1 /* 2 * Copyright © 2016 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 */ 24 25 #include <drm/drm_print.h> 26 27 #include "intel_device_info.h" 28 #include "i915_drv.h" 29 30 #define PLATFORM_NAME(x) [INTEL_##x] = #x 31 static const char * const platform_names[] = { 32 PLATFORM_NAME(I830), 33 PLATFORM_NAME(I845G), 34 PLATFORM_NAME(I85X), 35 PLATFORM_NAME(I865G), 36 PLATFORM_NAME(I915G), 37 PLATFORM_NAME(I915GM), 38 PLATFORM_NAME(I945G), 39 PLATFORM_NAME(I945GM), 40 PLATFORM_NAME(G33), 41 PLATFORM_NAME(PINEVIEW), 42 PLATFORM_NAME(I965G), 43 PLATFORM_NAME(I965GM), 44 PLATFORM_NAME(G45), 45 PLATFORM_NAME(GM45), 46 PLATFORM_NAME(IRONLAKE), 47 PLATFORM_NAME(SANDYBRIDGE), 48 PLATFORM_NAME(IVYBRIDGE), 49 PLATFORM_NAME(VALLEYVIEW), 50 PLATFORM_NAME(HASWELL), 51 PLATFORM_NAME(BROADWELL), 52 PLATFORM_NAME(CHERRYVIEW), 53 PLATFORM_NAME(SKYLAKE), 54 PLATFORM_NAME(BROXTON), 55 PLATFORM_NAME(KABYLAKE), 56 PLATFORM_NAME(GEMINILAKE), 57 PLATFORM_NAME(COFFEELAKE), 58 PLATFORM_NAME(CANNONLAKE), 59 PLATFORM_NAME(ICELAKE), 60 PLATFORM_NAME(ELKHARTLAKE), 61 PLATFORM_NAME(TIGERLAKE), 62 }; 63 #undef PLATFORM_NAME 64 65 const char *intel_platform_name(enum intel_platform platform) 66 { 67 BUILD_BUG_ON(ARRAY_SIZE(platform_names) != INTEL_MAX_PLATFORMS); 68 69 if (WARN_ON_ONCE(platform >= ARRAY_SIZE(platform_names) || 70 platform_names[platform] == NULL)) 71 return "<unknown>"; 72 73 return platform_names[platform]; 74 } 75 76 void intel_device_info_dump_flags(const struct intel_device_info *info, 77 struct drm_printer *p) 78 { 79 #define PRINT_FLAG(name) drm_printf(p, "%s: %s\n", #name, yesno(info->name)); 80 DEV_INFO_FOR_EACH_FLAG(PRINT_FLAG); 81 #undef PRINT_FLAG 82 83 #define PRINT_FLAG(name) drm_printf(p, "%s: %s\n", #name, yesno(info->display.name)); 84 DEV_INFO_DISPLAY_FOR_EACH_FLAG(PRINT_FLAG); 85 #undef PRINT_FLAG 86 } 87 88 static void sseu_dump(const struct sseu_dev_info *sseu, struct drm_printer *p) 89 { 90 int s; 91 92 drm_printf(p, "slice total: %u, mask=%04x\n", 93 hweight8(sseu->slice_mask), sseu->slice_mask); 94 drm_printf(p, "subslice total: %u\n", intel_sseu_subslice_total(sseu)); 95 for (s = 0; s < sseu->max_slices; s++) { 96 drm_printf(p, "slice%d: %u subslices, mask=%04x\n", 97 s, intel_sseu_subslices_per_slice(sseu, s), 98 sseu->subslice_mask[s]); 99 } 100 drm_printf(p, "EU total: %u\n", sseu->eu_total); 101 drm_printf(p, "EU per subslice: %u\n", sseu->eu_per_subslice); 102 drm_printf(p, "has slice power gating: %s\n", 103 yesno(sseu->has_slice_pg)); 104 drm_printf(p, "has subslice power gating: %s\n", 105 yesno(sseu->has_subslice_pg)); 106 drm_printf(p, "has EU power gating: %s\n", yesno(sseu->has_eu_pg)); 107 } 108 109 void intel_device_info_dump_runtime(const struct intel_runtime_info *info, 110 struct drm_printer *p) 111 { 112 sseu_dump(&info->sseu, p); 113 114 drm_printf(p, "CS timestamp frequency: %u kHz\n", 115 info->cs_timestamp_frequency_khz); 116 } 117 118 static int sseu_eu_idx(const struct sseu_dev_info *sseu, int slice, 119 int subslice) 120 { 121 int subslice_stride = GEN_SSEU_STRIDE(sseu->max_eus_per_subslice); 122 int slice_stride = sseu->max_subslices * subslice_stride; 123 124 return slice * slice_stride + subslice * subslice_stride; 125 } 126 127 static u16 sseu_get_eus(const struct sseu_dev_info *sseu, int slice, 128 int subslice) 129 { 130 int i, offset = sseu_eu_idx(sseu, slice, subslice); 131 u16 eu_mask = 0; 132 133 for (i = 0; i < GEN_SSEU_STRIDE(sseu->max_eus_per_subslice); i++) { 134 eu_mask |= ((u16)sseu->eu_mask[offset + i]) << 135 (i * BITS_PER_BYTE); 136 } 137 138 return eu_mask; 139 } 140 141 static void sseu_set_eus(struct sseu_dev_info *sseu, int slice, int subslice, 142 u16 eu_mask) 143 { 144 int i, offset = sseu_eu_idx(sseu, slice, subslice); 145 146 for (i = 0; i < GEN_SSEU_STRIDE(sseu->max_eus_per_subslice); i++) { 147 sseu->eu_mask[offset + i] = 148 (eu_mask >> (BITS_PER_BYTE * i)) & 0xff; 149 } 150 } 151 152 void intel_device_info_dump_topology(const struct sseu_dev_info *sseu, 153 struct drm_printer *p) 154 { 155 int s, ss; 156 157 if (sseu->max_slices == 0) { 158 drm_printf(p, "Unavailable\n"); 159 return; 160 } 161 162 for (s = 0; s < sseu->max_slices; s++) { 163 drm_printf(p, "slice%d: %u subslice(s) (0x%hhx):\n", 164 s, intel_sseu_subslices_per_slice(sseu, s), 165 sseu->subslice_mask[s]); 166 167 for (ss = 0; ss < sseu->max_subslices; ss++) { 168 u16 enabled_eus = sseu_get_eus(sseu, s, ss); 169 170 drm_printf(p, "\tsubslice%d: %u EUs (0x%hx)\n", 171 ss, hweight16(enabled_eus), enabled_eus); 172 } 173 } 174 } 175 176 static u16 compute_eu_total(const struct sseu_dev_info *sseu) 177 { 178 u16 i, total = 0; 179 180 for (i = 0; i < ARRAY_SIZE(sseu->eu_mask); i++) 181 total += hweight8(sseu->eu_mask[i]); 182 183 return total; 184 } 185 186 static void gen11_sseu_info_init(struct drm_i915_private *dev_priv) 187 { 188 struct sseu_dev_info *sseu = &RUNTIME_INFO(dev_priv)->sseu; 189 u8 s_en; 190 u32 ss_en, ss_en_mask; 191 u8 eu_en; 192 int s; 193 194 if (IS_ELKHARTLAKE(dev_priv)) { 195 sseu->max_slices = 1; 196 sseu->max_subslices = 4; 197 sseu->max_eus_per_subslice = 8; 198 } else { 199 sseu->max_slices = 1; 200 sseu->max_subslices = 8; 201 sseu->max_eus_per_subslice = 8; 202 } 203 204 s_en = I915_READ(GEN11_GT_SLICE_ENABLE) & GEN11_GT_S_ENA_MASK; 205 ss_en = ~I915_READ(GEN11_GT_SUBSLICE_DISABLE); 206 ss_en_mask = BIT(sseu->max_subslices) - 1; 207 eu_en = ~(I915_READ(GEN11_EU_DISABLE) & GEN11_EU_DIS_MASK); 208 209 for (s = 0; s < sseu->max_slices; s++) { 210 if (s_en & BIT(s)) { 211 int ss_idx = sseu->max_subslices * s; 212 int ss; 213 214 sseu->slice_mask |= BIT(s); 215 sseu->subslice_mask[s] = (ss_en >> ss_idx) & ss_en_mask; 216 for (ss = 0; ss < sseu->max_subslices; ss++) { 217 if (sseu->subslice_mask[s] & BIT(ss)) 218 sseu_set_eus(sseu, s, ss, eu_en); 219 } 220 } 221 } 222 sseu->eu_per_subslice = hweight8(eu_en); 223 sseu->eu_total = compute_eu_total(sseu); 224 225 /* ICL has no power gating restrictions. */ 226 sseu->has_slice_pg = 1; 227 sseu->has_subslice_pg = 1; 228 sseu->has_eu_pg = 1; 229 } 230 231 static void gen10_sseu_info_init(struct drm_i915_private *dev_priv) 232 { 233 struct sseu_dev_info *sseu = &RUNTIME_INFO(dev_priv)->sseu; 234 const u32 fuse2 = I915_READ(GEN8_FUSE2); 235 int s, ss; 236 const int eu_mask = 0xff; 237 u32 subslice_mask, eu_en; 238 239 sseu->slice_mask = (fuse2 & GEN10_F2_S_ENA_MASK) >> 240 GEN10_F2_S_ENA_SHIFT; 241 sseu->max_slices = 6; 242 sseu->max_subslices = 4; 243 sseu->max_eus_per_subslice = 8; 244 245 subslice_mask = (1 << 4) - 1; 246 subslice_mask &= ~((fuse2 & GEN10_F2_SS_DIS_MASK) >> 247 GEN10_F2_SS_DIS_SHIFT); 248 249 /* 250 * Slice0 can have up to 3 subslices, but there are only 2 in 251 * slice1/2. 252 */ 253 sseu->subslice_mask[0] = subslice_mask; 254 for (s = 1; s < sseu->max_slices; s++) 255 sseu->subslice_mask[s] = subslice_mask & 0x3; 256 257 /* Slice0 */ 258 eu_en = ~I915_READ(GEN8_EU_DISABLE0); 259 for (ss = 0; ss < sseu->max_subslices; ss++) 260 sseu_set_eus(sseu, 0, ss, (eu_en >> (8 * ss)) & eu_mask); 261 /* Slice1 */ 262 sseu_set_eus(sseu, 1, 0, (eu_en >> 24) & eu_mask); 263 eu_en = ~I915_READ(GEN8_EU_DISABLE1); 264 sseu_set_eus(sseu, 1, 1, eu_en & eu_mask); 265 /* Slice2 */ 266 sseu_set_eus(sseu, 2, 0, (eu_en >> 8) & eu_mask); 267 sseu_set_eus(sseu, 2, 1, (eu_en >> 16) & eu_mask); 268 /* Slice3 */ 269 sseu_set_eus(sseu, 3, 0, (eu_en >> 24) & eu_mask); 270 eu_en = ~I915_READ(GEN8_EU_DISABLE2); 271 sseu_set_eus(sseu, 3, 1, eu_en & eu_mask); 272 /* Slice4 */ 273 sseu_set_eus(sseu, 4, 0, (eu_en >> 8) & eu_mask); 274 sseu_set_eus(sseu, 4, 1, (eu_en >> 16) & eu_mask); 275 /* Slice5 */ 276 sseu_set_eus(sseu, 5, 0, (eu_en >> 24) & eu_mask); 277 eu_en = ~I915_READ(GEN10_EU_DISABLE3); 278 sseu_set_eus(sseu, 5, 1, eu_en & eu_mask); 279 280 /* Do a second pass where we mark the subslices disabled if all their 281 * eus are off. 282 */ 283 for (s = 0; s < sseu->max_slices; s++) { 284 for (ss = 0; ss < sseu->max_subslices; ss++) { 285 if (sseu_get_eus(sseu, s, ss) == 0) 286 sseu->subslice_mask[s] &= ~BIT(ss); 287 } 288 } 289 290 sseu->eu_total = compute_eu_total(sseu); 291 292 /* 293 * CNL is expected to always have a uniform distribution 294 * of EU across subslices with the exception that any one 295 * EU in any one subslice may be fused off for die 296 * recovery. 297 */ 298 sseu->eu_per_subslice = intel_sseu_subslice_total(sseu) ? 299 DIV_ROUND_UP(sseu->eu_total, 300 intel_sseu_subslice_total(sseu)) : 301 0; 302 303 /* No restrictions on Power Gating */ 304 sseu->has_slice_pg = 1; 305 sseu->has_subslice_pg = 1; 306 sseu->has_eu_pg = 1; 307 } 308 309 static void cherryview_sseu_info_init(struct drm_i915_private *dev_priv) 310 { 311 struct sseu_dev_info *sseu = &RUNTIME_INFO(dev_priv)->sseu; 312 u32 fuse; 313 314 fuse = I915_READ(CHV_FUSE_GT); 315 316 sseu->slice_mask = BIT(0); 317 sseu->max_slices = 1; 318 sseu->max_subslices = 2; 319 sseu->max_eus_per_subslice = 8; 320 321 if (!(fuse & CHV_FGT_DISABLE_SS0)) { 322 u8 disabled_mask = 323 ((fuse & CHV_FGT_EU_DIS_SS0_R0_MASK) >> 324 CHV_FGT_EU_DIS_SS0_R0_SHIFT) | 325 (((fuse & CHV_FGT_EU_DIS_SS0_R1_MASK) >> 326 CHV_FGT_EU_DIS_SS0_R1_SHIFT) << 4); 327 328 sseu->subslice_mask[0] |= BIT(0); 329 sseu_set_eus(sseu, 0, 0, ~disabled_mask); 330 } 331 332 if (!(fuse & CHV_FGT_DISABLE_SS1)) { 333 u8 disabled_mask = 334 ((fuse & CHV_FGT_EU_DIS_SS1_R0_MASK) >> 335 CHV_FGT_EU_DIS_SS1_R0_SHIFT) | 336 (((fuse & CHV_FGT_EU_DIS_SS1_R1_MASK) >> 337 CHV_FGT_EU_DIS_SS1_R1_SHIFT) << 4); 338 339 sseu->subslice_mask[0] |= BIT(1); 340 sseu_set_eus(sseu, 0, 1, ~disabled_mask); 341 } 342 343 sseu->eu_total = compute_eu_total(sseu); 344 345 /* 346 * CHV expected to always have a uniform distribution of EU 347 * across subslices. 348 */ 349 sseu->eu_per_subslice = intel_sseu_subslice_total(sseu) ? 350 sseu->eu_total / 351 intel_sseu_subslice_total(sseu) : 352 0; 353 /* 354 * CHV supports subslice power gating on devices with more than 355 * one subslice, and supports EU power gating on devices with 356 * more than one EU pair per subslice. 357 */ 358 sseu->has_slice_pg = 0; 359 sseu->has_subslice_pg = intel_sseu_subslice_total(sseu) > 1; 360 sseu->has_eu_pg = (sseu->eu_per_subslice > 2); 361 } 362 363 static void gen9_sseu_info_init(struct drm_i915_private *dev_priv) 364 { 365 struct intel_device_info *info = mkwrite_device_info(dev_priv); 366 struct sseu_dev_info *sseu = &RUNTIME_INFO(dev_priv)->sseu; 367 int s, ss; 368 u32 fuse2, eu_disable, subslice_mask; 369 const u8 eu_mask = 0xff; 370 371 fuse2 = I915_READ(GEN8_FUSE2); 372 sseu->slice_mask = (fuse2 & GEN8_F2_S_ENA_MASK) >> GEN8_F2_S_ENA_SHIFT; 373 374 /* BXT has a single slice and at most 3 subslices. */ 375 sseu->max_slices = IS_GEN9_LP(dev_priv) ? 1 : 3; 376 sseu->max_subslices = IS_GEN9_LP(dev_priv) ? 3 : 4; 377 sseu->max_eus_per_subslice = 8; 378 379 /* 380 * The subslice disable field is global, i.e. it applies 381 * to each of the enabled slices. 382 */ 383 subslice_mask = (1 << sseu->max_subslices) - 1; 384 subslice_mask &= ~((fuse2 & GEN9_F2_SS_DIS_MASK) >> 385 GEN9_F2_SS_DIS_SHIFT); 386 387 /* 388 * Iterate through enabled slices and subslices to 389 * count the total enabled EU. 390 */ 391 for (s = 0; s < sseu->max_slices; s++) { 392 if (!(sseu->slice_mask & BIT(s))) 393 /* skip disabled slice */ 394 continue; 395 396 sseu->subslice_mask[s] = subslice_mask; 397 398 eu_disable = I915_READ(GEN9_EU_DISABLE(s)); 399 for (ss = 0; ss < sseu->max_subslices; ss++) { 400 int eu_per_ss; 401 u8 eu_disabled_mask; 402 403 if (!(sseu->subslice_mask[s] & BIT(ss))) 404 /* skip disabled subslice */ 405 continue; 406 407 eu_disabled_mask = (eu_disable >> (ss * 8)) & eu_mask; 408 409 sseu_set_eus(sseu, s, ss, ~eu_disabled_mask); 410 411 eu_per_ss = sseu->max_eus_per_subslice - 412 hweight8(eu_disabled_mask); 413 414 /* 415 * Record which subslice(s) has(have) 7 EUs. we 416 * can tune the hash used to spread work among 417 * subslices if they are unbalanced. 418 */ 419 if (eu_per_ss == 7) 420 sseu->subslice_7eu[s] |= BIT(ss); 421 } 422 } 423 424 sseu->eu_total = compute_eu_total(sseu); 425 426 /* 427 * SKL is expected to always have a uniform distribution 428 * of EU across subslices with the exception that any one 429 * EU in any one subslice may be fused off for die 430 * recovery. BXT is expected to be perfectly uniform in EU 431 * distribution. 432 */ 433 sseu->eu_per_subslice = intel_sseu_subslice_total(sseu) ? 434 DIV_ROUND_UP(sseu->eu_total, 435 intel_sseu_subslice_total(sseu)) : 436 0; 437 /* 438 * SKL+ supports slice power gating on devices with more than 439 * one slice, and supports EU power gating on devices with 440 * more than one EU pair per subslice. BXT+ supports subslice 441 * power gating on devices with more than one subslice, and 442 * supports EU power gating on devices with more than one EU 443 * pair per subslice. 444 */ 445 sseu->has_slice_pg = 446 !IS_GEN9_LP(dev_priv) && hweight8(sseu->slice_mask) > 1; 447 sseu->has_subslice_pg = 448 IS_GEN9_LP(dev_priv) && intel_sseu_subslice_total(sseu) > 1; 449 sseu->has_eu_pg = sseu->eu_per_subslice > 2; 450 451 if (IS_GEN9_LP(dev_priv)) { 452 #define IS_SS_DISABLED(ss) (!(sseu->subslice_mask[0] & BIT(ss))) 453 info->has_pooled_eu = hweight8(sseu->subslice_mask[0]) == 3; 454 455 sseu->min_eu_in_pool = 0; 456 if (info->has_pooled_eu) { 457 if (IS_SS_DISABLED(2) || IS_SS_DISABLED(0)) 458 sseu->min_eu_in_pool = 3; 459 else if (IS_SS_DISABLED(1)) 460 sseu->min_eu_in_pool = 6; 461 else 462 sseu->min_eu_in_pool = 9; 463 } 464 #undef IS_SS_DISABLED 465 } 466 } 467 468 static void broadwell_sseu_info_init(struct drm_i915_private *dev_priv) 469 { 470 struct sseu_dev_info *sseu = &RUNTIME_INFO(dev_priv)->sseu; 471 int s, ss; 472 u32 fuse2, subslice_mask, eu_disable[3]; /* s_max */ 473 474 fuse2 = I915_READ(GEN8_FUSE2); 475 sseu->slice_mask = (fuse2 & GEN8_F2_S_ENA_MASK) >> GEN8_F2_S_ENA_SHIFT; 476 sseu->max_slices = 3; 477 sseu->max_subslices = 3; 478 sseu->max_eus_per_subslice = 8; 479 480 /* 481 * The subslice disable field is global, i.e. it applies 482 * to each of the enabled slices. 483 */ 484 subslice_mask = GENMASK(sseu->max_subslices - 1, 0); 485 subslice_mask &= ~((fuse2 & GEN8_F2_SS_DIS_MASK) >> 486 GEN8_F2_SS_DIS_SHIFT); 487 488 eu_disable[0] = I915_READ(GEN8_EU_DISABLE0) & GEN8_EU_DIS0_S0_MASK; 489 eu_disable[1] = (I915_READ(GEN8_EU_DISABLE0) >> GEN8_EU_DIS0_S1_SHIFT) | 490 ((I915_READ(GEN8_EU_DISABLE1) & GEN8_EU_DIS1_S1_MASK) << 491 (32 - GEN8_EU_DIS0_S1_SHIFT)); 492 eu_disable[2] = (I915_READ(GEN8_EU_DISABLE1) >> GEN8_EU_DIS1_S2_SHIFT) | 493 ((I915_READ(GEN8_EU_DISABLE2) & GEN8_EU_DIS2_S2_MASK) << 494 (32 - GEN8_EU_DIS1_S2_SHIFT)); 495 496 /* 497 * Iterate through enabled slices and subslices to 498 * count the total enabled EU. 499 */ 500 for (s = 0; s < sseu->max_slices; s++) { 501 if (!(sseu->slice_mask & BIT(s))) 502 /* skip disabled slice */ 503 continue; 504 505 sseu->subslice_mask[s] = subslice_mask; 506 507 for (ss = 0; ss < sseu->max_subslices; ss++) { 508 u8 eu_disabled_mask; 509 u32 n_disabled; 510 511 if (!(sseu->subslice_mask[s] & BIT(ss))) 512 /* skip disabled subslice */ 513 continue; 514 515 eu_disabled_mask = 516 eu_disable[s] >> (ss * sseu->max_eus_per_subslice); 517 518 sseu_set_eus(sseu, s, ss, ~eu_disabled_mask); 519 520 n_disabled = hweight8(eu_disabled_mask); 521 522 /* 523 * Record which subslices have 7 EUs. 524 */ 525 if (sseu->max_eus_per_subslice - n_disabled == 7) 526 sseu->subslice_7eu[s] |= 1 << ss; 527 } 528 } 529 530 sseu->eu_total = compute_eu_total(sseu); 531 532 /* 533 * BDW is expected to always have a uniform distribution of EU across 534 * subslices with the exception that any one EU in any one subslice may 535 * be fused off for die recovery. 536 */ 537 sseu->eu_per_subslice = intel_sseu_subslice_total(sseu) ? 538 DIV_ROUND_UP(sseu->eu_total, 539 intel_sseu_subslice_total(sseu)) : 540 0; 541 542 /* 543 * BDW supports slice power gating on devices with more than 544 * one slice. 545 */ 546 sseu->has_slice_pg = hweight8(sseu->slice_mask) > 1; 547 sseu->has_subslice_pg = 0; 548 sseu->has_eu_pg = 0; 549 } 550 551 static void haswell_sseu_info_init(struct drm_i915_private *dev_priv) 552 { 553 struct sseu_dev_info *sseu = &RUNTIME_INFO(dev_priv)->sseu; 554 u32 fuse1; 555 int s, ss; 556 557 /* 558 * There isn't a register to tell us how many slices/subslices. We 559 * work off the PCI-ids here. 560 */ 561 switch (INTEL_INFO(dev_priv)->gt) { 562 default: 563 MISSING_CASE(INTEL_INFO(dev_priv)->gt); 564 /* fall through */ 565 case 1: 566 sseu->slice_mask = BIT(0); 567 sseu->subslice_mask[0] = BIT(0); 568 break; 569 case 2: 570 sseu->slice_mask = BIT(0); 571 sseu->subslice_mask[0] = BIT(0) | BIT(1); 572 break; 573 case 3: 574 sseu->slice_mask = BIT(0) | BIT(1); 575 sseu->subslice_mask[0] = BIT(0) | BIT(1); 576 sseu->subslice_mask[1] = BIT(0) | BIT(1); 577 break; 578 } 579 580 sseu->max_slices = hweight8(sseu->slice_mask); 581 sseu->max_subslices = hweight8(sseu->subslice_mask[0]); 582 583 fuse1 = I915_READ(HSW_PAVP_FUSE1); 584 switch ((fuse1 & HSW_F1_EU_DIS_MASK) >> HSW_F1_EU_DIS_SHIFT) { 585 default: 586 MISSING_CASE((fuse1 & HSW_F1_EU_DIS_MASK) >> 587 HSW_F1_EU_DIS_SHIFT); 588 /* fall through */ 589 case HSW_F1_EU_DIS_10EUS: 590 sseu->eu_per_subslice = 10; 591 break; 592 case HSW_F1_EU_DIS_8EUS: 593 sseu->eu_per_subslice = 8; 594 break; 595 case HSW_F1_EU_DIS_6EUS: 596 sseu->eu_per_subslice = 6; 597 break; 598 } 599 sseu->max_eus_per_subslice = sseu->eu_per_subslice; 600 601 for (s = 0; s < sseu->max_slices; s++) { 602 for (ss = 0; ss < sseu->max_subslices; ss++) { 603 sseu_set_eus(sseu, s, ss, 604 (1UL << sseu->eu_per_subslice) - 1); 605 } 606 } 607 608 sseu->eu_total = compute_eu_total(sseu); 609 610 /* No powergating for you. */ 611 sseu->has_slice_pg = 0; 612 sseu->has_subslice_pg = 0; 613 sseu->has_eu_pg = 0; 614 } 615 616 static u32 read_reference_ts_freq(struct drm_i915_private *dev_priv) 617 { 618 u32 ts_override = I915_READ(GEN9_TIMESTAMP_OVERRIDE); 619 u32 base_freq, frac_freq; 620 621 base_freq = ((ts_override & GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DIVIDER_MASK) >> 622 GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DIVIDER_SHIFT) + 1; 623 base_freq *= 1000; 624 625 frac_freq = ((ts_override & 626 GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DENOMINATOR_MASK) >> 627 GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DENOMINATOR_SHIFT); 628 frac_freq = 1000 / (frac_freq + 1); 629 630 return base_freq + frac_freq; 631 } 632 633 static u32 gen10_get_crystal_clock_freq(struct drm_i915_private *dev_priv, 634 u32 rpm_config_reg) 635 { 636 u32 f19_2_mhz = 19200; 637 u32 f24_mhz = 24000; 638 u32 crystal_clock = (rpm_config_reg & 639 GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_MASK) >> 640 GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_SHIFT; 641 642 switch (crystal_clock) { 643 case GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_19_2_MHZ: 644 return f19_2_mhz; 645 case GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_24_MHZ: 646 return f24_mhz; 647 default: 648 MISSING_CASE(crystal_clock); 649 return 0; 650 } 651 } 652 653 static u32 gen11_get_crystal_clock_freq(struct drm_i915_private *dev_priv, 654 u32 rpm_config_reg) 655 { 656 u32 f19_2_mhz = 19200; 657 u32 f24_mhz = 24000; 658 u32 f25_mhz = 25000; 659 u32 f38_4_mhz = 38400; 660 u32 crystal_clock = (rpm_config_reg & 661 GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_MASK) >> 662 GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_SHIFT; 663 664 switch (crystal_clock) { 665 case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_24_MHZ: 666 return f24_mhz; 667 case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_19_2_MHZ: 668 return f19_2_mhz; 669 case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_38_4_MHZ: 670 return f38_4_mhz; 671 case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_25_MHZ: 672 return f25_mhz; 673 default: 674 MISSING_CASE(crystal_clock); 675 return 0; 676 } 677 } 678 679 static u32 read_timestamp_frequency(struct drm_i915_private *dev_priv) 680 { 681 u32 f12_5_mhz = 12500; 682 u32 f19_2_mhz = 19200; 683 u32 f24_mhz = 24000; 684 685 if (INTEL_GEN(dev_priv) <= 4) { 686 /* PRMs say: 687 * 688 * "The value in this register increments once every 16 689 * hclks." (through the “Clocking Configuration” 690 * (“CLKCFG”) MCHBAR register) 691 */ 692 return dev_priv->rawclk_freq / 16; 693 } else if (INTEL_GEN(dev_priv) <= 8) { 694 /* PRMs say: 695 * 696 * "The PCU TSC counts 10ns increments; this timestamp 697 * reflects bits 38:3 of the TSC (i.e. 80ns granularity, 698 * rolling over every 1.5 hours). 699 */ 700 return f12_5_mhz; 701 } else if (INTEL_GEN(dev_priv) <= 9) { 702 u32 ctc_reg = I915_READ(CTC_MODE); 703 u32 freq = 0; 704 705 if ((ctc_reg & CTC_SOURCE_PARAMETER_MASK) == CTC_SOURCE_DIVIDE_LOGIC) { 706 freq = read_reference_ts_freq(dev_priv); 707 } else { 708 freq = IS_GEN9_LP(dev_priv) ? f19_2_mhz : f24_mhz; 709 710 /* Now figure out how the command stream's timestamp 711 * register increments from this frequency (it might 712 * increment only every few clock cycle). 713 */ 714 freq >>= 3 - ((ctc_reg & CTC_SHIFT_PARAMETER_MASK) >> 715 CTC_SHIFT_PARAMETER_SHIFT); 716 } 717 718 return freq; 719 } else if (INTEL_GEN(dev_priv) <= 11) { 720 u32 ctc_reg = I915_READ(CTC_MODE); 721 u32 freq = 0; 722 723 /* First figure out the reference frequency. There are 2 ways 724 * we can compute the frequency, either through the 725 * TIMESTAMP_OVERRIDE register or through RPM_CONFIG. CTC_MODE 726 * tells us which one we should use. 727 */ 728 if ((ctc_reg & CTC_SOURCE_PARAMETER_MASK) == CTC_SOURCE_DIVIDE_LOGIC) { 729 freq = read_reference_ts_freq(dev_priv); 730 } else { 731 u32 rpm_config_reg = I915_READ(RPM_CONFIG0); 732 733 if (INTEL_GEN(dev_priv) <= 10) 734 freq = gen10_get_crystal_clock_freq(dev_priv, 735 rpm_config_reg); 736 else 737 freq = gen11_get_crystal_clock_freq(dev_priv, 738 rpm_config_reg); 739 740 /* Now figure out how the command stream's timestamp 741 * register increments from this frequency (it might 742 * increment only every few clock cycle). 743 */ 744 freq >>= 3 - ((rpm_config_reg & 745 GEN10_RPM_CONFIG0_CTC_SHIFT_PARAMETER_MASK) >> 746 GEN10_RPM_CONFIG0_CTC_SHIFT_PARAMETER_SHIFT); 747 } 748 749 return freq; 750 } 751 752 MISSING_CASE("Unknown gen, unable to read command streamer timestamp frequency\n"); 753 return 0; 754 } 755 756 #undef INTEL_VGA_DEVICE 757 #define INTEL_VGA_DEVICE(id, info) (id) 758 759 static const u16 subplatform_ult_ids[] = { 760 INTEL_HSW_ULT_GT1_IDS(0), 761 INTEL_HSW_ULT_GT2_IDS(0), 762 INTEL_HSW_ULT_GT3_IDS(0), 763 INTEL_BDW_ULT_GT1_IDS(0), 764 INTEL_BDW_ULT_GT2_IDS(0), 765 INTEL_BDW_ULT_GT3_IDS(0), 766 INTEL_BDW_ULT_RSVD_IDS(0), 767 INTEL_SKL_ULT_GT1_IDS(0), 768 INTEL_SKL_ULT_GT2_IDS(0), 769 INTEL_SKL_ULT_GT3_IDS(0), 770 INTEL_KBL_ULT_GT1_IDS(0), 771 INTEL_KBL_ULT_GT2_IDS(0), 772 INTEL_KBL_ULT_GT3_IDS(0), 773 INTEL_CFL_U_GT2_IDS(0), 774 INTEL_CFL_U_GT3_IDS(0), 775 INTEL_WHL_U_GT1_IDS(0), 776 INTEL_WHL_U_GT2_IDS(0), 777 INTEL_WHL_U_GT3_IDS(0), 778 }; 779 780 static const u16 subplatform_ulx_ids[] = { 781 INTEL_HSW_ULX_GT1_IDS(0), 782 INTEL_HSW_ULX_GT2_IDS(0), 783 INTEL_BDW_ULX_GT1_IDS(0), 784 INTEL_BDW_ULX_GT2_IDS(0), 785 INTEL_BDW_ULX_GT3_IDS(0), 786 INTEL_BDW_ULX_RSVD_IDS(0), 787 INTEL_SKL_ULX_GT1_IDS(0), 788 INTEL_SKL_ULX_GT2_IDS(0), 789 INTEL_KBL_ULX_GT1_IDS(0), 790 INTEL_KBL_ULX_GT2_IDS(0), 791 INTEL_AML_KBL_GT2_IDS(0), 792 INTEL_AML_CFL_GT2_IDS(0), 793 }; 794 795 static const u16 subplatform_portf_ids[] = { 796 INTEL_CNL_PORT_F_IDS(0), 797 INTEL_ICL_PORT_F_IDS(0), 798 }; 799 800 static bool find_devid(u16 id, const u16 *p, unsigned int num) 801 { 802 for (; num; num--, p++) { 803 if (*p == id) 804 return true; 805 } 806 807 return false; 808 } 809 810 void intel_device_info_subplatform_init(struct drm_i915_private *i915) 811 { 812 const struct intel_device_info *info = INTEL_INFO(i915); 813 const struct intel_runtime_info *rinfo = RUNTIME_INFO(i915); 814 const unsigned int pi = __platform_mask_index(rinfo, info->platform); 815 const unsigned int pb = __platform_mask_bit(rinfo, info->platform); 816 u16 devid = INTEL_DEVID(i915); 817 u32 mask = 0; 818 819 /* Make sure IS_<platform> checks are working. */ 820 RUNTIME_INFO(i915)->platform_mask[pi] = BIT(pb); 821 822 /* Find and mark subplatform bits based on the PCI device id. */ 823 if (find_devid(devid, subplatform_ult_ids, 824 ARRAY_SIZE(subplatform_ult_ids))) { 825 mask = BIT(INTEL_SUBPLATFORM_ULT); 826 } else if (find_devid(devid, subplatform_ulx_ids, 827 ARRAY_SIZE(subplatform_ulx_ids))) { 828 mask = BIT(INTEL_SUBPLATFORM_ULX); 829 if (IS_HASWELL(i915) || IS_BROADWELL(i915)) { 830 /* ULX machines are also considered ULT. */ 831 mask |= BIT(INTEL_SUBPLATFORM_ULT); 832 } 833 } else if (find_devid(devid, subplatform_portf_ids, 834 ARRAY_SIZE(subplatform_portf_ids))) { 835 mask = BIT(INTEL_SUBPLATFORM_PORTF); 836 } 837 838 GEM_BUG_ON(mask & ~INTEL_SUBPLATFORM_BITS); 839 840 RUNTIME_INFO(i915)->platform_mask[pi] |= mask; 841 } 842 843 /** 844 * intel_device_info_runtime_init - initialize runtime info 845 * @dev_priv: the i915 device 846 * 847 * Determine various intel_device_info fields at runtime. 848 * 849 * Use it when either: 850 * - it's judged too laborious to fill n static structures with the limit 851 * when a simple if statement does the job, 852 * - run-time checks (eg read fuse/strap registers) are needed. 853 * 854 * This function needs to be called: 855 * - after the MMIO has been setup as we are reading registers, 856 * - after the PCH has been detected, 857 * - before the first usage of the fields it can tweak. 858 */ 859 void intel_device_info_runtime_init(struct drm_i915_private *dev_priv) 860 { 861 struct intel_device_info *info = mkwrite_device_info(dev_priv); 862 struct intel_runtime_info *runtime = RUNTIME_INFO(dev_priv); 863 enum pipe pipe; 864 865 if (INTEL_GEN(dev_priv) >= 10) { 866 for_each_pipe(dev_priv, pipe) 867 runtime->num_scalers[pipe] = 2; 868 } else if (IS_GEN(dev_priv, 9)) { 869 runtime->num_scalers[PIPE_A] = 2; 870 runtime->num_scalers[PIPE_B] = 2; 871 runtime->num_scalers[PIPE_C] = 1; 872 } 873 874 BUILD_BUG_ON(BITS_PER_TYPE(intel_engine_mask_t) < I915_NUM_ENGINES); 875 876 if (INTEL_GEN(dev_priv) >= 11) 877 for_each_pipe(dev_priv, pipe) 878 runtime->num_sprites[pipe] = 6; 879 else if (IS_GEN(dev_priv, 10) || IS_GEMINILAKE(dev_priv)) 880 for_each_pipe(dev_priv, pipe) 881 runtime->num_sprites[pipe] = 3; 882 else if (IS_BROXTON(dev_priv)) { 883 /* 884 * Skylake and Broxton currently don't expose the topmost plane as its 885 * use is exclusive with the legacy cursor and we only want to expose 886 * one of those, not both. Until we can safely expose the topmost plane 887 * as a DRM_PLANE_TYPE_CURSOR with all the features exposed/supported, 888 * we don't expose the topmost plane at all to prevent ABI breakage 889 * down the line. 890 */ 891 892 runtime->num_sprites[PIPE_A] = 2; 893 runtime->num_sprites[PIPE_B] = 2; 894 runtime->num_sprites[PIPE_C] = 1; 895 } else if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) { 896 for_each_pipe(dev_priv, pipe) 897 runtime->num_sprites[pipe] = 2; 898 } else if (INTEL_GEN(dev_priv) >= 5 || IS_G4X(dev_priv)) { 899 for_each_pipe(dev_priv, pipe) 900 runtime->num_sprites[pipe] = 1; 901 } 902 903 if (i915_modparams.disable_display) { 904 DRM_INFO("Display disabled (module parameter)\n"); 905 info->num_pipes = 0; 906 } else if (HAS_DISPLAY(dev_priv) && 907 (IS_GEN_RANGE(dev_priv, 7, 8)) && 908 HAS_PCH_SPLIT(dev_priv)) { 909 u32 fuse_strap = I915_READ(FUSE_STRAP); 910 u32 sfuse_strap = I915_READ(SFUSE_STRAP); 911 912 /* 913 * SFUSE_STRAP is supposed to have a bit signalling the display 914 * is fused off. Unfortunately it seems that, at least in 915 * certain cases, fused off display means that PCH display 916 * reads don't land anywhere. In that case, we read 0s. 917 * 918 * On CPT/PPT, we can detect this case as SFUSE_STRAP_FUSE_LOCK 919 * should be set when taking over after the firmware. 920 */ 921 if (fuse_strap & ILK_INTERNAL_DISPLAY_DISABLE || 922 sfuse_strap & SFUSE_STRAP_DISPLAY_DISABLED || 923 (HAS_PCH_CPT(dev_priv) && 924 !(sfuse_strap & SFUSE_STRAP_FUSE_LOCK))) { 925 DRM_INFO("Display fused off, disabling\n"); 926 info->num_pipes = 0; 927 } else if (fuse_strap & IVB_PIPE_C_DISABLE) { 928 DRM_INFO("PipeC fused off\n"); 929 info->num_pipes -= 1; 930 } 931 } else if (HAS_DISPLAY(dev_priv) && INTEL_GEN(dev_priv) >= 9) { 932 u32 dfsm = I915_READ(SKL_DFSM); 933 u8 enabled_mask = BIT(info->num_pipes) - 1; 934 935 if (dfsm & SKL_DFSM_PIPE_A_DISABLE) 936 enabled_mask &= ~BIT(PIPE_A); 937 if (dfsm & SKL_DFSM_PIPE_B_DISABLE) 938 enabled_mask &= ~BIT(PIPE_B); 939 if (dfsm & SKL_DFSM_PIPE_C_DISABLE) 940 enabled_mask &= ~BIT(PIPE_C); 941 if (INTEL_GEN(dev_priv) >= 12 && 942 (dfsm & TGL_DFSM_PIPE_D_DISABLE)) 943 enabled_mask &= ~BIT(PIPE_D); 944 945 /* 946 * At least one pipe should be enabled and if there are 947 * disabled pipes, they should be the last ones, with no holes 948 * in the mask. 949 */ 950 if (enabled_mask == 0 || !is_power_of_2(enabled_mask + 1)) 951 DRM_ERROR("invalid pipe fuse configuration: enabled_mask=0x%x\n", 952 enabled_mask); 953 else 954 info->num_pipes = hweight8(enabled_mask); 955 } 956 957 /* Initialize slice/subslice/EU info */ 958 if (IS_HASWELL(dev_priv)) 959 haswell_sseu_info_init(dev_priv); 960 else if (IS_CHERRYVIEW(dev_priv)) 961 cherryview_sseu_info_init(dev_priv); 962 else if (IS_BROADWELL(dev_priv)) 963 broadwell_sseu_info_init(dev_priv); 964 else if (IS_GEN(dev_priv, 9)) 965 gen9_sseu_info_init(dev_priv); 966 else if (IS_GEN(dev_priv, 10)) 967 gen10_sseu_info_init(dev_priv); 968 else if (INTEL_GEN(dev_priv) >= 11) 969 gen11_sseu_info_init(dev_priv); 970 971 if (IS_GEN(dev_priv, 6) && intel_vtd_active()) { 972 DRM_INFO("Disabling ppGTT for VT-d support\n"); 973 info->ppgtt_type = INTEL_PPGTT_NONE; 974 } 975 976 /* Initialize command stream timestamp frequency */ 977 runtime->cs_timestamp_frequency_khz = read_timestamp_frequency(dev_priv); 978 } 979 980 void intel_driver_caps_print(const struct intel_driver_caps *caps, 981 struct drm_printer *p) 982 { 983 drm_printf(p, "Has logical contexts? %s\n", 984 yesno(caps->has_logical_contexts)); 985 drm_printf(p, "scheduler: %x\n", caps->scheduler); 986 } 987 988 /* 989 * Determine which engines are fused off in our particular hardware. Since the 990 * fuse register is in the blitter powerwell, we need forcewake to be ready at 991 * this point (but later we need to prune the forcewake domains for engines that 992 * are indeed fused off). 993 */ 994 void intel_device_info_init_mmio(struct drm_i915_private *dev_priv) 995 { 996 struct intel_device_info *info = mkwrite_device_info(dev_priv); 997 unsigned int logical_vdbox = 0; 998 unsigned int i; 999 u32 media_fuse; 1000 u16 vdbox_mask; 1001 u16 vebox_mask; 1002 1003 if (INTEL_GEN(dev_priv) < 11) 1004 return; 1005 1006 media_fuse = ~I915_READ(GEN11_GT_VEBOX_VDBOX_DISABLE); 1007 1008 vdbox_mask = media_fuse & GEN11_GT_VDBOX_DISABLE_MASK; 1009 vebox_mask = (media_fuse & GEN11_GT_VEBOX_DISABLE_MASK) >> 1010 GEN11_GT_VEBOX_DISABLE_SHIFT; 1011 1012 for (i = 0; i < I915_MAX_VCS; i++) { 1013 if (!HAS_ENGINE(dev_priv, _VCS(i))) 1014 continue; 1015 1016 if (!(BIT(i) & vdbox_mask)) { 1017 info->engine_mask &= ~BIT(_VCS(i)); 1018 DRM_DEBUG_DRIVER("vcs%u fused off\n", i); 1019 continue; 1020 } 1021 1022 /* 1023 * In Gen11, only even numbered logical VDBOXes are 1024 * hooked up to an SFC (Scaler & Format Converter) unit. 1025 */ 1026 if (logical_vdbox++ % 2 == 0) 1027 RUNTIME_INFO(dev_priv)->vdbox_sfc_access |= BIT(i); 1028 } 1029 DRM_DEBUG_DRIVER("vdbox enable: %04x, instances: %04lx\n", 1030 vdbox_mask, VDBOX_MASK(dev_priv)); 1031 GEM_BUG_ON(vdbox_mask != VDBOX_MASK(dev_priv)); 1032 1033 for (i = 0; i < I915_MAX_VECS; i++) { 1034 if (!HAS_ENGINE(dev_priv, _VECS(i))) 1035 continue; 1036 1037 if (!(BIT(i) & vebox_mask)) { 1038 info->engine_mask &= ~BIT(_VECS(i)); 1039 DRM_DEBUG_DRIVER("vecs%u fused off\n", i); 1040 } 1041 } 1042 DRM_DEBUG_DRIVER("vebox enable: %04x, instances: %04lx\n", 1043 vebox_mask, VEBOX_MASK(dev_priv)); 1044 GEM_BUG_ON(vebox_mask != VEBOX_MASK(dev_priv)); 1045 } 1046