xref: /openbmc/linux/kernel/padata.c (revision ca79522c)
1 /*
2  * padata.c - generic interface to process data streams in parallel
3  *
4  * See Documentation/padata.txt for an api documentation.
5  *
6  * Copyright (C) 2008, 2009 secunet Security Networks AG
7  * Copyright (C) 2008, 2009 Steffen Klassert <steffen.klassert@secunet.com>
8  *
9  * This program is free software; you can redistribute it and/or modify it
10  * under the terms and conditions of the GNU General Public License,
11  * version 2, as published by the Free Software Foundation.
12  *
13  * This program is distributed in the hope it will be useful, but WITHOUT
14  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
15  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
16  * more details.
17  *
18  * You should have received a copy of the GNU General Public License along with
19  * this program; if not, write to the Free Software Foundation, Inc.,
20  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
21  */
22 
23 #include <linux/export.h>
24 #include <linux/cpumask.h>
25 #include <linux/err.h>
26 #include <linux/cpu.h>
27 #include <linux/padata.h>
28 #include <linux/mutex.h>
29 #include <linux/sched.h>
30 #include <linux/slab.h>
31 #include <linux/sysfs.h>
32 #include <linux/rcupdate.h>
33 
34 #define MAX_OBJ_NUM 1000
35 
36 static int padata_index_to_cpu(struct parallel_data *pd, int cpu_index)
37 {
38 	int cpu, target_cpu;
39 
40 	target_cpu = cpumask_first(pd->cpumask.pcpu);
41 	for (cpu = 0; cpu < cpu_index; cpu++)
42 		target_cpu = cpumask_next(target_cpu, pd->cpumask.pcpu);
43 
44 	return target_cpu;
45 }
46 
47 static int padata_cpu_hash(struct parallel_data *pd)
48 {
49 	int cpu_index;
50 
51 	/*
52 	 * Hash the sequence numbers to the cpus by taking
53 	 * seq_nr mod. number of cpus in use.
54 	 */
55 
56 	spin_lock(&pd->seq_lock);
57 	cpu_index =  pd->seq_nr % cpumask_weight(pd->cpumask.pcpu);
58 	pd->seq_nr++;
59 	spin_unlock(&pd->seq_lock);
60 
61 	return padata_index_to_cpu(pd, cpu_index);
62 }
63 
64 static void padata_parallel_worker(struct work_struct *parallel_work)
65 {
66 	struct padata_parallel_queue *pqueue;
67 	struct parallel_data *pd;
68 	struct padata_instance *pinst;
69 	LIST_HEAD(local_list);
70 
71 	local_bh_disable();
72 	pqueue = container_of(parallel_work,
73 			      struct padata_parallel_queue, work);
74 	pd = pqueue->pd;
75 	pinst = pd->pinst;
76 
77 	spin_lock(&pqueue->parallel.lock);
78 	list_replace_init(&pqueue->parallel.list, &local_list);
79 	spin_unlock(&pqueue->parallel.lock);
80 
81 	while (!list_empty(&local_list)) {
82 		struct padata_priv *padata;
83 
84 		padata = list_entry(local_list.next,
85 				    struct padata_priv, list);
86 
87 		list_del_init(&padata->list);
88 
89 		padata->parallel(padata);
90 	}
91 
92 	local_bh_enable();
93 }
94 
95 /**
96  * padata_do_parallel - padata parallelization function
97  *
98  * @pinst: padata instance
99  * @padata: object to be parallelized
100  * @cb_cpu: cpu the serialization callback function will run on,
101  *          must be in the serial cpumask of padata(i.e. cpumask.cbcpu).
102  *
103  * The parallelization callback function will run with BHs off.
104  * Note: Every object which is parallelized by padata_do_parallel
105  * must be seen by padata_do_serial.
106  */
107 int padata_do_parallel(struct padata_instance *pinst,
108 		       struct padata_priv *padata, int cb_cpu)
109 {
110 	int target_cpu, err;
111 	struct padata_parallel_queue *queue;
112 	struct parallel_data *pd;
113 
114 	rcu_read_lock_bh();
115 
116 	pd = rcu_dereference(pinst->pd);
117 
118 	err = -EINVAL;
119 	if (!(pinst->flags & PADATA_INIT) || pinst->flags & PADATA_INVALID)
120 		goto out;
121 
122 	if (!cpumask_test_cpu(cb_cpu, pd->cpumask.cbcpu))
123 		goto out;
124 
125 	err =  -EBUSY;
126 	if ((pinst->flags & PADATA_RESET))
127 		goto out;
128 
129 	if (atomic_read(&pd->refcnt) >= MAX_OBJ_NUM)
130 		goto out;
131 
132 	err = 0;
133 	atomic_inc(&pd->refcnt);
134 	padata->pd = pd;
135 	padata->cb_cpu = cb_cpu;
136 
137 	target_cpu = padata_cpu_hash(pd);
138 	queue = per_cpu_ptr(pd->pqueue, target_cpu);
139 
140 	spin_lock(&queue->parallel.lock);
141 	list_add_tail(&padata->list, &queue->parallel.list);
142 	spin_unlock(&queue->parallel.lock);
143 
144 	queue_work_on(target_cpu, pinst->wq, &queue->work);
145 
146 out:
147 	rcu_read_unlock_bh();
148 
149 	return err;
150 }
151 EXPORT_SYMBOL(padata_do_parallel);
152 
153 /*
154  * padata_get_next - Get the next object that needs serialization.
155  *
156  * Return values are:
157  *
158  * A pointer to the control struct of the next object that needs
159  * serialization, if present in one of the percpu reorder queues.
160  *
161  * NULL, if all percpu reorder queues are empty.
162  *
163  * -EINPROGRESS, if the next object that needs serialization will
164  *  be parallel processed by another cpu and is not yet present in
165  *  the cpu's reorder queue.
166  *
167  * -ENODATA, if this cpu has to do the parallel processing for
168  *  the next object.
169  */
170 static struct padata_priv *padata_get_next(struct parallel_data *pd)
171 {
172 	int cpu, num_cpus;
173 	unsigned int next_nr, next_index;
174 	struct padata_parallel_queue *next_queue;
175 	struct padata_priv *padata;
176 	struct padata_list *reorder;
177 
178 	num_cpus = cpumask_weight(pd->cpumask.pcpu);
179 
180 	/*
181 	 * Calculate the percpu reorder queue and the sequence
182 	 * number of the next object.
183 	 */
184 	next_nr = pd->processed;
185 	next_index = next_nr % num_cpus;
186 	cpu = padata_index_to_cpu(pd, next_index);
187 	next_queue = per_cpu_ptr(pd->pqueue, cpu);
188 
189 	padata = NULL;
190 
191 	reorder = &next_queue->reorder;
192 
193 	if (!list_empty(&reorder->list)) {
194 		padata = list_entry(reorder->list.next,
195 				    struct padata_priv, list);
196 
197 		spin_lock(&reorder->lock);
198 		list_del_init(&padata->list);
199 		atomic_dec(&pd->reorder_objects);
200 		spin_unlock(&reorder->lock);
201 
202 		pd->processed++;
203 
204 		goto out;
205 	}
206 
207 	if (__this_cpu_read(pd->pqueue->cpu_index) == next_queue->cpu_index) {
208 		padata = ERR_PTR(-ENODATA);
209 		goto out;
210 	}
211 
212 	padata = ERR_PTR(-EINPROGRESS);
213 out:
214 	return padata;
215 }
216 
217 static void padata_reorder(struct parallel_data *pd)
218 {
219 	int cb_cpu;
220 	struct padata_priv *padata;
221 	struct padata_serial_queue *squeue;
222 	struct padata_instance *pinst = pd->pinst;
223 
224 	/*
225 	 * We need to ensure that only one cpu can work on dequeueing of
226 	 * the reorder queue the time. Calculating in which percpu reorder
227 	 * queue the next object will arrive takes some time. A spinlock
228 	 * would be highly contended. Also it is not clear in which order
229 	 * the objects arrive to the reorder queues. So a cpu could wait to
230 	 * get the lock just to notice that there is nothing to do at the
231 	 * moment. Therefore we use a trylock and let the holder of the lock
232 	 * care for all the objects enqueued during the holdtime of the lock.
233 	 */
234 	if (!spin_trylock_bh(&pd->lock))
235 		return;
236 
237 	while (1) {
238 		padata = padata_get_next(pd);
239 
240 		/*
241 		 * All reorder queues are empty, or the next object that needs
242 		 * serialization is parallel processed by another cpu and is
243 		 * still on it's way to the cpu's reorder queue, nothing to
244 		 * do for now.
245 		 */
246 		if (!padata || PTR_ERR(padata) == -EINPROGRESS)
247 			break;
248 
249 		/*
250 		 * This cpu has to do the parallel processing of the next
251 		 * object. It's waiting in the cpu's parallelization queue,
252 		 * so exit immediately.
253 		 */
254 		if (PTR_ERR(padata) == -ENODATA) {
255 			del_timer(&pd->timer);
256 			spin_unlock_bh(&pd->lock);
257 			return;
258 		}
259 
260 		cb_cpu = padata->cb_cpu;
261 		squeue = per_cpu_ptr(pd->squeue, cb_cpu);
262 
263 		spin_lock(&squeue->serial.lock);
264 		list_add_tail(&padata->list, &squeue->serial.list);
265 		spin_unlock(&squeue->serial.lock);
266 
267 		queue_work_on(cb_cpu, pinst->wq, &squeue->work);
268 	}
269 
270 	spin_unlock_bh(&pd->lock);
271 
272 	/*
273 	 * The next object that needs serialization might have arrived to
274 	 * the reorder queues in the meantime, we will be called again
275 	 * from the timer function if no one else cares for it.
276 	 */
277 	if (atomic_read(&pd->reorder_objects)
278 			&& !(pinst->flags & PADATA_RESET))
279 		mod_timer(&pd->timer, jiffies + HZ);
280 	else
281 		del_timer(&pd->timer);
282 
283 	return;
284 }
285 
286 static void padata_reorder_timer(unsigned long arg)
287 {
288 	struct parallel_data *pd = (struct parallel_data *)arg;
289 
290 	padata_reorder(pd);
291 }
292 
293 static void padata_serial_worker(struct work_struct *serial_work)
294 {
295 	struct padata_serial_queue *squeue;
296 	struct parallel_data *pd;
297 	LIST_HEAD(local_list);
298 
299 	local_bh_disable();
300 	squeue = container_of(serial_work, struct padata_serial_queue, work);
301 	pd = squeue->pd;
302 
303 	spin_lock(&squeue->serial.lock);
304 	list_replace_init(&squeue->serial.list, &local_list);
305 	spin_unlock(&squeue->serial.lock);
306 
307 	while (!list_empty(&local_list)) {
308 		struct padata_priv *padata;
309 
310 		padata = list_entry(local_list.next,
311 				    struct padata_priv, list);
312 
313 		list_del_init(&padata->list);
314 
315 		padata->serial(padata);
316 		atomic_dec(&pd->refcnt);
317 	}
318 	local_bh_enable();
319 }
320 
321 /**
322  * padata_do_serial - padata serialization function
323  *
324  * @padata: object to be serialized.
325  *
326  * padata_do_serial must be called for every parallelized object.
327  * The serialization callback function will run with BHs off.
328  */
329 void padata_do_serial(struct padata_priv *padata)
330 {
331 	int cpu;
332 	struct padata_parallel_queue *pqueue;
333 	struct parallel_data *pd;
334 
335 	pd = padata->pd;
336 
337 	cpu = get_cpu();
338 	pqueue = per_cpu_ptr(pd->pqueue, cpu);
339 
340 	spin_lock(&pqueue->reorder.lock);
341 	atomic_inc(&pd->reorder_objects);
342 	list_add_tail(&padata->list, &pqueue->reorder.list);
343 	spin_unlock(&pqueue->reorder.lock);
344 
345 	put_cpu();
346 
347 	padata_reorder(pd);
348 }
349 EXPORT_SYMBOL(padata_do_serial);
350 
351 static int padata_setup_cpumasks(struct parallel_data *pd,
352 				 const struct cpumask *pcpumask,
353 				 const struct cpumask *cbcpumask)
354 {
355 	if (!alloc_cpumask_var(&pd->cpumask.pcpu, GFP_KERNEL))
356 		return -ENOMEM;
357 
358 	cpumask_and(pd->cpumask.pcpu, pcpumask, cpu_online_mask);
359 	if (!alloc_cpumask_var(&pd->cpumask.cbcpu, GFP_KERNEL)) {
360 		free_cpumask_var(pd->cpumask.cbcpu);
361 		return -ENOMEM;
362 	}
363 
364 	cpumask_and(pd->cpumask.cbcpu, cbcpumask, cpu_online_mask);
365 	return 0;
366 }
367 
368 static void __padata_list_init(struct padata_list *pd_list)
369 {
370 	INIT_LIST_HEAD(&pd_list->list);
371 	spin_lock_init(&pd_list->lock);
372 }
373 
374 /* Initialize all percpu queues used by serial workers */
375 static void padata_init_squeues(struct parallel_data *pd)
376 {
377 	int cpu;
378 	struct padata_serial_queue *squeue;
379 
380 	for_each_cpu(cpu, pd->cpumask.cbcpu) {
381 		squeue = per_cpu_ptr(pd->squeue, cpu);
382 		squeue->pd = pd;
383 		__padata_list_init(&squeue->serial);
384 		INIT_WORK(&squeue->work, padata_serial_worker);
385 	}
386 }
387 
388 /* Initialize all percpu queues used by parallel workers */
389 static void padata_init_pqueues(struct parallel_data *pd)
390 {
391 	int cpu_index, cpu;
392 	struct padata_parallel_queue *pqueue;
393 
394 	cpu_index = 0;
395 	for_each_cpu(cpu, pd->cpumask.pcpu) {
396 		pqueue = per_cpu_ptr(pd->pqueue, cpu);
397 		pqueue->pd = pd;
398 		pqueue->cpu_index = cpu_index;
399 		cpu_index++;
400 
401 		__padata_list_init(&pqueue->reorder);
402 		__padata_list_init(&pqueue->parallel);
403 		INIT_WORK(&pqueue->work, padata_parallel_worker);
404 		atomic_set(&pqueue->num_obj, 0);
405 	}
406 }
407 
408 /* Allocate and initialize the internal cpumask dependend resources. */
409 static struct parallel_data *padata_alloc_pd(struct padata_instance *pinst,
410 					     const struct cpumask *pcpumask,
411 					     const struct cpumask *cbcpumask)
412 {
413 	struct parallel_data *pd;
414 
415 	pd = kzalloc(sizeof(struct parallel_data), GFP_KERNEL);
416 	if (!pd)
417 		goto err;
418 
419 	pd->pqueue = alloc_percpu(struct padata_parallel_queue);
420 	if (!pd->pqueue)
421 		goto err_free_pd;
422 
423 	pd->squeue = alloc_percpu(struct padata_serial_queue);
424 	if (!pd->squeue)
425 		goto err_free_pqueue;
426 	if (padata_setup_cpumasks(pd, pcpumask, cbcpumask) < 0)
427 		goto err_free_squeue;
428 
429 	padata_init_pqueues(pd);
430 	padata_init_squeues(pd);
431 	setup_timer(&pd->timer, padata_reorder_timer, (unsigned long)pd);
432 	pd->seq_nr = 0;
433 	atomic_set(&pd->reorder_objects, 0);
434 	atomic_set(&pd->refcnt, 0);
435 	pd->pinst = pinst;
436 	spin_lock_init(&pd->lock);
437 
438 	return pd;
439 
440 err_free_squeue:
441 	free_percpu(pd->squeue);
442 err_free_pqueue:
443 	free_percpu(pd->pqueue);
444 err_free_pd:
445 	kfree(pd);
446 err:
447 	return NULL;
448 }
449 
450 static void padata_free_pd(struct parallel_data *pd)
451 {
452 	free_cpumask_var(pd->cpumask.pcpu);
453 	free_cpumask_var(pd->cpumask.cbcpu);
454 	free_percpu(pd->pqueue);
455 	free_percpu(pd->squeue);
456 	kfree(pd);
457 }
458 
459 /* Flush all objects out of the padata queues. */
460 static void padata_flush_queues(struct parallel_data *pd)
461 {
462 	int cpu;
463 	struct padata_parallel_queue *pqueue;
464 	struct padata_serial_queue *squeue;
465 
466 	for_each_cpu(cpu, pd->cpumask.pcpu) {
467 		pqueue = per_cpu_ptr(pd->pqueue, cpu);
468 		flush_work(&pqueue->work);
469 	}
470 
471 	del_timer_sync(&pd->timer);
472 
473 	if (atomic_read(&pd->reorder_objects))
474 		padata_reorder(pd);
475 
476 	for_each_cpu(cpu, pd->cpumask.cbcpu) {
477 		squeue = per_cpu_ptr(pd->squeue, cpu);
478 		flush_work(&squeue->work);
479 	}
480 
481 	BUG_ON(atomic_read(&pd->refcnt) != 0);
482 }
483 
484 static void __padata_start(struct padata_instance *pinst)
485 {
486 	pinst->flags |= PADATA_INIT;
487 }
488 
489 static void __padata_stop(struct padata_instance *pinst)
490 {
491 	if (!(pinst->flags & PADATA_INIT))
492 		return;
493 
494 	pinst->flags &= ~PADATA_INIT;
495 
496 	synchronize_rcu();
497 
498 	get_online_cpus();
499 	padata_flush_queues(pinst->pd);
500 	put_online_cpus();
501 }
502 
503 /* Replace the internal control structure with a new one. */
504 static void padata_replace(struct padata_instance *pinst,
505 			   struct parallel_data *pd_new)
506 {
507 	struct parallel_data *pd_old = pinst->pd;
508 	int notification_mask = 0;
509 
510 	pinst->flags |= PADATA_RESET;
511 
512 	rcu_assign_pointer(pinst->pd, pd_new);
513 
514 	synchronize_rcu();
515 
516 	if (!cpumask_equal(pd_old->cpumask.pcpu, pd_new->cpumask.pcpu))
517 		notification_mask |= PADATA_CPU_PARALLEL;
518 	if (!cpumask_equal(pd_old->cpumask.cbcpu, pd_new->cpumask.cbcpu))
519 		notification_mask |= PADATA_CPU_SERIAL;
520 
521 	padata_flush_queues(pd_old);
522 	padata_free_pd(pd_old);
523 
524 	if (notification_mask)
525 		blocking_notifier_call_chain(&pinst->cpumask_change_notifier,
526 					     notification_mask,
527 					     &pd_new->cpumask);
528 
529 	pinst->flags &= ~PADATA_RESET;
530 }
531 
532 /**
533  * padata_register_cpumask_notifier - Registers a notifier that will be called
534  *                             if either pcpu or cbcpu or both cpumasks change.
535  *
536  * @pinst: A poineter to padata instance
537  * @nblock: A pointer to notifier block.
538  */
539 int padata_register_cpumask_notifier(struct padata_instance *pinst,
540 				     struct notifier_block *nblock)
541 {
542 	return blocking_notifier_chain_register(&pinst->cpumask_change_notifier,
543 						nblock);
544 }
545 EXPORT_SYMBOL(padata_register_cpumask_notifier);
546 
547 /**
548  * padata_unregister_cpumask_notifier - Unregisters cpumask notifier
549  *        registered earlier  using padata_register_cpumask_notifier
550  *
551  * @pinst: A pointer to data instance.
552  * @nlock: A pointer to notifier block.
553  */
554 int padata_unregister_cpumask_notifier(struct padata_instance *pinst,
555 				       struct notifier_block *nblock)
556 {
557 	return blocking_notifier_chain_unregister(
558 		&pinst->cpumask_change_notifier,
559 		nblock);
560 }
561 EXPORT_SYMBOL(padata_unregister_cpumask_notifier);
562 
563 
564 /* If cpumask contains no active cpu, we mark the instance as invalid. */
565 static bool padata_validate_cpumask(struct padata_instance *pinst,
566 				    const struct cpumask *cpumask)
567 {
568 	if (!cpumask_intersects(cpumask, cpu_online_mask)) {
569 		pinst->flags |= PADATA_INVALID;
570 		return false;
571 	}
572 
573 	pinst->flags &= ~PADATA_INVALID;
574 	return true;
575 }
576 
577 static int __padata_set_cpumasks(struct padata_instance *pinst,
578 				 cpumask_var_t pcpumask,
579 				 cpumask_var_t cbcpumask)
580 {
581 	int valid;
582 	struct parallel_data *pd;
583 
584 	valid = padata_validate_cpumask(pinst, pcpumask);
585 	if (!valid) {
586 		__padata_stop(pinst);
587 		goto out_replace;
588 	}
589 
590 	valid = padata_validate_cpumask(pinst, cbcpumask);
591 	if (!valid)
592 		__padata_stop(pinst);
593 
594 out_replace:
595 	pd = padata_alloc_pd(pinst, pcpumask, cbcpumask);
596 	if (!pd)
597 		return -ENOMEM;
598 
599 	cpumask_copy(pinst->cpumask.pcpu, pcpumask);
600 	cpumask_copy(pinst->cpumask.cbcpu, cbcpumask);
601 
602 	padata_replace(pinst, pd);
603 
604 	if (valid)
605 		__padata_start(pinst);
606 
607 	return 0;
608 }
609 
610 /**
611  * padata_set_cpumasks - Set both parallel and serial cpumasks. The first
612  *                       one is used by parallel workers and the second one
613  *                       by the wokers doing serialization.
614  *
615  * @pinst: padata instance
616  * @pcpumask: the cpumask to use for parallel workers
617  * @cbcpumask: the cpumsak to use for serial workers
618  */
619 int padata_set_cpumasks(struct padata_instance *pinst, cpumask_var_t pcpumask,
620 			cpumask_var_t cbcpumask)
621 {
622 	int err;
623 
624 	mutex_lock(&pinst->lock);
625 	get_online_cpus();
626 
627 	err = __padata_set_cpumasks(pinst, pcpumask, cbcpumask);
628 
629 	put_online_cpus();
630 	mutex_unlock(&pinst->lock);
631 
632 	return err;
633 
634 }
635 EXPORT_SYMBOL(padata_set_cpumasks);
636 
637 /**
638  * padata_set_cpumask: Sets specified by @cpumask_type cpumask to the value
639  *                     equivalent to @cpumask.
640  *
641  * @pinst: padata instance
642  * @cpumask_type: PADATA_CPU_SERIAL or PADATA_CPU_PARALLEL corresponding
643  *                to parallel and serial cpumasks respectively.
644  * @cpumask: the cpumask to use
645  */
646 int padata_set_cpumask(struct padata_instance *pinst, int cpumask_type,
647 		       cpumask_var_t cpumask)
648 {
649 	struct cpumask *serial_mask, *parallel_mask;
650 	int err = -EINVAL;
651 
652 	mutex_lock(&pinst->lock);
653 	get_online_cpus();
654 
655 	switch (cpumask_type) {
656 	case PADATA_CPU_PARALLEL:
657 		serial_mask = pinst->cpumask.cbcpu;
658 		parallel_mask = cpumask;
659 		break;
660 	case PADATA_CPU_SERIAL:
661 		parallel_mask = pinst->cpumask.pcpu;
662 		serial_mask = cpumask;
663 		break;
664 	default:
665 		 goto out;
666 	}
667 
668 	err =  __padata_set_cpumasks(pinst, parallel_mask, serial_mask);
669 
670 out:
671 	put_online_cpus();
672 	mutex_unlock(&pinst->lock);
673 
674 	return err;
675 }
676 EXPORT_SYMBOL(padata_set_cpumask);
677 
678 static int __padata_add_cpu(struct padata_instance *pinst, int cpu)
679 {
680 	struct parallel_data *pd;
681 
682 	if (cpumask_test_cpu(cpu, cpu_online_mask)) {
683 		pd = padata_alloc_pd(pinst, pinst->cpumask.pcpu,
684 				     pinst->cpumask.cbcpu);
685 		if (!pd)
686 			return -ENOMEM;
687 
688 		padata_replace(pinst, pd);
689 
690 		if (padata_validate_cpumask(pinst, pinst->cpumask.pcpu) &&
691 		    padata_validate_cpumask(pinst, pinst->cpumask.cbcpu))
692 			__padata_start(pinst);
693 	}
694 
695 	return 0;
696 }
697 
698  /**
699  * padata_add_cpu - add a cpu to one or both(parallel and serial)
700  *                  padata cpumasks.
701  *
702  * @pinst: padata instance
703  * @cpu: cpu to add
704  * @mask: bitmask of flags specifying to which cpumask @cpu shuld be added.
705  *        The @mask may be any combination of the following flags:
706  *          PADATA_CPU_SERIAL   - serial cpumask
707  *          PADATA_CPU_PARALLEL - parallel cpumask
708  */
709 
710 int padata_add_cpu(struct padata_instance *pinst, int cpu, int mask)
711 {
712 	int err;
713 
714 	if (!(mask & (PADATA_CPU_SERIAL | PADATA_CPU_PARALLEL)))
715 		return -EINVAL;
716 
717 	mutex_lock(&pinst->lock);
718 
719 	get_online_cpus();
720 	if (mask & PADATA_CPU_SERIAL)
721 		cpumask_set_cpu(cpu, pinst->cpumask.cbcpu);
722 	if (mask & PADATA_CPU_PARALLEL)
723 		cpumask_set_cpu(cpu, pinst->cpumask.pcpu);
724 
725 	err = __padata_add_cpu(pinst, cpu);
726 	put_online_cpus();
727 
728 	mutex_unlock(&pinst->lock);
729 
730 	return err;
731 }
732 EXPORT_SYMBOL(padata_add_cpu);
733 
734 static int __padata_remove_cpu(struct padata_instance *pinst, int cpu)
735 {
736 	struct parallel_data *pd = NULL;
737 
738 	if (cpumask_test_cpu(cpu, cpu_online_mask)) {
739 
740 		if (!padata_validate_cpumask(pinst, pinst->cpumask.pcpu) ||
741 		    !padata_validate_cpumask(pinst, pinst->cpumask.cbcpu))
742 			__padata_stop(pinst);
743 
744 		pd = padata_alloc_pd(pinst, pinst->cpumask.pcpu,
745 				     pinst->cpumask.cbcpu);
746 		if (!pd)
747 			return -ENOMEM;
748 
749 		padata_replace(pinst, pd);
750 
751 		cpumask_clear_cpu(cpu, pd->cpumask.cbcpu);
752 		cpumask_clear_cpu(cpu, pd->cpumask.pcpu);
753 	}
754 
755 	return 0;
756 }
757 
758  /**
759  * padata_remove_cpu - remove a cpu from the one or both(serial and parallel)
760  *                     padata cpumasks.
761  *
762  * @pinst: padata instance
763  * @cpu: cpu to remove
764  * @mask: bitmask specifying from which cpumask @cpu should be removed
765  *        The @mask may be any combination of the following flags:
766  *          PADATA_CPU_SERIAL   - serial cpumask
767  *          PADATA_CPU_PARALLEL - parallel cpumask
768  */
769 int padata_remove_cpu(struct padata_instance *pinst, int cpu, int mask)
770 {
771 	int err;
772 
773 	if (!(mask & (PADATA_CPU_SERIAL | PADATA_CPU_PARALLEL)))
774 		return -EINVAL;
775 
776 	mutex_lock(&pinst->lock);
777 
778 	get_online_cpus();
779 	if (mask & PADATA_CPU_SERIAL)
780 		cpumask_clear_cpu(cpu, pinst->cpumask.cbcpu);
781 	if (mask & PADATA_CPU_PARALLEL)
782 		cpumask_clear_cpu(cpu, pinst->cpumask.pcpu);
783 
784 	err = __padata_remove_cpu(pinst, cpu);
785 	put_online_cpus();
786 
787 	mutex_unlock(&pinst->lock);
788 
789 	return err;
790 }
791 EXPORT_SYMBOL(padata_remove_cpu);
792 
793 /**
794  * padata_start - start the parallel processing
795  *
796  * @pinst: padata instance to start
797  */
798 int padata_start(struct padata_instance *pinst)
799 {
800 	int err = 0;
801 
802 	mutex_lock(&pinst->lock);
803 
804 	if (pinst->flags & PADATA_INVALID)
805 		err =-EINVAL;
806 
807 	 __padata_start(pinst);
808 
809 	mutex_unlock(&pinst->lock);
810 
811 	return err;
812 }
813 EXPORT_SYMBOL(padata_start);
814 
815 /**
816  * padata_stop - stop the parallel processing
817  *
818  * @pinst: padata instance to stop
819  */
820 void padata_stop(struct padata_instance *pinst)
821 {
822 	mutex_lock(&pinst->lock);
823 	__padata_stop(pinst);
824 	mutex_unlock(&pinst->lock);
825 }
826 EXPORT_SYMBOL(padata_stop);
827 
828 #ifdef CONFIG_HOTPLUG_CPU
829 
830 static inline int pinst_has_cpu(struct padata_instance *pinst, int cpu)
831 {
832 	return cpumask_test_cpu(cpu, pinst->cpumask.pcpu) ||
833 		cpumask_test_cpu(cpu, pinst->cpumask.cbcpu);
834 }
835 
836 
837 static int padata_cpu_callback(struct notifier_block *nfb,
838 			       unsigned long action, void *hcpu)
839 {
840 	int err;
841 	struct padata_instance *pinst;
842 	int cpu = (unsigned long)hcpu;
843 
844 	pinst = container_of(nfb, struct padata_instance, cpu_notifier);
845 
846 	switch (action) {
847 	case CPU_ONLINE:
848 	case CPU_ONLINE_FROZEN:
849 		if (!pinst_has_cpu(pinst, cpu))
850 			break;
851 		mutex_lock(&pinst->lock);
852 		err = __padata_add_cpu(pinst, cpu);
853 		mutex_unlock(&pinst->lock);
854 		if (err)
855 			return notifier_from_errno(err);
856 		break;
857 
858 	case CPU_DOWN_PREPARE:
859 	case CPU_DOWN_PREPARE_FROZEN:
860 		if (!pinst_has_cpu(pinst, cpu))
861 			break;
862 		mutex_lock(&pinst->lock);
863 		err = __padata_remove_cpu(pinst, cpu);
864 		mutex_unlock(&pinst->lock);
865 		if (err)
866 			return notifier_from_errno(err);
867 		break;
868 
869 	case CPU_UP_CANCELED:
870 	case CPU_UP_CANCELED_FROZEN:
871 		if (!pinst_has_cpu(pinst, cpu))
872 			break;
873 		mutex_lock(&pinst->lock);
874 		__padata_remove_cpu(pinst, cpu);
875 		mutex_unlock(&pinst->lock);
876 
877 	case CPU_DOWN_FAILED:
878 	case CPU_DOWN_FAILED_FROZEN:
879 		if (!pinst_has_cpu(pinst, cpu))
880 			break;
881 		mutex_lock(&pinst->lock);
882 		__padata_add_cpu(pinst, cpu);
883 		mutex_unlock(&pinst->lock);
884 	}
885 
886 	return NOTIFY_OK;
887 }
888 #endif
889 
890 static void __padata_free(struct padata_instance *pinst)
891 {
892 #ifdef CONFIG_HOTPLUG_CPU
893 	unregister_hotcpu_notifier(&pinst->cpu_notifier);
894 #endif
895 
896 	padata_stop(pinst);
897 	padata_free_pd(pinst->pd);
898 	free_cpumask_var(pinst->cpumask.pcpu);
899 	free_cpumask_var(pinst->cpumask.cbcpu);
900 	kfree(pinst);
901 }
902 
903 #define kobj2pinst(_kobj)					\
904 	container_of(_kobj, struct padata_instance, kobj)
905 #define attr2pentry(_attr)					\
906 	container_of(_attr, struct padata_sysfs_entry, attr)
907 
908 static void padata_sysfs_release(struct kobject *kobj)
909 {
910 	struct padata_instance *pinst = kobj2pinst(kobj);
911 	__padata_free(pinst);
912 }
913 
914 struct padata_sysfs_entry {
915 	struct attribute attr;
916 	ssize_t (*show)(struct padata_instance *, struct attribute *, char *);
917 	ssize_t (*store)(struct padata_instance *, struct attribute *,
918 			 const char *, size_t);
919 };
920 
921 static ssize_t show_cpumask(struct padata_instance *pinst,
922 			    struct attribute *attr,  char *buf)
923 {
924 	struct cpumask *cpumask;
925 	ssize_t len;
926 
927 	mutex_lock(&pinst->lock);
928 	if (!strcmp(attr->name, "serial_cpumask"))
929 		cpumask = pinst->cpumask.cbcpu;
930 	else
931 		cpumask = pinst->cpumask.pcpu;
932 
933 	len = bitmap_scnprintf(buf, PAGE_SIZE, cpumask_bits(cpumask),
934 			       nr_cpu_ids);
935 	if (PAGE_SIZE - len < 2)
936 		len = -EINVAL;
937 	else
938 		len += sprintf(buf + len, "\n");
939 
940 	mutex_unlock(&pinst->lock);
941 	return len;
942 }
943 
944 static ssize_t store_cpumask(struct padata_instance *pinst,
945 			     struct attribute *attr,
946 			     const char *buf, size_t count)
947 {
948 	cpumask_var_t new_cpumask;
949 	ssize_t ret;
950 	int mask_type;
951 
952 	if (!alloc_cpumask_var(&new_cpumask, GFP_KERNEL))
953 		return -ENOMEM;
954 
955 	ret = bitmap_parse(buf, count, cpumask_bits(new_cpumask),
956 			   nr_cpumask_bits);
957 	if (ret < 0)
958 		goto out;
959 
960 	mask_type = !strcmp(attr->name, "serial_cpumask") ?
961 		PADATA_CPU_SERIAL : PADATA_CPU_PARALLEL;
962 	ret = padata_set_cpumask(pinst, mask_type, new_cpumask);
963 	if (!ret)
964 		ret = count;
965 
966 out:
967 	free_cpumask_var(new_cpumask);
968 	return ret;
969 }
970 
971 #define PADATA_ATTR_RW(_name, _show_name, _store_name)		\
972 	static struct padata_sysfs_entry _name##_attr =		\
973 		__ATTR(_name, 0644, _show_name, _store_name)
974 #define PADATA_ATTR_RO(_name, _show_name)		\
975 	static struct padata_sysfs_entry _name##_attr = \
976 		__ATTR(_name, 0400, _show_name, NULL)
977 
978 PADATA_ATTR_RW(serial_cpumask, show_cpumask, store_cpumask);
979 PADATA_ATTR_RW(parallel_cpumask, show_cpumask, store_cpumask);
980 
981 /*
982  * Padata sysfs provides the following objects:
983  * serial_cpumask   [RW] - cpumask for serial workers
984  * parallel_cpumask [RW] - cpumask for parallel workers
985  */
986 static struct attribute *padata_default_attrs[] = {
987 	&serial_cpumask_attr.attr,
988 	&parallel_cpumask_attr.attr,
989 	NULL,
990 };
991 
992 static ssize_t padata_sysfs_show(struct kobject *kobj,
993 				 struct attribute *attr, char *buf)
994 {
995 	struct padata_instance *pinst;
996 	struct padata_sysfs_entry *pentry;
997 	ssize_t ret = -EIO;
998 
999 	pinst = kobj2pinst(kobj);
1000 	pentry = attr2pentry(attr);
1001 	if (pentry->show)
1002 		ret = pentry->show(pinst, attr, buf);
1003 
1004 	return ret;
1005 }
1006 
1007 static ssize_t padata_sysfs_store(struct kobject *kobj, struct attribute *attr,
1008 				  const char *buf, size_t count)
1009 {
1010 	struct padata_instance *pinst;
1011 	struct padata_sysfs_entry *pentry;
1012 	ssize_t ret = -EIO;
1013 
1014 	pinst = kobj2pinst(kobj);
1015 	pentry = attr2pentry(attr);
1016 	if (pentry->show)
1017 		ret = pentry->store(pinst, attr, buf, count);
1018 
1019 	return ret;
1020 }
1021 
1022 static const struct sysfs_ops padata_sysfs_ops = {
1023 	.show = padata_sysfs_show,
1024 	.store = padata_sysfs_store,
1025 };
1026 
1027 static struct kobj_type padata_attr_type = {
1028 	.sysfs_ops = &padata_sysfs_ops,
1029 	.default_attrs = padata_default_attrs,
1030 	.release = padata_sysfs_release,
1031 };
1032 
1033 /**
1034  * padata_alloc_possible - Allocate and initialize padata instance.
1035  *                         Use the cpu_possible_mask for serial and
1036  *                         parallel workers.
1037  *
1038  * @wq: workqueue to use for the allocated padata instance
1039  */
1040 struct padata_instance *padata_alloc_possible(struct workqueue_struct *wq)
1041 {
1042 	return padata_alloc(wq, cpu_possible_mask, cpu_possible_mask);
1043 }
1044 EXPORT_SYMBOL(padata_alloc_possible);
1045 
1046 /**
1047  * padata_alloc - allocate and initialize a padata instance and specify
1048  *                cpumasks for serial and parallel workers.
1049  *
1050  * @wq: workqueue to use for the allocated padata instance
1051  * @pcpumask: cpumask that will be used for padata parallelization
1052  * @cbcpumask: cpumask that will be used for padata serialization
1053  */
1054 struct padata_instance *padata_alloc(struct workqueue_struct *wq,
1055 				     const struct cpumask *pcpumask,
1056 				     const struct cpumask *cbcpumask)
1057 {
1058 	struct padata_instance *pinst;
1059 	struct parallel_data *pd = NULL;
1060 
1061 	pinst = kzalloc(sizeof(struct padata_instance), GFP_KERNEL);
1062 	if (!pinst)
1063 		goto err;
1064 
1065 	get_online_cpus();
1066 	if (!alloc_cpumask_var(&pinst->cpumask.pcpu, GFP_KERNEL))
1067 		goto err_free_inst;
1068 	if (!alloc_cpumask_var(&pinst->cpumask.cbcpu, GFP_KERNEL)) {
1069 		free_cpumask_var(pinst->cpumask.pcpu);
1070 		goto err_free_inst;
1071 	}
1072 	if (!padata_validate_cpumask(pinst, pcpumask) ||
1073 	    !padata_validate_cpumask(pinst, cbcpumask))
1074 		goto err_free_masks;
1075 
1076 	pd = padata_alloc_pd(pinst, pcpumask, cbcpumask);
1077 	if (!pd)
1078 		goto err_free_masks;
1079 
1080 	rcu_assign_pointer(pinst->pd, pd);
1081 
1082 	pinst->wq = wq;
1083 
1084 	cpumask_copy(pinst->cpumask.pcpu, pcpumask);
1085 	cpumask_copy(pinst->cpumask.cbcpu, cbcpumask);
1086 
1087 	pinst->flags = 0;
1088 
1089 #ifdef CONFIG_HOTPLUG_CPU
1090 	pinst->cpu_notifier.notifier_call = padata_cpu_callback;
1091 	pinst->cpu_notifier.priority = 0;
1092 	register_hotcpu_notifier(&pinst->cpu_notifier);
1093 #endif
1094 
1095 	put_online_cpus();
1096 
1097 	BLOCKING_INIT_NOTIFIER_HEAD(&pinst->cpumask_change_notifier);
1098 	kobject_init(&pinst->kobj, &padata_attr_type);
1099 	mutex_init(&pinst->lock);
1100 
1101 	return pinst;
1102 
1103 err_free_masks:
1104 	free_cpumask_var(pinst->cpumask.pcpu);
1105 	free_cpumask_var(pinst->cpumask.cbcpu);
1106 err_free_inst:
1107 	kfree(pinst);
1108 	put_online_cpus();
1109 err:
1110 	return NULL;
1111 }
1112 EXPORT_SYMBOL(padata_alloc);
1113 
1114 /**
1115  * padata_free - free a padata instance
1116  *
1117  * @padata_inst: padata instance to free
1118  */
1119 void padata_free(struct padata_instance *pinst)
1120 {
1121 	kobject_put(&pinst->kobj);
1122 }
1123 EXPORT_SYMBOL(padata_free);
1124