arch/powerpc/platforms/cell/spufs/sched.c

   1 /* sched.c - SPU scheduler.
   2  *
   3  * Copyright (C) IBM 2005
   4  * Author: Mark Nutter <mnutter@us.ibm.com>
   5  *
   6  * 2006-03-31   NUMA domains added.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License as published by
  10  * the Free Software Foundation; either version 2, or (at your option)
  11  * any later version.
  12  *
  13  * This program is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  16  * GNU General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * along with this program; if not, write to the Free Software
  20  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  21  */
  22
  23 #undef DEBUG
  24
  25 #include <linux/module.h>
  26 #include <linux/errno.h>
  27 #include <linux/sched.h>
  28 #include <linux/kernel.h>
  29 #include <linux/mm.h>
  30 #include <linux/completion.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/smp.h>
  33 #include <linux/stddef.h>
  34 #include <linux/unistd.h>
  35 #include <linux/numa.h>
  36 #include <linux/mutex.h>
  37 #include <linux/notifier.h>
  38 #include <linux/kthread.h>
  39
  40 #include <asm/io.h>
  41 #include <asm/mmu_context.h>
  42 #include <asm/spu.h>
  43 #include <asm/spu_csa.h>
  44 #include <asm/spu_priv1.h>
  45 #include "spufs.h"
  46
  47 struct spu_prio_array {
  48         DECLARE_BITMAP(bitmap, MAX_PRIO);
  49         struct list_head runq[MAX_PRIO];
  50         spinlock_t runq_lock;
  51         struct list_head active_list[MAX_NUMNODES];
  52         struct mutex active_mutex[MAX_NUMNODES];
  53 };
  54
  55 static struct spu_prio_array *spu_prio;
  56 static struct task_struct *spusched_task;
  57 static struct timer_list spusched_timer;
  58
  59 /*
  60  * Priority of a normal, non-rt, non-niced'd process (aka nice level 0).
  61  */
  62 #define NORMAL_PRIO             120
  63
  64 /*
  65  * Frequency of the spu scheduler tick.  By default we do one SPU scheduler
  66  * tick for every 10 CPU scheduler ticks.
  67  */
  68 #define SPUSCHED_TICK           (10)
  69
  70 /*
  71  * These are the 'tuning knobs' of the scheduler:
  72  *
  73  * Minimum timeslice is 5 msecs (or 1 spu scheduler tick, whichever is
  74  * larger), default timeslice is 100 msecs, maximum timeslice is 800 msecs.
  75  */
  76 #define MIN_SPU_TIMESLICE       max(5 * HZ / (1000 * SPUSCHED_TICK), 1)
  77 #define DEF_SPU_TIMESLICE       (100 * HZ / (1000 * SPUSCHED_TICK))
  78
  79 #define MAX_USER_PRIO           (MAX_PRIO - MAX_RT_PRIO)
  80 #define SCALE_PRIO(x, prio) \
  81         max(x * (MAX_PRIO - prio) / (MAX_USER_PRIO / 2), MIN_SPU_TIMESLICE)
  82
  83 /*
  84  * scale user-nice values [ -20 ... 0 ... 19 ] to time slice values:
  85  * [800ms ... 100ms ... 5ms]
  86  *
  87  * The higher a thread's priority, the bigger timeslices
  88  * it gets during one round of execution. But even the lowest
  89  * priority thread gets MIN_TIMESLICE worth of execution time.
  90  */
  91 void spu_set_timeslice(struct spu_context *ctx)
  92 {
  93         if (ctx->prio < NORMAL_PRIO)
  94                 ctx->time_slice = SCALE_PRIO(DEF_SPU_TIMESLICE * 4, ctx->prio);
  95         else
  96                 ctx->time_slice = SCALE_PRIO(DEF_SPU_TIMESLICE, ctx->prio);
  97 }
  98
  99 static inline int node_allowed(int node)
 100 {
 101         cpumask_t mask;
 102
 103         if (!nr_cpus_node(node))
 104                 return 0;
 105         mask = node_to_cpumask(node);
 106         if (!cpus_intersects(mask, current->cpus_allowed))
 107                 return 0;
 108         return 1;
 109 }
 110
 111 /**
 112  * spu_add_to_active_list - add spu to active list
 113  * @spu:        spu to add to the active list
 114  */
 115 static void spu_add_to_active_list(struct spu *spu)
 116 {
 117         mutex_lock(&spu_prio->active_mutex[spu->node]);
 118         list_add_tail(&spu->list, &spu_prio->active_list[spu->node]);
 119         mutex_unlock(&spu_prio->active_mutex[spu->node]);
 120 }
 121
 122 static void __spu_remove_from_active_list(struct spu *spu)
 123 {
 124         list_del_init(&spu->list);
 125 }
 126
 127 /**
 128  * spu_remove_from_active_list - remove spu from active list
 129  * @spu:       spu to remove from the active list
 130  */
 131 static void spu_remove_from_active_list(struct spu *spu)
 132 {
 133         int node = spu->node;
 134
 135         mutex_lock(&spu_prio->active_mutex[node]);
 136         __spu_remove_from_active_list(spu);
 137         mutex_unlock(&spu_prio->active_mutex[node]);
 138 }
 139
 140 static BLOCKING_NOTIFIER_HEAD(spu_switch_notifier);
 141
 142 static void spu_switch_notify(struct spu *spu, struct spu_context *ctx)
 143 {
 144         blocking_notifier_call_chain(&spu_switch_notifier,
 145                             ctx ? ctx->object_id : 0, spu);
 146 }
 147
 148 int spu_switch_event_register(struct notifier_block * n)
 149 {
 150         return blocking_notifier_chain_register(&spu_switch_notifier, n);
 151 }
 152
 153 int spu_switch_event_unregister(struct notifier_block * n)
 154 {
 155         return blocking_notifier_chain_unregister(&spu_switch_notifier, n);
 156 }
 157
 158 /**
 159  * spu_bind_context - bind spu context to physical spu
 160  * @spu:        physical spu to bind to
 161  * @ctx:        context to bind
 162  */
 163 static void spu_bind_context(struct spu *spu, struct spu_context *ctx)
 164 {
 165         pr_debug("%s: pid=%d SPU=%d NODE=%d\n", __FUNCTION__, current->pid,
 166                  spu->number, spu->node);
 167         spu->ctx = ctx;
 168         spu->flags = 0;
 169         ctx->spu = spu;
 170         ctx->ops = &spu_hw_ops;
 171         spu->pid = current->pid;
 172         spu_associate_mm(spu, ctx->owner);
 173         spu->ibox_callback = spufs_ibox_callback;
 174         spu->wbox_callback = spufs_wbox_callback;
 175         spu->stop_callback = spufs_stop_callback;
 176         spu->mfc_callback = spufs_mfc_callback;
 177         spu->dma_callback = spufs_dma_callback;
 178         mb();
 179         spu_unmap_mappings(ctx);
 180         spu_restore(&ctx->csa, spu);
 181         spu->timestamp = jiffies;
 182         spu_cpu_affinity_set(spu, raw_smp_processor_id());
 183         spu_switch_notify(spu, ctx);
 184         ctx->state = SPU_STATE_RUNNABLE;
 185 }
 186
 187 /**
 188  * spu_unbind_context - unbind spu context from physical spu
 189  * @spu:        physical spu to unbind from
 190  * @ctx:        context to unbind
 191  */
 192 static void spu_unbind_context(struct spu *spu, struct spu_context *ctx)
 193 {
 194         pr_debug("%s: unbind pid=%d SPU=%d NODE=%d\n", __FUNCTION__,
 195                  spu->pid, spu->number, spu->node);
 196
 197         spu_switch_notify(spu, NULL);
 198         spu_unmap_mappings(ctx);
 199         spu_save(&ctx->csa, spu);
 200         spu->timestamp = jiffies;
 201         ctx->state = SPU_STATE_SAVED;
 202         spu->ibox_callback = NULL;
 203         spu->wbox_callback = NULL;
 204         spu->stop_callback = NULL;
 205         spu->mfc_callback = NULL;
 206         spu->dma_callback = NULL;
 207         spu_associate_mm(spu, NULL);
 208         spu->pid = 0;
 209         ctx->ops = &spu_backing_ops;
 210         ctx->spu = NULL;
 211         spu->flags = 0;
 212         spu->ctx = NULL;
 213 }
 214
 215 /**
 216  * spu_add_to_rq - add a context to the runqueue
 217  * @ctx:       context to add
 218  */
 219 static void __spu_add_to_rq(struct spu_context *ctx)
 220 {
 221         int prio = ctx->prio;
 222
 223         list_add_tail(&ctx->rq, &spu_prio->runq[prio]);
 224         set_bit(prio, spu_prio->bitmap);
 225 }
 226
 227 static void __spu_del_from_rq(struct spu_context *ctx)
 228 {
 229         int prio = ctx->prio;
 230
 231         if (!list_empty(&ctx->rq))
 232                 list_del_init(&ctx->rq);
 233         if (list_empty(&spu_prio->runq[prio]))
 234                 clear_bit(prio, spu_prio->bitmap);
 235 }
 236
 237 static void spu_prio_wait(struct spu_context *ctx)
 238 {
 239         DEFINE_WAIT(wait);
 240
 241         spin_lock(&spu_prio->runq_lock);
 242         prepare_to_wait_exclusive(&ctx->stop_wq, &wait, TASK_INTERRUPTIBLE);
 243         if (!signal_pending(current)) {
 244                 __spu_add_to_rq(ctx);
 245                 spin_unlock(&spu_prio->runq_lock);
 246                 mutex_unlock(&ctx->state_mutex);
 247                 schedule();
 248                 mutex_lock(&ctx->state_mutex);
 249                 spin_lock(&spu_prio->runq_lock);
 250                 __spu_del_from_rq(ctx);
 251         }
 252         spin_unlock(&spu_prio->runq_lock);
 253         __set_current_state(TASK_RUNNING);
 254         remove_wait_queue(&ctx->stop_wq, &wait);
 255 }
 256
 257 static struct spu *spu_get_idle(struct spu_context *ctx)
 258 {
 259         struct spu *spu = NULL;
 260         int node = cpu_to_node(raw_smp_processor_id());
 261         int n;
 262
 263         for (n = 0; n < MAX_NUMNODES; n++, node++) {
 264                 node = (node < MAX_NUMNODES) ? node : 0;
 265                 if (!node_allowed(node))
 266                         continue;
 267                 spu = spu_alloc_node(node);
 268                 if (spu)
 269                         break;
 270         }
 271         return spu;
 272 }
 273
 274 /**
 275  * find_victim - find a lower priority context to preempt
 276  * @ctx:        canidate context for running
 277  *
 278  * Returns the freed physical spu to run the new context on.
 279  */
 280 static struct spu *find_victim(struct spu_context *ctx)
 281 {
 282         struct spu_context *victim = NULL;
 283         struct spu *spu;
 284         int node, n;
 285
 286         /*
 287          * Look for a possible preemption candidate on the local node first.
 288          * If there is no candidate look at the other nodes.  This isn't
 289          * exactly fair, but so far the whole spu schedule tries to keep
 290          * a strong node affinity.  We might want to fine-tune this in
 291          * the future.
 292          */
 293  restart:
 294         node = cpu_to_node(raw_smp_processor_id());
 295         for (n = 0; n < MAX_NUMNODES; n++, node++) {
 296                 node = (node < MAX_NUMNODES) ? node : 0;
 297                 if (!node_allowed(node))
 298                         continue;
 299
 300                 mutex_lock(&spu_prio->active_mutex[node]);
 301                 list_for_each_entry(spu, &spu_prio->active_list[node], list) {
 302                         struct spu_context *tmp = spu->ctx;
 303
 304                         if (tmp->prio > ctx->prio &&
 305                             (!victim || tmp->prio > victim->prio))
 306                                 victim = spu->ctx;
 307                 }
 308                 mutex_unlock(&spu_prio->active_mutex[node]);
 309
 310                 if (victim) {
 311                         /*
 312                          * This nests ctx->state_mutex, but we always lock
 313                          * higher priority contexts before lower priority
 314                          * ones, so this is safe until we introduce
 315                          * priority inheritance schemes.
 316                          */
 317                         if (!mutex_trylock(&victim->state_mutex)) {
 318                                 victim = NULL;
 319                                 goto restart;
 320                         }
 321
 322                         spu = victim->spu;
 323                         if (!spu) {
 324                                 /*
 325                                  * This race can happen because we've dropped
 326                                  * the active list mutex.  No a problem, just
 327                                  * restart the search.
 328                                  */
 329                                 mutex_unlock(&victim->state_mutex);
 330                                 victim = NULL;
 331                                 goto restart;
 332                         }
 333                         spu_remove_from_active_list(spu);
 334                         spu_unbind_context(spu, victim);
 335                         mutex_unlock(&victim->state_mutex);
 336                         /*
 337                          * We need to break out of the wait loop in spu_run
 338                          * manually to ensure this context gets put on the
 339                          * runqueue again ASAP.
 340                          */
 341                         wake_up(&victim->stop_wq);
 342                         return spu;
 343                 }
 344         }
 345
 346         return NULL;
 347 }
 348
 349 /**
 350  * spu_activate - find a free spu for a context and execute it
 351  * @ctx:        spu context to schedule
 352  * @flags:      flags (currently ignored)
 353  *
 354  * Tries to find a free spu to run @ctx.  If no free spu is available
 355  * add the context to the runqueue so it gets woken up once an spu
 356  * is available.
 357  */
 358 int spu_activate(struct spu_context *ctx, unsigned long flags)
 359 {
 360
 361         if (ctx->spu)
 362                 return 0;
 363
 364         do {
 365                 struct spu *spu;
 366
 367                 spu = spu_get_idle(ctx);
 368                 /*
 369                  * If this is a realtime thread we try to get it running by
 370                  * preempting a lower priority thread.
 371                  */
 372                 if (!spu && rt_prio(ctx->prio))
 373                         spu = find_victim(ctx);
 374                 if (spu) {
 375                         spu_bind_context(spu, ctx);
 376                         spu_add_to_active_list(spu);
 377                         return 0;
 378                 }
 379
 380                 spu_prio_wait(ctx);
 381         } while (!signal_pending(current));
 382
 383         return -ERESTARTSYS;
 384 }
 385
 386 /**
 387  * grab_runnable_context - try to find a runnable context
 388  *
 389  * Remove the highest priority context on the runqueue and return it
 390  * to the caller.  Returns %NULL if no runnable context was found.
 391  */
 392 static struct spu_context *grab_runnable_context(int prio)
 393 {
 394         struct spu_context *ctx = NULL;
 395         int best;
 396
 397         spin_lock(&spu_prio->runq_lock);
 398         best = sched_find_first_bit(spu_prio->bitmap);
 399         if (best < prio) {
 400                 struct list_head *rq = &spu_prio->runq[best];
 401
 402                 BUG_ON(list_empty(rq));
 403
 404                 ctx = list_entry(rq->next, struct spu_context, rq);
 405                 __spu_del_from_rq(ctx);
 406         }
 407         spin_unlock(&spu_prio->runq_lock);
 408
 409         return ctx;
 410 }
 411
 412 static int __spu_deactivate(struct spu_context *ctx, int force, int max_prio)
 413 {
 414         struct spu *spu = ctx->spu;
 415         struct spu_context *new = NULL;
 416
 417         if (spu) {
 418                 new = grab_runnable_context(max_prio);
 419                 if (new || force) {
 420                         spu_remove_from_active_list(spu);
 421                         spu_unbind_context(spu, ctx);
 422                         spu_free(spu);
 423                         if (new)
 424                                 wake_up(&new->stop_wq);
 425                 }
 426
 427         }
 428
 429         return new != NULL;
 430 }
 431
 432 /**
 433  * spu_deactivate - unbind a context from it's physical spu
 434  * @ctx:        spu context to unbind
 435  *
 436  * Unbind @ctx from the physical spu it is running on and schedule
 437  * the highest priority context to run on the freed physical spu.
 438  */
 439 void spu_deactivate(struct spu_context *ctx)
 440 {
 441         __spu_deactivate(ctx, 1, MAX_PRIO);
 442 }
 443
 444 /**
 445  * spu_yield -  yield a physical spu if others are waiting
 446  * @ctx:        spu context to yield
 447  *
 448  * Check if there is a higher priority context waiting and if yes
 449  * unbind @ctx from the physical spu and schedule the highest
 450  * priority context to run on the freed physical spu instead.
 451  */
 452 void spu_yield(struct spu_context *ctx)
 453 {
 454         if (!(ctx->flags & SPU_CREATE_NOSCHED)) {
 455                 mutex_lock(&ctx->state_mutex);
 456                 __spu_deactivate(ctx, 0, MAX_PRIO);
 457                 mutex_unlock(&ctx->state_mutex);
 458         }
 459 }
 460
 461 static void spusched_tick(struct spu_context *ctx)
 462 {
 463         if (ctx->policy == SCHED_FIFO || --ctx->time_slice)
 464                 return;
 465
 466         /*
 467          * Unfortunately active_mutex ranks outside of state_mutex, so
 468          * we have to trylock here.  If we fail give the context another
 469          * tick and try again.
 470          */
 471         if (mutex_trylock(&ctx->state_mutex)) {
 472                 struct spu_context *new = grab_runnable_context(ctx->prio + 1);
 473                 if (new) {
 474                         struct spu *spu = ctx->spu;
 475
 476                         __spu_remove_from_active_list(spu);
 477                         spu_unbind_context(spu, ctx);
 478                         spu_free(spu);
 479                         wake_up(&new->stop_wq);
 480                         /*
 481                          * We need to break out of the wait loop in
 482                          * spu_run manually to ensure this context
 483                          * gets put on the runqueue again ASAP.
 484                          */
 485                         wake_up(&ctx->stop_wq);
 486                 }
 487                 spu_set_timeslice(ctx);
 488                 mutex_unlock(&ctx->state_mutex);
 489         } else {
 490                 ctx->time_slice++;
 491         }
 492 }
 493
 494 static void spusched_wake(unsigned long data)
 495 {
 496         mod_timer(&spusched_timer, jiffies + SPUSCHED_TICK);
 497         wake_up_process(spusched_task);
 498 }
 499
 500 static int spusched_thread(void *unused)
 501 {
 502         struct spu *spu, *next;
 503         int node;
 504
 505         setup_timer(&spusched_timer, spusched_wake, 0);
 506         __mod_timer(&spusched_timer, jiffies + SPUSCHED_TICK);
 507
 508         while (!kthread_should_stop()) {
 509                 set_current_state(TASK_INTERRUPTIBLE);
 510                 schedule();
 511                 for (node = 0; node < MAX_NUMNODES; node++) {
 512                         mutex_lock(&spu_prio->active_mutex[node]);
 513                         list_for_each_entry_safe(spu, next,
 514                                                  &spu_prio->active_list[node],
 515                                                  list)
 516                                 spusched_tick(spu->ctx);
 517                         mutex_unlock(&spu_prio->active_mutex[node]);
 518                 }
 519         }
 520
 521         del_timer_sync(&spusched_timer);
 522         return 0;
 523 }
 524
 525 int __init spu_sched_init(void)
 526 {
 527         int i;
 528
 529         spu_prio = kzalloc(sizeof(struct spu_prio_array), GFP_KERNEL);
 530         if (!spu_prio)
 531                 return -ENOMEM;
 532
 533         for (i = 0; i < MAX_PRIO; i++) {
 534                 INIT_LIST_HEAD(&spu_prio->runq[i]);
 535                 __clear_bit(i, spu_prio->bitmap);
 536         }
 537         __set_bit(MAX_PRIO, spu_prio->bitmap);
 538         for (i = 0; i < MAX_NUMNODES; i++) {
 539                 mutex_init(&spu_prio->active_mutex[i]);
 540                 INIT_LIST_HEAD(&spu_prio->active_list[i]);
 541         }
 542         spin_lock_init(&spu_prio->runq_lock);
 543
 544         spusched_task = kthread_run(spusched_thread, NULL, "spusched");
 545         if (IS_ERR(spusched_task)) {
 546                 kfree(spu_prio);
 547                 return PTR_ERR(spusched_task);
 548         }
 549
 550         pr_debug("spusched: tick: %d, min ticks: %d, default ticks: %d\n",
 551                         SPUSCHED_TICK, MIN_SPU_TIMESLICE, DEF_SPU_TIMESLICE);
 552         return 0;
 553
 554 }
 555
 556 void __exit spu_sched_exit(void)
 557 {
 558         struct spu *spu, *tmp;
 559         int node;
 560
 561         kthread_stop(spusched_task);
 562
 563         for (node = 0; node < MAX_NUMNODES; node++) {
 564                 mutex_lock(&spu_prio->active_mutex[node]);
 565                 list_for_each_entry_safe(spu, tmp, &spu_prio->active_list[node],
 566                                          list) {
 567                         list_del_init(&spu->list);
 568                         spu_free(spu);
 569                 }
 570                 mutex_unlock(&spu_prio->active_mutex[node]);
 571         }
 572         kfree(spu_prio);
 573 }