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 "i915_drv.h"
  32
  33 static const char *ring_str(int ring)
  34 {
  35         switch (ring) {
  36         case RCS: return "render";
  37         case VCS: return "bsd";
  38         case BCS: return "blt";
  39         case VECS: return "vebox";
  40         case VCS2: return "bsd2";
  41         default: return "";
  42         }
  43 }
  44
  45 static const char *pin_flag(int pinned)
  46 {
  47         if (pinned > 0)
  48                 return " P";
  49         else if (pinned < 0)
  50                 return " p";
  51         else
  52                 return "";
  53 }
  54
  55 static const char *tiling_flag(int tiling)
  56 {
  57         switch (tiling) {
  58         default:
  59         case I915_TILING_NONE: return "";
  60         case I915_TILING_X: return " X";
  61         case I915_TILING_Y: return " Y";
  62         }
  63 }
  64
  65 static const char *dirty_flag(int dirty)
  66 {
  67         return dirty ? " dirty" : "";
  68 }
  69
  70 static const char *purgeable_flag(int purgeable)
  71 {
  72         return purgeable ? " purgeable" : "";
  73 }
  74
  75 static bool __i915_error_ok(struct drm_i915_error_state_buf *e)
  76 {
  77
  78         if (!e->err && WARN(e->bytes > (e->size - 1), "overflow")) {
  79                 e->err = -ENOSPC;
  80                 return false;
  81         }
  82
  83         if (e->bytes == e->size - 1 || e->err)
  84                 return false;
  85
  86         return true;
  87 }
  88
  89 static bool __i915_error_seek(struct drm_i915_error_state_buf *e,
  90                               unsigned len)
  91 {
  92         if (e->pos + len <= e->start) {
  93                 e->pos += len;
  94                 return false;
  95         }
  96
  97         /* First vsnprintf needs to fit in its entirety for memmove */
  98         if (len >= e->size) {
  99                 e->err = -EIO;
 100                 return false;
 101         }
 102
 103         return true;
 104 }
 105
 106 static void __i915_error_advance(struct drm_i915_error_state_buf *e,
 107                                  unsigned len)
 108 {
 109         /* If this is first printf in this window, adjust it so that
 110          * start position matches start of the buffer
 111          */
 112
 113         if (e->pos < e->start) {
 114                 const size_t off = e->start - e->pos;
 115
 116                 /* Should not happen but be paranoid */
 117                 if (off > len || e->bytes) {
 118                         e->err = -EIO;
 119                         return;
 120                 }
 121
 122                 memmove(e->buf, e->buf + off, len - off);
 123                 e->bytes = len - off;
 124                 e->pos = e->start;
 125                 return;
 126         }
 127
 128         e->bytes += len;
 129         e->pos += len;
 130 }
 131
 132 static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
 133                                const char *f, va_list args)
 134 {
 135         unsigned len;
 136
 137         if (!__i915_error_ok(e))
 138                 return;
 139
 140         /* Seek the first printf which is hits start position */
 141         if (e->pos < e->start) {
 142                 va_list tmp;
 143
 144                 va_copy(tmp, args);
 145                 len = vsnprintf(NULL, 0, f, tmp);
 146                 va_end(tmp);
 147
 148                 if (!__i915_error_seek(e, len))
 149                         return;
 150         }
 151
 152         len = vsnprintf(e->buf + e->bytes, e->size - e->bytes, f, args);
 153         if (len >= e->size - e->bytes)
 154                 len = e->size - e->bytes - 1;
 155
 156         __i915_error_advance(e, len);
 157 }
 158
 159 static void i915_error_puts(struct drm_i915_error_state_buf *e,
 160                             const char *str)
 161 {
 162         unsigned len;
 163
 164         if (!__i915_error_ok(e))
 165                 return;
 166
 167         len = strlen(str);
 168
 169         /* Seek the first printf which is hits start position */
 170         if (e->pos < e->start) {
 171                 if (!__i915_error_seek(e, len))
 172                         return;
 173         }
 174
 175         if (len >= e->size - e->bytes)
 176                 len = e->size - e->bytes - 1;
 177         memcpy(e->buf + e->bytes, str, len);
 178
 179         __i915_error_advance(e, len);
 180 }
 181
 182 #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
 183 #define err_puts(e, s) i915_error_puts(e, s)
 184
 185 static void print_error_buffers(struct drm_i915_error_state_buf *m,
 186                                 const char *name,
 187                                 struct drm_i915_error_buffer *err,
 188                                 int count)
 189 {
 190         int i;
 191
 192         err_printf(m, "  %s [%d]:\n", name, count);
 193
 194         while (count--) {
 195                 err_printf(m, "    %08x_%08x %8u %02x %02x [ ",
 196                            upper_32_bits(err->gtt_offset),
 197                            lower_32_bits(err->gtt_offset),
 198                            err->size,
 199                            err->read_domains,
 200                            err->write_domain);
 201                 for (i = 0; i < I915_NUM_ENGINES; i++)
 202                         err_printf(m, "%02x ", err->rseqno[i]);
 203
 204                 err_printf(m, "] %02x", err->wseqno);
 205                 err_puts(m, pin_flag(err->pinned));
 206                 err_puts(m, tiling_flag(err->tiling));
 207                 err_puts(m, dirty_flag(err->dirty));
 208                 err_puts(m, purgeable_flag(err->purgeable));
 209                 err_puts(m, err->userptr ? " userptr" : "");
 210                 err_puts(m, err->ring != -1 ? " " : "");
 211                 err_puts(m, ring_str(err->ring));
 212                 err_puts(m, i915_cache_level_str(m->i915, err->cache_level));
 213
 214                 if (err->name)
 215                         err_printf(m, " (name: %d)", err->name);
 216                 if (err->fence_reg != I915_FENCE_REG_NONE)
 217                         err_printf(m, " (fence: %d)", err->fence_reg);
 218
 219                 err_puts(m, "\n");
 220                 err++;
 221         }
 222 }
 223
 224 static const char *hangcheck_action_to_str(enum intel_ring_hangcheck_action a)
 225 {
 226         switch (a) {
 227         case HANGCHECK_IDLE:
 228                 return "idle";
 229         case HANGCHECK_WAIT:
 230                 return "wait";
 231         case HANGCHECK_ACTIVE:
 232                 return "active";
 233         case HANGCHECK_KICK:
 234                 return "kick";
 235         case HANGCHECK_HUNG:
 236                 return "hung";
 237         }
 238
 239         return "unknown";
 240 }
 241
 242 static void i915_ring_error_state(struct drm_i915_error_state_buf *m,
 243                                   struct drm_device *dev,
 244                                   struct drm_i915_error_state *error,
 245                                   int ring_idx)
 246 {
 247         struct drm_i915_error_ring *ring = &error->ring[ring_idx];
 248
 249         if (!ring->valid)
 250                 return;
 251
 252         err_printf(m, "%s command stream:\n", ring_str(ring_idx));
 253         err_printf(m, "  START: 0x%08x\n", ring->start);
 254         err_printf(m, "  HEAD:  0x%08x\n", ring->head);
 255         err_printf(m, "  TAIL:  0x%08x\n", ring->tail);
 256         err_printf(m, "  CTL:   0x%08x\n", ring->ctl);
 257         err_printf(m, "  HWS:   0x%08x\n", ring->hws);
 258         err_printf(m, "  ACTHD: 0x%08x %08x\n", (u32)(ring->acthd>>32), (u32)ring->acthd);
 259         err_printf(m, "  IPEIR: 0x%08x\n", ring->ipeir);
 260         err_printf(m, "  IPEHR: 0x%08x\n", ring->ipehr);
 261         err_printf(m, "  INSTDONE: 0x%08x\n", ring->instdone);
 262         if (INTEL_INFO(dev)->gen >= 4) {
 263                 err_printf(m, "  BBADDR: 0x%08x %08x\n", (u32)(ring->bbaddr>>32), (u32)ring->bbaddr);
 264                 err_printf(m, "  BB_STATE: 0x%08x\n", ring->bbstate);
 265                 err_printf(m, "  INSTPS: 0x%08x\n", ring->instps);
 266         }
 267         err_printf(m, "  INSTPM: 0x%08x\n", ring->instpm);
 268         err_printf(m, "  FADDR: 0x%08x %08x\n", upper_32_bits(ring->faddr),
 269                    lower_32_bits(ring->faddr));
 270         if (INTEL_INFO(dev)->gen >= 6) {
 271                 err_printf(m, "  RC PSMI: 0x%08x\n", ring->rc_psmi);
 272                 err_printf(m, "  FAULT_REG: 0x%08x\n", ring->fault_reg);
 273                 err_printf(m, "  SYNC_0: 0x%08x [last synced 0x%08x]\n",
 274                            ring->semaphore_mboxes[0],
 275                            ring->semaphore_seqno[0]);
 276                 err_printf(m, "  SYNC_1: 0x%08x [last synced 0x%08x]\n",
 277                            ring->semaphore_mboxes[1],
 278                            ring->semaphore_seqno[1]);
 279                 if (HAS_VEBOX(dev)) {
 280                         err_printf(m, "  SYNC_2: 0x%08x [last synced 0x%08x]\n",
 281                                    ring->semaphore_mboxes[2],
 282                                    ring->semaphore_seqno[2]);
 283                 }
 284         }
 285         if (USES_PPGTT(dev)) {
 286                 err_printf(m, "  GFX_MODE: 0x%08x\n", ring->vm_info.gfx_mode);
 287
 288                 if (INTEL_INFO(dev)->gen >= 8) {
 289                         int i;
 290                         for (i = 0; i < 4; i++)
 291                                 err_printf(m, "  PDP%d: 0x%016llx\n",
 292                                            i, ring->vm_info.pdp[i]);
 293                 } else {
 294                         err_printf(m, "  PP_DIR_BASE: 0x%08x\n",
 295                                    ring->vm_info.pp_dir_base);
 296                 }
 297         }
 298         err_printf(m, "  seqno: 0x%08x\n", ring->seqno);
 299         err_printf(m, "  last_seqno: 0x%08x\n", ring->last_seqno);
 300         err_printf(m, "  waiting: %s\n", yesno(ring->waiting));
 301         err_printf(m, "  ring->head: 0x%08x\n", ring->cpu_ring_head);
 302         err_printf(m, "  ring->tail: 0x%08x\n", ring->cpu_ring_tail);
 303         err_printf(m, "  hangcheck: %s [%d]\n",
 304                    hangcheck_action_to_str(ring->hangcheck_action),
 305                    ring->hangcheck_score);
 306 }
 307
 308 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
 309 {
 310         va_list args;
 311
 312         va_start(args, f);
 313         i915_error_vprintf(e, f, args);
 314         va_end(args);
 315 }
 316
 317 static void print_error_obj(struct drm_i915_error_state_buf *m,
 318                             struct drm_i915_error_object *obj)
 319 {
 320         int page, offset, elt;
 321
 322         for (page = offset = 0; page < obj->page_count; page++) {
 323                 for (elt = 0; elt < PAGE_SIZE/4; elt++) {
 324                         err_printf(m, "%08x :  %08x\n", offset,
 325                                    obj->pages[page][elt]);
 326                         offset += 4;
 327                 }
 328         }
 329 }
 330
 331 int i915_error_state_to_str(struct drm_i915_error_state_buf *m,
 332                             const struct i915_error_state_file_priv *error_priv)
 333 {
 334         struct drm_device *dev = error_priv->dev;
 335         struct drm_i915_private *dev_priv = dev->dev_private;
 336         struct drm_i915_error_state *error = error_priv->error;
 337         struct drm_i915_error_object *obj;
 338         int i, j, offset, elt;
 339         int max_hangcheck_score;
 340
 341         if (!error) {
 342                 err_printf(m, "no error state collected\n");
 343                 goto out;
 344         }
 345
 346         err_printf(m, "%s\n", error->error_msg);
 347         err_printf(m, "Time: %ld s %ld us\n", error->time.tv_sec,
 348                    error->time.tv_usec);
 349         err_printf(m, "Kernel: " UTS_RELEASE "\n");
 350         max_hangcheck_score = 0;
 351         for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
 352                 if (error->ring[i].hangcheck_score > max_hangcheck_score)
 353                         max_hangcheck_score = error->ring[i].hangcheck_score;
 354         }
 355         for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
 356                 if (error->ring[i].hangcheck_score == max_hangcheck_score &&
 357                     error->ring[i].pid != -1) {
 358                         err_printf(m, "Active process (on ring %s): %s [%d]\n",
 359                                    ring_str(i),
 360                                    error->ring[i].comm,
 361                                    error->ring[i].pid);
 362                 }
 363         }
 364         err_printf(m, "Reset count: %u\n", error->reset_count);
 365         err_printf(m, "Suspend count: %u\n", error->suspend_count);
 366         err_printf(m, "PCI ID: 0x%04x\n", dev->pdev->device);
 367         err_printf(m, "PCI Revision: 0x%02x\n", dev->pdev->revision);
 368         err_printf(m, "PCI Subsystem: %04x:%04x\n",
 369                    dev->pdev->subsystem_vendor,
 370                    dev->pdev->subsystem_device);
 371         err_printf(m, "IOMMU enabled?: %d\n", error->iommu);
 372
 373         if (HAS_CSR(dev)) {
 374                 struct intel_csr *csr = &dev_priv->csr;
 375
 376                 err_printf(m, "DMC loaded: %s\n",
 377                            yesno(csr->dmc_payload != NULL));
 378                 err_printf(m, "DMC fw version: %d.%d\n",
 379                            CSR_VERSION_MAJOR(csr->version),
 380                            CSR_VERSION_MINOR(csr->version));
 381         }
 382
 383         err_printf(m, "EIR: 0x%08x\n", error->eir);
 384         err_printf(m, "IER: 0x%08x\n", error->ier);
 385         if (INTEL_INFO(dev)->gen >= 8) {
 386                 for (i = 0; i < 4; i++)
 387                         err_printf(m, "GTIER gt %d: 0x%08x\n", i,
 388                                    error->gtier[i]);
 389         } else if (HAS_PCH_SPLIT(dev) || IS_VALLEYVIEW(dev))
 390                 err_printf(m, "GTIER: 0x%08x\n", error->gtier[0]);
 391         err_printf(m, "PGTBL_ER: 0x%08x\n", error->pgtbl_er);
 392         err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake);
 393         err_printf(m, "DERRMR: 0x%08x\n", error->derrmr);
 394         err_printf(m, "CCID: 0x%08x\n", error->ccid);
 395         err_printf(m, "Missed interrupts: 0x%08lx\n", dev_priv->gpu_error.missed_irq_rings);
 396
 397         for (i = 0; i < dev_priv->num_fence_regs; i++)
 398                 err_printf(m, "  fence[%d] = %08llx\n", i, error->fence[i]);
 399
 400         for (i = 0; i < ARRAY_SIZE(error->extra_instdone); i++)
 401                 err_printf(m, "  INSTDONE_%d: 0x%08x\n", i,
 402                            error->extra_instdone[i]);
 403
 404         if (INTEL_INFO(dev)->gen >= 6) {
 405                 err_printf(m, "ERROR: 0x%08x\n", error->error);
 406
 407                 if (INTEL_INFO(dev)->gen >= 8)
 408                         err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n",
 409                                    error->fault_data1, error->fault_data0);
 410
 411                 err_printf(m, "DONE_REG: 0x%08x\n", error->done_reg);
 412         }
 413
 414         if (INTEL_INFO(dev)->gen == 7)
 415                 err_printf(m, "ERR_INT: 0x%08x\n", error->err_int);
 416
 417         for (i = 0; i < ARRAY_SIZE(error->ring); i++)
 418                 i915_ring_error_state(m, dev, error, i);
 419
 420         for (i = 0; i < error->vm_count; i++) {
 421                 err_printf(m, "vm[%d]\n", i);
 422
 423                 print_error_buffers(m, "Active",
 424                                     error->active_bo[i],
 425                                     error->active_bo_count[i]);
 426
 427                 print_error_buffers(m, "Pinned",
 428                                     error->pinned_bo[i],
 429                                     error->pinned_bo_count[i]);
 430         }
 431
 432         for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
 433                 obj = error->ring[i].batchbuffer;
 434                 if (obj) {
 435                         err_puts(m, dev_priv->engine[i].name);
 436                         if (error->ring[i].pid != -1)
 437                                 err_printf(m, " (submitted by %s [%d])",
 438                                            error->ring[i].comm,
 439                                            error->ring[i].pid);
 440                         err_printf(m, " --- gtt_offset = 0x%08x %08x\n",
 441                                    upper_32_bits(obj->gtt_offset),
 442                                    lower_32_bits(obj->gtt_offset));
 443                         print_error_obj(m, obj);
 444                 }
 445
 446                 obj = error->ring[i].wa_batchbuffer;
 447                 if (obj) {
 448                         err_printf(m, "%s (w/a) --- gtt_offset = 0x%08x\n",
 449                                    dev_priv->engine[i].name,
 450                                    lower_32_bits(obj->gtt_offset));
 451                         print_error_obj(m, obj);
 452                 }
 453
 454                 if (error->ring[i].num_requests) {
 455                         err_printf(m, "%s --- %d requests\n",
 456                                    dev_priv->engine[i].name,
 457                                    error->ring[i].num_requests);
 458                         for (j = 0; j < error->ring[i].num_requests; j++) {
 459                                 err_printf(m, "  seqno 0x%08x, emitted %ld, tail 0x%08x\n",
 460                                            error->ring[i].requests[j].seqno,
 461                                            error->ring[i].requests[j].jiffies,
 462                                            error->ring[i].requests[j].tail);
 463                         }
 464                 }
 465
 466                 if ((obj = error->ring[i].ringbuffer)) {
 467                         err_printf(m, "%s --- ringbuffer = 0x%08x\n",
 468                                    dev_priv->engine[i].name,
 469                                    lower_32_bits(obj->gtt_offset));
 470                         print_error_obj(m, obj);
 471                 }
 472
 473                 if ((obj = error->ring[i].hws_page)) {
 474                         u64 hws_offset = obj->gtt_offset;
 475                         u32 *hws_page = &obj->pages[0][0];
 476
 477                         if (i915.enable_execlists) {
 478                                 hws_offset += LRC_PPHWSP_PN * PAGE_SIZE;
 479                                 hws_page = &obj->pages[LRC_PPHWSP_PN][0];
 480                         }
 481                         err_printf(m, "%s --- HW Status = 0x%08llx\n",
 482                                    dev_priv->engine[i].name, hws_offset);
 483                         offset = 0;
 484                         for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
 485                                 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
 486                                            offset,
 487                                            hws_page[elt],
 488                                            hws_page[elt+1],
 489                                            hws_page[elt+2],
 490                                            hws_page[elt+3]);
 491                                         offset += 16;
 492                         }
 493                 }
 494
 495                 obj = error->ring[i].wa_ctx;
 496                 if (obj) {
 497                         u64 wa_ctx_offset = obj->gtt_offset;
 498                         u32 *wa_ctx_page = &obj->pages[0][0];
 499                         struct intel_engine_cs *engine = &dev_priv->engine[RCS];
 500                         u32 wa_ctx_size = (engine->wa_ctx.indirect_ctx.size +
 501                                            engine->wa_ctx.per_ctx.size);
 502
 503                         err_printf(m, "%s --- WA ctx batch buffer = 0x%08llx\n",
 504                                    dev_priv->engine[i].name, wa_ctx_offset);
 505                         offset = 0;
 506                         for (elt = 0; elt < wa_ctx_size; elt += 4) {
 507                                 err_printf(m, "[%04x] %08x %08x %08x %08x\n",
 508                                            offset,
 509                                            wa_ctx_page[elt + 0],
 510                                            wa_ctx_page[elt + 1],
 511                                            wa_ctx_page[elt + 2],
 512                                            wa_ctx_page[elt + 3]);
 513                                 offset += 16;
 514                         }
 515                 }
 516
 517                 if ((obj = error->ring[i].ctx)) {
 518                         err_printf(m, "%s --- HW Context = 0x%08x\n",
 519                                    dev_priv->engine[i].name,
 520                                    lower_32_bits(obj->gtt_offset));
 521                         print_error_obj(m, obj);
 522                 }
 523         }
 524
 525         if ((obj = error->semaphore_obj)) {
 526                 err_printf(m, "Semaphore page = 0x%08x\n",
 527                            lower_32_bits(obj->gtt_offset));
 528                 for (elt = 0; elt < PAGE_SIZE/16; elt += 4) {
 529                         err_printf(m, "[%04x] %08x %08x %08x %08x\n",
 530                                    elt * 4,
 531                                    obj->pages[0][elt],
 532                                    obj->pages[0][elt+1],
 533                                    obj->pages[0][elt+2],
 534                                    obj->pages[0][elt+3]);
 535                 }
 536         }
 537
 538         if (error->overlay)
 539                 intel_overlay_print_error_state(m, error->overlay);
 540
 541         if (error->display)
 542                 intel_display_print_error_state(m, dev, error->display);
 543
 544 out:
 545         if (m->bytes == 0 && m->err)
 546                 return m->err;
 547
 548         return 0;
 549 }
 550
 551 int i915_error_state_buf_init(struct drm_i915_error_state_buf *ebuf,
 552                               struct drm_i915_private *i915,
 553                               size_t count, loff_t pos)
 554 {
 555         memset(ebuf, 0, sizeof(*ebuf));
 556         ebuf->i915 = i915;
 557
 558         /* We need to have enough room to store any i915_error_state printf
 559          * so that we can move it to start position.
 560          */
 561         ebuf->size = count + 1 > PAGE_SIZE ? count + 1 : PAGE_SIZE;
 562         ebuf->buf = kmalloc(ebuf->size,
 563                                 GFP_TEMPORARY | __GFP_NORETRY | __GFP_NOWARN);
 564
 565         if (ebuf->buf == NULL) {
 566                 ebuf->size = PAGE_SIZE;
 567                 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
 568         }
 569
 570         if (ebuf->buf == NULL) {
 571                 ebuf->size = 128;
 572                 ebuf->buf = kmalloc(ebuf->size, GFP_TEMPORARY);
 573         }
 574
 575         if (ebuf->buf == NULL)
 576                 return -ENOMEM;
 577
 578         ebuf->start = pos;
 579
 580         return 0;
 581 }
 582
 583 static void i915_error_object_free(struct drm_i915_error_object *obj)
 584 {
 585         int page;
 586
 587         if (obj == NULL)
 588                 return;
 589
 590         for (page = 0; page < obj->page_count; page++)
 591                 kfree(obj->pages[page]);
 592
 593         kfree(obj);
 594 }
 595
 596 static void i915_error_state_free(struct kref *error_ref)
 597 {
 598         struct drm_i915_error_state *error = container_of(error_ref,
 599                                                           typeof(*error), ref);
 600         int i;
 601
 602         for (i = 0; i < ARRAY_SIZE(error->ring); i++) {
 603                 i915_error_object_free(error->ring[i].batchbuffer);
 604                 i915_error_object_free(error->ring[i].wa_batchbuffer);
 605                 i915_error_object_free(error->ring[i].ringbuffer);
 606                 i915_error_object_free(error->ring[i].hws_page);
 607                 i915_error_object_free(error->ring[i].ctx);
 608                 kfree(error->ring[i].requests);
 609                 i915_error_object_free(error->ring[i].wa_ctx);
 610         }
 611
 612         i915_error_object_free(error->semaphore_obj);
 613
 614         for (i = 0; i < error->vm_count; i++)
 615                 kfree(error->active_bo[i]);
 616
 617         kfree(error->active_bo);
 618         kfree(error->active_bo_count);
 619         kfree(error->pinned_bo);
 620         kfree(error->pinned_bo_count);
 621         kfree(error->overlay);
 622         kfree(error->display);
 623         kfree(error);
 624 }
 625
 626 static struct drm_i915_error_object *
 627 i915_error_object_create(struct drm_i915_private *dev_priv,
 628                          struct drm_i915_gem_object *src,
 629                          struct i915_address_space *vm)
 630 {
 631         struct i915_ggtt *ggtt = &dev_priv->ggtt;
 632         struct drm_i915_error_object *dst;
 633         struct i915_vma *vma = NULL;
 634         int num_pages;
 635         bool use_ggtt;
 636         int i = 0;
 637         u64 reloc_offset;
 638
 639         if (src == NULL || src->pages == NULL)
 640                 return NULL;
 641
 642         num_pages = src->base.size >> PAGE_SHIFT;
 643
 644         dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), GFP_ATOMIC);
 645         if (dst == NULL)
 646                 return NULL;
 647
 648         if (i915_gem_obj_bound(src, vm))
 649                 dst->gtt_offset = i915_gem_obj_offset(src, vm);
 650         else
 651                 dst->gtt_offset = -1;
 652
 653         reloc_offset = dst->gtt_offset;
 654         if (i915_is_ggtt(vm))
 655                 vma = i915_gem_obj_to_ggtt(src);
 656         use_ggtt = (src->cache_level == I915_CACHE_NONE &&
 657                    vma && (vma->bound & GLOBAL_BIND) &&
 658                    reloc_offset + num_pages * PAGE_SIZE <= ggtt->mappable_end);
 659
 660         /* Cannot access stolen address directly, try to use the aperture */
 661         if (src->stolen) {
 662                 use_ggtt = true;
 663
 664                 if (!(vma && vma->bound & GLOBAL_BIND))
 665                         goto unwind;
 666
 667                 reloc_offset = i915_gem_obj_ggtt_offset(src);
 668                 if (reloc_offset + num_pages * PAGE_SIZE > ggtt->mappable_end)
 669                         goto unwind;
 670         }
 671
 672         /* Cannot access snooped pages through the aperture */
 673         if (use_ggtt && src->cache_level != I915_CACHE_NONE &&
 674             !HAS_LLC(dev_priv))
 675                 goto unwind;
 676
 677         dst->page_count = num_pages;
 678         while (num_pages--) {
 679                 unsigned long flags;
 680                 void *d;
 681
 682                 d = kmalloc(PAGE_SIZE, GFP_ATOMIC);
 683                 if (d == NULL)
 684                         goto unwind;
 685
 686                 local_irq_save(flags);
 687                 if (use_ggtt) {
 688                         void __iomem *s;
 689
 690                         /* Simply ignore tiling or any overlapping fence.
 691                          * It's part of the error state, and this hopefully
 692                          * captures what the GPU read.
 693                          */
 694
 695                         s = io_mapping_map_atomic_wc(ggtt->mappable,
 696                                                      reloc_offset);
 697                         memcpy_fromio(d, s, PAGE_SIZE);
 698                         io_mapping_unmap_atomic(s);
 699                 } else {
 700                         struct page *page;
 701                         void *s;
 702
 703                         page = i915_gem_object_get_page(src, i);
 704
 705                         drm_clflush_pages(&page, 1);
 706
 707                         s = kmap_atomic(page);
 708                         memcpy(d, s, PAGE_SIZE);
 709                         kunmap_atomic(s);
 710
 711                         drm_clflush_pages(&page, 1);
 712                 }
 713                 local_irq_restore(flags);
 714
 715                 dst->pages[i++] = d;
 716                 reloc_offset += PAGE_SIZE;
 717         }
 718
 719         return dst;
 720
 721 unwind:
 722         while (i--)
 723                 kfree(dst->pages[i]);
 724         kfree(dst);
 725         return NULL;
 726 }
 727 #define i915_error_ggtt_object_create(dev_priv, src) \
 728         i915_error_object_create((dev_priv), (src), &(dev_priv)->ggtt.base)
 729
 730 static void capture_bo(struct drm_i915_error_buffer *err,
 731                        struct i915_vma *vma)
 732 {
 733         struct drm_i915_gem_object *obj = vma->obj;
 734         int i;
 735
 736         err->size = obj->base.size;
 737         err->name = obj->base.name;
 738         for (i = 0; i < I915_NUM_ENGINES; i++)
 739                 err->rseqno[i] = i915_gem_request_get_seqno(obj->last_read_req[i]);
 740         err->wseqno = i915_gem_request_get_seqno(obj->last_write_req);
 741         err->gtt_offset = vma->node.start;
 742         err->read_domains = obj->base.read_domains;
 743         err->write_domain = obj->base.write_domain;
 744         err->fence_reg = obj->fence_reg;
 745         err->pinned = 0;
 746         if (i915_gem_obj_is_pinned(obj))
 747                 err->pinned = 1;
 748         err->tiling = obj->tiling_mode;
 749         err->dirty = obj->dirty;
 750         err->purgeable = obj->madv != I915_MADV_WILLNEED;
 751         err->userptr = obj->userptr.mm != NULL;
 752         err->ring = obj->last_write_req ?
 753                         i915_gem_request_get_engine(obj->last_write_req)->id : -1;
 754         err->cache_level = obj->cache_level;
 755 }
 756
 757 static u32 capture_active_bo(struct drm_i915_error_buffer *err,
 758                              int count, struct list_head *head)
 759 {
 760         struct i915_vma *vma;
 761         int i = 0;
 762
 763         list_for_each_entry(vma, head, vm_link) {
 764                 capture_bo(err++, vma);
 765                 if (++i == count)
 766                         break;
 767         }
 768
 769         return i;
 770 }
 771
 772 static u32 capture_pinned_bo(struct drm_i915_error_buffer *err,
 773                              int count, struct list_head *head,
 774                              struct i915_address_space *vm)
 775 {
 776         struct drm_i915_gem_object *obj;
 777         struct drm_i915_error_buffer * const first = err;
 778         struct drm_i915_error_buffer * const last = err + count;
 779
 780         list_for_each_entry(obj, head, global_list) {
 781                 struct i915_vma *vma;
 782
 783                 if (err == last)
 784                         break;
 785
 786                 list_for_each_entry(vma, &obj->vma_list, obj_link)
 787                         if (vma->vm == vm && vma->pin_count > 0)
 788                                 capture_bo(err++, vma);
 789         }
 790
 791         return err - first;
 792 }
 793
 794 /* Generate a semi-unique error code. The code is not meant to have meaning, The
 795  * code's only purpose is to try to prevent false duplicated bug reports by
 796  * grossly estimating a GPU error state.
 797  *
 798  * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
 799  * the hang if we could strip the GTT offset information from it.
 800  *
 801  * It's only a small step better than a random number in its current form.
 802  */
 803 static uint32_t i915_error_generate_code(struct drm_i915_private *dev_priv,
 804                                          struct drm_i915_error_state *error,
 805                                          int *ring_id)
 806 {
 807         uint32_t error_code = 0;
 808         int i;
 809
 810         /* IPEHR would be an ideal way to detect errors, as it's the gross
 811          * measure of "the command that hung." However, has some very common
 812          * synchronization commands which almost always appear in the case
 813          * strictly a client bug. Use instdone to differentiate those some.
 814          */
 815         for (i = 0; i < I915_NUM_ENGINES; i++) {
 816                 if (error->ring[i].hangcheck_action == HANGCHECK_HUNG) {
 817                         if (ring_id)
 818                                 *ring_id = i;
 819
 820                         return error->ring[i].ipehr ^ error->ring[i].instdone;
 821                 }
 822         }
 823
 824         return error_code;
 825 }
 826
 827 static void i915_gem_record_fences(struct drm_device *dev,
 828                                    struct drm_i915_error_state *error)
 829 {
 830         struct drm_i915_private *dev_priv = dev->dev_private;
 831         int i;
 832
 833         if (IS_GEN3(dev) || IS_GEN2(dev)) {
 834                 for (i = 0; i < dev_priv->num_fence_regs; i++)
 835                         error->fence[i] = I915_READ(FENCE_REG(i));
 836         } else if (IS_GEN5(dev) || IS_GEN4(dev)) {
 837                 for (i = 0; i < dev_priv->num_fence_regs; i++)
 838                         error->fence[i] = I915_READ64(FENCE_REG_965_LO(i));
 839         } else if (INTEL_INFO(dev)->gen >= 6) {
 840                 for (i = 0; i < dev_priv->num_fence_regs; i++)
 841                         error->fence[i] = I915_READ64(FENCE_REG_GEN6_LO(i));
 842         }
 843 }
 844
 845
 846 static void gen8_record_semaphore_state(struct drm_i915_private *dev_priv,
 847                                         struct drm_i915_error_state *error,
 848                                         struct intel_engine_cs *engine,
 849                                         struct drm_i915_error_ring *ering)
 850 {
 851         struct intel_engine_cs *to;
 852         enum intel_engine_id id;
 853
 854         if (!i915_semaphore_is_enabled(dev_priv->dev))
 855                 return;
 856
 857         if (!error->semaphore_obj)
 858                 error->semaphore_obj =
 859                         i915_error_ggtt_object_create(dev_priv,
 860                                                       dev_priv->semaphore_obj);
 861
 862         for_each_engine_id(to, dev_priv, id) {
 863                 int idx;
 864                 u16 signal_offset;
 865                 u32 *tmp;
 866
 867                 if (engine == to)
 868                         continue;
 869
 870                 signal_offset = (GEN8_SIGNAL_OFFSET(engine, id) & (PAGE_SIZE - 1))
 871                                 / 4;
 872                 tmp = error->semaphore_obj->pages[0];
 873                 idx = intel_ring_sync_index(engine, to);
 874
 875                 ering->semaphore_mboxes[idx] = tmp[signal_offset];
 876                 ering->semaphore_seqno[idx] = engine->semaphore.sync_seqno[idx];
 877         }
 878 }
 879
 880 static void gen6_record_semaphore_state(struct drm_i915_private *dev_priv,
 881                                         struct intel_engine_cs *engine,
 882                                         struct drm_i915_error_ring *ering)
 883 {
 884         ering->semaphore_mboxes[0] = I915_READ(RING_SYNC_0(engine->mmio_base));
 885         ering->semaphore_mboxes[1] = I915_READ(RING_SYNC_1(engine->mmio_base));
 886         ering->semaphore_seqno[0] = engine->semaphore.sync_seqno[0];
 887         ering->semaphore_seqno[1] = engine->semaphore.sync_seqno[1];
 888
 889         if (HAS_VEBOX(dev_priv)) {
 890                 ering->semaphore_mboxes[2] =
 891                         I915_READ(RING_SYNC_2(engine->mmio_base));
 892                 ering->semaphore_seqno[2] = engine->semaphore.sync_seqno[2];
 893         }
 894 }
 895
 896 static void i915_record_ring_state(struct drm_device *dev,
 897                                    struct drm_i915_error_state *error,
 898                                    struct intel_engine_cs *engine,
 899                                    struct drm_i915_error_ring *ering)
 900 {
 901         struct drm_i915_private *dev_priv = dev->dev_private;
 902
 903         if (INTEL_INFO(dev)->gen >= 6) {
 904                 ering->rc_psmi = I915_READ(RING_PSMI_CTL(engine->mmio_base));
 905                 ering->fault_reg = I915_READ(RING_FAULT_REG(engine));
 906                 if (INTEL_INFO(dev)->gen >= 8)
 907                         gen8_record_semaphore_state(dev_priv, error, engine,
 908                                                     ering);
 909                 else
 910                         gen6_record_semaphore_state(dev_priv, engine, ering);
 911         }
 912
 913         if (INTEL_INFO(dev)->gen >= 4) {
 914                 ering->faddr = I915_READ(RING_DMA_FADD(engine->mmio_base));
 915                 ering->ipeir = I915_READ(RING_IPEIR(engine->mmio_base));
 916                 ering->ipehr = I915_READ(RING_IPEHR(engine->mmio_base));
 917                 ering->instdone = I915_READ(RING_INSTDONE(engine->mmio_base));
 918                 ering->instps = I915_READ(RING_INSTPS(engine->mmio_base));
 919                 ering->bbaddr = I915_READ(RING_BBADDR(engine->mmio_base));
 920                 if (INTEL_INFO(dev)->gen >= 8) {
 921                         ering->faddr |= (u64) I915_READ(RING_DMA_FADD_UDW(engine->mmio_base)) << 32;
 922                         ering->bbaddr |= (u64) I915_READ(RING_BBADDR_UDW(engine->mmio_base)) << 32;
 923                 }
 924                 ering->bbstate = I915_READ(RING_BBSTATE(engine->mmio_base));
 925         } else {
 926                 ering->faddr = I915_READ(DMA_FADD_I8XX);
 927                 ering->ipeir = I915_READ(IPEIR);
 928                 ering->ipehr = I915_READ(IPEHR);
 929                 ering->instdone = I915_READ(GEN2_INSTDONE);
 930         }
 931
 932         ering->waiting = waitqueue_active(&engine->irq_queue);
 933         ering->instpm = I915_READ(RING_INSTPM(engine->mmio_base));
 934         ering->acthd = intel_ring_get_active_head(engine);
 935         ering->seqno = engine->get_seqno(engine);
 936         ering->last_seqno = engine->last_submitted_seqno;
 937         ering->start = I915_READ_START(engine);
 938         ering->head = I915_READ_HEAD(engine);
 939         ering->tail = I915_READ_TAIL(engine);
 940         ering->ctl = I915_READ_CTL(engine);
 941
 942         if (I915_NEED_GFX_HWS(dev)) {
 943                 i915_reg_t mmio;
 944
 945                 if (IS_GEN7(dev)) {
 946                         switch (engine->id) {
 947                         default:
 948                         case RCS:
 949                                 mmio = RENDER_HWS_PGA_GEN7;
 950                                 break;
 951                         case BCS:
 952                                 mmio = BLT_HWS_PGA_GEN7;
 953                                 break;
 954                         case VCS:
 955                                 mmio = BSD_HWS_PGA_GEN7;
 956                                 break;
 957                         case VECS:
 958                                 mmio = VEBOX_HWS_PGA_GEN7;
 959                                 break;
 960                         }
 961                 } else if (IS_GEN6(engine->dev)) {
 962                         mmio = RING_HWS_PGA_GEN6(engine->mmio_base);
 963                 } else {
 964                         /* XXX: gen8 returns to sanity */
 965                         mmio = RING_HWS_PGA(engine->mmio_base);
 966                 }
 967
 968                 ering->hws = I915_READ(mmio);
 969         }
 970
 971         ering->hangcheck_score = engine->hangcheck.score;
 972         ering->hangcheck_action = engine->hangcheck.action;
 973
 974         if (USES_PPGTT(dev)) {
 975                 int i;
 976
 977                 ering->vm_info.gfx_mode = I915_READ(RING_MODE_GEN7(engine));
 978
 979                 if (IS_GEN6(dev))
 980                         ering->vm_info.pp_dir_base =
 981                                 I915_READ(RING_PP_DIR_BASE_READ(engine));
 982                 else if (IS_GEN7(dev))
 983                         ering->vm_info.pp_dir_base =
 984                                 I915_READ(RING_PP_DIR_BASE(engine));
 985                 else if (INTEL_INFO(dev)->gen >= 8)
 986                         for (i = 0; i < 4; i++) {
 987                                 ering->vm_info.pdp[i] =
 988                                         I915_READ(GEN8_RING_PDP_UDW(engine, i));
 989                                 ering->vm_info.pdp[i] <<= 32;
 990                                 ering->vm_info.pdp[i] |=
 991                                         I915_READ(GEN8_RING_PDP_LDW(engine, i));
 992                         }
 993         }
 994 }
 995
 996
 997 static void i915_gem_record_active_context(struct intel_engine_cs *engine,
 998                                            struct drm_i915_error_state *error,
 999                                            struct drm_i915_error_ring *ering)
1000 {
1001         struct drm_i915_private *dev_priv = engine->dev->dev_private;
1002         struct drm_i915_gem_object *obj;
1003
1004         /* Currently render ring is the only HW context user */
1005         if (engine->id != RCS || !error->ccid)
1006                 return;
1007
1008         list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
1009                 if (!i915_gem_obj_ggtt_bound(obj))
1010                         continue;
1011
1012                 if ((error->ccid & PAGE_MASK) == i915_gem_obj_ggtt_offset(obj)) {
1013                         ering->ctx = i915_error_ggtt_object_create(dev_priv, obj);
1014                         break;
1015                 }
1016         }
1017 }
1018
1019 static void i915_gem_record_rings(struct drm_device *dev,
1020                                   struct drm_i915_error_state *error)
1021 {
1022         struct drm_i915_private *dev_priv = to_i915(dev);
1023         struct i915_ggtt *ggtt = &dev_priv->ggtt;
1024         struct drm_i915_gem_request *request;
1025         int i, count;
1026
1027         for (i = 0; i < I915_NUM_ENGINES; i++) {
1028                 struct intel_engine_cs *engine = &dev_priv->engine[i];
1029                 struct intel_ringbuffer *rbuf;
1030
1031                 error->ring[i].pid = -1;
1032
1033                 if (engine->dev == NULL)
1034                         continue;
1035
1036                 error->ring[i].valid = true;
1037
1038                 i915_record_ring_state(dev, error, engine, &error->ring[i]);
1039
1040                 request = i915_gem_find_active_request(engine);
1041                 if (request) {
1042                         struct i915_address_space *vm;
1043
1044                         vm = request->ctx && request->ctx->ppgtt ?
1045                                 &request->ctx->ppgtt->base :
1046                                 &ggtt->base;
1047
1048                         /* We need to copy these to an anonymous buffer
1049                          * as the simplest method to avoid being overwritten
1050                          * by userspace.
1051                          */
1052                         error->ring[i].batchbuffer =
1053                                 i915_error_object_create(dev_priv,
1054                                                          request->batch_obj,
1055                                                          vm);
1056
1057                         if (HAS_BROKEN_CS_TLB(dev_priv))
1058                                 error->ring[i].wa_batchbuffer =
1059                                         i915_error_ggtt_object_create(dev_priv,
1060                                                              engine->scratch.obj);
1061
1062                         if (request->pid) {
1063                                 struct task_struct *task;
1064
1065                                 rcu_read_lock();
1066                                 task = pid_task(request->pid, PIDTYPE_PID);
1067                                 if (task) {
1068                                         strcpy(error->ring[i].comm, task->comm);
1069                                         error->ring[i].pid = task->pid;
1070                                 }
1071                                 rcu_read_unlock();
1072                         }
1073                 }
1074
1075                 if (i915.enable_execlists) {
1076                         /* TODO: This is only a small fix to keep basic error
1077                          * capture working, but we need to add more information
1078                          * for it to be useful (e.g. dump the context being
1079                          * executed).
1080                          */
1081                         if (request)
1082                                 rbuf = request->ctx->engine[engine->id].ringbuf;
1083                         else
1084                                 rbuf = dev_priv->kernel_context->engine[engine->id].ringbuf;
1085                 } else
1086                         rbuf = engine->buffer;
1087
1088                 error->ring[i].cpu_ring_head = rbuf->head;
1089                 error->ring[i].cpu_ring_tail = rbuf->tail;
1090
1091                 error->ring[i].ringbuffer =
1092                         i915_error_ggtt_object_create(dev_priv, rbuf->obj);
1093
1094                 error->ring[i].hws_page =
1095                         i915_error_ggtt_object_create(dev_priv,
1096                                                       engine->status_page.obj);
1097
1098                 if (engine->wa_ctx.obj) {
1099                         error->ring[i].wa_ctx =
1100                                 i915_error_ggtt_object_create(dev_priv,
1101                                                               engine->wa_ctx.obj);
1102                 }
1103
1104                 i915_gem_record_active_context(engine, error, &error->ring[i]);
1105
1106                 count = 0;
1107                 list_for_each_entry(request, &engine->request_list, list)
1108                         count++;
1109
1110                 error->ring[i].num_requests = count;
1111                 error->ring[i].requests =
1112                         kcalloc(count, sizeof(*error->ring[i].requests),
1113                                 GFP_ATOMIC);
1114                 if (error->ring[i].requests == NULL) {
1115                         error->ring[i].num_requests = 0;
1116                         continue;
1117                 }
1118
1119                 count = 0;
1120                 list_for_each_entry(request, &engine->request_list, list) {
1121                         struct drm_i915_error_request *erq;
1122
1123                         if (count >= error->ring[i].num_requests) {
1124                                 /*
1125                                  * If the ring request list was changed in
1126                                  * between the point where the error request
1127                                  * list was created and dimensioned and this
1128                                  * point then just exit early to avoid crashes.
1129                                  *
1130                                  * We don't need to communicate that the
1131                                  * request list changed state during error
1132                                  * state capture and that the error state is
1133                                  * slightly incorrect as a consequence since we
1134                                  * are typically only interested in the request
1135                                  * list state at the point of error state
1136                                  * capture, not in any changes happening during
1137                                  * the capture.
1138                                  */
1139                                 break;
1140                         }
1141
1142                         erq = &error->ring[i].requests[count++];
1143                         erq->seqno = request->seqno;
1144                         erq->jiffies = request->emitted_jiffies;
1145                         erq->tail = request->postfix;
1146                 }
1147         }
1148 }
1149
1150 /* FIXME: Since pin count/bound list is global, we duplicate what we capture per
1151  * VM.
1152  */
1153 static void i915_gem_capture_vm(struct drm_i915_private *dev_priv,
1154                                 struct drm_i915_error_state *error,
1155                                 struct i915_address_space *vm,
1156                                 const int ndx)
1157 {
1158         struct drm_i915_error_buffer *active_bo = NULL, *pinned_bo = NULL;
1159         struct drm_i915_gem_object *obj;
1160         struct i915_vma *vma;
1161         int i;
1162
1163         i = 0;
1164         list_for_each_entry(vma, &vm->active_list, vm_link)
1165                 i++;
1166         error->active_bo_count[ndx] = i;
1167
1168         list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
1169                 list_for_each_entry(vma, &obj->vma_list, obj_link)
1170                         if (vma->vm == vm && vma->pin_count > 0)
1171                                 i++;
1172         }
1173         error->pinned_bo_count[ndx] = i - error->active_bo_count[ndx];
1174
1175         if (i) {
1176                 active_bo = kcalloc(i, sizeof(*active_bo), GFP_ATOMIC);
1177                 if (active_bo)
1178                         pinned_bo = active_bo + error->active_bo_count[ndx];
1179         }
1180
1181         if (active_bo)
1182                 error->active_bo_count[ndx] =
1183                         capture_active_bo(active_bo,
1184                                           error->active_bo_count[ndx],
1185                                           &vm->active_list);
1186
1187         if (pinned_bo)
1188                 error->pinned_bo_count[ndx] =
1189                         capture_pinned_bo(pinned_bo,
1190                                           error->pinned_bo_count[ndx],
1191                                           &dev_priv->mm.bound_list, vm);
1192         error->active_bo[ndx] = active_bo;
1193         error->pinned_bo[ndx] = pinned_bo;
1194 }
1195
1196 static void i915_gem_capture_buffers(struct drm_i915_private *dev_priv,
1197                                      struct drm_i915_error_state *error)
1198 {
1199         struct i915_address_space *vm;
1200         int cnt = 0, i = 0;
1201
1202         list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1203                 cnt++;
1204
1205         error->active_bo = kcalloc(cnt, sizeof(*error->active_bo), GFP_ATOMIC);
1206         error->pinned_bo = kcalloc(cnt, sizeof(*error->pinned_bo), GFP_ATOMIC);
1207         error->active_bo_count = kcalloc(cnt, sizeof(*error->active_bo_count),
1208                                          GFP_ATOMIC);
1209         error->pinned_bo_count = kcalloc(cnt, sizeof(*error->pinned_bo_count),
1210                                          GFP_ATOMIC);
1211
1212         if (error->active_bo == NULL ||
1213             error->pinned_bo == NULL ||
1214             error->active_bo_count == NULL ||
1215             error->pinned_bo_count == NULL) {
1216                 kfree(error->active_bo);
1217                 kfree(error->active_bo_count);
1218                 kfree(error->pinned_bo);
1219                 kfree(error->pinned_bo_count);
1220
1221                 error->active_bo = NULL;
1222                 error->active_bo_count = NULL;
1223                 error->pinned_bo = NULL;
1224                 error->pinned_bo_count = NULL;
1225         } else {
1226                 list_for_each_entry(vm, &dev_priv->vm_list, global_link)
1227                         i915_gem_capture_vm(dev_priv, error, vm, i++);
1228
1229                 error->vm_count = cnt;
1230         }
1231 }
1232
1233 /* Capture all registers which don't fit into another category. */
1234 static void i915_capture_reg_state(struct drm_i915_private *dev_priv,
1235                                    struct drm_i915_error_state *error)
1236 {
1237         struct drm_device *dev = dev_priv->dev;
1238         int i;
1239
1240         /* General organization
1241          * 1. Registers specific to a single generation
1242          * 2. Registers which belong to multiple generations
1243          * 3. Feature specific registers.
1244          * 4. Everything else
1245          * Please try to follow the order.
1246          */
1247
1248         /* 1: Registers specific to a single generation */
1249         if (IS_VALLEYVIEW(dev)) {
1250                 error->gtier[0] = I915_READ(GTIER);
1251                 error->ier = I915_READ(VLV_IER);
1252                 error->forcewake = I915_READ_FW(FORCEWAKE_VLV);
1253         }
1254
1255         if (IS_GEN7(dev))
1256                 error->err_int = I915_READ(GEN7_ERR_INT);
1257
1258         if (INTEL_INFO(dev)->gen >= 8) {
1259                 error->fault_data0 = I915_READ(GEN8_FAULT_TLB_DATA0);
1260                 error->fault_data1 = I915_READ(GEN8_FAULT_TLB_DATA1);
1261         }
1262
1263         if (IS_GEN6(dev)) {
1264                 error->forcewake = I915_READ_FW(FORCEWAKE);
1265                 error->gab_ctl = I915_READ(GAB_CTL);
1266                 error->gfx_mode = I915_READ(GFX_MODE);
1267         }
1268
1269         /* 2: Registers which belong to multiple generations */
1270         if (INTEL_INFO(dev)->gen >= 7)
1271                 error->forcewake = I915_READ_FW(FORCEWAKE_MT);
1272
1273         if (INTEL_INFO(dev)->gen >= 6) {
1274                 error->derrmr = I915_READ(DERRMR);
1275                 error->error = I915_READ(ERROR_GEN6);
1276                 error->done_reg = I915_READ(DONE_REG);
1277         }
1278
1279         /* 3: Feature specific registers */
1280         if (IS_GEN6(dev) || IS_GEN7(dev)) {
1281                 error->gam_ecochk = I915_READ(GAM_ECOCHK);
1282                 error->gac_eco = I915_READ(GAC_ECO_BITS);
1283         }
1284
1285         /* 4: Everything else */
1286         if (HAS_HW_CONTEXTS(dev))
1287                 error->ccid = I915_READ(CCID);
1288
1289         if (INTEL_INFO(dev)->gen >= 8) {
1290                 error->ier = I915_READ(GEN8_DE_MISC_IER);
1291                 for (i = 0; i < 4; i++)
1292                         error->gtier[i] = I915_READ(GEN8_GT_IER(i));
1293         } else if (HAS_PCH_SPLIT(dev)) {
1294                 error->ier = I915_READ(DEIER);
1295                 error->gtier[0] = I915_READ(GTIER);
1296         } else if (IS_GEN2(dev)) {
1297                 error->ier = I915_READ16(IER);
1298         } else if (!IS_VALLEYVIEW(dev)) {
1299                 error->ier = I915_READ(IER);
1300         }
1301         error->eir = I915_READ(EIR);
1302         error->pgtbl_er = I915_READ(PGTBL_ER);
1303
1304         i915_get_extra_instdone(dev, error->extra_instdone);
1305 }
1306
1307 static void i915_error_capture_msg(struct drm_device *dev,
1308                                    struct drm_i915_error_state *error,
1309                                    u32 engine_mask,
1310                                    const char *error_msg)
1311 {
1312         struct drm_i915_private *dev_priv = dev->dev_private;
1313         u32 ecode;
1314         int ring_id = -1, len;
1315
1316         ecode = i915_error_generate_code(dev_priv, error, &ring_id);
1317
1318         len = scnprintf(error->error_msg, sizeof(error->error_msg),
1319                         "GPU HANG: ecode %d:%d:0x%08x",
1320                         INTEL_INFO(dev)->gen, ring_id, ecode);
1321
1322         if (ring_id != -1 && error->ring[ring_id].pid != -1)
1323                 len += scnprintf(error->error_msg + len,
1324                                  sizeof(error->error_msg) - len,
1325                                  ", in %s [%d]",
1326                                  error->ring[ring_id].comm,
1327                                  error->ring[ring_id].pid);
1328
1329         scnprintf(error->error_msg + len, sizeof(error->error_msg) - len,
1330                   ", reason: %s, action: %s",
1331                   error_msg,
1332                   engine_mask ? "reset" : "continue");
1333 }
1334
1335 static void i915_capture_gen_state(struct drm_i915_private *dev_priv,
1336                                    struct drm_i915_error_state *error)
1337 {
1338         error->iommu = -1;
1339 #ifdef CONFIG_INTEL_IOMMU
1340         error->iommu = intel_iommu_gfx_mapped;
1341 #endif
1342         error->reset_count = i915_reset_count(&dev_priv->gpu_error);
1343         error->suspend_count = dev_priv->suspend_count;
1344 }
1345
1346 /**
1347  * i915_capture_error_state - capture an error record for later analysis
1348  * @dev: drm device
1349  *
1350  * Should be called when an error is detected (either a hang or an error
1351  * interrupt) to capture error state from the time of the error.  Fills
1352  * out a structure which becomes available in debugfs for user level tools
1353  * to pick up.
1354  */
1355 void i915_capture_error_state(struct drm_device *dev, u32 engine_mask,
1356                               const char *error_msg)
1357 {
1358         static bool warned;
1359         struct drm_i915_private *dev_priv = dev->dev_private;
1360         struct drm_i915_error_state *error;
1361         unsigned long flags;
1362
1363         /* Account for pipe specific data like PIPE*STAT */
1364         error = kzalloc(sizeof(*error), GFP_ATOMIC);
1365         if (!error) {
1366                 DRM_DEBUG_DRIVER("out of memory, not capturing error state\n");
1367                 return;
1368         }
1369
1370         kref_init(&error->ref);
1371
1372         i915_capture_gen_state(dev_priv, error);
1373         i915_capture_reg_state(dev_priv, error);
1374         i915_gem_capture_buffers(dev_priv, error);
1375         i915_gem_record_fences(dev, error);
1376         i915_gem_record_rings(dev, error);
1377
1378         do_gettimeofday(&error->time);
1379
1380         error->overlay = intel_overlay_capture_error_state(dev);
1381         error->display = intel_display_capture_error_state(dev);
1382
1383         i915_error_capture_msg(dev, error, engine_mask, error_msg);
1384         DRM_INFO("%s\n", error->error_msg);
1385
1386         spin_lock_irqsave(&dev_priv->gpu_error.lock, flags);
1387         if (dev_priv->gpu_error.first_error == NULL) {
1388                 dev_priv->gpu_error.first_error = error;
1389                 error = NULL;
1390         }
1391         spin_unlock_irqrestore(&dev_priv->gpu_error.lock, flags);
1392
1393         if (error) {
1394                 i915_error_state_free(&error->ref);
1395                 return;
1396         }
1397
1398         if (!warned) {
1399                 DRM_INFO("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1400                 DRM_INFO("Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel\n");
1401                 DRM_INFO("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1402                 DRM_INFO("The gpu crash dump is required to analyze gpu hangs, so please always attach it.\n");
1403                 DRM_INFO("GPU crash dump saved to /sys/class/drm/card%d/error\n", dev->primary->index);
1404                 warned = true;
1405         }
1406 }
1407
1408 void i915_error_state_get(struct drm_device *dev,
1409                           struct i915_error_state_file_priv *error_priv)
1410 {
1411         struct drm_i915_private *dev_priv = dev->dev_private;
1412
1413         spin_lock_irq(&dev_priv->gpu_error.lock);
1414         error_priv->error = dev_priv->gpu_error.first_error;
1415         if (error_priv->error)
1416                 kref_get(&error_priv->error->ref);
1417         spin_unlock_irq(&dev_priv->gpu_error.lock);
1418
1419 }
1420
1421 void i915_error_state_put(struct i915_error_state_file_priv *error_priv)
1422 {
1423         if (error_priv->error)
1424                 kref_put(&error_priv->error->ref, i915_error_state_free);
1425 }
1426
1427 void i915_destroy_error_state(struct drm_device *dev)
1428 {
1429         struct drm_i915_private *dev_priv = dev->dev_private;
1430         struct drm_i915_error_state *error;
1431
1432         spin_lock_irq(&dev_priv->gpu_error.lock);
1433         error = dev_priv->gpu_error.first_error;
1434         dev_priv->gpu_error.first_error = NULL;
1435         spin_unlock_irq(&dev_priv->gpu_error.lock);
1436
1437         if (error)
1438                 kref_put(&error->ref, i915_error_state_free);
1439 }
1440
1441 const char *i915_cache_level_str(struct drm_i915_private *i915, int type)
1442 {
1443         switch (type) {
1444         case I915_CACHE_NONE: return " uncached";
1445         case I915_CACHE_LLC: return HAS_LLC(i915) ? " LLC" : " snooped";
1446         case I915_CACHE_L3_LLC: return " L3+LLC";
1447         case I915_CACHE_WT: return " WT";
1448         default: return "";
1449         }
1450 }
1451
1452 /* NB: please notice the memset */
1453 void i915_get_extra_instdone(struct drm_device *dev, uint32_t *instdone)
1454 {
1455         struct drm_i915_private *dev_priv = dev->dev_private;
1456         memset(instdone, 0, sizeof(*instdone) * I915_NUM_INSTDONE_REG);
1457
1458         if (IS_GEN2(dev) || IS_GEN3(dev))
1459                 instdone[0] = I915_READ(GEN2_INSTDONE);
1460         else if (IS_GEN4(dev) || IS_GEN5(dev) || IS_GEN6(dev)) {
1461                 instdone[0] = I915_READ(RING_INSTDONE(RENDER_RING_BASE));
1462                 instdone[1] = I915_READ(GEN4_INSTDONE1);
1463         } else if (INTEL_INFO(dev)->gen >= 7) {
1464                 instdone[0] = I915_READ(RING_INSTDONE(RENDER_RING_BASE));
1465                 instdone[1] = I915_READ(GEN7_SC_INSTDONE);
1466                 instdone[2] = I915_READ(GEN7_SAMPLER_INSTDONE);
1467                 instdone[3] = I915_READ(GEN7_ROW_INSTDONE);
1468         }
1469 }