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/highmem.h> 32 #include <linux/nmi.h> 33 #include <linux/pagevec.h> 34 #include <linux/scatterlist.h> 35 #include <linux/string_helpers.h> 36 #include <linux/utsname.h> 37 #include <linux/zlib.h> 38 39 #include <drm/drm_cache.h> 40 #include <drm/drm_print.h> 41 42 #include "display/intel_dmc.h" 43 #include "display/intel_overlay.h" 44 45 #include "gem/i915_gem_context.h" 46 #include "gem/i915_gem_lmem.h" 47 #include "gt/intel_engine_regs.h" 48 #include "gt/intel_gt.h" 49 #include "gt/intel_gt_mcr.h" 50 #include "gt/intel_gt_pm.h" 51 #include "gt/intel_gt_regs.h" 52 #include "gt/uc/intel_guc_capture.h" 53 54 #include "i915_driver.h" 55 #include "i915_drv.h" 56 #include "i915_gpu_error.h" 57 #include "i915_memcpy.h" 58 #include "i915_reg.h" 59 #include "i915_scatterlist.h" 60 #include "i915_utils.h" 61 62 #define ALLOW_FAIL (__GFP_KSWAPD_RECLAIM | __GFP_RETRY_MAYFAIL | __GFP_NOWARN) 63 #define ATOMIC_MAYFAIL (GFP_ATOMIC | __GFP_NOWARN) 64 65 static void __sg_set_buf(struct scatterlist *sg, 66 void *addr, unsigned int len, loff_t it) 67 { 68 sg->page_link = (unsigned long)virt_to_page(addr); 69 sg->offset = offset_in_page(addr); 70 sg->length = len; 71 sg->dma_address = it; 72 } 73 74 static bool __i915_error_grow(struct drm_i915_error_state_buf *e, size_t len) 75 { 76 if (!len) 77 return false; 78 79 if (e->bytes + len + 1 <= e->size) 80 return true; 81 82 if (e->bytes) { 83 __sg_set_buf(e->cur++, e->buf, e->bytes, e->iter); 84 e->iter += e->bytes; 85 e->buf = NULL; 86 e->bytes = 0; 87 } 88 89 if (e->cur == e->end) { 90 struct scatterlist *sgl; 91 92 sgl = (typeof(sgl))__get_free_page(ALLOW_FAIL); 93 if (!sgl) { 94 e->err = -ENOMEM; 95 return false; 96 } 97 98 if (e->cur) { 99 e->cur->offset = 0; 100 e->cur->length = 0; 101 e->cur->page_link = 102 (unsigned long)sgl | SG_CHAIN; 103 } else { 104 e->sgl = sgl; 105 } 106 107 e->cur = sgl; 108 e->end = sgl + SG_MAX_SINGLE_ALLOC - 1; 109 } 110 111 e->size = ALIGN(len + 1, SZ_64K); 112 e->buf = kmalloc(e->size, ALLOW_FAIL); 113 if (!e->buf) { 114 e->size = PAGE_ALIGN(len + 1); 115 e->buf = kmalloc(e->size, GFP_KERNEL); 116 } 117 if (!e->buf) { 118 e->err = -ENOMEM; 119 return false; 120 } 121 122 return true; 123 } 124 125 __printf(2, 0) 126 static void i915_error_vprintf(struct drm_i915_error_state_buf *e, 127 const char *fmt, va_list args) 128 { 129 va_list ap; 130 int len; 131 132 if (e->err) 133 return; 134 135 va_copy(ap, args); 136 len = vsnprintf(NULL, 0, fmt, ap); 137 va_end(ap); 138 if (len <= 0) { 139 e->err = len; 140 return; 141 } 142 143 if (!__i915_error_grow(e, len)) 144 return; 145 146 GEM_BUG_ON(e->bytes >= e->size); 147 len = vscnprintf(e->buf + e->bytes, e->size - e->bytes, fmt, args); 148 if (len < 0) { 149 e->err = len; 150 return; 151 } 152 e->bytes += len; 153 } 154 155 static void i915_error_puts(struct drm_i915_error_state_buf *e, const char *str) 156 { 157 unsigned len; 158 159 if (e->err || !str) 160 return; 161 162 len = strlen(str); 163 if (!__i915_error_grow(e, len)) 164 return; 165 166 GEM_BUG_ON(e->bytes + len > e->size); 167 memcpy(e->buf + e->bytes, str, len); 168 e->bytes += len; 169 } 170 171 #define err_printf(e, ...) i915_error_printf(e, __VA_ARGS__) 172 #define err_puts(e, s) i915_error_puts(e, s) 173 174 static void __i915_printfn_error(struct drm_printer *p, struct va_format *vaf) 175 { 176 i915_error_vprintf(p->arg, vaf->fmt, *vaf->va); 177 } 178 179 static inline struct drm_printer 180 i915_error_printer(struct drm_i915_error_state_buf *e) 181 { 182 struct drm_printer p = { 183 .printfn = __i915_printfn_error, 184 .arg = e, 185 }; 186 return p; 187 } 188 189 /* single threaded page allocator with a reserved stash for emergencies */ 190 static void pool_fini(struct folio_batch *fbatch) 191 { 192 folio_batch_release(fbatch); 193 } 194 195 static int pool_refill(struct folio_batch *fbatch, gfp_t gfp) 196 { 197 while (folio_batch_space(fbatch)) { 198 struct folio *folio; 199 200 folio = folio_alloc(gfp, 0); 201 if (!folio) 202 return -ENOMEM; 203 204 folio_batch_add(fbatch, folio); 205 } 206 207 return 0; 208 } 209 210 static int pool_init(struct folio_batch *fbatch, gfp_t gfp) 211 { 212 int err; 213 214 folio_batch_init(fbatch); 215 216 err = pool_refill(fbatch, gfp); 217 if (err) 218 pool_fini(fbatch); 219 220 return err; 221 } 222 223 static void *pool_alloc(struct folio_batch *fbatch, gfp_t gfp) 224 { 225 struct folio *folio; 226 227 folio = folio_alloc(gfp, 0); 228 if (!folio && folio_batch_count(fbatch)) 229 folio = fbatch->folios[--fbatch->nr]; 230 231 return folio ? folio_address(folio) : NULL; 232 } 233 234 static void pool_free(struct folio_batch *fbatch, void *addr) 235 { 236 struct folio *folio = virt_to_folio(addr); 237 238 if (folio_batch_space(fbatch)) 239 folio_batch_add(fbatch, folio); 240 else 241 folio_put(folio); 242 } 243 244 #ifdef CONFIG_DRM_I915_COMPRESS_ERROR 245 246 struct i915_vma_compress { 247 struct folio_batch pool; 248 struct z_stream_s zstream; 249 void *tmp; 250 }; 251 252 static bool compress_init(struct i915_vma_compress *c) 253 { 254 struct z_stream_s *zstream = &c->zstream; 255 256 if (pool_init(&c->pool, ALLOW_FAIL)) 257 return false; 258 259 zstream->workspace = 260 kmalloc(zlib_deflate_workspacesize(MAX_WBITS, MAX_MEM_LEVEL), 261 ALLOW_FAIL); 262 if (!zstream->workspace) { 263 pool_fini(&c->pool); 264 return false; 265 } 266 267 c->tmp = NULL; 268 if (i915_has_memcpy_from_wc()) 269 c->tmp = pool_alloc(&c->pool, ALLOW_FAIL); 270 271 return true; 272 } 273 274 static bool compress_start(struct i915_vma_compress *c) 275 { 276 struct z_stream_s *zstream = &c->zstream; 277 void *workspace = zstream->workspace; 278 279 memset(zstream, 0, sizeof(*zstream)); 280 zstream->workspace = workspace; 281 282 return zlib_deflateInit(zstream, Z_DEFAULT_COMPRESSION) == Z_OK; 283 } 284 285 static void *compress_next_page(struct i915_vma_compress *c, 286 struct i915_vma_coredump *dst) 287 { 288 void *page_addr; 289 struct page *page; 290 291 page_addr = pool_alloc(&c->pool, ALLOW_FAIL); 292 if (!page_addr) 293 return ERR_PTR(-ENOMEM); 294 295 page = virt_to_page(page_addr); 296 list_add_tail(&page->lru, &dst->page_list); 297 return page_addr; 298 } 299 300 static int compress_page(struct i915_vma_compress *c, 301 void *src, 302 struct i915_vma_coredump *dst, 303 bool wc) 304 { 305 struct z_stream_s *zstream = &c->zstream; 306 307 zstream->next_in = src; 308 if (wc && c->tmp && i915_memcpy_from_wc(c->tmp, src, PAGE_SIZE)) 309 zstream->next_in = c->tmp; 310 zstream->avail_in = PAGE_SIZE; 311 312 do { 313 if (zstream->avail_out == 0) { 314 zstream->next_out = compress_next_page(c, dst); 315 if (IS_ERR(zstream->next_out)) 316 return PTR_ERR(zstream->next_out); 317 318 zstream->avail_out = PAGE_SIZE; 319 } 320 321 if (zlib_deflate(zstream, Z_NO_FLUSH) != Z_OK) 322 return -EIO; 323 324 cond_resched(); 325 } while (zstream->avail_in); 326 327 /* Fallback to uncompressed if we increase size? */ 328 if (0 && zstream->total_out > zstream->total_in) 329 return -E2BIG; 330 331 return 0; 332 } 333 334 static int compress_flush(struct i915_vma_compress *c, 335 struct i915_vma_coredump *dst) 336 { 337 struct z_stream_s *zstream = &c->zstream; 338 339 do { 340 switch (zlib_deflate(zstream, Z_FINISH)) { 341 case Z_OK: /* more space requested */ 342 zstream->next_out = compress_next_page(c, dst); 343 if (IS_ERR(zstream->next_out)) 344 return PTR_ERR(zstream->next_out); 345 346 zstream->avail_out = PAGE_SIZE; 347 break; 348 349 case Z_STREAM_END: 350 goto end; 351 352 default: /* any error */ 353 return -EIO; 354 } 355 } while (1); 356 357 end: 358 memset(zstream->next_out, 0, zstream->avail_out); 359 dst->unused = zstream->avail_out; 360 return 0; 361 } 362 363 static void compress_finish(struct i915_vma_compress *c) 364 { 365 zlib_deflateEnd(&c->zstream); 366 } 367 368 static void compress_fini(struct i915_vma_compress *c) 369 { 370 kfree(c->zstream.workspace); 371 if (c->tmp) 372 pool_free(&c->pool, c->tmp); 373 pool_fini(&c->pool); 374 } 375 376 static void err_compression_marker(struct drm_i915_error_state_buf *m) 377 { 378 err_puts(m, ":"); 379 } 380 381 #else 382 383 struct i915_vma_compress { 384 struct folio_batch pool; 385 }; 386 387 static bool compress_init(struct i915_vma_compress *c) 388 { 389 return pool_init(&c->pool, ALLOW_FAIL) == 0; 390 } 391 392 static bool compress_start(struct i915_vma_compress *c) 393 { 394 return true; 395 } 396 397 static int compress_page(struct i915_vma_compress *c, 398 void *src, 399 struct i915_vma_coredump *dst, 400 bool wc) 401 { 402 void *ptr; 403 404 ptr = pool_alloc(&c->pool, ALLOW_FAIL); 405 if (!ptr) 406 return -ENOMEM; 407 408 if (!(wc && i915_memcpy_from_wc(ptr, src, PAGE_SIZE))) 409 memcpy(ptr, src, PAGE_SIZE); 410 list_add_tail(&virt_to_page(ptr)->lru, &dst->page_list); 411 cond_resched(); 412 413 return 0; 414 } 415 416 static int compress_flush(struct i915_vma_compress *c, 417 struct i915_vma_coredump *dst) 418 { 419 return 0; 420 } 421 422 static void compress_finish(struct i915_vma_compress *c) 423 { 424 } 425 426 static void compress_fini(struct i915_vma_compress *c) 427 { 428 pool_fini(&c->pool); 429 } 430 431 static void err_compression_marker(struct drm_i915_error_state_buf *m) 432 { 433 err_puts(m, "~"); 434 } 435 436 #endif 437 438 static void error_print_instdone(struct drm_i915_error_state_buf *m, 439 const struct intel_engine_coredump *ee) 440 { 441 int slice; 442 int subslice; 443 int iter; 444 445 err_printf(m, " INSTDONE: 0x%08x\n", 446 ee->instdone.instdone); 447 448 if (ee->engine->class != RENDER_CLASS || GRAPHICS_VER(m->i915) <= 3) 449 return; 450 451 err_printf(m, " SC_INSTDONE: 0x%08x\n", 452 ee->instdone.slice_common); 453 454 if (GRAPHICS_VER(m->i915) <= 6) 455 return; 456 457 for_each_ss_steering(iter, ee->engine->gt, slice, subslice) 458 err_printf(m, " SAMPLER_INSTDONE[%d][%d]: 0x%08x\n", 459 slice, subslice, 460 ee->instdone.sampler[slice][subslice]); 461 462 for_each_ss_steering(iter, ee->engine->gt, slice, subslice) 463 err_printf(m, " ROW_INSTDONE[%d][%d]: 0x%08x\n", 464 slice, subslice, 465 ee->instdone.row[slice][subslice]); 466 467 if (GRAPHICS_VER(m->i915) < 12) 468 return; 469 470 if (GRAPHICS_VER_FULL(m->i915) >= IP_VER(12, 55)) { 471 for_each_ss_steering(iter, ee->engine->gt, slice, subslice) 472 err_printf(m, " GEOM_SVGUNIT_INSTDONE[%d][%d]: 0x%08x\n", 473 slice, subslice, 474 ee->instdone.geom_svg[slice][subslice]); 475 } 476 477 err_printf(m, " SC_INSTDONE_EXTRA: 0x%08x\n", 478 ee->instdone.slice_common_extra[0]); 479 err_printf(m, " SC_INSTDONE_EXTRA2: 0x%08x\n", 480 ee->instdone.slice_common_extra[1]); 481 } 482 483 static void error_print_request(struct drm_i915_error_state_buf *m, 484 const char *prefix, 485 const struct i915_request_coredump *erq) 486 { 487 if (!erq->seqno) 488 return; 489 490 err_printf(m, "%s pid %d, seqno %8x:%08x%s%s, prio %d, head %08x, tail %08x\n", 491 prefix, erq->pid, erq->context, erq->seqno, 492 test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, 493 &erq->flags) ? "!" : "", 494 test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, 495 &erq->flags) ? "+" : "", 496 erq->sched_attr.priority, 497 erq->head, erq->tail); 498 } 499 500 static void error_print_context(struct drm_i915_error_state_buf *m, 501 const char *header, 502 const struct i915_gem_context_coredump *ctx) 503 { 504 err_printf(m, "%s%s[%d] prio %d, guilty %d active %d, runtime total %lluns, avg %lluns\n", 505 header, ctx->comm, ctx->pid, ctx->sched_attr.priority, 506 ctx->guilty, ctx->active, 507 ctx->total_runtime, ctx->avg_runtime); 508 err_printf(m, " context timeline seqno %u\n", ctx->hwsp_seqno); 509 } 510 511 static struct i915_vma_coredump * 512 __find_vma(struct i915_vma_coredump *vma, const char *name) 513 { 514 while (vma) { 515 if (strcmp(vma->name, name) == 0) 516 return vma; 517 vma = vma->next; 518 } 519 520 return NULL; 521 } 522 523 struct i915_vma_coredump * 524 intel_gpu_error_find_batch(const struct intel_engine_coredump *ee) 525 { 526 return __find_vma(ee->vma, "batch"); 527 } 528 529 static void error_print_engine(struct drm_i915_error_state_buf *m, 530 const struct intel_engine_coredump *ee) 531 { 532 struct i915_vma_coredump *batch; 533 int n; 534 535 err_printf(m, "%s command stream:\n", ee->engine->name); 536 err_printf(m, " CCID: 0x%08x\n", ee->ccid); 537 err_printf(m, " START: 0x%08x\n", ee->start); 538 err_printf(m, " HEAD: 0x%08x [0x%08x]\n", ee->head, ee->rq_head); 539 err_printf(m, " TAIL: 0x%08x [0x%08x, 0x%08x]\n", 540 ee->tail, ee->rq_post, ee->rq_tail); 541 err_printf(m, " CTL: 0x%08x\n", ee->ctl); 542 err_printf(m, " MODE: 0x%08x\n", ee->mode); 543 err_printf(m, " HWS: 0x%08x\n", ee->hws); 544 err_printf(m, " ACTHD: 0x%08x %08x\n", 545 (u32)(ee->acthd>>32), (u32)ee->acthd); 546 err_printf(m, " IPEIR: 0x%08x\n", ee->ipeir); 547 err_printf(m, " IPEHR: 0x%08x\n", ee->ipehr); 548 err_printf(m, " ESR: 0x%08x\n", ee->esr); 549 550 error_print_instdone(m, ee); 551 552 batch = intel_gpu_error_find_batch(ee); 553 if (batch) { 554 u64 start = batch->gtt_offset; 555 u64 end = start + batch->gtt_size; 556 557 err_printf(m, " batch: [0x%08x_%08x, 0x%08x_%08x]\n", 558 upper_32_bits(start), lower_32_bits(start), 559 upper_32_bits(end), lower_32_bits(end)); 560 } 561 if (GRAPHICS_VER(m->i915) >= 4) { 562 err_printf(m, " BBADDR: 0x%08x_%08x\n", 563 (u32)(ee->bbaddr>>32), (u32)ee->bbaddr); 564 err_printf(m, " BB_STATE: 0x%08x\n", ee->bbstate); 565 err_printf(m, " INSTPS: 0x%08x\n", ee->instps); 566 } 567 err_printf(m, " INSTPM: 0x%08x\n", ee->instpm); 568 err_printf(m, " FADDR: 0x%08x %08x\n", upper_32_bits(ee->faddr), 569 lower_32_bits(ee->faddr)); 570 if (GRAPHICS_VER(m->i915) >= 6) { 571 err_printf(m, " RC PSMI: 0x%08x\n", ee->rc_psmi); 572 err_printf(m, " FAULT_REG: 0x%08x\n", ee->fault_reg); 573 } 574 if (GRAPHICS_VER(m->i915) >= 11) { 575 err_printf(m, " NOPID: 0x%08x\n", ee->nopid); 576 err_printf(m, " EXCC: 0x%08x\n", ee->excc); 577 err_printf(m, " CMD_CCTL: 0x%08x\n", ee->cmd_cctl); 578 err_printf(m, " CSCMDOP: 0x%08x\n", ee->cscmdop); 579 err_printf(m, " CTX_SR_CTL: 0x%08x\n", ee->ctx_sr_ctl); 580 err_printf(m, " DMA_FADDR_HI: 0x%08x\n", ee->dma_faddr_hi); 581 err_printf(m, " DMA_FADDR_LO: 0x%08x\n", ee->dma_faddr_lo); 582 } 583 if (HAS_PPGTT(m->i915)) { 584 err_printf(m, " GFX_MODE: 0x%08x\n", ee->vm_info.gfx_mode); 585 586 if (GRAPHICS_VER(m->i915) >= 8) { 587 int i; 588 for (i = 0; i < 4; i++) 589 err_printf(m, " PDP%d: 0x%016llx\n", 590 i, ee->vm_info.pdp[i]); 591 } else { 592 err_printf(m, " PP_DIR_BASE: 0x%08x\n", 593 ee->vm_info.pp_dir_base); 594 } 595 } 596 597 for (n = 0; n < ee->num_ports; n++) { 598 err_printf(m, " ELSP[%d]:", n); 599 error_print_request(m, " ", &ee->execlist[n]); 600 } 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 void intel_gpu_error_print_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 struct page *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 list_for_each_entry(page, &vma->page_list, lru) { 632 int i, len; 633 const u32 *addr = page_address(page); 634 635 len = PAGE_SIZE; 636 if (page == list_last_entry(&vma->page_list, typeof(*page), lru)) 637 len -= vma->unused; 638 len = ascii85_encode_len(len); 639 640 for (i = 0; i < len; i++) 641 err_puts(m, ascii85_encode(addr[i], out)); 642 } 643 err_puts(m, "\n"); 644 } 645 646 static void err_print_capabilities(struct drm_i915_error_state_buf *m, 647 struct i915_gpu_coredump *error) 648 { 649 struct drm_printer p = i915_error_printer(m); 650 651 intel_device_info_print(&error->device_info, &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_guc_ctb(struct drm_i915_error_state_buf *m, 676 const char *name, 677 const struct intel_ctb_coredump *ctb) 678 { 679 if (!ctb->size) 680 return; 681 682 err_printf(m, "GuC %s CTB: raw: 0x%08X, 0x%08X/%08X, cached: 0x%08X/%08X, desc = 0x%08X, buf = 0x%08X x 0x%08X\n", 683 name, ctb->raw_status, ctb->raw_head, ctb->raw_tail, 684 ctb->head, ctb->tail, ctb->desc_offset, ctb->cmds_offset, ctb->size); 685 } 686 687 static void err_print_uc(struct drm_i915_error_state_buf *m, 688 const struct intel_uc_coredump *error_uc) 689 { 690 struct drm_printer p = i915_error_printer(m); 691 692 intel_uc_fw_dump(&error_uc->guc_fw, &p); 693 intel_uc_fw_dump(&error_uc->huc_fw, &p); 694 err_printf(m, "GuC timestamp: 0x%08x\n", error_uc->guc.timestamp); 695 intel_gpu_error_print_vma(m, NULL, error_uc->guc.vma_log); 696 err_printf(m, "GuC CTB fence: %d\n", error_uc->guc.last_fence); 697 err_print_guc_ctb(m, "Send", error_uc->guc.ctb + 0); 698 err_print_guc_ctb(m, "Recv", error_uc->guc.ctb + 1); 699 intel_gpu_error_print_vma(m, NULL, error_uc->guc.vma_ctb); 700 } 701 702 static void err_free_sgl(struct scatterlist *sgl) 703 { 704 while (sgl) { 705 struct scatterlist *sg; 706 707 for (sg = sgl; !sg_is_chain(sg); sg++) { 708 kfree(sg_virt(sg)); 709 if (sg_is_last(sg)) 710 break; 711 } 712 713 sg = sg_is_last(sg) ? NULL : sg_chain_ptr(sg); 714 free_page((unsigned long)sgl); 715 sgl = sg; 716 } 717 } 718 719 static void err_print_gt_info(struct drm_i915_error_state_buf *m, 720 struct intel_gt_coredump *gt) 721 { 722 struct drm_printer p = i915_error_printer(m); 723 724 intel_gt_info_print(>->info, &p); 725 intel_sseu_print_topology(gt->_gt->i915, >->info.sseu, &p); 726 } 727 728 static void err_print_gt_display(struct drm_i915_error_state_buf *m, 729 struct intel_gt_coredump *gt) 730 { 731 err_printf(m, "IER: 0x%08x\n", gt->ier); 732 err_printf(m, "DERRMR: 0x%08x\n", gt->derrmr); 733 } 734 735 static void err_print_gt_global_nonguc(struct drm_i915_error_state_buf *m, 736 struct intel_gt_coredump *gt) 737 { 738 int i; 739 740 err_printf(m, "GT awake: %s\n", str_yes_no(gt->awake)); 741 err_printf(m, "CS timestamp frequency: %u Hz, %d ns\n", 742 gt->clock_frequency, gt->clock_period_ns); 743 err_printf(m, "EIR: 0x%08x\n", gt->eir); 744 err_printf(m, "PGTBL_ER: 0x%08x\n", gt->pgtbl_er); 745 746 for (i = 0; i < gt->ngtier; i++) 747 err_printf(m, "GTIER[%d]: 0x%08x\n", i, gt->gtier[i]); 748 } 749 750 static void err_print_gt_global(struct drm_i915_error_state_buf *m, 751 struct intel_gt_coredump *gt) 752 { 753 err_printf(m, "FORCEWAKE: 0x%08x\n", gt->forcewake); 754 755 if (IS_GRAPHICS_VER(m->i915, 6, 11)) { 756 err_printf(m, "ERROR: 0x%08x\n", gt->error); 757 err_printf(m, "DONE_REG: 0x%08x\n", gt->done_reg); 758 } 759 760 if (GRAPHICS_VER(m->i915) >= 8) 761 err_printf(m, "FAULT_TLB_DATA: 0x%08x 0x%08x\n", 762 gt->fault_data1, gt->fault_data0); 763 764 if (GRAPHICS_VER(m->i915) == 7) 765 err_printf(m, "ERR_INT: 0x%08x\n", gt->err_int); 766 767 if (IS_GRAPHICS_VER(m->i915, 8, 11)) 768 err_printf(m, "GTT_CACHE_EN: 0x%08x\n", gt->gtt_cache); 769 770 if (GRAPHICS_VER(m->i915) == 12) 771 err_printf(m, "AUX_ERR_DBG: 0x%08x\n", gt->aux_err); 772 773 if (GRAPHICS_VER(m->i915) >= 12) { 774 int i; 775 776 for (i = 0; i < I915_MAX_SFC; i++) { 777 /* 778 * SFC_DONE resides in the VD forcewake domain, so it 779 * only exists if the corresponding VCS engine is 780 * present. 781 */ 782 if ((gt->_gt->info.sfc_mask & BIT(i)) == 0 || 783 !HAS_ENGINE(gt->_gt, _VCS(i * 2))) 784 continue; 785 786 err_printf(m, " SFC_DONE[%d]: 0x%08x\n", i, 787 gt->sfc_done[i]); 788 } 789 790 err_printf(m, " GAM_DONE: 0x%08x\n", gt->gam_done); 791 } 792 } 793 794 static void err_print_gt_fences(struct drm_i915_error_state_buf *m, 795 struct intel_gt_coredump *gt) 796 { 797 int i; 798 799 for (i = 0; i < gt->nfence; i++) 800 err_printf(m, " fence[%d] = %08llx\n", i, gt->fence[i]); 801 } 802 803 static void err_print_gt_engines(struct drm_i915_error_state_buf *m, 804 struct intel_gt_coredump *gt) 805 { 806 const struct intel_engine_coredump *ee; 807 808 for (ee = gt->engine; ee; ee = ee->next) { 809 const struct i915_vma_coredump *vma; 810 811 if (gt->uc && gt->uc->guc.is_guc_capture) { 812 if (ee->guc_capture_node) 813 intel_guc_capture_print_engine_node(m, ee); 814 else 815 err_printf(m, " Missing GuC capture node for %s\n", 816 ee->engine->name); 817 } else { 818 error_print_engine(m, ee); 819 } 820 821 err_printf(m, " hung: %u\n", ee->hung); 822 err_printf(m, " engine reset count: %u\n", ee->reset_count); 823 error_print_context(m, " Active context: ", &ee->context); 824 825 for (vma = ee->vma; vma; vma = vma->next) 826 intel_gpu_error_print_vma(m, ee->engine, vma); 827 } 828 829 } 830 831 static void __err_print_to_sgl(struct drm_i915_error_state_buf *m, 832 struct i915_gpu_coredump *error) 833 { 834 const struct intel_engine_coredump *ee; 835 struct timespec64 ts; 836 837 if (*error->error_msg) 838 err_printf(m, "%s\n", error->error_msg); 839 err_printf(m, "Kernel: %s %s\n", 840 init_utsname()->release, 841 init_utsname()->machine); 842 err_printf(m, "Driver: %s\n", DRIVER_DATE); 843 ts = ktime_to_timespec64(error->time); 844 err_printf(m, "Time: %lld s %ld us\n", 845 (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC); 846 ts = ktime_to_timespec64(error->boottime); 847 err_printf(m, "Boottime: %lld s %ld us\n", 848 (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC); 849 ts = ktime_to_timespec64(error->uptime); 850 err_printf(m, "Uptime: %lld s %ld us\n", 851 (s64)ts.tv_sec, ts.tv_nsec / NSEC_PER_USEC); 852 err_printf(m, "Capture: %lu jiffies; %d ms ago\n", 853 error->capture, jiffies_to_msecs(jiffies - error->capture)); 854 855 for (ee = error->gt ? error->gt->engine : NULL; ee; ee = ee->next) 856 err_printf(m, "Active process (on ring %s): %s [%d]\n", 857 ee->engine->name, 858 ee->context.comm, 859 ee->context.pid); 860 861 err_printf(m, "Reset count: %u\n", error->reset_count); 862 err_printf(m, "Suspend count: %u\n", error->suspend_count); 863 err_printf(m, "Platform: %s\n", intel_platform_name(error->device_info.platform)); 864 err_printf(m, "Subplatform: 0x%x\n", 865 intel_subplatform(&error->runtime_info, 866 error->device_info.platform)); 867 err_print_pciid(m, m->i915); 868 869 err_printf(m, "IOMMU enabled?: %d\n", error->iommu); 870 871 intel_dmc_print_error_state(m, m->i915); 872 873 err_printf(m, "RPM wakelock: %s\n", str_yes_no(error->wakelock)); 874 err_printf(m, "PM suspended: %s\n", str_yes_no(error->suspended)); 875 876 if (error->gt) { 877 bool print_guc_capture = false; 878 879 if (error->gt->uc && error->gt->uc->guc.is_guc_capture) 880 print_guc_capture = true; 881 882 err_print_gt_display(m, error->gt); 883 err_print_gt_global_nonguc(m, error->gt); 884 err_print_gt_fences(m, error->gt); 885 886 /* 887 * GuC dumped global, eng-class and eng-instance registers together 888 * as part of engine state dump so we print in err_print_gt_engines 889 */ 890 if (!print_guc_capture) 891 err_print_gt_global(m, error->gt); 892 893 err_print_gt_engines(m, error->gt); 894 895 if (error->gt->uc) 896 err_print_uc(m, error->gt->uc); 897 898 err_print_gt_info(m, error->gt); 899 } 900 901 if (error->overlay) 902 intel_overlay_print_error_state(m, error->overlay); 903 904 err_print_capabilities(m, error); 905 err_print_params(m, &error->params); 906 } 907 908 static int err_print_to_sgl(struct i915_gpu_coredump *error) 909 { 910 struct drm_i915_error_state_buf m; 911 912 if (IS_ERR(error)) 913 return PTR_ERR(error); 914 915 if (READ_ONCE(error->sgl)) 916 return 0; 917 918 memset(&m, 0, sizeof(m)); 919 m.i915 = error->i915; 920 921 __err_print_to_sgl(&m, error); 922 923 if (m.buf) { 924 __sg_set_buf(m.cur++, m.buf, m.bytes, m.iter); 925 m.bytes = 0; 926 m.buf = NULL; 927 } 928 if (m.cur) { 929 GEM_BUG_ON(m.end < m.cur); 930 sg_mark_end(m.cur - 1); 931 } 932 GEM_BUG_ON(m.sgl && !m.cur); 933 934 if (m.err) { 935 err_free_sgl(m.sgl); 936 return m.err; 937 } 938 939 if (cmpxchg(&error->sgl, NULL, m.sgl)) 940 err_free_sgl(m.sgl); 941 942 return 0; 943 } 944 945 ssize_t i915_gpu_coredump_copy_to_buffer(struct i915_gpu_coredump *error, 946 char *buf, loff_t off, size_t rem) 947 { 948 struct scatterlist *sg; 949 size_t count; 950 loff_t pos; 951 int err; 952 953 if (!error || !rem) 954 return 0; 955 956 err = err_print_to_sgl(error); 957 if (err) 958 return err; 959 960 sg = READ_ONCE(error->fit); 961 if (!sg || off < sg->dma_address) 962 sg = error->sgl; 963 if (!sg) 964 return 0; 965 966 pos = sg->dma_address; 967 count = 0; 968 do { 969 size_t len, start; 970 971 if (sg_is_chain(sg)) { 972 sg = sg_chain_ptr(sg); 973 GEM_BUG_ON(sg_is_chain(sg)); 974 } 975 976 len = sg->length; 977 if (pos + len <= off) { 978 pos += len; 979 continue; 980 } 981 982 start = sg->offset; 983 if (pos < off) { 984 GEM_BUG_ON(off - pos > len); 985 len -= off - pos; 986 start += off - pos; 987 pos = off; 988 } 989 990 len = min(len, rem); 991 GEM_BUG_ON(!len || len > sg->length); 992 993 memcpy(buf, page_address(sg_page(sg)) + start, len); 994 995 count += len; 996 pos += len; 997 998 buf += len; 999 rem -= len; 1000 if (!rem) { 1001 WRITE_ONCE(error->fit, sg); 1002 break; 1003 } 1004 } while (!sg_is_last(sg++)); 1005 1006 return count; 1007 } 1008 1009 static void i915_vma_coredump_free(struct i915_vma_coredump *vma) 1010 { 1011 while (vma) { 1012 struct i915_vma_coredump *next = vma->next; 1013 struct page *page, *n; 1014 1015 list_for_each_entry_safe(page, n, &vma->page_list, lru) { 1016 list_del_init(&page->lru); 1017 __free_page(page); 1018 } 1019 1020 kfree(vma); 1021 vma = next; 1022 } 1023 } 1024 1025 static void cleanup_params(struct i915_gpu_coredump *error) 1026 { 1027 i915_params_free(&error->params); 1028 } 1029 1030 static void cleanup_uc(struct intel_uc_coredump *uc) 1031 { 1032 kfree(uc->guc_fw.file_selected.path); 1033 kfree(uc->huc_fw.file_selected.path); 1034 kfree(uc->guc_fw.file_wanted.path); 1035 kfree(uc->huc_fw.file_wanted.path); 1036 i915_vma_coredump_free(uc->guc.vma_log); 1037 i915_vma_coredump_free(uc->guc.vma_ctb); 1038 1039 kfree(uc); 1040 } 1041 1042 static void cleanup_gt(struct intel_gt_coredump *gt) 1043 { 1044 while (gt->engine) { 1045 struct intel_engine_coredump *ee = gt->engine; 1046 1047 gt->engine = ee->next; 1048 1049 i915_vma_coredump_free(ee->vma); 1050 intel_guc_capture_free_node(ee); 1051 kfree(ee); 1052 } 1053 1054 if (gt->uc) 1055 cleanup_uc(gt->uc); 1056 1057 kfree(gt); 1058 } 1059 1060 void __i915_gpu_coredump_free(struct kref *error_ref) 1061 { 1062 struct i915_gpu_coredump *error = 1063 container_of(error_ref, typeof(*error), ref); 1064 1065 while (error->gt) { 1066 struct intel_gt_coredump *gt = error->gt; 1067 1068 error->gt = gt->next; 1069 cleanup_gt(gt); 1070 } 1071 1072 kfree(error->overlay); 1073 1074 cleanup_params(error); 1075 1076 err_free_sgl(error->sgl); 1077 kfree(error); 1078 } 1079 1080 static struct i915_vma_coredump * 1081 i915_vma_coredump_create(const struct intel_gt *gt, 1082 const struct i915_vma_resource *vma_res, 1083 struct i915_vma_compress *compress, 1084 const char *name) 1085 1086 { 1087 struct i915_ggtt *ggtt = gt->ggtt; 1088 const u64 slot = ggtt->error_capture.start; 1089 struct i915_vma_coredump *dst; 1090 struct sgt_iter iter; 1091 int ret; 1092 1093 might_sleep(); 1094 1095 if (!vma_res || !vma_res->bi.pages || !compress) 1096 return NULL; 1097 1098 dst = kmalloc(sizeof(*dst), ALLOW_FAIL); 1099 if (!dst) 1100 return NULL; 1101 1102 if (!compress_start(compress)) { 1103 kfree(dst); 1104 return NULL; 1105 } 1106 1107 INIT_LIST_HEAD(&dst->page_list); 1108 strcpy(dst->name, name); 1109 dst->next = NULL; 1110 1111 dst->gtt_offset = vma_res->start; 1112 dst->gtt_size = vma_res->node_size; 1113 dst->gtt_page_sizes = vma_res->page_sizes_gtt; 1114 dst->unused = 0; 1115 1116 ret = -EINVAL; 1117 if (drm_mm_node_allocated(&ggtt->error_capture)) { 1118 void __iomem *s; 1119 dma_addr_t dma; 1120 1121 for_each_sgt_daddr(dma, iter, vma_res->bi.pages) { 1122 mutex_lock(&ggtt->error_mutex); 1123 if (ggtt->vm.raw_insert_page) 1124 ggtt->vm.raw_insert_page(&ggtt->vm, dma, slot, 1125 i915_gem_get_pat_index(gt->i915, 1126 I915_CACHE_NONE), 1127 0); 1128 else 1129 ggtt->vm.insert_page(&ggtt->vm, dma, slot, 1130 i915_gem_get_pat_index(gt->i915, 1131 I915_CACHE_NONE), 1132 0); 1133 mb(); 1134 1135 s = io_mapping_map_wc(&ggtt->iomap, slot, PAGE_SIZE); 1136 ret = compress_page(compress, 1137 (void __force *)s, dst, 1138 true); 1139 io_mapping_unmap(s); 1140 1141 mb(); 1142 ggtt->vm.clear_range(&ggtt->vm, slot, PAGE_SIZE); 1143 mutex_unlock(&ggtt->error_mutex); 1144 if (ret) 1145 break; 1146 } 1147 } else if (vma_res->bi.lmem) { 1148 struct intel_memory_region *mem = vma_res->mr; 1149 dma_addr_t dma; 1150 1151 for_each_sgt_daddr(dma, iter, vma_res->bi.pages) { 1152 dma_addr_t offset = dma - mem->region.start; 1153 void __iomem *s; 1154 1155 if (offset + PAGE_SIZE > mem->io_size) { 1156 ret = -EINVAL; 1157 break; 1158 } 1159 1160 s = io_mapping_map_wc(&mem->iomap, offset, PAGE_SIZE); 1161 ret = compress_page(compress, 1162 (void __force *)s, dst, 1163 true); 1164 io_mapping_unmap(s); 1165 if (ret) 1166 break; 1167 } 1168 } else { 1169 struct page *page; 1170 1171 for_each_sgt_page(page, iter, vma_res->bi.pages) { 1172 void *s; 1173 1174 drm_clflush_pages(&page, 1); 1175 1176 s = kmap(page); 1177 ret = compress_page(compress, s, dst, false); 1178 kunmap(page); 1179 1180 drm_clflush_pages(&page, 1); 1181 1182 if (ret) 1183 break; 1184 } 1185 } 1186 1187 if (ret || compress_flush(compress, dst)) { 1188 struct page *page, *n; 1189 1190 list_for_each_entry_safe_reverse(page, n, &dst->page_list, lru) { 1191 list_del_init(&page->lru); 1192 pool_free(&compress->pool, page_address(page)); 1193 } 1194 1195 kfree(dst); 1196 dst = NULL; 1197 } 1198 compress_finish(compress); 1199 1200 return dst; 1201 } 1202 1203 static void gt_record_fences(struct intel_gt_coredump *gt) 1204 { 1205 struct i915_ggtt *ggtt = gt->_gt->ggtt; 1206 struct intel_uncore *uncore = gt->_gt->uncore; 1207 int i; 1208 1209 if (GRAPHICS_VER(uncore->i915) >= 6) { 1210 for (i = 0; i < ggtt->num_fences; i++) 1211 gt->fence[i] = 1212 intel_uncore_read64(uncore, 1213 FENCE_REG_GEN6_LO(i)); 1214 } else if (GRAPHICS_VER(uncore->i915) >= 4) { 1215 for (i = 0; i < ggtt->num_fences; i++) 1216 gt->fence[i] = 1217 intel_uncore_read64(uncore, 1218 FENCE_REG_965_LO(i)); 1219 } else { 1220 for (i = 0; i < ggtt->num_fences; i++) 1221 gt->fence[i] = 1222 intel_uncore_read(uncore, FENCE_REG(i)); 1223 } 1224 gt->nfence = i; 1225 } 1226 1227 static void engine_record_registers(struct intel_engine_coredump *ee) 1228 { 1229 const struct intel_engine_cs *engine = ee->engine; 1230 struct drm_i915_private *i915 = engine->i915; 1231 1232 if (GRAPHICS_VER(i915) >= 6) { 1233 ee->rc_psmi = ENGINE_READ(engine, RING_PSMI_CTL); 1234 1235 if (GRAPHICS_VER_FULL(i915) >= IP_VER(12, 50)) 1236 ee->fault_reg = intel_gt_mcr_read_any(engine->gt, 1237 XEHP_RING_FAULT_REG); 1238 else if (GRAPHICS_VER(i915) >= 12) 1239 ee->fault_reg = intel_uncore_read(engine->uncore, 1240 GEN12_RING_FAULT_REG); 1241 else if (GRAPHICS_VER(i915) >= 8) 1242 ee->fault_reg = intel_uncore_read(engine->uncore, 1243 GEN8_RING_FAULT_REG); 1244 else 1245 ee->fault_reg = GEN6_RING_FAULT_REG_READ(engine); 1246 } 1247 1248 if (GRAPHICS_VER(i915) >= 4) { 1249 ee->esr = ENGINE_READ(engine, RING_ESR); 1250 ee->faddr = ENGINE_READ(engine, RING_DMA_FADD); 1251 ee->ipeir = ENGINE_READ(engine, RING_IPEIR); 1252 ee->ipehr = ENGINE_READ(engine, RING_IPEHR); 1253 ee->instps = ENGINE_READ(engine, RING_INSTPS); 1254 ee->bbaddr = ENGINE_READ(engine, RING_BBADDR); 1255 ee->ccid = ENGINE_READ(engine, CCID); 1256 if (GRAPHICS_VER(i915) >= 8) { 1257 ee->faddr |= (u64)ENGINE_READ(engine, RING_DMA_FADD_UDW) << 32; 1258 ee->bbaddr |= (u64)ENGINE_READ(engine, RING_BBADDR_UDW) << 32; 1259 } 1260 ee->bbstate = ENGINE_READ(engine, RING_BBSTATE); 1261 } else { 1262 ee->faddr = ENGINE_READ(engine, DMA_FADD_I8XX); 1263 ee->ipeir = ENGINE_READ(engine, IPEIR); 1264 ee->ipehr = ENGINE_READ(engine, IPEHR); 1265 } 1266 1267 if (GRAPHICS_VER(i915) >= 11) { 1268 ee->cmd_cctl = ENGINE_READ(engine, RING_CMD_CCTL); 1269 ee->cscmdop = ENGINE_READ(engine, RING_CSCMDOP); 1270 ee->ctx_sr_ctl = ENGINE_READ(engine, RING_CTX_SR_CTL); 1271 ee->dma_faddr_hi = ENGINE_READ(engine, RING_DMA_FADD_UDW); 1272 ee->dma_faddr_lo = ENGINE_READ(engine, RING_DMA_FADD); 1273 ee->nopid = ENGINE_READ(engine, RING_NOPID); 1274 ee->excc = ENGINE_READ(engine, RING_EXCC); 1275 } 1276 1277 intel_engine_get_instdone(engine, &ee->instdone); 1278 1279 ee->instpm = ENGINE_READ(engine, RING_INSTPM); 1280 ee->acthd = intel_engine_get_active_head(engine); 1281 ee->start = ENGINE_READ(engine, RING_START); 1282 ee->head = ENGINE_READ(engine, RING_HEAD); 1283 ee->tail = ENGINE_READ(engine, RING_TAIL); 1284 ee->ctl = ENGINE_READ(engine, RING_CTL); 1285 if (GRAPHICS_VER(i915) > 2) 1286 ee->mode = ENGINE_READ(engine, RING_MI_MODE); 1287 1288 if (!HWS_NEEDS_PHYSICAL(i915)) { 1289 i915_reg_t mmio; 1290 1291 if (GRAPHICS_VER(i915) == 7) { 1292 switch (engine->id) { 1293 default: 1294 MISSING_CASE(engine->id); 1295 fallthrough; 1296 case RCS0: 1297 mmio = RENDER_HWS_PGA_GEN7; 1298 break; 1299 case BCS0: 1300 mmio = BLT_HWS_PGA_GEN7; 1301 break; 1302 case VCS0: 1303 mmio = BSD_HWS_PGA_GEN7; 1304 break; 1305 case VECS0: 1306 mmio = VEBOX_HWS_PGA_GEN7; 1307 break; 1308 } 1309 } else if (GRAPHICS_VER(engine->i915) == 6) { 1310 mmio = RING_HWS_PGA_GEN6(engine->mmio_base); 1311 } else { 1312 /* XXX: gen8 returns to sanity */ 1313 mmio = RING_HWS_PGA(engine->mmio_base); 1314 } 1315 1316 ee->hws = intel_uncore_read(engine->uncore, mmio); 1317 } 1318 1319 ee->reset_count = i915_reset_engine_count(&i915->gpu_error, engine); 1320 1321 if (HAS_PPGTT(i915)) { 1322 int i; 1323 1324 ee->vm_info.gfx_mode = ENGINE_READ(engine, RING_MODE_GEN7); 1325 1326 if (GRAPHICS_VER(i915) == 6) { 1327 ee->vm_info.pp_dir_base = 1328 ENGINE_READ(engine, RING_PP_DIR_BASE_READ); 1329 } else if (GRAPHICS_VER(i915) == 7) { 1330 ee->vm_info.pp_dir_base = 1331 ENGINE_READ(engine, RING_PP_DIR_BASE); 1332 } else if (GRAPHICS_VER(i915) >= 8) { 1333 u32 base = engine->mmio_base; 1334 1335 for (i = 0; i < 4; i++) { 1336 ee->vm_info.pdp[i] = 1337 intel_uncore_read(engine->uncore, 1338 GEN8_RING_PDP_UDW(base, i)); 1339 ee->vm_info.pdp[i] <<= 32; 1340 ee->vm_info.pdp[i] |= 1341 intel_uncore_read(engine->uncore, 1342 GEN8_RING_PDP_LDW(base, i)); 1343 } 1344 } 1345 } 1346 } 1347 1348 static void record_request(const struct i915_request *request, 1349 struct i915_request_coredump *erq) 1350 { 1351 erq->flags = request->fence.flags; 1352 erq->context = request->fence.context; 1353 erq->seqno = request->fence.seqno; 1354 erq->sched_attr = request->sched.attr; 1355 erq->head = request->head; 1356 erq->tail = request->tail; 1357 1358 erq->pid = 0; 1359 rcu_read_lock(); 1360 if (!intel_context_is_closed(request->context)) { 1361 const struct i915_gem_context *ctx; 1362 1363 ctx = rcu_dereference(request->context->gem_context); 1364 if (ctx) 1365 erq->pid = pid_nr(ctx->pid); 1366 } 1367 rcu_read_unlock(); 1368 } 1369 1370 static void engine_record_execlists(struct intel_engine_coredump *ee) 1371 { 1372 const struct intel_engine_execlists * const el = &ee->engine->execlists; 1373 struct i915_request * const *port = el->active; 1374 unsigned int n = 0; 1375 1376 while (*port) 1377 record_request(*port++, &ee->execlist[n++]); 1378 1379 ee->num_ports = n; 1380 } 1381 1382 static bool record_context(struct i915_gem_context_coredump *e, 1383 struct intel_context *ce) 1384 { 1385 struct i915_gem_context *ctx; 1386 struct task_struct *task; 1387 bool simulated; 1388 1389 rcu_read_lock(); 1390 ctx = rcu_dereference(ce->gem_context); 1391 if (ctx && !kref_get_unless_zero(&ctx->ref)) 1392 ctx = NULL; 1393 rcu_read_unlock(); 1394 if (!ctx) 1395 return true; 1396 1397 rcu_read_lock(); 1398 task = pid_task(ctx->pid, PIDTYPE_PID); 1399 if (task) { 1400 strcpy(e->comm, task->comm); 1401 e->pid = task->pid; 1402 } 1403 rcu_read_unlock(); 1404 1405 e->sched_attr = ctx->sched; 1406 e->guilty = atomic_read(&ctx->guilty_count); 1407 e->active = atomic_read(&ctx->active_count); 1408 e->hwsp_seqno = (ce->timeline && ce->timeline->hwsp_seqno) ? 1409 *ce->timeline->hwsp_seqno : ~0U; 1410 1411 e->total_runtime = intel_context_get_total_runtime_ns(ce); 1412 e->avg_runtime = intel_context_get_avg_runtime_ns(ce); 1413 1414 simulated = i915_gem_context_no_error_capture(ctx); 1415 1416 i915_gem_context_put(ctx); 1417 return simulated; 1418 } 1419 1420 struct intel_engine_capture_vma { 1421 struct intel_engine_capture_vma *next; 1422 struct i915_vma_resource *vma_res; 1423 char name[16]; 1424 bool lockdep_cookie; 1425 }; 1426 1427 static struct intel_engine_capture_vma * 1428 capture_vma_snapshot(struct intel_engine_capture_vma *next, 1429 struct i915_vma_resource *vma_res, 1430 gfp_t gfp, const char *name) 1431 { 1432 struct intel_engine_capture_vma *c; 1433 1434 if (!vma_res) 1435 return next; 1436 1437 c = kmalloc(sizeof(*c), gfp); 1438 if (!c) 1439 return next; 1440 1441 if (!i915_vma_resource_hold(vma_res, &c->lockdep_cookie)) { 1442 kfree(c); 1443 return next; 1444 } 1445 1446 strcpy(c->name, name); 1447 c->vma_res = i915_vma_resource_get(vma_res); 1448 1449 c->next = next; 1450 return c; 1451 } 1452 1453 static struct intel_engine_capture_vma * 1454 capture_vma(struct intel_engine_capture_vma *next, 1455 struct i915_vma *vma, 1456 const char *name, 1457 gfp_t gfp) 1458 { 1459 if (!vma) 1460 return next; 1461 1462 /* 1463 * If the vma isn't pinned, then the vma should be snapshotted 1464 * to a struct i915_vma_snapshot at command submission time. 1465 * Not here. 1466 */ 1467 if (GEM_WARN_ON(!i915_vma_is_pinned(vma))) 1468 return next; 1469 1470 next = capture_vma_snapshot(next, vma->resource, gfp, name); 1471 1472 return next; 1473 } 1474 1475 static struct intel_engine_capture_vma * 1476 capture_user(struct intel_engine_capture_vma *capture, 1477 const struct i915_request *rq, 1478 gfp_t gfp) 1479 { 1480 struct i915_capture_list *c; 1481 1482 for (c = rq->capture_list; c; c = c->next) 1483 capture = capture_vma_snapshot(capture, c->vma_res, gfp, 1484 "user"); 1485 1486 return capture; 1487 } 1488 1489 static void add_vma(struct intel_engine_coredump *ee, 1490 struct i915_vma_coredump *vma) 1491 { 1492 if (vma) { 1493 vma->next = ee->vma; 1494 ee->vma = vma; 1495 } 1496 } 1497 1498 static struct i915_vma_coredump * 1499 create_vma_coredump(const struct intel_gt *gt, struct i915_vma *vma, 1500 const char *name, struct i915_vma_compress *compress) 1501 { 1502 struct i915_vma_coredump *ret = NULL; 1503 struct i915_vma_resource *vma_res; 1504 bool lockdep_cookie; 1505 1506 if (!vma) 1507 return NULL; 1508 1509 vma_res = vma->resource; 1510 1511 if (i915_vma_resource_hold(vma_res, &lockdep_cookie)) { 1512 ret = i915_vma_coredump_create(gt, vma_res, compress, name); 1513 i915_vma_resource_unhold(vma_res, lockdep_cookie); 1514 } 1515 1516 return ret; 1517 } 1518 1519 static void add_vma_coredump(struct intel_engine_coredump *ee, 1520 const struct intel_gt *gt, 1521 struct i915_vma *vma, 1522 const char *name, 1523 struct i915_vma_compress *compress) 1524 { 1525 add_vma(ee, create_vma_coredump(gt, vma, name, compress)); 1526 } 1527 1528 struct intel_engine_coredump * 1529 intel_engine_coredump_alloc(struct intel_engine_cs *engine, gfp_t gfp, u32 dump_flags) 1530 { 1531 struct intel_engine_coredump *ee; 1532 1533 ee = kzalloc(sizeof(*ee), gfp); 1534 if (!ee) 1535 return NULL; 1536 1537 ee->engine = engine; 1538 1539 if (!(dump_flags & CORE_DUMP_FLAG_IS_GUC_CAPTURE)) { 1540 engine_record_registers(ee); 1541 engine_record_execlists(ee); 1542 } 1543 1544 return ee; 1545 } 1546 1547 static struct intel_engine_capture_vma * 1548 engine_coredump_add_context(struct intel_engine_coredump *ee, 1549 struct intel_context *ce, 1550 gfp_t gfp) 1551 { 1552 struct intel_engine_capture_vma *vma = NULL; 1553 1554 ee->simulated |= record_context(&ee->context, ce); 1555 if (ee->simulated) 1556 return NULL; 1557 1558 /* 1559 * We need to copy these to an anonymous buffer 1560 * as the simplest method to avoid being overwritten 1561 * by userspace. 1562 */ 1563 vma = capture_vma(vma, ce->ring->vma, "ring", gfp); 1564 vma = capture_vma(vma, ce->state, "HW context", gfp); 1565 1566 return vma; 1567 } 1568 1569 struct intel_engine_capture_vma * 1570 intel_engine_coredump_add_request(struct intel_engine_coredump *ee, 1571 struct i915_request *rq, 1572 gfp_t gfp) 1573 { 1574 struct intel_engine_capture_vma *vma; 1575 1576 vma = engine_coredump_add_context(ee, rq->context, gfp); 1577 if (!vma) 1578 return NULL; 1579 1580 /* 1581 * We need to copy these to an anonymous buffer 1582 * as the simplest method to avoid being overwritten 1583 * by userspace. 1584 */ 1585 vma = capture_vma_snapshot(vma, rq->batch_res, gfp, "batch"); 1586 vma = capture_user(vma, rq, gfp); 1587 1588 ee->rq_head = rq->head; 1589 ee->rq_post = rq->postfix; 1590 ee->rq_tail = rq->tail; 1591 1592 return vma; 1593 } 1594 1595 void 1596 intel_engine_coredump_add_vma(struct intel_engine_coredump *ee, 1597 struct intel_engine_capture_vma *capture, 1598 struct i915_vma_compress *compress) 1599 { 1600 const struct intel_engine_cs *engine = ee->engine; 1601 1602 while (capture) { 1603 struct intel_engine_capture_vma *this = capture; 1604 struct i915_vma_resource *vma_res = this->vma_res; 1605 1606 add_vma(ee, 1607 i915_vma_coredump_create(engine->gt, vma_res, 1608 compress, this->name)); 1609 1610 i915_vma_resource_unhold(vma_res, this->lockdep_cookie); 1611 i915_vma_resource_put(vma_res); 1612 1613 capture = this->next; 1614 kfree(this); 1615 } 1616 1617 add_vma_coredump(ee, engine->gt, engine->status_page.vma, 1618 "HW Status", compress); 1619 1620 add_vma_coredump(ee, engine->gt, engine->wa_ctx.vma, 1621 "WA context", compress); 1622 } 1623 1624 static struct intel_engine_coredump * 1625 capture_engine(struct intel_engine_cs *engine, 1626 struct i915_vma_compress *compress, 1627 u32 dump_flags) 1628 { 1629 struct intel_engine_capture_vma *capture = NULL; 1630 struct intel_engine_coredump *ee; 1631 struct intel_context *ce = NULL; 1632 struct i915_request *rq = NULL; 1633 1634 ee = intel_engine_coredump_alloc(engine, ALLOW_FAIL, dump_flags); 1635 if (!ee) 1636 return NULL; 1637 1638 intel_engine_get_hung_entity(engine, &ce, &rq); 1639 if (rq && !i915_request_started(rq)) 1640 drm_info(&engine->gt->i915->drm, "Got hung context on %s with active request %lld:%lld [0x%04X] not yet started\n", 1641 engine->name, rq->fence.context, rq->fence.seqno, ce->guc_id.id); 1642 1643 if (rq) { 1644 capture = intel_engine_coredump_add_request(ee, rq, ATOMIC_MAYFAIL); 1645 i915_request_put(rq); 1646 } else if (ce) { 1647 capture = engine_coredump_add_context(ee, ce, ATOMIC_MAYFAIL); 1648 } 1649 1650 if (capture) { 1651 intel_engine_coredump_add_vma(ee, capture, compress); 1652 1653 if (dump_flags & CORE_DUMP_FLAG_IS_GUC_CAPTURE) 1654 intel_guc_capture_get_matching_node(engine->gt, ee, ce); 1655 } else { 1656 kfree(ee); 1657 ee = NULL; 1658 } 1659 1660 return ee; 1661 } 1662 1663 static void 1664 gt_record_engines(struct intel_gt_coredump *gt, 1665 intel_engine_mask_t engine_mask, 1666 struct i915_vma_compress *compress, 1667 u32 dump_flags) 1668 { 1669 struct intel_engine_cs *engine; 1670 enum intel_engine_id id; 1671 1672 for_each_engine(engine, gt->_gt, id) { 1673 struct intel_engine_coredump *ee; 1674 1675 /* Refill our page pool before entering atomic section */ 1676 pool_refill(&compress->pool, ALLOW_FAIL); 1677 1678 ee = capture_engine(engine, compress, dump_flags); 1679 if (!ee) 1680 continue; 1681 1682 ee->hung = engine->mask & engine_mask; 1683 1684 gt->simulated |= ee->simulated; 1685 if (ee->simulated) { 1686 if (dump_flags & CORE_DUMP_FLAG_IS_GUC_CAPTURE) 1687 intel_guc_capture_free_node(ee); 1688 kfree(ee); 1689 continue; 1690 } 1691 1692 ee->next = gt->engine; 1693 gt->engine = ee; 1694 } 1695 } 1696 1697 static void gt_record_guc_ctb(struct intel_ctb_coredump *saved, 1698 const struct intel_guc_ct_buffer *ctb, 1699 const void *blob_ptr, struct intel_guc *guc) 1700 { 1701 if (!ctb || !ctb->desc) 1702 return; 1703 1704 saved->raw_status = ctb->desc->status; 1705 saved->raw_head = ctb->desc->head; 1706 saved->raw_tail = ctb->desc->tail; 1707 saved->head = ctb->head; 1708 saved->tail = ctb->tail; 1709 saved->size = ctb->size; 1710 saved->desc_offset = ((void *)ctb->desc) - blob_ptr; 1711 saved->cmds_offset = ((void *)ctb->cmds) - blob_ptr; 1712 } 1713 1714 static struct intel_uc_coredump * 1715 gt_record_uc(struct intel_gt_coredump *gt, 1716 struct i915_vma_compress *compress) 1717 { 1718 const struct intel_uc *uc = >->_gt->uc; 1719 struct intel_uc_coredump *error_uc; 1720 1721 error_uc = kzalloc(sizeof(*error_uc), ALLOW_FAIL); 1722 if (!error_uc) 1723 return NULL; 1724 1725 memcpy(&error_uc->guc_fw, &uc->guc.fw, sizeof(uc->guc.fw)); 1726 memcpy(&error_uc->huc_fw, &uc->huc.fw, sizeof(uc->huc.fw)); 1727 1728 error_uc->guc_fw.file_selected.path = kstrdup(uc->guc.fw.file_selected.path, ALLOW_FAIL); 1729 error_uc->huc_fw.file_selected.path = kstrdup(uc->huc.fw.file_selected.path, ALLOW_FAIL); 1730 error_uc->guc_fw.file_wanted.path = kstrdup(uc->guc.fw.file_wanted.path, ALLOW_FAIL); 1731 error_uc->huc_fw.file_wanted.path = kstrdup(uc->huc.fw.file_wanted.path, ALLOW_FAIL); 1732 1733 /* 1734 * Save the GuC log and include a timestamp reference for converting the 1735 * log times to system times (in conjunction with the error->boottime and 1736 * gt->clock_frequency fields saved elsewhere). 1737 */ 1738 error_uc->guc.timestamp = intel_uncore_read(gt->_gt->uncore, GUCPMTIMESTAMP); 1739 error_uc->guc.vma_log = create_vma_coredump(gt->_gt, uc->guc.log.vma, 1740 "GuC log buffer", compress); 1741 error_uc->guc.vma_ctb = create_vma_coredump(gt->_gt, uc->guc.ct.vma, 1742 "GuC CT buffer", compress); 1743 error_uc->guc.last_fence = uc->guc.ct.requests.last_fence; 1744 gt_record_guc_ctb(error_uc->guc.ctb + 0, &uc->guc.ct.ctbs.send, 1745 uc->guc.ct.ctbs.send.desc, (struct intel_guc *)&uc->guc); 1746 gt_record_guc_ctb(error_uc->guc.ctb + 1, &uc->guc.ct.ctbs.recv, 1747 uc->guc.ct.ctbs.send.desc, (struct intel_guc *)&uc->guc); 1748 1749 return error_uc; 1750 } 1751 1752 /* Capture display registers. */ 1753 static void gt_record_display_regs(struct intel_gt_coredump *gt) 1754 { 1755 struct intel_uncore *uncore = gt->_gt->uncore; 1756 struct drm_i915_private *i915 = uncore->i915; 1757 1758 if (GRAPHICS_VER(i915) >= 6) 1759 gt->derrmr = intel_uncore_read(uncore, DERRMR); 1760 1761 if (GRAPHICS_VER(i915) >= 8) 1762 gt->ier = intel_uncore_read(uncore, GEN8_DE_MISC_IER); 1763 else if (IS_VALLEYVIEW(i915)) 1764 gt->ier = intel_uncore_read(uncore, VLV_IER); 1765 else if (HAS_PCH_SPLIT(i915)) 1766 gt->ier = intel_uncore_read(uncore, DEIER); 1767 else if (GRAPHICS_VER(i915) == 2) 1768 gt->ier = intel_uncore_read16(uncore, GEN2_IER); 1769 else 1770 gt->ier = intel_uncore_read(uncore, GEN2_IER); 1771 } 1772 1773 /* Capture all other registers that GuC doesn't capture. */ 1774 static void gt_record_global_nonguc_regs(struct intel_gt_coredump *gt) 1775 { 1776 struct intel_uncore *uncore = gt->_gt->uncore; 1777 struct drm_i915_private *i915 = uncore->i915; 1778 int i; 1779 1780 if (IS_VALLEYVIEW(i915)) { 1781 gt->gtier[0] = intel_uncore_read(uncore, GTIER); 1782 gt->ngtier = 1; 1783 } else if (GRAPHICS_VER(i915) >= 11) { 1784 gt->gtier[0] = 1785 intel_uncore_read(uncore, 1786 GEN11_RENDER_COPY_INTR_ENABLE); 1787 gt->gtier[1] = 1788 intel_uncore_read(uncore, GEN11_VCS_VECS_INTR_ENABLE); 1789 gt->gtier[2] = 1790 intel_uncore_read(uncore, GEN11_GUC_SG_INTR_ENABLE); 1791 gt->gtier[3] = 1792 intel_uncore_read(uncore, 1793 GEN11_GPM_WGBOXPERF_INTR_ENABLE); 1794 gt->gtier[4] = 1795 intel_uncore_read(uncore, 1796 GEN11_CRYPTO_RSVD_INTR_ENABLE); 1797 gt->gtier[5] = 1798 intel_uncore_read(uncore, 1799 GEN11_GUNIT_CSME_INTR_ENABLE); 1800 gt->ngtier = 6; 1801 } else if (GRAPHICS_VER(i915) >= 8) { 1802 for (i = 0; i < 4; i++) 1803 gt->gtier[i] = 1804 intel_uncore_read(uncore, GEN8_GT_IER(i)); 1805 gt->ngtier = 4; 1806 } else if (HAS_PCH_SPLIT(i915)) { 1807 gt->gtier[0] = intel_uncore_read(uncore, GTIER); 1808 gt->ngtier = 1; 1809 } 1810 1811 gt->eir = intel_uncore_read(uncore, EIR); 1812 gt->pgtbl_er = intel_uncore_read(uncore, PGTBL_ER); 1813 } 1814 1815 /* 1816 * Capture all registers that relate to workload submission. 1817 * NOTE: In GuC submission, when GuC resets an engine, it can dump these for us 1818 */ 1819 static void gt_record_global_regs(struct intel_gt_coredump *gt) 1820 { 1821 struct intel_uncore *uncore = gt->_gt->uncore; 1822 struct drm_i915_private *i915 = uncore->i915; 1823 int i; 1824 1825 /* 1826 * General organization 1827 * 1. Registers specific to a single generation 1828 * 2. Registers which belong to multiple generations 1829 * 3. Feature specific registers. 1830 * 4. Everything else 1831 * Please try to follow the order. 1832 */ 1833 1834 /* 1: Registers specific to a single generation */ 1835 if (IS_VALLEYVIEW(i915)) 1836 gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_VLV); 1837 1838 if (GRAPHICS_VER(i915) == 7) 1839 gt->err_int = intel_uncore_read(uncore, GEN7_ERR_INT); 1840 1841 if (GRAPHICS_VER_FULL(i915) >= IP_VER(12, 50)) { 1842 gt->fault_data0 = intel_gt_mcr_read_any((struct intel_gt *)gt->_gt, 1843 XEHP_FAULT_TLB_DATA0); 1844 gt->fault_data1 = intel_gt_mcr_read_any((struct intel_gt *)gt->_gt, 1845 XEHP_FAULT_TLB_DATA1); 1846 } else if (GRAPHICS_VER(i915) >= 12) { 1847 gt->fault_data0 = intel_uncore_read(uncore, 1848 GEN12_FAULT_TLB_DATA0); 1849 gt->fault_data1 = intel_uncore_read(uncore, 1850 GEN12_FAULT_TLB_DATA1); 1851 } else if (GRAPHICS_VER(i915) >= 8) { 1852 gt->fault_data0 = intel_uncore_read(uncore, 1853 GEN8_FAULT_TLB_DATA0); 1854 gt->fault_data1 = intel_uncore_read(uncore, 1855 GEN8_FAULT_TLB_DATA1); 1856 } 1857 1858 if (GRAPHICS_VER(i915) == 6) { 1859 gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE); 1860 gt->gab_ctl = intel_uncore_read(uncore, GAB_CTL); 1861 gt->gfx_mode = intel_uncore_read(uncore, GFX_MODE); 1862 } 1863 1864 /* 2: Registers which belong to multiple generations */ 1865 if (GRAPHICS_VER(i915) >= 7) 1866 gt->forcewake = intel_uncore_read_fw(uncore, FORCEWAKE_MT); 1867 1868 if (GRAPHICS_VER(i915) >= 6) { 1869 if (GRAPHICS_VER(i915) < 12) { 1870 gt->error = intel_uncore_read(uncore, ERROR_GEN6); 1871 gt->done_reg = intel_uncore_read(uncore, DONE_REG); 1872 } 1873 } 1874 1875 /* 3: Feature specific registers */ 1876 if (IS_GRAPHICS_VER(i915, 6, 7)) { 1877 gt->gam_ecochk = intel_uncore_read(uncore, GAM_ECOCHK); 1878 gt->gac_eco = intel_uncore_read(uncore, GAC_ECO_BITS); 1879 } 1880 1881 if (IS_GRAPHICS_VER(i915, 8, 11)) 1882 gt->gtt_cache = intel_uncore_read(uncore, HSW_GTT_CACHE_EN); 1883 1884 if (GRAPHICS_VER(i915) == 12) 1885 gt->aux_err = intel_uncore_read(uncore, GEN12_AUX_ERR_DBG); 1886 1887 if (GRAPHICS_VER(i915) >= 12) { 1888 for (i = 0; i < I915_MAX_SFC; i++) { 1889 /* 1890 * SFC_DONE resides in the VD forcewake domain, so it 1891 * only exists if the corresponding VCS engine is 1892 * present. 1893 */ 1894 if ((gt->_gt->info.sfc_mask & BIT(i)) == 0 || 1895 !HAS_ENGINE(gt->_gt, _VCS(i * 2))) 1896 continue; 1897 1898 gt->sfc_done[i] = 1899 intel_uncore_read(uncore, GEN12_SFC_DONE(i)); 1900 } 1901 1902 gt->gam_done = intel_uncore_read(uncore, GEN12_GAM_DONE); 1903 } 1904 } 1905 1906 static void gt_record_info(struct intel_gt_coredump *gt) 1907 { 1908 memcpy(>->info, >->_gt->info, sizeof(struct intel_gt_info)); 1909 gt->clock_frequency = gt->_gt->clock_frequency; 1910 gt->clock_period_ns = gt->_gt->clock_period_ns; 1911 } 1912 1913 /* 1914 * Generate a semi-unique error code. The code is not meant to have meaning, The 1915 * code's only purpose is to try to prevent false duplicated bug reports by 1916 * grossly estimating a GPU error state. 1917 * 1918 * TODO Ideally, hashing the batchbuffer would be a very nice way to determine 1919 * the hang if we could strip the GTT offset information from it. 1920 * 1921 * It's only a small step better than a random number in its current form. 1922 */ 1923 static u32 generate_ecode(const struct intel_engine_coredump *ee) 1924 { 1925 /* 1926 * IPEHR would be an ideal way to detect errors, as it's the gross 1927 * measure of "the command that hung." However, has some very common 1928 * synchronization commands which almost always appear in the case 1929 * strictly a client bug. Use instdone to differentiate those some. 1930 */ 1931 return ee ? ee->ipehr ^ ee->instdone.instdone : 0; 1932 } 1933 1934 static const char *error_msg(struct i915_gpu_coredump *error) 1935 { 1936 struct intel_engine_coredump *first = NULL; 1937 unsigned int hung_classes = 0; 1938 struct intel_gt_coredump *gt; 1939 int len; 1940 1941 for (gt = error->gt; gt; gt = gt->next) { 1942 struct intel_engine_coredump *cs; 1943 1944 for (cs = gt->engine; cs; cs = cs->next) { 1945 if (cs->hung) { 1946 hung_classes |= BIT(cs->engine->uabi_class); 1947 if (!first) 1948 first = cs; 1949 } 1950 } 1951 } 1952 1953 len = scnprintf(error->error_msg, sizeof(error->error_msg), 1954 "GPU HANG: ecode %d:%x:%08x", 1955 GRAPHICS_VER(error->i915), hung_classes, 1956 generate_ecode(first)); 1957 if (first && first->context.pid) { 1958 /* Just show the first executing process, more is confusing */ 1959 len += scnprintf(error->error_msg + len, 1960 sizeof(error->error_msg) - len, 1961 ", in %s [%d]", 1962 first->context.comm, first->context.pid); 1963 } 1964 1965 return error->error_msg; 1966 } 1967 1968 static void capture_gen(struct i915_gpu_coredump *error) 1969 { 1970 struct drm_i915_private *i915 = error->i915; 1971 1972 error->wakelock = atomic_read(&i915->runtime_pm.wakeref_count); 1973 error->suspended = i915->runtime_pm.suspended; 1974 1975 error->iommu = i915_vtd_active(i915); 1976 error->reset_count = i915_reset_count(&i915->gpu_error); 1977 error->suspend_count = i915->suspend_count; 1978 1979 i915_params_copy(&error->params, &i915->params); 1980 memcpy(&error->device_info, 1981 INTEL_INFO(i915), 1982 sizeof(error->device_info)); 1983 memcpy(&error->runtime_info, 1984 RUNTIME_INFO(i915), 1985 sizeof(error->runtime_info)); 1986 error->driver_caps = i915->caps; 1987 } 1988 1989 struct i915_gpu_coredump * 1990 i915_gpu_coredump_alloc(struct drm_i915_private *i915, gfp_t gfp) 1991 { 1992 struct i915_gpu_coredump *error; 1993 1994 if (!i915->params.error_capture) 1995 return NULL; 1996 1997 error = kzalloc(sizeof(*error), gfp); 1998 if (!error) 1999 return NULL; 2000 2001 kref_init(&error->ref); 2002 error->i915 = i915; 2003 2004 error->time = ktime_get_real(); 2005 error->boottime = ktime_get_boottime(); 2006 error->uptime = ktime_sub(ktime_get(), to_gt(i915)->last_init_time); 2007 error->capture = jiffies; 2008 2009 capture_gen(error); 2010 2011 return error; 2012 } 2013 2014 #define DAY_AS_SECONDS(x) (24 * 60 * 60 * (x)) 2015 2016 struct intel_gt_coredump * 2017 intel_gt_coredump_alloc(struct intel_gt *gt, gfp_t gfp, u32 dump_flags) 2018 { 2019 struct intel_gt_coredump *gc; 2020 2021 gc = kzalloc(sizeof(*gc), gfp); 2022 if (!gc) 2023 return NULL; 2024 2025 gc->_gt = gt; 2026 gc->awake = intel_gt_pm_is_awake(gt); 2027 2028 gt_record_display_regs(gc); 2029 gt_record_global_nonguc_regs(gc); 2030 2031 /* 2032 * GuC dumps global, eng-class and eng-instance registers 2033 * (that can change as part of engine state during execution) 2034 * before an engine is reset due to a hung context. 2035 * GuC captures and reports all three groups of registers 2036 * together as a single set before the engine is reset. 2037 * Thus, if GuC triggered the context reset we retrieve 2038 * the register values as part of gt_record_engines. 2039 */ 2040 if (!(dump_flags & CORE_DUMP_FLAG_IS_GUC_CAPTURE)) 2041 gt_record_global_regs(gc); 2042 2043 gt_record_fences(gc); 2044 2045 return gc; 2046 } 2047 2048 struct i915_vma_compress * 2049 i915_vma_capture_prepare(struct intel_gt_coredump *gt) 2050 { 2051 struct i915_vma_compress *compress; 2052 2053 compress = kmalloc(sizeof(*compress), ALLOW_FAIL); 2054 if (!compress) 2055 return NULL; 2056 2057 if (!compress_init(compress)) { 2058 kfree(compress); 2059 return NULL; 2060 } 2061 2062 return compress; 2063 } 2064 2065 void i915_vma_capture_finish(struct intel_gt_coredump *gt, 2066 struct i915_vma_compress *compress) 2067 { 2068 if (!compress) 2069 return; 2070 2071 compress_fini(compress); 2072 kfree(compress); 2073 } 2074 2075 static struct i915_gpu_coredump * 2076 __i915_gpu_coredump(struct intel_gt *gt, intel_engine_mask_t engine_mask, u32 dump_flags) 2077 { 2078 struct drm_i915_private *i915 = gt->i915; 2079 struct i915_gpu_coredump *error; 2080 2081 /* Check if GPU capture has been disabled */ 2082 error = READ_ONCE(i915->gpu_error.first_error); 2083 if (IS_ERR(error)) 2084 return error; 2085 2086 error = i915_gpu_coredump_alloc(i915, ALLOW_FAIL); 2087 if (!error) 2088 return ERR_PTR(-ENOMEM); 2089 2090 error->gt = intel_gt_coredump_alloc(gt, ALLOW_FAIL, dump_flags); 2091 if (error->gt) { 2092 struct i915_vma_compress *compress; 2093 2094 compress = i915_vma_capture_prepare(error->gt); 2095 if (!compress) { 2096 kfree(error->gt); 2097 kfree(error); 2098 return ERR_PTR(-ENOMEM); 2099 } 2100 2101 if (INTEL_INFO(i915)->has_gt_uc) { 2102 error->gt->uc = gt_record_uc(error->gt, compress); 2103 if (error->gt->uc) { 2104 if (dump_flags & CORE_DUMP_FLAG_IS_GUC_CAPTURE) 2105 error->gt->uc->guc.is_guc_capture = true; 2106 else 2107 GEM_BUG_ON(error->gt->uc->guc.is_guc_capture); 2108 } 2109 } 2110 2111 gt_record_info(error->gt); 2112 gt_record_engines(error->gt, engine_mask, compress, dump_flags); 2113 2114 2115 i915_vma_capture_finish(error->gt, compress); 2116 2117 error->simulated |= error->gt->simulated; 2118 } 2119 2120 error->overlay = intel_overlay_capture_error_state(i915); 2121 2122 return error; 2123 } 2124 2125 struct i915_gpu_coredump * 2126 i915_gpu_coredump(struct intel_gt *gt, intel_engine_mask_t engine_mask, u32 dump_flags) 2127 { 2128 static DEFINE_MUTEX(capture_mutex); 2129 int ret = mutex_lock_interruptible(&capture_mutex); 2130 struct i915_gpu_coredump *dump; 2131 2132 if (ret) 2133 return ERR_PTR(ret); 2134 2135 dump = __i915_gpu_coredump(gt, engine_mask, dump_flags); 2136 mutex_unlock(&capture_mutex); 2137 2138 return dump; 2139 } 2140 2141 void i915_error_state_store(struct i915_gpu_coredump *error) 2142 { 2143 struct drm_i915_private *i915; 2144 static bool warned; 2145 2146 if (IS_ERR_OR_NULL(error)) 2147 return; 2148 2149 i915 = error->i915; 2150 drm_info(&i915->drm, "%s\n", error_msg(error)); 2151 2152 if (error->simulated || 2153 cmpxchg(&i915->gpu_error.first_error, NULL, error)) 2154 return; 2155 2156 i915_gpu_coredump_get(error); 2157 2158 if (!xchg(&warned, true) && 2159 ktime_get_real_seconds() - DRIVER_TIMESTAMP < DAY_AS_SECONDS(180)) { 2160 pr_info("GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.\n"); 2161 pr_info("Please file a _new_ bug report at https://gitlab.freedesktop.org/drm/intel/issues/new.\n"); 2162 pr_info("Please see https://gitlab.freedesktop.org/drm/intel/-/wikis/How-to-file-i915-bugs for details.\n"); 2163 pr_info("drm/i915 developers can then reassign to the right component if it's not a kernel issue.\n"); 2164 pr_info("The GPU crash dump is required to analyze GPU hangs, so please always attach it.\n"); 2165 pr_info("GPU crash dump saved to /sys/class/drm/card%d/error\n", 2166 i915->drm.primary->index); 2167 } 2168 } 2169 2170 /** 2171 * i915_capture_error_state - capture an error record for later analysis 2172 * @gt: intel_gt which originated the hang 2173 * @engine_mask: hung engines 2174 * @dump_flags: dump flags 2175 * 2176 * Should be called when an error is detected (either a hang or an error 2177 * interrupt) to capture error state from the time of the error. Fills 2178 * out a structure which becomes available in debugfs for user level tools 2179 * to pick up. 2180 */ 2181 void i915_capture_error_state(struct intel_gt *gt, 2182 intel_engine_mask_t engine_mask, u32 dump_flags) 2183 { 2184 struct i915_gpu_coredump *error; 2185 2186 error = i915_gpu_coredump(gt, engine_mask, dump_flags); 2187 if (IS_ERR(error)) { 2188 cmpxchg(>->i915->gpu_error.first_error, NULL, error); 2189 return; 2190 } 2191 2192 i915_error_state_store(error); 2193 i915_gpu_coredump_put(error); 2194 } 2195 2196 struct i915_gpu_coredump * 2197 i915_first_error_state(struct drm_i915_private *i915) 2198 { 2199 struct i915_gpu_coredump *error; 2200 2201 spin_lock_irq(&i915->gpu_error.lock); 2202 error = i915->gpu_error.first_error; 2203 if (!IS_ERR_OR_NULL(error)) 2204 i915_gpu_coredump_get(error); 2205 spin_unlock_irq(&i915->gpu_error.lock); 2206 2207 return error; 2208 } 2209 2210 void i915_reset_error_state(struct drm_i915_private *i915) 2211 { 2212 struct i915_gpu_coredump *error; 2213 2214 spin_lock_irq(&i915->gpu_error.lock); 2215 error = i915->gpu_error.first_error; 2216 if (error != ERR_PTR(-ENODEV)) /* if disabled, always disabled */ 2217 i915->gpu_error.first_error = NULL; 2218 spin_unlock_irq(&i915->gpu_error.lock); 2219 2220 if (!IS_ERR_OR_NULL(error)) 2221 i915_gpu_coredump_put(error); 2222 } 2223 2224 void i915_disable_error_state(struct drm_i915_private *i915, int err) 2225 { 2226 spin_lock_irq(&i915->gpu_error.lock); 2227 if (!i915->gpu_error.first_error) 2228 i915->gpu_error.first_error = ERR_PTR(err); 2229 spin_unlock_irq(&i915->gpu_error.lock); 2230 } 2231 2232 #if IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM) 2233 void intel_klog_error_capture(struct intel_gt *gt, 2234 intel_engine_mask_t engine_mask) 2235 { 2236 static int g_count; 2237 struct drm_i915_private *i915 = gt->i915; 2238 struct i915_gpu_coredump *error; 2239 intel_wakeref_t wakeref; 2240 size_t buf_size = PAGE_SIZE * 128; 2241 size_t pos_err; 2242 char *buf, *ptr, *next; 2243 int l_count = g_count++; 2244 int line = 0; 2245 2246 /* Can't allocate memory during a reset */ 2247 if (test_bit(I915_RESET_BACKOFF, >->reset.flags)) { 2248 drm_err(>->i915->drm, "[Capture/%d.%d] Inside GT reset, skipping error capture :(\n", 2249 l_count, line++); 2250 return; 2251 } 2252 2253 error = READ_ONCE(i915->gpu_error.first_error); 2254 if (error) { 2255 drm_err(&i915->drm, "[Capture/%d.%d] Clearing existing error capture first...\n", 2256 l_count, line++); 2257 i915_reset_error_state(i915); 2258 } 2259 2260 with_intel_runtime_pm(&i915->runtime_pm, wakeref) 2261 error = i915_gpu_coredump(gt, engine_mask, CORE_DUMP_FLAG_NONE); 2262 2263 if (IS_ERR(error)) { 2264 drm_err(&i915->drm, "[Capture/%d.%d] Failed to capture error capture: %ld!\n", 2265 l_count, line++, PTR_ERR(error)); 2266 return; 2267 } 2268 2269 buf = kvmalloc(buf_size, GFP_KERNEL); 2270 if (!buf) { 2271 drm_err(&i915->drm, "[Capture/%d.%d] Failed to allocate buffer for error capture!\n", 2272 l_count, line++); 2273 i915_gpu_coredump_put(error); 2274 return; 2275 } 2276 2277 drm_info(&i915->drm, "[Capture/%d.%d] Dumping i915 error capture for %ps...\n", 2278 l_count, line++, __builtin_return_address(0)); 2279 2280 /* Largest string length safe to print via dmesg */ 2281 # define MAX_CHUNK 800 2282 2283 pos_err = 0; 2284 while (1) { 2285 ssize_t got = i915_gpu_coredump_copy_to_buffer(error, buf, pos_err, buf_size - 1); 2286 2287 if (got <= 0) 2288 break; 2289 2290 buf[got] = 0; 2291 pos_err += got; 2292 2293 ptr = buf; 2294 while (got > 0) { 2295 size_t count; 2296 char tag[2]; 2297 2298 next = strnchr(ptr, got, '\n'); 2299 if (next) { 2300 count = next - ptr; 2301 *next = 0; 2302 tag[0] = '>'; 2303 tag[1] = '<'; 2304 } else { 2305 count = got; 2306 tag[0] = '}'; 2307 tag[1] = '{'; 2308 } 2309 2310 if (count > MAX_CHUNK) { 2311 size_t pos; 2312 char *ptr2 = ptr; 2313 2314 for (pos = MAX_CHUNK; pos < count; pos += MAX_CHUNK) { 2315 char chr = ptr[pos]; 2316 2317 ptr[pos] = 0; 2318 drm_info(&i915->drm, "[Capture/%d.%d] }%s{\n", 2319 l_count, line++, ptr2); 2320 ptr[pos] = chr; 2321 ptr2 = ptr + pos; 2322 2323 /* 2324 * If spewing large amounts of data via a serial console, 2325 * this can be a very slow process. So be friendly and try 2326 * not to cause 'softlockup on CPU' problems. 2327 */ 2328 cond_resched(); 2329 } 2330 2331 if (ptr2 < (ptr + count)) 2332 drm_info(&i915->drm, "[Capture/%d.%d] %c%s%c\n", 2333 l_count, line++, tag[0], ptr2, tag[1]); 2334 else if (tag[0] == '>') 2335 drm_info(&i915->drm, "[Capture/%d.%d] ><\n", 2336 l_count, line++); 2337 } else { 2338 drm_info(&i915->drm, "[Capture/%d.%d] %c%s%c\n", 2339 l_count, line++, tag[0], ptr, tag[1]); 2340 } 2341 2342 ptr = next; 2343 got -= count; 2344 if (next) { 2345 ptr++; 2346 got--; 2347 } 2348 2349 /* As above. */ 2350 cond_resched(); 2351 } 2352 2353 if (got) 2354 drm_info(&i915->drm, "[Capture/%d.%d] Got %zd bytes remaining!\n", 2355 l_count, line++, got); 2356 } 2357 2358 kvfree(buf); 2359 2360 drm_info(&i915->drm, "[Capture/%d.%d] Dumped %zd bytes\n", l_count, line++, pos_err); 2361 } 2362 #endif 2363