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 <linux/ascii85.h>
  31 #include <linux/nmi.h>
  32 #include <linux/pagevec.h>
  33 #include <linux/scatterlist.h>
  34 #include <linux/utsname.h>
  35 #include <linux/zlib.h>
  36
  37 #include <drm/drm_print.h>
  38
  39 #include "display/intel_dmc.h"
  40 #include "display/intel_overlay.h"
  41
  42 #include "gem/i915_gem_context.h"
  43 #include "gem/i915_gem_lmem.h"
  44 #include "gt/intel_gt.h"
  45 #include "gt/intel_gt_pm.h"
  46
  47 #include "i915_drv.h"
  48 #include "i915_gpu_error.h"
  49 #include "i915_memcpy.h"
  50 #include "i915_scatterlist.h"
  51
  52 #define ALLOW_FAIL (GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN)
  53 #define ATOMIC_MAYFAIL (GFP_ATOMIC | __GFP_NOWARN)
  54
  55 static void __sg_set_buf(struct scatterlist *sg,
  56                          void *addr, unsigned int len, loff_t it)
  57 {
  58         sg->page_link = (unsigned long)virt_to_page(addr);
  59         sg->offset = offset_in_page(addr);
  60         sg->length = len;
  61         sg->dma_address = it;
  62 }
  63
  64 static bool __i915_error_grow(struct drm_i915_error_state_buf *e, size_t len)
  65 {
  66         if (!len)
  67                 return false;
  68
  69         if (e->bytes + len + 1 <= e->size)
  70                 return true;
  71
  72         if (e->bytes) {
  73                 __sg_set_buf(e->cur++, e->buf, e->bytes, e->iter);
  74                 e->iter += e->bytes;
  75                 e->buf = NULL;
  76                 e->bytes = 0;
  77         }
  78
  79         if (e->cur == e->end) {
  80                 struct scatterlist *sgl;
  81
  82                 sgl = (typeof(sgl))__get_free_page(ALLOW_FAIL);
  83                 if (!sgl) {
  84                         e->err = -ENOMEM;
  85                         return false;
  86                 }
  87
  88                 if (e->cur) {
  89                         e->cur->offset = 0;
  90                         e->cur->length = 0;
  91                         e->cur->page_link =
  92                                 (unsigned long)sgl | SG_CHAIN;
  93                 } else {
  94                         e->sgl = sgl;
  95                 }
  96
  97                 e->cur = sgl;
  98                 e->end = sgl + SG_MAX_SINGLE_ALLOC - 1;
  99         }
 100
 101         e->size = ALIGN(len + 1, SZ_64K);
 102         e->buf = kmalloc(e->size, ALLOW_FAIL);
 103         if (!e->buf) {
 104                 e->size = PAGE_ALIGN(len + 1);
 105                 e->buf = kmalloc(e->size, GFP_KERNEL);
 106         }
 107         if (!e->buf) {
 108                 e->err = -ENOMEM;
 109                 return false;
 110         }
 111
 112         return true;
 113 }
 114
 115 __printf(2, 0)
 116 static void i915_error_vprintf(struct drm_i915_error_state_buf *e,
 117                                const char *fmt, va_list args)
 118 {
 119         va_list ap;
 120         int len;
 121
 122         if (e->err)
 123                 return;
 124
 125         va_copy(ap, args);
 126         len = vsnprintf(NULL, 0, fmt, ap);
 127         va_end(ap);
 128         if (len <= 0) {
 129                 e->err = len;
 130                 return;
 131         }
 132
 133         if (!__i915_error_grow(e, len))
 134                 return;
 135
 136         GEM_BUG_ON(e->bytes >= e->size);
 137         len = vscnprintf(e->buf + e->bytes, e->size - e->bytes, fmt, args);
 138         if (len < 0) {
 139                 e->err = len;
 140                 return;
 141         }
 142         e->bytes += len;
 143 }
 144
 145 static void i915_error_puts(struct drm_i915_error_state_buf *e, const char *str)
 146 {
 147         unsigned len;
 148
 149         if (e->err || !str)
 150                 return;
 151
 152         len = strlen(str);
 153         if (!__i915_error_grow(e, len))
 154                 return;
 155
 156         GEM_BUG_ON(e->bytes + len > e->size);
 157         memcpy(e->buf + e->bytes, str, len);
 158         e->bytes += len;
 159 }
 160
 161 #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__)
 162 #define err_puts(e, s) i915_error_puts(e, s)
 163
 164 static void __i915_printfn_error(struct drm_printer *p, struct va_format *vaf)
 165 {
 166         i915_error_vprintf(p->arg, vaf->fmt, *vaf->va);
 167 }
 168
 169 static inline struct drm_printer
 170 i915_error_printer(struct drm_i915_error_state_buf *e)
 171 {
 172         struct drm_printer p = {
 173                 .printfn = __i915_printfn_error,
 174                 .arg = e,
 175         };
 176         return p;
 177 }
 178
 179 /* single threaded page allocator with a reserved stash for emergencies */
 180 static void pool_fini(struct pagevec *pv)
 181 {
 182         pagevec_release(pv);
 183 }
 184
 185 static int pool_refill(struct pagevec *pv, gfp_t gfp)
 186 {
 187         while (pagevec_space(pv)) {
 188                 struct page *p;
 189
 190                 p = alloc_page(gfp);
 191                 if (!p)
 192                         return -ENOMEM;
 193
 194                 pagevec_add(pv, p);
 195         }
 196
 197         return 0;
 198 }
 199
 200 static int pool_init(struct pagevec *pv, gfp_t gfp)
 201 {
 202         int err;
 203
 204         pagevec_init(pv);
 205
 206         err = pool_refill(pv, gfp);
 207         if (err)
 208                 pool_fini(pv);
 209
 210         return err;
 211 }
 212
 213 static void *pool_alloc(struct pagevec *pv, gfp_t gfp)
 214 {
 215         struct page *p;
 216
 217         p = alloc_page(gfp);
 218         if (!p && pagevec_count(pv))
 219                 p = pv->pages[--pv->nr];
 220
 221         return p ? page_address(p) : NULL;
 222 }
 223
 224 static void pool_free(struct pagevec *pv, void *addr)
 225 {
 226         struct page *p = virt_to_page(addr);
 227
 228         if (pagevec_space(pv))
 229                 pagevec_add(pv, p);
 230         else
 231                 __free_page(p);
 232 }
 233
 234 #ifdef CONFIG_DRM_I915_COMPRESS_ERROR
 235
 236 struct i915_vma_compress {
 237         struct pagevec pool;
 238         struct z_stream_s zstream;
 239         void *tmp;
 240 };
 241
 242 static bool compress_init(struct i915_vma_compress *c)
 243 {
 244         struct z_stream_s *zstream = &c->zstream;
 245
 246         if (pool_init(&c->pool, ALLOW_FAIL))
 247                 return false;
 248
 249         zstream->workspace =
 250                 kmalloc(zlib_deflate_workspacesize(MAX_WBITS, MAX_MEM_LEVEL),
 251                         ALLOW_FAIL);
 252         if (!zstream->workspace) {
 253                 pool_fini(&c->pool);
 254                 return false;
 255         }
 256
 257         c->tmp = NULL;
 258         if (i915_has_memcpy_from_wc())
 259                 c->tmp = pool_alloc(&c->pool, ALLOW_FAIL);
 260
 261         return true;
 262 }
 263
 264 static bool compress_start(struct i915_vma_compress *c)
 265 {
 266         struct z_stream_s *zstream = &c->zstream;
 267         void *workspace = zstream->workspace;
 268
 269         memset(zstream, 0, sizeof(*zstream));
 270         zstream->workspace = workspace;
 271
 272         return zlib_deflateInit(zstream, Z_DEFAULT_COMPRESSION) == Z_OK;
 273 }
 274
 275 static void *compress_next_page(struct i915_vma_compress *c,
 276                                 struct i915_vma_coredump *dst)
 277 {
 278         void *page;
 279
 280         if (dst->page_count >= dst->num_pages)
 281                 return ERR_PTR(-ENOSPC);
 282
 283         page = pool_alloc(&c->pool, ALLOW_FAIL);
 284         if (!page)
 285                 return ERR_PTR(-ENOMEM);
 286
 287         return dst->pages[dst->page_count++] = page;
 288 }
 289
 290 static int compress_page(struct i915_vma_compress *c,
 291                          void *src,
 292                          struct i915_vma_coredump *dst,
 293                          bool wc)
 294 {
 295         struct z_stream_s *zstream = &c->zstream;
 296
 297         zstream->next_in = src;
 298         if (wc && c->tmp && i915_memcpy_from_wc(c->tmp, src, PAGE_SIZE))
 299                 zstream->next_in = c->tmp;
 300         zstream->avail_in = PAGE_SIZE;
 301
 302         do {
 303                 if (zstream->avail_out == 0) {
 304                         zstream->next_out = compress_next_page(c, dst);
 305                         if (IS_ERR(zstream->next_out))
 306                                 return PTR_ERR(zstream->next_out);
 307
 308                         zstream->avail_out = PAGE_SIZE;
 309                 }
 310
 311                 if (zlib_deflate(zstream, Z_NO_FLUSH) != Z_OK)
 312                         return -EIO;
 313
 314                 cond_resched();
 315         } while (zstream->avail_in);
 316
 317         /* Fallback to uncompressed if we increase size? */
 318         if (0 && zstream->total_out > zstream->total_in)
 319                 return -E2BIG;
 320
 321         return 0;
 322 }
 323
 324 static int compress_flush(struct i915_vma_compress *c,
 325                           struct i915_vma_coredump *dst)
 326 {
 327         struct z_stream_s *zstream = &c->zstream;
 328
 329         do {
 330                 switch (zlib_deflate(zstream, Z_FINISH)) {
 331                 case Z_OK: /* more space requested */
 332                         zstream->next_out = compress_next_page(c, dst);
 333                         if (IS_ERR(zstream->next_out))
 334                                 return PTR_ERR(zstream->next_out);
 335
 336                         zstream->avail_out = PAGE_SIZE;
 337                         break;
 338
 339                 case Z_STREAM_END:
 340                         goto end;
 341
 342                 default: /* any error */
 343                         return -EIO;
 344                 }
 345         } while (1);
 346
 347 end:
 348         memset(zstream->next_out, 0, zstream->avail_out);
 349         dst->unused = zstream->avail_out;
 350         return 0;
 351 }
 352
 353 static void compress_finish(struct i915_vma_compress *c)
 354 {
 355         zlib_deflateEnd(&c->zstream);
 356 }
 357
 358 static void compress_fini(struct i915_vma_compress *c)
 359 {
 360         kfree(c->zstream.workspace);
 361         if (c->tmp)
 362                 pool_free(&c->pool, c->tmp);
 363         pool_fini(&c->pool);
 364 }
 365
 366 static void err_compression_marker(struct drm_i915_error_state_buf *m)
 367 {
 368         err_puts(m, ":");
 369 }
 370
 371 #else
 372
 373 struct i915_vma_compress {
 374         struct pagevec pool;
 375 };
 376
 377 static bool compress_init(struct i915_vma_compress *c)
 378 {
 379         return pool_init(&c->pool, ALLOW_FAIL) == 0;
 380 }
 381
 382 static bool compress_start(struct i915_vma_compress *c)
 383 {
 384         return true;
 385 }
 386
 387 static int compress_page(struct i915_vma_compress *c,
 388                          void *src,
 389                          struct i915_vma_coredump *dst,
 390                          bool wc)
 391 {
 392         void *ptr;
 393
 394         ptr = pool_alloc(&c->pool, ALLOW_FAIL);
 395         if (!ptr)
 396                 return -ENOMEM;
 397
 398         if (!(wc && i915_memcpy_from_wc(ptr, src, PAGE_SIZE)))
 399                 memcpy(ptr, src, PAGE_SIZE);
 400         dst->pages[dst->page_count++] = ptr;
 401         cond_resched();
 402
 403         return 0;
 404 }
 405
 406 static int compress_flush(struct i915_vma_compress *c,
 407                           struct i915_vma_coredump *dst)
 408 {
 409         return 0;
 410 }
 411
 412 static void compress_finish(struct i915_vma_compress *c)
 413 {
 414 }
 415
 416 static void compress_fini(struct i915_vma_compress *c)
 417 {
 418         pool_fini(&c->pool);
 419 }
 420
 421 static void err_compression_marker(struct drm_i915_error_state_buf *m)
 422 {
 423         err_puts(m, "~");
 424 }
 425
 426 #endif
 427
 428 static void error_print_instdone(struct drm_i915_error_state_buf *m,
 429                                  const struct intel_engine_coredump *ee)
 430 {
 431         const struct sseu_dev_info *sseu = &ee->engine->gt->info.sseu;
 432         int slice;
 433         int subslice;
 434         int iter;
 435
 436         err_printf(m, "  INSTDONE: 0x%08x\n",
 437                    ee->instdone.instdone);
 438
 439         if (ee->engine->class != RENDER_CLASS || GRAPHICS_VER(m->i915) <= 3)
 440                 return;
 441
 442         err_printf(m, "  SC_INSTDONE: 0x%08x\n",
 443                    ee->instdone.slice_common);
 444
 445         if (GRAPHICS_VER(m->i915) <= 6)
 446                 return;
 447
 448         if (GRAPHICS_VER_FULL(m->i915) >= IP_VER(12, 50)) {
 449                 for_each_instdone_gslice_dss_xehp(m->i915, sseu, iter, slice, subslice)
 450                         err_printf(m, "  SAMPLER_INSTDONE[%d][%d]: 0x%08x\n",
 451                                    slice, subslice,
 452                                    ee->instdone.sampler[slice][subslice]);
 453
 454                 for_each_instdone_gslice_dss_xehp(m->i915, sseu, iter, slice, subslice)
 455                         err_printf(m, "  ROW_INSTDONE[%d][%d]: 0x%08x\n",
 456                                    slice, subslice,
 457                                    ee->instdone.row[slice][subslice]);
 458         } else {
 459                 for_each_instdone_slice_subslice(m->i915, sseu, slice, subslice)
 460                         err_printf(m, "  SAMPLER_INSTDONE[%d][%d]: 0x%08x\n",
 461                                    slice, subslice,
 462                                    ee->instdone.sampler[slice][subslice]);
 463
 464                 for_each_instdone_slice_subslice(m->i915, sseu, slice, subslice)
 465                         err_printf(m, "  ROW_INSTDONE[%d][%d]: 0x%08x\n",
 466                                    slice, subslice,
 467                                    ee->instdone.row[slice][subslice]);
 468         }
 469
 470         if (GRAPHICS_VER(m->i915) < 12)
 471                 return;
 472
 473         if (GRAPHICS_VER_FULL(m->i915) >= IP_VER(12, 55)) {
 474                 for_each_instdone_gslice_dss_xehp(m->i915, sseu, iter, slice, subslice)
 475                         err_printf(m, "  GEOM_SVGUNIT_INSTDONE[%d][%d]: 0x%08x\n",
 476                                    slice, subslice,
 477                                    ee->instdone.geom_svg[slice][subslice]);
 478         }
 479
 480         err_printf(m, "  SC_INSTDONE_EXTRA: 0x%08x\n",
 481                    ee->instdone.slice_common_extra[0]);
 482         err_printf(m, "  SC_INSTDONE_EXTRA2: 0x%08x\n",
 483                    ee->instdone.slice_common_extra[1]);
 484 }
 485
 486 static void error_print_request(struct drm_i915_error_state_buf *m,
 487                                 const char *prefix,
 488                                 const struct i915_request_coredump *erq)
 489 {
 490         if (!erq->seqno)
 491                 return;
 492
 493         err_printf(m, "%s pid %d, seqno %8x:%08x%s%s, prio %d, head %08x, tail %08x\n",
 494                    prefix, erq->pid, erq->context, erq->seqno,
 495                    test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
 496                             &erq->flags) ? "!" : "",
 497                    test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT,
 498                             &erq->flags) ? "+" : "",
 499                    erq->sched_attr.priority,
 500                    erq->head, erq->tail);
 501 }
 502
 503 static void error_print_context(struct drm_i915_error_state_buf *m,
 504                                 const char *header,
 505                                 const struct i915_gem_context_coredump *ctx)
 506 {
 507         const u32 period = m->i915->gt.clock_period_ns;
 508
 509         err_printf(m, "%s%s[%d] prio %d, guilty %d active %d, runtime total %lluns, avg %lluns\n",
 510                    header, ctx->comm, ctx->pid, ctx->sched_attr.priority,
 511                    ctx->guilty, ctx->active,
 512                    ctx->total_runtime * period,
 513                    mul_u32_u32(ctx->avg_runtime, period));
 514 }
 515
 516 static struct i915_vma_coredump *
 517 __find_vma(struct i915_vma_coredump *vma, const char *name)
 518 {
 519         while (vma) {
 520                 if (strcmp(vma->name, name) == 0)
 521                         return vma;
 522                 vma = vma->next;
 523         }
 524
 525         return NULL;
 526 }
 527
 528 static struct i915_vma_coredump *
 529 find_batch(const struct intel_engine_coredump *ee)
 530 {
 531         return __find_vma(ee->vma, "batch");
 532 }
 533
 534 static void error_print_engine(struct drm_i915_error_state_buf *m,
 535                                const struct intel_engine_coredump *ee)
 536 {
 537         struct i915_vma_coredump *batch;
 538         int n;
 539
 540         err_printf(m, "%s command stream:\n", ee->engine->name);
 541         err_printf(m, "  CCID:  0x%08x\n", ee->ccid);
 542         err_printf(m, "  START: 0x%08x\n", ee->start);
 543         err_printf(m, "  HEAD:  0x%08x [0x%08x]\n", ee->head, ee->rq_head);
 544         err_printf(m, "  TAIL:  0x%08x [0x%08x, 0x%08x]\n",
 545                    ee->tail, ee->rq_post, ee->rq_tail);
 546         err_printf(m, "  CTL:   0x%08x\n", ee->ctl);
 547         err_printf(m, "  MODE:  0x%08x\n", ee->mode);
 548         err_printf(m, "  HWS:   0x%08x\n", ee->hws);
 549         err_printf(m, "  ACTHD: 0x%08x %08x\n",
 550                    (u32)(ee->acthd>>32), (u32)ee->acthd);
 551         err_printf(m, "  IPEIR: 0x%08x\n", ee->ipeir);
 552         err_printf(m, "  IPEHR: 0x%08x\n", ee->ipehr);
 553         err_printf(m, "  ESR:   0x%08x\n", ee->esr);
 554
 555         error_print_instdone(m, ee);
 556
 557         batch = find_batch(ee);
 558         if (batch) {
 559                 u64 start = batch->gtt_offset;
 560                 u64 end = start + batch->gtt_size;
 561
 562                 err_printf(m, "  batch: [0x%08x_%08x, 0x%08x_%08x]\n",
 563                            upper_32_bits(start), lower_32_bits(start),
 564                            upper_32_bits(end), lower_32_bits(end));
 565         }
 566         if (GRAPHICS_VER(m->i915) >= 4) {
 567                 err_printf(m, "  BBADDR: 0x%08x_%08x\n",
 568                            (u32)(ee->bbaddr>>32), (u32)ee->bbaddr);
 569                 err_printf(m, "  BB_STATE: 0x%08x\n", ee->bbstate);
 570                 err_printf(m, "  INSTPS: 0x%08x\n", ee->instps);
 571         }
 572         err_printf(m, "  INSTPM: 0x%08x\n", ee->instpm);
 573         err_printf(m, "  FADDR: 0x%08x %08x\n", upper_32_bits(ee->faddr),
 574                    lower_32_bits(ee->faddr));
 575         if (GRAPHICS_VER(m->i915) >= 6) {
 576                 err_printf(m, "  RC PSMI: 0x%08x\n", ee->rc_psmi);
 577                 err_printf(m, "  FAULT_REG: 0x%08x\n", ee->fault_reg);
 578         }
 579         if (HAS_PPGTT(m->i915)) {
 580                 err_printf(m, "  GFX_MODE: 0x%08x\n", ee->vm_info.gfx_mode);
 581
 582                 if (GRAPHICS_VER(m->i915) >= 8) {
 583                         int i;
 584                         for (i = 0; i < 4; i++)
 585                                 err_printf(m, "  PDP%d: 0x%016llx\n",
 586                                            i, ee->vm_info.pdp[i]);
 587                 } else {
 588                         err_printf(m, "  PP_DIR_BASE: 0x%08x\n",
 589                                    ee->vm_info.pp_dir_base);
 590                 }
 591         }
 592         err_printf(m, "  hung: %u\n", ee->hung);
 593         err_printf(m, "  engine reset count: %u\n", ee->reset_count);
 594
 595         for (n = 0; n < ee->num_ports; n++) {
 596                 err_printf(m, "  ELSP[%d]:", n);
 597                 error_print_request(m, " ", &ee->execlist[n]);
 598         }
 599
 600         error_print_context(m, "  Active context: ", &ee->context);
 601 }
 602
 603 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
 604 {
 605         va_list args;
 606
 607         va_start(args, f);
 608         i915_error_vprintf(e, f, args);
 609         va_end(args);
 610 }
 611
 612 static void print_error_vma(struct drm_i915_error_state_buf *m,
 613                             const struct intel_engine_cs *engine,
 614                             const struct i915_vma_coredump *vma)
 615 {
 616         char out[ASCII85_BUFSZ];
 617         int page;
 618
 619         if (!vma)
 620                 return;
 621
 622         err_printf(m, "%s --- %s = 0x%08x %08x\n",
 623                    engine ? engine->name : "global", vma->name,
 624                    upper_32_bits(vma->gtt_offset),
 625                    lower_32_bits(vma->gtt_offset));
 626
 627         if (vma->gtt_page_sizes > I915_GTT_PAGE_SIZE_4K)
 628                 err_printf(m, "gtt_page_sizes = 0x%08x\n", vma->gtt_page_sizes);
 629
 630         err_compression_marker(m);
 631         for (page = 0; page < vma->page_count; page++) {
 632                 int i, len;
 633
 634                 len = PAGE_SIZE;
 635                 if (page == vma->page_count - 1)
 636                         len -= vma->unused;
 637                 len = ascii85_encode_len(len);
 638
 639                 for (i = 0; i < len; i++)
 640                         err_puts(m, ascii85_encode(vma->pages[page][i], out));
 641         }
 642         err_puts(m, "\n");
 643 }
 644
 645 static void err_print_capabilities(struct drm_i915_error_state_buf *m,
 646                                    struct i915_gpu_coredump *error)
 647 {
 648         struct drm_printer p = i915_error_printer(m);
 649
 650         intel_device_info_print_static(&error->device_info, &p);
 651         intel_device_info_print_runtime(&error->runtime_info, &p);
 652         intel_driver_caps_print(&error->driver_caps, &p);
 653 }
 654
 655 static void err_print_params(struct drm_i915_error_state_buf *m,
 656                              const struct i915_params *params)
 657 {
 658         struct drm_printer p = i915_error_printer(m);
 659
 660         i915_params_dump(params, &p);
 661 }
 662
 663 static void err_print_pciid(struct drm_i915_error_state_buf *m,
 664                             struct drm_i915_private *i915)
 665 {
 666         struct pci_dev *pdev = to_pci_dev(i915->drm.dev);
 667
 668         err_printf(m, "PCI ID: 0x%04x\n", pdev->device);
 669         err_printf(m, "PCI Revision: 0x%02x\n", pdev->revision);
 670         err_printf(m, "PCI Subsystem: %04x:%04x\n",
 671                    pdev->subsystem_vendor,
 672                    pdev->subsystem_device);
 673 }
 674
 675 static void err_print_uc(struct drm_i915_error_state_buf *m,
 676                          const struct intel_uc_coredump *error_uc)
 677 {
 678         struct drm_printer p = i915_error_printer(m);
 679
 680         intel_uc_fw_dump(&error_uc->guc_fw, &p);
 681         intel_uc_fw_dump(&error_uc->huc_fw, &p);
 682         print_error_vma(m, NULL, error_uc->guc_log);
 683 }
 684
 685 static void err_free_sgl(struct scatterlist *sgl)
 686 {
 687         while (sgl) {
 688                 struct scatterlist *sg;
 689
 690                 for (sg = sgl; !sg_is_chain(sg); sg++) {
 691                         kfree(sg_virt(sg));
 692                         if (sg_is_last(sg))
 693                                 break;
 694                 }
 695
 696                 sg = sg_is_last(sg) ? NULL : sg_chain_ptr(sg);
 697                 free_page((unsigned long)sgl);
 698                 sgl = sg;
 699         }
 700 }
 701
 702 static void err_print_gt_info(struct drm_i915_error_state_buf *m,
 703                               struct intel_gt_coredump *gt)
 704 {
 705         struct drm_printer p = i915_error_printer(m);
 706
 707         intel_gt_info_print(&gt->info, &p);
 708         intel_sseu_print_topology(&gt->info.sseu, &p);
 709 }
 710
 711 static void err_print_gt(struct drm_i915_error_state_buf *m,
 712                          struct intel_gt_coredump *gt)
 713 {
 714         const struct intel_engine_coredump *ee;
 715         int i;
 716
 717         err_printf(m, "GT awake: %s\n", yesno(gt->awake));
 718         err_printf(m, "EIR: 0x%08x\n", gt->eir);
 719         err_printf(m, "IER: 0x%08x\n", gt->ier);
 720         for (i = 0; i < gt->ngtier; i++)
 721                 err_printf(m, "GTIER[%d]: 0x%08x\n", i, gt->gtier[i]);
 722         err_printf(m, "PGTBL_ER: 0x%08x\n", gt->pgtbl_er);
 723         err_printf(m, "FORCEWAKE: 0x%08x\n", gt->forcewake);
 724         err_printf(m, "DERRMR: 0x%08x\n", gt->derrmr);
 725
 726         for (i = 0; i < gt->nfence; i++)
 727                 err_printf(m, "  fence[%d] = %08llx\n", i, gt->fence[i]);
 728
 729         if (IS_GRAPHICS_VER(m->i915, 6, 11)) {
 730                 err_printf(m, "ERROR: 0x%08x\n", gt->error);
 731                 err_printf(m, "DONE_REG: 0x%08x\n", gt->done_reg);
 732         }
 733
 734         if (GRAPHICS_VER(m->i915) >= 8)
 735                 err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n",
 736                            gt->fault_data1, gt->fault_data0);
 737
 738         if (GRAPHICS_VER(m->i915) == 7)
 739                 err_printf(m, "ERR_INT: 0x%08x\n", gt->err_int);
 740
 741         if (IS_GRAPHICS_VER(m->i915, 8, 11))
 742                 err_printf(m, "GTT_CACHE_EN: 0x%08x\n", gt->gtt_cache);
 743
 744         if (GRAPHICS_VER(m->i915) == 12)
 745                 err_printf(m, "AUX_ERR_DBG: 0x%08x\n", gt->aux_err);
 746
 747         if (GRAPHICS_VER(m->i915) >= 12) {
 748                 int i;
 749
 750                 for (i = 0; i < GEN12_SFC_DONE_MAX; i++) {
 751                         /*
 752                          * SFC_DONE resides in the VD forcewake domain, so it
 753                          * only exists if the corresponding VCS engine is
 754                          * present.
 755                          */
 756                         if (!HAS_ENGINE(gt->_gt, _VCS(i * 2)))
 757                                 continue;
 758
 759                         err_printf(m, "  SFC_DONE[%d]: 0x%08x\n", i,
 760                                    gt->sfc_done[i]);
 761                 }
 762
 763                 err_printf(m, "  GAM_DONE: 0x%08x\n", gt->gam_done);
 764         }
 765
 766         for (ee = gt->engine; ee; ee = ee->next) {
 767                 const struct i915_vma_coredump *vma;
 768
 769                 error_print_engine(m, ee);
 770                 for (vma = ee->vma; vma; vma = vma->next)
 771                         print_error_vma(m, ee->engine, vma);
 772         }
 773
 774         if (gt->uc)
 775                 err_print_uc(m, gt->uc);
 776
 777         err_print_gt_info(m, gt);
 778 }
 779
 780 static void __err_print_to_sgl(struct drm_i915_error_state_buf *m,
 781                                struct i915_gpu_coredump *error)
 782 {
 783         const struct intel_engine_coredump *ee;
 784         struct timespec64 ts;
 785
 786         if (*error->error_msg)
 787                 err_printf(m, "%s\n", error->error_msg);
 788         err_printf(m, "Kernel: %s %s\n",
 789                    init_utsname()->release,
 790                    init_utsname()->machine);
 791         err_printf(m, "Driver: %s\n", DRIVER_DATE);
 792         ts = ktime_to_timespec64(error->time);
 793         err_printf(m, "Time: %lld s %ld us\n",
 794                    (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
 795         ts = ktime_to_timespec64(error->boottime);
 796         err_printf(m, "Boottime: %lld s %ld us\n",
 797                    (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
 798         ts = ktime_to_timespec64(error->uptime);
 799         err_printf(m, "Uptime: %lld s %ld us\n",
 800                    (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC);
 801         err_printf(m, "Capture: %lu jiffies; %d ms ago\n",
 802                    error->capture, jiffies_to_msecs(jiffies - error->capture));
 803
 804         for (ee = error->gt ? error->gt->engine : NULL; ee; ee = ee->next)
 805                 err_printf(m, "Active process (on ring %s): %s [%d]\n",
 806                            ee->engine->name,
 807                            ee->context.comm,
 808                            ee->context.pid);
 809
 810         err_printf(m, "Reset count: %u\n", error->reset_count);
 811         err_printf(m, "Suspend count: %u\n", error->suspend_count);
 812         err_printf(m, "Platform: %s\n", intel_platform_name(error->device_info.platform));
 813         err_printf(m, "Subplatform: 0x%x\n",
 814                    intel_subplatform(&error->runtime_info,
 815                                      error->device_info.platform));
 816         err_print_pciid(m, m->i915);
 817
 818         err_printf(m, "IOMMU enabled?: %d\n", error->iommu);
 819
 820         if (HAS_DMC(m->i915)) {
 821                 struct intel_dmc *dmc = &m->i915->dmc;
 822
 823                 err_printf(m, "DMC loaded: %s\n",
 824                            yesno(intel_dmc_has_payload(m->i915) != 0));
 825                 err_printf(m, "DMC fw version: %d.%d\n",
 826                            DMC_VERSION_MAJOR(dmc->version),
 827                            DMC_VERSION_MINOR(dmc->version));
 828         }
 829
 830         err_printf(m, "RPM wakelock: %s\n", yesno(error->wakelock));
 831         err_printf(m, "PM suspended: %s\n", yesno(error->suspended));
 832
 833         if (error->gt)
 834                 err_print_gt(m, error->gt);
 835
 836         if (error->overlay)
 837                 intel_overlay_print_error_state(m, error->overlay);
 838
 839         err_print_capabilities(m, error);
 840         err_print_params(m, &error->params);
 841 }
 842
 843 static int err_print_to_sgl(struct i915_gpu_coredump *error)
 844 {
 845         struct drm_i915_error_state_buf m;
 846
 847         if (IS_ERR(error))
 848                 return PTR_ERR(error);
 849
 850         if (READ_ONCE(error->sgl))
 851                 return 0;
 852
 853         memset(&m, 0, sizeof(m));
 854         m.i915 = error->i915;
 855
 856         __err_print_to_sgl(&m, error);
 857
 858         if (m.buf) {
 859                 __sg_set_buf(m.cur++, m.buf, m.bytes, m.iter);
 860                 m.bytes = 0;
 861                 m.buf = NULL;
 862         }
 863         if (m.cur) {
 864                 GEM_BUG_ON(m.end < m.cur);
 865                 sg_mark_end(m.cur - 1);
 866         }
 867         GEM_BUG_ON(m.sgl && !m.cur);
 868
 869         if (m.err) {
 870                 err_free_sgl(m.sgl);
 871                 return m.err;
 872         }
 873
 874         if (cmpxchg(&error->sgl, NULL, m.sgl))
 875                 err_free_sgl(m.sgl);
 876
 877         return 0;
 878 }
 879
 880 ssize_t i915_gpu_coredump_copy_to_buffer(struct i915_gpu_coredump *error,
 881                                          char *buf, loff_t off, size_t rem)
 882 {
 883         struct scatterlist *sg;
 884         size_t count;
 885         loff_t pos;
 886         int err;
 887
 888         if (!error || !rem)
 889                 return 0;
 890
 891         err = err_print_to_sgl(error);
 892         if (err)
 893                 return err;
 894
 895         sg = READ_ONCE(error->fit);
 896         if (!sg || off < sg->dma_address)
 897                 sg = error->sgl;
 898         if (!sg)
 899                 return 0;
 900
 901         pos = sg->dma_address;
 902         count = 0;
 903         do {
 904                 size_t len, start;
 905
 906                 if (sg_is_chain(sg)) {
 907                         sg = sg_chain_ptr(sg);
 908                         GEM_BUG_ON(sg_is_chain(sg));
 909                 }
 910
 911                 len = sg->length;
 912                 if (pos + len <= off) {
 913                         pos += len;
 914                         continue;
 915                 }
 916
 917                 start = sg->offset;
 918                 if (pos < off) {
 919                         GEM_BUG_ON(off - pos > len);
 920                         len -= off - pos;
 921                         start += off - pos;
 922                         pos = off;
 923                 }
 924
 925                 len = min(len, rem);
 926                 GEM_BUG_ON(!len || len > sg->length);
 927
 928                 memcpy(buf, page_address(sg_page(sg)) + start, len);
 929
 930                 count += len;
 931                 pos += len;
 932
 933                 buf += len;
 934                 rem -= len;
 935                 if (!rem) {
 936                         WRITE_ONCE(error->fit, sg);
 937                         break;
 938                 }
 939         } while (!sg_is_last(sg++));
 940
 941         return count;
 942 }
 943
 944 static void i915_vma_coredump_free(struct i915_vma_coredump *vma)
 945 {
 946         while (vma) {
 947                 struct i915_vma_coredump *next = vma->next;
 948                 int page;
 949
 950                 for (page = 0; page < vma->page_count; page++)
 951                         free_page((unsigned long)vma->pages[page]);
 952
 953                 kfree(vma);
 954                 vma = next;
 955         }
 956 }
 957
 958 static void cleanup_params(struct i915_gpu_coredump *error)
 959 {
 960         i915_params_free(&error->params);
 961 }
 962
 963 static void cleanup_uc(struct intel_uc_coredump *uc)
 964 {
 965         kfree(uc->guc_fw.path);
 966         kfree(uc->huc_fw.path);
 967         i915_vma_coredump_free(uc->guc_log);
 968
 969         kfree(uc);
 970 }
 971
 972 static void cleanup_gt(struct intel_gt_coredump *gt)
 973 {
 974         while (gt->engine) {
 975                 struct intel_engine_coredump *ee = gt->engine;
 976
 977                 gt->engine = ee->next;
 978
 979                 i915_vma_coredump_free(ee->vma);
 980                 kfree(ee);
 981         }
 982
 983         if (gt->uc)
 984                 cleanup_uc(gt->uc);
 985
 986         kfree(gt);
 987 }
 988
 989 void __i915_gpu_coredump_free(struct kref *error_ref)
 990 {
 991         struct i915_gpu_coredump *error =
 992                 container_of(error_ref, typeof(*error), ref);
 993
 994         while (error->gt) {
 995                 struct intel_gt_coredump *gt = error->gt;
 996
 997                 error->gt = gt->next;
 998                 cleanup_gt(gt);
 999         }
1000
1001         kfree(error->overlay);
1002
1003         cleanup_params(error);
1004
1005         err_free_sgl(error->sgl);
1006         kfree(error);
1007 }
1008
1009 static struct i915_vma_coredump *
1010 i915_vma_coredump_create(const struct intel_gt *gt,
1011                          const struct i915_vma *vma,
1012                          const char *name,
1013                          struct i915_vma_compress *compress)
1014 {
1015         struct i915_ggtt *ggtt = gt->ggtt;
1016         const u64 slot = ggtt->error_capture.start;
1017         struct i915_vma_coredump *dst;
1018         unsigned long num_pages;
1019         struct sgt_iter iter;
1020         int ret;
1021
1022         might_sleep();
1023
1024         if (!vma || !vma->pages || !compress)
1025                 return NULL;
1026
1027         num_pages = min_t(u64, vma->size, vma->obj->base.size) >> PAGE_SHIFT;
1028         num_pages = DIV_ROUND_UP(10 * num_pages, 8); /* worstcase zlib growth */
1029         dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *), ALLOW_FAIL);
1030         if (!dst)
1031                 return NULL;
1032
1033         if (!compress_start(compress)) {
1034                 kfree(dst);
1035                 return NULL;
1036         }
1037
1038         strcpy(dst->name, name);
1039         dst->next = NULL;
1040
1041         dst->gtt_offset = vma->node.start;
1042         dst->gtt_size = vma->node.size;
1043         dst->gtt_page_sizes = vma->page_sizes.gtt;
1044         dst->num_pages = num_pages;
1045         dst->page_count = 0;
1046         dst->unused = 0;
1047
1048         ret = -EINVAL;
1049         if (drm_mm_node_allocated(&ggtt->error_capture)) {
1050                 void __iomem *s;
1051                 dma_addr_t dma;
1052
1053                 for_each_sgt_daddr(dma, iter, vma->pages) {
1054                         mutex_lock(&ggtt->error_mutex);
1055                         ggtt->vm.insert_page(&ggtt->vm, dma, slot,
1056                                              I915_CACHE_NONE, 0);
1057                         mb();
1058
1059                         s = io_mapping_map_wc(&ggtt->iomap, slot, PAGE_SIZE);
1060                         ret = compress_page(compress,
1061                                             (void  __force *)s, dst,
1062                                             true);
1063                         io_mapping_unmap(s);
1064
1065                         mb();
1066                         ggtt->vm.clear_range(&ggtt->vm, slot, PAGE_SIZE);
1067                         mutex_unlock(&ggtt->error_mutex);
1068                         if (ret)
1069                                 break;
1070                 }
1071         } else if (__i915_gem_object_is_lmem(vma->obj)) {
1072                 struct intel_memory_region *mem = vma->obj->mm.region;
1073                 dma_addr_t dma;
1074
1075                 for_each_sgt_daddr(dma, iter, vma->pages) {
1076                         void __iomem *s;
1077
1078                         s = io_mapping_map_wc(&mem->iomap,
1079                                               dma - mem->region.start,
1080                                               PAGE_SIZE);
1081                         ret = compress_page(compress,
1082                                             (void __force *)s, dst,
1083                                             true);
1084                         io_mapping_unmap(s);
1085                         if (ret)
1086                                 break;
1087                 }
1088         } else {
1089                 struct page *page;
1090
1091                 for_each_sgt_page(page, iter, vma->pages) {
1092                         void *s;
1093
1094                         drm_clflush_pages(&page, 1);
1095
1096                         s = kmap(page);
1097                         ret = compress_page(compress, s, dst, false);
1098                         kunmap(page);
1099
1100                         drm_clflush_pages(&page, 1);
1101
1102                         if (ret)
1103                                 break;
1104                 }
1105         }
1106
1107         if (ret || compress_flush(compress, dst)) {
1108                 while (dst->page_count--)
1109                         pool_free(&compress->pool, dst->pages[dst->page_count]);
1110                 kfree(dst);
1111                 dst = NULL;
1112         }
1113         compress_finish(compress);
1114
1115         return dst;
1116 }
1117
1118 static void gt_record_fences(struct intel_gt_coredump *gt)
1119 {
1120         struct i915_ggtt *ggtt = gt->_gt->ggtt;
1121         struct intel_uncore *uncore = gt->_gt->uncore;
1122         int i;
1123
1124         if (GRAPHICS_VER(uncore->i915) >= 6) {
1125                 for (i = 0; i < ggtt->num_fences; i++)
1126                         gt->fence[i] =
1127                                 intel_uncore_read64(uncore,
1128                                                     FENCE_REG_GEN6_LO(i));
1129         } else if (GRAPHICS_VER(uncore->i915) >= 4) {
1130                 for (i = 0; i < ggtt->num_fences; i++)
1131                         gt->fence[i] =
1132                                 intel_uncore_read64(uncore,
1133                                                     FENCE_REG_965_LO(i));
1134         } else {
1135                 for (i = 0; i < ggtt->num_fences; i++)
1136                         gt->fence[i] =
1137                                 intel_uncore_read(uncore, FENCE_REG(i));
1138         }
1139         gt->nfence = i;
1140 }
1141
1142 static void engine_record_registers(struct intel_engine_coredump *ee)
1143 {
1144         const struct intel_engine_cs *engine = ee->engine;
1145         struct drm_i915_private *i915 = engine->i915;
1146
1147         if (GRAPHICS_VER(i915) >= 6) {
1148                 ee->rc_psmi = ENGINE_READ(engine, RING_PSMI_CTL);
1149
1150                 if (GRAPHICS_VER(i915) >= 12)
1151                         ee->fault_reg = intel_uncore_read(engine->uncore,
1152                                                           GEN12_RING_FAULT_REG);
1153                 else if (GRAPHICS_VER(i915) >= 8)
1154                         ee->fault_reg = intel_uncore_read(engine->uncore,
1155                                                           GEN8_RING_FAULT_REG);
1156                 else
1157                         ee->fault_reg = GEN6_RING_FAULT_REG_READ(engine);
1158         }
1159
1160         if (GRAPHICS_VER(i915) >= 4) {
1161                 ee->esr = ENGINE_READ(engine, RING_ESR);
1162                 ee->faddr = ENGINE_READ(engine, RING_DMA_FADD);
1163                 ee->ipeir = ENGINE_READ(engine, RING_IPEIR);
1164                 ee->ipehr = ENGINE_READ(engine, RING_IPEHR);
1165                 ee->instps = ENGINE_READ(engine, RING_INSTPS);
1166                 ee->bbaddr = ENGINE_READ(engine, RING_BBADDR);
1167                 ee->ccid = ENGINE_READ(engine, CCID);
1168                 if (GRAPHICS_VER(i915) >= 8) {
1169                         ee->faddr |= (u64)ENGINE_READ(engine, RING_DMA_FADD_UDW) << 32;
1170                         ee->bbaddr |= (u64)ENGINE_READ(engine, RING_BBADDR_UDW) << 32;
1171                 }
1172                 ee->bbstate = ENGINE_READ(engine, RING_BBSTATE);
1173         } else {
1174                 ee->faddr = ENGINE_READ(engine, DMA_FADD_I8XX);
1175                 ee->ipeir = ENGINE_READ(engine, IPEIR);
1176                 ee->ipehr = ENGINE_READ(engine, IPEHR);
1177         }
1178
1179         intel_engine_get_instdone(engine, &ee->instdone);
1180
1181         ee->instpm = ENGINE_READ(engine, RING_INSTPM);
1182         ee->acthd = intel_engine_get_active_head(engine);
1183         ee->start = ENGINE_READ(engine, RING_START);
1184         ee->head = ENGINE_READ(engine, RING_HEAD);
1185         ee->tail = ENGINE_READ(engine, RING_TAIL);
1186         ee->ctl = ENGINE_READ(engine, RING_CTL);
1187         if (GRAPHICS_VER(i915) > 2)
1188                 ee->mode = ENGINE_READ(engine, RING_MI_MODE);
1189
1190         if (!HWS_NEEDS_PHYSICAL(i915)) {
1191                 i915_reg_t mmio;
1192
1193                 if (GRAPHICS_VER(i915) == 7) {
1194                         switch (engine->id) {
1195                         default:
1196                                 MISSING_CASE(engine->id);
1197                                 fallthrough;
1198                         case RCS0:
1199                                 mmio = RENDER_HWS_PGA_GEN7;
1200                                 break;
1201                         case BCS0:
1202                                 mmio = BLT_HWS_PGA_GEN7;
1203                                 break;
1204                         case VCS0:
1205                                 mmio = BSD_HWS_PGA_GEN7;
1206                                 break;
1207                         case VECS0:
1208                                 mmio = VEBOX_HWS_PGA_GEN7;
1209                                 break;
1210                         }
1211                 } else if (GRAPHICS_VER(engine->i915) == 6) {
1212                         mmio = RING_HWS_PGA_GEN6(engine->mmio_base);
1213                 } else {
1214                         /* XXX: gen8 returns to sanity */
1215                         mmio = RING_HWS_PGA(engine->mmio_base);
1216                 }
1217
1218                 ee->hws = intel_uncore_read(engine->uncore, mmio);
1219         }
1220
1221         ee->reset_count = i915_reset_engine_count(&i915->gpu_error, engine);
1222
1223         if (HAS_PPGTT(i915)) {
1224                 int i;
1225
1226                 ee->vm_info.gfx_mode = ENGINE_READ(engine, RING_MODE_GEN7);
1227
1228                 if (GRAPHICS_VER(i915) == 6) {
1229                         ee->vm_info.pp_dir_base =
1230                                 ENGINE_READ(engine, RING_PP_DIR_BASE_READ);
1231                 } else if (GRAPHICS_VER(i915) == 7) {
1232                         ee->vm_info.pp_dir_base =
1233                                 ENGINE_READ(engine, RING_PP_DIR_BASE);
1234                 } else if (GRAPHICS_VER(i915) >= 8) {
1235                         u32 base = engine->mmio_base;
1236
1237                         for (i = 0; i < 4; i++) {
1238                                 ee->vm_info.pdp[i] =
1239                                         intel_uncore_read(engine->uncore,
1240                                                           GEN8_RING_PDP_UDW(base, i));
1241                                 ee->vm_info.pdp[i] <<= 32;
1242                                 ee->vm_info.pdp[i] |=
1243                                         intel_uncore_read(engine->uncore,
1244                                                           GEN8_RING_PDP_LDW(base, i));
1245                         }
1246                 }
1247         }
1248 }
1249
1250 static void record_request(const struct i915_request *request,
1251                            struct i915_request_coredump *erq)
1252 {
1253         erq->flags = request->fence.flags;
1254         erq->context = request->fence.context;
1255         erq->seqno = request->fence.seqno;
1256         erq->sched_attr = request->sched.attr;
1257         erq->head = request->head;
1258         erq->tail = request->tail;
1259
1260         erq->pid = 0;
1261         rcu_read_lock();
1262         if (!intel_context_is_closed(request->context)) {
1263                 const struct i915_gem_context *ctx;
1264
1265                 ctx = rcu_dereference(request->context->gem_context);
1266                 if (ctx)
1267                         erq->pid = pid_nr(ctx->pid);
1268         }
1269         rcu_read_unlock();
1270 }
1271
1272 static void engine_record_execlists(struct intel_engine_coredump *ee)
1273 {
1274         const struct intel_engine_execlists * const el = &ee->engine->execlists;
1275         struct i915_request * const *port = el->active;
1276         unsigned int n = 0;
1277
1278         while (*port)
1279                 record_request(*port++, &ee->execlist[n++]);
1280
1281         ee->num_ports = n;
1282 }
1283
1284 static bool record_context(struct i915_gem_context_coredump *e,
1285                            const struct i915_request *rq)
1286 {
1287         struct i915_gem_context *ctx;
1288         struct task_struct *task;
1289         bool simulated;
1290
1291         rcu_read_lock();
1292         ctx = rcu_dereference(rq->context->gem_context);
1293         if (ctx && !kref_get_unless_zero(&ctx->ref))
1294                 ctx = NULL;
1295         rcu_read_unlock();
1296         if (!ctx)
1297                 return true;
1298
1299         rcu_read_lock();
1300         task = pid_task(ctx->pid, PIDTYPE_PID);
1301         if (task) {
1302                 strcpy(e->comm, task->comm);
1303                 e->pid = task->pid;
1304         }
1305         rcu_read_unlock();
1306
1307         e->sched_attr = ctx->sched;
1308         e->guilty = atomic_read(&ctx->guilty_count);
1309         e->active = atomic_read(&ctx->active_count);
1310
1311         e->total_runtime = rq->context->runtime.total;
1312         e->avg_runtime = ewma_runtime_read(&rq->context->runtime.avg);
1313
1314         simulated = i915_gem_context_no_error_capture(ctx);
1315
1316         i915_gem_context_put(ctx);
1317         return simulated;
1318 }
1319
1320 struct intel_engine_capture_vma {
1321         struct intel_engine_capture_vma *next;
1322         struct i915_vma *vma;
1323         char name[16];
1324 };
1325
1326 static struct intel_engine_capture_vma *
1327 capture_vma(struct intel_engine_capture_vma *next,
1328             struct i915_vma *vma,
1329             const char *name,
1330             gfp_t gfp)
1331 {
1332         struct intel_engine_capture_vma *c;
1333
1334         if (!vma)
1335                 return next;
1336
1337         c = kmalloc(sizeof(*c), gfp);
1338         if (!c)
1339                 return next;
1340
1341         if (!i915_active_acquire_if_busy(&vma->active)) {
1342                 kfree(c);
1343                 return next;
1344         }
1345
1346         strcpy(c->name, name);
1347         c->vma = vma; /* reference held while active */
1348
1349         c->next = next;
1350         return c;
1351 }
1352
1353 static struct intel_engine_capture_vma *
1354 capture_user(struct intel_engine_capture_vma *capture,
1355              const struct i915_request *rq,
1356              gfp_t gfp)
1357 {
1358         struct i915_capture_list *c;
1359
1360         for (c = rq->capture_list; c; c = c->next)
1361                 capture = capture_vma(capture, c->vma, "user", gfp);
1362
1363         return capture;
1364 }
1365
1366 static void add_vma(struct intel_engine_coredump *ee,
1367                     struct i915_vma_coredump *vma)
1368 {
1369         if (vma) {
1370                 vma->next = ee->vma;
1371                 ee->vma = vma;
1372         }
1373 }
1374
1375 struct intel_engine_coredump *
1376 intel_engine_coredump_alloc(struct intel_engine_cs *engine, gfp_t gfp)
1377 {
1378         struct intel_engine_coredump *ee;
1379
1380         ee = kzalloc(sizeof(*ee), gfp);
1381         if (!ee)
1382                 return NULL;
1383
1384         ee->engine = engine;
1385
1386         engine_record_registers(ee);
1387         engine_record_execlists(ee);
1388
1389         return ee;
1390 }
1391
1392 struct intel_engine_capture_vma *
1393 intel_engine_coredump_add_request(struct intel_engine_coredump *ee,
1394                                   struct i915_request *rq,
1395                                   gfp_t gfp)
1396 {
1397         struct intel_engine_capture_vma *vma = NULL;
1398
1399         ee->simulated |= record_context(&ee->context, rq);
1400         if (ee->simulated)
1401                 return NULL;
1402
1403         /*
1404          * We need to copy these to an anonymous buffer
1405          * as the simplest method to avoid being overwritten
1406          * by userspace.
1407          */
1408         vma = capture_vma(vma, rq->batch, "batch", gfp);
1409         vma = capture_user(vma, rq, gfp);
1410         vma = capture_vma(vma, rq->ring->vma, "ring", gfp);
1411         vma = capture_vma(vma, rq->context->state, "HW context", gfp);
1412
1413         ee->rq_head = rq->head;
1414         ee->rq_post = rq->postfix;
1415         ee->rq_tail = rq->tail;
1416
1417         return vma;
1418 }
1419
1420 void
1421 intel_engine_coredump_add_vma(struct intel_engine_coredump *ee,
1422                               struct intel_engine_capture_vma *capture,
1423                               struct i915_vma_compress *compress)
1424 {
1425         const struct intel_engine_cs *engine = ee->engine;
1426
1427         while (capture) {
1428                 struct intel_engine_capture_vma *this = capture;
1429                 struct i915_vma *vma = this->vma;
1430
1431                 add_vma(ee,
1432                         i915_vma_coredump_create(engine->gt,
1433                                                  vma, this->name,
1434                                                  compress));
1435
1436                 i915_active_release(&vma->active);
1437
1438                 capture = this->next;
1439                 kfree(this);
1440         }
1441
1442         add_vma(ee,
1443                 i915_vma_coredump_create(engine->gt,
1444                                          engine->status_page.vma,
1445                                          "HW Status",
1446                                          compress));
1447
1448         add_vma(ee,
1449                 i915_vma_coredump_create(engine->gt,
1450                                          engine->wa_ctx.vma,
1451                                          "WA context",
1452                                          compress));
1453 }
1454
1455 static struct intel_engine_coredump *
1456 capture_engine(struct intel_engine_cs *engine,
1457                struct i915_vma_compress *compress)
1458 {
1459         struct intel_engine_capture_vma *capture = NULL;
1460         struct intel_engine_coredump *ee;
1461         struct intel_context *ce;
1462         struct i915_request *rq = NULL;
1463         unsigned long flags;
1464
1465         ee = intel_engine_coredump_alloc(engine, GFP_KERNEL);
1466         if (!ee)
1467                 return NULL;
1468
1469         ce = intel_engine_get_hung_context(engine);
1470         if (ce) {
1471                 intel_engine_clear_hung_context(engine);
1472                 rq = intel_context_find_active_request(ce);
1473                 if (!rq || !i915_request_started(rq))
1474                         goto no_request_capture;
1475         } else {
1476                 /*
1477                  * Getting here with GuC enabled means it is a forced error capture
1478                  * with no actual hang. So, no need to attempt the execlist search.
1479                  */
1480                 if (!intel_uc_uses_guc_submission(&engine->gt->uc)) {
1481                         spin_lock_irqsave(&engine->sched_engine->lock, flags);
1482                         rq = intel_engine_execlist_find_hung_request(engine);
1483                         spin_unlock_irqrestore(&engine->sched_engine->lock,
1484                                                flags);
1485                 }
1486         }
1487         if (rq)
1488                 capture = intel_engine_coredump_add_request(ee, rq,
1489                                                             ATOMIC_MAYFAIL);
1490         if (!capture) {
1491 no_request_capture:
1492                 kfree(ee);
1493                 return NULL;
1494         }
1495
1496         intel_engine_coredump_add_vma(ee, capture, compress);
1497
1498         return ee;
1499 }
1500
1501 static void
1502 gt_record_engines(struct intel_gt_coredump *gt,
1503                   intel_engine_mask_t engine_mask,
1504                   struct i915_vma_compress *compress)
1505 {
1506         struct intel_engine_cs *engine;
1507         enum intel_engine_id id;
1508
1509         for_each_engine(engine, gt->_gt, id) {
1510                 struct intel_engine_coredump *ee;
1511
1512                 /* Refill our page pool before entering atomic section */
1513                 pool_refill(&compress->pool, ALLOW_FAIL);
1514
1515                 ee = capture_engine(engine, compress);
1516                 if (!ee)
1517                         continue;
1518
1519                 ee->hung = engine->mask & engine_mask;
1520
1521                 gt->simulated |= ee->simulated;
1522                 if (ee->simulated) {
1523                         kfree(ee);
1524                         continue;
1525                 }
1526
1527                 ee->next = gt->engine;
1528                 gt->engine = ee;
1529         }
1530 }
1531
1532 static struct intel_uc_coredump *
1533 gt_record_uc(struct intel_gt_coredump *gt,
1534              struct i915_vma_compress *compress)
1535 {
1536         const struct intel_uc *uc = &gt->_gt->uc;
1537         struct intel_uc_coredump *error_uc;
1538
1539         error_uc = kzalloc(sizeof(*error_uc), ALLOW_FAIL);
1540         if (!error_uc)
1541                 return NULL;
1542
1543         memcpy(&error_uc->guc_fw, &uc->guc.fw, sizeof(uc->guc.fw));
1544         memcpy(&error_uc->huc_fw, &uc->huc.fw, sizeof(uc->huc.fw));
1545
1546         /* Non-default firmware paths will be specified by the modparam.
1547          * As modparams are generally accesible from the userspace make
1548          * explicit copies of the firmware paths.
1549          */
1550         error_uc->guc_fw.path = kstrdup(uc->guc.fw.path, ALLOW_FAIL);
1551         error_uc->huc_fw.path = kstrdup(uc->huc.fw.path, ALLOW_FAIL);
1552         error_uc->guc_log =
1553                 i915_vma_coredump_create(gt->_gt,
1554                                          uc->guc.log.vma, "GuC log buffer",
1555                                          compress);
1556
1557         return error_uc;
1558 }
1559
1560 /* Capture all registers which don't fit into another category. */
1561 static void gt_record_regs(struct intel_gt_coredump *gt)
1562 {
1563         struct intel_uncore *uncore = gt->_gt->uncore;
1564         struct drm_i915_private *i915 = uncore->i915;
1565         int i;
1566
1567         /*
1568          * General organization
1569          * 1. Registers specific to a single generation
1570          * 2. Registers which belong to multiple generations
1571          * 3. Feature specific registers.
1572          * 4. Everything else
1573          * Please try to follow the order.
1574          */
1575
1576         /* 1: Registers specific to a single generation */
1577         if (IS_VALLEYVIEW(i915)) {
1578                 gt->gtier[0] = intel_uncore_read(uncore, GTIER);
1579                 gt->ier = intel_uncore_read(uncore, VLV_IER);
1580                 gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_VLV);
1581         }
1582
1583         if (GRAPHICS_VER(i915) == 7)
1584                 gt->err_int = intel_uncore_read(uncore, GEN7_ERR_INT);
1585
1586         if (GRAPHICS_VER(i915) >= 12) {
1587                 gt->fault_data0 = intel_uncore_read(uncore,
1588                                                     GEN12_FAULT_TLB_DATA0);
1589                 gt->fault_data1 = intel_uncore_read(uncore,
1590                                                     GEN12_FAULT_TLB_DATA1);
1591         } else if (GRAPHICS_VER(i915) >= 8) {
1592                 gt->fault_data0 = intel_uncore_read(uncore,
1593                                                     GEN8_FAULT_TLB_DATA0);
1594                 gt->fault_data1 = intel_uncore_read(uncore,
1595                                                     GEN8_FAULT_TLB_DATA1);
1596         }
1597
1598         if (GRAPHICS_VER(i915) == 6) {
1599                 gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE);
1600                 gt->gab_ctl = intel_uncore_read(uncore, GAB_CTL);
1601                 gt->gfx_mode = intel_uncore_read(uncore, GFX_MODE);
1602         }
1603
1604         /* 2: Registers which belong to multiple generations */
1605         if (GRAPHICS_VER(i915) >= 7)
1606                 gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_MT);
1607
1608         if (GRAPHICS_VER(i915) >= 6) {
1609                 gt->derrmr = intel_uncore_read(uncore, DERRMR);
1610                 if (GRAPHICS_VER(i915) < 12) {
1611                         gt->error = intel_uncore_read(uncore, ERROR_GEN6);
1612                         gt->done_reg = intel_uncore_read(uncore, DONE_REG);
1613                 }
1614         }
1615
1616         /* 3: Feature specific registers */
1617         if (IS_GRAPHICS_VER(i915, 6, 7)) {
1618                 gt->gam_ecochk = intel_uncore_read(uncore, GAM_ECOCHK);
1619                 gt->gac_eco = intel_uncore_read(uncore, GAC_ECO_BITS);
1620         }
1621
1622         if (IS_GRAPHICS_VER(i915, 8, 11))
1623                 gt->gtt_cache = intel_uncore_read(uncore, HSW_GTT_CACHE_EN);
1624
1625         if (GRAPHICS_VER(i915) == 12)
1626                 gt->aux_err = intel_uncore_read(uncore, GEN12_AUX_ERR_DBG);
1627
1628         if (GRAPHICS_VER(i915) >= 12) {
1629                 for (i = 0; i < GEN12_SFC_DONE_MAX; i++) {
1630                         /*
1631                          * SFC_DONE resides in the VD forcewake domain, so it
1632                          * only exists if the corresponding VCS engine is
1633                          * present.
1634                          */
1635                         if (!HAS_ENGINE(gt->_gt, _VCS(i * 2)))
1636                                 continue;
1637
1638                         gt->sfc_done[i] =
1639                                 intel_uncore_read(uncore, GEN12_SFC_DONE(i));
1640                 }
1641
1642                 gt->gam_done = intel_uncore_read(uncore, GEN12_GAM_DONE);
1643         }
1644
1645         /* 4: Everything else */
1646         if (GRAPHICS_VER(i915) >= 11) {
1647                 gt->ier = intel_uncore_read(uncore, GEN8_DE_MISC_IER);
1648                 gt->gtier[0] =
1649                         intel_uncore_read(uncore,
1650                                           GEN11_RENDER_COPY_INTR_ENABLE);
1651                 gt->gtier[1] =
1652                         intel_uncore_read(uncore, GEN11_VCS_VECS_INTR_ENABLE);
1653                 gt->gtier[2] =
1654                         intel_uncore_read(uncore, GEN11_GUC_SG_INTR_ENABLE);
1655                 gt->gtier[3] =
1656                         intel_uncore_read(uncore,
1657                                           GEN11_GPM_WGBOXPERF_INTR_ENABLE);
1658                 gt->gtier[4] =
1659                         intel_uncore_read(uncore,
1660                                           GEN11_CRYPTO_RSVD_INTR_ENABLE);
1661                 gt->gtier[5] =
1662                         intel_uncore_read(uncore,
1663                                           GEN11_GUNIT_CSME_INTR_ENABLE);
1664                 gt->ngtier = 6;
1665         } else if (GRAPHICS_VER(i915) >= 8) {
1666                 gt->ier = intel_uncore_read(uncore, GEN8_DE_MISC_IER);
1667                 for (i = 0; i < 4; i++)
1668                         gt->gtier[i] =
1669                                 intel_uncore_read(uncore, GEN8_GT_IER(i));
1670                 gt->ngtier = 4;
1671         } else if (HAS_PCH_SPLIT(i915)) {
1672                 gt->ier = intel_uncore_read(uncore, DEIER);
1673                 gt->gtier[0] = intel_uncore_read(uncore, GTIER);
1674                 gt->ngtier = 1;
1675         } else if (GRAPHICS_VER(i915) == 2) {
1676                 gt->ier = intel_uncore_read16(uncore, GEN2_IER);
1677         } else if (!IS_VALLEYVIEW(i915)) {
1678                 gt->ier = intel_uncore_read(uncore, GEN2_IER);
1679         }
1680         gt->eir = intel_uncore_read(uncore, EIR);
1681         gt->pgtbl_er = intel_uncore_read(uncore, PGTBL_ER);
1682 }
1683
1684 static void gt_record_info(struct intel_gt_coredump *gt)
1685 {
1686         memcpy(&gt->info, &gt->_gt->info, sizeof(struct intel_gt_info));
1687 }
1688
1689 /*
1690  * Generate a semi-unique error code. The code is not meant to have meaning, The
1691  * code's only purpose is to try to prevent false duplicated bug reports by
1692  * grossly estimating a GPU error state.
1693  *
1694  * TODO Ideally, hashing the batchbuffer would be a very nice way to determine
1695  * the hang if we could strip the GTT offset information from it.
1696  *
1697  * It's only a small step better than a random number in its current form.
1698  */
1699 static u32 generate_ecode(const struct intel_engine_coredump *ee)
1700 {
1701         /*
1702          * IPEHR would be an ideal way to detect errors, as it's the gross
1703          * measure of "the command that hung." However, has some very common
1704          * synchronization commands which almost always appear in the case
1705          * strictly a client bug. Use instdone to differentiate those some.
1706          */
1707         return ee ? ee->ipehr ^ ee->instdone.instdone : 0;
1708 }
1709
1710 static const char *error_msg(struct i915_gpu_coredump *error)
1711 {
1712         struct intel_engine_coredump *first = NULL;
1713         unsigned int hung_classes = 0;
1714         struct intel_gt_coredump *gt;
1715         int len;
1716
1717         for (gt = error->gt; gt; gt = gt->next) {
1718                 struct intel_engine_coredump *cs;
1719
1720                 for (cs = gt->engine; cs; cs = cs->next) {
1721                         if (cs->hung) {
1722                                 hung_classes |= BIT(cs->engine->uabi_class);
1723                                 if (!first)
1724                                         first = cs;
1725                         }
1726                 }
1727         }
1728
1729         len = scnprintf(error->error_msg, sizeof(error->error_msg),
1730                         "GPU HANG: ecode %d:%x:%08x",
1731                         GRAPHICS_VER(error->i915), hung_classes,
1732                         generate_ecode(first));
1733         if (first && first->context.pid) {
1734                 /* Just show the first executing process, more is confusing */
1735                 len += scnprintf(error->error_msg + len,
1736                                  sizeof(error->error_msg) - len,
1737                                  ", in %s [%d]",
1738                                  first->context.comm, first->context.pid);
1739         }
1740
1741         return error->error_msg;
1742 }
1743
1744 static void capture_gen(struct i915_gpu_coredump *error)
1745 {
1746         struct drm_i915_private *i915 = error->i915;
1747
1748         error->wakelock = atomic_read(&i915->runtime_pm.wakeref_count);
1749         error->suspended = i915->runtime_pm.suspended;
1750
1751         error->iommu = -1;
1752 #ifdef CONFIG_INTEL_IOMMU
1753         error->iommu = intel_iommu_gfx_mapped;
1754 #endif
1755         error->reset_count = i915_reset_count(&i915->gpu_error);
1756         error->suspend_count = i915->suspend_count;
1757
1758         i915_params_copy(&error->params, &i915->params);
1759         memcpy(&error->device_info,
1760                INTEL_INFO(i915),
1761                sizeof(error->device_info));
1762         memcpy(&error->runtime_info,
1763                RUNTIME_INFO(i915),
1764                sizeof(error->runtime_info));
1765         error->driver_caps = i915->caps;
1766 }
1767
1768 struct i915_gpu_coredump *
1769 i915_gpu_coredump_alloc(struct drm_i915_private *i915, gfp_t gfp)
1770 {
1771         struct i915_gpu_coredump *error;
1772
1773         if (!i915->params.error_capture)
1774                 return NULL;
1775
1776         error = kzalloc(sizeof(*error), gfp);
1777         if (!error)
1778                 return NULL;
1779
1780         kref_init(&error->ref);
1781         error->i915 = i915;
1782
1783         error->time = ktime_get_real();
1784         error->boottime = ktime_get_boottime();
1785         error->uptime = ktime_sub(ktime_get(), i915->gt.last_init_time);
1786         error->capture = jiffies;
1787
1788         capture_gen(error);
1789
1790         return error;
1791 }
1792
1793 #define DAY_AS_SECONDS(x) (24 * 60 * 60 * (x))
1794
1795 struct intel_gt_coredump *
1796 intel_gt_coredump_alloc(struct intel_gt *gt, gfp_t gfp)
1797 {
1798         struct intel_gt_coredump *gc;
1799
1800         gc = kzalloc(sizeof(*gc), gfp);
1801         if (!gc)
1802                 return NULL;
1803
1804         gc->_gt = gt;
1805         gc->awake = intel_gt_pm_is_awake(gt);
1806
1807         gt_record_regs(gc);
1808         gt_record_fences(gc);
1809
1810         return gc;
1811 }
1812
1813 struct i915_vma_compress *
1814 i915_vma_capture_prepare(struct intel_gt_coredump *gt)
1815 {
1816         struct i915_vma_compress *compress;
1817
1818         compress = kmalloc(sizeof(*compress), ALLOW_FAIL);
1819         if (!compress)
1820                 return NULL;
1821
1822         if (!compress_init(compress)) {
1823                 kfree(compress);
1824                 return NULL;
1825         }
1826
1827         return compress;
1828 }
1829
1830 void i915_vma_capture_finish(struct intel_gt_coredump *gt,
1831                              struct i915_vma_compress *compress)
1832 {
1833         if (!compress)
1834                 return;
1835
1836         compress_fini(compress);
1837         kfree(compress);
1838 }
1839
1840 struct i915_gpu_coredump *
1841 i915_gpu_coredump(struct intel_gt *gt, intel_engine_mask_t engine_mask)
1842 {
1843         struct drm_i915_private *i915 = gt->i915;
1844         struct i915_gpu_coredump *error;
1845
1846         /* Check if GPU capture has been disabled */
1847         error = READ_ONCE(i915->gpu_error.first_error);
1848         if (IS_ERR(error))
1849                 return error;
1850
1851         error = i915_gpu_coredump_alloc(i915, ALLOW_FAIL);
1852         if (!error)
1853                 return ERR_PTR(-ENOMEM);
1854
1855         error->gt = intel_gt_coredump_alloc(gt, ALLOW_FAIL);
1856         if (error->gt) {
1857                 struct i915_vma_compress *compress;
1858
1859                 compress = i915_vma_capture_prepare(error->gt);
1860                 if (!compress) {
1861                         kfree(error->gt);
1862                         kfree(error);
1863                         return ERR_PTR(-ENOMEM);
1864                 }
1865
1866                 gt_record_info(error->gt);
1867                 gt_record_engines(error->gt, engine_mask, compress);
1868
1869                 if (INTEL_INFO(i915)->has_gt_uc)
1870                         error->gt->uc = gt_record_uc(error->gt, compress);
1871
1872                 i915_vma_capture_finish(error->gt, compress);
1873
1874                 error->simulated |= error->gt->simulated;
1875         }
1876
1877         error->overlay = intel_overlay_capture_error_state(i915);
1878
1879         return error;
1880 }
1881
1882 void i915_error_state_store(struct i915_gpu_coredump *error)
1883 {
1884         struct drm_i915_private *i915;
1885         static bool warned;
1886
1887         if (IS_ERR_OR_NULL(error))
1888                 return;
1889
1890         i915 = error->i915;
1891         drm_info(&i915->drm, "%s\n", error_msg(error));
1892
1893         if (error->simulated ||
1894             cmpxchg(&i915->gpu_error.first_error, NULL, error))
1895                 return;
1896
1897         i915_gpu_coredump_get(error);
1898
1899         if (!xchg(&warned, true) &&
1900             ktime_get_real_seconds() - DRIVER_TIMESTAMP < DAY_AS_SECONDS(180)) {
1901                 pr_info("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n");
1902                 pr_info("Please file a _new_ bug report at https://gitlab.freedesktop.org/drm/intel/issues/new.\n");
1903                 pr_info("Please see https://gitlab.freedesktop.org/drm/intel/-/wikis/How-to-file-i915-bugs for details.\n");
1904                 pr_info("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n");
1905                 pr_info("The GPU crash dump is required to analyze GPU hangs, so please always attach it.\n");
1906                 pr_info("GPU crash dump saved to /sys/class/drm/card%d/error\n",
1907                         i915->drm.primary->index);
1908         }
1909 }
1910
1911 /**
1912  * i915_capture_error_state - capture an error record for later analysis
1913  * @gt: intel_gt which originated the hang
1914  * @engine_mask: hung engines
1915  *
1916  *
1917  * Should be called when an error is detected (either a hang or an error
1918  * interrupt) to capture error state from the time of the error.  Fills
1919  * out a structure which becomes available in debugfs for user level tools
1920  * to pick up.
1921  */
1922 void i915_capture_error_state(struct intel_gt *gt,
1923                               intel_engine_mask_t engine_mask)
1924 {
1925         struct i915_gpu_coredump *error;
1926
1927         error = i915_gpu_coredump(gt, engine_mask);
1928         if (IS_ERR(error)) {
1929                 cmpxchg(&gt->i915->gpu_error.first_error, NULL, error);
1930                 return;
1931         }
1932
1933         i915_error_state_store(error);
1934         i915_gpu_coredump_put(error);
1935 }
1936
1937 struct i915_gpu_coredump *
1938 i915_first_error_state(struct drm_i915_private *i915)
1939 {
1940         struct i915_gpu_coredump *error;
1941
1942         spin_lock_irq(&i915->gpu_error.lock);
1943         error = i915->gpu_error.first_error;
1944         if (!IS_ERR_OR_NULL(error))
1945                 i915_gpu_coredump_get(error);
1946         spin_unlock_irq(&i915->gpu_error.lock);
1947
1948         return error;
1949 }
1950
1951 void i915_reset_error_state(struct drm_i915_private *i915)
1952 {
1953         struct i915_gpu_coredump *error;
1954
1955         spin_lock_irq(&i915->gpu_error.lock);
1956         error = i915->gpu_error.first_error;
1957         if (error != ERR_PTR(-ENODEV)) /* if disabled, always disabled */
1958                 i915->gpu_error.first_error = NULL;
1959         spin_unlock_irq(&i915->gpu_error.lock);
1960
1961         if (!IS_ERR_OR_NULL(error))
1962                 i915_gpu_coredump_put(error);
1963 }
1964
1965 void i915_disable_error_state(struct drm_i915_private *i915, int err)
1966 {
1967         spin_lock_irq(&i915->gpu_error.lock);
1968         if (!i915->gpu_error.first_error)
1969                 i915->gpu_error.first_error = ERR_PTR(err);
1970         spin_unlock_irq(&i915->gpu_error.lock);
1971 }