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