drivers/gpu/drm/i915/intel_engine_cs.c

   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 "i915_drv.h"
  26 #include "intel_ringbuffer.h"
  27 #include "intel_lrc.h"
  28
  29 /* Haswell does have the CXT_SIZE register however it does not appear to be
  30  * valid. Now, docs explain in dwords what is in the context object. The full
  31  * size is 70720 bytes, however, the power context and execlist context will
  32  * never be saved (power context is stored elsewhere, and execlists don't work
  33  * on HSW) - so the final size, including the extra state required for the
  34  * Resource Streamer, is 66944 bytes, which rounds to 17 pages.
  35  */
  36 #define HSW_CXT_TOTAL_SIZE              (17 * PAGE_SIZE)
  37 /* Same as Haswell, but 72064 bytes now. */
  38 #define GEN8_CXT_TOTAL_SIZE             (18 * PAGE_SIZE)
  39
  40 #define GEN8_LR_CONTEXT_RENDER_SIZE     (20 * PAGE_SIZE)
  41 #define GEN9_LR_CONTEXT_RENDER_SIZE     (22 * PAGE_SIZE)
  42
  43 #define GEN8_LR_CONTEXT_OTHER_SIZE      ( 2 * PAGE_SIZE)
  44
  45 struct engine_class_info {
  46         const char *name;
  47         int (*init_legacy)(struct intel_engine_cs *engine);
  48         int (*init_execlists)(struct intel_engine_cs *engine);
  49 };
  50
  51 static const struct engine_class_info intel_engine_classes[] = {
  52         [RENDER_CLASS] = {
  53                 .name = "rcs",
  54                 .init_execlists = logical_render_ring_init,
  55                 .init_legacy = intel_init_render_ring_buffer,
  56         },
  57         [COPY_ENGINE_CLASS] = {
  58                 .name = "bcs",
  59                 .init_execlists = logical_xcs_ring_init,
  60                 .init_legacy = intel_init_blt_ring_buffer,
  61         },
  62         [VIDEO_DECODE_CLASS] = {
  63                 .name = "vcs",
  64                 .init_execlists = logical_xcs_ring_init,
  65                 .init_legacy = intel_init_bsd_ring_buffer,
  66         },
  67         [VIDEO_ENHANCEMENT_CLASS] = {
  68                 .name = "vecs",
  69                 .init_execlists = logical_xcs_ring_init,
  70                 .init_legacy = intel_init_vebox_ring_buffer,
  71         },
  72 };
  73
  74 struct engine_info {
  75         unsigned int hw_id;
  76         unsigned int uabi_id;
  77         u8 class;
  78         u8 instance;
  79         u32 mmio_base;
  80         unsigned irq_shift;
  81 };
  82
  83 static const struct engine_info intel_engines[] = {
  84         [RCS] = {
  85                 .hw_id = RCS_HW,
  86                 .uabi_id = I915_EXEC_RENDER,
  87                 .class = RENDER_CLASS,
  88                 .instance = 0,
  89                 .mmio_base = RENDER_RING_BASE,
  90                 .irq_shift = GEN8_RCS_IRQ_SHIFT,
  91         },
  92         [BCS] = {
  93                 .hw_id = BCS_HW,
  94                 .uabi_id = I915_EXEC_BLT,
  95                 .class = COPY_ENGINE_CLASS,
  96                 .instance = 0,
  97                 .mmio_base = BLT_RING_BASE,
  98                 .irq_shift = GEN8_BCS_IRQ_SHIFT,
  99         },
 100         [VCS] = {
 101                 .hw_id = VCS_HW,
 102                 .uabi_id = I915_EXEC_BSD,
 103                 .class = VIDEO_DECODE_CLASS,
 104                 .instance = 0,
 105                 .mmio_base = GEN6_BSD_RING_BASE,
 106                 .irq_shift = GEN8_VCS1_IRQ_SHIFT,
 107         },
 108         [VCS2] = {
 109                 .hw_id = VCS2_HW,
 110                 .uabi_id = I915_EXEC_BSD,
 111                 .class = VIDEO_DECODE_CLASS,
 112                 .instance = 1,
 113                 .mmio_base = GEN8_BSD2_RING_BASE,
 114                 .irq_shift = GEN8_VCS2_IRQ_SHIFT,
 115         },
 116         [VECS] = {
 117                 .hw_id = VECS_HW,
 118                 .uabi_id = I915_EXEC_VEBOX,
 119                 .class = VIDEO_ENHANCEMENT_CLASS,
 120                 .instance = 0,
 121                 .mmio_base = VEBOX_RING_BASE,
 122                 .irq_shift = GEN8_VECS_IRQ_SHIFT,
 123         },
 124 };
 125
 126 /**
 127  * ___intel_engine_context_size() - return the size of the context for an engine
 128  * @dev_priv: i915 device private
 129  * @class: engine class
 130  *
 131  * Each engine class may require a different amount of space for a context
 132  * image.
 133  *
 134  * Return: size (in bytes) of an engine class specific context image
 135  *
 136  * Note: this size includes the HWSP, which is part of the context image
 137  * in LRC mode, but does not include the "shared data page" used with
 138  * GuC submission. The caller should account for this if using the GuC.
 139  */
 140 static u32
 141 __intel_engine_context_size(struct drm_i915_private *dev_priv, u8 class)
 142 {
 143         u32 cxt_size;
 144
 145         BUILD_BUG_ON(I915_GTT_PAGE_SIZE != PAGE_SIZE);
 146
 147         switch (class) {
 148         case RENDER_CLASS:
 149                 switch (INTEL_GEN(dev_priv)) {
 150                 default:
 151                         MISSING_CASE(INTEL_GEN(dev_priv));
 152                 case 9:
 153                         return GEN9_LR_CONTEXT_RENDER_SIZE;
 154                 case 8:
 155                         return i915.enable_execlists ?
 156                                GEN8_LR_CONTEXT_RENDER_SIZE :
 157                                GEN8_CXT_TOTAL_SIZE;
 158                 case 7:
 159                         if (IS_HASWELL(dev_priv))
 160                                 return HSW_CXT_TOTAL_SIZE;
 161
 162                         cxt_size = I915_READ(GEN7_CXT_SIZE);
 163                         return round_up(GEN7_CXT_TOTAL_SIZE(cxt_size) * 64,
 164                                         PAGE_SIZE);
 165                 case 6:
 166                         cxt_size = I915_READ(CXT_SIZE);
 167                         return round_up(GEN6_CXT_TOTAL_SIZE(cxt_size) * 64,
 168                                         PAGE_SIZE);
 169                 case 5:
 170                 case 4:
 171                 case 3:
 172                 case 2:
 173                 /* For the special day when i810 gets merged. */
 174                 case 1:
 175                         return 0;
 176                 }
 177                 break;
 178         default:
 179                 MISSING_CASE(class);
 180         case VIDEO_DECODE_CLASS:
 181         case VIDEO_ENHANCEMENT_CLASS:
 182         case COPY_ENGINE_CLASS:
 183                 if (INTEL_GEN(dev_priv) < 8)
 184                         return 0;
 185                 return GEN8_LR_CONTEXT_OTHER_SIZE;
 186         }
 187 }
 188
 189 static int
 190 intel_engine_setup(struct drm_i915_private *dev_priv,
 191                    enum intel_engine_id id)
 192 {
 193         const struct engine_info *info = &intel_engines[id];
 194         const struct engine_class_info *class_info;
 195         struct intel_engine_cs *engine;
 196
 197         GEM_BUG_ON(info->class >= ARRAY_SIZE(intel_engine_classes));
 198         class_info = &intel_engine_classes[info->class];
 199
 200         GEM_BUG_ON(dev_priv->engine[id]);
 201         engine = kzalloc(sizeof(*engine), GFP_KERNEL);
 202         if (!engine)
 203                 return -ENOMEM;
 204
 205         engine->id = id;
 206         engine->i915 = dev_priv;
 207         WARN_ON(snprintf(engine->name, sizeof(engine->name), "%s%u",
 208                          class_info->name, info->instance) >=
 209                 sizeof(engine->name));
 210         engine->uabi_id = info->uabi_id;
 211         engine->hw_id = engine->guc_id = info->hw_id;
 212         engine->mmio_base = info->mmio_base;
 213         engine->irq_shift = info->irq_shift;
 214         engine->class = info->class;
 215         engine->instance = info->instance;
 216
 217         engine->context_size = __intel_engine_context_size(dev_priv,
 218                                                            engine->class);
 219         if (WARN_ON(engine->context_size > BIT(20)))
 220                 engine->context_size = 0;
 221
 222         /* Nothing to do here, execute in order of dependencies */
 223         engine->schedule = NULL;
 224
 225         ATOMIC_INIT_NOTIFIER_HEAD(&engine->context_status_notifier);
 226
 227         dev_priv->engine[id] = engine;
 228         return 0;
 229 }
 230
 231 /**
 232  * intel_engines_init_mmio() - allocate and prepare the Engine Command Streamers
 233  * @dev_priv: i915 device private
 234  *
 235  * Return: non-zero if the initialization failed.
 236  */
 237 int intel_engines_init_mmio(struct drm_i915_private *dev_priv)
 238 {
 239         struct intel_device_info *device_info = mkwrite_device_info(dev_priv);
 240         const unsigned int ring_mask = INTEL_INFO(dev_priv)->ring_mask;
 241         struct intel_engine_cs *engine;
 242         enum intel_engine_id id;
 243         unsigned int mask = 0;
 244         unsigned int i;
 245         int err;
 246
 247         WARN_ON(ring_mask == 0);
 248         WARN_ON(ring_mask &
 249                 GENMASK(sizeof(mask) * BITS_PER_BYTE - 1, I915_NUM_ENGINES));
 250
 251         for (i = 0; i < ARRAY_SIZE(intel_engines); i++) {
 252                 if (!HAS_ENGINE(dev_priv, i))
 253                         continue;
 254
 255                 err = intel_engine_setup(dev_priv, i);
 256                 if (err)
 257                         goto cleanup;
 258
 259                 mask |= ENGINE_MASK(i);
 260         }
 261
 262         /*
 263          * Catch failures to update intel_engines table when the new engines
 264          * are added to the driver by a warning and disabling the forgotten
 265          * engines.
 266          */
 267         if (WARN_ON(mask != ring_mask))
 268                 device_info->ring_mask = mask;
 269
 270         /* We always presume we have at least RCS available for later probing */
 271         if (WARN_ON(!HAS_ENGINE(dev_priv, RCS))) {
 272                 err = -ENODEV;
 273                 goto cleanup;
 274         }
 275
 276         device_info->num_rings = hweight32(mask);
 277
 278         return 0;
 279
 280 cleanup:
 281         for_each_engine(engine, dev_priv, id)
 282                 kfree(engine);
 283         return err;
 284 }
 285
 286 /**
 287  * intel_engines_init() - init the Engine Command Streamers
 288  * @dev_priv: i915 device private
 289  *
 290  * Return: non-zero if the initialization failed.
 291  */
 292 int intel_engines_init(struct drm_i915_private *dev_priv)
 293 {
 294         struct intel_device_info *device_info = mkwrite_device_info(dev_priv);
 295         struct intel_engine_cs *engine;
 296         enum intel_engine_id id, err_id;
 297         unsigned int mask = 0;
 298         int err = 0;
 299
 300         for_each_engine(engine, dev_priv, id) {
 301                 const struct engine_class_info *class_info =
 302                         &intel_engine_classes[engine->class];
 303                 int (*init)(struct intel_engine_cs *engine);
 304
 305                 if (i915.enable_execlists)
 306                         init = class_info->init_execlists;
 307                 else
 308                         init = class_info->init_legacy;
 309                 if (!init) {
 310                         kfree(engine);
 311                         dev_priv->engine[id] = NULL;
 312                         continue;
 313                 }
 314
 315                 err = init(engine);
 316                 if (err) {
 317                         err_id = id;
 318                         goto cleanup;
 319                 }
 320
 321                 GEM_BUG_ON(!engine->submit_request);
 322                 mask |= ENGINE_MASK(id);
 323         }
 324
 325         /*
 326          * Catch failures to update intel_engines table when the new engines
 327          * are added to the driver by a warning and disabling the forgotten
 328          * engines.
 329          */
 330         if (WARN_ON(mask != INTEL_INFO(dev_priv)->ring_mask))
 331                 device_info->ring_mask = mask;
 332
 333         device_info->num_rings = hweight32(mask);
 334
 335         return 0;
 336
 337 cleanup:
 338         for_each_engine(engine, dev_priv, id) {
 339                 if (id >= err_id)
 340                         kfree(engine);
 341                 else
 342                         dev_priv->gt.cleanup_engine(engine);
 343         }
 344         return err;
 345 }
 346
 347 void intel_engine_init_global_seqno(struct intel_engine_cs *engine, u32 seqno)
 348 {
 349         struct drm_i915_private *dev_priv = engine->i915;
 350
 351         GEM_BUG_ON(!intel_engine_is_idle(engine));
 352         GEM_BUG_ON(i915_gem_active_isset(&engine->timeline->last_request));
 353
 354         /* Our semaphore implementation is strictly monotonic (i.e. we proceed
 355          * so long as the semaphore value in the register/page is greater
 356          * than the sync value), so whenever we reset the seqno,
 357          * so long as we reset the tracking semaphore value to 0, it will
 358          * always be before the next request's seqno. If we don't reset
 359          * the semaphore value, then when the seqno moves backwards all
 360          * future waits will complete instantly (causing rendering corruption).
 361          */
 362         if (IS_GEN6(dev_priv) || IS_GEN7(dev_priv)) {
 363                 I915_WRITE(RING_SYNC_0(engine->mmio_base), 0);
 364                 I915_WRITE(RING_SYNC_1(engine->mmio_base), 0);
 365                 if (HAS_VEBOX(dev_priv))
 366                         I915_WRITE(RING_SYNC_2(engine->mmio_base), 0);
 367         }
 368         if (dev_priv->semaphore) {
 369                 struct page *page = i915_vma_first_page(dev_priv->semaphore);
 370                 void *semaphores;
 371
 372                 /* Semaphores are in noncoherent memory, flush to be safe */
 373                 semaphores = kmap_atomic(page);
 374                 memset(semaphores + GEN8_SEMAPHORE_OFFSET(engine->id, 0),
 375                        0, I915_NUM_ENGINES * gen8_semaphore_seqno_size);
 376                 drm_clflush_virt_range(semaphores + GEN8_SEMAPHORE_OFFSET(engine->id, 0),
 377                                        I915_NUM_ENGINES * gen8_semaphore_seqno_size);
 378                 kunmap_atomic(semaphores);
 379         }
 380
 381         intel_write_status_page(engine, I915_GEM_HWS_INDEX, seqno);
 382         clear_bit(ENGINE_IRQ_BREADCRUMB, &engine->irq_posted);
 383
 384         /* After manually advancing the seqno, fake the interrupt in case
 385          * there are any waiters for that seqno.
 386          */
 387         intel_engine_wakeup(engine);
 388
 389         GEM_BUG_ON(intel_engine_get_seqno(engine) != seqno);
 390 }
 391
 392 static void intel_engine_init_timeline(struct intel_engine_cs *engine)
 393 {
 394         engine->timeline = &engine->i915->gt.global_timeline.engine[engine->id];
 395 }
 396
 397 /**
 398  * intel_engines_setup_common - setup engine state not requiring hw access
 399  * @engine: Engine to setup.
 400  *
 401  * Initializes @engine@ structure members shared between legacy and execlists
 402  * submission modes which do not require hardware access.
 403  *
 404  * Typically done early in the submission mode specific engine setup stage.
 405  */
 406 void intel_engine_setup_common(struct intel_engine_cs *engine)
 407 {
 408         engine->execlist_queue = RB_ROOT;
 409         engine->execlist_first = NULL;
 410
 411         intel_engine_init_timeline(engine);
 412         intel_engine_init_hangcheck(engine);
 413         i915_gem_batch_pool_init(engine, &engine->batch_pool);
 414
 415         intel_engine_init_cmd_parser(engine);
 416 }
 417
 418 int intel_engine_create_scratch(struct intel_engine_cs *engine, int size)
 419 {
 420         struct drm_i915_gem_object *obj;
 421         struct i915_vma *vma;
 422         int ret;
 423
 424         WARN_ON(engine->scratch);
 425
 426         obj = i915_gem_object_create_stolen(engine->i915, size);
 427         if (!obj)
 428                 obj = i915_gem_object_create_internal(engine->i915, size);
 429         if (IS_ERR(obj)) {
 430                 DRM_ERROR("Failed to allocate scratch page\n");
 431                 return PTR_ERR(obj);
 432         }
 433
 434         vma = i915_vma_instance(obj, &engine->i915->ggtt.base, NULL);
 435         if (IS_ERR(vma)) {
 436                 ret = PTR_ERR(vma);
 437                 goto err_unref;
 438         }
 439
 440         ret = i915_vma_pin(vma, 0, 4096, PIN_GLOBAL | PIN_HIGH);
 441         if (ret)
 442                 goto err_unref;
 443
 444         engine->scratch = vma;
 445         DRM_DEBUG_DRIVER("%s pipe control offset: 0x%08x\n",
 446                          engine->name, i915_ggtt_offset(vma));
 447         return 0;
 448
 449 err_unref:
 450         i915_gem_object_put(obj);
 451         return ret;
 452 }
 453
 454 static void intel_engine_cleanup_scratch(struct intel_engine_cs *engine)
 455 {
 456         i915_vma_unpin_and_release(&engine->scratch);
 457 }
 458
 459 /**
 460  * intel_engines_init_common - initialize cengine state which might require hw access
 461  * @engine: Engine to initialize.
 462  *
 463  * Initializes @engine@ structure members shared between legacy and execlists
 464  * submission modes which do require hardware access.
 465  *
 466  * Typcally done at later stages of submission mode specific engine setup.
 467  *
 468  * Returns zero on success or an error code on failure.
 469  */
 470 int intel_engine_init_common(struct intel_engine_cs *engine)
 471 {
 472         struct intel_ring *ring;
 473         int ret;
 474
 475         engine->set_default_submission(engine);
 476
 477         /* We may need to do things with the shrinker which
 478          * require us to immediately switch back to the default
 479          * context. This can cause a problem as pinning the
 480          * default context also requires GTT space which may not
 481          * be available. To avoid this we always pin the default
 482          * context.
 483          */
 484         ring = engine->context_pin(engine, engine->i915->kernel_context);
 485         if (IS_ERR(ring))
 486                 return PTR_ERR(ring);
 487
 488         ret = intel_engine_init_breadcrumbs(engine);
 489         if (ret)
 490                 goto err_unpin;
 491
 492         ret = i915_gem_render_state_init(engine);
 493         if (ret)
 494                 goto err_unpin;
 495
 496         return 0;
 497
 498 err_unpin:
 499         engine->context_unpin(engine, engine->i915->kernel_context);
 500         return ret;
 501 }
 502
 503 /**
 504  * intel_engines_cleanup_common - cleans up the engine state created by
 505  *                                the common initiailizers.
 506  * @engine: Engine to cleanup.
 507  *
 508  * This cleans up everything created by the common helpers.
 509  */
 510 void intel_engine_cleanup_common(struct intel_engine_cs *engine)
 511 {
 512         intel_engine_cleanup_scratch(engine);
 513
 514         i915_gem_render_state_fini(engine);
 515         intel_engine_fini_breadcrumbs(engine);
 516         intel_engine_cleanup_cmd_parser(engine);
 517         i915_gem_batch_pool_fini(&engine->batch_pool);
 518
 519         engine->context_unpin(engine, engine->i915->kernel_context);
 520 }
 521
 522 u64 intel_engine_get_active_head(struct intel_engine_cs *engine)
 523 {
 524         struct drm_i915_private *dev_priv = engine->i915;
 525         u64 acthd;
 526
 527         if (INTEL_GEN(dev_priv) >= 8)
 528                 acthd = I915_READ64_2x32(RING_ACTHD(engine->mmio_base),
 529                                          RING_ACTHD_UDW(engine->mmio_base));
 530         else if (INTEL_GEN(dev_priv) >= 4)
 531                 acthd = I915_READ(RING_ACTHD(engine->mmio_base));
 532         else
 533                 acthd = I915_READ(ACTHD);
 534
 535         return acthd;
 536 }
 537
 538 u64 intel_engine_get_last_batch_head(struct intel_engine_cs *engine)
 539 {
 540         struct drm_i915_private *dev_priv = engine->i915;
 541         u64 bbaddr;
 542
 543         if (INTEL_GEN(dev_priv) >= 8)
 544                 bbaddr = I915_READ64_2x32(RING_BBADDR(engine->mmio_base),
 545                                           RING_BBADDR_UDW(engine->mmio_base));
 546         else
 547                 bbaddr = I915_READ(RING_BBADDR(engine->mmio_base));
 548
 549         return bbaddr;
 550 }
 551
 552 const char *i915_cache_level_str(struct drm_i915_private *i915, int type)
 553 {
 554         switch (type) {
 555         case I915_CACHE_NONE: return " uncached";
 556         case I915_CACHE_LLC: return HAS_LLC(i915) ? " LLC" : " snooped";
 557         case I915_CACHE_L3_LLC: return " L3+LLC";
 558         case I915_CACHE_WT: return " WT";
 559         default: return "";
 560         }
 561 }
 562
 563 static inline uint32_t
 564 read_subslice_reg(struct drm_i915_private *dev_priv, int slice,
 565                   int subslice, i915_reg_t reg)
 566 {
 567         uint32_t mcr;
 568         uint32_t ret;
 569         enum forcewake_domains fw_domains;
 570
 571         fw_domains = intel_uncore_forcewake_for_reg(dev_priv, reg,
 572                                                     FW_REG_READ);
 573         fw_domains |= intel_uncore_forcewake_for_reg(dev_priv,
 574                                                      GEN8_MCR_SELECTOR,
 575                                                      FW_REG_READ | FW_REG_WRITE);
 576
 577         spin_lock_irq(&dev_priv->uncore.lock);
 578         intel_uncore_forcewake_get__locked(dev_priv, fw_domains);
 579
 580         mcr = I915_READ_FW(GEN8_MCR_SELECTOR);
 581         /*
 582          * The HW expects the slice and sublice selectors to be reset to 0
 583          * after reading out the registers.
 584          */
 585         WARN_ON_ONCE(mcr & (GEN8_MCR_SLICE_MASK | GEN8_MCR_SUBSLICE_MASK));
 586         mcr &= ~(GEN8_MCR_SLICE_MASK | GEN8_MCR_SUBSLICE_MASK);
 587         mcr |= GEN8_MCR_SLICE(slice) | GEN8_MCR_SUBSLICE(subslice);
 588         I915_WRITE_FW(GEN8_MCR_SELECTOR, mcr);
 589
 590         ret = I915_READ_FW(reg);
 591
 592         mcr &= ~(GEN8_MCR_SLICE_MASK | GEN8_MCR_SUBSLICE_MASK);
 593         I915_WRITE_FW(GEN8_MCR_SELECTOR, mcr);
 594
 595         intel_uncore_forcewake_put__locked(dev_priv, fw_domains);
 596         spin_unlock_irq(&dev_priv->uncore.lock);
 597
 598         return ret;
 599 }
 600
 601 /* NB: please notice the memset */
 602 void intel_engine_get_instdone(struct intel_engine_cs *engine,
 603                                struct intel_instdone *instdone)
 604 {
 605         struct drm_i915_private *dev_priv = engine->i915;
 606         u32 mmio_base = engine->mmio_base;
 607         int slice;
 608         int subslice;
 609
 610         memset(instdone, 0, sizeof(*instdone));
 611
 612         switch (INTEL_GEN(dev_priv)) {
 613         default:
 614                 instdone->instdone = I915_READ(RING_INSTDONE(mmio_base));
 615
 616                 if (engine->id != RCS)
 617                         break;
 618
 619                 instdone->slice_common = I915_READ(GEN7_SC_INSTDONE);
 620                 for_each_instdone_slice_subslice(dev_priv, slice, subslice) {
 621                         instdone->sampler[slice][subslice] =
 622                                 read_subslice_reg(dev_priv, slice, subslice,
 623                                                   GEN7_SAMPLER_INSTDONE);
 624                         instdone->row[slice][subslice] =
 625                                 read_subslice_reg(dev_priv, slice, subslice,
 626                                                   GEN7_ROW_INSTDONE);
 627                 }
 628                 break;
 629         case 7:
 630                 instdone->instdone = I915_READ(RING_INSTDONE(mmio_base));
 631
 632                 if (engine->id != RCS)
 633                         break;
 634
 635                 instdone->slice_common = I915_READ(GEN7_SC_INSTDONE);
 636                 instdone->sampler[0][0] = I915_READ(GEN7_SAMPLER_INSTDONE);
 637                 instdone->row[0][0] = I915_READ(GEN7_ROW_INSTDONE);
 638
 639                 break;
 640         case 6:
 641         case 5:
 642         case 4:
 643                 instdone->instdone = I915_READ(RING_INSTDONE(mmio_base));
 644
 645                 if (engine->id == RCS)
 646                         /* HACK: Using the wrong struct member */
 647                         instdone->slice_common = I915_READ(GEN4_INSTDONE1);
 648                 break;
 649         case 3:
 650         case 2:
 651                 instdone->instdone = I915_READ(GEN2_INSTDONE);
 652                 break;
 653         }
 654 }
 655
 656 static int wa_add(struct drm_i915_private *dev_priv,
 657                   i915_reg_t addr,
 658                   const u32 mask, const u32 val)
 659 {
 660         const u32 idx = dev_priv->workarounds.count;
 661
 662         if (WARN_ON(idx >= I915_MAX_WA_REGS))
 663                 return -ENOSPC;
 664
 665         dev_priv->workarounds.reg[idx].addr = addr;
 666         dev_priv->workarounds.reg[idx].value = val;
 667         dev_priv->workarounds.reg[idx].mask = mask;
 668
 669         dev_priv->workarounds.count++;
 670
 671         return 0;
 672 }
 673
 674 #define WA_REG(addr, mask, val) do { \
 675                 const int r = wa_add(dev_priv, (addr), (mask), (val)); \
 676                 if (r) \
 677                         return r; \
 678         } while (0)
 679
 680 #define WA_SET_BIT_MASKED(addr, mask) \
 681         WA_REG(addr, (mask), _MASKED_BIT_ENABLE(mask))
 682
 683 #define WA_CLR_BIT_MASKED(addr, mask) \
 684         WA_REG(addr, (mask), _MASKED_BIT_DISABLE(mask))
 685
 686 #define WA_SET_FIELD_MASKED(addr, mask, value) \
 687         WA_REG(addr, mask, _MASKED_FIELD(mask, value))
 688
 689 #define WA_SET_BIT(addr, mask) WA_REG(addr, mask, I915_READ(addr) | (mask))
 690 #define WA_CLR_BIT(addr, mask) WA_REG(addr, mask, I915_READ(addr) & ~(mask))
 691
 692 #define WA_WRITE(addr, val) WA_REG(addr, 0xffffffff, val)
 693
 694 static int wa_ring_whitelist_reg(struct intel_engine_cs *engine,
 695                                  i915_reg_t reg)
 696 {
 697         struct drm_i915_private *dev_priv = engine->i915;
 698         struct i915_workarounds *wa = &dev_priv->workarounds;
 699         const uint32_t index = wa->hw_whitelist_count[engine->id];
 700
 701         if (WARN_ON(index >= RING_MAX_NONPRIV_SLOTS))
 702                 return -EINVAL;
 703
 704         WA_WRITE(RING_FORCE_TO_NONPRIV(engine->mmio_base, index),
 705                  i915_mmio_reg_offset(reg));
 706         wa->hw_whitelist_count[engine->id]++;
 707
 708         return 0;
 709 }
 710
 711 static int gen8_init_workarounds(struct intel_engine_cs *engine)
 712 {
 713         struct drm_i915_private *dev_priv = engine->i915;
 714
 715         WA_SET_BIT_MASKED(INSTPM, INSTPM_FORCE_ORDERING);
 716
 717         /* WaDisableAsyncFlipPerfMode:bdw,chv */
 718         WA_SET_BIT_MASKED(MI_MODE, ASYNC_FLIP_PERF_DISABLE);
 719
 720         /* WaDisablePartialInstShootdown:bdw,chv */
 721         WA_SET_BIT_MASKED(GEN8_ROW_CHICKEN,
 722                           PARTIAL_INSTRUCTION_SHOOTDOWN_DISABLE);
 723
 724         /* Use Force Non-Coherent whenever executing a 3D context. This is a
 725          * workaround for for a possible hang in the unlikely event a TLB
 726          * invalidation occurs during a PSD flush.
 727          */
 728         /* WaForceEnableNonCoherent:bdw,chv */
 729         /* WaHdcDisableFetchWhenMasked:bdw,chv */
 730         WA_SET_BIT_MASKED(HDC_CHICKEN0,
 731                           HDC_DONOT_FETCH_MEM_WHEN_MASKED |
 732                           HDC_FORCE_NON_COHERENT);
 733
 734         /* From the Haswell PRM, Command Reference: Registers, CACHE_MODE_0:
 735          * "The Hierarchical Z RAW Stall Optimization allows non-overlapping
 736          *  polygons in the same 8x4 pixel/sample area to be processed without
 737          *  stalling waiting for the earlier ones to write to Hierarchical Z
 738          *  buffer."
 739          *
 740          * This optimization is off by default for BDW and CHV; turn it on.
 741          */
 742         WA_CLR_BIT_MASKED(CACHE_MODE_0_GEN7, HIZ_RAW_STALL_OPT_DISABLE);
 743
 744         /* Wa4x4STCOptimizationDisable:bdw,chv */
 745         WA_SET_BIT_MASKED(CACHE_MODE_1, GEN8_4x4_STC_OPTIMIZATION_DISABLE);
 746
 747         /*
 748          * BSpec recommends 8x4 when MSAA is used,
 749          * however in practice 16x4 seems fastest.
 750          *
 751          * Note that PS/WM thread counts depend on the WIZ hashing
 752          * disable bit, which we don't touch here, but it's good
 753          * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
 754          */
 755         WA_SET_FIELD_MASKED(GEN7_GT_MODE,
 756                             GEN6_WIZ_HASHING_MASK,
 757                             GEN6_WIZ_HASHING_16x4);
 758
 759         return 0;
 760 }
 761
 762 static int bdw_init_workarounds(struct intel_engine_cs *engine)
 763 {
 764         struct drm_i915_private *dev_priv = engine->i915;
 765         int ret;
 766
 767         ret = gen8_init_workarounds(engine);
 768         if (ret)
 769                 return ret;
 770
 771         /* WaDisableThreadStallDopClockGating:bdw (pre-production) */
 772         WA_SET_BIT_MASKED(GEN8_ROW_CHICKEN, STALL_DOP_GATING_DISABLE);
 773
 774         /* WaDisableDopClockGating:bdw
 775          *
 776          * Also see the related UCGTCL1 write in broadwell_init_clock_gating()
 777          * to disable EUTC clock gating.
 778          */
 779         WA_SET_BIT_MASKED(GEN7_ROW_CHICKEN2,
 780                           DOP_CLOCK_GATING_DISABLE);
 781
 782         WA_SET_BIT_MASKED(HALF_SLICE_CHICKEN3,
 783                           GEN8_SAMPLER_POWER_BYPASS_DIS);
 784
 785         WA_SET_BIT_MASKED(HDC_CHICKEN0,
 786                           /* WaForceContextSaveRestoreNonCoherent:bdw */
 787                           HDC_FORCE_CONTEXT_SAVE_RESTORE_NON_COHERENT |
 788                           /* WaDisableFenceDestinationToSLM:bdw (pre-prod) */
 789                           (IS_BDW_GT3(dev_priv) ? HDC_FENCE_DEST_SLM_DISABLE : 0));
 790
 791         return 0;
 792 }
 793
 794 static int chv_init_workarounds(struct intel_engine_cs *engine)
 795 {
 796         struct drm_i915_private *dev_priv = engine->i915;
 797         int ret;
 798
 799         ret = gen8_init_workarounds(engine);
 800         if (ret)
 801                 return ret;
 802
 803         /* WaDisableThreadStallDopClockGating:chv */
 804         WA_SET_BIT_MASKED(GEN8_ROW_CHICKEN, STALL_DOP_GATING_DISABLE);
 805
 806         /* Improve HiZ throughput on CHV. */
 807         WA_SET_BIT_MASKED(HIZ_CHICKEN, CHV_HZ_8X8_MODE_IN_1X);
 808
 809         return 0;
 810 }
 811
 812 static int gen9_init_workarounds(struct intel_engine_cs *engine)
 813 {
 814         struct drm_i915_private *dev_priv = engine->i915;
 815         int ret;
 816
 817         /* WaConextSwitchWithConcurrentTLBInvalidate:skl,bxt,kbl,glk,cfl */
 818         I915_WRITE(GEN9_CSFE_CHICKEN1_RCS, _MASKED_BIT_ENABLE(GEN9_PREEMPT_GPGPU_SYNC_SWITCH_DISABLE));
 819
 820         /* WaEnableLbsSlaRetryTimerDecrement:skl,bxt,kbl,glk,cfl */
 821         I915_WRITE(BDW_SCRATCH1, I915_READ(BDW_SCRATCH1) |
 822                    GEN9_LBS_SLA_RETRY_TIMER_DECREMENT_ENABLE);
 823
 824         /* WaDisableKillLogic:bxt,skl,kbl,cfl */
 825         I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) |
 826                    ECOCHK_DIS_TLB);
 827
 828         /* WaClearFlowControlGpgpuContextSave:skl,bxt,kbl,glk,cfl */
 829         /* WaDisablePartialInstShootdown:skl,bxt,kbl,glk,cfl */
 830         WA_SET_BIT_MASKED(GEN8_ROW_CHICKEN,
 831                           FLOW_CONTROL_ENABLE |
 832                           PARTIAL_INSTRUCTION_SHOOTDOWN_DISABLE);
 833
 834         /* Syncing dependencies between camera and graphics:skl,bxt,kbl */
 835         if (!IS_COFFEELAKE(dev_priv))
 836                 WA_SET_BIT_MASKED(HALF_SLICE_CHICKEN3,
 837                                   GEN9_DISABLE_OCL_OOB_SUPPRESS_LOGIC);
 838
 839         /* WaDisableDgMirrorFixInHalfSliceChicken5:bxt */
 840         if (IS_BXT_REVID(dev_priv, 0, BXT_REVID_A1))
 841                 WA_CLR_BIT_MASKED(GEN9_HALF_SLICE_CHICKEN5,
 842                                   GEN9_DG_MIRROR_FIX_ENABLE);
 843
 844         /* WaSetDisablePixMaskCammingAndRhwoInCommonSliceChicken:bxt */
 845         if (IS_BXT_REVID(dev_priv, 0, BXT_REVID_A1)) {
 846                 WA_SET_BIT_MASKED(GEN7_COMMON_SLICE_CHICKEN1,
 847                                   GEN9_RHWO_OPTIMIZATION_DISABLE);
 848                 /*
 849                  * WA also requires GEN9_SLICE_COMMON_ECO_CHICKEN0[14:14] to be set
 850                  * but we do that in per ctx batchbuffer as there is an issue
 851                  * with this register not getting restored on ctx restore
 852                  */
 853         }
 854
 855         /* WaEnableYV12BugFixInHalfSliceChicken7:skl,bxt,kbl,glk,cfl */
 856         /* WaEnableSamplerGPGPUPreemptionSupport:skl,bxt,kbl,cfl */
 857         WA_SET_BIT_MASKED(GEN9_HALF_SLICE_CHICKEN7,
 858                           GEN9_ENABLE_YV12_BUGFIX |
 859                           GEN9_ENABLE_GPGPU_PREEMPTION);
 860
 861         /* Wa4x4STCOptimizationDisable:skl,bxt,kbl,glk,cfl */
 862         /* WaDisablePartialResolveInVc:skl,bxt,kbl,cfl */
 863         WA_SET_BIT_MASKED(CACHE_MODE_1, (GEN8_4x4_STC_OPTIMIZATION_DISABLE |
 864                                          GEN9_PARTIAL_RESOLVE_IN_VC_DISABLE));
 865
 866         /* WaCcsTlbPrefetchDisable:skl,bxt,kbl,glk,cfl */
 867         WA_CLR_BIT_MASKED(GEN9_HALF_SLICE_CHICKEN5,
 868                           GEN9_CCS_TLB_PREFETCH_ENABLE);
 869
 870         /* WaDisableMaskBasedCammingInRCC:bxt */
 871         if (IS_BXT_REVID(dev_priv, 0, BXT_REVID_A1))
 872                 WA_SET_BIT_MASKED(SLICE_ECO_CHICKEN0,
 873                                   PIXEL_MASK_CAMMING_DISABLE);
 874
 875         /* WaForceContextSaveRestoreNonCoherent:skl,bxt,kbl,cfl */
 876         WA_SET_BIT_MASKED(HDC_CHICKEN0,
 877                           HDC_FORCE_CONTEXT_SAVE_RESTORE_NON_COHERENT |
 878                           HDC_FORCE_CSR_NON_COHERENT_OVR_DISABLE);
 879
 880         /* WaForceEnableNonCoherent and WaDisableHDCInvalidation are
 881          * both tied to WaForceContextSaveRestoreNonCoherent
 882          * in some hsds for skl. We keep the tie for all gen9. The
 883          * documentation is a bit hazy and so we want to get common behaviour,
 884          * even though there is no clear evidence we would need both on kbl/bxt.
 885          * This area has been source of system hangs so we play it safe
 886          * and mimic the skl regardless of what bspec says.
 887          *
 888          * Use Force Non-Coherent whenever executing a 3D context. This
 889          * is a workaround for a possible hang in the unlikely event
 890          * a TLB invalidation occurs during a PSD flush.
 891          */
 892
 893         /* WaForceEnableNonCoherent:skl,bxt,kbl,cfl */
 894         WA_SET_BIT_MASKED(HDC_CHICKEN0,
 895                           HDC_FORCE_NON_COHERENT);
 896
 897         /* WaDisableHDCInvalidation:skl,bxt,kbl */
 898         if (!IS_COFFEELAKE(dev_priv))
 899                 I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) |
 900                            BDW_DISABLE_HDC_INVALIDATION);
 901
 902         /* WaDisableSamplerPowerBypassForSOPingPong:skl,bxt,kbl,cfl */
 903         if (IS_SKYLAKE(dev_priv) ||
 904             IS_KABYLAKE(dev_priv) ||
 905             IS_COFFEELAKE(dev_priv) ||
 906             IS_BXT_REVID(dev_priv, 0, BXT_REVID_B0))
 907                 WA_SET_BIT_MASKED(HALF_SLICE_CHICKEN3,
 908                                   GEN8_SAMPLER_POWER_BYPASS_DIS);
 909
 910         /* WaDisableSTUnitPowerOptimization:skl,bxt,kbl,glk,cfl */
 911         WA_SET_BIT_MASKED(HALF_SLICE_CHICKEN2, GEN8_ST_PO_DISABLE);
 912
 913         /* WaOCLCoherentLineFlush:skl,bxt,kbl,cfl */
 914         I915_WRITE(GEN8_L3SQCREG4, (I915_READ(GEN8_L3SQCREG4) |
 915                                     GEN8_LQSC_FLUSH_COHERENT_LINES));
 916
 917         /* WaVFEStateAfterPipeControlwithMediaStateClear:skl,bxt,glk,cfl */
 918         ret = wa_ring_whitelist_reg(engine, GEN9_CTX_PREEMPT_REG);
 919         if (ret)
 920                 return ret;
 921
 922         /* WaEnablePreemptionGranularityControlByUMD:skl,bxt,kbl,cfl */
 923         ret= wa_ring_whitelist_reg(engine, GEN8_CS_CHICKEN1);
 924         if (ret)
 925                 return ret;
 926
 927         /* WaAllowUMDToModifyHDCChicken1:skl,bxt,kbl,glk,cfl */
 928         ret = wa_ring_whitelist_reg(engine, GEN8_HDC_CHICKEN1);
 929         if (ret)
 930                 return ret;
 931
 932         return 0;
 933 }
 934
 935 static int skl_tune_iz_hashing(struct intel_engine_cs *engine)
 936 {
 937         struct drm_i915_private *dev_priv = engine->i915;
 938         u8 vals[3] = { 0, 0, 0 };
 939         unsigned int i;
 940
 941         for (i = 0; i < 3; i++) {
 942                 u8 ss;
 943
 944                 /*
 945                  * Only consider slices where one, and only one, subslice has 7
 946                  * EUs
 947                  */
 948                 if (!is_power_of_2(INTEL_INFO(dev_priv)->sseu.subslice_7eu[i]))
 949                         continue;
 950
 951                 /*
 952                  * subslice_7eu[i] != 0 (because of the check above) and
 953                  * ss_max == 4 (maximum number of subslices possible per slice)
 954                  *
 955                  * ->    0 <= ss <= 3;
 956                  */
 957                 ss = ffs(INTEL_INFO(dev_priv)->sseu.subslice_7eu[i]) - 1;
 958                 vals[i] = 3 - ss;
 959         }
 960
 961         if (vals[0] == 0 && vals[1] == 0 && vals[2] == 0)
 962                 return 0;
 963
 964         /* Tune IZ hashing. See intel_device_info_runtime_init() */
 965         WA_SET_FIELD_MASKED(GEN7_GT_MODE,
 966                             GEN9_IZ_HASHING_MASK(2) |
 967                             GEN9_IZ_HASHING_MASK(1) |
 968                             GEN9_IZ_HASHING_MASK(0),
 969                             GEN9_IZ_HASHING(2, vals[2]) |
 970                             GEN9_IZ_HASHING(1, vals[1]) |
 971                             GEN9_IZ_HASHING(0, vals[0]));
 972
 973         return 0;
 974 }
 975
 976 static int skl_init_workarounds(struct intel_engine_cs *engine)
 977 {
 978         struct drm_i915_private *dev_priv = engine->i915;
 979         int ret;
 980
 981         ret = gen9_init_workarounds(engine);
 982         if (ret)
 983                 return ret;
 984
 985         /*
 986          * Actual WA is to disable percontext preemption granularity control
 987          * until D0 which is the default case so this is equivalent to
 988          * !WaDisablePerCtxtPreemptionGranularityControl:skl
 989          */
 990         I915_WRITE(GEN7_FF_SLICE_CS_CHICKEN1,
 991                    _MASKED_BIT_ENABLE(GEN9_FFSC_PERCTX_PREEMPT_CTRL));
 992
 993         /* WaEnableGapsTsvCreditFix:skl */
 994         I915_WRITE(GEN8_GARBCNTL, (I915_READ(GEN8_GARBCNTL) |
 995                                    GEN9_GAPS_TSV_CREDIT_DISABLE));
 996
 997         /* WaDisableGafsUnitClkGating:skl */
 998         WA_SET_BIT(GEN7_UCGCTL4, GEN8_EU_GAUNIT_CLOCK_GATE_DISABLE);
 999
1000         /* WaInPlaceDecompressionHang:skl */
1001         if (IS_SKL_REVID(dev_priv, SKL_REVID_H0, REVID_FOREVER))
1002                 WA_SET_BIT(GEN9_GAMT_ECO_REG_RW_IA,
1003                            GAMT_ECO_ENABLE_IN_PLACE_DECOMPRESS);
1004
1005         /* WaDisableLSQCROPERFforOCL:skl */
1006         ret = wa_ring_whitelist_reg(engine, GEN8_L3SQCREG4);
1007         if (ret)
1008                 return ret;
1009
1010         return skl_tune_iz_hashing(engine);
1011 }
1012
1013 static int bxt_init_workarounds(struct intel_engine_cs *engine)
1014 {
1015         struct drm_i915_private *dev_priv = engine->i915;
1016         int ret;
1017
1018         ret = gen9_init_workarounds(engine);
1019         if (ret)
1020                 return ret;
1021
1022         /* WaStoreMultiplePTEenable:bxt */
1023         /* This is a requirement according to Hardware specification */
1024         if (IS_BXT_REVID(dev_priv, 0, BXT_REVID_A1))
1025                 I915_WRITE(TILECTL, I915_READ(TILECTL) | TILECTL_TLBPF);
1026
1027         /* WaSetClckGatingDisableMedia:bxt */
1028         if (IS_BXT_REVID(dev_priv, 0, BXT_REVID_A1)) {
1029                 I915_WRITE(GEN7_MISCCPCTL, (I915_READ(GEN7_MISCCPCTL) &
1030                                             ~GEN8_DOP_CLOCK_GATE_MEDIA_ENABLE));
1031         }
1032
1033         /* WaDisableThreadStallDopClockGating:bxt */
1034         WA_SET_BIT_MASKED(GEN8_ROW_CHICKEN,
1035                           STALL_DOP_GATING_DISABLE);
1036
1037         /* WaDisablePooledEuLoadBalancingFix:bxt */
1038         if (IS_BXT_REVID(dev_priv, BXT_REVID_B0, REVID_FOREVER)) {
1039                 WA_SET_BIT_MASKED(FF_SLICE_CS_CHICKEN2,
1040                                   GEN9_POOLED_EU_LOAD_BALANCING_FIX_DISABLE);
1041         }
1042
1043         /* WaDisableSbeCacheDispatchPortSharing:bxt */
1044         if (IS_BXT_REVID(dev_priv, 0, BXT_REVID_B0)) {
1045                 WA_SET_BIT_MASKED(
1046                         GEN7_HALF_SLICE_CHICKEN1,
1047                         GEN7_SBE_SS_CACHE_DISPATCH_PORT_SHARING_DISABLE);
1048         }
1049
1050         /* WaDisableObjectLevelPreemptionForTrifanOrPolygon:bxt */
1051         /* WaDisableObjectLevelPreemptionForInstancedDraw:bxt */
1052         /* WaDisableObjectLevelPreemtionForInstanceId:bxt */
1053         /* WaDisableLSQCROPERFforOCL:bxt */
1054         if (IS_BXT_REVID(dev_priv, 0, BXT_REVID_A1)) {
1055                 ret = wa_ring_whitelist_reg(engine, GEN9_CS_DEBUG_MODE1);
1056                 if (ret)
1057                         return ret;
1058
1059                 ret = wa_ring_whitelist_reg(engine, GEN8_L3SQCREG4);
1060                 if (ret)
1061                         return ret;
1062         }
1063
1064         /* WaProgramL3SqcReg1DefaultForPerf:bxt */
1065         if (IS_BXT_REVID(dev_priv, BXT_REVID_B0, REVID_FOREVER))
1066                 I915_WRITE(GEN8_L3SQCREG1, L3_GENERAL_PRIO_CREDITS(62) |
1067                                            L3_HIGH_PRIO_CREDITS(2));
1068
1069         /* WaToEnableHwFixForPushConstHWBug:bxt */
1070         if (IS_BXT_REVID(dev_priv, BXT_REVID_C0, REVID_FOREVER))
1071                 WA_SET_BIT_MASKED(COMMON_SLICE_CHICKEN2,
1072                                   GEN8_SBE_DISABLE_REPLAY_BUF_OPTIMIZATION);
1073
1074         /* WaInPlaceDecompressionHang:bxt */
1075         if (IS_BXT_REVID(dev_priv, BXT_REVID_C0, REVID_FOREVER))
1076                 WA_SET_BIT(GEN9_GAMT_ECO_REG_RW_IA,
1077                            GAMT_ECO_ENABLE_IN_PLACE_DECOMPRESS);
1078
1079         return 0;
1080 }
1081
1082 static int kbl_init_workarounds(struct intel_engine_cs *engine)
1083 {
1084         struct drm_i915_private *dev_priv = engine->i915;
1085         int ret;
1086
1087         ret = gen9_init_workarounds(engine);
1088         if (ret)
1089                 return ret;
1090
1091         /* WaEnableGapsTsvCreditFix:kbl */
1092         I915_WRITE(GEN8_GARBCNTL, (I915_READ(GEN8_GARBCNTL) |
1093                                    GEN9_GAPS_TSV_CREDIT_DISABLE));
1094
1095         /* WaDisableDynamicCreditSharing:kbl */
1096         if (IS_KBL_REVID(dev_priv, 0, KBL_REVID_B0))
1097                 WA_SET_BIT(GAMT_CHKN_BIT_REG,
1098                            GAMT_CHKN_DISABLE_DYNAMIC_CREDIT_SHARING);
1099
1100         /* WaDisableFenceDestinationToSLM:kbl (pre-prod) */
1101         if (IS_KBL_REVID(dev_priv, KBL_REVID_A0, KBL_REVID_A0))
1102                 WA_SET_BIT_MASKED(HDC_CHICKEN0,
1103                                   HDC_FENCE_DEST_SLM_DISABLE);
1104
1105         /* WaToEnableHwFixForPushConstHWBug:kbl */
1106         if (IS_KBL_REVID(dev_priv, KBL_REVID_C0, REVID_FOREVER))
1107                 WA_SET_BIT_MASKED(COMMON_SLICE_CHICKEN2,
1108                                   GEN8_SBE_DISABLE_REPLAY_BUF_OPTIMIZATION);
1109
1110         /* WaDisableGafsUnitClkGating:kbl */
1111         WA_SET_BIT(GEN7_UCGCTL4, GEN8_EU_GAUNIT_CLOCK_GATE_DISABLE);
1112
1113         /* WaDisableSbeCacheDispatchPortSharing:kbl */
1114         WA_SET_BIT_MASKED(
1115                 GEN7_HALF_SLICE_CHICKEN1,
1116                 GEN7_SBE_SS_CACHE_DISPATCH_PORT_SHARING_DISABLE);
1117
1118         /* WaInPlaceDecompressionHang:kbl */
1119         WA_SET_BIT(GEN9_GAMT_ECO_REG_RW_IA,
1120                    GAMT_ECO_ENABLE_IN_PLACE_DECOMPRESS);
1121
1122         /* WaDisableLSQCROPERFforOCL:kbl */
1123         ret = wa_ring_whitelist_reg(engine, GEN8_L3SQCREG4);
1124         if (ret)
1125                 return ret;
1126
1127         return 0;
1128 }
1129
1130 static int glk_init_workarounds(struct intel_engine_cs *engine)
1131 {
1132         struct drm_i915_private *dev_priv = engine->i915;
1133         int ret;
1134
1135         ret = gen9_init_workarounds(engine);
1136         if (ret)
1137                 return ret;
1138
1139         /* WaToEnableHwFixForPushConstHWBug:glk */
1140         WA_SET_BIT_MASKED(COMMON_SLICE_CHICKEN2,
1141                           GEN8_SBE_DISABLE_REPLAY_BUF_OPTIMIZATION);
1142
1143         return 0;
1144 }
1145
1146 static int cfl_init_workarounds(struct intel_engine_cs *engine)
1147 {
1148         struct drm_i915_private *dev_priv = engine->i915;
1149         int ret;
1150
1151         ret = gen9_init_workarounds(engine);
1152         if (ret)
1153                 return ret;
1154
1155         /* WaEnableGapsTsvCreditFix:cfl */
1156         I915_WRITE(GEN8_GARBCNTL, (I915_READ(GEN8_GARBCNTL) |
1157                                    GEN9_GAPS_TSV_CREDIT_DISABLE));
1158
1159         /* WaToEnableHwFixForPushConstHWBug:cfl */
1160         WA_SET_BIT_MASKED(COMMON_SLICE_CHICKEN2,
1161                           GEN8_SBE_DISABLE_REPLAY_BUF_OPTIMIZATION);
1162
1163         /* WaDisableGafsUnitClkGating:cfl */
1164         WA_SET_BIT(GEN7_UCGCTL4, GEN8_EU_GAUNIT_CLOCK_GATE_DISABLE);
1165
1166         /* WaDisableSbeCacheDispatchPortSharing:cfl */
1167         WA_SET_BIT_MASKED(
1168                 GEN7_HALF_SLICE_CHICKEN1,
1169                 GEN7_SBE_SS_CACHE_DISPATCH_PORT_SHARING_DISABLE);
1170
1171         /* WaInPlaceDecompressionHang:cfl */
1172         WA_SET_BIT(GEN9_GAMT_ECO_REG_RW_IA,
1173                    GAMT_ECO_ENABLE_IN_PLACE_DECOMPRESS);
1174
1175         return 0;
1176 }
1177
1178 int init_workarounds_ring(struct intel_engine_cs *engine)
1179 {
1180         struct drm_i915_private *dev_priv = engine->i915;
1181         int err;
1182
1183         WARN_ON(engine->id != RCS);
1184
1185         dev_priv->workarounds.count = 0;
1186         dev_priv->workarounds.hw_whitelist_count[engine->id] = 0;
1187
1188         if (IS_BROADWELL(dev_priv))
1189                 err = bdw_init_workarounds(engine);
1190         else if (IS_CHERRYVIEW(dev_priv))
1191                 err = chv_init_workarounds(engine);
1192         else if (IS_SKYLAKE(dev_priv))
1193                 err =  skl_init_workarounds(engine);
1194         else if (IS_BROXTON(dev_priv))
1195                 err = bxt_init_workarounds(engine);
1196         else if (IS_KABYLAKE(dev_priv))
1197                 err = kbl_init_workarounds(engine);
1198         else if (IS_GEMINILAKE(dev_priv))
1199                 err =  glk_init_workarounds(engine);
1200         else if (IS_COFFEELAKE(dev_priv))
1201                 err = cfl_init_workarounds(engine);
1202         else
1203                 err = 0;
1204         if (err)
1205                 return err;
1206
1207         DRM_DEBUG_DRIVER("%s: Number of context specific w/a: %d\n",
1208                          engine->name, dev_priv->workarounds.count);
1209         return 0;
1210 }
1211
1212 int intel_ring_workarounds_emit(struct drm_i915_gem_request *req)
1213 {
1214         struct i915_workarounds *w = &req->i915->workarounds;
1215         u32 *cs;
1216         int ret, i;
1217
1218         if (w->count == 0)
1219                 return 0;
1220
1221         ret = req->engine->emit_flush(req, EMIT_BARRIER);
1222         if (ret)
1223                 return ret;
1224
1225         cs = intel_ring_begin(req, (w->count * 2 + 2));
1226         if (IS_ERR(cs))
1227                 return PTR_ERR(cs);
1228
1229         *cs++ = MI_LOAD_REGISTER_IMM(w->count);
1230         for (i = 0; i < w->count; i++) {
1231                 *cs++ = i915_mmio_reg_offset(w->reg[i].addr);
1232                 *cs++ = w->reg[i].value;
1233         }
1234         *cs++ = MI_NOOP;
1235
1236         intel_ring_advance(req, cs);
1237
1238         ret = req->engine->emit_flush(req, EMIT_BARRIER);
1239         if (ret)
1240                 return ret;
1241
1242         return 0;
1243 }
1244
1245 static bool ring_is_idle(struct intel_engine_cs *engine)
1246 {
1247         struct drm_i915_private *dev_priv = engine->i915;
1248         bool idle = true;
1249
1250         intel_runtime_pm_get(dev_priv);
1251
1252         /* First check that no commands are left in the ring */
1253         if ((I915_READ_HEAD(engine) & HEAD_ADDR) !=
1254             (I915_READ_TAIL(engine) & TAIL_ADDR))
1255                 idle = false;
1256
1257         /* No bit for gen2, so assume the CS parser is idle */
1258         if (INTEL_GEN(dev_priv) > 2 && !(I915_READ_MODE(engine) & MODE_IDLE))
1259                 idle = false;
1260
1261         intel_runtime_pm_put(dev_priv);
1262
1263         return idle;
1264 }
1265
1266 /**
1267  * intel_engine_is_idle() - Report if the engine has finished process all work
1268  * @engine: the intel_engine_cs
1269  *
1270  * Return true if there are no requests pending, nothing left to be submitted
1271  * to hardware, and that the engine is idle.
1272  */
1273 bool intel_engine_is_idle(struct intel_engine_cs *engine)
1274 {
1275         struct drm_i915_private *dev_priv = engine->i915;
1276
1277         /* More white lies, if wedged, hw state is inconsistent */
1278         if (i915_terminally_wedged(&dev_priv->gpu_error))
1279                 return true;
1280
1281         /* Any inflight/incomplete requests? */
1282         if (!i915_seqno_passed(intel_engine_get_seqno(engine),
1283                                intel_engine_last_submit(engine)))
1284                 return false;
1285
1286         if (I915_SELFTEST_ONLY(engine->breadcrumbs.mock))
1287                 return true;
1288
1289         /* Interrupt/tasklet pending? */
1290         if (test_bit(ENGINE_IRQ_EXECLIST, &engine->irq_posted))
1291                 return false;
1292
1293         /* Both ports drained, no more ELSP submission? */
1294         if (port_request(&engine->execlist_port[0]))
1295                 return false;
1296
1297         /* Ring stopped? */
1298         if (!ring_is_idle(engine))
1299                 return false;
1300
1301         return true;
1302 }
1303
1304 bool intel_engines_are_idle(struct drm_i915_private *dev_priv)
1305 {
1306         struct intel_engine_cs *engine;
1307         enum intel_engine_id id;
1308
1309         if (READ_ONCE(dev_priv->gt.active_requests))
1310                 return false;
1311
1312         /* If the driver is wedged, HW state may be very inconsistent and
1313          * report that it is still busy, even though we have stopped using it.
1314          */
1315         if (i915_terminally_wedged(&dev_priv->gpu_error))
1316                 return true;
1317
1318         for_each_engine(engine, dev_priv, id) {
1319                 if (!intel_engine_is_idle(engine))
1320                         return false;
1321         }
1322
1323         return true;
1324 }
1325
1326 void intel_engines_reset_default_submission(struct drm_i915_private *i915)
1327 {
1328         struct intel_engine_cs *engine;
1329         enum intel_engine_id id;
1330
1331         for_each_engine(engine, i915, id)
1332                 engine->set_default_submission(engine);
1333 }
1334
1335 void intel_engines_mark_idle(struct drm_i915_private *i915)
1336 {
1337         struct intel_engine_cs *engine;
1338         enum intel_engine_id id;
1339
1340         for_each_engine(engine, i915, id) {
1341                 intel_engine_disarm_breadcrumbs(engine);
1342                 i915_gem_batch_pool_fini(&engine->batch_pool);
1343                 engine->no_priolist = false;
1344         }
1345 }
1346
1347 #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
1348 #include "selftests/mock_engine.c"
1349 #endif