1 /*-
2 * Copyright (c) 2000 Doug Rabson
3 * Copyright (c) 2014 Jeff Roberson
4 * Copyright (c) 2016 Matthew Macy
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/bus.h>
35 #include <sys/cpuset.h>
36 #include <sys/interrupt.h>
37 #include <sys/kernel.h>
38 #include <sys/kthread.h>
39 #include <sys/libkern.h>
40 #include <sys/limits.h>
41 #include <sys/lock.h>
42 #include <sys/malloc.h>
43 #include <sys/mutex.h>
44 #include <sys/proc.h>
45 #include <sys/sched.h>
46 #include <sys/smp.h>
47 #include <sys/gtaskqueue.h>
48 #include <sys/unistd.h>
49 #include <machine/stdarg.h>
50
51 static MALLOC_DEFINE(M_GTASKQUEUE, "gtaskqueue", "Group Task Queues");
52 static void gtaskqueue_thread_enqueue(void *);
53 static void gtaskqueue_thread_loop(void *arg);
54
55 TASKQGROUP_DEFINE(softirq, mp_ncpus, 1);
56
57 struct gtaskqueue_busy {
58 struct gtask *tb_running;
59 u_int tb_seq;
60 LIST_ENTRY(gtaskqueue_busy) tb_link;
61 };
62
63 struct gtaskqueue {
64 STAILQ_HEAD(, gtask) tq_queue;
65 LIST_HEAD(, gtaskqueue_busy) tq_active;
66 u_int tq_seq;
67 int tq_callouts;
68 struct mtx_padalign tq_mutex;
69 gtaskqueue_enqueue_fn tq_enqueue;
70 void *tq_context;
71 char *tq_name;
72 struct thread **tq_threads;
73 int tq_tcount;
74 int tq_spin;
75 int tq_flags;
76 taskqueue_callback_fn tq_callbacks[TASKQUEUE_NUM_CALLBACKS];
77 void *tq_cb_contexts[TASKQUEUE_NUM_CALLBACKS];
78 };
79
80 #define TQ_FLAGS_ACTIVE (1 << 0)
81 #define TQ_FLAGS_BLOCKED (1 << 1)
82 #define TQ_FLAGS_UNLOCKED_ENQUEUE (1 << 2)
83
84 #define DT_CALLOUT_ARMED (1 << 0)
85
86 #define TQ_LOCK(tq) \
87 do { \
88 if ((tq)->tq_spin) \
89 mtx_lock_spin(&(tq)->tq_mutex); \
90 else \
91 mtx_lock(&(tq)->tq_mutex); \
92 } while (0)
93 #define TQ_ASSERT_LOCKED(tq) mtx_assert(&(tq)->tq_mutex, MA_OWNED)
94
95 #define TQ_UNLOCK(tq) \
96 do { \
97 if ((tq)->tq_spin) \
98 mtx_unlock_spin(&(tq)->tq_mutex); \
99 else \
100 mtx_unlock(&(tq)->tq_mutex); \
101 } while (0)
102 #define TQ_ASSERT_UNLOCKED(tq) mtx_assert(&(tq)->tq_mutex, MA_NOTOWNED)
103
104 #ifdef INVARIANTS
105 static void
106 gtask_dump(struct gtask *gtask)
107 {
108 printf("gtask: %p ta_flags=%x ta_priority=%d ta_func=%p ta_context=%p\n",
109 gtask, gtask->ta_flags, gtask->ta_priority, gtask->ta_func, gtask->ta_context);
110 }
111 #endif
112
113 static __inline int
114 TQ_SLEEP(struct gtaskqueue *tq, void *p, const char *wm)
115 {
116 if (tq->tq_spin)
117 return (msleep_spin(p, (struct mtx *)&tq->tq_mutex, wm, 0));
118 return (msleep(p, &tq->tq_mutex, 0, wm, 0));
119 }
120
121 static struct gtaskqueue *
122 _gtaskqueue_create(const char *name, int mflags,
123 taskqueue_enqueue_fn enqueue, void *context,
124 int mtxflags, const char *mtxname __unused)
125 {
126 struct gtaskqueue *queue;
127 char *tq_name;
128
129 tq_name = malloc(TASKQUEUE_NAMELEN, M_GTASKQUEUE, mflags | M_ZERO);
130 if (!tq_name)
131 return (NULL);
132
133 snprintf(tq_name, TASKQUEUE_NAMELEN, "%s", (name) ? name : "taskqueue");
134
135 queue = malloc(sizeof(struct gtaskqueue), M_GTASKQUEUE, mflags | M_ZERO);
136 if (!queue) {
137 free(tq_name, M_GTASKQUEUE);
138 return (NULL);
139 }
140
141 STAILQ_INIT(&queue->tq_queue);
142 LIST_INIT(&queue->tq_active);
143 queue->tq_enqueue = enqueue;
144 queue->tq_context = context;
145 queue->tq_name = tq_name;
146 queue->tq_spin = (mtxflags & MTX_SPIN) != 0;
147 queue->tq_flags |= TQ_FLAGS_ACTIVE;
148 if (enqueue == gtaskqueue_thread_enqueue)
149 queue->tq_flags |= TQ_FLAGS_UNLOCKED_ENQUEUE;
150 mtx_init(&queue->tq_mutex, tq_name, NULL, mtxflags);
151
152 return (queue);
153 }
154
155
156 /*
157 * Signal a taskqueue thread to terminate.
158 */
159 static void
160 gtaskqueue_terminate(struct thread **pp, struct gtaskqueue *tq)
161 {
162
163 while (tq->tq_tcount > 0 || tq->tq_callouts > 0) {
164 wakeup(tq);
165 TQ_SLEEP(tq, pp, "gtq_destroy");
166 }
167 }
168
169 static void
170 gtaskqueue_free(struct gtaskqueue *queue)
171 {
172
173 TQ_LOCK(queue);
174 queue->tq_flags &= ~TQ_FLAGS_ACTIVE;
175 gtaskqueue_terminate(queue->tq_threads, queue);
176 KASSERT(LIST_EMPTY(&queue->tq_active), ("Tasks still running?"));
177 KASSERT(queue->tq_callouts == 0, ("Armed timeout tasks"));
178 mtx_destroy(&queue->tq_mutex);
179 free(queue->tq_threads, M_GTASKQUEUE);
180 free(queue->tq_name, M_GTASKQUEUE);
181 free(queue, M_GTASKQUEUE);
182 }
183
184 int
185 grouptaskqueue_enqueue(struct gtaskqueue *queue, struct gtask *gtask)
186 {
187 #ifdef INVARIANTS
188 if (queue == NULL) {
189 gtask_dump(gtask);
190 panic("queue == NULL");
191 }
192 #endif
193 TQ_LOCK(queue);
194 if (gtask->ta_flags & TASK_ENQUEUED) {
195 TQ_UNLOCK(queue);
196 return (0);
197 }
198 STAILQ_INSERT_TAIL(&queue->tq_queue, gtask, ta_link);
199 gtask->ta_flags |= TASK_ENQUEUED;
200 TQ_UNLOCK(queue);
201 if ((queue->tq_flags & TQ_FLAGS_BLOCKED) == 0)
202 queue->tq_enqueue(queue->tq_context);
203 return (0);
204 }
205
206 static void
207 gtaskqueue_task_nop_fn(void *context)
208 {
209 }
210
211 /*
212 * Block until all currently queued tasks in this taskqueue
213 * have begun execution. Tasks queued during execution of
214 * this function are ignored.
215 */
216 static void
217 gtaskqueue_drain_tq_queue(struct gtaskqueue *queue)
218 {
219 struct gtask t_barrier;
220
221 if (STAILQ_EMPTY(&queue->tq_queue))
222 return;
223
224 /*
225 * Enqueue our barrier after all current tasks, but with
226 * the highest priority so that newly queued tasks cannot
227 * pass it. Because of the high priority, we can not use
228 * taskqueue_enqueue_locked directly (which drops the lock
229 * anyway) so just insert it at tail while we have the
230 * queue lock.
231 */
232 GTASK_INIT(&t_barrier, 0, USHRT_MAX, gtaskqueue_task_nop_fn, &t_barrier);
233 STAILQ_INSERT_TAIL(&queue->tq_queue, &t_barrier, ta_link);
234 t_barrier.ta_flags |= TASK_ENQUEUED;
235
236 /*
237 * Once the barrier has executed, all previously queued tasks
238 * have completed or are currently executing.
239 */
240 while (t_barrier.ta_flags & TASK_ENQUEUED)
241 TQ_SLEEP(queue, &t_barrier, "gtq_qdrain");
242 }
243
244 /*
245 * Block until all currently executing tasks for this taskqueue
246 * complete. Tasks that begin execution during the execution
247 * of this function are ignored.
248 */
249 static void
250 gtaskqueue_drain_tq_active(struct gtaskqueue *queue)
251 {
252 struct gtaskqueue_busy *tb;
253 u_int seq;
254
255 if (LIST_EMPTY(&queue->tq_active))
256 return;
257
258 /* Block taskq_terminate().*/
259 queue->tq_callouts++;
260
261 /* Wait for any active task with sequence from the past. */
262 seq = queue->tq_seq;
263 restart:
264 LIST_FOREACH(tb, &queue->tq_active, tb_link) {
265 if ((int)(tb->tb_seq - seq) <= 0) {
266 TQ_SLEEP(queue, tb->tb_running, "gtq_adrain");
267 goto restart;
268 }
269 }
270
271 /* Release taskqueue_terminate(). */
272 queue->tq_callouts--;
273 if ((queue->tq_flags & TQ_FLAGS_ACTIVE) == 0)
274 wakeup_one(queue->tq_threads);
275 }
276
277 void
278 gtaskqueue_block(struct gtaskqueue *queue)
279 {
280
281 TQ_LOCK(queue);
282 queue->tq_flags |= TQ_FLAGS_BLOCKED;
283 TQ_UNLOCK(queue);
284 }
285
286 void
287 gtaskqueue_unblock(struct gtaskqueue *queue)
288 {
289
290 TQ_LOCK(queue);
291 queue->tq_flags &= ~TQ_FLAGS_BLOCKED;
292 if (!STAILQ_EMPTY(&queue->tq_queue))
293 queue->tq_enqueue(queue->tq_context);
294 TQ_UNLOCK(queue);
295 }
296
297 static void
298 gtaskqueue_run_locked(struct gtaskqueue *queue)
299 {
300 struct gtaskqueue_busy tb;
301 struct gtask *gtask;
302
303 KASSERT(queue != NULL, ("tq is NULL"));
304 TQ_ASSERT_LOCKED(queue);
305 tb.tb_running = NULL;
306 LIST_INSERT_HEAD(&queue->tq_active, &tb, tb_link);
307
308 while ((gtask = STAILQ_FIRST(&queue->tq_queue)) != NULL) {
309 STAILQ_REMOVE_HEAD(&queue->tq_queue, ta_link);
310 gtask->ta_flags &= ~TASK_ENQUEUED;
311 tb.tb_running = gtask;
312 tb.tb_seq = ++queue->tq_seq;
313 TQ_UNLOCK(queue);
314
315 KASSERT(gtask->ta_func != NULL, ("task->ta_func is NULL"));
316 gtask->ta_func(gtask->ta_context);
317
318 TQ_LOCK(queue);
319 wakeup(gtask);
320 }
321 LIST_REMOVE(&tb, tb_link);
322 }
323
324 static int
325 task_is_running(struct gtaskqueue *queue, struct gtask *gtask)
326 {
327 struct gtaskqueue_busy *tb;
328
329 TQ_ASSERT_LOCKED(queue);
330 LIST_FOREACH(tb, &queue->tq_active, tb_link) {
331 if (tb->tb_running == gtask)
332 return (1);
333 }
334 return (0);
335 }
336
337 static int
338 gtaskqueue_cancel_locked(struct gtaskqueue *queue, struct gtask *gtask)
339 {
340
341 if (gtask->ta_flags & TASK_ENQUEUED)
342 STAILQ_REMOVE(&queue->tq_queue, gtask, gtask, ta_link);
343 gtask->ta_flags &= ~TASK_ENQUEUED;
344 return (task_is_running(queue, gtask) ? EBUSY : 0);
345 }
346
347 int
348 gtaskqueue_cancel(struct gtaskqueue *queue, struct gtask *gtask)
349 {
350 int error;
351
352 TQ_LOCK(queue);
353 error = gtaskqueue_cancel_locked(queue, gtask);
354 TQ_UNLOCK(queue);
355
356 return (error);
357 }
358
359 void
360 gtaskqueue_drain(struct gtaskqueue *queue, struct gtask *gtask)
361 {
362
363 if (!queue->tq_spin)
364 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, __func__);
365
366 TQ_LOCK(queue);
367 while ((gtask->ta_flags & TASK_ENQUEUED) || task_is_running(queue, gtask))
368 TQ_SLEEP(queue, gtask, "gtq_drain");
369 TQ_UNLOCK(queue);
370 }
371
372 void
373 gtaskqueue_drain_all(struct gtaskqueue *queue)
374 {
375
376 if (!queue->tq_spin)
377 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, __func__);
378
379 TQ_LOCK(queue);
380 gtaskqueue_drain_tq_queue(queue);
381 gtaskqueue_drain_tq_active(queue);
382 TQ_UNLOCK(queue);
383 }
384
385 static int
386 _gtaskqueue_start_threads(struct gtaskqueue **tqp, int count, int pri,
387 cpuset_t *mask, const char *name, va_list ap)
388 {
389 char ktname[MAXCOMLEN + 1];
390 struct thread *td;
391 struct gtaskqueue *tq;
392 int i, error;
393
394 if (count <= 0)
395 return (EINVAL);
396
397 vsnprintf(ktname, sizeof(ktname), name, ap);
398 tq = *tqp;
399
400 tq->tq_threads = malloc(sizeof(struct thread *) * count, M_GTASKQUEUE,
401 M_NOWAIT | M_ZERO);
402 if (tq->tq_threads == NULL) {
403 printf("%s: no memory for %s threads\n", __func__, ktname);
404 return (ENOMEM);
405 }
406
407 for (i = 0; i < count; i++) {
408 if (count == 1)
409 error = kthread_add(gtaskqueue_thread_loop, tqp, NULL,
410 &tq->tq_threads[i], RFSTOPPED, 0, "%s", ktname);
411 else
412 error = kthread_add(gtaskqueue_thread_loop, tqp, NULL,
413 &tq->tq_threads[i], RFSTOPPED, 0,
414 "%s_%d", ktname, i);
415 if (error) {
416 /* should be ok to continue, taskqueue_free will dtrt */
417 printf("%s: kthread_add(%s): error %d", __func__,
418 ktname, error);
419 tq->tq_threads[i] = NULL; /* paranoid */
420 } else
421 tq->tq_tcount++;
422 }
423 for (i = 0; i < count; i++) {
424 if (tq->tq_threads[i] == NULL)
425 continue;
426 td = tq->tq_threads[i];
427 if (mask) {
428 error = cpuset_setthread(td->td_tid, mask);
429 /*
430 * Failing to pin is rarely an actual fatal error;
431 * it'll just affect performance.
432 */
433 if (error)
434 printf("%s: curthread=%llu: can't pin; "
435 "error=%d\n",
436 __func__,
437 (unsigned long long) td->td_tid,
438 error);
439 }
440 thread_lock(td);
441 sched_prio(td, pri);
442 sched_add(td, SRQ_BORING);
443 thread_unlock(td);
444 }
445
446 return (0);
447 }
448
449 static int
450 gtaskqueue_start_threads(struct gtaskqueue **tqp, int count, int pri,
451 const char *name, ...)
452 {
453 va_list ap;
454 int error;
455
456 va_start(ap, name);
457 error = _gtaskqueue_start_threads(tqp, count, pri, NULL, name, ap);
458 va_end(ap);
459 return (error);
460 }
461
462 static inline void
463 gtaskqueue_run_callback(struct gtaskqueue *tq,
464 enum taskqueue_callback_type cb_type)
465 {
466 taskqueue_callback_fn tq_callback;
467
468 TQ_ASSERT_UNLOCKED(tq);
469 tq_callback = tq->tq_callbacks[cb_type];
470 if (tq_callback != NULL)
471 tq_callback(tq->tq_cb_contexts[cb_type]);
472 }
473
474 static void
475 gtaskqueue_thread_loop(void *arg)
476 {
477 struct gtaskqueue **tqp, *tq;
478
479 tqp = arg;
480 tq = *tqp;
481 gtaskqueue_run_callback(tq, TASKQUEUE_CALLBACK_TYPE_INIT);
482 TQ_LOCK(tq);
483 while ((tq->tq_flags & TQ_FLAGS_ACTIVE) != 0) {
484 /* XXX ? */
485 gtaskqueue_run_locked(tq);
486 /*
487 * Because taskqueue_run() can drop tq_mutex, we need to
488 * check if the TQ_FLAGS_ACTIVE flag wasn't removed in the
489 * meantime, which means we missed a wakeup.
490 */
491 if ((tq->tq_flags & TQ_FLAGS_ACTIVE) == 0)
492 break;
493 TQ_SLEEP(tq, tq, "-");
494 }
495 gtaskqueue_run_locked(tq);
496 /*
497 * This thread is on its way out, so just drop the lock temporarily
498 * in order to call the shutdown callback. This allows the callback
499 * to look at the taskqueue, even just before it dies.
500 */
501 TQ_UNLOCK(tq);
502 gtaskqueue_run_callback(tq, TASKQUEUE_CALLBACK_TYPE_SHUTDOWN);
503 TQ_LOCK(tq);
504
505 /* rendezvous with thread that asked us to terminate */
506 tq->tq_tcount--;
507 wakeup_one(tq->tq_threads);
508 TQ_UNLOCK(tq);
509 kthread_exit();
510 }
511
512 static void
513 gtaskqueue_thread_enqueue(void *context)
514 {
515 struct gtaskqueue **tqp, *tq;
516
517 tqp = context;
518 tq = *tqp;
519 wakeup_any(tq);
520 }
521
522
523 static struct gtaskqueue *
524 gtaskqueue_create_fast(const char *name, int mflags,
525 taskqueue_enqueue_fn enqueue, void *context)
526 {
527 return _gtaskqueue_create(name, mflags, enqueue, context,
528 MTX_SPIN, "fast_taskqueue");
529 }
530
531
532 struct taskqgroup_cpu {
533 LIST_HEAD(, grouptask) tgc_tasks;
534 struct gtaskqueue *tgc_taskq;
535 int tgc_cnt;
536 int tgc_cpu;
537 };
538
539 struct taskqgroup {
540 struct taskqgroup_cpu tqg_queue[MAXCPU];
541 struct mtx tqg_lock;
542 char * tqg_name;
543 int tqg_adjusting;
544 int tqg_stride;
545 int tqg_cnt;
546 };
547
548 struct taskq_bind_task {
549 struct gtask bt_task;
550 int bt_cpuid;
551 };
552
553 static void
554 taskqgroup_cpu_create(struct taskqgroup *qgroup, int idx, int cpu)
555 {
556 struct taskqgroup_cpu *qcpu;
557
558 qcpu = &qgroup->tqg_queue[idx];
559 LIST_INIT(&qcpu->tgc_tasks);
560 qcpu->tgc_taskq = gtaskqueue_create_fast(NULL, M_WAITOK,
561 taskqueue_thread_enqueue, &qcpu->tgc_taskq);
562 gtaskqueue_start_threads(&qcpu->tgc_taskq, 1, PI_SOFT,
563 "%s_%d", qgroup->tqg_name, idx);
564 qcpu->tgc_cpu = cpu;
565 }
566
567 static void
568 taskqgroup_cpu_remove(struct taskqgroup *qgroup, int idx)
569 {
570
571 gtaskqueue_free(qgroup->tqg_queue[idx].tgc_taskq);
572 }
573
574 /*
575 * Find the taskq with least # of tasks that doesn't currently have any
576 * other queues from the uniq identifier.
577 */
578 static int
579 taskqgroup_find(struct taskqgroup *qgroup, void *uniq)
580 {
581 struct grouptask *n;
582 int i, idx, mincnt;
583 int strict;
584
585 mtx_assert(&qgroup->tqg_lock, MA_OWNED);
586 if (qgroup->tqg_cnt == 0)
587 return (0);
588 idx = -1;
589 mincnt = INT_MAX;
590 /*
591 * Two passes; First scan for a queue with the least tasks that
592 * does not already service this uniq id. If that fails simply find
593 * the queue with the least total tasks;
594 */
595 for (strict = 1; mincnt == INT_MAX; strict = 0) {
596 for (i = 0; i < qgroup->tqg_cnt; i++) {
597 if (qgroup->tqg_queue[i].tgc_cnt > mincnt)
598 continue;
599 if (strict) {
600 LIST_FOREACH(n,
601 &qgroup->tqg_queue[i].tgc_tasks, gt_list)
602 if (n->gt_uniq == uniq)
603 break;
604 if (n != NULL)
605 continue;
606 }
607 mincnt = qgroup->tqg_queue[i].tgc_cnt;
608 idx = i;
609 }
610 }
611 if (idx == -1)
612 panic("taskqgroup_find: Failed to pick a qid.");
613
614 return (idx);
615 }
616
617 /*
618 * smp_started is unusable since it is not set for UP kernels or even for
619 * SMP kernels when there is 1 CPU. This is usually handled by adding a
620 * (mp_ncpus == 1) test, but that would be broken here since we need to
621 * to synchronize with the SI_SUB_SMP ordering. Even in the pure SMP case
622 * smp_started only gives a fuzzy ordering relative to SI_SUB_SMP.
623 *
624 * So maintain our own flag. It must be set after all CPUs are started
625 * and before SI_SUB_SMP:SI_ORDER_ANY so that the SYSINIT for delayed
626 * adjustment is properly delayed. SI_ORDER_FOURTH is clearly before
627 * SI_ORDER_ANY and unclearly after the CPUs are started. It would be
628 * simpler for adjustment to pass a flag indicating if it is delayed.
629 */
630
631 static int tqg_smp_started;
632
633 static void
634 tqg_record_smp_started(void *arg)
635 {
636 tqg_smp_started = 1;
637 }
638
639 SYSINIT(tqg_record_smp_started, SI_SUB_SMP, SI_ORDER_FOURTH,
640 tqg_record_smp_started, NULL);
641
642 void
643 taskqgroup_attach(struct taskqgroup *qgroup, struct grouptask *gtask,
644 void *uniq, int irq, char *name)
645 {
646 cpuset_t mask;
647 int qid, error;
648
649 gtask->gt_uniq = uniq;
650 snprintf(gtask->gt_name, GROUPTASK_NAMELEN, "%s", name ? name : "grouptask");
651 gtask->gt_irq = irq;
652 gtask->gt_cpu = -1;
653 mtx_lock(&qgroup->tqg_lock);
654 qid = taskqgroup_find(qgroup, uniq);
655 qgroup->tqg_queue[qid].tgc_cnt++;
656 LIST_INSERT_HEAD(&qgroup->tqg_queue[qid].tgc_tasks, gtask, gt_list);
657 gtask->gt_taskqueue = qgroup->tqg_queue[qid].tgc_taskq;
658 if (irq != -1 && tqg_smp_started) {
659 gtask->gt_cpu = qgroup->tqg_queue[qid].tgc_cpu;
660 CPU_ZERO(&mask);
661 CPU_SET(qgroup->tqg_queue[qid].tgc_cpu, &mask);
662 mtx_unlock(&qgroup->tqg_lock);
663 error = intr_setaffinity(irq, CPU_WHICH_IRQ, &mask);
664 if (error)
665 printf("%s: setaffinity failed for %s: %d\n", __func__, gtask->gt_name, error);
666 } else
667 mtx_unlock(&qgroup->tqg_lock);
668 }
669
670 static void
671 taskqgroup_attach_deferred(struct taskqgroup *qgroup, struct grouptask *gtask)
672 {
673 cpuset_t mask;
674 int qid, cpu, error;
675
676 mtx_lock(&qgroup->tqg_lock);
677 qid = taskqgroup_find(qgroup, gtask->gt_uniq);
678 cpu = qgroup->tqg_queue[qid].tgc_cpu;
679 if (gtask->gt_irq != -1) {
680 mtx_unlock(&qgroup->tqg_lock);
681
682 CPU_ZERO(&mask);
683 CPU_SET(cpu, &mask);
684 error = intr_setaffinity(gtask->gt_irq, CPU_WHICH_IRQ, &mask);
685 mtx_lock(&qgroup->tqg_lock);
686 if (error)
687 printf("%s: %s setaffinity failed: %d\n", __func__, gtask->gt_name, error);
688
689 }
690 qgroup->tqg_queue[qid].tgc_cnt++;
691
692 LIST_INSERT_HEAD(&qgroup->tqg_queue[qid].tgc_tasks, gtask,
693 gt_list);
694 MPASS(qgroup->tqg_queue[qid].tgc_taskq != NULL);
695 gtask->gt_taskqueue = qgroup->tqg_queue[qid].tgc_taskq;
696 mtx_unlock(&qgroup->tqg_lock);
697 }
698
699 int
700 taskqgroup_attach_cpu(struct taskqgroup *qgroup, struct grouptask *gtask,
701 void *uniq, int cpu, int irq, char *name)
702 {
703 cpuset_t mask;
704 int i, qid, error;
705
706 qid = -1;
707 gtask->gt_uniq = uniq;
708 snprintf(gtask->gt_name, GROUPTASK_NAMELEN, "%s", name ? name : "grouptask");
709 gtask->gt_irq = irq;
710 gtask->gt_cpu = cpu;
711 mtx_lock(&qgroup->tqg_lock);
712 if (tqg_smp_started) {
713 for (i = 0; i < qgroup->tqg_cnt; i++)
714 if (qgroup->tqg_queue[i].tgc_cpu == cpu) {
715 qid = i;
716 break;
717 }
718 if (qid == -1) {
719 mtx_unlock(&qgroup->tqg_lock);
720 printf("%s: qid not found for %s cpu=%d\n", __func__, gtask->gt_name, cpu);
721 return (EINVAL);
722 }
723 } else
724 qid = 0;
725 qgroup->tqg_queue[qid].tgc_cnt++;
726 LIST_INSERT_HEAD(&qgroup->tqg_queue[qid].tgc_tasks, gtask, gt_list);
727 gtask->gt_taskqueue = qgroup->tqg_queue[qid].tgc_taskq;
728 cpu = qgroup->tqg_queue[qid].tgc_cpu;
729 mtx_unlock(&qgroup->tqg_lock);
730
731 CPU_ZERO(&mask);
732 CPU_SET(cpu, &mask);
733 if (irq != -1 && tqg_smp_started) {
734 error = intr_setaffinity(irq, CPU_WHICH_IRQ, &mask);
735 if (error)
736 printf("%s: setaffinity failed: %d\n", __func__, error);
737 }
738 return (0);
739 }
740
741 static int
742 taskqgroup_attach_cpu_deferred(struct taskqgroup *qgroup, struct grouptask *gtask)
743 {
744 cpuset_t mask;
745 int i, qid, irq, cpu, error;
746
747 qid = -1;
748 irq = gtask->gt_irq;
749 cpu = gtask->gt_cpu;
750 MPASS(tqg_smp_started);
751 mtx_lock(&qgroup->tqg_lock);
752 for (i = 0; i < qgroup->tqg_cnt; i++)
753 if (qgroup->tqg_queue[i].tgc_cpu == cpu) {
754 qid = i;
755 break;
756 }
757 if (qid == -1) {
758 mtx_unlock(&qgroup->tqg_lock);
759 printf("%s: qid not found for %s cpu=%d\n", __func__, gtask->gt_name, cpu);
760 return (EINVAL);
761 }
762 qgroup->tqg_queue[qid].tgc_cnt++;
763 LIST_INSERT_HEAD(&qgroup->tqg_queue[qid].tgc_tasks, gtask, gt_list);
764 MPASS(qgroup->tqg_queue[qid].tgc_taskq != NULL);
765 gtask->gt_taskqueue = qgroup->tqg_queue[qid].tgc_taskq;
766 mtx_unlock(&qgroup->tqg_lock);
767
768 CPU_ZERO(&mask);
769 CPU_SET(cpu, &mask);
770
771 if (irq != -1) {
772 error = intr_setaffinity(irq, CPU_WHICH_IRQ, &mask);
773 if (error)
774 printf("%s: setaffinity failed: %d\n", __func__, error);
775 }
776 return (0);
777 }
778
779 void
780 taskqgroup_detach(struct taskqgroup *qgroup, struct grouptask *gtask)
781 {
782 int i;
783
784 mtx_lock(&qgroup->tqg_lock);
785 for (i = 0; i < qgroup->tqg_cnt; i++)
786 if (qgroup->tqg_queue[i].tgc_taskq == gtask->gt_taskqueue)
787 break;
788 if (i == qgroup->tqg_cnt)
789 panic("taskqgroup_detach: task %s not in group\n", gtask->gt_name);
790 qgroup->tqg_queue[i].tgc_cnt--;
791 LIST_REMOVE(gtask, gt_list);
792 mtx_unlock(&qgroup->tqg_lock);
793 gtask->gt_taskqueue = NULL;
794 }
795
796 static void
797 taskqgroup_binder(void *ctx)
798 {
799 struct taskq_bind_task *gtask = (struct taskq_bind_task *)ctx;
800 cpuset_t mask;
801 int error;
802
803 CPU_ZERO(&mask);
804 CPU_SET(gtask->bt_cpuid, &mask);
805 error = cpuset_setthread(curthread->td_tid, &mask);
806 thread_lock(curthread);
807 sched_bind(curthread, gtask->bt_cpuid);
808 thread_unlock(curthread);
809
810 if (error)
811 printf("%s: setaffinity failed: %d\n", __func__,
812 error);
813 free(gtask, M_DEVBUF);
814 }
815
816 static void
817 taskqgroup_bind(struct taskqgroup *qgroup)
818 {
819 struct taskq_bind_task *gtask;
820 int i;
821
822 /*
823 * Bind taskqueue threads to specific CPUs, if they have been assigned
824 * one.
825 */
826 if (qgroup->tqg_cnt == 1)
827 return;
828
829 for (i = 0; i < qgroup->tqg_cnt; i++) {
830 gtask = malloc(sizeof (*gtask), M_DEVBUF, M_WAITOK);
831 GTASK_INIT(>ask->bt_task, 0, 0, taskqgroup_binder, gtask);
832 gtask->bt_cpuid = qgroup->tqg_queue[i].tgc_cpu;
833 grouptaskqueue_enqueue(qgroup->tqg_queue[i].tgc_taskq,
834 >ask->bt_task);
835 }
836 }
837
838 static int
839 _taskqgroup_adjust(struct taskqgroup *qgroup, int cnt, int stride)
840 {
841 LIST_HEAD(, grouptask) gtask_head = LIST_HEAD_INITIALIZER(NULL);
842 struct grouptask *gtask;
843 int i, k, old_cnt, old_cpu, cpu;
844
845 mtx_assert(&qgroup->tqg_lock, MA_OWNED);
846
847 if (cnt < 1 || cnt * stride > mp_ncpus || !tqg_smp_started) {
848 printf("%s: failed cnt: %d stride: %d "
849 "mp_ncpus: %d tqg_smp_started: %d\n",
850 __func__, cnt, stride, mp_ncpus, tqg_smp_started);
851 return (EINVAL);
852 }
853 if (qgroup->tqg_adjusting) {
854 printf("%s failed: adjusting\n", __func__);
855 return (EBUSY);
856 }
857 qgroup->tqg_adjusting = 1;
858 old_cnt = qgroup->tqg_cnt;
859 old_cpu = 0;
860 if (old_cnt < cnt)
861 old_cpu = qgroup->tqg_queue[old_cnt].tgc_cpu;
862 mtx_unlock(&qgroup->tqg_lock);
863 /*
864 * Set up queue for tasks added before boot.
865 */
866 if (old_cnt == 0) {
867 LIST_SWAP(>ask_head, &qgroup->tqg_queue[0].tgc_tasks,
868 grouptask, gt_list);
869 qgroup->tqg_queue[0].tgc_cnt = 0;
870 }
871
872 /*
873 * If new taskq threads have been added.
874 */
875 cpu = old_cpu;
876 for (i = old_cnt; i < cnt; i++) {
877 taskqgroup_cpu_create(qgroup, i, cpu);
878
879 for (k = 0; k < stride; k++)
880 cpu = CPU_NEXT(cpu);
881 }
882 mtx_lock(&qgroup->tqg_lock);
883 qgroup->tqg_cnt = cnt;
884 qgroup->tqg_stride = stride;
885
886 /*
887 * Adjust drivers to use new taskqs.
888 */
889 for (i = 0; i < old_cnt; i++) {
890 while ((gtask = LIST_FIRST(&qgroup->tqg_queue[i].tgc_tasks))) {
891 LIST_REMOVE(gtask, gt_list);
892 qgroup->tqg_queue[i].tgc_cnt--;
893 LIST_INSERT_HEAD(>ask_head, gtask, gt_list);
894 }
895 }
896 mtx_unlock(&qgroup->tqg_lock);
897
898 while ((gtask = LIST_FIRST(>ask_head))) {
899 LIST_REMOVE(gtask, gt_list);
900 if (gtask->gt_cpu == -1)
901 taskqgroup_attach_deferred(qgroup, gtask);
902 else if (taskqgroup_attach_cpu_deferred(qgroup, gtask))
903 taskqgroup_attach_deferred(qgroup, gtask);
904 }
905
906 #ifdef INVARIANTS
907 mtx_lock(&qgroup->tqg_lock);
908 for (i = 0; i < qgroup->tqg_cnt; i++) {
909 MPASS(qgroup->tqg_queue[i].tgc_taskq != NULL);
910 LIST_FOREACH(gtask, &qgroup->tqg_queue[i].tgc_tasks, gt_list)
911 MPASS(gtask->gt_taskqueue != NULL);
912 }
913 mtx_unlock(&qgroup->tqg_lock);
914 #endif
915 /*
916 * If taskq thread count has been reduced.
917 */
918 for (i = cnt; i < old_cnt; i++)
919 taskqgroup_cpu_remove(qgroup, i);
920
921 taskqgroup_bind(qgroup);
922
923 mtx_lock(&qgroup->tqg_lock);
924 qgroup->tqg_adjusting = 0;
925
926 return (0);
927 }
928
929 int
930 taskqgroup_adjust(struct taskqgroup *qgroup, int cnt, int stride)
931 {
932 int error;
933
934 mtx_lock(&qgroup->tqg_lock);
935 error = _taskqgroup_adjust(qgroup, cnt, stride);
936 mtx_unlock(&qgroup->tqg_lock);
937
938 return (error);
939 }
940
941 struct taskqgroup *
942 taskqgroup_create(char *name)
943 {
944 struct taskqgroup *qgroup;
945
946 qgroup = malloc(sizeof(*qgroup), M_GTASKQUEUE, M_WAITOK | M_ZERO);
947 mtx_init(&qgroup->tqg_lock, "taskqgroup", NULL, MTX_DEF);
948 qgroup->tqg_name = name;
949 LIST_INIT(&qgroup->tqg_queue[0].tgc_tasks);
950
951 return (qgroup);
952 }
953
954 void
955 taskqgroup_destroy(struct taskqgroup *qgroup)
956 {
957
958 }
Cache object: 6005332184ec73c22635817fee62b222
|