1 // SPDX-License-Identifier: MIT 2 /* 3 * Copyright © 2021-2022 Intel Corporation 4 */ 5 6 #include <linux/types.h> 7 8 #include <drm/drm_print.h> 9 10 #include "gt/intel_engine_regs.h" 11 #include "gt/intel_gt.h" 12 #include "gt/intel_gt_mcr.h" 13 #include "gt/intel_gt_regs.h" 14 #include "gt/intel_lrc.h" 15 #include "guc_capture_fwif.h" 16 #include "intel_guc_capture.h" 17 #include "intel_guc_fwif.h" 18 #include "i915_drv.h" 19 #include "i915_gpu_error.h" 20 #include "i915_irq.h" 21 #include "i915_memcpy.h" 22 #include "i915_reg.h" 23 24 /* 25 * Define all device tables of GuC error capture register lists 26 * NOTE: For engine-registers, GuC only needs the register offsets 27 * from the engine-mmio-base 28 */ 29 #define COMMON_BASE_GLOBAL \ 30 { FORCEWAKE_MT, 0, 0, "FORCEWAKE" } 31 32 #define COMMON_GEN9BASE_GLOBAL \ 33 { GEN8_FAULT_TLB_DATA0, 0, 0, "GEN8_FAULT_TLB_DATA0" }, \ 34 { GEN8_FAULT_TLB_DATA1, 0, 0, "GEN8_FAULT_TLB_DATA1" }, \ 35 { ERROR_GEN6, 0, 0, "ERROR_GEN6" }, \ 36 { DONE_REG, 0, 0, "DONE_REG" }, \ 37 { HSW_GTT_CACHE_EN, 0, 0, "HSW_GTT_CACHE_EN" } 38 39 #define COMMON_GEN12BASE_GLOBAL \ 40 { GEN12_FAULT_TLB_DATA0, 0, 0, "GEN12_FAULT_TLB_DATA0" }, \ 41 { GEN12_FAULT_TLB_DATA1, 0, 0, "GEN12_FAULT_TLB_DATA1" }, \ 42 { GEN12_AUX_ERR_DBG, 0, 0, "AUX_ERR_DBG" }, \ 43 { GEN12_GAM_DONE, 0, 0, "GAM_DONE" }, \ 44 { GEN12_RING_FAULT_REG, 0, 0, "FAULT_REG" } 45 46 #define COMMON_BASE_ENGINE_INSTANCE \ 47 { RING_PSMI_CTL(0), 0, 0, "RC PSMI" }, \ 48 { RING_ESR(0), 0, 0, "ESR" }, \ 49 { RING_DMA_FADD(0), 0, 0, "RING_DMA_FADD_LDW" }, \ 50 { RING_DMA_FADD_UDW(0), 0, 0, "RING_DMA_FADD_UDW" }, \ 51 { RING_IPEIR(0), 0, 0, "IPEIR" }, \ 52 { RING_IPEHR(0), 0, 0, "IPEHR" }, \ 53 { RING_INSTPS(0), 0, 0, "INSTPS" }, \ 54 { RING_BBADDR(0), 0, 0, "RING_BBADDR_LOW32" }, \ 55 { RING_BBADDR_UDW(0), 0, 0, "RING_BBADDR_UP32" }, \ 56 { RING_BBSTATE(0), 0, 0, "BB_STATE" }, \ 57 { CCID(0), 0, 0, "CCID" }, \ 58 { RING_ACTHD(0), 0, 0, "ACTHD_LDW" }, \ 59 { RING_ACTHD_UDW(0), 0, 0, "ACTHD_UDW" }, \ 60 { RING_INSTPM(0), 0, 0, "INSTPM" }, \ 61 { RING_INSTDONE(0), 0, 0, "INSTDONE" }, \ 62 { RING_NOPID(0), 0, 0, "RING_NOPID" }, \ 63 { RING_START(0), 0, 0, "START" }, \ 64 { RING_HEAD(0), 0, 0, "HEAD" }, \ 65 { RING_TAIL(0), 0, 0, "TAIL" }, \ 66 { RING_CTL(0), 0, 0, "CTL" }, \ 67 { RING_MI_MODE(0), 0, 0, "MODE" }, \ 68 { RING_CONTEXT_CONTROL(0), 0, 0, "RING_CONTEXT_CONTROL" }, \ 69 { RING_HWS_PGA(0), 0, 0, "HWS" }, \ 70 { RING_MODE_GEN7(0), 0, 0, "GFX_MODE" }, \ 71 { GEN8_RING_PDP_LDW(0, 0), 0, 0, "PDP0_LDW" }, \ 72 { GEN8_RING_PDP_UDW(0, 0), 0, 0, "PDP0_UDW" }, \ 73 { GEN8_RING_PDP_LDW(0, 1), 0, 0, "PDP1_LDW" }, \ 74 { GEN8_RING_PDP_UDW(0, 1), 0, 0, "PDP1_UDW" }, \ 75 { GEN8_RING_PDP_LDW(0, 2), 0, 0, "PDP2_LDW" }, \ 76 { GEN8_RING_PDP_UDW(0, 2), 0, 0, "PDP2_UDW" }, \ 77 { GEN8_RING_PDP_LDW(0, 3), 0, 0, "PDP3_LDW" }, \ 78 { GEN8_RING_PDP_UDW(0, 3), 0, 0, "PDP3_UDW" } 79 80 #define COMMON_BASE_HAS_EU \ 81 { EIR, 0, 0, "EIR" } 82 83 #define COMMON_BASE_RENDER \ 84 { GEN7_SC_INSTDONE, 0, 0, "GEN7_SC_INSTDONE" } 85 86 #define COMMON_GEN12BASE_RENDER \ 87 { GEN12_SC_INSTDONE_EXTRA, 0, 0, "GEN12_SC_INSTDONE_EXTRA" }, \ 88 { GEN12_SC_INSTDONE_EXTRA2, 0, 0, "GEN12_SC_INSTDONE_EXTRA2" } 89 90 #define COMMON_GEN12BASE_VEC \ 91 { GEN12_SFC_DONE(0), 0, 0, "SFC_DONE[0]" }, \ 92 { GEN12_SFC_DONE(1), 0, 0, "SFC_DONE[1]" }, \ 93 { GEN12_SFC_DONE(2), 0, 0, "SFC_DONE[2]" }, \ 94 { GEN12_SFC_DONE(3), 0, 0, "SFC_DONE[3]" } 95 96 /* XE_LPD - Global */ 97 static const struct __guc_mmio_reg_descr xe_lpd_global_regs[] = { 98 COMMON_BASE_GLOBAL, 99 COMMON_GEN9BASE_GLOBAL, 100 COMMON_GEN12BASE_GLOBAL, 101 }; 102 103 /* XE_LPD - Render / Compute Per-Class */ 104 static const struct __guc_mmio_reg_descr xe_lpd_rc_class_regs[] = { 105 COMMON_BASE_HAS_EU, 106 COMMON_BASE_RENDER, 107 COMMON_GEN12BASE_RENDER, 108 }; 109 110 /* GEN9/XE_LPD - Render / Compute Per-Engine-Instance */ 111 static const struct __guc_mmio_reg_descr xe_lpd_rc_inst_regs[] = { 112 COMMON_BASE_ENGINE_INSTANCE, 113 }; 114 115 /* GEN9/XE_LPD - Media Decode/Encode Per-Engine-Instance */ 116 static const struct __guc_mmio_reg_descr xe_lpd_vd_inst_regs[] = { 117 COMMON_BASE_ENGINE_INSTANCE, 118 }; 119 120 /* XE_LPD - Video Enhancement Per-Class */ 121 static const struct __guc_mmio_reg_descr xe_lpd_vec_class_regs[] = { 122 COMMON_GEN12BASE_VEC, 123 }; 124 125 /* GEN9/XE_LPD - Video Enhancement Per-Engine-Instance */ 126 static const struct __guc_mmio_reg_descr xe_lpd_vec_inst_regs[] = { 127 COMMON_BASE_ENGINE_INSTANCE, 128 }; 129 130 /* GEN9/XE_LPD - Blitter Per-Engine-Instance */ 131 static const struct __guc_mmio_reg_descr xe_lpd_blt_inst_regs[] = { 132 COMMON_BASE_ENGINE_INSTANCE, 133 }; 134 135 /* XE_LPD - GSC Per-Engine-Instance */ 136 static const struct __guc_mmio_reg_descr xe_lpd_gsc_inst_regs[] = { 137 COMMON_BASE_ENGINE_INSTANCE, 138 }; 139 140 /* GEN9 - Global */ 141 static const struct __guc_mmio_reg_descr default_global_regs[] = { 142 COMMON_BASE_GLOBAL, 143 COMMON_GEN9BASE_GLOBAL, 144 }; 145 146 static const struct __guc_mmio_reg_descr default_rc_class_regs[] = { 147 COMMON_BASE_HAS_EU, 148 COMMON_BASE_RENDER, 149 }; 150 151 /* 152 * Empty lists: 153 * GEN9/XE_LPD - Blitter Per-Class 154 * GEN9/XE_LPD - Media Decode/Encode Per-Class 155 * GEN9 - VEC Class 156 */ 157 static const struct __guc_mmio_reg_descr empty_regs_list[] = { 158 }; 159 160 #define TO_GCAP_DEF_OWNER(x) (GUC_CAPTURE_LIST_INDEX_##x) 161 #define TO_GCAP_DEF_TYPE(x) (GUC_CAPTURE_LIST_TYPE_##x) 162 #define MAKE_REGLIST(regslist, regsowner, regstype, class) \ 163 { \ 164 regslist, \ 165 ARRAY_SIZE(regslist), \ 166 TO_GCAP_DEF_OWNER(regsowner), \ 167 TO_GCAP_DEF_TYPE(regstype), \ 168 class, \ 169 NULL, \ 170 } 171 172 /* List of lists */ 173 static const struct __guc_mmio_reg_descr_group default_lists[] = { 174 MAKE_REGLIST(default_global_regs, PF, GLOBAL, 0), 175 MAKE_REGLIST(default_rc_class_regs, PF, ENGINE_CLASS, GUC_RENDER_CLASS), 176 MAKE_REGLIST(xe_lpd_rc_inst_regs, PF, ENGINE_INSTANCE, GUC_RENDER_CLASS), 177 MAKE_REGLIST(default_rc_class_regs, PF, ENGINE_CLASS, GUC_COMPUTE_CLASS), 178 MAKE_REGLIST(xe_lpd_rc_inst_regs, PF, ENGINE_INSTANCE, GUC_COMPUTE_CLASS), 179 MAKE_REGLIST(empty_regs_list, PF, ENGINE_CLASS, GUC_VIDEO_CLASS), 180 MAKE_REGLIST(xe_lpd_vd_inst_regs, PF, ENGINE_INSTANCE, GUC_VIDEO_CLASS), 181 MAKE_REGLIST(empty_regs_list, PF, ENGINE_CLASS, GUC_VIDEOENHANCE_CLASS), 182 MAKE_REGLIST(xe_lpd_vec_inst_regs, PF, ENGINE_INSTANCE, GUC_VIDEOENHANCE_CLASS), 183 MAKE_REGLIST(empty_regs_list, PF, ENGINE_CLASS, GUC_BLITTER_CLASS), 184 MAKE_REGLIST(xe_lpd_blt_inst_regs, PF, ENGINE_INSTANCE, GUC_BLITTER_CLASS), 185 MAKE_REGLIST(empty_regs_list, PF, ENGINE_CLASS, GUC_GSC_OTHER_CLASS), 186 MAKE_REGLIST(xe_lpd_gsc_inst_regs, PF, ENGINE_INSTANCE, GUC_GSC_OTHER_CLASS), 187 {} 188 }; 189 190 static const struct __guc_mmio_reg_descr_group xe_lpd_lists[] = { 191 MAKE_REGLIST(xe_lpd_global_regs, PF, GLOBAL, 0), 192 MAKE_REGLIST(xe_lpd_rc_class_regs, PF, ENGINE_CLASS, GUC_RENDER_CLASS), 193 MAKE_REGLIST(xe_lpd_rc_inst_regs, PF, ENGINE_INSTANCE, GUC_RENDER_CLASS), 194 MAKE_REGLIST(xe_lpd_rc_class_regs, PF, ENGINE_CLASS, GUC_COMPUTE_CLASS), 195 MAKE_REGLIST(xe_lpd_rc_inst_regs, PF, ENGINE_INSTANCE, GUC_COMPUTE_CLASS), 196 MAKE_REGLIST(empty_regs_list, PF, ENGINE_CLASS, GUC_VIDEO_CLASS), 197 MAKE_REGLIST(xe_lpd_vd_inst_regs, PF, ENGINE_INSTANCE, GUC_VIDEO_CLASS), 198 MAKE_REGLIST(xe_lpd_vec_class_regs, PF, ENGINE_CLASS, GUC_VIDEOENHANCE_CLASS), 199 MAKE_REGLIST(xe_lpd_vec_inst_regs, PF, ENGINE_INSTANCE, GUC_VIDEOENHANCE_CLASS), 200 MAKE_REGLIST(empty_regs_list, PF, ENGINE_CLASS, GUC_BLITTER_CLASS), 201 MAKE_REGLIST(xe_lpd_blt_inst_regs, PF, ENGINE_INSTANCE, GUC_BLITTER_CLASS), 202 MAKE_REGLIST(empty_regs_list, PF, ENGINE_CLASS, GUC_GSC_OTHER_CLASS), 203 MAKE_REGLIST(xe_lpd_gsc_inst_regs, PF, ENGINE_INSTANCE, GUC_GSC_OTHER_CLASS), 204 {} 205 }; 206 207 static const struct __guc_mmio_reg_descr_group * 208 guc_capture_get_one_list(const struct __guc_mmio_reg_descr_group *reglists, 209 u32 owner, u32 type, u32 id) 210 { 211 int i; 212 213 if (!reglists) 214 return NULL; 215 216 for (i = 0; reglists[i].list; ++i) { 217 if (reglists[i].owner == owner && reglists[i].type == type && 218 (reglists[i].engine == id || reglists[i].type == GUC_CAPTURE_LIST_TYPE_GLOBAL)) 219 return ®lists[i]; 220 } 221 222 return NULL; 223 } 224 225 static struct __guc_mmio_reg_descr_group * 226 guc_capture_get_one_ext_list(struct __guc_mmio_reg_descr_group *reglists, 227 u32 owner, u32 type, u32 id) 228 { 229 int i; 230 231 if (!reglists) 232 return NULL; 233 234 for (i = 0; reglists[i].extlist; ++i) { 235 if (reglists[i].owner == owner && reglists[i].type == type && 236 (reglists[i].engine == id || reglists[i].type == GUC_CAPTURE_LIST_TYPE_GLOBAL)) 237 return ®lists[i]; 238 } 239 240 return NULL; 241 } 242 243 static void guc_capture_free_extlists(struct __guc_mmio_reg_descr_group *reglists) 244 { 245 int i = 0; 246 247 if (!reglists) 248 return; 249 250 while (reglists[i].extlist) 251 kfree(reglists[i++].extlist); 252 } 253 254 struct __ext_steer_reg { 255 const char *name; 256 i915_mcr_reg_t reg; 257 }; 258 259 static const struct __ext_steer_reg xe_extregs[] = { 260 {"GEN8_SAMPLER_INSTDONE", GEN8_SAMPLER_INSTDONE}, 261 {"GEN8_ROW_INSTDONE", GEN8_ROW_INSTDONE} 262 }; 263 264 static void __fill_ext_reg(struct __guc_mmio_reg_descr *ext, 265 const struct __ext_steer_reg *extlist, 266 int slice_id, int subslice_id) 267 { 268 ext->reg = _MMIO(i915_mmio_reg_offset(extlist->reg)); 269 ext->flags = FIELD_PREP(GUC_REGSET_STEERING_GROUP, slice_id); 270 ext->flags |= FIELD_PREP(GUC_REGSET_STEERING_INSTANCE, subslice_id); 271 ext->regname = extlist->name; 272 } 273 274 static int 275 __alloc_ext_regs(struct __guc_mmio_reg_descr_group *newlist, 276 const struct __guc_mmio_reg_descr_group *rootlist, int num_regs) 277 { 278 struct __guc_mmio_reg_descr *list; 279 280 list = kcalloc(num_regs, sizeof(struct __guc_mmio_reg_descr), GFP_KERNEL); 281 if (!list) 282 return -ENOMEM; 283 284 newlist->extlist = list; 285 newlist->num_regs = num_regs; 286 newlist->owner = rootlist->owner; 287 newlist->engine = rootlist->engine; 288 newlist->type = rootlist->type; 289 290 return 0; 291 } 292 293 static void 294 guc_capture_alloc_steered_lists_xe_lpd(struct intel_guc *guc, 295 const struct __guc_mmio_reg_descr_group *lists) 296 { 297 struct intel_gt *gt = guc_to_gt(guc); 298 int slice, subslice, iter, i, num_steer_regs, num_tot_regs = 0; 299 const struct __guc_mmio_reg_descr_group *list; 300 struct __guc_mmio_reg_descr_group *extlists; 301 struct __guc_mmio_reg_descr *extarray; 302 struct sseu_dev_info *sseu; 303 304 /* In XE_LPD we only have steered registers for the render-class */ 305 list = guc_capture_get_one_list(lists, GUC_CAPTURE_LIST_INDEX_PF, 306 GUC_CAPTURE_LIST_TYPE_ENGINE_CLASS, GUC_RENDER_CLASS); 307 /* skip if extlists was previously allocated */ 308 if (!list || guc->capture->extlists) 309 return; 310 311 num_steer_regs = ARRAY_SIZE(xe_extregs); 312 313 sseu = >->info.sseu; 314 for_each_ss_steering(iter, gt, slice, subslice) 315 num_tot_regs += num_steer_regs; 316 317 if (!num_tot_regs) 318 return; 319 320 /* allocate an extra for an end marker */ 321 extlists = kcalloc(2, sizeof(struct __guc_mmio_reg_descr_group), GFP_KERNEL); 322 if (!extlists) 323 return; 324 325 if (__alloc_ext_regs(&extlists[0], list, num_tot_regs)) { 326 kfree(extlists); 327 return; 328 } 329 330 extarray = extlists[0].extlist; 331 for_each_ss_steering(iter, gt, slice, subslice) { 332 for (i = 0; i < num_steer_regs; ++i) { 333 __fill_ext_reg(extarray, &xe_extregs[i], slice, subslice); 334 ++extarray; 335 } 336 } 337 338 guc->capture->extlists = extlists; 339 } 340 341 static const struct __ext_steer_reg xehpg_extregs[] = { 342 {"XEHPG_INSTDONE_GEOM_SVG", XEHPG_INSTDONE_GEOM_SVG} 343 }; 344 345 static bool __has_xehpg_extregs(u32 ipver) 346 { 347 return (ipver >= IP_VER(12, 55)); 348 } 349 350 static void 351 guc_capture_alloc_steered_lists_xe_hpg(struct intel_guc *guc, 352 const struct __guc_mmio_reg_descr_group *lists, 353 u32 ipver) 354 { 355 struct intel_gt *gt = guc_to_gt(guc); 356 struct drm_i915_private *i915 = guc_to_gt(guc)->i915; 357 struct sseu_dev_info *sseu; 358 int slice, subslice, i, iter, num_steer_regs, num_tot_regs = 0; 359 const struct __guc_mmio_reg_descr_group *list; 360 struct __guc_mmio_reg_descr_group *extlists; 361 struct __guc_mmio_reg_descr *extarray; 362 363 /* In XE_LP / HPG we only have render-class steering registers during error-capture */ 364 list = guc_capture_get_one_list(lists, GUC_CAPTURE_LIST_INDEX_PF, 365 GUC_CAPTURE_LIST_TYPE_ENGINE_CLASS, GUC_RENDER_CLASS); 366 /* skip if extlists was previously allocated */ 367 if (!list || guc->capture->extlists) 368 return; 369 370 num_steer_regs = ARRAY_SIZE(xe_extregs); 371 if (__has_xehpg_extregs(ipver)) 372 num_steer_regs += ARRAY_SIZE(xehpg_extregs); 373 374 sseu = >->info.sseu; 375 for_each_ss_steering(iter, gt, slice, subslice) 376 num_tot_regs += num_steer_regs; 377 378 if (!num_tot_regs) 379 return; 380 381 /* allocate an extra for an end marker */ 382 extlists = kcalloc(2, sizeof(struct __guc_mmio_reg_descr_group), GFP_KERNEL); 383 if (!extlists) 384 return; 385 386 if (__alloc_ext_regs(&extlists[0], list, num_tot_regs)) { 387 kfree(extlists); 388 return; 389 } 390 391 extarray = extlists[0].extlist; 392 for_each_ss_steering(iter, gt, slice, subslice) { 393 for (i = 0; i < ARRAY_SIZE(xe_extregs); ++i) { 394 __fill_ext_reg(extarray, &xe_extregs[i], slice, subslice); 395 ++extarray; 396 } 397 if (__has_xehpg_extregs(ipver)) { 398 for (i = 0; i < ARRAY_SIZE(xehpg_extregs); ++i) { 399 __fill_ext_reg(extarray, &xehpg_extregs[i], slice, subslice); 400 ++extarray; 401 } 402 } 403 } 404 405 drm_dbg(&i915->drm, "GuC-capture found %d-ext-regs.\n", num_tot_regs); 406 guc->capture->extlists = extlists; 407 } 408 409 static const struct __guc_mmio_reg_descr_group * 410 guc_capture_get_device_reglist(struct intel_guc *guc) 411 { 412 struct drm_i915_private *i915 = guc_to_gt(guc)->i915; 413 414 if (GRAPHICS_VER(i915) > 11) { 415 /* 416 * For certain engine classes, there are slice and subslice 417 * level registers requiring steering. We allocate and populate 418 * these at init time based on hw config add it as an extension 419 * list at the end of the pre-populated render list. 420 */ 421 if (IS_DG2(i915)) 422 guc_capture_alloc_steered_lists_xe_hpg(guc, xe_lpd_lists, IP_VER(12, 55)); 423 else if (IS_XEHPSDV(i915)) 424 guc_capture_alloc_steered_lists_xe_hpg(guc, xe_lpd_lists, IP_VER(12, 50)); 425 else 426 guc_capture_alloc_steered_lists_xe_lpd(guc, xe_lpd_lists); 427 428 return xe_lpd_lists; 429 } 430 431 /* if GuC submission is enabled on a non-POR platform, just use a common baseline */ 432 return default_lists; 433 } 434 435 static const char * 436 __stringify_type(u32 type) 437 { 438 switch (type) { 439 case GUC_CAPTURE_LIST_TYPE_GLOBAL: 440 return "Global"; 441 case GUC_CAPTURE_LIST_TYPE_ENGINE_CLASS: 442 return "Class"; 443 case GUC_CAPTURE_LIST_TYPE_ENGINE_INSTANCE: 444 return "Instance"; 445 default: 446 break; 447 } 448 449 return "unknown"; 450 } 451 452 static const char * 453 __stringify_engclass(u32 class) 454 { 455 switch (class) { 456 case GUC_RENDER_CLASS: 457 return "Render"; 458 case GUC_VIDEO_CLASS: 459 return "Video"; 460 case GUC_VIDEOENHANCE_CLASS: 461 return "VideoEnhance"; 462 case GUC_BLITTER_CLASS: 463 return "Blitter"; 464 case GUC_COMPUTE_CLASS: 465 return "Compute"; 466 case GUC_GSC_OTHER_CLASS: 467 return "GSC-Other"; 468 default: 469 break; 470 } 471 472 return "unknown"; 473 } 474 475 static int 476 guc_capture_list_init(struct intel_guc *guc, u32 owner, u32 type, u32 classid, 477 struct guc_mmio_reg *ptr, u16 num_entries) 478 { 479 u32 i = 0, j = 0; 480 struct drm_i915_private *i915 = guc_to_gt(guc)->i915; 481 const struct __guc_mmio_reg_descr_group *reglists = guc->capture->reglists; 482 struct __guc_mmio_reg_descr_group *extlists = guc->capture->extlists; 483 const struct __guc_mmio_reg_descr_group *match; 484 struct __guc_mmio_reg_descr_group *matchext; 485 486 if (!reglists) 487 return -ENODEV; 488 489 match = guc_capture_get_one_list(reglists, owner, type, classid); 490 if (!match) 491 return -ENODATA; 492 493 for (i = 0; i < num_entries && i < match->num_regs; ++i) { 494 ptr[i].offset = match->list[i].reg.reg; 495 ptr[i].value = 0xDEADF00D; 496 ptr[i].flags = match->list[i].flags; 497 ptr[i].mask = match->list[i].mask; 498 } 499 500 matchext = guc_capture_get_one_ext_list(extlists, owner, type, classid); 501 if (matchext) { 502 for (i = match->num_regs, j = 0; i < num_entries && 503 i < (match->num_regs + matchext->num_regs) && 504 j < matchext->num_regs; ++i, ++j) { 505 ptr[i].offset = matchext->extlist[j].reg.reg; 506 ptr[i].value = 0xDEADF00D; 507 ptr[i].flags = matchext->extlist[j].flags; 508 ptr[i].mask = matchext->extlist[j].mask; 509 } 510 } 511 if (i < num_entries) 512 drm_dbg(&i915->drm, "GuC-capture: Init reglist short %d out %d.\n", 513 (int)i, (int)num_entries); 514 515 return 0; 516 } 517 518 static int 519 guc_cap_list_num_regs(struct intel_guc_state_capture *gc, u32 owner, u32 type, u32 classid) 520 { 521 const struct __guc_mmio_reg_descr_group *match; 522 struct __guc_mmio_reg_descr_group *matchext; 523 int num_regs; 524 525 match = guc_capture_get_one_list(gc->reglists, owner, type, classid); 526 if (!match) 527 return 0; 528 529 num_regs = match->num_regs; 530 531 matchext = guc_capture_get_one_ext_list(gc->extlists, owner, type, classid); 532 if (matchext) 533 num_regs += matchext->num_regs; 534 535 return num_regs; 536 } 537 538 static int 539 guc_capture_getlistsize(struct intel_guc *guc, u32 owner, u32 type, u32 classid, 540 size_t *size, bool is_purpose_est) 541 { 542 struct intel_guc_state_capture *gc = guc->capture; 543 struct drm_i915_private *i915 = guc_to_gt(guc)->i915; 544 struct __guc_capture_ads_cache *cache = &gc->ads_cache[owner][type][classid]; 545 int num_regs; 546 547 if (!gc->reglists) { 548 drm_warn(&i915->drm, "GuC-capture: No reglist on this device\n"); 549 return -ENODEV; 550 } 551 552 if (cache->is_valid) { 553 *size = cache->size; 554 return cache->status; 555 } 556 557 if (!is_purpose_est && owner == GUC_CAPTURE_LIST_INDEX_PF && 558 !guc_capture_get_one_list(gc->reglists, owner, type, classid)) { 559 if (type == GUC_CAPTURE_LIST_TYPE_GLOBAL) 560 drm_warn(&i915->drm, "Missing GuC-Err-Cap reglist Global!\n"); 561 else 562 drm_warn(&i915->drm, "Missing GuC-Err-Cap reglist %s(%u):%s(%u)!\n", 563 __stringify_type(type), type, 564 __stringify_engclass(classid), classid); 565 return -ENODATA; 566 } 567 568 num_regs = guc_cap_list_num_regs(gc, owner, type, classid); 569 /* intentional empty lists can exist depending on hw config */ 570 if (!num_regs) 571 return -ENODATA; 572 573 if (size) 574 *size = PAGE_ALIGN((sizeof(struct guc_debug_capture_list)) + 575 (num_regs * sizeof(struct guc_mmio_reg))); 576 577 return 0; 578 } 579 580 int 581 intel_guc_capture_getlistsize(struct intel_guc *guc, u32 owner, u32 type, u32 classid, 582 size_t *size) 583 { 584 return guc_capture_getlistsize(guc, owner, type, classid, size, false); 585 } 586 587 static void guc_capture_create_prealloc_nodes(struct intel_guc *guc); 588 589 int 590 intel_guc_capture_getlist(struct intel_guc *guc, u32 owner, u32 type, u32 classid, 591 void **outptr) 592 { 593 struct intel_guc_state_capture *gc = guc->capture; 594 struct __guc_capture_ads_cache *cache = &gc->ads_cache[owner][type][classid]; 595 struct drm_i915_private *i915 = guc_to_gt(guc)->i915; 596 struct guc_debug_capture_list *listnode; 597 int ret, num_regs; 598 u8 *caplist, *tmp; 599 size_t size = 0; 600 601 if (!gc->reglists) 602 return -ENODEV; 603 604 if (cache->is_valid) { 605 *outptr = cache->ptr; 606 return cache->status; 607 } 608 609 /* 610 * ADS population of input registers is a good 611 * time to pre-allocate cachelist output nodes 612 */ 613 guc_capture_create_prealloc_nodes(guc); 614 615 ret = intel_guc_capture_getlistsize(guc, owner, type, classid, &size); 616 if (ret) { 617 cache->is_valid = true; 618 cache->ptr = NULL; 619 cache->size = 0; 620 cache->status = ret; 621 return ret; 622 } 623 624 caplist = kzalloc(size, GFP_KERNEL); 625 if (!caplist) { 626 drm_dbg(&i915->drm, "GuC-capture: failed to alloc cached caplist"); 627 return -ENOMEM; 628 } 629 630 /* populate capture list header */ 631 tmp = caplist; 632 num_regs = guc_cap_list_num_regs(guc->capture, owner, type, classid); 633 listnode = (struct guc_debug_capture_list *)tmp; 634 listnode->header.info = FIELD_PREP(GUC_CAPTURELISTHDR_NUMDESCR, (u32)num_regs); 635 636 /* populate list of register descriptor */ 637 tmp += sizeof(struct guc_debug_capture_list); 638 guc_capture_list_init(guc, owner, type, classid, (struct guc_mmio_reg *)tmp, num_regs); 639 640 /* cache this list */ 641 cache->is_valid = true; 642 cache->ptr = caplist; 643 cache->size = size; 644 cache->status = 0; 645 646 *outptr = caplist; 647 648 return 0; 649 } 650 651 int 652 intel_guc_capture_getnullheader(struct intel_guc *guc, 653 void **outptr, size_t *size) 654 { 655 struct intel_guc_state_capture *gc = guc->capture; 656 struct drm_i915_private *i915 = guc_to_gt(guc)->i915; 657 int tmp = sizeof(u32) * 4; 658 void *null_header; 659 660 if (gc->ads_null_cache) { 661 *outptr = gc->ads_null_cache; 662 *size = tmp; 663 return 0; 664 } 665 666 null_header = kzalloc(tmp, GFP_KERNEL); 667 if (!null_header) { 668 drm_dbg(&i915->drm, "GuC-capture: failed to alloc cached nulllist"); 669 return -ENOMEM; 670 } 671 672 gc->ads_null_cache = null_header; 673 *outptr = null_header; 674 *size = tmp; 675 676 return 0; 677 } 678 679 static int 680 guc_capture_output_min_size_est(struct intel_guc *guc) 681 { 682 struct intel_gt *gt = guc_to_gt(guc); 683 struct intel_engine_cs *engine; 684 enum intel_engine_id id; 685 int worst_min_size = 0; 686 size_t tmp = 0; 687 688 if (!guc->capture) 689 return -ENODEV; 690 691 /* 692 * If every single engine-instance suffered a failure in quick succession but 693 * were all unrelated, then a burst of multiple error-capture events would dump 694 * registers for every one engine instance, one at a time. In this case, GuC 695 * would even dump the global-registers repeatedly. 696 * 697 * For each engine instance, there would be 1 x guc_state_capture_group_t output 698 * followed by 3 x guc_state_capture_t lists. The latter is how the register 699 * dumps are split across different register types (where the '3' are global vs class 700 * vs instance). 701 */ 702 for_each_engine(engine, gt, id) { 703 worst_min_size += sizeof(struct guc_state_capture_group_header_t) + 704 (3 * sizeof(struct guc_state_capture_header_t)); 705 706 if (!guc_capture_getlistsize(guc, 0, GUC_CAPTURE_LIST_TYPE_GLOBAL, 0, &tmp, true)) 707 worst_min_size += tmp; 708 709 if (!guc_capture_getlistsize(guc, 0, GUC_CAPTURE_LIST_TYPE_ENGINE_CLASS, 710 engine->class, &tmp, true)) { 711 worst_min_size += tmp; 712 } 713 if (!guc_capture_getlistsize(guc, 0, GUC_CAPTURE_LIST_TYPE_ENGINE_INSTANCE, 714 engine->class, &tmp, true)) { 715 worst_min_size += tmp; 716 } 717 } 718 719 return worst_min_size; 720 } 721 722 /* 723 * Add on a 3x multiplier to allow for multiple back-to-back captures occurring 724 * before the i915 can read the data out and process it 725 */ 726 #define GUC_CAPTURE_OVERBUFFER_MULTIPLIER 3 727 728 static void check_guc_capture_size(struct intel_guc *guc) 729 { 730 struct drm_i915_private *i915 = guc_to_gt(guc)->i915; 731 int min_size = guc_capture_output_min_size_est(guc); 732 int spare_size = min_size * GUC_CAPTURE_OVERBUFFER_MULTIPLIER; 733 u32 buffer_size = intel_guc_log_section_size_capture(&guc->log); 734 735 /* 736 * NOTE: min_size is much smaller than the capture region allocation (DG2: <80K vs 1MB) 737 * Additionally, its based on space needed to fit all engines getting reset at once 738 * within the same G2H handler task slot. This is very unlikely. However, if GuC really 739 * does run out of space for whatever reason, we will see an separate warning message 740 * when processing the G2H event capture-notification, search for: 741 * INTEL_GUC_STATE_CAPTURE_EVENT_STATUS_NOSPACE. 742 */ 743 if (min_size < 0) 744 drm_warn(&i915->drm, "Failed to calculate GuC error state capture buffer minimum size: %d!\n", 745 min_size); 746 else if (min_size > buffer_size) 747 drm_warn(&i915->drm, "GuC error state capture buffer maybe small: %d < %d\n", 748 buffer_size, min_size); 749 else if (spare_size > buffer_size) 750 drm_dbg(&i915->drm, "GuC error state capture buffer lacks spare size: %d < %d (min = %d)\n", 751 buffer_size, spare_size, min_size); 752 } 753 754 /* 755 * KMD Init time flows: 756 * -------------------- 757 * --> alloc A: GuC input capture regs lists (registered to GuC via ADS). 758 * intel_guc_ads acquires the register lists by calling 759 * intel_guc_capture_list_size and intel_guc_capture_list_get 'n' times, 760 * where n = 1 for global-reg-list + 761 * num_engine_classes for class-reg-list + 762 * num_engine_classes for instance-reg-list 763 * (since all instances of the same engine-class type 764 * have an identical engine-instance register-list). 765 * ADS module also calls separately for PF vs VF. 766 * 767 * --> alloc B: GuC output capture buf (registered via guc_init_params(log_param)) 768 * Size = #define CAPTURE_BUFFER_SIZE (warns if on too-small) 769 * Note2: 'x 3' to hold multiple capture groups 770 * 771 * GUC Runtime notify capture: 772 * -------------------------- 773 * --> G2H STATE_CAPTURE_NOTIFICATION 774 * L--> intel_guc_capture_process 775 * L--> Loop through B (head..tail) and for each engine instance's 776 * err-state-captured register-list we find, we alloc 'C': 777 * --> alloc C: A capture-output-node structure that includes misc capture info along 778 * with 3 register list dumps (global, engine-class and engine-instance) 779 * This node is created from a pre-allocated list of blank nodes in 780 * guc->capture->cachelist and populated with the error-capture 781 * data from GuC and then it's added into guc->capture->outlist linked 782 * list. This list is used for matchup and printout by i915_gpu_coredump 783 * and err_print_gt, (when user invokes the error capture sysfs). 784 * 785 * GUC --> notify context reset: 786 * ----------------------------- 787 * --> G2H CONTEXT RESET 788 * L--> guc_handle_context_reset --> i915_capture_error_state 789 * L--> i915_gpu_coredump(..IS_GUC_CAPTURE) --> gt_record_engines 790 * --> capture_engine(..IS_GUC_CAPTURE) 791 * L--> intel_guc_capture_get_matching_node is where 792 * detach C from internal linked list and add it into 793 * intel_engine_coredump struct (if the context and 794 * engine of the event notification matches a node 795 * in the link list). 796 * 797 * User Sysfs / Debugfs 798 * -------------------- 799 * --> i915_gpu_coredump_copy_to_buffer-> 800 * L--> err_print_to_sgl --> err_print_gt 801 * L--> error_print_guc_captures 802 * L--> intel_guc_capture_print_node prints the 803 * register lists values of the attached node 804 * on the error-engine-dump being reported. 805 * L--> i915_reset_error_state ... -->__i915_gpu_coredump_free 806 * L--> ... cleanup_gt --> 807 * L--> intel_guc_capture_free_node returns the 808 * capture-output-node back to the internal 809 * cachelist for reuse. 810 * 811 */ 812 813 static int guc_capture_buf_cnt(struct __guc_capture_bufstate *buf) 814 { 815 if (buf->wr >= buf->rd) 816 return (buf->wr - buf->rd); 817 return (buf->size - buf->rd) + buf->wr; 818 } 819 820 static int guc_capture_buf_cnt_to_end(struct __guc_capture_bufstate *buf) 821 { 822 if (buf->rd > buf->wr) 823 return (buf->size - buf->rd); 824 return (buf->wr - buf->rd); 825 } 826 827 /* 828 * GuC's error-capture output is a ring buffer populated in a byte-stream fashion: 829 * 830 * The GuC Log buffer region for error-capture is managed like a ring buffer. 831 * The GuC firmware dumps error capture logs into this ring in a byte-stream flow. 832 * Additionally, as per the current and foreseeable future, all packed error- 833 * capture output structures are dword aligned. 834 * 835 * That said, if the GuC firmware is in the midst of writing a structure that is larger 836 * than one dword but the tail end of the err-capture buffer-region has lesser space left, 837 * we would need to extract that structure one dword at a time straddled across the end, 838 * onto the start of the ring. 839 * 840 * Below function, guc_capture_log_remove_dw is a helper for that. All callers of this 841 * function would typically do a straight-up memcpy from the ring contents and will only 842 * call this helper if their structure-extraction is straddling across the end of the 843 * ring. GuC firmware does not add any padding. The reason for the no-padding is to ease 844 * scalability for future expansion of output data types without requiring a redesign 845 * of the flow controls. 846 */ 847 static int 848 guc_capture_log_remove_dw(struct intel_guc *guc, struct __guc_capture_bufstate *buf, 849 u32 *dw) 850 { 851 struct drm_i915_private *i915 = guc_to_gt(guc)->i915; 852 int tries = 2; 853 int avail = 0; 854 u32 *src_data; 855 856 if (!guc_capture_buf_cnt(buf)) 857 return 0; 858 859 while (tries--) { 860 avail = guc_capture_buf_cnt_to_end(buf); 861 if (avail >= sizeof(u32)) { 862 src_data = (u32 *)(buf->data + buf->rd); 863 *dw = *src_data; 864 buf->rd += 4; 865 return 4; 866 } 867 if (avail) 868 drm_dbg(&i915->drm, "GuC-Cap-Logs not dword aligned, skipping.\n"); 869 buf->rd = 0; 870 } 871 872 return 0; 873 } 874 875 static bool 876 guc_capture_data_extracted(struct __guc_capture_bufstate *b, 877 int size, void *dest) 878 { 879 if (guc_capture_buf_cnt_to_end(b) >= size) { 880 memcpy(dest, (b->data + b->rd), size); 881 b->rd += size; 882 return true; 883 } 884 return false; 885 } 886 887 static int 888 guc_capture_log_get_group_hdr(struct intel_guc *guc, struct __guc_capture_bufstate *buf, 889 struct guc_state_capture_group_header_t *ghdr) 890 { 891 int read = 0; 892 int fullsize = sizeof(struct guc_state_capture_group_header_t); 893 894 if (fullsize > guc_capture_buf_cnt(buf)) 895 return -1; 896 897 if (guc_capture_data_extracted(buf, fullsize, (void *)ghdr)) 898 return 0; 899 900 read += guc_capture_log_remove_dw(guc, buf, &ghdr->owner); 901 read += guc_capture_log_remove_dw(guc, buf, &ghdr->info); 902 if (read != fullsize) 903 return -1; 904 905 return 0; 906 } 907 908 static int 909 guc_capture_log_get_data_hdr(struct intel_guc *guc, struct __guc_capture_bufstate *buf, 910 struct guc_state_capture_header_t *hdr) 911 { 912 int read = 0; 913 int fullsize = sizeof(struct guc_state_capture_header_t); 914 915 if (fullsize > guc_capture_buf_cnt(buf)) 916 return -1; 917 918 if (guc_capture_data_extracted(buf, fullsize, (void *)hdr)) 919 return 0; 920 921 read += guc_capture_log_remove_dw(guc, buf, &hdr->owner); 922 read += guc_capture_log_remove_dw(guc, buf, &hdr->info); 923 read += guc_capture_log_remove_dw(guc, buf, &hdr->lrca); 924 read += guc_capture_log_remove_dw(guc, buf, &hdr->guc_id); 925 read += guc_capture_log_remove_dw(guc, buf, &hdr->num_mmios); 926 if (read != fullsize) 927 return -1; 928 929 return 0; 930 } 931 932 static int 933 guc_capture_log_get_register(struct intel_guc *guc, struct __guc_capture_bufstate *buf, 934 struct guc_mmio_reg *reg) 935 { 936 int read = 0; 937 int fullsize = sizeof(struct guc_mmio_reg); 938 939 if (fullsize > guc_capture_buf_cnt(buf)) 940 return -1; 941 942 if (guc_capture_data_extracted(buf, fullsize, (void *)reg)) 943 return 0; 944 945 read += guc_capture_log_remove_dw(guc, buf, ®->offset); 946 read += guc_capture_log_remove_dw(guc, buf, ®->value); 947 read += guc_capture_log_remove_dw(guc, buf, ®->flags); 948 read += guc_capture_log_remove_dw(guc, buf, ®->mask); 949 if (read != fullsize) 950 return -1; 951 952 return 0; 953 } 954 955 static void 956 guc_capture_delete_one_node(struct intel_guc *guc, struct __guc_capture_parsed_output *node) 957 { 958 int i; 959 960 for (i = 0; i < GUC_CAPTURE_LIST_TYPE_MAX; ++i) 961 kfree(node->reginfo[i].regs); 962 list_del(&node->link); 963 kfree(node); 964 } 965 966 static void 967 guc_capture_delete_prealloc_nodes(struct intel_guc *guc) 968 { 969 struct __guc_capture_parsed_output *n, *ntmp; 970 971 /* 972 * NOTE: At the end of driver operation, we must assume that we 973 * have prealloc nodes in both the cachelist as well as outlist 974 * if unclaimed error capture events occurred prior to shutdown. 975 */ 976 list_for_each_entry_safe(n, ntmp, &guc->capture->outlist, link) 977 guc_capture_delete_one_node(guc, n); 978 979 list_for_each_entry_safe(n, ntmp, &guc->capture->cachelist, link) 980 guc_capture_delete_one_node(guc, n); 981 } 982 983 static void 984 guc_capture_add_node_to_list(struct __guc_capture_parsed_output *node, 985 struct list_head *list) 986 { 987 list_add_tail(&node->link, list); 988 } 989 990 static void 991 guc_capture_add_node_to_outlist(struct intel_guc_state_capture *gc, 992 struct __guc_capture_parsed_output *node) 993 { 994 guc_capture_add_node_to_list(node, &gc->outlist); 995 } 996 997 static void 998 guc_capture_add_node_to_cachelist(struct intel_guc_state_capture *gc, 999 struct __guc_capture_parsed_output *node) 1000 { 1001 guc_capture_add_node_to_list(node, &gc->cachelist); 1002 } 1003 1004 static void 1005 guc_capture_init_node(struct intel_guc *guc, struct __guc_capture_parsed_output *node) 1006 { 1007 struct guc_mmio_reg *tmp[GUC_CAPTURE_LIST_TYPE_MAX]; 1008 int i; 1009 1010 for (i = 0; i < GUC_CAPTURE_LIST_TYPE_MAX; ++i) { 1011 tmp[i] = node->reginfo[i].regs; 1012 memset(tmp[i], 0, sizeof(struct guc_mmio_reg) * 1013 guc->capture->max_mmio_per_node); 1014 } 1015 memset(node, 0, sizeof(*node)); 1016 for (i = 0; i < GUC_CAPTURE_LIST_TYPE_MAX; ++i) 1017 node->reginfo[i].regs = tmp[i]; 1018 1019 INIT_LIST_HEAD(&node->link); 1020 } 1021 1022 static struct __guc_capture_parsed_output * 1023 guc_capture_get_prealloc_node(struct intel_guc *guc) 1024 { 1025 struct __guc_capture_parsed_output *found = NULL; 1026 1027 if (!list_empty(&guc->capture->cachelist)) { 1028 struct __guc_capture_parsed_output *n, *ntmp; 1029 1030 /* get first avail node from the cache list */ 1031 list_for_each_entry_safe(n, ntmp, &guc->capture->cachelist, link) { 1032 found = n; 1033 list_del(&n->link); 1034 break; 1035 } 1036 } else { 1037 struct __guc_capture_parsed_output *n, *ntmp; 1038 1039 /* traverse down and steal back the oldest node already allocated */ 1040 list_for_each_entry_safe(n, ntmp, &guc->capture->outlist, link) { 1041 found = n; 1042 } 1043 if (found) 1044 list_del(&found->link); 1045 } 1046 if (found) 1047 guc_capture_init_node(guc, found); 1048 1049 return found; 1050 } 1051 1052 static struct __guc_capture_parsed_output * 1053 guc_capture_alloc_one_node(struct intel_guc *guc) 1054 { 1055 struct __guc_capture_parsed_output *new; 1056 int i; 1057 1058 new = kzalloc(sizeof(*new), GFP_KERNEL); 1059 if (!new) 1060 return NULL; 1061 1062 for (i = 0; i < GUC_CAPTURE_LIST_TYPE_MAX; ++i) { 1063 new->reginfo[i].regs = kcalloc(guc->capture->max_mmio_per_node, 1064 sizeof(struct guc_mmio_reg), GFP_KERNEL); 1065 if (!new->reginfo[i].regs) { 1066 while (i) 1067 kfree(new->reginfo[--i].regs); 1068 kfree(new); 1069 return NULL; 1070 } 1071 } 1072 guc_capture_init_node(guc, new); 1073 1074 return new; 1075 } 1076 1077 static struct __guc_capture_parsed_output * 1078 guc_capture_clone_node(struct intel_guc *guc, struct __guc_capture_parsed_output *original, 1079 u32 keep_reglist_mask) 1080 { 1081 struct __guc_capture_parsed_output *new; 1082 int i; 1083 1084 new = guc_capture_get_prealloc_node(guc); 1085 if (!new) 1086 return NULL; 1087 if (!original) 1088 return new; 1089 1090 new->is_partial = original->is_partial; 1091 1092 /* copy reg-lists that we want to clone */ 1093 for (i = 0; i < GUC_CAPTURE_LIST_TYPE_MAX; ++i) { 1094 if (keep_reglist_mask & BIT(i)) { 1095 GEM_BUG_ON(original->reginfo[i].num_regs > 1096 guc->capture->max_mmio_per_node); 1097 1098 memcpy(new->reginfo[i].regs, original->reginfo[i].regs, 1099 original->reginfo[i].num_regs * sizeof(struct guc_mmio_reg)); 1100 1101 new->reginfo[i].num_regs = original->reginfo[i].num_regs; 1102 new->reginfo[i].vfid = original->reginfo[i].vfid; 1103 1104 if (i == GUC_CAPTURE_LIST_TYPE_ENGINE_CLASS) { 1105 new->eng_class = original->eng_class; 1106 } else if (i == GUC_CAPTURE_LIST_TYPE_ENGINE_INSTANCE) { 1107 new->eng_inst = original->eng_inst; 1108 new->guc_id = original->guc_id; 1109 new->lrca = original->lrca; 1110 } 1111 } 1112 } 1113 1114 return new; 1115 } 1116 1117 static void 1118 __guc_capture_create_prealloc_nodes(struct intel_guc *guc) 1119 { 1120 struct __guc_capture_parsed_output *node = NULL; 1121 struct drm_i915_private *i915 = guc_to_gt(guc)->i915; 1122 int i; 1123 1124 for (i = 0; i < PREALLOC_NODES_MAX_COUNT; ++i) { 1125 node = guc_capture_alloc_one_node(guc); 1126 if (!node) { 1127 drm_warn(&i915->drm, "GuC Capture pre-alloc-cache failure\n"); 1128 /* dont free the priors, use what we got and cleanup at shutdown */ 1129 return; 1130 } 1131 guc_capture_add_node_to_cachelist(guc->capture, node); 1132 } 1133 } 1134 1135 static int 1136 guc_get_max_reglist_count(struct intel_guc *guc) 1137 { 1138 int i, j, k, tmp, maxregcount = 0; 1139 1140 for (i = 0; i < GUC_CAPTURE_LIST_INDEX_MAX; ++i) { 1141 for (j = 0; j < GUC_CAPTURE_LIST_TYPE_MAX; ++j) { 1142 for (k = 0; k < GUC_MAX_ENGINE_CLASSES; ++k) { 1143 if (j == GUC_CAPTURE_LIST_TYPE_GLOBAL && k > 0) 1144 continue; 1145 1146 tmp = guc_cap_list_num_regs(guc->capture, i, j, k); 1147 if (tmp > maxregcount) 1148 maxregcount = tmp; 1149 } 1150 } 1151 } 1152 if (!maxregcount) 1153 maxregcount = PREALLOC_NODES_DEFAULT_NUMREGS; 1154 1155 return maxregcount; 1156 } 1157 1158 static void 1159 guc_capture_create_prealloc_nodes(struct intel_guc *guc) 1160 { 1161 /* skip if we've already done the pre-alloc */ 1162 if (guc->capture->max_mmio_per_node) 1163 return; 1164 1165 guc->capture->max_mmio_per_node = guc_get_max_reglist_count(guc); 1166 __guc_capture_create_prealloc_nodes(guc); 1167 } 1168 1169 static int 1170 guc_capture_extract_reglists(struct intel_guc *guc, struct __guc_capture_bufstate *buf) 1171 { 1172 struct drm_i915_private *i915 = guc_to_gt(guc)->i915; 1173 struct guc_state_capture_group_header_t ghdr = {0}; 1174 struct guc_state_capture_header_t hdr = {0}; 1175 struct __guc_capture_parsed_output *node = NULL; 1176 struct guc_mmio_reg *regs = NULL; 1177 int i, numlists, numregs, ret = 0; 1178 enum guc_capture_type datatype; 1179 struct guc_mmio_reg tmp; 1180 bool is_partial = false; 1181 1182 i = guc_capture_buf_cnt(buf); 1183 if (!i) 1184 return -ENODATA; 1185 if (i % sizeof(u32)) { 1186 drm_warn(&i915->drm, "GuC Capture new entries unaligned\n"); 1187 ret = -EIO; 1188 goto bailout; 1189 } 1190 1191 /* first get the capture group header */ 1192 if (guc_capture_log_get_group_hdr(guc, buf, &ghdr)) { 1193 ret = -EIO; 1194 goto bailout; 1195 } 1196 /* 1197 * we would typically expect a layout as below where n would be expected to be 1198 * anywhere between 3 to n where n > 3 if we are seeing multiple dependent engine 1199 * instances being reset together. 1200 * ____________________________________________ 1201 * | Capture Group | 1202 * | ________________________________________ | 1203 * | | Capture Group Header: | | 1204 * | | - num_captures = 5 | | 1205 * | |______________________________________| | 1206 * | ________________________________________ | 1207 * | | Capture1: | | 1208 * | | Hdr: GLOBAL, numregs=a | | 1209 * | | ____________________________________ | | 1210 * | | | Reglist | | | 1211 * | | | - reg1, reg2, ... rega | | | 1212 * | | |__________________________________| | | 1213 * | |______________________________________| | 1214 * | ________________________________________ | 1215 * | | Capture2: | | 1216 * | | Hdr: CLASS=RENDER/COMPUTE, numregs=b| | 1217 * | | ____________________________________ | | 1218 * | | | Reglist | | | 1219 * | | | - reg1, reg2, ... regb | | | 1220 * | | |__________________________________| | | 1221 * | |______________________________________| | 1222 * | ________________________________________ | 1223 * | | Capture3: | | 1224 * | | Hdr: INSTANCE=RCS, numregs=c | | 1225 * | | ____________________________________ | | 1226 * | | | Reglist | | | 1227 * | | | - reg1, reg2, ... regc | | | 1228 * | | |__________________________________| | | 1229 * | |______________________________________| | 1230 * | ________________________________________ | 1231 * | | Capture4: | | 1232 * | | Hdr: CLASS=RENDER/COMPUTE, numregs=d| | 1233 * | | ____________________________________ | | 1234 * | | | Reglist | | | 1235 * | | | - reg1, reg2, ... regd | | | 1236 * | | |__________________________________| | | 1237 * | |______________________________________| | 1238 * | ________________________________________ | 1239 * | | Capture5: | | 1240 * | | Hdr: INSTANCE=CCS0, numregs=e | | 1241 * | | ____________________________________ | | 1242 * | | | Reglist | | | 1243 * | | | - reg1, reg2, ... rege | | | 1244 * | | |__________________________________| | | 1245 * | |______________________________________| | 1246 * |__________________________________________| 1247 */ 1248 is_partial = FIELD_GET(CAP_GRP_HDR_CAPTURE_TYPE, ghdr.info); 1249 numlists = FIELD_GET(CAP_GRP_HDR_NUM_CAPTURES, ghdr.info); 1250 1251 while (numlists--) { 1252 if (guc_capture_log_get_data_hdr(guc, buf, &hdr)) { 1253 ret = -EIO; 1254 break; 1255 } 1256 1257 datatype = FIELD_GET(CAP_HDR_CAPTURE_TYPE, hdr.info); 1258 if (datatype > GUC_CAPTURE_LIST_TYPE_ENGINE_INSTANCE) { 1259 /* unknown capture type - skip over to next capture set */ 1260 numregs = FIELD_GET(CAP_HDR_NUM_MMIOS, hdr.num_mmios); 1261 while (numregs--) { 1262 if (guc_capture_log_get_register(guc, buf, &tmp)) { 1263 ret = -EIO; 1264 break; 1265 } 1266 } 1267 continue; 1268 } else if (node) { 1269 /* 1270 * Based on the current capture type and what we have so far, 1271 * decide if we should add the current node into the internal 1272 * linked list for match-up when i915_gpu_coredump calls later 1273 * (and alloc a blank node for the next set of reglists) 1274 * or continue with the same node or clone the current node 1275 * but only retain the global or class registers (such as the 1276 * case of dependent engine resets). 1277 */ 1278 if (datatype == GUC_CAPTURE_LIST_TYPE_GLOBAL) { 1279 guc_capture_add_node_to_outlist(guc->capture, node); 1280 node = NULL; 1281 } else if (datatype == GUC_CAPTURE_LIST_TYPE_ENGINE_CLASS && 1282 node->reginfo[GUC_CAPTURE_LIST_TYPE_ENGINE_CLASS].num_regs) { 1283 /* Add to list, clone node and duplicate global list */ 1284 guc_capture_add_node_to_outlist(guc->capture, node); 1285 node = guc_capture_clone_node(guc, node, 1286 GCAP_PARSED_REGLIST_INDEX_GLOBAL); 1287 } else if (datatype == GUC_CAPTURE_LIST_TYPE_ENGINE_INSTANCE && 1288 node->reginfo[GUC_CAPTURE_LIST_TYPE_ENGINE_INSTANCE].num_regs) { 1289 /* Add to list, clone node and duplicate global + class lists */ 1290 guc_capture_add_node_to_outlist(guc->capture, node); 1291 node = guc_capture_clone_node(guc, node, 1292 (GCAP_PARSED_REGLIST_INDEX_GLOBAL | 1293 GCAP_PARSED_REGLIST_INDEX_ENGCLASS)); 1294 } 1295 } 1296 1297 if (!node) { 1298 node = guc_capture_get_prealloc_node(guc); 1299 if (!node) { 1300 ret = -ENOMEM; 1301 break; 1302 } 1303 if (datatype != GUC_CAPTURE_LIST_TYPE_GLOBAL) 1304 drm_dbg(&i915->drm, "GuC Capture missing global dump: %08x!\n", 1305 datatype); 1306 } 1307 node->is_partial = is_partial; 1308 node->reginfo[datatype].vfid = FIELD_GET(CAP_HDR_CAPTURE_VFID, hdr.owner); 1309 switch (datatype) { 1310 case GUC_CAPTURE_LIST_TYPE_ENGINE_INSTANCE: 1311 node->eng_class = FIELD_GET(CAP_HDR_ENGINE_CLASS, hdr.info); 1312 node->eng_inst = FIELD_GET(CAP_HDR_ENGINE_INSTANCE, hdr.info); 1313 node->lrca = hdr.lrca; 1314 node->guc_id = hdr.guc_id; 1315 break; 1316 case GUC_CAPTURE_LIST_TYPE_ENGINE_CLASS: 1317 node->eng_class = FIELD_GET(CAP_HDR_ENGINE_CLASS, hdr.info); 1318 break; 1319 default: 1320 break; 1321 } 1322 1323 numregs = FIELD_GET(CAP_HDR_NUM_MMIOS, hdr.num_mmios); 1324 if (numregs > guc->capture->max_mmio_per_node) { 1325 drm_dbg(&i915->drm, "GuC Capture list extraction clipped by prealloc!\n"); 1326 numregs = guc->capture->max_mmio_per_node; 1327 } 1328 node->reginfo[datatype].num_regs = numregs; 1329 regs = node->reginfo[datatype].regs; 1330 i = 0; 1331 while (numregs--) { 1332 if (guc_capture_log_get_register(guc, buf, ®s[i++])) { 1333 ret = -EIO; 1334 break; 1335 } 1336 } 1337 } 1338 1339 bailout: 1340 if (node) { 1341 /* If we have data, add to linked list for match-up when i915_gpu_coredump calls */ 1342 for (i = GUC_CAPTURE_LIST_TYPE_GLOBAL; i < GUC_CAPTURE_LIST_TYPE_MAX; ++i) { 1343 if (node->reginfo[i].regs) { 1344 guc_capture_add_node_to_outlist(guc->capture, node); 1345 node = NULL; 1346 break; 1347 } 1348 } 1349 if (node) /* else return it back to cache list */ 1350 guc_capture_add_node_to_cachelist(guc->capture, node); 1351 } 1352 return ret; 1353 } 1354 1355 static int __guc_capture_flushlog_complete(struct intel_guc *guc) 1356 { 1357 u32 action[] = { 1358 INTEL_GUC_ACTION_LOG_BUFFER_FILE_FLUSH_COMPLETE, 1359 GUC_CAPTURE_LOG_BUFFER 1360 }; 1361 1362 return intel_guc_send_nb(guc, action, ARRAY_SIZE(action), 0); 1363 1364 } 1365 1366 static void __guc_capture_process_output(struct intel_guc *guc) 1367 { 1368 unsigned int buffer_size, read_offset, write_offset, full_count; 1369 struct intel_uc *uc = container_of(guc, typeof(*uc), guc); 1370 struct drm_i915_private *i915 = guc_to_gt(guc)->i915; 1371 struct guc_log_buffer_state log_buf_state_local; 1372 struct guc_log_buffer_state *log_buf_state; 1373 struct __guc_capture_bufstate buf; 1374 void *src_data = NULL; 1375 bool new_overflow; 1376 int ret; 1377 1378 log_buf_state = guc->log.buf_addr + 1379 (sizeof(struct guc_log_buffer_state) * GUC_CAPTURE_LOG_BUFFER); 1380 src_data = guc->log.buf_addr + 1381 intel_guc_get_log_buffer_offset(&guc->log, GUC_CAPTURE_LOG_BUFFER); 1382 1383 /* 1384 * Make a copy of the state structure, inside GuC log buffer 1385 * (which is uncached mapped), on the stack to avoid reading 1386 * from it multiple times. 1387 */ 1388 memcpy(&log_buf_state_local, log_buf_state, sizeof(struct guc_log_buffer_state)); 1389 buffer_size = intel_guc_get_log_buffer_size(&guc->log, GUC_CAPTURE_LOG_BUFFER); 1390 read_offset = log_buf_state_local.read_ptr; 1391 write_offset = log_buf_state_local.sampled_write_ptr; 1392 full_count = log_buf_state_local.buffer_full_cnt; 1393 1394 /* Bookkeeping stuff */ 1395 guc->log.stats[GUC_CAPTURE_LOG_BUFFER].flush += log_buf_state_local.flush_to_file; 1396 new_overflow = intel_guc_check_log_buf_overflow(&guc->log, GUC_CAPTURE_LOG_BUFFER, 1397 full_count); 1398 1399 /* Now copy the actual logs. */ 1400 if (unlikely(new_overflow)) { 1401 /* copy the whole buffer in case of overflow */ 1402 read_offset = 0; 1403 write_offset = buffer_size; 1404 } else if (unlikely((read_offset > buffer_size) || 1405 (write_offset > buffer_size))) { 1406 drm_err(&i915->drm, "invalid GuC log capture buffer state!\n"); 1407 /* copy whole buffer as offsets are unreliable */ 1408 read_offset = 0; 1409 write_offset = buffer_size; 1410 } 1411 1412 buf.size = buffer_size; 1413 buf.rd = read_offset; 1414 buf.wr = write_offset; 1415 buf.data = src_data; 1416 1417 if (!uc->reset_in_progress) { 1418 do { 1419 ret = guc_capture_extract_reglists(guc, &buf); 1420 } while (ret >= 0); 1421 } 1422 1423 /* Update the state of log buffer err-cap state */ 1424 log_buf_state->read_ptr = write_offset; 1425 log_buf_state->flush_to_file = 0; 1426 __guc_capture_flushlog_complete(guc); 1427 } 1428 1429 #if IS_ENABLED(CONFIG_DRM_I915_CAPTURE_ERROR) 1430 1431 static const char * 1432 guc_capture_reg_to_str(const struct intel_guc *guc, u32 owner, u32 type, 1433 u32 class, u32 id, u32 offset, u32 *is_ext) 1434 { 1435 const struct __guc_mmio_reg_descr_group *reglists = guc->capture->reglists; 1436 struct __guc_mmio_reg_descr_group *extlists = guc->capture->extlists; 1437 const struct __guc_mmio_reg_descr_group *match; 1438 struct __guc_mmio_reg_descr_group *matchext; 1439 int j; 1440 1441 *is_ext = 0; 1442 if (!reglists) 1443 return NULL; 1444 1445 match = guc_capture_get_one_list(reglists, owner, type, id); 1446 if (!match) 1447 return NULL; 1448 1449 for (j = 0; j < match->num_regs; ++j) { 1450 if (offset == match->list[j].reg.reg) 1451 return match->list[j].regname; 1452 } 1453 if (extlists) { 1454 matchext = guc_capture_get_one_ext_list(extlists, owner, type, id); 1455 if (!matchext) 1456 return NULL; 1457 for (j = 0; j < matchext->num_regs; ++j) { 1458 if (offset == matchext->extlist[j].reg.reg) { 1459 *is_ext = 1; 1460 return matchext->extlist[j].regname; 1461 } 1462 } 1463 } 1464 1465 return NULL; 1466 } 1467 1468 #define GCAP_PRINT_INTEL_ENG_INFO(ebuf, eng) \ 1469 do { \ 1470 i915_error_printf(ebuf, " i915-Eng-Name: %s command stream\n", \ 1471 (eng)->name); \ 1472 i915_error_printf(ebuf, " i915-Eng-Inst-Class: 0x%02x\n", (eng)->class); \ 1473 i915_error_printf(ebuf, " i915-Eng-Inst-Id: 0x%02x\n", (eng)->instance); \ 1474 i915_error_printf(ebuf, " i915-Eng-LogicalMask: 0x%08x\n", \ 1475 (eng)->logical_mask); \ 1476 } while (0) 1477 1478 #define GCAP_PRINT_GUC_INST_INFO(ebuf, node) \ 1479 do { \ 1480 i915_error_printf(ebuf, " GuC-Engine-Inst-Id: 0x%08x\n", \ 1481 (node)->eng_inst); \ 1482 i915_error_printf(ebuf, " GuC-Context-Id: 0x%08x\n", (node)->guc_id); \ 1483 i915_error_printf(ebuf, " LRCA: 0x%08x\n", (node)->lrca); \ 1484 } while (0) 1485 1486 int intel_guc_capture_print_engine_node(struct drm_i915_error_state_buf *ebuf, 1487 const struct intel_engine_coredump *ee) 1488 { 1489 const char *grptype[GUC_STATE_CAPTURE_GROUP_TYPE_MAX] = { 1490 "full-capture", 1491 "partial-capture" 1492 }; 1493 const char *datatype[GUC_CAPTURE_LIST_TYPE_MAX] = { 1494 "Global", 1495 "Engine-Class", 1496 "Engine-Instance" 1497 }; 1498 struct intel_guc_state_capture *cap; 1499 struct __guc_capture_parsed_output *node; 1500 struct intel_engine_cs *eng; 1501 struct guc_mmio_reg *regs; 1502 struct intel_guc *guc; 1503 const char *str; 1504 int numregs, i, j; 1505 u32 is_ext; 1506 1507 if (!ebuf || !ee) 1508 return -EINVAL; 1509 cap = ee->guc_capture; 1510 if (!cap || !ee->engine) 1511 return -ENODEV; 1512 1513 guc = &ee->engine->gt->uc.guc; 1514 1515 i915_error_printf(ebuf, "global --- GuC Error Capture on %s command stream:\n", 1516 ee->engine->name); 1517 1518 node = ee->guc_capture_node; 1519 if (!node) { 1520 i915_error_printf(ebuf, " No matching ee-node\n"); 1521 return 0; 1522 } 1523 1524 i915_error_printf(ebuf, "Coverage: %s\n", grptype[node->is_partial]); 1525 1526 for (i = GUC_CAPTURE_LIST_TYPE_GLOBAL; i < GUC_CAPTURE_LIST_TYPE_MAX; ++i) { 1527 i915_error_printf(ebuf, " RegListType: %s\n", 1528 datatype[i % GUC_CAPTURE_LIST_TYPE_MAX]); 1529 i915_error_printf(ebuf, " Owner-Id: %d\n", node->reginfo[i].vfid); 1530 1531 switch (i) { 1532 case GUC_CAPTURE_LIST_TYPE_GLOBAL: 1533 default: 1534 break; 1535 case GUC_CAPTURE_LIST_TYPE_ENGINE_CLASS: 1536 i915_error_printf(ebuf, " GuC-Eng-Class: %d\n", node->eng_class); 1537 i915_error_printf(ebuf, " i915-Eng-Class: %d\n", 1538 guc_class_to_engine_class(node->eng_class)); 1539 break; 1540 case GUC_CAPTURE_LIST_TYPE_ENGINE_INSTANCE: 1541 eng = intel_guc_lookup_engine(guc, node->eng_class, node->eng_inst); 1542 if (eng) 1543 GCAP_PRINT_INTEL_ENG_INFO(ebuf, eng); 1544 else 1545 i915_error_printf(ebuf, " i915-Eng-Lookup Fail!\n"); 1546 GCAP_PRINT_GUC_INST_INFO(ebuf, node); 1547 break; 1548 } 1549 1550 numregs = node->reginfo[i].num_regs; 1551 i915_error_printf(ebuf, " NumRegs: %d\n", numregs); 1552 j = 0; 1553 while (numregs--) { 1554 regs = node->reginfo[i].regs; 1555 str = guc_capture_reg_to_str(guc, GUC_CAPTURE_LIST_INDEX_PF, i, 1556 node->eng_class, 0, regs[j].offset, &is_ext); 1557 if (!str) 1558 i915_error_printf(ebuf, " REG-0x%08x", regs[j].offset); 1559 else 1560 i915_error_printf(ebuf, " %s", str); 1561 if (is_ext) 1562 i915_error_printf(ebuf, "[%ld][%ld]", 1563 FIELD_GET(GUC_REGSET_STEERING_GROUP, regs[j].flags), 1564 FIELD_GET(GUC_REGSET_STEERING_INSTANCE, regs[j].flags)); 1565 i915_error_printf(ebuf, ": 0x%08x\n", regs[j].value); 1566 ++j; 1567 } 1568 } 1569 return 0; 1570 } 1571 1572 #endif //CONFIG_DRM_I915_CAPTURE_ERROR 1573 1574 void intel_guc_capture_free_node(struct intel_engine_coredump *ee) 1575 { 1576 if (!ee || !ee->guc_capture_node) 1577 return; 1578 1579 guc_capture_add_node_to_cachelist(ee->guc_capture, ee->guc_capture_node); 1580 ee->guc_capture = NULL; 1581 ee->guc_capture_node = NULL; 1582 } 1583 1584 void intel_guc_capture_get_matching_node(struct intel_gt *gt, 1585 struct intel_engine_coredump *ee, 1586 struct intel_context *ce) 1587 { 1588 struct __guc_capture_parsed_output *n, *ntmp; 1589 struct drm_i915_private *i915; 1590 struct intel_guc *guc; 1591 1592 if (!gt || !ee || !ce) 1593 return; 1594 1595 i915 = gt->i915; 1596 guc = >->uc.guc; 1597 if (!guc->capture) 1598 return; 1599 1600 GEM_BUG_ON(ee->guc_capture_node); 1601 /* 1602 * Look for a matching GuC reported error capture node from 1603 * the internal output link-list based on lrca, guc-id and engine 1604 * identification. 1605 */ 1606 list_for_each_entry_safe(n, ntmp, &guc->capture->outlist, link) { 1607 if (n->eng_inst == GUC_ID_TO_ENGINE_INSTANCE(ee->engine->guc_id) && 1608 n->eng_class == GUC_ID_TO_ENGINE_CLASS(ee->engine->guc_id) && 1609 n->guc_id && n->guc_id == ce->guc_id.id && 1610 (n->lrca & CTX_GTT_ADDRESS_MASK) && (n->lrca & CTX_GTT_ADDRESS_MASK) == 1611 (ce->lrc.lrca & CTX_GTT_ADDRESS_MASK)) { 1612 list_del(&n->link); 1613 ee->guc_capture_node = n; 1614 ee->guc_capture = guc->capture; 1615 return; 1616 } 1617 } 1618 drm_dbg(&i915->drm, "GuC capture can't match ee to node\n"); 1619 } 1620 1621 void intel_guc_capture_process(struct intel_guc *guc) 1622 { 1623 if (guc->capture) 1624 __guc_capture_process_output(guc); 1625 } 1626 1627 static void 1628 guc_capture_free_ads_cache(struct intel_guc_state_capture *gc) 1629 { 1630 int i, j, k; 1631 struct __guc_capture_ads_cache *cache; 1632 1633 for (i = 0; i < GUC_CAPTURE_LIST_INDEX_MAX; ++i) { 1634 for (j = 0; j < GUC_CAPTURE_LIST_TYPE_MAX; ++j) { 1635 for (k = 0; k < GUC_MAX_ENGINE_CLASSES; ++k) { 1636 cache = &gc->ads_cache[i][j][k]; 1637 if (cache->is_valid) 1638 kfree(cache->ptr); 1639 } 1640 } 1641 } 1642 kfree(gc->ads_null_cache); 1643 } 1644 1645 void intel_guc_capture_destroy(struct intel_guc *guc) 1646 { 1647 if (!guc->capture) 1648 return; 1649 1650 guc_capture_free_ads_cache(guc->capture); 1651 1652 guc_capture_delete_prealloc_nodes(guc); 1653 1654 guc_capture_free_extlists(guc->capture->extlists); 1655 kfree(guc->capture->extlists); 1656 1657 kfree(guc->capture); 1658 guc->capture = NULL; 1659 } 1660 1661 int intel_guc_capture_init(struct intel_guc *guc) 1662 { 1663 guc->capture = kzalloc(sizeof(*guc->capture), GFP_KERNEL); 1664 if (!guc->capture) 1665 return -ENOMEM; 1666 1667 guc->capture->reglists = guc_capture_get_device_reglist(guc); 1668 1669 INIT_LIST_HEAD(&guc->capture->outlist); 1670 INIT_LIST_HEAD(&guc->capture->cachelist); 1671 1672 check_guc_capture_size(guc); 1673 1674 return 0; 1675 } 1676