctx, f, R_008210_CP_CPC_STATUS);
 505         si_dump_mmapped_reg(sctx, f, R_008214_CP_CPC_BUSY_STAT);
 506         si_dump_mmapped_reg(sctx, f, R_008218_CP_CPC_STALLED_STAT1);
 507         si_dump_mmapped_reg(sctx, f, R_00821C_CP_CPF_STATUS);
 508         si_dump_mmapped_reg(sctx, f, R_008220_CP_CPF_BUSY_STAT);
 509         si_dump_mmapped_reg(sctx, f, R_008224_CP_CPF_STALLED_STAT1);
 510         fprintf(f, "\n");
 511 }
 512
 513 static void si_dump_last_ib(struct si_context *sctx, FILE *f)
 514 {
 515         int last_trace_id = -1;
 516
 517         if (!sctx->last_gfx.ib)
 518                 return;
 519
 520         if (sctx->last_trace_buf) {
 521                 /* We are expecting that the ddebug pipe has already
 522                  * waited for the context, so this buffer should be idle.
 523                  * If the GPU is hung, there is no point in waiting for it.
 524                  */
 525                 uint32_t *map = sctx->b.ws->buffer_map(sctx->last_trace_buf->buf,
 526                                                        NULL,
 527                                                        PIPE_TRANSFER_UNSYNCHRONIZED |
 528                                                        PIPE_TRANSFER_READ);
 529                 if (map)
 530                         last_trace_id = *map;
 531         }
 532
 533         if (sctx->init_config)
 534                 si_parse_ib(f, sctx->init_config->pm4, sctx->init_config->ndw,
 535                             -1, "IB2: Init config", sctx->b.chip_class);
 536
 537         if (sctx->init_config_gs_rings)
 538                 si_parse_ib(f, sctx->init_config_gs_rings->pm4,
 539                             sctx->init_config_gs_rings->ndw,
 540                             -1, "IB2: Init GS rings", sctx->b.chip_class);
 541
 542         si_parse_ib(f, sctx->last_gfx.ib, sctx->last_gfx.num_dw,
 543                     last_trace_id, "IB", sctx->b.chip_class);
 544 }
 545
 546 static const char *priority_to_string(enum radeon_bo_priority priority)
 547 {
 548 #define ITEM(x) [RADEON_PRIO_##x] = #x
 549         static const char *table[64] = {
 550                 ITEM(FENCE),
 551                 ITEM(TRACE),
 552                 ITEM(SO_FILLED_SIZE),
 553                 ITEM(QUERY),
 554                 ITEM(IB1),
 555                 ITEM(IB2),
 556                 ITEM(DRAW_INDIRECT),
 557                 ITEM(INDEX_BUFFER),
 558                 ITEM(VCE),
 559                 ITEM(UVD),
 560                 ITEM(SDMA_BUFFER),
 561                 ITEM(SDMA_TEXTURE),
 562                 ITEM(CP_DMA),
 563                 ITEM(CONST_BUFFER),
 564                 ITEM(DESCRIPTORS),
 565                 ITEM(BORDER_COLORS),
 566                 ITEM(SAMPLER_BUFFER),
 567                 ITEM(VERTEX_BUFFER),
 568                 ITEM(SHADER_RW_BUFFER),
 569                 ITEM(COMPUTE_GLOBAL),
 570                 ITEM(SAMPLER_TEXTURE),
 571                 ITEM(SHADER_RW_IMAGE),
 572                 ITEM(SAMPLER_TEXTURE_MSAA),
 573                 ITEM(COLOR_BUFFER),
 574                 ITEM(DEPTH_BUFFER),
 575                 ITEM(COLOR_BUFFER_MSAA),
 576                 ITEM(DEPTH_BUFFER_MSAA),
 577                 ITEM(CMASK),
 578                 ITEM(DCC),
 579                 ITEM(HTILE),
 580                 ITEM(SHADER_BINARY),
 581                 ITEM(SHADER_RINGS),
 582                 ITEM(SCRATCH_BUFFER),
 583         };
 584 #undef ITEM
 585
 586         assert(priority < ARRAY_SIZE(table));
 587         return table[priority];
 588 }
 589
 590 static int bo_list_compare_va(const struct radeon_bo_list_item *a,
 591                                    const struct radeon_bo_list_item *b)
 592 {
 593         return a->vm_address < b->vm_address ? -1 :
 594                a->vm_address > b->vm_address ? 1 : 0;
 595 }
 596
 597 static void si_dump_bo_list(struct si_context *sctx,
 598                             const struct radeon_saved_cs *saved, FILE *f)
 599 {
 600         unsigned i,j;
 601
 602         if (!saved->bo_list)
 603                 return;
 604
 605         /* Sort the list according to VM adddresses first. */
 606         qsort(saved->bo_list, saved->bo_count,
 607               sizeof(saved->bo_list[0]), (void*)bo_list_compare_va);
 608
 609         fprintf(f, "Buffer list (in units of pages = 4kB):\n"
 610                 COLOR_YELLOW "        Size    VM start page         "
 611                 "VM end page           Usage" COLOR_RESET "\n");
 612
 613         for (i = 0; i < saved->bo_count; i++) {
 614                 /* Note: Buffer sizes are expected to be aligned to 4k by the winsys. */
 615                 const unsigned page_size = sctx->b.screen->info.gart_page_size;
 616                 uint64_t va = saved->bo_list[i].vm_address;
 617                 uint64_t size = saved->bo_list[i].bo_size;
 618                 bool hit = false;
 619
 620                 /* If there's unused virtual memory between 2 buffers, print it. */
 621                 if (i) {
 622                         uint64_t previous_va_end = saved->bo_list[i-1].vm_address +
 623                                                    saved->bo_list[i-1].bo_size;
 624
 625                         if (va > previous_va_end) {
 626                                 fprintf(f, "  %10"PRIu64"    -- hole --\n",
 627                                         (va - previous_va_end) / page_size);
 628                         }
 629                 }
 630
 631                 /* Print the buffer. */
 632                 fprintf(f, "  %10"PRIu64"    0x%013"PRIx64"       0x%013"PRIx64"       ",
 633                         size / page_size, va / page_size, (va + size) / page_size);
 634
 635                 /* Print the usage. */
 636                 for (j = 0; j < 64; j++) {
 637                         if (!(saved->bo_list[i].priority_usage & (1llu << j)))
 638                                 continue;
 639
 640                         fprintf(f, "%s%s", !hit ? "" : ", ", priority_to_string(j));
 641                         hit = true;
 642                 }
 643                 fprintf(f, "\n");
 644         }
 645         fprintf(f, "\nNote: The holes represent memory not used by the IB.\n"
 646                    "      Other buffers can still be allocated there.\n\n");
 647 }
 648
 649 static void si_dump_framebuffer(struct si_context *sctx, FILE *f)
 650 {
 651         struct pipe_framebuffer_state *state = &sctx->framebuffer.state;
 652         struct r600_texture *rtex;
 653         int i;
 654
 655         for (i = 0; i < state->nr_cbufs; i++) {
 656                 if (!state->cbufs[i])
 657                         continue;
 658
 659                 rtex = (struct r600_texture*)state->cbufs[i]->texture;
 660                 fprintf(f, COLOR_YELLOW "Color buffer %i:" COLOR_RESET "\n", i);
 661                 r600_print_texture_info(rtex, f);
 662                 fprintf(f, "\n");
 663         }
 664
 665         if (state->zsbuf) {
 666                 rtex = (struct r600_texture*)state->zsbuf->texture;
 667                 fprintf(f, COLOR_YELLOW "Depth-stencil buffer:" COLOR_RESET "\n");
 668                 r600_print_texture_info(rtex, f);
 669                 fprintf(f, "\n");
 670         }
 671 }
 672
 673 static void si_dump_debug_state(struct pipe_context *ctx, FILE *f,
 674                                 unsigned flags)
 675 {
 676         struct si_context *sctx = (struct si_context*)ctx;
 677
 678         if (flags & PIPE_DUMP_DEVICE_STATUS_REGISTERS)
 679                 si_dump_debug_registers(sctx, f);
 680
 681         if (flags & PIPE_DUMP_CURRENT_STATES)
 682                 si_dump_framebuffer(sctx, f);
 683
 684         if (flags & PIPE_DUMP_CURRENT_SHADERS) {
 685                 si_dump_shader(sctx->screen, &sctx->vs_shader, f);
 686                 si_dump_shader(sctx->screen, &sctx->tcs_shader, f);
 687                 si_dump_shader(sctx->screen, &sctx->tes_shader, f);
 688                 si_dump_shader(sctx->screen, &sctx->gs_shader, f);
 689                 si_dump_shader(sctx->screen, &sctx->ps_shader, f);
 690         }
 691
 692         if (flags & PIPE_DUMP_LAST_COMMAND_BUFFER) {
 693                 si_dump_bo_list(sctx, &sctx->last_gfx, f);
 694                 si_dump_last_ib(sctx, f);
 695
 696                 fprintf(f, "Done.\n");
 697
 698                 /* dump only once */
 699                 radeon_clear_saved_cs(&sctx->last_gfx);
 700                 r600_resource_reference(&sctx->last_trace_buf, NULL);
 701         }
 702 }
 703
 704 static void si_dump_dma(struct si_context *sctx,
 705                         struct radeon_saved_cs *saved, FILE *f)
 706 {
 707         static const char ib_name[] = "sDMA IB";
 708         unsigned i;
 709
 710         si_dump_bo_list(sctx, saved, f);
 711
 712         fprintf(f, "------------------ %s begin ------------------\n", ib_name);
 713
 714         for (i = 0; i < saved->num_dw; ++i) {
 715                 fprintf(f, " %08x\n", saved->ib[i]);
 716         }
 717
 718         fprintf(f, "------------------- %s end -------------------\n", ib_name);
 719         fprintf(f, "\n");
 720
 721         fprintf(f, "SDMA Dump Done.\n");
 722 }
 723
 724 static bool si_vm_fault_occured(struct si_context *sctx, uint32_t *out_addr)
 725 {
 726         char line[2000];
 727         unsigned sec, usec;
 728         int progress = 0;
 729         uint64_t timestamp = 0;
 730         bool fault = false;
 731
 732         FILE *p = popen("dmesg", "r");
 733         if (!p)
 734                 return false;
 735
 736         while (fgets(line, sizeof(line), p)) {
 737                 char *msg, len;
 738
 739                 if (!line[0] || line[0] == '\n')
 740                         continue;
 741
 742                 /* Get the timestamp. */
 743                 if (sscanf(line, "[%u.%u]", &sec, &usec) != 2) {
 744                         static bool hit = false;
 745                         if (!hit) {
 746                                 fprintf(stderr, "%s: failed to parse line '%s'\n",
 747                                         __func__, line);
 748                                 hit = true;
 749                         }
 750                         continue;
 751                 }
 752                 timestamp = sec * 1000000llu + usec;
 753
 754                 /* If just updating the timestamp. */
 755                 if (!out_addr)
 756                         continue;
 757
 758                 /* Process messages only if the timestamp is newer. */
 759                 if (timestamp <= sctx->dmesg_timestamp)
 760                         continue;
 761
 762                 /* Only process the first VM fault. */
 763                 if (fault)
 764                         continue;
 765
 766                 /* Remove trailing \n */
 767                 len = strlen(line);
 768                 if (len && line[len-1] == '\n')
 769                         line[len-1] = 0;
 770
 771                 /* Get the message part. */
 772                 msg = strchr(line, ']');
 773                 if (!msg) {
 774                         assert(0);
 775                         continue;
 776                 }
 777                 msg++;
 778
 779                 switch (progress) {
 780                 case 0:
 781                         if (strstr(msg, "GPU fault detected:"))
 782                                 progress = 1;
 783                         break;
 784                 case 1:
 785                         msg = strstr(msg, "VM_CONTEXT1_PROTECTION_FAULT_ADDR");
 786                         if (msg) {
 787                                 msg = strstr(msg, "0x");
 788                                 if (msg) {
 789                                         msg += 2;
 790                                         if (sscanf(msg, "%X", out_addr) == 1)
 791                                                 fault = true;
 792                                 }
 793                         }
 794                         progress = 0;
 795                         break;
 796                 default:
 797                         progress = 0;
 798                 }
 799         }
 800         pclose(p);
 801
 802         if (timestamp > sctx->dmesg_timestamp)
 803                 sctx->dmesg_timestamp = timestamp;
 804         return fault;
 805 }
 806
 807 void si_check_vm_faults(struct r600_common_context *ctx,
 808                         struct radeon_saved_cs *saved, enum ring_type ring)
 809 {
 810         struct si_context *sctx = (struct si_context *)ctx;
 811         struct pipe_screen *screen = sctx->b.b.screen;
 812         FILE *f;
 813         uint32_t addr;
 814         char cmd_line[4096];
 815
 816         if (!si_vm_fault_occured(sctx, &addr))
 817                 return;
 818
 819         f = dd_get_debug_file(false);
 820         if (!f)
 821                 return;
 822
 823         fprintf(f, "VM fault report.\n\n");
 824         if (os_get_command_line(cmd_line, sizeof(cmd_line)))
 825                 fprintf(f, "Command: %s\n", cmd_line);
 826         fprintf(f, "Driver vendor: %s\n", screen->get_vendor(screen));
 827         fprintf(f, "Device vendor: %s\n", screen->get_device_vendor(screen));
 828         fprintf(f, "Device name: %s\n\n", screen->get_name(screen));
 829         fprintf(f, "Failing VM page: 0x%08x\n\n", addr);
 830
 831         if (sctx->apitrace_call_number)
 832                 fprintf(f, "Last apitrace call: %u\n\n",
 833                         sctx->apitrace_call_number);
 834
 835         switch (ring) {
 836         case RING_GFX:
 837                 si_dump_debug_state(&sctx->b.b, f,
 838                                     PIPE_DUMP_CURRENT_STATES |
 839                                     PIPE_DUMP_CURRENT_SHADERS |
 840                                     PIPE_DUMP_LAST_COMMAND_BUFFER);
 841                 break;
 842
 843         case RING_DMA:
 844                 si_dump_dma(sctx, saved, f);
 845                 break;
 846
 847         default:
 848                 break;
 849         }
 850
 851         fclose(f);
 852
 853         fprintf(stderr, "Detected a VM fault, exiting...\n");
 854         exit(0);
 855 }
 856
 857 void si_init_debug_functions(struct si_context *sctx)
 858 {
 859         sctx->b.b.dump_debug_state = si_dump_debug_state;
 860         sctx->b.check_vm_faults = si_check_vm_faults;
 861
 862         /* Set the initial dmesg timestamp for this context, so that
 863          * only new messages will be checked for VM faults.
 864          */
 865         if (sctx->screen->b.debug_flags & DBG_CHECK_VM)
 866                 si_vm_fault_occured(sctx, NULL);
 867 }