drivers/gpu/drm/i915/i915_gpu_error.c

   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_device *dev = error_priv->dev;
 532         struct drm_i915_private *dev_priv = to_i915(dev);
 533         struct pci_dev *pdev = dev_priv->drm.pdev;
 534         struct drm_i915_error_state *error = error_priv->error;
 535         struct drm_i915_error_object *obj;
 536         int max_hangcheck_score;
 537         int i, j;
 538
 539         if (!error) {
 540                 err_printf(m, "no error state collected\n");
 541                 goto out;
 542         }
 543
 544         err_printf(m, "%s\n", error->error_msg);
 545         err_printf(m, "Kernel: " UTS_RELEASE "\n");
 546         err_printf(m, "Time: %ld s %ld us\n",
 547                    error->time.tv_sec, error->time.tv_usec);
 548         err_printf(m, "Boottime: %ld s %ld us\n",
 549                    error->boottime.tv_sec, error->boottime.tv_usec);
 550         err_printf(m, "Uptime: %ld s %ld us\n",
 551                    error->uptime.tv_sec, error->uptime.tv_usec);
 552         err_print_capabilities(m, &error->device_info);
 553         max_hangcheck_score = 0;
 554         for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
 555                 if (error->engine[i].hangcheck_score > max_hangcheck_score)
 556                         max_hangcheck_score = error->engine[i].hangcheck_score;
 557         }
 558         for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
 559                 if (error->engine[i].hangcheck_score == max_hangcheck_score &&
 560                     error->engine[i].pid != -1) {
 561                         err_printf(m, "Active process (on ring %s): %s [%d]\n",
 562                                    engine_str(i),
 563                                    error->engine[i].comm,
 564                                    error->engine[i].pid);
 565                 }
 566         }
 567         err_printf(m, "Reset count: %u\n", error->reset_count);
 568         err_printf(m, "Suspend count: %u\n", error->suspend_count);
 569         err_printf(m, "PCI ID: 0x%04x\n", pdev->device);
 570         err_printf(m, "PCI Revision: 0x%02x\n", pdev->revision);
 571         err_printf(m, "PCI Subsystem: %04x:%04x\n",
 572                    pdev->subsystem_vendor,
 573                    pdev->subsystem_device);
 574         err_printf(m, "IOMMU enabled?: %d\n", error->iommu);
 575
 576         if (HAS_CSR(dev)) {
 577                 struct intel_csr *csr = &dev_priv->csr;
 578
 579                 err_printf(m, "DMC loaded: %s\n",
 580                            yesno(csr->dmc_payload != NULL));
 581                 err_printf(m, "DMC fw version: %d.%d\n",
 582                            CSR_VERSION_MAJOR(csr->version),
 583                            CSR_VERSION_MINOR(csr->version));
 584         }
 585
 586         err_printf(m, "EIR: 0x%08x\n", error->eir);
 587         err_printf(m, "IER: 0x%08x\n", error->ier);
 588         if (INTEL_INFO(dev)->gen >= 8) {
 589                 for (i = 0; i < 4; i++)
 590                         err_printf(m, "GTIER gt %d: 0x%08x\n", i,
 591                                    error->gtier[i]);
 592         } else if (HAS_PCH_SPLIT(dev_priv) || IS_VALLEYVIEW(dev_priv))
 593                 err_printf(m, "GTIER: 0x%08x\n", error->gtier[0]);
 594         err_printf(m, "PGTBL_ER: 0x%08x\n", error->pgtbl_er);
 595         err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake);
 596         err_printf(m, "DERRMR: 0x%08x\n", error->derrmr);
 597         err_printf(m, "CCID: 0x%08x\n", error->ccid);
 598         err_printf(m, "Missed interrupts: 0x%08lx\n", dev_priv->gpu_error.missed_irq_rings);
 599
 600         for (i = 0; i < dev_priv->num_fence_regs; i++)
 601                 err_printf(m, "  fence[%d] = %08llx\n", i, error->fence[i]);
 602
 603         if (INTEL_INFO(dev)->gen >= 6) {
 604                 err_printf(m, "ERROR: 0x%08x\n", error->error);
 605
 606                 if (INTEL_INFO(dev)->gen >= 8)
 607                         err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n",
 608                                    error->fault_data1, error->fault_data0);
 609
 610                 err_printf(m, "DONE_REG: 0x%08x\n", error->done_reg);
 611         }
 612
 613         if (IS_GEN7(dev_priv))
 614                 err_printf(m, "ERR_INT: 0x%08x\n", error->err_int);
 615
 616         for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
 617                 if (error->engine[i].engine_id != -1)
 618                         error_print_engine(m, &error->engine[i]);
 619         }
 620
 621         for (i = 0; i < ARRAY_SIZE(error->active_vm); i++) {
 622                 char buf[128];
 623                 int len, first = 1;
 624
 625                 if (!error->active_vm[i])
 626                         break;
 627
 628                 len = scnprintf(buf, sizeof(buf), "Active (");
 629                 for (j = 0; j < ARRAY_SIZE(error->engine); j++) {
 630                         if (error->engine[j].vm != error->active_vm[i])
 631                                 continue;
 632
 633                         len += scnprintf(buf + len, sizeof(buf), "%s%s",
 634                                          first ? "" : ", ",
 635                                          dev_priv->engine[j]->name);
 636                         first = 0;
 637                 }
 638                 scnprintf(buf + len, sizeof(buf), ")");
 639                 print_error_buffers(m, buf,
 640                                     error->active_bo[i],
 641                                     error->active_bo_count[i]);
 642         }
 643
 644         print_error_buffers(m, "Pinned (global)",
 645                             error->pinned_bo,
 646                             error->pinned_bo_count);
 647
 648         for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
 649                 struct drm_i915_error_engine *ee = &error->engine[i];
 650
 651                 obj = ee->batchbuffer;
 652                 if (obj) {
 653                         err_puts(m, dev_priv->engine[i]->name);
 654                         if (ee->pid != -1)
 655                                 err_printf(m, " (submitted by %s [%d])",
 656                                            ee->comm,
 657                                            ee->pid);
 658                         err_printf(m, " --- gtt_offset = 0x%08x %08x\n",
 659                                    upper_32_bits(obj->gtt_offset),
 660                                    lower_32_bits(obj->gtt_offset));
 661                         print_error_obj(m, dev_priv->engine[i], NULL, obj);
 662                 }
 663
 664                 if (ee->num_requests) {
 665                         err_printf(m, "%s --- %d requests\n",
 666                                    dev_priv->engine[i]->name,
 667                                    ee->num_requests);
 668                         for (j = 0; j < ee->num_requests; j++)
 669                                 error_print_request(m, " ", &ee->requests[j]);
 670                 }
 671
 672                 if (IS_ERR(ee->waiters)) {
 673                         err_printf(m, "%s --- ? waiters [unable to acquire spinlock]\n",
 674                                    dev_priv->engine[i]->name);
 675                 } else if (ee->num_waiters) {
 676                         err_printf(m, "%s --- %d waiters\n",
 677                                    dev_priv->engine[i]->name,
 678                                    ee->num_waiters);
 679                         for (j = 0; j < ee->num_waiters; j++) {
 680                                 err_printf(m, " seqno 0x%08x for %s [%d]\n",
 681                                            ee->waiters[j].seqno,
 682                                            ee->waiters[j].comm,
 683                                            ee->waiters[j].pid);
 684                         }
 685                 }
 686
 687                 print_error_obj(m, dev_priv->engine[i],
 688                                 "ringbuffer", ee->ringbuffer);
 689
 690                 print_error_obj(m, dev_priv->engine[i],
 691                                 "HW Status", ee->hws_page);
 692
 693                 print_error_obj(m, dev_priv->engine[i],
 694                                 "HW context", ee->ctx);
 695
 696                 print_error_obj(m, dev_priv->engine[i],
 697                                 "WA context", ee->wa_ctx);
 698
 699                 print_error_obj(m, dev_priv->engine[i],
 700                                 "WA batchbuffer", ee->wa_batchbuffer);
 701         }
 702
 703         print_error_obj(m, NULL, "Semaphores", error->semaphore);
 704
 705         print_error_obj(m, NULL, "GuC log buffer", error->guc_log);
 706
 707         if (error->overlay)
 708                 intel_overlay_print_error_state(m, error->overlay);
 709
 710         if (error->display)
 711                 intel_display_print_error_state(m, dev, error->display);
 712
 713 out:
 714         if (m->bytes == 0 && m->err)
 715                 return m->err;
 716
 717         return 0;
 718 }
 719
 720 int i915_error_state_buf_init(struct drm_i915_error_state_buf *ebuf,
 721                               struct drm_i915_private *i915,
 722                               size_t count, loff_t pos)
 723 {
 724         memset(ebuf, 0, sizeof(*ebuf));
 725         ebuf->i915 = i915;
 726
 727         /* We need to have enough room to store any i915_error_state printf
 728          * so that we can move it to start position.
 729          */
 730         ebuf->size = count + 1 > PAGE_SIZE ? count + 1 : PAGE_SIZE;
 731         ebuf->buf = kmalloc(ebuf->size,
 732                                 GFP_TEMPORARY | __GFP_NORETRY | __GFP_NOWARN);
 733
 734         if (ebuf->buf == NULL) {
 735                 ebuf->size = PAGE_SIZE;
 736                 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
 737         }
 738
 739         if (ebuf->buf == NULL) {
 740                 ebuf->size = 128;
 741                 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
 742         }
 743
 744         if (ebuf->buf == NULL)
 745                 return -ENOMEM;
 746
 747         ebuf->start = pos;
 748
 749         return 0;
 750 }
 751
 752 static void i915_error_object_free(struct drm_i915_error_object *obj)
 753 {
 754         int page;
 755
 756         if (obj == NULL)
 757                 return;
 758
 759         for (page = 0; page < obj->page_count; page++)
 760                 free_page((unsigned long)obj->pages[page]);
 761
 762         kfree(obj);
 763 }
 764
 765 static void i915_error_state_free(struct kref *error_ref)
 766 {
 767         struct drm_i915_error_state *error = container_of(error_ref,
 768                                                           typeof(*error), ref);
 769         int i;
 770
 771         for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
 772                 struct drm_i915_error_engine *ee = &error->engine[i];
 773
 774                 i915_error_object_free(ee->batchbuffer);
 775                 i915_error_object_free(ee->wa_batchbuffer);
 776                 i915_error_object_free(ee->ringbuffer);
 777                 i915_error_object_free(ee->hws_page);
 778                 i915_error_object_free(ee->ctx);
 779                 i915_error_object_free(ee->wa_ctx);
 780
 781                 kfree(ee->requests);
 782                 if (!IS_ERR_OR_NULL(ee->waiters))
 783                         kfree(ee->waiters);
 784         }
 785
 786         i915_error_object_free(error->semaphore);
 787         i915_error_object_free(error->guc_log);
 788
 789         for (i = 0; i < ARRAY_SIZE(error->active_bo); i++)
 790                 kfree(error->active_bo[i]);
 791         kfree(error->pinned_bo);
 792
 793         kfree(error->overlay);
 794         kfree(error->display);
 795         kfree(error);
 796 }
 797
 798 static struct drm_i915_error_object *
 799 i915_error_object_create(struct drm_i915_private *i915,
 800                          struct i915_vma *vma)
 801 {
 802         struct i915_ggtt *ggtt = &i915->ggtt;
 803         const u64 slot = ggtt->error_capture.start;
 804         struct drm_i915_error_object *dst;
 805         struct z_stream_s zstream;
 806         unsigned long num_pages;
 807         struct sgt_iter iter;
 808         dma_addr_t dma;
 809
 810         if (!vma)
 811                 return NULL;
 812
 813         num_pages = min_t(u64, vma->size, vma->obj->base.size) >> PAGE_SHIFT;
 814         num_pages = DIV_ROUND_UP(10 * num_pages, 8); /* worstcase zlib growth */
 815         dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *),
 816                       GFP_ATOMIC | __GFP_NOWARN);
 817         if (!dst)
 818                 return NULL;
 819
 820         dst->gtt_offset = vma->node.start;
 821         dst->gtt_size = vma->node.size;
 822         dst->page_count = 0;
 823         dst->unused = 0;
 824
 825         if (!compress_init(&zstream)) {
 826                 kfree(dst);
 827                 return NULL;
 828         }
 829
 830         for_each_sgt_dma(dma, iter, vma->pages) {
 831                 void __iomem *s;
 832                 int ret;
 833
 834                 ggtt->base.insert_page(&ggtt->base, dma, slot,
 835                                        I915_CACHE_NONE, 0);
 836
 837                 s = io_mapping_map_atomic_wc(&ggtt->mappable, slot);
 838                 ret = compress_page(&zstream, (void  __force *)s, dst);
 839                 io_mapping_unmap_atomic(s);
 840
 841                 if (ret)
 842                         goto unwind;
 843         }
 844         goto out;
 845
 846 unwind:
 847         while (dst->page_count--)
 848                 free_page((unsigned long)dst->pages[dst->page_count]);
 849         kfree(dst);
 850         dst = NULL;
 851
 852 out:
 853         compress_fini(&zstream, dst);
 854         ggtt->base.clear_range(&ggtt->base, slot, PAGE_SIZE);
 855         return dst;
 856 }
 857
 858 /* The error capture is special as tries to run underneath the normal
 859  * locking rules - so we use the raw version of the i915_gem_active lookup.
 860  */
 861 static inline uint32_t
 862 __active_get_seqno(struct i915_gem_active *active)
 863 {
 864         return i915_gem_request_get_seqno(__i915_gem_active_peek(active));
 865 }
 866
 867 static inline int
 868 __active_get_engine_id(struct i915_gem_active *active)
 869 {
 870         struct intel_engine_cs *engine;
 871
 872         engine = i915_gem_request_get_engine(__i915_gem_active_peek(active));
 873         return engine ? engine->id : -1;
 874 }
 875
 876 static void capture_bo(struct drm_i915_error_buffer *err,
 877                        struct i915_vma *vma)
 878 {
 879         struct drm_i915_gem_object *obj = vma->obj;
 880         int i;
 881
 882         err->size = obj->base.size;
 883         err->name = obj->base.name;
 884
 885         for (i = 0; i < I915_NUM_ENGINES; i++)
 886                 err->rseqno[i] = __active_get_seqno(&vma->last_read[i]);
 887         err->wseqno = __active_get_seqno(&vma->last_write);
 888         err->engine = __active_get_engine_id(&vma->last_write);
 889
 890         err->gtt_offset = vma->node.start;
 891         err->read_domains = obj->base.read_domains;
 892         err->write_domain = obj->base.write_domain;
 893         err->fence_reg = vma->fence ? vma->fence->id : -1;
 894         err->tiling = i915_gem_object_get_tiling(obj);
 895         err->dirty = obj->mm.dirty;
 896         err->purgeable = obj->mm.madv != I915_MADV_WILLNEED;
 897         err->userptr = obj->userptr.mm != NULL;
 898         err->cache_level = obj->cache_level;
 899 }
 900
 901 static u32 capture_error_bo(struct drm_i915_error_buffer *err,
 902                             int count, struct list_head *head,
 903                             bool pinned_only)
 904 {
 905         struct i915_vma *vma;
 906         int i = 0;
 907
 908         list_for_each_entry(vma, head, vm_link) {
 909                 if (pinned_only && !i915_vma_is_pinned(vma))
 910                         continue;
 911
 912                 capture_bo(err++, vma);
 913                 if (++i == count)
 914                         break;
 915         }
 916
 917         return i;
 918 }
 919
 920 /* Generate a semi-unique error code. The code is not meant to have meaning, The
 921  * code's only purpose is to try to prevent false duplicated bug reports by
 922  * grossly estimating a GPU error state.
 923  *
 924  * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
 925  * the hang if we could strip the GTT offset information from it.
 926  *
 927  * It's only a small step better than a random number in its current form.
 928  */
 929 static uint32_t i915_error_generate_code(struct drm_i915_private *dev_priv,
 930                                          struct drm_i915_error_state *error,
 931                                          int *engine_id)
 932 {
 933         uint32_t error_code = 0;
 934         int i;
 935
 936         /* IPEHR would be an ideal way to detect errors, as it's the gross
 937          * measure of "the command that hung." However, has some very common
 938          * synchronization commands which almost always appear in the case
 939          * strictly a client bug. Use instdone to differentiate those some.
 940          */
 941         for (i = 0; i < I915_NUM_ENGINES; i++) {
 942                 if (error->engine[i].hangcheck_action == HANGCHECK_HUNG) {
 943                         if (engine_id)
 944                                 *engine_id = i;
 945
 946                         return error->engine[i].ipehr ^
 947                                error->engine[i].instdone.instdone;
 948                 }
 949         }
 950
 951         return error_code;
 952 }
 953
 954 static void i915_gem_record_fences(struct drm_i915_private *dev_priv,
 955                                    struct drm_i915_error_state *error)
 956 {
 957         int i;
 958
 959         if (IS_GEN3(dev_priv) || IS_GEN2(dev_priv)) {
 960                 for (i = 0; i < dev_priv->num_fence_regs; i++)
 961                         error->fence[i] = I915_READ(FENCE_REG(i));
 962         } else if (IS_GEN5(dev_priv) || IS_GEN4(dev_priv)) {
 963                 for (i = 0; i < dev_priv->num_fence_regs; i++)
 964                         error->fence[i] = I915_READ64(FENCE_REG_965_LO(i));
 965         } else if (INTEL_GEN(dev_priv) >= 6) {
 966                 for (i = 0; i < dev_priv->num_fence_regs; i++)
 967                         error->fence[i] = I915_READ64(FENCE_REG_GEN6_LO(i));
 968         }
 969 }
 970
 971 static inline u32
 972 gen8_engine_sync_index(struct intel_engine_cs *engine,
 973                        struct intel_engine_cs *other)
 974 {
 975         int idx;
 976
 977         /*
 978          * rcs -> 0 = vcs, 1 = bcs, 2 = vecs, 3 = vcs2;
 979          * vcs -> 0 = bcs, 1 = vecs, 2 = vcs2, 3 = rcs;
 980          * bcs -> 0 = vecs, 1 = vcs2. 2 = rcs, 3 = vcs;
 981          * vecs -> 0 = vcs2, 1 = rcs, 2 = vcs, 3 = bcs;
 982          * vcs2 -> 0 = rcs, 1 = vcs, 2 = bcs, 3 = vecs;
 983          */
 984
 985         idx = (other - engine) - 1;
 986         if (idx < 0)
 987                 idx += I915_NUM_ENGINES;
 988
 989         return idx;
 990 }
 991
 992 static void gen8_record_semaphore_state(struct drm_i915_error_state *error,
 993                                         struct intel_engine_cs *engine,
 994                                         struct drm_i915_error_engine *ee)
 995 {
 996         struct drm_i915_private *dev_priv = engine->i915;
 997         struct intel_engine_cs *to;
 998         enum intel_engine_id id;
 999
1000         if (!error->semaphore)
1001                 return;
1002
1003         for_each_engine(to, dev_priv, id) {
1004                 int idx;
1005                 u16 signal_offset;
1006                 u32 *tmp;
1007
1008                 if (engine == to)
1009                         continue;
1010
1011                 signal_offset =
1012                         (GEN8_SIGNAL_OFFSET(engine, id) & (PAGE_SIZE - 1)) / 4;
1013                 tmp = error->semaphore->pages[0];
1014                 idx = gen8_engine_sync_index(engine, to);
1015
1016                 ee->semaphore_mboxes[idx] = tmp[signal_offset];
1017         }
1018 }
1019
1020 static void gen6_record_semaphore_state(struct intel_engine_cs *engine,
1021                                         struct drm_i915_error_engine *ee)
1022 {
1023         struct drm_i915_private *dev_priv = engine->i915;
1024
1025         ee->semaphore_mboxes[0] = I915_READ(RING_SYNC_0(engine->mmio_base));
1026         ee->semaphore_mboxes[1] = I915_READ(RING_SYNC_1(engine->mmio_base));
1027         if (HAS_VEBOX(dev_priv))
1028                 ee->semaphore_mboxes[2] =
1029                         I915_READ(RING_SYNC_2(engine->mmio_base));
1030 }
1031
1032 static void error_record_engine_waiters(struct intel_engine_cs *engine,
1033                                         struct drm_i915_error_engine *ee)
1034 {
1035         struct intel_breadcrumbs *b = &engine->breadcrumbs;
1036         struct drm_i915_error_waiter *waiter;
1037         struct rb_node *rb;
1038         int count;
1039
1040         ee->num_waiters = 0;
1041         ee->waiters = NULL;
1042
1043         if (RB_EMPTY_ROOT(&b->waiters))
1044                 return;
1045
1046         if (!spin_trylock_irq(&b->lock)) {
1047                 ee->waiters = ERR_PTR(-EDEADLK);
1048                 return;
1049         }
1050
1051         count = 0;
1052         for (rb = rb_first(&b->waiters); rb != NULL; rb = rb_next(rb))
1053                 count++;
1054         spin_unlock_irq(&b->lock);
1055
1056         waiter = NULL;
1057         if (count)
1058                 waiter = kmalloc_array(count,
1059                                        sizeof(struct drm_i915_error_waiter),
1060                                        GFP_ATOMIC);
1061         if (!waiter)
1062                 return;
1063
1064         if (!spin_trylock_irq(&b->lock)) {
1065                 kfree(waiter);
1066                 ee->waiters = ERR_PTR(-EDEADLK);
1067                 return;
1068         }
1069
1070         ee->waiters = waiter;
1071         for (rb = rb_first(&b->waiters); rb; rb = rb_next(rb)) {
1072                 struct intel_wait *w = container_of(rb, typeof(*w), node);
1073
1074                 strcpy(waiter->comm, w->tsk->comm);
1075                 waiter->pid = w->tsk->pid;
1076                 waiter->seqno = w->seqno;
1077                 waiter++;
1078
1079                 if (++ee->num_waiters == count)
1080                         break;
1081         }
1082         spin_unlock_irq(&b->lock);
1083 }
1084
1085 static void error_record_engine_registers(struct drm_i915_error_state *error,
1086                                           struct intel_engine_cs *engine,
1087                                           struct drm_i915_error_engine *ee)
1088 {
1089         struct drm_i915_private *dev_priv = engine->i915;
1090
1091         if (INTEL_GEN(dev_priv) >= 6) {
1092                 ee->rc_psmi = I915_READ(RING_PSMI_CTL(engine->mmio_base));
1093                 ee->fault_reg = I915_READ(RING_FAULT_REG(engine));
1094                 if (INTEL_GEN(dev_priv) >= 8)
1095                         gen8_record_semaphore_state(error, engine, ee);
1096                 else
1097                         gen6_record_semaphore_state(engine, ee);
1098         }
1099
1100         if (INTEL_GEN(dev_priv) >= 4) {
1101                 ee->faddr = I915_READ(RING_DMA_FADD(engine->mmio_base));
1102                 ee->ipeir = I915_READ(RING_IPEIR(engine->mmio_base));
1103                 ee->ipehr = I915_READ(RING_IPEHR(engine->mmio_base));
1104                 ee->instps = I915_READ(RING_INSTPS(engine->mmio_base));
1105                 ee->bbaddr = I915_READ(RING_BBADDR(engine->mmio_base));
1106                 if (INTEL_GEN(dev_priv) >= 8) {
1107                         ee->faddr |= (u64) I915_READ(RING_DMA_FADD_UDW(engine->mmio_base)) << 32;
1108                         ee->bbaddr |= (u64) I915_READ(RING_BBADDR_UDW(engine->mmio_base)) << 32;
1109                 }
1110                 ee->bbstate = I915_READ(RING_BBSTATE(engine->mmio_base));
1111         } else {
1112                 ee->faddr = I915_READ(DMA_FADD_I8XX);
1113                 ee->ipeir = I915_READ(IPEIR);
1114                 ee->ipehr = I915_READ(IPEHR);
1115         }
1116
1117         intel_engine_get_instdone(engine, &ee->instdone);
1118
1119         ee->waiting = intel_engine_has_waiter(engine);
1120         ee->instpm = I915_READ(RING_INSTPM(engine->mmio_base));
1121         ee->acthd = intel_engine_get_active_head(engine);
1122         ee->seqno = intel_engine_get_seqno(engine);
1123         ee->last_seqno = engine->timeline->last_submitted_seqno;
1124         ee->start = I915_READ_START(engine);
1125         ee->head = I915_READ_HEAD(engine);
1126         ee->tail = I915_READ_TAIL(engine);
1127         ee->ctl = I915_READ_CTL(engine);
1128         if (INTEL_GEN(dev_priv) > 2)
1129                 ee->mode = I915_READ_MODE(engine);
1130
1131         if (!HWS_NEEDS_PHYSICAL(dev_priv)) {
1132                 i915_reg_t mmio;
1133
1134                 if (IS_GEN7(dev_priv)) {
1135                         switch (engine->id) {
1136                         default:
1137                         case RCS:
1138                                 mmio = RENDER_HWS_PGA_GEN7;
1139                                 break;
1140                         case BCS:
1141                                 mmio = BLT_HWS_PGA_GEN7;
1142                                 break;
1143                         case VCS:
1144                                 mmio = BSD_HWS_PGA_GEN7;
1145                                 break;
1146                         case VECS:
1147                                 mmio = VEBOX_HWS_PGA_GEN7;
1148                                 break;
1149                         }
1150                 } else if (IS_GEN6(engine->i915)) {
1151                         mmio = RING_HWS_PGA_GEN6(engine->mmio_base);
1152                 } else {
1153                         /* XXX: gen8 returns to sanity */
1154                         mmio = RING_HWS_PGA(engine->mmio_base);
1155                 }
1156
1157                 ee->hws = I915_READ(mmio);
1158         }
1159
1160         ee->hangcheck_score = engine->hangcheck.score;
1161         ee->hangcheck_action = engine->hangcheck.action;
1162
1163         if (USES_PPGTT(dev_priv)) {
1164                 int i;
1165
1166                 ee->vm_info.gfx_mode = I915_READ(RING_MODE_GEN7(engine));
1167
1168                 if (IS_GEN6(dev_priv))
1169                         ee->vm_info.pp_dir_base =
1170                                 I915_READ(RING_PP_DIR_BASE_READ(engine));
1171                 else if (IS_GEN7(dev_priv))
1172                         ee->vm_info.pp_dir_base =
1173                                 I915_READ(RING_PP_DIR_BASE(engine));
1174                 else if (INTEL_GEN(dev_priv) >= 8)
1175                         for (i = 0; i < 4; i++) {
1176                                 ee->vm_info.pdp[i] =
1177                                         I915_READ(GEN8_RING_PDP_UDW(engine, i));
1178                                 ee->vm_info.pdp[i] <<= 32;
1179                                 ee->vm_info.pdp[i] |=
1180                                         I915_READ(GEN8_RING_PDP_LDW(engine, i));
1181                         }
1182         }
1183 }
1184
1185 static void record_request(struct drm_i915_gem_request *request,
1186                            struct drm_i915_error_request *erq)
1187 {
1188         erq->context = request->ctx->hw_id;
1189         erq->seqno = request->global_seqno;
1190         erq->jiffies = request->emitted_jiffies;
1191         erq->head = request->head;
1192         erq->tail = request->tail;
1193
1194         rcu_read_lock();
1195         erq->pid = request->ctx->pid ? pid_nr(request->ctx->pid) : 0;
1196         rcu_read_unlock();
1197 }
1198
1199 static void engine_record_requests(struct intel_engine_cs *engine,
1200                                    struct drm_i915_gem_request *first,
1201                                    struct drm_i915_error_engine *ee)
1202 {
1203         struct drm_i915_gem_request *request;
1204         int count;
1205
1206         count = 0;
1207         request = first;
1208         list_for_each_entry_from(request, &engine->timeline->requests, link)
1209                 count++;
1210         if (!count)
1211                 return;
1212
1213         ee->requests = kcalloc(count, sizeof(*ee->requests), GFP_ATOMIC);
1214         if (!ee->requests)
1215                 return;
1216
1217         ee->num_requests = count;
1218
1219         count = 0;
1220         request = first;
1221         list_for_each_entry_from(request, &engine->timeline->requests, link) {
1222                 if (count >= ee->num_requests) {
1223                         /*
1224                          * If the ring request list was changed in
1225                          * between the point where the error request
1226                          * list was created and dimensioned and this
1227                          * point then just exit early to avoid crashes.
1228                          *
1229                          * We don't need to communicate that the
1230                          * request list changed state during error
1231                          * state capture and that the error state is
1232                          * slightly incorrect as a consequence since we
1233                          * are typically only interested in the request
1234                          * list state at the point of error state
1235                          * capture, not in any changes happening during
1236                          * the capture.
1237                          */
1238                         break;
1239                 }
1240
1241                 record_request(request, &ee->requests[count++]);
1242         }
1243         ee->num_requests = count;
1244 }
1245
1246 static void error_record_engine_execlists(struct intel_engine_cs *engine,
1247                                           struct drm_i915_error_engine *ee)
1248 {
1249         unsigned int n;
1250
1251         for (n = 0; n < ARRAY_SIZE(engine->execlist_port); n++)
1252                 if (engine->execlist_port[n].request)
1253                         record_request(engine->execlist_port[n].request,
1254                                        &ee->execlist[n]);
1255 }
1256
1257 static void i915_gem_record_rings(struct drm_i915_private *dev_priv,
1258                                   struct drm_i915_error_state *error)
1259 {
1260         struct i915_ggtt *ggtt = &dev_priv->ggtt;
1261         int i;
1262
1263         error->semaphore =
1264                 i915_error_object_create(dev_priv, dev_priv->semaphore);
1265
1266         for (i = 0; i < I915_NUM_ENGINES; i++) {
1267                 struct intel_engine_cs *engine = dev_priv->engine[i];
1268                 struct drm_i915_error_engine *ee = &error->engine[i];
1269                 struct drm_i915_gem_request *request;
1270
1271                 ee->pid = -1;
1272                 ee->engine_id = -1;
1273
1274                 if (!engine)
1275                         continue;
1276
1277                 ee->engine_id = i;
1278
1279                 error_record_engine_registers(error, engine, ee);
1280                 error_record_engine_waiters(engine, ee);
1281                 error_record_engine_execlists(engine, ee);
1282
1283                 request = i915_gem_find_active_request(engine);
1284                 if (request) {
1285                         struct intel_ring *ring;
1286                         struct pid *pid;
1287
1288                         ee->vm = request->ctx->ppgtt ?
1289                                 &request->ctx->ppgtt->base : &ggtt->base;
1290
1291                         /* We need to copy these to an anonymous buffer
1292                          * as the simplest method to avoid being overwritten
1293                          * by userspace.
1294                          */
1295                         ee->batchbuffer =
1296                                 i915_error_object_create(dev_priv,
1297                                                          request->batch);
1298
1299                         if (HAS_BROKEN_CS_TLB(dev_priv))
1300                                 ee->wa_batchbuffer =
1301                                         i915_error_object_create(dev_priv,
1302                                                                  engine->scratch);
1303
1304                         ee->ctx =
1305                                 i915_error_object_create(dev_priv,
1306                                                          request->ctx->engine[i].state);
1307
1308                         pid = request->ctx->pid;
1309                         if (pid) {
1310                                 struct task_struct *task;
1311
1312                                 rcu_read_lock();
1313                                 task = pid_task(pid, PIDTYPE_PID);
1314                                 if (task) {
1315                                         strcpy(ee->comm, task->comm);
1316                                         ee->pid = task->pid;
1317                                 }
1318                                 rcu_read_unlock();
1319                         }
1320
1321                         error->simulated |=
1322                                 request->ctx->flags & CONTEXT_NO_ERROR_CAPTURE;
1323
1324                         ee->rq_head = request->head;
1325                         ee->rq_post = request->postfix;
1326                         ee->rq_tail = request->tail;
1327
1328                         ring = request->ring;
1329                         ee->cpu_ring_head = ring->head;
1330                         ee->cpu_ring_tail = ring->tail;
1331                         ee->ringbuffer =
1332                                 i915_error_object_create(dev_priv, ring->vma);
1333
1334                         engine_record_requests(engine, request, ee);
1335                 }
1336
1337                 ee->hws_page =
1338                         i915_error_object_create(dev_priv,
1339                                                  engine->status_page.vma);
1340
1341                 ee->wa_ctx =
1342                         i915_error_object_create(dev_priv, engine->wa_ctx.vma);
1343         }
1344 }
1345
1346 static void i915_gem_capture_vm(struct drm_i915_private *dev_priv,
1347                                 struct drm_i915_error_state *error,
1348                                 struct i915_address_space *vm,
1349                                 int idx)
1350 {
1351         struct drm_i915_error_buffer *active_bo;
1352         struct i915_vma *vma;
1353         int count;
1354
1355         count = 0;
1356         list_for_each_entry(vma, &vm->active_list, vm_link)
1357                 count++;
1358
1359         active_bo = NULL;
1360         if (count)
1361                 active_bo = kcalloc(count, sizeof(*active_bo), GFP_ATOMIC);
1362         if (active_bo)
1363                 count = capture_error_bo(active_bo, count, &vm->active_list, false);
1364         else
1365                 count = 0;
1366
1367         error->active_vm[idx] = vm;
1368         error->active_bo[idx] = active_bo;
1369         error->active_bo_count[idx] = count;
1370 }
1371
1372 static void i915_capture_active_buffers(struct drm_i915_private *dev_priv,
1373                                         struct drm_i915_error_state *error)
1374 {
1375         int cnt = 0, i, j;
1376
1377         BUILD_BUG_ON(ARRAY_SIZE(error->engine) > ARRAY_SIZE(error->active_bo));
1378         BUILD_BUG_ON(ARRAY_SIZE(error->active_bo) != ARRAY_SIZE(error->active_vm));
1379         BUILD_BUG_ON(ARRAY_SIZE(error->active_bo) != ARRAY_SIZE(error->active_bo_count));
1380
1381         /* Scan each engine looking for unique active contexts/vm */
1382         for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
1383                 struct drm_i915_error_engine *ee = &error->engine[i];
1384                 bool found;
1385
1386                 if (!ee->vm)
1387                         continue;
1388
1389                 found = false;
1390                 for (j = 0; j < i && !found; j++)
1391                         found = error->engine[j].vm == ee->vm;
1392                 if (!found)
1393                         i915_gem_capture_vm(dev_priv, error, ee->vm, cnt++);
1394         }
1395 }
1396
1397 static void i915_capture_pinned_buffers(struct drm_i915_private *dev_priv,
1398                                         struct drm_i915_error_state *error)
1399 {
1400         struct i915_address_space *vm = &dev_priv->ggtt.base;
1401         struct drm_i915_error_buffer *bo;
1402         struct i915_vma *vma;
1403         int count_inactive, count_active;
1404
1405         count_inactive = 0;
1406         list_for_each_entry(vma, &vm->active_list, vm_link)
1407                 count_inactive++;
1408
1409         count_active = 0;
1410         list_for_each_entry(vma, &vm->inactive_list, vm_link)
1411                 count_active++;
1412
1413         bo = NULL;
1414         if (count_inactive + count_active)
1415                 bo = kcalloc(count_inactive + count_active,
1416                              sizeof(*bo), GFP_ATOMIC);
1417         if (!bo)
1418                 return;
1419
1420         count_inactive = capture_error_bo(bo, count_inactive,
1421                                           &vm->active_list, true);
1422         count_active = capture_error_bo(bo + count_inactive, count_active,
1423                                         &vm->inactive_list, true);
1424         error->pinned_bo_count = count_inactive + count_active;
1425         error->pinned_bo = bo;
1426 }
1427
1428 static void i915_gem_capture_guc_log_buffer(struct drm_i915_private *dev_priv,
1429                                             struct drm_i915_error_state *error)
1430 {
1431         /* Capturing log buf contents won't be useful if logging was disabled */
1432         if (!dev_priv->guc.log.vma || (i915.guc_log_level < 0))
1433                 return;
1434
1435         error->guc_log = i915_error_object_create(dev_priv,
1436                                                   dev_priv->guc.log.vma);
1437 }
1438
1439 /* Capture all registers which don't fit into another category. */
1440 static void i915_capture_reg_state(struct drm_i915_private *dev_priv,
1441                                    struct drm_i915_error_state *error)
1442 {
1443         struct drm_device *dev = &dev_priv->drm;
1444         int i;
1445
1446         /* General organization
1447          * 1. Registers specific to a single generation
1448          * 2. Registers which belong to multiple generations
1449          * 3. Feature specific registers.
1450          * 4. Everything else
1451          * Please try to follow the order.
1452          */
1453
1454         /* 1: Registers specific to a single generation */
1455         if (IS_VALLEYVIEW(dev_priv)) {
1456                 error->gtier[0] = I915_READ(GTIER);
1457                 error->ier = I915_READ(VLV_IER);
1458                 error->forcewake = I915_READ_FW(FORCEWAKE_VLV);
1459         }
1460
1461         if (IS_GEN7(dev_priv))
1462                 error->err_int = I915_READ(GEN7_ERR_INT);
1463
1464         if (INTEL_INFO(dev)->gen >= 8) {
1465                 error->fault_data0 = I915_READ(GEN8_FAULT_TLB_DATA0);
1466                 error->fault_data1 = I915_READ(GEN8_FAULT_TLB_DATA1);
1467         }
1468
1469         if (IS_GEN6(dev_priv)) {
1470                 error->forcewake = I915_READ_FW(FORCEWAKE);
1471                 error->gab_ctl = I915_READ(GAB_CTL);
1472                 error->gfx_mode = I915_READ(GFX_MODE);
1473         }
1474
1475         /* 2: Registers which belong to multiple generations */
1476         if (INTEL_INFO(dev)->gen >= 7)
1477                 error->forcewake = I915_READ_FW(FORCEWAKE_MT);
1478
1479         if (INTEL_INFO(dev)->gen >= 6) {
1480                 error->derrmr = I915_READ(DERRMR);
1481                 error->error = I915_READ(ERROR_GEN6);
1482                 error->done_reg = I915_READ(DONE_REG);
1483         }
1484
1485         /* 3: Feature specific registers */
1486         if (IS_GEN6(dev_priv) || IS_GEN7(dev_priv)) {
1487                 error->gam_ecochk = I915_READ(GAM_ECOCHK);
1488                 error->gac_eco = I915_READ(GAC_ECO_BITS);
1489         }
1490
1491         /* 4: Everything else */
1492         if (HAS_HW_CONTEXTS(dev))
1493                 error->ccid = I915_READ(CCID);
1494
1495         if (INTEL_INFO(dev)->gen >= 8) {
1496                 error->ier = I915_READ(GEN8_DE_MISC_IER);
1497                 for (i = 0; i < 4; i++)
1498                         error->gtier[i] = I915_READ(GEN8_GT_IER(i));
1499         } else if (HAS_PCH_SPLIT(dev_priv)) {
1500                 error->ier = I915_READ(DEIER);
1501                 error->gtier[0] = I915_READ(GTIER);
1502         } else if (IS_GEN2(dev_priv)) {
1503                 error->ier = I915_READ16(IER);
1504         } else if (!IS_VALLEYVIEW(dev_priv)) {
1505                 error->ier = I915_READ(IER);
1506         }
1507         error->eir = I915_READ(EIR);
1508         error->pgtbl_er = I915_READ(PGTBL_ER);
1509 }
1510
1511 static void i915_error_capture_msg(struct drm_i915_private *dev_priv,
1512                                    struct drm_i915_error_state *error,
1513                                    u32 engine_mask,
1514                                    const char *error_msg)
1515 {
1516         u32 ecode;
1517         int engine_id = -1, len;
1518
1519         ecode = i915_error_generate_code(dev_priv, error, &engine_id);
1520
1521         len = scnprintf(error->error_msg, sizeof(error->error_msg),
1522                         "GPU HANG: ecode %d:%d:0x%08x",
1523                         INTEL_GEN(dev_priv), engine_id, ecode);
1524
1525         if (engine_id != -1 && error->engine[engine_id].pid != -1)
1526                 len += scnprintf(error->error_msg + len,
1527                                  sizeof(error->error_msg) - len,
1528                                  ", in %s [%d]",
1529                                  error->engine[engine_id].comm,
1530                                  error->engine[engine_id].pid);
1531
1532         scnprintf(error->error_msg + len, sizeof(error->error_msg) - len,
1533                   ", reason: %s, action: %s",
1534                   error_msg,
1535                   engine_mask ? "reset" : "continue");
1536 }
1537
1538 static void i915_capture_gen_state(struct drm_i915_private *dev_priv,
1539                                    struct drm_i915_error_state *error)
1540 {
1541         error->iommu = -1;
1542 #ifdef CONFIG_INTEL_IOMMU
1543         error->iommu = intel_iommu_gfx_mapped;
1544 #endif
1545         error->reset_count = i915_reset_count(&dev_priv->gpu_error);
1546         error->suspend_count = dev_priv->suspend_count;
1547
1548         memcpy(&error->device_info,
1549                INTEL_INFO(dev_priv),
1550                sizeof(error->device_info));
1551 }
1552
1553 static int capture(void *data)
1554 {
1555         struct drm_i915_error_state *error = data;
1556
1557         i915_capture_gen_state(error->i915, error);
1558         i915_capture_reg_state(error->i915, error);
1559         i915_gem_record_fences(error->i915, error);
1560         i915_gem_record_rings(error->i915, error);
1561         i915_capture_active_buffers(error->i915, error);
1562         i915_capture_pinned_buffers(error->i915, error);
1563         i915_gem_capture_guc_log_buffer(error->i915, error);
1564
1565         do_gettimeofday(&error->time);
1566         error->boottime = ktime_to_timeval(ktime_get_boottime());
1567         error->uptime =
1568                 ktime_to_timeval(ktime_sub(ktime_get(),
1569                                            error->i915->gt.last_init_time));
1570
1571         error->overlay = intel_overlay_capture_error_state(error->i915);
1572         error->display = intel_display_capture_error_state(error->i915);
1573
1574         return 0;
1575 }
1576
1577 #define DAY_AS_SECONDS(x) (24 * 60 * 60 * (x))
1578
1579 /**
1580  * i915_capture_error_state - capture an error record for later analysis
1581  * @dev: drm device
1582  *
1583  * Should be called when an error is detected (either a hang or an error
1584  * interrupt) to capture error state from the time of the error.  Fills
1585  * out a structure which becomes available in debugfs for user level tools
1586  * to pick up.
1587  */
1588 void i915_capture_error_state(struct drm_i915_private *dev_priv,
1589                               u32 engine_mask,
1590                               const char *error_msg)
1591 {
1592         static bool warned;
1593         struct drm_i915_error_state *error;
1594         unsigned long flags;
1595
1596         if (!i915.error_capture)
1597                 return;
1598
1599         if (READ_ONCE(dev_priv->gpu_error.first_error))
1600                 return;
1601
1602         /* Account for pipe specific data like PIPE*STAT */
1603         error = kzalloc(sizeof(*error), GFP_ATOMIC);
1604         if (!error) {
1605                 DRM_DEBUG_DRIVER("out of memory, not capturing error state\n");
1606                 return;
1607         }
1608
1609         kref_init(&error->ref);
1610         error->i915 = dev_priv;
1611
1612         stop_machine(capture, error, NULL);
1613
1614         i915_error_capture_msg(dev_priv, error, engine_mask, error_msg);
1615         DRM_INFO("%s\n", error->error_msg);
1616
1617         if (!error->simulated) {
1618                 spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1619                 if (!dev_priv->gpu_error.first_error) {
1620                         dev_priv->gpu_error.first_error = error;
1621                         error = NULL;
1622                 }
1623                 spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1624         }
1625
1626         if (error) {
1627                 i915_error_state_free(&error->ref);
1628                 return;
1629         }
1630
1631         if (!warned &&
1632             ktime_get_real_seconds() - DRIVER_TIMESTAMP < DAY_AS_SECONDS(180)) {
1633                 DRM_INFO("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1634                 DRM_INFO("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n");
1635                 DRM_INFO("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1636                 DRM_INFO("The gpu crash dump is required to analyze gpu hangs, so please always attach it.\n");
1637                 DRM_INFO("GPU crash dump saved to /sys/class/drm/card%d/error\n",
1638                          dev_priv->drm.primary->index);
1639                 warned = true;
1640         }
1641 }
1642
1643 void i915_error_state_get(struct drm_device *dev,
1644                           struct i915_error_state_file_priv *error_priv)
1645 {
1646         struct drm_i915_private *dev_priv = to_i915(dev);
1647
1648         spin_lock_irq(&dev_priv->gpu_error.lock);
1649         error_priv->error = dev_priv->gpu_error.first_error;
1650         if (error_priv->error)
1651                 kref_get(&error_priv->error->ref);
1652         spin_unlock_irq(&dev_priv->gpu_error.lock);
1653 }
1654
1655 void i915_error_state_put(struct i915_error_state_file_priv *error_priv)
1656 {
1657         if (error_priv->error)
1658                 kref_put(&error_priv->error->ref, i915_error_state_free);
1659 }
1660
1661 void i915_destroy_error_state(struct drm_device *dev)
1662 {
1663         struct drm_i915_private *dev_priv = to_i915(dev);
1664         struct drm_i915_error_state *error;
1665
1666         spin_lock_irq(&dev_priv->gpu_error.lock);
1667         error = dev_priv->gpu_error.first_error;
1668         dev_priv->gpu_error.first_error = NULL;
1669         spin_unlock_irq(&dev_priv->gpu_error.lock);
1670
1671         if (error)
1672                 kref_put(&error->ref, i915_error_state_free);
1673 }