Commit | Line | Data |
---|---|---|
16295bec SK |
1 | /* |
2 | * padata.c - generic interface to process data streams in parallel | |
3 | * | |
4 | * Copyright (C) 2008, 2009 secunet Security Networks AG | |
5 | * Copyright (C) 2008, 2009 Steffen Klassert <steffen.klassert@secunet.com> | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify it | |
8 | * under the terms and conditions of the GNU General Public License, | |
9 | * version 2, as published by the Free Software Foundation. | |
10 | * | |
11 | * This program is distributed in the hope it will be useful, but WITHOUT | |
12 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
13 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
14 | * more details. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License along with | |
17 | * this program; if not, write to the Free Software Foundation, Inc., | |
18 | * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. | |
19 | */ | |
20 | ||
21 | #include <linux/module.h> | |
22 | #include <linux/cpumask.h> | |
23 | #include <linux/err.h> | |
24 | #include <linux/cpu.h> | |
25 | #include <linux/padata.h> | |
26 | #include <linux/mutex.h> | |
27 | #include <linux/sched.h> | |
5a0e3ad6 | 28 | #include <linux/slab.h> |
16295bec SK |
29 | #include <linux/rcupdate.h> |
30 | ||
749d811f | 31 | #define MAX_SEQ_NR (INT_MAX - NR_CPUS) |
97e3d94a | 32 | #define MAX_OBJ_NUM 1000 |
16295bec SK |
33 | |
34 | static int padata_index_to_cpu(struct parallel_data *pd, int cpu_index) | |
35 | { | |
36 | int cpu, target_cpu; | |
37 | ||
38 | target_cpu = cpumask_first(pd->cpumask); | |
39 | for (cpu = 0; cpu < cpu_index; cpu++) | |
40 | target_cpu = cpumask_next(target_cpu, pd->cpumask); | |
41 | ||
42 | return target_cpu; | |
43 | } | |
44 | ||
45 | static int padata_cpu_hash(struct padata_priv *padata) | |
46 | { | |
47 | int cpu_index; | |
48 | struct parallel_data *pd; | |
49 | ||
50 | pd = padata->pd; | |
51 | ||
52 | /* | |
53 | * Hash the sequence numbers to the cpus by taking | |
54 | * seq_nr mod. number of cpus in use. | |
55 | */ | |
56 | cpu_index = padata->seq_nr % cpumask_weight(pd->cpumask); | |
57 | ||
58 | return padata_index_to_cpu(pd, cpu_index); | |
59 | } | |
60 | ||
61 | static void padata_parallel_worker(struct work_struct *work) | |
62 | { | |
63 | struct padata_queue *queue; | |
64 | struct parallel_data *pd; | |
65 | struct padata_instance *pinst; | |
66 | LIST_HEAD(local_list); | |
67 | ||
68 | local_bh_disable(); | |
69 | queue = container_of(work, struct padata_queue, pwork); | |
70 | pd = queue->pd; | |
71 | pinst = pd->pinst; | |
72 | ||
73 | spin_lock(&queue->parallel.lock); | |
74 | list_replace_init(&queue->parallel.list, &local_list); | |
75 | spin_unlock(&queue->parallel.lock); | |
76 | ||
77 | while (!list_empty(&local_list)) { | |
78 | struct padata_priv *padata; | |
79 | ||
80 | padata = list_entry(local_list.next, | |
81 | struct padata_priv, list); | |
82 | ||
83 | list_del_init(&padata->list); | |
84 | ||
85 | padata->parallel(padata); | |
86 | } | |
87 | ||
88 | local_bh_enable(); | |
89 | } | |
90 | ||
0198ffd1 | 91 | /** |
16295bec SK |
92 | * padata_do_parallel - padata parallelization function |
93 | * | |
94 | * @pinst: padata instance | |
95 | * @padata: object to be parallelized | |
96 | * @cb_cpu: cpu the serialization callback function will run on, | |
97 | * must be in the cpumask of padata. | |
98 | * | |
99 | * The parallelization callback function will run with BHs off. | |
100 | * Note: Every object which is parallelized by padata_do_parallel | |
101 | * must be seen by padata_do_serial. | |
102 | */ | |
103 | int padata_do_parallel(struct padata_instance *pinst, | |
104 | struct padata_priv *padata, int cb_cpu) | |
105 | { | |
106 | int target_cpu, err; | |
107 | struct padata_queue *queue; | |
108 | struct parallel_data *pd; | |
109 | ||
110 | rcu_read_lock_bh(); | |
111 | ||
112 | pd = rcu_dereference(pinst->pd); | |
113 | ||
83f619f3 | 114 | err = -EINVAL; |
16295bec SK |
115 | if (!(pinst->flags & PADATA_INIT)) |
116 | goto out; | |
117 | ||
83f619f3 SK |
118 | if (!cpumask_test_cpu(cb_cpu, pd->cpumask)) |
119 | goto out; | |
120 | ||
16295bec SK |
121 | err = -EBUSY; |
122 | if ((pinst->flags & PADATA_RESET)) | |
123 | goto out; | |
124 | ||
125 | if (atomic_read(&pd->refcnt) >= MAX_OBJ_NUM) | |
126 | goto out; | |
127 | ||
83f619f3 | 128 | err = 0; |
16295bec SK |
129 | atomic_inc(&pd->refcnt); |
130 | padata->pd = pd; | |
131 | padata->cb_cpu = cb_cpu; | |
132 | ||
133 | if (unlikely(atomic_read(&pd->seq_nr) == pd->max_seq_nr)) | |
134 | atomic_set(&pd->seq_nr, -1); | |
135 | ||
136 | padata->seq_nr = atomic_inc_return(&pd->seq_nr); | |
137 | ||
138 | target_cpu = padata_cpu_hash(padata); | |
139 | queue = per_cpu_ptr(pd->queue, target_cpu); | |
140 | ||
141 | spin_lock(&queue->parallel.lock); | |
142 | list_add_tail(&padata->list, &queue->parallel.list); | |
143 | spin_unlock(&queue->parallel.lock); | |
144 | ||
145 | queue_work_on(target_cpu, pinst->wq, &queue->pwork); | |
146 | ||
147 | out: | |
148 | rcu_read_unlock_bh(); | |
149 | ||
150 | return err; | |
151 | } | |
152 | EXPORT_SYMBOL(padata_do_parallel); | |
153 | ||
0198ffd1 SK |
154 | /* |
155 | * padata_get_next - Get the next object that needs serialization. | |
156 | * | |
157 | * Return values are: | |
158 | * | |
159 | * A pointer to the control struct of the next object that needs | |
160 | * serialization, if present in one of the percpu reorder queues. | |
161 | * | |
162 | * NULL, if all percpu reorder queues are empty. | |
163 | * | |
164 | * -EINPROGRESS, if the next object that needs serialization will | |
165 | * be parallel processed by another cpu and is not yet present in | |
166 | * the cpu's reorder queue. | |
167 | * | |
168 | * -ENODATA, if this cpu has to do the parallel processing for | |
169 | * the next object. | |
170 | */ | |
16295bec SK |
171 | static struct padata_priv *padata_get_next(struct parallel_data *pd) |
172 | { | |
5f1a8c1b SK |
173 | int cpu, num_cpus; |
174 | int next_nr, next_index; | |
16295bec SK |
175 | struct padata_queue *queue, *next_queue; |
176 | struct padata_priv *padata; | |
177 | struct padata_list *reorder; | |
178 | ||
16295bec SK |
179 | num_cpus = cpumask_weight(pd->cpumask); |
180 | ||
5f1a8c1b SK |
181 | /* |
182 | * Calculate the percpu reorder queue and the sequence | |
183 | * number of the next object. | |
184 | */ | |
185 | next_nr = pd->processed; | |
186 | next_index = next_nr % num_cpus; | |
187 | cpu = padata_index_to_cpu(pd, next_index); | |
188 | next_queue = per_cpu_ptr(pd->queue, cpu); | |
189 | ||
190 | if (unlikely(next_nr > pd->max_seq_nr)) { | |
191 | next_nr = next_nr - pd->max_seq_nr - 1; | |
192 | next_index = next_nr % num_cpus; | |
193 | cpu = padata_index_to_cpu(pd, next_index); | |
194 | next_queue = per_cpu_ptr(pd->queue, cpu); | |
195 | pd->processed = 0; | |
16295bec SK |
196 | } |
197 | ||
198 | padata = NULL; | |
199 | ||
16295bec SK |
200 | reorder = &next_queue->reorder; |
201 | ||
202 | if (!list_empty(&reorder->list)) { | |
203 | padata = list_entry(reorder->list.next, | |
204 | struct padata_priv, list); | |
205 | ||
5f1a8c1b | 206 | BUG_ON(next_nr != padata->seq_nr); |
16295bec SK |
207 | |
208 | spin_lock(&reorder->lock); | |
209 | list_del_init(&padata->list); | |
210 | atomic_dec(&pd->reorder_objects); | |
211 | spin_unlock(&reorder->lock); | |
212 | ||
5f1a8c1b | 213 | pd->processed++; |
16295bec SK |
214 | |
215 | goto out; | |
216 | } | |
217 | ||
d46a5ac7 SK |
218 | queue = per_cpu_ptr(pd->queue, smp_processor_id()); |
219 | if (queue->cpu_index == next_queue->cpu_index) { | |
16295bec SK |
220 | padata = ERR_PTR(-ENODATA); |
221 | goto out; | |
222 | } | |
223 | ||
224 | padata = ERR_PTR(-EINPROGRESS); | |
225 | out: | |
226 | return padata; | |
227 | } | |
228 | ||
229 | static void padata_reorder(struct parallel_data *pd) | |
230 | { | |
231 | struct padata_priv *padata; | |
232 | struct padata_queue *queue; | |
233 | struct padata_instance *pinst = pd->pinst; | |
234 | ||
0198ffd1 SK |
235 | /* |
236 | * We need to ensure that only one cpu can work on dequeueing of | |
237 | * the reorder queue the time. Calculating in which percpu reorder | |
238 | * queue the next object will arrive takes some time. A spinlock | |
239 | * would be highly contended. Also it is not clear in which order | |
240 | * the objects arrive to the reorder queues. So a cpu could wait to | |
241 | * get the lock just to notice that there is nothing to do at the | |
242 | * moment. Therefore we use a trylock and let the holder of the lock | |
243 | * care for all the objects enqueued during the holdtime of the lock. | |
244 | */ | |
16295bec | 245 | if (!spin_trylock_bh(&pd->lock)) |
d46a5ac7 | 246 | return; |
16295bec SK |
247 | |
248 | while (1) { | |
249 | padata = padata_get_next(pd); | |
250 | ||
0198ffd1 SK |
251 | /* |
252 | * All reorder queues are empty, or the next object that needs | |
253 | * serialization is parallel processed by another cpu and is | |
254 | * still on it's way to the cpu's reorder queue, nothing to | |
255 | * do for now. | |
256 | */ | |
16295bec SK |
257 | if (!padata || PTR_ERR(padata) == -EINPROGRESS) |
258 | break; | |
259 | ||
0198ffd1 SK |
260 | /* |
261 | * This cpu has to do the parallel processing of the next | |
262 | * object. It's waiting in the cpu's parallelization queue, | |
263 | * so exit imediately. | |
264 | */ | |
16295bec | 265 | if (PTR_ERR(padata) == -ENODATA) { |
d46a5ac7 | 266 | del_timer(&pd->timer); |
16295bec | 267 | spin_unlock_bh(&pd->lock); |
d46a5ac7 | 268 | return; |
16295bec SK |
269 | } |
270 | ||
271 | queue = per_cpu_ptr(pd->queue, padata->cb_cpu); | |
272 | ||
273 | spin_lock(&queue->serial.lock); | |
274 | list_add_tail(&padata->list, &queue->serial.list); | |
275 | spin_unlock(&queue->serial.lock); | |
276 | ||
277 | queue_work_on(padata->cb_cpu, pinst->wq, &queue->swork); | |
278 | } | |
279 | ||
280 | spin_unlock_bh(&pd->lock); | |
281 | ||
0198ffd1 SK |
282 | /* |
283 | * The next object that needs serialization might have arrived to | |
284 | * the reorder queues in the meantime, we will be called again | |
285 | * from the timer function if noone else cares for it. | |
286 | */ | |
d46a5ac7 SK |
287 | if (atomic_read(&pd->reorder_objects) |
288 | && !(pinst->flags & PADATA_RESET)) | |
289 | mod_timer(&pd->timer, jiffies + HZ); | |
290 | else | |
291 | del_timer(&pd->timer); | |
16295bec | 292 | |
16295bec SK |
293 | return; |
294 | } | |
295 | ||
d46a5ac7 SK |
296 | static void padata_reorder_timer(unsigned long arg) |
297 | { | |
298 | struct parallel_data *pd = (struct parallel_data *)arg; | |
299 | ||
300 | padata_reorder(pd); | |
301 | } | |
302 | ||
16295bec SK |
303 | static void padata_serial_worker(struct work_struct *work) |
304 | { | |
305 | struct padata_queue *queue; | |
306 | struct parallel_data *pd; | |
307 | LIST_HEAD(local_list); | |
308 | ||
309 | local_bh_disable(); | |
310 | queue = container_of(work, struct padata_queue, swork); | |
311 | pd = queue->pd; | |
312 | ||
313 | spin_lock(&queue->serial.lock); | |
314 | list_replace_init(&queue->serial.list, &local_list); | |
315 | spin_unlock(&queue->serial.lock); | |
316 | ||
317 | while (!list_empty(&local_list)) { | |
318 | struct padata_priv *padata; | |
319 | ||
320 | padata = list_entry(local_list.next, | |
321 | struct padata_priv, list); | |
322 | ||
323 | list_del_init(&padata->list); | |
324 | ||
325 | padata->serial(padata); | |
326 | atomic_dec(&pd->refcnt); | |
327 | } | |
328 | local_bh_enable(); | |
329 | } | |
330 | ||
0198ffd1 | 331 | /** |
16295bec SK |
332 | * padata_do_serial - padata serialization function |
333 | * | |
334 | * @padata: object to be serialized. | |
335 | * | |
336 | * padata_do_serial must be called for every parallelized object. | |
337 | * The serialization callback function will run with BHs off. | |
338 | */ | |
339 | void padata_do_serial(struct padata_priv *padata) | |
340 | { | |
341 | int cpu; | |
342 | struct padata_queue *queue; | |
343 | struct parallel_data *pd; | |
344 | ||
345 | pd = padata->pd; | |
346 | ||
347 | cpu = get_cpu(); | |
348 | queue = per_cpu_ptr(pd->queue, cpu); | |
349 | ||
350 | spin_lock(&queue->reorder.lock); | |
351 | atomic_inc(&pd->reorder_objects); | |
352 | list_add_tail(&padata->list, &queue->reorder.list); | |
353 | spin_unlock(&queue->reorder.lock); | |
354 | ||
355 | put_cpu(); | |
356 | ||
357 | padata_reorder(pd); | |
358 | } | |
359 | EXPORT_SYMBOL(padata_do_serial); | |
360 | ||
0198ffd1 | 361 | /* Allocate and initialize the internal cpumask dependend resources. */ |
16295bec SK |
362 | static struct parallel_data *padata_alloc_pd(struct padata_instance *pinst, |
363 | const struct cpumask *cpumask) | |
364 | { | |
365 | int cpu, cpu_index, num_cpus; | |
366 | struct padata_queue *queue; | |
367 | struct parallel_data *pd; | |
368 | ||
369 | cpu_index = 0; | |
370 | ||
371 | pd = kzalloc(sizeof(struct parallel_data), GFP_KERNEL); | |
372 | if (!pd) | |
373 | goto err; | |
374 | ||
375 | pd->queue = alloc_percpu(struct padata_queue); | |
376 | if (!pd->queue) | |
377 | goto err_free_pd; | |
378 | ||
379 | if (!alloc_cpumask_var(&pd->cpumask, GFP_KERNEL)) | |
380 | goto err_free_queue; | |
381 | ||
7b389b2c SK |
382 | cpumask_and(pd->cpumask, cpumask, cpu_active_mask); |
383 | ||
384 | for_each_cpu(cpu, pd->cpumask) { | |
16295bec SK |
385 | queue = per_cpu_ptr(pd->queue, cpu); |
386 | ||
387 | queue->pd = pd; | |
388 | ||
7b389b2c SK |
389 | queue->cpu_index = cpu_index; |
390 | cpu_index++; | |
16295bec SK |
391 | |
392 | INIT_LIST_HEAD(&queue->reorder.list); | |
393 | INIT_LIST_HEAD(&queue->parallel.list); | |
394 | INIT_LIST_HEAD(&queue->serial.list); | |
395 | spin_lock_init(&queue->reorder.lock); | |
396 | spin_lock_init(&queue->parallel.lock); | |
397 | spin_lock_init(&queue->serial.lock); | |
398 | ||
399 | INIT_WORK(&queue->pwork, padata_parallel_worker); | |
400 | INIT_WORK(&queue->swork, padata_serial_worker); | |
16295bec SK |
401 | } |
402 | ||
16295bec SK |
403 | num_cpus = cpumask_weight(pd->cpumask); |
404 | pd->max_seq_nr = (MAX_SEQ_NR / num_cpus) * num_cpus - 1; | |
405 | ||
d46a5ac7 | 406 | setup_timer(&pd->timer, padata_reorder_timer, (unsigned long)pd); |
16295bec SK |
407 | atomic_set(&pd->seq_nr, -1); |
408 | atomic_set(&pd->reorder_objects, 0); | |
409 | atomic_set(&pd->refcnt, 0); | |
410 | pd->pinst = pinst; | |
411 | spin_lock_init(&pd->lock); | |
412 | ||
413 | return pd; | |
414 | ||
415 | err_free_queue: | |
416 | free_percpu(pd->queue); | |
417 | err_free_pd: | |
418 | kfree(pd); | |
419 | err: | |
420 | return NULL; | |
421 | } | |
422 | ||
423 | static void padata_free_pd(struct parallel_data *pd) | |
424 | { | |
425 | free_cpumask_var(pd->cpumask); | |
426 | free_percpu(pd->queue); | |
427 | kfree(pd); | |
428 | } | |
429 | ||
0198ffd1 | 430 | /* Flush all objects out of the padata queues. */ |
2b73b07a SK |
431 | static void padata_flush_queues(struct parallel_data *pd) |
432 | { | |
433 | int cpu; | |
434 | struct padata_queue *queue; | |
435 | ||
436 | for_each_cpu(cpu, pd->cpumask) { | |
437 | queue = per_cpu_ptr(pd->queue, cpu); | |
438 | flush_work(&queue->pwork); | |
439 | } | |
440 | ||
441 | del_timer_sync(&pd->timer); | |
442 | ||
443 | if (atomic_read(&pd->reorder_objects)) | |
444 | padata_reorder(pd); | |
445 | ||
446 | for_each_cpu(cpu, pd->cpumask) { | |
447 | queue = per_cpu_ptr(pd->queue, cpu); | |
448 | flush_work(&queue->swork); | |
449 | } | |
450 | ||
451 | BUG_ON(atomic_read(&pd->refcnt) != 0); | |
452 | } | |
453 | ||
4c879170 SK |
454 | static void __padata_start(struct padata_instance *pinst) |
455 | { | |
456 | pinst->flags |= PADATA_INIT; | |
457 | } | |
458 | ||
ee836555 SK |
459 | static void __padata_stop(struct padata_instance *pinst) |
460 | { | |
461 | if (!(pinst->flags & PADATA_INIT)) | |
462 | return; | |
463 | ||
464 | pinst->flags &= ~PADATA_INIT; | |
465 | ||
466 | synchronize_rcu(); | |
467 | ||
468 | get_online_cpus(); | |
469 | padata_flush_queues(pinst->pd); | |
470 | put_online_cpus(); | |
471 | } | |
472 | ||
0198ffd1 | 473 | /* Replace the internal control stucture with a new one. */ |
16295bec SK |
474 | static void padata_replace(struct padata_instance *pinst, |
475 | struct parallel_data *pd_new) | |
476 | { | |
477 | struct parallel_data *pd_old = pinst->pd; | |
478 | ||
479 | pinst->flags |= PADATA_RESET; | |
480 | ||
481 | rcu_assign_pointer(pinst->pd, pd_new); | |
482 | ||
483 | synchronize_rcu(); | |
484 | ||
33e54450 SK |
485 | if (pd_old) { |
486 | padata_flush_queues(pd_old); | |
487 | padata_free_pd(pd_old); | |
488 | } | |
16295bec SK |
489 | |
490 | pinst->flags &= ~PADATA_RESET; | |
491 | } | |
492 | ||
33e54450 SK |
493 | /* If cpumask contains no active cpu, we mark the instance as invalid. */ |
494 | static bool padata_validate_cpumask(struct padata_instance *pinst, | |
495 | const struct cpumask *cpumask) | |
496 | { | |
497 | if (!cpumask_intersects(cpumask, cpu_active_mask)) { | |
498 | pinst->flags |= PADATA_INVALID; | |
499 | return false; | |
500 | } | |
501 | ||
502 | pinst->flags &= ~PADATA_INVALID; | |
503 | return true; | |
504 | } | |
505 | ||
0198ffd1 | 506 | /** |
16295bec SK |
507 | * padata_set_cpumask - set the cpumask that padata should use |
508 | * | |
509 | * @pinst: padata instance | |
510 | * @cpumask: the cpumask to use | |
511 | */ | |
512 | int padata_set_cpumask(struct padata_instance *pinst, | |
513 | cpumask_var_t cpumask) | |
514 | { | |
33e54450 | 515 | int valid; |
16295bec | 516 | int err = 0; |
33e54450 | 517 | struct parallel_data *pd = NULL; |
16295bec | 518 | |
16295bec SK |
519 | mutex_lock(&pinst->lock); |
520 | ||
33e54450 SK |
521 | valid = padata_validate_cpumask(pinst, cpumask); |
522 | if (!valid) { | |
523 | __padata_stop(pinst); | |
524 | goto out_replace; | |
525 | } | |
526 | ||
6751fb3c SK |
527 | get_online_cpus(); |
528 | ||
16295bec SK |
529 | pd = padata_alloc_pd(pinst, cpumask); |
530 | if (!pd) { | |
531 | err = -ENOMEM; | |
532 | goto out; | |
533 | } | |
534 | ||
33e54450 | 535 | out_replace: |
16295bec SK |
536 | cpumask_copy(pinst->cpumask, cpumask); |
537 | ||
538 | padata_replace(pinst, pd); | |
539 | ||
33e54450 SK |
540 | if (valid) |
541 | __padata_start(pinst); | |
542 | ||
16295bec | 543 | out: |
6751fb3c SK |
544 | put_online_cpus(); |
545 | ||
16295bec SK |
546 | mutex_unlock(&pinst->lock); |
547 | ||
548 | return err; | |
549 | } | |
550 | EXPORT_SYMBOL(padata_set_cpumask); | |
551 | ||
552 | static int __padata_add_cpu(struct padata_instance *pinst, int cpu) | |
553 | { | |
554 | struct parallel_data *pd; | |
555 | ||
556 | if (cpumask_test_cpu(cpu, cpu_active_mask)) { | |
557 | pd = padata_alloc_pd(pinst, pinst->cpumask); | |
558 | if (!pd) | |
559 | return -ENOMEM; | |
560 | ||
561 | padata_replace(pinst, pd); | |
33e54450 SK |
562 | |
563 | if (padata_validate_cpumask(pinst, pinst->cpumask)) | |
564 | __padata_start(pinst); | |
16295bec SK |
565 | } |
566 | ||
567 | return 0; | |
568 | } | |
569 | ||
0198ffd1 | 570 | /** |
16295bec SK |
571 | * padata_add_cpu - add a cpu to the padata cpumask |
572 | * | |
573 | * @pinst: padata instance | |
574 | * @cpu: cpu to add | |
575 | */ | |
576 | int padata_add_cpu(struct padata_instance *pinst, int cpu) | |
577 | { | |
578 | int err; | |
579 | ||
16295bec SK |
580 | mutex_lock(&pinst->lock); |
581 | ||
6751fb3c | 582 | get_online_cpus(); |
16295bec SK |
583 | cpumask_set_cpu(cpu, pinst->cpumask); |
584 | err = __padata_add_cpu(pinst, cpu); | |
6751fb3c | 585 | put_online_cpus(); |
16295bec SK |
586 | |
587 | mutex_unlock(&pinst->lock); | |
588 | ||
589 | return err; | |
590 | } | |
591 | EXPORT_SYMBOL(padata_add_cpu); | |
592 | ||
593 | static int __padata_remove_cpu(struct padata_instance *pinst, int cpu) | |
594 | { | |
33e54450 | 595 | struct parallel_data *pd = NULL; |
16295bec SK |
596 | |
597 | if (cpumask_test_cpu(cpu, cpu_online_mask)) { | |
33e54450 SK |
598 | |
599 | if (!padata_validate_cpumask(pinst, pinst->cpumask)) { | |
600 | __padata_stop(pinst); | |
601 | padata_replace(pinst, pd); | |
602 | goto out; | |
603 | } | |
604 | ||
16295bec SK |
605 | pd = padata_alloc_pd(pinst, pinst->cpumask); |
606 | if (!pd) | |
607 | return -ENOMEM; | |
608 | ||
609 | padata_replace(pinst, pd); | |
610 | } | |
611 | ||
33e54450 | 612 | out: |
16295bec SK |
613 | return 0; |
614 | } | |
615 | ||
0198ffd1 | 616 | /** |
16295bec SK |
617 | * padata_remove_cpu - remove a cpu from the padata cpumask |
618 | * | |
619 | * @pinst: padata instance | |
620 | * @cpu: cpu to remove | |
621 | */ | |
622 | int padata_remove_cpu(struct padata_instance *pinst, int cpu) | |
623 | { | |
624 | int err; | |
625 | ||
16295bec SK |
626 | mutex_lock(&pinst->lock); |
627 | ||
6751fb3c | 628 | get_online_cpus(); |
16295bec SK |
629 | cpumask_clear_cpu(cpu, pinst->cpumask); |
630 | err = __padata_remove_cpu(pinst, cpu); | |
6751fb3c | 631 | put_online_cpus(); |
16295bec SK |
632 | |
633 | mutex_unlock(&pinst->lock); | |
634 | ||
635 | return err; | |
636 | } | |
637 | EXPORT_SYMBOL(padata_remove_cpu); | |
638 | ||
0198ffd1 | 639 | /** |
16295bec SK |
640 | * padata_start - start the parallel processing |
641 | * | |
642 | * @pinst: padata instance to start | |
643 | */ | |
4c879170 | 644 | int padata_start(struct padata_instance *pinst) |
16295bec | 645 | { |
4c879170 SK |
646 | int err = 0; |
647 | ||
16295bec | 648 | mutex_lock(&pinst->lock); |
4c879170 SK |
649 | |
650 | if (pinst->flags & PADATA_INVALID) | |
651 | err =-EINVAL; | |
652 | ||
653 | __padata_start(pinst); | |
654 | ||
16295bec | 655 | mutex_unlock(&pinst->lock); |
4c879170 SK |
656 | |
657 | return err; | |
16295bec SK |
658 | } |
659 | EXPORT_SYMBOL(padata_start); | |
660 | ||
0198ffd1 | 661 | /** |
16295bec SK |
662 | * padata_stop - stop the parallel processing |
663 | * | |
664 | * @pinst: padata instance to stop | |
665 | */ | |
666 | void padata_stop(struct padata_instance *pinst) | |
667 | { | |
16295bec | 668 | mutex_lock(&pinst->lock); |
ee836555 | 669 | __padata_stop(pinst); |
16295bec SK |
670 | mutex_unlock(&pinst->lock); |
671 | } | |
672 | EXPORT_SYMBOL(padata_stop); | |
673 | ||
e2cb2f1c | 674 | #ifdef CONFIG_HOTPLUG_CPU |
975d2603 HK |
675 | static int padata_cpu_callback(struct notifier_block *nfb, |
676 | unsigned long action, void *hcpu) | |
16295bec SK |
677 | { |
678 | int err; | |
679 | struct padata_instance *pinst; | |
680 | int cpu = (unsigned long)hcpu; | |
681 | ||
682 | pinst = container_of(nfb, struct padata_instance, cpu_notifier); | |
683 | ||
684 | switch (action) { | |
685 | case CPU_ONLINE: | |
686 | case CPU_ONLINE_FROZEN: | |
687 | if (!cpumask_test_cpu(cpu, pinst->cpumask)) | |
688 | break; | |
689 | mutex_lock(&pinst->lock); | |
690 | err = __padata_add_cpu(pinst, cpu); | |
691 | mutex_unlock(&pinst->lock); | |
692 | if (err) | |
693 | return NOTIFY_BAD; | |
694 | break; | |
695 | ||
696 | case CPU_DOWN_PREPARE: | |
697 | case CPU_DOWN_PREPARE_FROZEN: | |
698 | if (!cpumask_test_cpu(cpu, pinst->cpumask)) | |
699 | break; | |
700 | mutex_lock(&pinst->lock); | |
701 | err = __padata_remove_cpu(pinst, cpu); | |
702 | mutex_unlock(&pinst->lock); | |
703 | if (err) | |
704 | return NOTIFY_BAD; | |
705 | break; | |
706 | ||
707 | case CPU_UP_CANCELED: | |
708 | case CPU_UP_CANCELED_FROZEN: | |
709 | if (!cpumask_test_cpu(cpu, pinst->cpumask)) | |
710 | break; | |
711 | mutex_lock(&pinst->lock); | |
712 | __padata_remove_cpu(pinst, cpu); | |
713 | mutex_unlock(&pinst->lock); | |
714 | ||
715 | case CPU_DOWN_FAILED: | |
716 | case CPU_DOWN_FAILED_FROZEN: | |
717 | if (!cpumask_test_cpu(cpu, pinst->cpumask)) | |
718 | break; | |
719 | mutex_lock(&pinst->lock); | |
720 | __padata_add_cpu(pinst, cpu); | |
721 | mutex_unlock(&pinst->lock); | |
722 | } | |
723 | ||
724 | return NOTIFY_OK; | |
725 | } | |
e2cb2f1c | 726 | #endif |
16295bec | 727 | |
0198ffd1 | 728 | /** |
16295bec SK |
729 | * padata_alloc - allocate and initialize a padata instance |
730 | * | |
731 | * @cpumask: cpumask that padata uses for parallelization | |
732 | * @wq: workqueue to use for the allocated padata instance | |
733 | */ | |
734 | struct padata_instance *padata_alloc(const struct cpumask *cpumask, | |
735 | struct workqueue_struct *wq) | |
736 | { | |
16295bec | 737 | struct padata_instance *pinst; |
33e54450 | 738 | struct parallel_data *pd = NULL; |
16295bec SK |
739 | |
740 | pinst = kzalloc(sizeof(struct padata_instance), GFP_KERNEL); | |
741 | if (!pinst) | |
742 | goto err; | |
743 | ||
6751fb3c SK |
744 | get_online_cpus(); |
745 | ||
33e54450 | 746 | if (!alloc_cpumask_var(&pinst->cpumask, GFP_KERNEL)) |
16295bec SK |
747 | goto err_free_inst; |
748 | ||
33e54450 SK |
749 | if (padata_validate_cpumask(pinst, cpumask)) { |
750 | pd = padata_alloc_pd(pinst, cpumask); | |
751 | if (!pd) | |
752 | goto err_free_mask; | |
753 | } | |
74781387 | 754 | |
16295bec SK |
755 | rcu_assign_pointer(pinst->pd, pd); |
756 | ||
757 | pinst->wq = wq; | |
758 | ||
759 | cpumask_copy(pinst->cpumask, cpumask); | |
760 | ||
761 | pinst->flags = 0; | |
762 | ||
e2cb2f1c | 763 | #ifdef CONFIG_HOTPLUG_CPU |
16295bec SK |
764 | pinst->cpu_notifier.notifier_call = padata_cpu_callback; |
765 | pinst->cpu_notifier.priority = 0; | |
e2cb2f1c SK |
766 | register_hotcpu_notifier(&pinst->cpu_notifier); |
767 | #endif | |
16295bec | 768 | |
6751fb3c SK |
769 | put_online_cpus(); |
770 | ||
16295bec SK |
771 | mutex_init(&pinst->lock); |
772 | ||
773 | return pinst; | |
774 | ||
33e54450 SK |
775 | err_free_mask: |
776 | free_cpumask_var(pinst->cpumask); | |
16295bec SK |
777 | err_free_inst: |
778 | kfree(pinst); | |
6751fb3c | 779 | put_online_cpus(); |
16295bec SK |
780 | err: |
781 | return NULL; | |
782 | } | |
783 | EXPORT_SYMBOL(padata_alloc); | |
784 | ||
0198ffd1 | 785 | /** |
16295bec SK |
786 | * padata_free - free a padata instance |
787 | * | |
0198ffd1 | 788 | * @padata_inst: padata instance to free |
16295bec SK |
789 | */ |
790 | void padata_free(struct padata_instance *pinst) | |
791 | { | |
e2cb2f1c | 792 | #ifdef CONFIG_HOTPLUG_CPU |
16295bec | 793 | unregister_hotcpu_notifier(&pinst->cpu_notifier); |
e2cb2f1c | 794 | #endif |
3789ae7d | 795 | |
ee836555 | 796 | padata_stop(pinst); |
16295bec | 797 | padata_free_pd(pinst->pd); |
74781387 | 798 | free_cpumask_var(pinst->cpumask); |
16295bec SK |
799 | kfree(pinst); |
800 | } | |
801 | EXPORT_SYMBOL(padata_free); |