1 /* 2 * Copyright (c) 2008 Intel Corporation 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice (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 NONINFRINGEMENT. IN NO EVENT SHALL 18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 21 * IN THE SOFTWARE. 22 * 23 * Authors: 24 * Eric Anholt <eric@anholt.net> 25 * Keith Packard <keithp@keithp.com> 26 * Mika Kuoppala <mika.kuoppala@intel.com> 27 * 28 */ 29 30 #include <generated/utsrelease.h> 31 #include <linux/stop_machine.h> 32 #include <linux/zlib.h> 33 #include "i915_drv.h" 34 35 static const char *engine_str(int engine) 36 { 37 switch (engine) { 38 case RCS: return "render"; 39 case VCS: return "bsd"; 40 case BCS: return "blt"; 41 case VECS: return "vebox"; 42 case VCS2: return "bsd2"; 43 default: return ""; 44 } 45 } 46 47 static const char *tiling_flag(int tiling) 48 { 49 switch (tiling) { 50 default: 51 case I915_TILING_NONE: return ""; 52 case I915_TILING_X: return " X"; 53 case I915_TILING_Y: return " Y"; 54 } 55 } 56 57 static const char *dirty_flag(int dirty) 58 { 59 return dirty ? " dirty" : ""; 60 } 61 62 static const char *purgeable_flag(int purgeable) 63 { 64 return purgeable ? " purgeable" : ""; 65 } 66 67 static bool __i915_error_ok(struct drm_i915_error_state_buf *e) 68 { 69 70 if (!e->err && WARN(e->bytes > (e->size - 1), "overflow")) { 71 e->err = -ENOSPC; 72 return false; 73 } 74 75 if (e->bytes == e->size - 1 || e->err) 76 return false; 77 78 return true; 79 } 80 81 static bool __i915_error_seek(struct drm_i915_error_state_buf *e, 82 unsigned len) 83 { 84 if (e->pos + len <= e->start) { 85 e->pos += len; 86 return false; 87 } 88 89 /* First vsnprintf needs to fit in its entirety for memmove */ 90 if (len >= e->size) { 91 e->err = -EIO; 92 return false; 93 } 94 95 return true; 96 } 97 98 static void __i915_error_advance(struct drm_i915_error_state_buf *e, 99 unsigned len) 100 { 101 /* If this is first printf in this window, adjust it so that 102 * start position matches start of the buffer 103 */ 104 105 if (e->pos < e->start) { 106 const size_t off = e->start - e->pos; 107 108 /* Should not happen but be paranoid */ 109 if (off > len || e->bytes) { 110 e->err = -EIO; 111 return; 112 } 113 114 memmove(e->buf, e->buf + off, len - off); 115 e->bytes = len - off; 116 e->pos = e->start; 117 return; 118 } 119 120 e->bytes += len; 121 e->pos += len; 122 } 123 124 static void i915_error_vprintf(struct drm_i915_error_state_buf *e, 125 const char *f, va_list args) 126 { 127 unsigned len; 128 129 if (!__i915_error_ok(e)) 130 return; 131 132 /* Seek the first printf which is hits start position */ 133 if (e->pos < e->start) { 134 va_list tmp; 135 136 va_copy(tmp, args); 137 len = vsnprintf(NULL, 0, f, tmp); 138 va_end(tmp); 139 140 if (!__i915_error_seek(e, len)) 141 return; 142 } 143 144 len = vsnprintf(e->buf + e->bytes, e->size - e->bytes, f, args); 145 if (len >= e->size - e->bytes) 146 len = e->size - e->bytes - 1; 147 148 __i915_error_advance(e, len); 149 } 150 151 static void i915_error_puts(struct drm_i915_error_state_buf *e, 152 const char *str) 153 { 154 unsigned len; 155 156 if (!__i915_error_ok(e)) 157 return; 158 159 len = strlen(str); 160 161 /* Seek the first printf which is hits start position */ 162 if (e->pos < e->start) { 163 if (!__i915_error_seek(e, len)) 164 return; 165 } 166 167 if (len >= e->size - e->bytes) 168 len = e->size - e->bytes - 1; 169 memcpy(e->buf + e->bytes, str, len); 170 171 __i915_error_advance(e, len); 172 } 173 174 #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__) 175 #define err_puts(e, s) i915_error_puts(e, s) 176 177 #ifdef CONFIG_DRM_I915_COMPRESS_ERROR 178 179 static bool compress_init(struct z_stream_s *zstream) 180 { 181 memset(zstream, 0, sizeof(*zstream)); 182 183 zstream->workspace = 184 kmalloc(zlib_deflate_workspacesize(MAX_WBITS, MAX_MEM_LEVEL), 185 GFP_ATOMIC | __GFP_NOWARN); 186 if (!zstream->workspace) 187 return false; 188 189 if (zlib_deflateInit(zstream, Z_DEFAULT_COMPRESSION) != Z_OK) { 190 kfree(zstream->workspace); 191 return false; 192 } 193 194 return true; 195 } 196 197 static int compress_page(struct z_stream_s *zstream, 198 void *src, 199 struct drm_i915_error_object *dst) 200 { 201 zstream->next_in = src; 202 zstream->avail_in = PAGE_SIZE; 203 204 do { 205 if (zstream->avail_out == 0) { 206 unsigned long page; 207 208 page = __get_free_page(GFP_ATOMIC | __GFP_NOWARN); 209 if (!page) 210 return -ENOMEM; 211 212 dst->pages[dst->page_count++] = (void *)page; 213 214 zstream->next_out = (void *)page; 215 zstream->avail_out = PAGE_SIZE; 216 } 217 218 if (zlib_deflate(zstream, Z_SYNC_FLUSH) != Z_OK) 219 return -EIO; 220 } while (zstream->avail_in); 221 222 /* Fallback to uncompressed if we increase size? */ 223 if (0 && zstream->total_out > zstream->total_in) 224 return -E2BIG; 225 226 return 0; 227 } 228 229 static void compress_fini(struct z_stream_s *zstream, 230 struct drm_i915_error_object *dst) 231 { 232 if (dst) { 233 zlib_deflate(zstream, Z_FINISH); 234 dst->unused = zstream->avail_out; 235 } 236 237 zlib_deflateEnd(zstream); 238 kfree(zstream->workspace); 239 } 240 241 static void err_compression_marker(struct drm_i915_error_state_buf *m) 242 { 243 err_puts(m, ":"); 244 } 245 246 #else 247 248 static bool compress_init(struct z_stream_s *zstream) 249 { 250 return true; 251 } 252 253 static int compress_page(struct z_stream_s *zstream, 254 void *src, 255 struct drm_i915_error_object *dst) 256 { 257 unsigned long page; 258 259 page = __get_free_page(GFP_ATOMIC | __GFP_NOWARN); 260 if (!page) 261 return -ENOMEM; 262 263 dst->pages[dst->page_count++] = 264 memcpy((void *)page, src, PAGE_SIZE); 265 266 return 0; 267 } 268 269 static void compress_fini(struct z_stream_s *zstream, 270 struct drm_i915_error_object *dst) 271 { 272 } 273 274 static void err_compression_marker(struct drm_i915_error_state_buf *m) 275 { 276 err_puts(m, "~"); 277 } 278 279 #endif 280 281 static void print_error_buffers(struct drm_i915_error_state_buf *m, 282 const char *name, 283 struct drm_i915_error_buffer *err, 284 int count) 285 { 286 int i; 287 288 err_printf(m, "%s [%d]:\n", name, count); 289 290 while (count--) { 291 err_printf(m, " %08x_%08x %8u %02x %02x [ ", 292 upper_32_bits(err->gtt_offset), 293 lower_32_bits(err->gtt_offset), 294 err->size, 295 err->read_domains, 296 err->write_domain); 297 for (i = 0; i < I915_NUM_ENGINES; i++) 298 err_printf(m, "%02x ", err->rseqno[i]); 299 300 err_printf(m, "] %02x", err->wseqno); 301 err_puts(m, tiling_flag(err->tiling)); 302 err_puts(m, dirty_flag(err->dirty)); 303 err_puts(m, purgeable_flag(err->purgeable)); 304 err_puts(m, err->userptr ? " userptr" : ""); 305 err_puts(m, err->engine != -1 ? " " : ""); 306 err_puts(m, engine_str(err->engine)); 307 err_puts(m, i915_cache_level_str(m->i915, err->cache_level)); 308 309 if (err->name) 310 err_printf(m, " (name: %d)", err->name); 311 if (err->fence_reg != I915_FENCE_REG_NONE) 312 err_printf(m, " (fence: %d)", err->fence_reg); 313 314 err_puts(m, "\n"); 315 err++; 316 } 317 } 318 319 static const char *hangcheck_action_to_str(enum intel_engine_hangcheck_action a) 320 { 321 switch (a) { 322 case HANGCHECK_IDLE: 323 return "idle"; 324 case HANGCHECK_WAIT: 325 return "wait"; 326 case HANGCHECK_ACTIVE: 327 return "active"; 328 case HANGCHECK_KICK: 329 return "kick"; 330 case HANGCHECK_HUNG: 331 return "hung"; 332 } 333 334 return "unknown"; 335 } 336 337 static void error_print_instdone(struct drm_i915_error_state_buf *m, 338 struct drm_i915_error_engine *ee) 339 { 340 int slice; 341 int subslice; 342 343 err_printf(m, " INSTDONE: 0x%08x\n", 344 ee->instdone.instdone); 345 346 if (ee->engine_id != RCS || INTEL_GEN(m->i915) <= 3) 347 return; 348 349 err_printf(m, " SC_INSTDONE: 0x%08x\n", 350 ee->instdone.slice_common); 351 352 if (INTEL_GEN(m->i915) <= 6) 353 return; 354 355 for_each_instdone_slice_subslice(m->i915, slice, subslice) 356 err_printf(m, " SAMPLER_INSTDONE[%d][%d]: 0x%08x\n", 357 slice, subslice, 358 ee->instdone.sampler[slice][subslice]); 359 360 for_each_instdone_slice_subslice(m->i915, slice, subslice) 361 err_printf(m, " ROW_INSTDONE[%d][%d]: 0x%08x\n", 362 slice, subslice, 363 ee->instdone.row[slice][subslice]); 364 } 365 366 static void error_print_request(struct drm_i915_error_state_buf *m, 367 const char *prefix, 368 struct drm_i915_error_request *erq) 369 { 370 if (!erq->seqno) 371 return; 372 373 err_printf(m, "%s pid %d, seqno %8x:%08x, emitted %dms ago, head %08x, tail %08x\n", 374 prefix, erq->pid, 375 erq->context, erq->seqno, 376 jiffies_to_msecs(jiffies - erq->jiffies), 377 erq->head, erq->tail); 378 } 379 380 static void error_print_engine(struct drm_i915_error_state_buf *m, 381 struct drm_i915_error_engine *ee) 382 { 383 err_printf(m, "%s command stream:\n", engine_str(ee->engine_id)); 384 err_printf(m, " START: 0x%08x\n", ee->start); 385 err_printf(m, " HEAD: 0x%08x [0x%08x]\n", ee->head, ee->rq_head); 386 err_printf(m, " TAIL: 0x%08x [0x%08x, 0x%08x]\n", 387 ee->tail, ee->rq_post, ee->rq_tail); 388 err_printf(m, " CTL: 0x%08x\n", ee->ctl); 389 err_printf(m, " MODE: 0x%08x\n", ee->mode); 390 err_printf(m, " HWS: 0x%08x\n", ee->hws); 391 err_printf(m, " ACTHD: 0x%08x %08x\n", 392 (u32)(ee->acthd>>32), (u32)ee->acthd); 393 err_printf(m, " IPEIR: 0x%08x\n", ee->ipeir); 394 err_printf(m, " IPEHR: 0x%08x\n", ee->ipehr); 395 396 error_print_instdone(m, ee); 397 398 if (ee->batchbuffer) { 399 u64 start = ee->batchbuffer->gtt_offset; 400 u64 end = start + ee->batchbuffer->gtt_size; 401 402 err_printf(m, " batch: [0x%08x_%08x, 0x%08x_%08x]\n", 403 upper_32_bits(start), lower_32_bits(start), 404 upper_32_bits(end), lower_32_bits(end)); 405 } 406 if (INTEL_GEN(m->i915) >= 4) { 407 err_printf(m, " BBADDR: 0x%08x_%08x\n", 408 (u32)(ee->bbaddr>>32), (u32)ee->bbaddr); 409 err_printf(m, " BB_STATE: 0x%08x\n", ee->bbstate); 410 err_printf(m, " INSTPS: 0x%08x\n", ee->instps); 411 } 412 err_printf(m, " INSTPM: 0x%08x\n", ee->instpm); 413 err_printf(m, " FADDR: 0x%08x %08x\n", upper_32_bits(ee->faddr), 414 lower_32_bits(ee->faddr)); 415 if (INTEL_GEN(m->i915) >= 6) { 416 err_printf(m, " RC PSMI: 0x%08x\n", ee->rc_psmi); 417 err_printf(m, " FAULT_REG: 0x%08x\n", ee->fault_reg); 418 err_printf(m, " SYNC_0: 0x%08x\n", 419 ee->semaphore_mboxes[0]); 420 err_printf(m, " SYNC_1: 0x%08x\n", 421 ee->semaphore_mboxes[1]); 422 if (HAS_VEBOX(m->i915)) 423 err_printf(m, " SYNC_2: 0x%08x\n", 424 ee->semaphore_mboxes[2]); 425 } 426 if (USES_PPGTT(m->i915)) { 427 err_printf(m, " GFX_MODE: 0x%08x\n", ee->vm_info.gfx_mode); 428 429 if (INTEL_GEN(m->i915) >= 8) { 430 int i; 431 for (i = 0; i < 4; i++) 432 err_printf(m, " PDP%d: 0x%016llx\n", 433 i, ee->vm_info.pdp[i]); 434 } else { 435 err_printf(m, " PP_DIR_BASE: 0x%08x\n", 436 ee->vm_info.pp_dir_base); 437 } 438 } 439 err_printf(m, " seqno: 0x%08x\n", ee->seqno); 440 err_printf(m, " last_seqno: 0x%08x\n", ee->last_seqno); 441 err_printf(m, " waiting: %s\n", yesno(ee->waiting)); 442 err_printf(m, " ring->head: 0x%08x\n", ee->cpu_ring_head); 443 err_printf(m, " ring->tail: 0x%08x\n", ee->cpu_ring_tail); 444 err_printf(m, " hangcheck: %s [%d]\n", 445 hangcheck_action_to_str(ee->hangcheck_action), 446 ee->hangcheck_score); 447 error_print_request(m, " ELSP[0]: ", &ee->execlist[0]); 448 error_print_request(m, " ELSP[1]: ", &ee->execlist[1]); 449 } 450 451 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...) 452 { 453 va_list args; 454 455 va_start(args, f); 456 i915_error_vprintf(e, f, args); 457 va_end(args); 458 } 459 460 static int 461 ascii85_encode_len(int len) 462 { 463 return DIV_ROUND_UP(len, 4); 464 } 465 466 static bool 467 ascii85_encode(u32 in, char *out) 468 { 469 int i; 470 471 if (in == 0) 472 return false; 473 474 out[5] = '\0'; 475 for (i = 5; i--; ) { 476 out[i] = '!' + in % 85; 477 in /= 85; 478 } 479 480 return true; 481 } 482 483 static void print_error_obj(struct drm_i915_error_state_buf *m, 484 struct intel_engine_cs *engine, 485 const char *name, 486 struct drm_i915_error_object *obj) 487 { 488 char out[6]; 489 int page; 490 491 if (!obj) 492 return; 493 494 if (name) { 495 err_printf(m, "%s --- %s = 0x%08x %08x\n", 496 engine ? engine->name : "global", name, 497 upper_32_bits(obj->gtt_offset), 498 lower_32_bits(obj->gtt_offset)); 499 } 500 501 err_compression_marker(m); 502 for (page = 0; page < obj->page_count; page++) { 503 int i, len; 504 505 len = PAGE_SIZE; 506 if (page == obj->page_count - 1) 507 len -= obj->unused; 508 len = ascii85_encode_len(len); 509 510 for (i = 0; i < len; i++) { 511 if (ascii85_encode(obj->pages[page][i], out)) 512 err_puts(m, out); 513 else 514 err_puts(m, "z"); 515 } 516 } 517 err_puts(m, "\n"); 518 } 519 520 static void err_print_capabilities(struct drm_i915_error_state_buf *m, 521 const struct intel_device_info *info) 522 { 523 #define PRINT_FLAG(x) err_printf(m, #x ": %s\n", yesno(info->x)) 524 DEV_INFO_FOR_EACH_FLAG(PRINT_FLAG); 525 #undef PRINT_FLAG 526 } 527 528 int i915_error_state_to_str(struct drm_i915_error_state_buf *m, 529 const struct i915_error_state_file_priv *error_priv) 530 { 531 struct drm_i915_private *dev_priv = to_i915(error_priv->dev); 532 struct pci_dev *pdev = dev_priv->drm.pdev; 533 struct drm_i915_error_state *error = error_priv->error; 534 struct drm_i915_error_object *obj; 535 int max_hangcheck_score; 536 int i, j; 537 538 if (!error) { 539 err_printf(m, "no error state collected\n"); 540 goto out; 541 } 542 543 err_printf(m, "%s\n", error->error_msg); 544 err_printf(m, "Kernel: " UTS_RELEASE "\n"); 545 err_printf(m, "Time: %ld s %ld us\n", 546 error->time.tv_sec, error->time.tv_usec); 547 err_printf(m, "Boottime: %ld s %ld us\n", 548 error->boottime.tv_sec, error->boottime.tv_usec); 549 err_printf(m, "Uptime: %ld s %ld us\n", 550 error->uptime.tv_sec, error->uptime.tv_usec); 551 err_print_capabilities(m, &error->device_info); 552 max_hangcheck_score = 0; 553 for (i = 0; i < ARRAY_SIZE(error->engine); i++) { 554 if (error->engine[i].hangcheck_score > max_hangcheck_score) 555 max_hangcheck_score = error->engine[i].hangcheck_score; 556 } 557 for (i = 0; i < ARRAY_SIZE(error->engine); i++) { 558 if (error->engine[i].hangcheck_score == max_hangcheck_score && 559 error->engine[i].pid != -1) { 560 err_printf(m, "Active process (on ring %s): %s [%d]\n", 561 engine_str(i), 562 error->engine[i].comm, 563 error->engine[i].pid); 564 } 565 } 566 err_printf(m, "Reset count: %u\n", error->reset_count); 567 err_printf(m, "Suspend count: %u\n", error->suspend_count); 568 err_printf(m, "PCI ID: 0x%04x\n", pdev->device); 569 err_printf(m, "PCI Revision: 0x%02x\n", pdev->revision); 570 err_printf(m, "PCI Subsystem: %04x:%04x\n", 571 pdev->subsystem_vendor, 572 pdev->subsystem_device); 573 err_printf(m, "IOMMU enabled?: %d\n", error->iommu); 574 575 if (HAS_CSR(dev_priv)) { 576 struct intel_csr *csr = &dev_priv->csr; 577 578 err_printf(m, "DMC loaded: %s\n", 579 yesno(csr->dmc_payload != NULL)); 580 err_printf(m, "DMC fw version: %d.%d\n", 581 CSR_VERSION_MAJOR(csr->version), 582 CSR_VERSION_MINOR(csr->version)); 583 } 584 585 err_printf(m, "EIR: 0x%08x\n", error->eir); 586 err_printf(m, "IER: 0x%08x\n", error->ier); 587 if (INTEL_GEN(dev_priv) >= 8) { 588 for (i = 0; i < 4; i++) 589 err_printf(m, "GTIER gt %d: 0x%08x\n", i, 590 error->gtier[i]); 591 } else if (HAS_PCH_SPLIT(dev_priv) || IS_VALLEYVIEW(dev_priv)) 592 err_printf(m, "GTIER: 0x%08x\n", error->gtier[0]); 593 err_printf(m, "PGTBL_ER: 0x%08x\n", error->pgtbl_er); 594 err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake); 595 err_printf(m, "DERRMR: 0x%08x\n", error->derrmr); 596 err_printf(m, "CCID: 0x%08x\n", error->ccid); 597 err_printf(m, "Missed interrupts: 0x%08lx\n", dev_priv->gpu_error.missed_irq_rings); 598 599 for (i = 0; i < dev_priv->num_fence_regs; i++) 600 err_printf(m, " fence[%d] = %08llx\n", i, error->fence[i]); 601 602 if (INTEL_GEN(dev_priv) >= 6) { 603 err_printf(m, "ERROR: 0x%08x\n", error->error); 604 605 if (INTEL_GEN(dev_priv) >= 8) 606 err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n", 607 error->fault_data1, error->fault_data0); 608 609 err_printf(m, "DONE_REG: 0x%08x\n", error->done_reg); 610 } 611 612 if (IS_GEN7(dev_priv)) 613 err_printf(m, "ERR_INT: 0x%08x\n", error->err_int); 614 615 for (i = 0; i < ARRAY_SIZE(error->engine); i++) { 616 if (error->engine[i].engine_id != -1) 617 error_print_engine(m, &error->engine[i]); 618 } 619 620 for (i = 0; i < ARRAY_SIZE(error->active_vm); i++) { 621 char buf[128]; 622 int len, first = 1; 623 624 if (!error->active_vm[i]) 625 break; 626 627 len = scnprintf(buf, sizeof(buf), "Active ("); 628 for (j = 0; j < ARRAY_SIZE(error->engine); j++) { 629 if (error->engine[j].vm != error->active_vm[i]) 630 continue; 631 632 len += scnprintf(buf + len, sizeof(buf), "%s%s", 633 first ? "" : ", ", 634 dev_priv->engine[j]->name); 635 first = 0; 636 } 637 scnprintf(buf + len, sizeof(buf), ")"); 638 print_error_buffers(m, buf, 639 error->active_bo[i], 640 error->active_bo_count[i]); 641 } 642 643 print_error_buffers(m, "Pinned (global)", 644 error->pinned_bo, 645 error->pinned_bo_count); 646 647 for (i = 0; i < ARRAY_SIZE(error->engine); i++) { 648 struct drm_i915_error_engine *ee = &error->engine[i]; 649 650 obj = ee->batchbuffer; 651 if (obj) { 652 err_puts(m, dev_priv->engine[i]->name); 653 if (ee->pid != -1) 654 err_printf(m, " (submitted by %s [%d])", 655 ee->comm, 656 ee->pid); 657 err_printf(m, " --- gtt_offset = 0x%08x %08x\n", 658 upper_32_bits(obj->gtt_offset), 659 lower_32_bits(obj->gtt_offset)); 660 print_error_obj(m, dev_priv->engine[i], NULL, obj); 661 } 662 663 if (ee->num_requests) { 664 err_printf(m, "%s --- %d requests\n", 665 dev_priv->engine[i]->name, 666 ee->num_requests); 667 for (j = 0; j < ee->num_requests; j++) 668 error_print_request(m, " ", &ee->requests[j]); 669 } 670 671 if (IS_ERR(ee->waiters)) { 672 err_printf(m, "%s --- ? waiters [unable to acquire spinlock]\n", 673 dev_priv->engine[i]->name); 674 } else if (ee->num_waiters) { 675 err_printf(m, "%s --- %d waiters\n", 676 dev_priv->engine[i]->name, 677 ee->num_waiters); 678 for (j = 0; j < ee->num_waiters; j++) { 679 err_printf(m, " seqno 0x%08x for %s [%d]\n", 680 ee->waiters[j].seqno, 681 ee->waiters[j].comm, 682 ee->waiters[j].pid); 683 } 684 } 685 686 print_error_obj(m, dev_priv->engine[i], 687 "ringbuffer", ee->ringbuffer); 688 689 print_error_obj(m, dev_priv->engine[i], 690 "HW Status", ee->hws_page); 691 692 print_error_obj(m, dev_priv->engine[i], 693 "HW context", ee->ctx); 694 695 print_error_obj(m, dev_priv->engine[i], 696 "WA context", ee->wa_ctx); 697 698 print_error_obj(m, dev_priv->engine[i], 699 "WA batchbuffer", ee->wa_batchbuffer); 700 } 701 702 print_error_obj(m, NULL, "Semaphores", error->semaphore); 703 704 print_error_obj(m, NULL, "GuC log buffer", error->guc_log); 705 706 if (error->overlay) 707 intel_overlay_print_error_state(m, error->overlay); 708 709 if (error->display) 710 intel_display_print_error_state(m, dev_priv, error->display); 711 712 out: 713 if (m->bytes == 0 && m->err) 714 return m->err; 715 716 return 0; 717 } 718 719 int i915_error_state_buf_init(struct drm_i915_error_state_buf *ebuf, 720 struct drm_i915_private *i915, 721 size_t count, loff_t pos) 722 { 723 memset(ebuf, 0, sizeof(*ebuf)); 724 ebuf->i915 = i915; 725 726 /* We need to have enough room to store any i915_error_state printf 727 * so that we can move it to start position. 728 */ 729 ebuf->size = count + 1 > PAGE_SIZE ? count + 1 : PAGE_SIZE; 730 ebuf->buf = kmalloc(ebuf->size, 731 GFP_TEMPORARY | __GFP_NORETRY | __GFP_NOWARN); 732 733 if (ebuf->buf == NULL) { 734 ebuf->size = PAGE_SIZE; 735 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY); 736 } 737 738 if (ebuf->buf == NULL) { 739 ebuf->size = 128; 740 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY); 741 } 742 743 if (ebuf->buf == NULL) 744 return -ENOMEM; 745 746 ebuf->start = pos; 747 748 return 0; 749 } 750 751 static void i915_error_object_free(struct drm_i915_error_object *obj) 752 { 753 int page; 754 755 if (obj == NULL) 756 return; 757 758 for (page = 0; page < obj->page_count; page++) 759 free_page((unsigned long)obj->pages[page]); 760 761 kfree(obj); 762 } 763 764 static void i915_error_state_free(struct kref *error_ref) 765 { 766 struct drm_i915_error_state *error = container_of(error_ref, 767 typeof(*error), ref); 768 int i; 769 770 for (i = 0; i < ARRAY_SIZE(error->engine); i++) { 771 struct drm_i915_error_engine *ee = &error->engine[i]; 772 773 i915_error_object_free(ee->batchbuffer); 774 i915_error_object_free(ee->wa_batchbuffer); 775 i915_error_object_free(ee->ringbuffer); 776 i915_error_object_free(ee->hws_page); 777 i915_error_object_free(ee->ctx); 778 i915_error_object_free(ee->wa_ctx); 779 780 kfree(ee->requests); 781 if (!IS_ERR_OR_NULL(ee->waiters)) 782 kfree(ee->waiters); 783 } 784 785 i915_error_object_free(error->semaphore); 786 i915_error_object_free(error->guc_log); 787 788 for (i = 0; i < ARRAY_SIZE(error->active_bo); i++) 789 kfree(error->active_bo[i]); 790 kfree(error->pinned_bo); 791 792 kfree(error->overlay); 793 kfree(error->display); 794 kfree(error); 795 } 796 797 static struct drm_i915_error_object * 798 i915_error_object_create(struct drm_i915_private *i915, 799 struct i915_vma *vma) 800 { 801 struct i915_ggtt *ggtt = &i915->ggtt; 802 const u64 slot = ggtt->error_capture.start; 803 struct drm_i915_error_object *dst; 804 struct z_stream_s zstream; 805 unsigned long num_pages; 806 struct sgt_iter iter; 807 dma_addr_t dma; 808 809 if (!vma) 810 return NULL; 811 812 num_pages = min_t(u64, vma->size, vma->obj->base.size) >> PAGE_SHIFT; 813 num_pages = DIV_ROUND_UP(10 * num_pages, 8); /* worstcase zlib growth */ 814 dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), 815 GFP_ATOMIC | __GFP_NOWARN); 816 if (!dst) 817 return NULL; 818 819 dst->gtt_offset = vma->node.start; 820 dst->gtt_size = vma->node.size; 821 dst->page_count = 0; 822 dst->unused = 0; 823 824 if (!compress_init(&zstream)) { 825 kfree(dst); 826 return NULL; 827 } 828 829 for_each_sgt_dma(dma, iter, vma->pages) { 830 void __iomem *s; 831 int ret; 832 833 ggtt->base.insert_page(&ggtt->base, dma, slot, 834 I915_CACHE_NONE, 0); 835 836 s = io_mapping_map_atomic_wc(&ggtt->mappable, slot); 837 ret = compress_page(&zstream, (void __force *)s, dst); 838 io_mapping_unmap_atomic(s); 839 840 if (ret) 841 goto unwind; 842 } 843 goto out; 844 845 unwind: 846 while (dst->page_count--) 847 free_page((unsigned long)dst->pages[dst->page_count]); 848 kfree(dst); 849 dst = NULL; 850 851 out: 852 compress_fini(&zstream, dst); 853 ggtt->base.clear_range(&ggtt->base, slot, PAGE_SIZE); 854 return dst; 855 } 856 857 /* The error capture is special as tries to run underneath the normal 858 * locking rules - so we use the raw version of the i915_gem_active lookup. 859 */ 860 static inline uint32_t 861 __active_get_seqno(struct i915_gem_active *active) 862 { 863 struct drm_i915_gem_request *request; 864 865 request = __i915_gem_active_peek(active); 866 return request ? request->global_seqno : 0; 867 } 868 869 static inline int 870 __active_get_engine_id(struct i915_gem_active *active) 871 { 872 struct drm_i915_gem_request *request; 873 874 request = __i915_gem_active_peek(active); 875 return request ? request->engine->id : -1; 876 } 877 878 static void capture_bo(struct drm_i915_error_buffer *err, 879 struct i915_vma *vma) 880 { 881 struct drm_i915_gem_object *obj = vma->obj; 882 int i; 883 884 err->size = obj->base.size; 885 err->name = obj->base.name; 886 887 for (i = 0; i < I915_NUM_ENGINES; i++) 888 err->rseqno[i] = __active_get_seqno(&vma->last_read[i]); 889 err->wseqno = __active_get_seqno(&obj->frontbuffer_write); 890 err->engine = __active_get_engine_id(&obj->frontbuffer_write); 891 892 err->gtt_offset = vma->node.start; 893 err->read_domains = obj->base.read_domains; 894 err->write_domain = obj->base.write_domain; 895 err->fence_reg = vma->fence ? vma->fence->id : -1; 896 err->tiling = i915_gem_object_get_tiling(obj); 897 err->dirty = obj->mm.dirty; 898 err->purgeable = obj->mm.madv != I915_MADV_WILLNEED; 899 err->userptr = obj->userptr.mm != NULL; 900 err->cache_level = obj->cache_level; 901 } 902 903 static u32 capture_error_bo(struct drm_i915_error_buffer *err, 904 int count, struct list_head *head, 905 bool pinned_only) 906 { 907 struct i915_vma *vma; 908 int i = 0; 909 910 list_for_each_entry(vma, head, vm_link) { 911 if (pinned_only && !i915_vma_is_pinned(vma)) 912 continue; 913 914 capture_bo(err++, vma); 915 if (++i == count) 916 break; 917 } 918 919 return i; 920 } 921 922 /* Generate a semi-unique error code. The code is not meant to have meaning, The 923 * code's only purpose is to try to prevent false duplicated bug reports by 924 * grossly estimating a GPU error state. 925 * 926 * TODO Ideally, hashing the batchbuffer would be a very nice way to determine 927 * the hang if we could strip the GTT offset information from it. 928 * 929 * It's only a small step better than a random number in its current form. 930 */ 931 static uint32_t i915_error_generate_code(struct drm_i915_private *dev_priv, 932 struct drm_i915_error_state *error, 933 int *engine_id) 934 { 935 uint32_t error_code = 0; 936 int i; 937 938 /* IPEHR would be an ideal way to detect errors, as it's the gross 939 * measure of "the command that hung." However, has some very common 940 * synchronization commands which almost always appear in the case 941 * strictly a client bug. Use instdone to differentiate those some. 942 */ 943 for (i = 0; i < I915_NUM_ENGINES; i++) { 944 if (error->engine[i].hangcheck_action == HANGCHECK_HUNG) { 945 if (engine_id) 946 *engine_id = i; 947 948 return error->engine[i].ipehr ^ 949 error->engine[i].instdone.instdone; 950 } 951 } 952 953 return error_code; 954 } 955 956 static void i915_gem_record_fences(struct drm_i915_private *dev_priv, 957 struct drm_i915_error_state *error) 958 { 959 int i; 960 961 if (IS_GEN3(dev_priv) || IS_GEN2(dev_priv)) { 962 for (i = 0; i < dev_priv->num_fence_regs; i++) 963 error->fence[i] = I915_READ(FENCE_REG(i)); 964 } else if (IS_GEN5(dev_priv) || IS_GEN4(dev_priv)) { 965 for (i = 0; i < dev_priv->num_fence_regs; i++) 966 error->fence[i] = I915_READ64(FENCE_REG_965_LO(i)); 967 } else if (INTEL_GEN(dev_priv) >= 6) { 968 for (i = 0; i < dev_priv->num_fence_regs; i++) 969 error->fence[i] = I915_READ64(FENCE_REG_GEN6_LO(i)); 970 } 971 } 972 973 static inline u32 974 gen8_engine_sync_index(struct intel_engine_cs *engine, 975 struct intel_engine_cs *other) 976 { 977 int idx; 978 979 /* 980 * rcs -> 0 = vcs, 1 = bcs, 2 = vecs, 3 = vcs2; 981 * vcs -> 0 = bcs, 1 = vecs, 2 = vcs2, 3 = rcs; 982 * bcs -> 0 = vecs, 1 = vcs2. 2 = rcs, 3 = vcs; 983 * vecs -> 0 = vcs2, 1 = rcs, 2 = vcs, 3 = bcs; 984 * vcs2 -> 0 = rcs, 1 = vcs, 2 = bcs, 3 = vecs; 985 */ 986 987 idx = (other - engine) - 1; 988 if (idx < 0) 989 idx += I915_NUM_ENGINES; 990 991 return idx; 992 } 993 994 static void gen8_record_semaphore_state(struct drm_i915_error_state *error, 995 struct intel_engine_cs *engine, 996 struct drm_i915_error_engine *ee) 997 { 998 struct drm_i915_private *dev_priv = engine->i915; 999 struct intel_engine_cs *to; 1000 enum intel_engine_id id; 1001 1002 if (!error->semaphore) 1003 return; 1004 1005 for_each_engine(to, dev_priv, id) { 1006 int idx; 1007 u16 signal_offset; 1008 u32 *tmp; 1009 1010 if (engine == to) 1011 continue; 1012 1013 signal_offset = 1014 (GEN8_SIGNAL_OFFSET(engine, id) & (PAGE_SIZE - 1)) / 4; 1015 tmp = error->semaphore->pages[0]; 1016 idx = gen8_engine_sync_index(engine, to); 1017 1018 ee->semaphore_mboxes[idx] = tmp[signal_offset]; 1019 } 1020 } 1021 1022 static void gen6_record_semaphore_state(struct intel_engine_cs *engine, 1023 struct drm_i915_error_engine *ee) 1024 { 1025 struct drm_i915_private *dev_priv = engine->i915; 1026 1027 ee->semaphore_mboxes[0] = I915_READ(RING_SYNC_0(engine->mmio_base)); 1028 ee->semaphore_mboxes[1] = I915_READ(RING_SYNC_1(engine->mmio_base)); 1029 if (HAS_VEBOX(dev_priv)) 1030 ee->semaphore_mboxes[2] = 1031 I915_READ(RING_SYNC_2(engine->mmio_base)); 1032 } 1033 1034 static void error_record_engine_waiters(struct intel_engine_cs *engine, 1035 struct drm_i915_error_engine *ee) 1036 { 1037 struct intel_breadcrumbs *b = &engine->breadcrumbs; 1038 struct drm_i915_error_waiter *waiter; 1039 struct rb_node *rb; 1040 int count; 1041 1042 ee->num_waiters = 0; 1043 ee->waiters = NULL; 1044 1045 if (RB_EMPTY_ROOT(&b->waiters)) 1046 return; 1047 1048 if (!spin_trylock_irq(&b->lock)) { 1049 ee->waiters = ERR_PTR(-EDEADLK); 1050 return; 1051 } 1052 1053 count = 0; 1054 for (rb = rb_first(&b->waiters); rb != NULL; rb = rb_next(rb)) 1055 count++; 1056 spin_unlock_irq(&b->lock); 1057 1058 waiter = NULL; 1059 if (count) 1060 waiter = kmalloc_array(count, 1061 sizeof(struct drm_i915_error_waiter), 1062 GFP_ATOMIC); 1063 if (!waiter) 1064 return; 1065 1066 if (!spin_trylock_irq(&b->lock)) { 1067 kfree(waiter); 1068 ee->waiters = ERR_PTR(-EDEADLK); 1069 return; 1070 } 1071 1072 ee->waiters = waiter; 1073 for (rb = rb_first(&b->waiters); rb; rb = rb_next(rb)) { 1074 struct intel_wait *w = container_of(rb, typeof(*w), node); 1075 1076 strcpy(waiter->comm, w->tsk->comm); 1077 waiter->pid = w->tsk->pid; 1078 waiter->seqno = w->seqno; 1079 waiter++; 1080 1081 if (++ee->num_waiters == count) 1082 break; 1083 } 1084 spin_unlock_irq(&b->lock); 1085 } 1086 1087 static void error_record_engine_registers(struct drm_i915_error_state *error, 1088 struct intel_engine_cs *engine, 1089 struct drm_i915_error_engine *ee) 1090 { 1091 struct drm_i915_private *dev_priv = engine->i915; 1092 1093 if (INTEL_GEN(dev_priv) >= 6) { 1094 ee->rc_psmi = I915_READ(RING_PSMI_CTL(engine->mmio_base)); 1095 ee->fault_reg = I915_READ(RING_FAULT_REG(engine)); 1096 if (INTEL_GEN(dev_priv) >= 8) 1097 gen8_record_semaphore_state(error, engine, ee); 1098 else 1099 gen6_record_semaphore_state(engine, ee); 1100 } 1101 1102 if (INTEL_GEN(dev_priv) >= 4) { 1103 ee->faddr = I915_READ(RING_DMA_FADD(engine->mmio_base)); 1104 ee->ipeir = I915_READ(RING_IPEIR(engine->mmio_base)); 1105 ee->ipehr = I915_READ(RING_IPEHR(engine->mmio_base)); 1106 ee->instps = I915_READ(RING_INSTPS(engine->mmio_base)); 1107 ee->bbaddr = I915_READ(RING_BBADDR(engine->mmio_base)); 1108 if (INTEL_GEN(dev_priv) >= 8) { 1109 ee->faddr |= (u64) I915_READ(RING_DMA_FADD_UDW(engine->mmio_base)) << 32; 1110 ee->bbaddr |= (u64) I915_READ(RING_BBADDR_UDW(engine->mmio_base)) << 32; 1111 } 1112 ee->bbstate = I915_READ(RING_BBSTATE(engine->mmio_base)); 1113 } else { 1114 ee->faddr = I915_READ(DMA_FADD_I8XX); 1115 ee->ipeir = I915_READ(IPEIR); 1116 ee->ipehr = I915_READ(IPEHR); 1117 } 1118 1119 intel_engine_get_instdone(engine, &ee->instdone); 1120 1121 ee->waiting = intel_engine_has_waiter(engine); 1122 ee->instpm = I915_READ(RING_INSTPM(engine->mmio_base)); 1123 ee->acthd = intel_engine_get_active_head(engine); 1124 ee->seqno = intel_engine_get_seqno(engine); 1125 ee->last_seqno = intel_engine_last_submit(engine); 1126 ee->start = I915_READ_START(engine); 1127 ee->head = I915_READ_HEAD(engine); 1128 ee->tail = I915_READ_TAIL(engine); 1129 ee->ctl = I915_READ_CTL(engine); 1130 if (INTEL_GEN(dev_priv) > 2) 1131 ee->mode = I915_READ_MODE(engine); 1132 1133 if (!HWS_NEEDS_PHYSICAL(dev_priv)) { 1134 i915_reg_t mmio; 1135 1136 if (IS_GEN7(dev_priv)) { 1137 switch (engine->id) { 1138 default: 1139 case RCS: 1140 mmio = RENDER_HWS_PGA_GEN7; 1141 break; 1142 case BCS: 1143 mmio = BLT_HWS_PGA_GEN7; 1144 break; 1145 case VCS: 1146 mmio = BSD_HWS_PGA_GEN7; 1147 break; 1148 case VECS: 1149 mmio = VEBOX_HWS_PGA_GEN7; 1150 break; 1151 } 1152 } else if (IS_GEN6(engine->i915)) { 1153 mmio = RING_HWS_PGA_GEN6(engine->mmio_base); 1154 } else { 1155 /* XXX: gen8 returns to sanity */ 1156 mmio = RING_HWS_PGA(engine->mmio_base); 1157 } 1158 1159 ee->hws = I915_READ(mmio); 1160 } 1161 1162 ee->hangcheck_score = engine->hangcheck.score; 1163 ee->hangcheck_action = engine->hangcheck.action; 1164 1165 if (USES_PPGTT(dev_priv)) { 1166 int i; 1167 1168 ee->vm_info.gfx_mode = I915_READ(RING_MODE_GEN7(engine)); 1169 1170 if (IS_GEN6(dev_priv)) 1171 ee->vm_info.pp_dir_base = 1172 I915_READ(RING_PP_DIR_BASE_READ(engine)); 1173 else if (IS_GEN7(dev_priv)) 1174 ee->vm_info.pp_dir_base = 1175 I915_READ(RING_PP_DIR_BASE(engine)); 1176 else if (INTEL_GEN(dev_priv) >= 8) 1177 for (i = 0; i < 4; i++) { 1178 ee->vm_info.pdp[i] = 1179 I915_READ(GEN8_RING_PDP_UDW(engine, i)); 1180 ee->vm_info.pdp[i] <<= 32; 1181 ee->vm_info.pdp[i] |= 1182 I915_READ(GEN8_RING_PDP_LDW(engine, i)); 1183 } 1184 } 1185 } 1186 1187 static void record_request(struct drm_i915_gem_request *request, 1188 struct drm_i915_error_request *erq) 1189 { 1190 erq->context = request->ctx->hw_id; 1191 erq->seqno = request->global_seqno; 1192 erq->jiffies = request->emitted_jiffies; 1193 erq->head = request->head; 1194 erq->tail = request->tail; 1195 1196 rcu_read_lock(); 1197 erq->pid = request->ctx->pid ? pid_nr(request->ctx->pid) : 0; 1198 rcu_read_unlock(); 1199 } 1200 1201 static void engine_record_requests(struct intel_engine_cs *engine, 1202 struct drm_i915_gem_request *first, 1203 struct drm_i915_error_engine *ee) 1204 { 1205 struct drm_i915_gem_request *request; 1206 int count; 1207 1208 count = 0; 1209 request = first; 1210 list_for_each_entry_from(request, &engine->timeline->requests, link) 1211 count++; 1212 if (!count) 1213 return; 1214 1215 ee->requests = kcalloc(count, sizeof(*ee->requests), GFP_ATOMIC); 1216 if (!ee->requests) 1217 return; 1218 1219 ee->num_requests = count; 1220 1221 count = 0; 1222 request = first; 1223 list_for_each_entry_from(request, &engine->timeline->requests, link) { 1224 if (count >= ee->num_requests) { 1225 /* 1226 * If the ring request list was changed in 1227 * between the point where the error request 1228 * list was created and dimensioned and this 1229 * point then just exit early to avoid crashes. 1230 * 1231 * We don't need to communicate that the 1232 * request list changed state during error 1233 * state capture and that the error state is 1234 * slightly incorrect as a consequence since we 1235 * are typically only interested in the request 1236 * list state at the point of error state 1237 * capture, not in any changes happening during 1238 * the capture. 1239 */ 1240 break; 1241 } 1242 1243 record_request(request, &ee->requests[count++]); 1244 } 1245 ee->num_requests = count; 1246 } 1247 1248 static void error_record_engine_execlists(struct intel_engine_cs *engine, 1249 struct drm_i915_error_engine *ee) 1250 { 1251 unsigned int n; 1252 1253 for (n = 0; n < ARRAY_SIZE(engine->execlist_port); n++) 1254 if (engine->execlist_port[n].request) 1255 record_request(engine->execlist_port[n].request, 1256 &ee->execlist[n]); 1257 } 1258 1259 static void i915_gem_record_rings(struct drm_i915_private *dev_priv, 1260 struct drm_i915_error_state *error) 1261 { 1262 struct i915_ggtt *ggtt = &dev_priv->ggtt; 1263 int i; 1264 1265 error->semaphore = 1266 i915_error_object_create(dev_priv, dev_priv->semaphore); 1267 1268 for (i = 0; i < I915_NUM_ENGINES; i++) { 1269 struct intel_engine_cs *engine = dev_priv->engine[i]; 1270 struct drm_i915_error_engine *ee = &error->engine[i]; 1271 struct drm_i915_gem_request *request; 1272 1273 ee->pid = -1; 1274 ee->engine_id = -1; 1275 1276 if (!engine) 1277 continue; 1278 1279 ee->engine_id = i; 1280 1281 error_record_engine_registers(error, engine, ee); 1282 error_record_engine_waiters(engine, ee); 1283 error_record_engine_execlists(engine, ee); 1284 1285 request = i915_gem_find_active_request(engine); 1286 if (request) { 1287 struct intel_ring *ring; 1288 struct pid *pid; 1289 1290 ee->vm = request->ctx->ppgtt ? 1291 &request->ctx->ppgtt->base : &ggtt->base; 1292 1293 /* We need to copy these to an anonymous buffer 1294 * as the simplest method to avoid being overwritten 1295 * by userspace. 1296 */ 1297 ee->batchbuffer = 1298 i915_error_object_create(dev_priv, 1299 request->batch); 1300 1301 if (HAS_BROKEN_CS_TLB(dev_priv)) 1302 ee->wa_batchbuffer = 1303 i915_error_object_create(dev_priv, 1304 engine->scratch); 1305 1306 ee->ctx = 1307 i915_error_object_create(dev_priv, 1308 request->ctx->engine[i].state); 1309 1310 pid = request->ctx->pid; 1311 if (pid) { 1312 struct task_struct *task; 1313 1314 rcu_read_lock(); 1315 task = pid_task(pid, PIDTYPE_PID); 1316 if (task) { 1317 strcpy(ee->comm, task->comm); 1318 ee->pid = task->pid; 1319 } 1320 rcu_read_unlock(); 1321 } 1322 1323 error->simulated |= 1324 request->ctx->flags & CONTEXT_NO_ERROR_CAPTURE; 1325 1326 ee->rq_head = request->head; 1327 ee->rq_post = request->postfix; 1328 ee->rq_tail = request->tail; 1329 1330 ring = request->ring; 1331 ee->cpu_ring_head = ring->head; 1332 ee->cpu_ring_tail = ring->tail; 1333 ee->ringbuffer = 1334 i915_error_object_create(dev_priv, ring->vma); 1335 1336 engine_record_requests(engine, request, ee); 1337 } 1338 1339 ee->hws_page = 1340 i915_error_object_create(dev_priv, 1341 engine->status_page.vma); 1342 1343 ee->wa_ctx = 1344 i915_error_object_create(dev_priv, engine->wa_ctx.vma); 1345 } 1346 } 1347 1348 static void i915_gem_capture_vm(struct drm_i915_private *dev_priv, 1349 struct drm_i915_error_state *error, 1350 struct i915_address_space *vm, 1351 int idx) 1352 { 1353 struct drm_i915_error_buffer *active_bo; 1354 struct i915_vma *vma; 1355 int count; 1356 1357 count = 0; 1358 list_for_each_entry(vma, &vm->active_list, vm_link) 1359 count++; 1360 1361 active_bo = NULL; 1362 if (count) 1363 active_bo = kcalloc(count, sizeof(*active_bo), GFP_ATOMIC); 1364 if (active_bo) 1365 count = capture_error_bo(active_bo, count, &vm->active_list, false); 1366 else 1367 count = 0; 1368 1369 error->active_vm[idx] = vm; 1370 error->active_bo[idx] = active_bo; 1371 error->active_bo_count[idx] = count; 1372 } 1373 1374 static void i915_capture_active_buffers(struct drm_i915_private *dev_priv, 1375 struct drm_i915_error_state *error) 1376 { 1377 int cnt = 0, i, j; 1378 1379 BUILD_BUG_ON(ARRAY_SIZE(error->engine) > ARRAY_SIZE(error->active_bo)); 1380 BUILD_BUG_ON(ARRAY_SIZE(error->active_bo) != ARRAY_SIZE(error->active_vm)); 1381 BUILD_BUG_ON(ARRAY_SIZE(error->active_bo) != ARRAY_SIZE(error->active_bo_count)); 1382 1383 /* Scan each engine looking for unique active contexts/vm */ 1384 for (i = 0; i < ARRAY_SIZE(error->engine); i++) { 1385 struct drm_i915_error_engine *ee = &error->engine[i]; 1386 bool found; 1387 1388 if (!ee->vm) 1389 continue; 1390 1391 found = false; 1392 for (j = 0; j < i && !found; j++) 1393 found = error->engine[j].vm == ee->vm; 1394 if (!found) 1395 i915_gem_capture_vm(dev_priv, error, ee->vm, cnt++); 1396 } 1397 } 1398 1399 static void i915_capture_pinned_buffers(struct drm_i915_private *dev_priv, 1400 struct drm_i915_error_state *error) 1401 { 1402 struct i915_address_space *vm = &dev_priv->ggtt.base; 1403 struct drm_i915_error_buffer *bo; 1404 struct i915_vma *vma; 1405 int count_inactive, count_active; 1406 1407 count_inactive = 0; 1408 list_for_each_entry(vma, &vm->active_list, vm_link) 1409 count_inactive++; 1410 1411 count_active = 0; 1412 list_for_each_entry(vma, &vm->inactive_list, vm_link) 1413 count_active++; 1414 1415 bo = NULL; 1416 if (count_inactive + count_active) 1417 bo = kcalloc(count_inactive + count_active, 1418 sizeof(*bo), GFP_ATOMIC); 1419 if (!bo) 1420 return; 1421 1422 count_inactive = capture_error_bo(bo, count_inactive, 1423 &vm->active_list, true); 1424 count_active = capture_error_bo(bo + count_inactive, count_active, 1425 &vm->inactive_list, true); 1426 error->pinned_bo_count = count_inactive + count_active; 1427 error->pinned_bo = bo; 1428 } 1429 1430 static void i915_gem_capture_guc_log_buffer(struct drm_i915_private *dev_priv, 1431 struct drm_i915_error_state *error) 1432 { 1433 /* Capturing log buf contents won't be useful if logging was disabled */ 1434 if (!dev_priv->guc.log.vma || (i915.guc_log_level < 0)) 1435 return; 1436 1437 error->guc_log = i915_error_object_create(dev_priv, 1438 dev_priv->guc.log.vma); 1439 } 1440 1441 /* Capture all registers which don't fit into another category. */ 1442 static void i915_capture_reg_state(struct drm_i915_private *dev_priv, 1443 struct drm_i915_error_state *error) 1444 { 1445 int i; 1446 1447 /* General organization 1448 * 1. Registers specific to a single generation 1449 * 2. Registers which belong to multiple generations 1450 * 3. Feature specific registers. 1451 * 4. Everything else 1452 * Please try to follow the order. 1453 */ 1454 1455 /* 1: Registers specific to a single generation */ 1456 if (IS_VALLEYVIEW(dev_priv)) { 1457 error->gtier[0] = I915_READ(GTIER); 1458 error->ier = I915_READ(VLV_IER); 1459 error->forcewake = I915_READ_FW(FORCEWAKE_VLV); 1460 } 1461 1462 if (IS_GEN7(dev_priv)) 1463 error->err_int = I915_READ(GEN7_ERR_INT); 1464 1465 if (INTEL_GEN(dev_priv) >= 8) { 1466 error->fault_data0 = I915_READ(GEN8_FAULT_TLB_DATA0); 1467 error->fault_data1 = I915_READ(GEN8_FAULT_TLB_DATA1); 1468 } 1469 1470 if (IS_GEN6(dev_priv)) { 1471 error->forcewake = I915_READ_FW(FORCEWAKE); 1472 error->gab_ctl = I915_READ(GAB_CTL); 1473 error->gfx_mode = I915_READ(GFX_MODE); 1474 } 1475 1476 /* 2: Registers which belong to multiple generations */ 1477 if (INTEL_GEN(dev_priv) >= 7) 1478 error->forcewake = I915_READ_FW(FORCEWAKE_MT); 1479 1480 if (INTEL_GEN(dev_priv) >= 6) { 1481 error->derrmr = I915_READ(DERRMR); 1482 error->error = I915_READ(ERROR_GEN6); 1483 error->done_reg = I915_READ(DONE_REG); 1484 } 1485 1486 /* 3: Feature specific registers */ 1487 if (IS_GEN6(dev_priv) || IS_GEN7(dev_priv)) { 1488 error->gam_ecochk = I915_READ(GAM_ECOCHK); 1489 error->gac_eco = I915_READ(GAC_ECO_BITS); 1490 } 1491 1492 /* 4: Everything else */ 1493 if (HAS_HW_CONTEXTS(dev_priv)) 1494 error->ccid = I915_READ(CCID); 1495 1496 if (INTEL_GEN(dev_priv) >= 8) { 1497 error->ier = I915_READ(GEN8_DE_MISC_IER); 1498 for (i = 0; i < 4; i++) 1499 error->gtier[i] = I915_READ(GEN8_GT_IER(i)); 1500 } else if (HAS_PCH_SPLIT(dev_priv)) { 1501 error->ier = I915_READ(DEIER); 1502 error->gtier[0] = I915_READ(GTIER); 1503 } else if (IS_GEN2(dev_priv)) { 1504 error->ier = I915_READ16(IER); 1505 } else if (!IS_VALLEYVIEW(dev_priv)) { 1506 error->ier = I915_READ(IER); 1507 } 1508 error->eir = I915_READ(EIR); 1509 error->pgtbl_er = I915_READ(PGTBL_ER); 1510 } 1511 1512 static void i915_error_capture_msg(struct drm_i915_private *dev_priv, 1513 struct drm_i915_error_state *error, 1514 u32 engine_mask, 1515 const char *error_msg) 1516 { 1517 u32 ecode; 1518 int engine_id = -1, len; 1519 1520 ecode = i915_error_generate_code(dev_priv, error, &engine_id); 1521 1522 len = scnprintf(error->error_msg, sizeof(error->error_msg), 1523 "GPU HANG: ecode %d:%d:0x%08x", 1524 INTEL_GEN(dev_priv), engine_id, ecode); 1525 1526 if (engine_id != -1 && error->engine[engine_id].pid != -1) 1527 len += scnprintf(error->error_msg + len, 1528 sizeof(error->error_msg) - len, 1529 ", in %s [%d]", 1530 error->engine[engine_id].comm, 1531 error->engine[engine_id].pid); 1532 1533 scnprintf(error->error_msg + len, sizeof(error->error_msg) - len, 1534 ", reason: %s, action: %s", 1535 error_msg, 1536 engine_mask ? "reset" : "continue"); 1537 } 1538 1539 static void i915_capture_gen_state(struct drm_i915_private *dev_priv, 1540 struct drm_i915_error_state *error) 1541 { 1542 error->iommu = -1; 1543 #ifdef CONFIG_INTEL_IOMMU 1544 error->iommu = intel_iommu_gfx_mapped; 1545 #endif 1546 error->reset_count = i915_reset_count(&dev_priv->gpu_error); 1547 error->suspend_count = dev_priv->suspend_count; 1548 1549 memcpy(&error->device_info, 1550 INTEL_INFO(dev_priv), 1551 sizeof(error->device_info)); 1552 } 1553 1554 static int capture(void *data) 1555 { 1556 struct drm_i915_error_state *error = data; 1557 1558 i915_capture_gen_state(error->i915, error); 1559 i915_capture_reg_state(error->i915, error); 1560 i915_gem_record_fences(error->i915, error); 1561 i915_gem_record_rings(error->i915, error); 1562 i915_capture_active_buffers(error->i915, error); 1563 i915_capture_pinned_buffers(error->i915, error); 1564 i915_gem_capture_guc_log_buffer(error->i915, error); 1565 1566 do_gettimeofday(&error->time); 1567 error->boottime = ktime_to_timeval(ktime_get_boottime()); 1568 error->uptime = 1569 ktime_to_timeval(ktime_sub(ktime_get(), 1570 error->i915->gt.last_init_time)); 1571 1572 error->overlay = intel_overlay_capture_error_state(error->i915); 1573 error->display = intel_display_capture_error_state(error->i915); 1574 1575 return 0; 1576 } 1577 1578 #define DAY_AS_SECONDS(x) (24 * 60 * 60 * (x)) 1579 1580 /** 1581 * i915_capture_error_state - capture an error record for later analysis 1582 * @dev: drm device 1583 * 1584 * Should be called when an error is detected (either a hang or an error 1585 * interrupt) to capture error state from the time of the error. Fills 1586 * out a structure which becomes available in debugfs for user level tools 1587 * to pick up. 1588 */ 1589 void i915_capture_error_state(struct drm_i915_private *dev_priv, 1590 u32 engine_mask, 1591 const char *error_msg) 1592 { 1593 static bool warned; 1594 struct drm_i915_error_state *error; 1595 unsigned long flags; 1596 1597 if (!i915.error_capture) 1598 return; 1599 1600 if (READ_ONCE(dev_priv->gpu_error.first_error)) 1601 return; 1602 1603 /* Account for pipe specific data like PIPE*STAT */ 1604 error = kzalloc(sizeof(*error), GFP_ATOMIC); 1605 if (!error) { 1606 DRM_DEBUG_DRIVER("out of memory, not capturing error state\n"); 1607 return; 1608 } 1609 1610 kref_init(&error->ref); 1611 error->i915 = dev_priv; 1612 1613 stop_machine(capture, error, NULL); 1614 1615 i915_error_capture_msg(dev_priv, error, engine_mask, error_msg); 1616 DRM_INFO("%s\n", error->error_msg); 1617 1618 if (!error->simulated) { 1619 spin_lock_irqsave(&dev_priv->gpu_error.lock, flags); 1620 if (!dev_priv->gpu_error.first_error) { 1621 dev_priv->gpu_error.first_error = error; 1622 error = NULL; 1623 } 1624 spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags); 1625 } 1626 1627 if (error) { 1628 i915_error_state_free(&error->ref); 1629 return; 1630 } 1631 1632 if (!warned && 1633 ktime_get_real_seconds() - DRIVER_TIMESTAMP < DAY_AS_SECONDS(180)) { 1634 DRM_INFO("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n"); 1635 DRM_INFO("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n"); 1636 DRM_INFO("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n"); 1637 DRM_INFO("The gpu crash dump is required to analyze gpu hangs, so please always attach it.\n"); 1638 DRM_INFO("GPU crash dump saved to /sys/class/drm/card%d/error\n", 1639 dev_priv->drm.primary->index); 1640 warned = true; 1641 } 1642 } 1643 1644 void i915_error_state_get(struct drm_device *dev, 1645 struct i915_error_state_file_priv *error_priv) 1646 { 1647 struct drm_i915_private *dev_priv = to_i915(dev); 1648 1649 spin_lock_irq(&dev_priv->gpu_error.lock); 1650 error_priv->error = dev_priv->gpu_error.first_error; 1651 if (error_priv->error) 1652 kref_get(&error_priv->error->ref); 1653 spin_unlock_irq(&dev_priv->gpu_error.lock); 1654 } 1655 1656 void i915_error_state_put(struct i915_error_state_file_priv *error_priv) 1657 { 1658 if (error_priv->error) 1659 kref_put(&error_priv->error->ref, i915_error_state_free); 1660 } 1661 1662 void i915_destroy_error_state(struct drm_device *dev) 1663 { 1664 struct drm_i915_private *dev_priv = to_i915(dev); 1665 struct drm_i915_error_state *error; 1666 1667 spin_lock_irq(&dev_priv->gpu_error.lock); 1668 error = dev_priv->gpu_error.first_error; 1669 dev_priv->gpu_error.first_error = NULL; 1670 spin_unlock_irq(&dev_priv->gpu_error.lock); 1671 1672 if (error) 1673 kref_put(&error->ref, i915_error_state_free); 1674 } 1675