drivers/gpu/drm/i915/i915_scheduler.c

   1 /*
   2  * SPDX-License-Identifier: MIT
   3  *
   4  * Copyright © 2018 Intel Corporation
   5  */
   6
   7 #include <linux/mutex.h>
   8
   9 #include "i915_drv.h"
  10 #include "i915_globals.h"
  11 #include "i915_request.h"
  12 #include "i915_scheduler.h"
  13
  14 static struct i915_global_scheduler {
  15         struct i915_global base;
  16         struct kmem_cache *slab_dependencies;
  17         struct kmem_cache *slab_priorities;
  18 } global;
  19
  20 static DEFINE_SPINLOCK(schedule_lock);
  21
  22 static const struct i915_request *
  23 node_to_request(const struct i915_sched_node *node)
  24 {
  25         return container_of(node, const struct i915_request, sched);
  26 }
  27
  28 static inline bool node_started(const struct i915_sched_node *node)
  29 {
  30         return i915_request_started(node_to_request(node));
  31 }
  32
  33 static inline bool node_signaled(const struct i915_sched_node *node)
  34 {
  35         return i915_request_completed(node_to_request(node));
  36 }
  37
  38 static inline struct i915_priolist *to_priolist(struct rb_node *rb)
  39 {
  40         return rb_entry(rb, struct i915_priolist, node);
  41 }
  42
  43 static void assert_priolists(struct intel_engine_execlists * const execlists)
  44 {
  45         struct rb_node *rb;
  46         long last_prio, i;
  47
  48         if (!IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM))
  49                 return;
  50
  51         GEM_BUG_ON(rb_first_cached(&execlists->queue) !=
  52                    rb_first(&execlists->queue.rb_root));
  53
  54         last_prio = (INT_MAX >> I915_USER_PRIORITY_SHIFT) + 1;
  55         for (rb = rb_first_cached(&execlists->queue); rb; rb = rb_next(rb)) {
  56                 const struct i915_priolist *p = to_priolist(rb);
  57
  58                 GEM_BUG_ON(p->priority >= last_prio);
  59                 last_prio = p->priority;
  60
  61                 GEM_BUG_ON(!p->used);
  62                 for (i = 0; i < ARRAY_SIZE(p->requests); i++) {
  63                         if (list_empty(&p->requests[i]))
  64                                 continue;
  65
  66                         GEM_BUG_ON(!(p->used & BIT(i)));
  67                 }
  68         }
  69 }
  70
  71 struct list_head *
  72 i915_sched_lookup_priolist(struct intel_engine_cs *engine, int prio)
  73 {
  74         struct intel_engine_execlists * const execlists = &engine->execlists;
  75         struct i915_priolist *p;
  76         struct rb_node **parent, *rb;
  77         bool first = true;
  78         int idx, i;
  79
  80         lockdep_assert_held(&engine->timeline.lock);
  81         assert_priolists(execlists);
  82
  83         /* buckets sorted from highest [in slot 0] to lowest priority */
  84         idx = I915_PRIORITY_COUNT - (prio & I915_PRIORITY_MASK) - 1;
  85         prio >>= I915_USER_PRIORITY_SHIFT;
  86         if (unlikely(execlists->no_priolist))
  87                 prio = I915_PRIORITY_NORMAL;
  88
  89 find_priolist:
  90         /* most positive priority is scheduled first, equal priorities fifo */
  91         rb = NULL;
  92         parent = &execlists->queue.rb_root.rb_node;
  93         while (*parent) {
  94                 rb = *parent;
  95                 p = to_priolist(rb);
  96                 if (prio > p->priority) {
  97                         parent = &rb->rb_left;
  98                 } else if (prio < p->priority) {
  99                         parent = &rb->rb_right;
 100                         first = false;
 101                 } else {
 102                         goto out;
 103                 }
 104         }
 105
 106         if (prio == I915_PRIORITY_NORMAL) {
 107                 p = &execlists->default_priolist;
 108         } else {
 109                 p = kmem_cache_alloc(global.slab_priorities, GFP_ATOMIC);
 110                 /* Convert an allocation failure to a priority bump */
 111                 if (unlikely(!p)) {
 112                         prio = I915_PRIORITY_NORMAL; /* recurses just once */
 113
 114                         /* To maintain ordering with all rendering, after an
 115                          * allocation failure we have to disable all scheduling.
 116                          * Requests will then be executed in fifo, and schedule
 117                          * will ensure that dependencies are emitted in fifo.
 118                          * There will be still some reordering with existing
 119                          * requests, so if userspace lied about their
 120                          * dependencies that reordering may be visible.
 121                          */
 122                         execlists->no_priolist = true;
 123                         goto find_priolist;
 124                 }
 125         }
 126
 127         p->priority = prio;
 128         for (i = 0; i < ARRAY_SIZE(p->requests); i++)
 129                 INIT_LIST_HEAD(&p->requests[i]);
 130         rb_link_node(&p->node, rb, parent);
 131         rb_insert_color_cached(&p->node, &execlists->queue, first);
 132         p->used = 0;
 133
 134 out:
 135         p->used |= BIT(idx);
 136         return &p->requests[idx];
 137 }
 138
 139 void __i915_priolist_free(struct i915_priolist *p)
 140 {
 141         kmem_cache_free(global.slab_priorities, p);
 142 }
 143
 144 struct sched_cache {
 145         struct list_head *priolist;
 146 };
 147
 148 static struct intel_engine_cs *
 149 sched_lock_engine(const struct i915_sched_node *node,
 150                   struct intel_engine_cs *locked,
 151                   struct sched_cache *cache)
 152 {
 153         struct intel_engine_cs *engine = node_to_request(node)->engine;
 154
 155         GEM_BUG_ON(!locked);
 156
 157         if (engine != locked) {
 158                 spin_unlock(&locked->timeline.lock);
 159                 memset(cache, 0, sizeof(*cache));
 160                 spin_lock(&engine->timeline.lock);
 161         }
 162
 163         return engine;
 164 }
 165
 166 static bool inflight(const struct i915_request *rq,
 167                      const struct intel_engine_cs *engine)
 168 {
 169         const struct i915_request *active;
 170
 171         if (!i915_request_is_active(rq))
 172                 return false;
 173
 174         active = port_request(engine->execlists.port);
 175         return active->hw_context == rq->hw_context;
 176 }
 177
 178 static void __i915_schedule(struct i915_sched_node *node,
 179                             const struct i915_sched_attr *attr)
 180 {
 181         struct intel_engine_cs *engine;
 182         struct i915_dependency *dep, *p;
 183         struct i915_dependency stack;
 184         const int prio = attr->priority;
 185         struct sched_cache cache;
 186         LIST_HEAD(dfs);
 187
 188         /* Needed in order to use the temporary link inside i915_dependency */
 189         lockdep_assert_held(&schedule_lock);
 190         GEM_BUG_ON(prio == I915_PRIORITY_INVALID);
 191
 192         if (node_signaled(node))
 193                 return;
 194
 195         if (prio <= READ_ONCE(node->attr.priority))
 196                 return;
 197
 198         stack.signaler = node;
 199         list_add(&stack.dfs_link, &dfs);
 200
 201         /*
 202          * Recursively bump all dependent priorities to match the new request.
 203          *
 204          * A naive approach would be to use recursion:
 205          * static void update_priorities(struct i915_sched_node *node, prio) {
 206          *      list_for_each_entry(dep, &node->signalers_list, signal_link)
 207          *              update_priorities(dep->signal, prio)
 208          *      queue_request(node);
 209          * }
 210          * but that may have unlimited recursion depth and so runs a very
 211          * real risk of overunning the kernel stack. Instead, we build
 212          * a flat list of all dependencies starting with the current request.
 213          * As we walk the list of dependencies, we add all of its dependencies
 214          * to the end of the list (this may include an already visited
 215          * request) and continue to walk onwards onto the new dependencies. The
 216          * end result is a topological list of requests in reverse order, the
 217          * last element in the list is the request we must execute first.
 218          */
 219         list_for_each_entry(dep, &dfs, dfs_link) {
 220                 struct i915_sched_node *node = dep->signaler;
 221
 222                 /* If we are already flying, we know we have no signalers */
 223                 if (node_started(node))
 224                         continue;
 225
 226                 /*
 227                  * Within an engine, there can be no cycle, but we may
 228                  * refer to the same dependency chain multiple times
 229                  * (redundant dependencies are not eliminated) and across
 230                  * engines.
 231                  */
 232                 list_for_each_entry(p, &node->signalers_list, signal_link) {
 233                         GEM_BUG_ON(p == dep); /* no cycles! */
 234
 235                         if (node_signaled(p->signaler))
 236                                 continue;
 237
 238                         if (prio > READ_ONCE(p->signaler->attr.priority))
 239                                 list_move_tail(&p->dfs_link, &dfs);
 240                 }
 241         }
 242
 243         /*
 244          * If we didn't need to bump any existing priorities, and we haven't
 245          * yet submitted this request (i.e. there is no potential race with
 246          * execlists_submit_request()), we can set our own priority and skip
 247          * acquiring the engine locks.
 248          */
 249         if (node->attr.priority == I915_PRIORITY_INVALID) {
 250                 GEM_BUG_ON(!list_empty(&node->link));
 251                 node->attr = *attr;
 252
 253                 if (stack.dfs_link.next == stack.dfs_link.prev)
 254                         return;
 255
 256                 __list_del_entry(&stack.dfs_link);
 257         }
 258
 259         memset(&cache, 0, sizeof(cache));
 260         engine = node_to_request(node)->engine;
 261         spin_lock(&engine->timeline.lock);
 262
 263         /* Fifo and depth-first replacement ensure our deps execute before us */
 264         list_for_each_entry_safe_reverse(dep, p, &dfs, dfs_link) {
 265                 INIT_LIST_HEAD(&dep->dfs_link);
 266
 267                 node = dep->signaler;
 268                 engine = sched_lock_engine(node, engine, &cache);
 269                 lockdep_assert_held(&engine->timeline.lock);
 270
 271                 /* Recheck after acquiring the engine->timeline.lock */
 272                 if (prio <= node->attr.priority || node_signaled(node))
 273                         continue;
 274
 275                 node->attr.priority = prio;
 276                 if (!list_empty(&node->link)) {
 277                         if (!cache.priolist)
 278                                 cache.priolist =
 279                                         i915_sched_lookup_priolist(engine,
 280                                                                    prio);
 281                         list_move_tail(&node->link, cache.priolist);
 282                 } else {
 283                         /*
 284                          * If the request is not in the priolist queue because
 285                          * it is not yet runnable, then it doesn't contribute
 286                          * to our preemption decisions. On the other hand,
 287                          * if the request is on the HW, it too is not in the
 288                          * queue; but in that case we may still need to reorder
 289                          * the inflight requests.
 290                          */
 291                         if (!i915_sw_fence_done(&node_to_request(node)->submit))
 292                                 continue;
 293                 }
 294
 295                 if (prio <= engine->execlists.queue_priority_hint)
 296                         continue;
 297
 298                 engine->execlists.queue_priority_hint = prio;
 299
 300                 /*
 301                  * If we are already the currently executing context, don't
 302                  * bother evaluating if we should preempt ourselves.
 303                  */
 304                 if (inflight(node_to_request(node), engine))
 305                         continue;
 306
 307                 /* Defer (tasklet) submission until after all of our updates. */
 308                 tasklet_hi_schedule(&engine->execlists.tasklet);
 309         }
 310
 311         spin_unlock(&engine->timeline.lock);
 312 }
 313
 314 void i915_schedule(struct i915_request *rq, const struct i915_sched_attr *attr)
 315 {
 316         spin_lock_irq(&schedule_lock);
 317         __i915_schedule(&rq->sched, attr);
 318         spin_unlock_irq(&schedule_lock);
 319 }
 320
 321 static void __bump_priority(struct i915_sched_node *node, unsigned int bump)
 322 {
 323         struct i915_sched_attr attr = node->attr;
 324
 325         attr.priority |= bump;
 326         __i915_schedule(node, &attr);
 327 }
 328
 329 void i915_schedule_bump_priority(struct i915_request *rq, unsigned int bump)
 330 {
 331         unsigned long flags;
 332
 333         GEM_BUG_ON(bump & ~I915_PRIORITY_MASK);
 334
 335         if (READ_ONCE(rq->sched.attr.priority) == I915_PRIORITY_INVALID)
 336                 return;
 337
 338         spin_lock_irqsave(&schedule_lock, flags);
 339         __bump_priority(&rq->sched, bump);
 340         spin_unlock_irqrestore(&schedule_lock, flags);
 341 }
 342
 343 void i915_sched_node_init(struct i915_sched_node *node)
 344 {
 345         INIT_LIST_HEAD(&node->signalers_list);
 346         INIT_LIST_HEAD(&node->waiters_list);
 347         INIT_LIST_HEAD(&node->link);
 348         node->attr.priority = I915_PRIORITY_INVALID;
 349         node->semaphores = 0;
 350         node->flags = 0;
 351 }
 352
 353 static struct i915_dependency *
 354 i915_dependency_alloc(void)
 355 {
 356         return kmem_cache_alloc(global.slab_dependencies, GFP_KERNEL);
 357 }
 358
 359 static void
 360 i915_dependency_free(struct i915_dependency *dep)
 361 {
 362         kmem_cache_free(global.slab_dependencies, dep);
 363 }
 364
 365 bool __i915_sched_node_add_dependency(struct i915_sched_node *node,
 366                                       struct i915_sched_node *signal,
 367                                       struct i915_dependency *dep,
 368                                       unsigned long flags)
 369 {
 370         bool ret = false;
 371
 372         spin_lock_irq(&schedule_lock);
 373
 374         if (!node_signaled(signal)) {
 375                 INIT_LIST_HEAD(&dep->dfs_link);
 376                 list_add(&dep->wait_link, &signal->waiters_list);
 377                 list_add(&dep->signal_link, &node->signalers_list);
 378                 dep->signaler = signal;
 379                 dep->flags = flags;
 380
 381                 /* Keep track of whether anyone on this chain has a semaphore */
 382                 if (signal->flags & I915_SCHED_HAS_SEMAPHORE_CHAIN &&
 383                     !node_started(signal))
 384                         node->flags |= I915_SCHED_HAS_SEMAPHORE_CHAIN;
 385
 386                 /*
 387                  * As we do not allow WAIT to preempt inflight requests,
 388                  * once we have executed a request, along with triggering
 389                  * any execution callbacks, we must preserve its ordering
 390                  * within the non-preemptible FIFO.
 391                  */
 392                 BUILD_BUG_ON(__NO_PREEMPTION & ~I915_PRIORITY_MASK);
 393                 if (flags & I915_DEPENDENCY_EXTERNAL)
 394                         __bump_priority(signal, __NO_PREEMPTION);
 395
 396                 ret = true;
 397         }
 398
 399         spin_unlock_irq(&schedule_lock);
 400
 401         return ret;
 402 }
 403
 404 int i915_sched_node_add_dependency(struct i915_sched_node *node,
 405                                    struct i915_sched_node *signal)
 406 {
 407         struct i915_dependency *dep;
 408
 409         dep = i915_dependency_alloc();
 410         if (!dep)
 411                 return -ENOMEM;
 412
 413         if (!__i915_sched_node_add_dependency(node, signal, dep,
 414                                               I915_DEPENDENCY_EXTERNAL |
 415                                               I915_DEPENDENCY_ALLOC))
 416                 i915_dependency_free(dep);
 417
 418         return 0;
 419 }
 420
 421 void i915_sched_node_fini(struct i915_sched_node *node)
 422 {
 423         struct i915_dependency *dep, *tmp;
 424
 425         GEM_BUG_ON(!list_empty(&node->link));
 426
 427         spin_lock_irq(&schedule_lock);
 428
 429         /*
 430          * Everyone we depended upon (the fences we wait to be signaled)
 431          * should retire before us and remove themselves from our list.
 432          * However, retirement is run independently on each timeline and
 433          * so we may be called out-of-order.
 434          */
 435         list_for_each_entry_safe(dep, tmp, &node->signalers_list, signal_link) {
 436                 GEM_BUG_ON(!node_signaled(dep->signaler));
 437                 GEM_BUG_ON(!list_empty(&dep->dfs_link));
 438
 439                 list_del(&dep->wait_link);
 440                 if (dep->flags & I915_DEPENDENCY_ALLOC)
 441                         i915_dependency_free(dep);
 442         }
 443
 444         /* Remove ourselves from everyone who depends upon us */
 445         list_for_each_entry_safe(dep, tmp, &node->waiters_list, wait_link) {
 446                 GEM_BUG_ON(dep->signaler != node);
 447                 GEM_BUG_ON(!list_empty(&dep->dfs_link));
 448
 449                 list_del(&dep->signal_link);
 450                 if (dep->flags & I915_DEPENDENCY_ALLOC)
 451                         i915_dependency_free(dep);
 452         }
 453
 454         spin_unlock_irq(&schedule_lock);
 455 }
 456
 457 static void i915_global_scheduler_shrink(void)
 458 {
 459         kmem_cache_shrink(global.slab_dependencies);
 460         kmem_cache_shrink(global.slab_priorities);
 461 }
 462
 463 static void i915_global_scheduler_exit(void)
 464 {
 465         kmem_cache_destroy(global.slab_dependencies);
 466         kmem_cache_destroy(global.slab_priorities);
 467 }
 468
 469 static struct i915_global_scheduler global = { {
 470         .shrink = i915_global_scheduler_shrink,
 471         .exit = i915_global_scheduler_exit,
 472 } };
 473
 474 int __init i915_global_scheduler_init(void)
 475 {
 476         global.slab_dependencies = KMEM_CACHE(i915_dependency,
 477                                               SLAB_HWCACHE_ALIGN);
 478         if (!global.slab_dependencies)
 479                 return -ENOMEM;
 480
 481         global.slab_priorities = KMEM_CACHE(i915_priolist,
 482                                             SLAB_HWCACHE_ALIGN);
 483         if (!global.slab_priorities)
 484                 goto err_priorities;
 485
 486         i915_global_register(&global.base);
 487         return 0;
 488
 489 err_priorities:
 490         kmem_cache_destroy(global.slab_priorities);
 491         return -ENOMEM;
 492 }