1 /* 2 * Copyright (c) 2015 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 * The above copyright notice and this permission notice (including the next 11 * paragraph) shall be included in all copies or substantial portions of the 12 * Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 18 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 19 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 20 * SOFTWARE. 21 */ 22 23 #include "i915_drv.h" 24 25 #include "intel_engine.h" 26 #include "intel_gt.h" 27 #include "intel_mocs.h" 28 #include "intel_lrc.h" 29 #include "intel_ring.h" 30 31 /* structures required */ 32 struct drm_i915_mocs_entry { 33 u32 control_value; 34 u16 l3cc_value; 35 u16 used; 36 }; 37 38 struct drm_i915_mocs_table { 39 unsigned int size; 40 unsigned int n_entries; 41 const struct drm_i915_mocs_entry *table; 42 }; 43 44 /* Defines for the tables (XXX_MOCS_0 - XXX_MOCS_63) */ 45 #define _LE_CACHEABILITY(value) ((value) << 0) 46 #define _LE_TGT_CACHE(value) ((value) << 2) 47 #define LE_LRUM(value) ((value) << 4) 48 #define LE_AOM(value) ((value) << 6) 49 #define LE_RSC(value) ((value) << 7) 50 #define LE_SCC(value) ((value) << 8) 51 #define LE_PFM(value) ((value) << 11) 52 #define LE_SCF(value) ((value) << 14) 53 #define LE_COS(value) ((value) << 15) 54 #define LE_SSE(value) ((value) << 17) 55 56 /* Defines for the tables (LNCFMOCS0 - LNCFMOCS31) - two entries per word */ 57 #define L3_ESC(value) ((value) << 0) 58 #define L3_SCC(value) ((value) << 1) 59 #define _L3_CACHEABILITY(value) ((value) << 4) 60 61 /* Helper defines */ 62 #define GEN9_NUM_MOCS_ENTRIES 62 /* 62 out of 64 - 63 & 64 are reserved. */ 63 #define GEN11_NUM_MOCS_ENTRIES 64 /* 63-64 are reserved, but configured. */ 64 65 /* (e)LLC caching options */ 66 /* 67 * Note: LE_0_PAGETABLE works only up to Gen11; for newer gens it means 68 * the same as LE_UC 69 */ 70 #define LE_0_PAGETABLE _LE_CACHEABILITY(0) 71 #define LE_1_UC _LE_CACHEABILITY(1) 72 #define LE_2_WT _LE_CACHEABILITY(2) 73 #define LE_3_WB _LE_CACHEABILITY(3) 74 75 /* Target cache */ 76 #define LE_TC_0_PAGETABLE _LE_TGT_CACHE(0) 77 #define LE_TC_1_LLC _LE_TGT_CACHE(1) 78 #define LE_TC_2_LLC_ELLC _LE_TGT_CACHE(2) 79 #define LE_TC_3_LLC_ELLC_ALT _LE_TGT_CACHE(3) 80 81 /* L3 caching options */ 82 #define L3_0_DIRECT _L3_CACHEABILITY(0) 83 #define L3_1_UC _L3_CACHEABILITY(1) 84 #define L3_2_RESERVED _L3_CACHEABILITY(2) 85 #define L3_3_WB _L3_CACHEABILITY(3) 86 87 #define MOCS_ENTRY(__idx, __control_value, __l3cc_value) \ 88 [__idx] = { \ 89 .control_value = __control_value, \ 90 .l3cc_value = __l3cc_value, \ 91 .used = 1, \ 92 } 93 94 /* 95 * MOCS tables 96 * 97 * These are the MOCS tables that are programmed across all the rings. 98 * The control value is programmed to all the rings that support the 99 * MOCS registers. While the l3cc_values are only programmed to the 100 * LNCFCMOCS0 - LNCFCMOCS32 registers. 101 * 102 * These tables are intended to be kept reasonably consistent across 103 * HW platforms, and for ICL+, be identical across OSes. To achieve 104 * that, for Icelake and above, list of entries is published as part 105 * of bspec. 106 * 107 * Entries not part of the following tables are undefined as far as 108 * userspace is concerned and shouldn't be relied upon. For Gen < 12 109 * they will be initialized to PTE. Gen >= 12 onwards don't have a setting for 110 * PTE and will be initialized to an invalid value. 111 * 112 * The last few entries are reserved by the hardware. For ICL+ they 113 * should be initialized according to bspec and never used, for older 114 * platforms they should never be written to. 115 * 116 * NOTE: These tables are part of bspec and defined as part of hardware 117 * interface for ICL+. For older platforms, they are part of kernel 118 * ABI. It is expected that, for specific hardware platform, existing 119 * entries will remain constant and the table will only be updated by 120 * adding new entries, filling unused positions. 121 */ 122 #define GEN9_MOCS_ENTRIES \ 123 MOCS_ENTRY(I915_MOCS_UNCACHED, \ 124 LE_1_UC | LE_TC_2_LLC_ELLC, \ 125 L3_1_UC), \ 126 MOCS_ENTRY(I915_MOCS_PTE, \ 127 LE_0_PAGETABLE | LE_TC_2_LLC_ELLC | LE_LRUM(3), \ 128 L3_3_WB) 129 130 static const struct drm_i915_mocs_entry skl_mocs_table[] = { 131 GEN9_MOCS_ENTRIES, 132 MOCS_ENTRY(I915_MOCS_CACHED, 133 LE_3_WB | LE_TC_2_LLC_ELLC | LE_LRUM(3), 134 L3_3_WB) 135 }; 136 137 /* NOTE: the LE_TGT_CACHE is not used on Broxton */ 138 static const struct drm_i915_mocs_entry broxton_mocs_table[] = { 139 GEN9_MOCS_ENTRIES, 140 MOCS_ENTRY(I915_MOCS_CACHED, 141 LE_1_UC | LE_TC_2_LLC_ELLC | LE_LRUM(3), 142 L3_3_WB) 143 }; 144 145 #define GEN11_MOCS_ENTRIES \ 146 /* Entries 0 and 1 are defined per-platform */ \ 147 /* Base - L3 + LLC */ \ 148 MOCS_ENTRY(2, \ 149 LE_3_WB | LE_TC_1_LLC | LE_LRUM(3), \ 150 L3_3_WB), \ 151 /* Base - Uncached */ \ 152 MOCS_ENTRY(3, \ 153 LE_1_UC | LE_TC_1_LLC, \ 154 L3_1_UC), \ 155 /* Base - L3 */ \ 156 MOCS_ENTRY(4, \ 157 LE_1_UC | LE_TC_1_LLC, \ 158 L3_3_WB), \ 159 /* Base - LLC */ \ 160 MOCS_ENTRY(5, \ 161 LE_3_WB | LE_TC_1_LLC | LE_LRUM(3), \ 162 L3_1_UC), \ 163 /* Age 0 - LLC */ \ 164 MOCS_ENTRY(6, \ 165 LE_3_WB | LE_TC_1_LLC | LE_LRUM(1), \ 166 L3_1_UC), \ 167 /* Age 0 - L3 + LLC */ \ 168 MOCS_ENTRY(7, \ 169 LE_3_WB | LE_TC_1_LLC | LE_LRUM(1), \ 170 L3_3_WB), \ 171 /* Age: Don't Chg. - LLC */ \ 172 MOCS_ENTRY(8, \ 173 LE_3_WB | LE_TC_1_LLC | LE_LRUM(2), \ 174 L3_1_UC), \ 175 /* Age: Don't Chg. - L3 + LLC */ \ 176 MOCS_ENTRY(9, \ 177 LE_3_WB | LE_TC_1_LLC | LE_LRUM(2), \ 178 L3_3_WB), \ 179 /* No AOM - LLC */ \ 180 MOCS_ENTRY(10, \ 181 LE_3_WB | LE_TC_1_LLC | LE_LRUM(3) | LE_AOM(1), \ 182 L3_1_UC), \ 183 /* No AOM - L3 + LLC */ \ 184 MOCS_ENTRY(11, \ 185 LE_3_WB | LE_TC_1_LLC | LE_LRUM(3) | LE_AOM(1), \ 186 L3_3_WB), \ 187 /* No AOM; Age 0 - LLC */ \ 188 MOCS_ENTRY(12, \ 189 LE_3_WB | LE_TC_1_LLC | LE_LRUM(1) | LE_AOM(1), \ 190 L3_1_UC), \ 191 /* No AOM; Age 0 - L3 + LLC */ \ 192 MOCS_ENTRY(13, \ 193 LE_3_WB | LE_TC_1_LLC | LE_LRUM(1) | LE_AOM(1), \ 194 L3_3_WB), \ 195 /* No AOM; Age:DC - LLC */ \ 196 MOCS_ENTRY(14, \ 197 LE_3_WB | LE_TC_1_LLC | LE_LRUM(2) | LE_AOM(1), \ 198 L3_1_UC), \ 199 /* No AOM; Age:DC - L3 + LLC */ \ 200 MOCS_ENTRY(15, \ 201 LE_3_WB | LE_TC_1_LLC | LE_LRUM(2) | LE_AOM(1), \ 202 L3_3_WB), \ 203 /* Self-Snoop - L3 + LLC */ \ 204 MOCS_ENTRY(18, \ 205 LE_3_WB | LE_TC_1_LLC | LE_LRUM(3) | LE_SSE(3), \ 206 L3_3_WB), \ 207 /* Skip Caching - L3 + LLC(12.5%) */ \ 208 MOCS_ENTRY(19, \ 209 LE_3_WB | LE_TC_1_LLC | LE_LRUM(3) | LE_SCC(7), \ 210 L3_3_WB), \ 211 /* Skip Caching - L3 + LLC(25%) */ \ 212 MOCS_ENTRY(20, \ 213 LE_3_WB | LE_TC_1_LLC | LE_LRUM(3) | LE_SCC(3), \ 214 L3_3_WB), \ 215 /* Skip Caching - L3 + LLC(50%) */ \ 216 MOCS_ENTRY(21, \ 217 LE_3_WB | LE_TC_1_LLC | LE_LRUM(3) | LE_SCC(1), \ 218 L3_3_WB), \ 219 /* Skip Caching - L3 + LLC(75%) */ \ 220 MOCS_ENTRY(22, \ 221 LE_3_WB | LE_TC_1_LLC | LE_LRUM(3) | LE_RSC(1) | LE_SCC(3), \ 222 L3_3_WB), \ 223 /* Skip Caching - L3 + LLC(87.5%) */ \ 224 MOCS_ENTRY(23, \ 225 LE_3_WB | LE_TC_1_LLC | LE_LRUM(3) | LE_RSC(1) | LE_SCC(7), \ 226 L3_3_WB), \ 227 /* HW Reserved - SW program but never use */ \ 228 MOCS_ENTRY(62, \ 229 LE_3_WB | LE_TC_1_LLC | LE_LRUM(3), \ 230 L3_1_UC), \ 231 /* HW Reserved - SW program but never use */ \ 232 MOCS_ENTRY(63, \ 233 LE_3_WB | LE_TC_1_LLC | LE_LRUM(3), \ 234 L3_1_UC) 235 236 static const struct drm_i915_mocs_entry tgl_mocs_table[] = { 237 /* 238 * NOTE: 239 * Reserved and unspecified MOCS indices have been set to (L3 + LCC). 240 * These reserved entries should never be used, they may be changed 241 * to low performant variants with better coherency in the future if 242 * more entries are needed. We are programming index I915_MOCS_PTE(1) 243 * only, __init_mocs_table() take care to program unused index with 244 * this entry. 245 */ 246 MOCS_ENTRY(1, LE_3_WB | LE_TC_1_LLC | LE_LRUM(3), 247 L3_3_WB), 248 GEN11_MOCS_ENTRIES, 249 250 /* Implicitly enable L1 - HDC:L1 + L3 + LLC */ 251 MOCS_ENTRY(48, 252 LE_3_WB | LE_TC_1_LLC | LE_LRUM(3), 253 L3_3_WB), 254 /* Implicitly enable L1 - HDC:L1 + L3 */ 255 MOCS_ENTRY(49, 256 LE_1_UC | LE_TC_1_LLC, 257 L3_3_WB), 258 /* Implicitly enable L1 - HDC:L1 + LLC */ 259 MOCS_ENTRY(50, 260 LE_3_WB | LE_TC_1_LLC | LE_LRUM(3), 261 L3_1_UC), 262 /* Implicitly enable L1 - HDC:L1 */ 263 MOCS_ENTRY(51, 264 LE_1_UC | LE_TC_1_LLC, 265 L3_1_UC), 266 /* HW Special Case (CCS) */ 267 MOCS_ENTRY(60, 268 LE_3_WB | LE_TC_1_LLC | LE_LRUM(3), 269 L3_1_UC), 270 /* HW Special Case (Displayable) */ 271 MOCS_ENTRY(61, 272 LE_1_UC | LE_TC_1_LLC, 273 L3_3_WB), 274 }; 275 276 static const struct drm_i915_mocs_entry icl_mocs_table[] = { 277 /* Base - Uncached (Deprecated) */ 278 MOCS_ENTRY(I915_MOCS_UNCACHED, 279 LE_1_UC | LE_TC_1_LLC, 280 L3_1_UC), 281 /* Base - L3 + LeCC:PAT (Deprecated) */ 282 MOCS_ENTRY(I915_MOCS_PTE, 283 LE_0_PAGETABLE | LE_TC_1_LLC, 284 L3_3_WB), 285 286 GEN11_MOCS_ENTRIES 287 }; 288 289 static const struct drm_i915_mocs_entry dg1_mocs_table[] = { 290 /* Error */ 291 MOCS_ENTRY(0, 0, L3_0_DIRECT), 292 293 /* UC */ 294 MOCS_ENTRY(1, 0, L3_1_UC), 295 296 /* Reserved */ 297 MOCS_ENTRY(2, 0, L3_0_DIRECT), 298 MOCS_ENTRY(3, 0, L3_0_DIRECT), 299 MOCS_ENTRY(4, 0, L3_0_DIRECT), 300 301 /* WB - L3 */ 302 MOCS_ENTRY(5, 0, L3_3_WB), 303 /* WB - L3 50% */ 304 MOCS_ENTRY(6, 0, L3_ESC(1) | L3_SCC(1) | L3_3_WB), 305 /* WB - L3 25% */ 306 MOCS_ENTRY(7, 0, L3_ESC(1) | L3_SCC(3) | L3_3_WB), 307 /* WB - L3 12.5% */ 308 MOCS_ENTRY(8, 0, L3_ESC(1) | L3_SCC(7) | L3_3_WB), 309 310 /* HDC:L1 + L3 */ 311 MOCS_ENTRY(48, 0, L3_3_WB), 312 /* HDC:L1 */ 313 MOCS_ENTRY(49, 0, L3_1_UC), 314 315 /* HW Reserved */ 316 MOCS_ENTRY(60, 0, L3_1_UC), 317 MOCS_ENTRY(61, 0, L3_1_UC), 318 MOCS_ENTRY(62, 0, L3_1_UC), 319 MOCS_ENTRY(63, 0, L3_1_UC), 320 }; 321 322 enum { 323 HAS_GLOBAL_MOCS = BIT(0), 324 HAS_ENGINE_MOCS = BIT(1), 325 HAS_RENDER_L3CC = BIT(2), 326 }; 327 328 static bool has_l3cc(const struct drm_i915_private *i915) 329 { 330 return true; 331 } 332 333 static bool has_global_mocs(const struct drm_i915_private *i915) 334 { 335 return HAS_GLOBAL_MOCS_REGISTERS(i915); 336 } 337 338 static bool has_mocs(const struct drm_i915_private *i915) 339 { 340 return !IS_DGFX(i915); 341 } 342 343 static unsigned int get_mocs_settings(const struct drm_i915_private *i915, 344 struct drm_i915_mocs_table *table) 345 { 346 unsigned int flags; 347 348 if (IS_DG1(i915)) { 349 table->size = ARRAY_SIZE(dg1_mocs_table); 350 table->table = dg1_mocs_table; 351 table->n_entries = GEN11_NUM_MOCS_ENTRIES; 352 } else if (INTEL_GEN(i915) >= 12) { 353 table->size = ARRAY_SIZE(tgl_mocs_table); 354 table->table = tgl_mocs_table; 355 table->n_entries = GEN11_NUM_MOCS_ENTRIES; 356 } else if (IS_GEN(i915, 11)) { 357 table->size = ARRAY_SIZE(icl_mocs_table); 358 table->table = icl_mocs_table; 359 table->n_entries = GEN11_NUM_MOCS_ENTRIES; 360 } else if (IS_GEN9_BC(i915) || IS_CANNONLAKE(i915)) { 361 table->size = ARRAY_SIZE(skl_mocs_table); 362 table->n_entries = GEN9_NUM_MOCS_ENTRIES; 363 table->table = skl_mocs_table; 364 } else if (IS_GEN9_LP(i915)) { 365 table->size = ARRAY_SIZE(broxton_mocs_table); 366 table->n_entries = GEN9_NUM_MOCS_ENTRIES; 367 table->table = broxton_mocs_table; 368 } else { 369 drm_WARN_ONCE(&i915->drm, INTEL_GEN(i915) >= 9, 370 "Platform that should have a MOCS table does not.\n"); 371 return 0; 372 } 373 374 if (GEM_DEBUG_WARN_ON(table->size > table->n_entries)) 375 return 0; 376 377 /* WaDisableSkipCaching:skl,bxt,kbl,glk */ 378 if (IS_GEN(i915, 9)) { 379 int i; 380 381 for (i = 0; i < table->size; i++) 382 if (GEM_DEBUG_WARN_ON(table->table[i].l3cc_value & 383 (L3_ESC(1) | L3_SCC(0x7)))) 384 return 0; 385 } 386 387 flags = 0; 388 if (has_mocs(i915)) { 389 if (has_global_mocs(i915)) 390 flags |= HAS_GLOBAL_MOCS; 391 else 392 flags |= HAS_ENGINE_MOCS; 393 } 394 if (has_l3cc(i915)) 395 flags |= HAS_RENDER_L3CC; 396 397 return flags; 398 } 399 400 /* 401 * Get control_value from MOCS entry taking into account when it's not used: 402 * I915_MOCS_PTE's value is returned in this case. 403 */ 404 static u32 get_entry_control(const struct drm_i915_mocs_table *table, 405 unsigned int index) 406 { 407 if (index < table->size && table->table[index].used) 408 return table->table[index].control_value; 409 410 return table->table[I915_MOCS_PTE].control_value; 411 } 412 413 #define for_each_mocs(mocs, t, i) \ 414 for (i = 0; \ 415 i < (t)->n_entries ? (mocs = get_entry_control((t), i)), 1 : 0;\ 416 i++) 417 418 static void __init_mocs_table(struct intel_uncore *uncore, 419 const struct drm_i915_mocs_table *table, 420 u32 addr) 421 { 422 unsigned int i; 423 u32 mocs; 424 425 for_each_mocs(mocs, table, i) 426 intel_uncore_write_fw(uncore, _MMIO(addr + i * 4), mocs); 427 } 428 429 static u32 mocs_offset(const struct intel_engine_cs *engine) 430 { 431 static const u32 offset[] = { 432 [RCS0] = __GEN9_RCS0_MOCS0, 433 [VCS0] = __GEN9_VCS0_MOCS0, 434 [VCS1] = __GEN9_VCS1_MOCS0, 435 [VECS0] = __GEN9_VECS0_MOCS0, 436 [BCS0] = __GEN9_BCS0_MOCS0, 437 [VCS2] = __GEN11_VCS2_MOCS0, 438 }; 439 440 GEM_BUG_ON(engine->id >= ARRAY_SIZE(offset)); 441 return offset[engine->id]; 442 } 443 444 static void init_mocs_table(struct intel_engine_cs *engine, 445 const struct drm_i915_mocs_table *table) 446 { 447 __init_mocs_table(engine->uncore, table, mocs_offset(engine)); 448 } 449 450 /* 451 * Get l3cc_value from MOCS entry taking into account when it's not used: 452 * I915_MOCS_PTE's value is returned in this case. 453 */ 454 static u16 get_entry_l3cc(const struct drm_i915_mocs_table *table, 455 unsigned int index) 456 { 457 if (index < table->size && table->table[index].used) 458 return table->table[index].l3cc_value; 459 460 return table->table[I915_MOCS_PTE].l3cc_value; 461 } 462 463 static inline u32 l3cc_combine(u16 low, u16 high) 464 { 465 return low | (u32)high << 16; 466 } 467 468 #define for_each_l3cc(l3cc, t, i) \ 469 for (i = 0; \ 470 i < ((t)->n_entries + 1) / 2 ? \ 471 (l3cc = l3cc_combine(get_entry_l3cc((t), 2 * i), \ 472 get_entry_l3cc((t), 2 * i + 1))), 1 : \ 473 0; \ 474 i++) 475 476 static void init_l3cc_table(struct intel_engine_cs *engine, 477 const struct drm_i915_mocs_table *table) 478 { 479 struct intel_uncore *uncore = engine->uncore; 480 unsigned int i; 481 u32 l3cc; 482 483 for_each_l3cc(l3cc, table, i) 484 intel_uncore_write_fw(uncore, GEN9_LNCFCMOCS(i), l3cc); 485 } 486 487 void intel_mocs_init_engine(struct intel_engine_cs *engine) 488 { 489 struct drm_i915_mocs_table table; 490 unsigned int flags; 491 492 /* Called under a blanket forcewake */ 493 assert_forcewakes_active(engine->uncore, FORCEWAKE_ALL); 494 495 flags = get_mocs_settings(engine->i915, &table); 496 if (!flags) 497 return; 498 499 /* Platforms with global MOCS do not need per-engine initialization. */ 500 if (flags & HAS_ENGINE_MOCS) 501 init_mocs_table(engine, &table); 502 503 if (flags & HAS_RENDER_L3CC && engine->class == RENDER_CLASS) 504 init_l3cc_table(engine, &table); 505 } 506 507 static u32 global_mocs_offset(void) 508 { 509 return i915_mmio_reg_offset(GEN12_GLOBAL_MOCS(0)); 510 } 511 512 void intel_mocs_init(struct intel_gt *gt) 513 { 514 struct drm_i915_mocs_table table; 515 unsigned int flags; 516 517 /* 518 * LLC and eDRAM control values are not applicable to dgfx 519 */ 520 flags = get_mocs_settings(gt->i915, &table); 521 if (flags & HAS_GLOBAL_MOCS) 522 __init_mocs_table(gt->uncore, &table, global_mocs_offset()); 523 } 524 525 #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST) 526 #include "selftest_mocs.c" 527 #endif 528