1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * padata.c - generic interface to process data streams in parallel
4 *
5 * See Documentation/core-api/padata.rst for more information.
6 *
7 * Copyright (C) 2008, 2009 secunet Security Networks AG
8 * Copyright (C) 2008, 2009 Steffen Klassert <steffen.klassert@secunet.com>
9 *
10 * Copyright (c) 2020 Oracle and/or its affiliates.
11 * Author: Daniel Jordan <daniel.m.jordan@oracle.com>
12 */
13
14 #include <linux/completion.h>
15 #include <linux/export.h>
16 #include <linux/cpumask.h>
17 #include <linux/err.h>
18 #include <linux/cpu.h>
19 #include <linux/padata.h>
20 #include <linux/mutex.h>
21 #include <linux/sched.h>
22 #include <linux/slab.h>
23 #include <linux/sysfs.h>
24 #include <linux/rcupdate.h>
25
26 #define PADATA_WORK_ONSTACK 1 /* Work's memory is on stack */
27
28 struct padata_work {
29 struct work_struct pw_work;
30 struct list_head pw_list; /* padata_free_works linkage */
31 void *pw_data;
32 };
33
34 static DEFINE_SPINLOCK(padata_works_lock);
35 static struct padata_work *padata_works;
36 static LIST_HEAD(padata_free_works);
37
38 struct padata_mt_job_state {
39 spinlock_t lock;
40 struct completion completion;
41 struct padata_mt_job *job;
42 int nworks;
43 int nworks_fini;
44 unsigned long chunk_size;
45 };
46
47 static void padata_free_pd(struct parallel_data *pd);
48 static void __init padata_mt_helper(struct work_struct *work);
49
padata_index_to_cpu(struct parallel_data * pd,int cpu_index)50 static int padata_index_to_cpu(struct parallel_data *pd, int cpu_index)
51 {
52 int cpu, target_cpu;
53
54 target_cpu = cpumask_first(pd->cpumask.pcpu);
55 for (cpu = 0; cpu < cpu_index; cpu++)
56 target_cpu = cpumask_next(target_cpu, pd->cpumask.pcpu);
57
58 return target_cpu;
59 }
60
padata_cpu_hash(struct parallel_data * pd,unsigned int seq_nr)61 static int padata_cpu_hash(struct parallel_data *pd, unsigned int seq_nr)
62 {
63 /*
64 * Hash the sequence numbers to the cpus by taking
65 * seq_nr mod. number of cpus in use.
66 */
67 int cpu_index = seq_nr % cpumask_weight(pd->cpumask.pcpu);
68
69 return padata_index_to_cpu(pd, cpu_index);
70 }
71
padata_work_alloc(void)72 static struct padata_work *padata_work_alloc(void)
73 {
74 struct padata_work *pw;
75
76 lockdep_assert_held(&padata_works_lock);
77
78 if (list_empty(&padata_free_works))
79 return NULL; /* No more work items allowed to be queued. */
80
81 pw = list_first_entry(&padata_free_works, struct padata_work, pw_list);
82 list_del(&pw->pw_list);
83 return pw;
84 }
85
padata_work_init(struct padata_work * pw,work_func_t work_fn,void * data,int flags)86 static void padata_work_init(struct padata_work *pw, work_func_t work_fn,
87 void *data, int flags)
88 {
89 if (flags & PADATA_WORK_ONSTACK)
90 INIT_WORK_ONSTACK(&pw->pw_work, work_fn);
91 else
92 INIT_WORK(&pw->pw_work, work_fn);
93 pw->pw_data = data;
94 }
95
padata_work_alloc_mt(int nworks,void * data,struct list_head * head)96 static int __init padata_work_alloc_mt(int nworks, void *data,
97 struct list_head *head)
98 {
99 int i;
100
101 spin_lock(&padata_works_lock);
102 /* Start at 1 because the current task participates in the job. */
103 for (i = 1; i < nworks; ++i) {
104 struct padata_work *pw = padata_work_alloc();
105
106 if (!pw)
107 break;
108 padata_work_init(pw, padata_mt_helper, data, 0);
109 list_add(&pw->pw_list, head);
110 }
111 spin_unlock(&padata_works_lock);
112
113 return i;
114 }
115
padata_work_free(struct padata_work * pw)116 static void padata_work_free(struct padata_work *pw)
117 {
118 lockdep_assert_held(&padata_works_lock);
119 list_add(&pw->pw_list, &padata_free_works);
120 }
121
padata_works_free(struct list_head * works)122 static void __init padata_works_free(struct list_head *works)
123 {
124 struct padata_work *cur, *next;
125
126 if (list_empty(works))
127 return;
128
129 spin_lock(&padata_works_lock);
130 list_for_each_entry_safe(cur, next, works, pw_list) {
131 list_del(&cur->pw_list);
132 padata_work_free(cur);
133 }
134 spin_unlock(&padata_works_lock);
135 }
136
padata_parallel_worker(struct work_struct * parallel_work)137 static void padata_parallel_worker(struct work_struct *parallel_work)
138 {
139 struct padata_work *pw = container_of(parallel_work, struct padata_work,
140 pw_work);
141 struct padata_priv *padata = pw->pw_data;
142
143 local_bh_disable();
144 padata->parallel(padata);
145 spin_lock(&padata_works_lock);
146 padata_work_free(pw);
147 spin_unlock(&padata_works_lock);
148 local_bh_enable();
149 }
150
151 /**
152 * padata_do_parallel - padata parallelization function
153 *
154 * @ps: padatashell
155 * @padata: object to be parallelized
156 * @cb_cpu: pointer to the CPU that the serialization callback function should
157 * run on. If it's not in the serial cpumask of @pinst
158 * (i.e. cpumask.cbcpu), this function selects a fallback CPU and if
159 * none found, returns -EINVAL.
160 *
161 * The parallelization callback function will run with BHs off.
162 * Note: Every object which is parallelized by padata_do_parallel
163 * must be seen by padata_do_serial.
164 *
165 * Return: 0 on success or else negative error code.
166 */
padata_do_parallel(struct padata_shell * ps,struct padata_priv * padata,int * cb_cpu)167 int padata_do_parallel(struct padata_shell *ps,
168 struct padata_priv *padata, int *cb_cpu)
169 {
170 struct padata_instance *pinst = ps->pinst;
171 int i, cpu, cpu_index, err;
172 struct parallel_data *pd;
173 struct padata_work *pw;
174
175 rcu_read_lock_bh();
176
177 pd = rcu_dereference_bh(ps->pd);
178
179 err = -EINVAL;
180 if (!(pinst->flags & PADATA_INIT) || pinst->flags & PADATA_INVALID)
181 goto out;
182
183 if (!cpumask_test_cpu(*cb_cpu, pd->cpumask.cbcpu)) {
184 if (cpumask_empty(pd->cpumask.cbcpu))
185 goto out;
186
187 /* Select an alternate fallback CPU and notify the caller. */
188 cpu_index = *cb_cpu % cpumask_weight(pd->cpumask.cbcpu);
189
190 cpu = cpumask_first(pd->cpumask.cbcpu);
191 for (i = 0; i < cpu_index; i++)
192 cpu = cpumask_next(cpu, pd->cpumask.cbcpu);
193
194 *cb_cpu = cpu;
195 }
196
197 err = -EBUSY;
198 if ((pinst->flags & PADATA_RESET))
199 goto out;
200
201 refcount_inc(&pd->refcnt);
202 padata->pd = pd;
203 padata->cb_cpu = *cb_cpu;
204
205 spin_lock(&padata_works_lock);
206 padata->seq_nr = ++pd->seq_nr;
207 pw = padata_work_alloc();
208 spin_unlock(&padata_works_lock);
209
210 rcu_read_unlock_bh();
211
212 if (pw) {
213 padata_work_init(pw, padata_parallel_worker, padata, 0);
214 queue_work(pinst->parallel_wq, &pw->pw_work);
215 } else {
216 /* Maximum works limit exceeded, run in the current task. */
217 padata->parallel(padata);
218 }
219
220 return 0;
221 out:
222 rcu_read_unlock_bh();
223
224 return err;
225 }
226 EXPORT_SYMBOL(padata_do_parallel);
227
228 /*
229 * padata_find_next - Find the next object that needs serialization.
230 *
231 * Return:
232 * * A pointer to the control struct of the next object that needs
233 * serialization, if present in one of the percpu reorder queues.
234 * * NULL, if the next object that needs serialization will
235 * be parallel processed by another cpu and is not yet present in
236 * the cpu's reorder queue.
237 */
padata_find_next(struct parallel_data * pd,bool remove_object)238 static struct padata_priv *padata_find_next(struct parallel_data *pd,
239 bool remove_object)
240 {
241 struct padata_priv *padata;
242 struct padata_list *reorder;
243 int cpu = pd->cpu;
244
245 reorder = per_cpu_ptr(pd->reorder_list, cpu);
246
247 spin_lock(&reorder->lock);
248 if (list_empty(&reorder->list)) {
249 spin_unlock(&reorder->lock);
250 return NULL;
251 }
252
253 padata = list_entry(reorder->list.next, struct padata_priv, list);
254
255 /*
256 * Checks the rare case where two or more parallel jobs have hashed to
257 * the same CPU and one of the later ones finishes first.
258 */
259 if (padata->seq_nr != pd->processed) {
260 spin_unlock(&reorder->lock);
261 return NULL;
262 }
263
264 if (remove_object) {
265 list_del_init(&padata->list);
266 ++pd->processed;
267 pd->cpu = cpumask_next_wrap(cpu, pd->cpumask.pcpu, -1, false);
268 }
269
270 spin_unlock(&reorder->lock);
271 return padata;
272 }
273
padata_reorder(struct parallel_data * pd)274 static void padata_reorder(struct parallel_data *pd)
275 {
276 struct padata_instance *pinst = pd->ps->pinst;
277 int cb_cpu;
278 struct padata_priv *padata;
279 struct padata_serial_queue *squeue;
280 struct padata_list *reorder;
281
282 /*
283 * We need to ensure that only one cpu can work on dequeueing of
284 * the reorder queue the time. Calculating in which percpu reorder
285 * queue the next object will arrive takes some time. A spinlock
286 * would be highly contended. Also it is not clear in which order
287 * the objects arrive to the reorder queues. So a cpu could wait to
288 * get the lock just to notice that there is nothing to do at the
289 * moment. Therefore we use a trylock and let the holder of the lock
290 * care for all the objects enqueued during the holdtime of the lock.
291 */
292 if (!spin_trylock_bh(&pd->lock))
293 return;
294
295 while (1) {
296 padata = padata_find_next(pd, true);
297
298 /*
299 * If the next object that needs serialization is parallel
300 * processed by another cpu and is still on it's way to the
301 * cpu's reorder queue, nothing to do for now.
302 */
303 if (!padata)
304 break;
305
306 cb_cpu = padata->cb_cpu;
307 squeue = per_cpu_ptr(pd->squeue, cb_cpu);
308
309 spin_lock(&squeue->serial.lock);
310 list_add_tail(&padata->list, &squeue->serial.list);
311 spin_unlock(&squeue->serial.lock);
312
313 queue_work_on(cb_cpu, pinst->serial_wq, &squeue->work);
314 }
315
316 spin_unlock_bh(&pd->lock);
317
318 /*
319 * The next object that needs serialization might have arrived to
320 * the reorder queues in the meantime.
321 *
322 * Ensure reorder queue is read after pd->lock is dropped so we see
323 * new objects from another task in padata_do_serial. Pairs with
324 * smp_mb in padata_do_serial.
325 */
326 smp_mb();
327
328 reorder = per_cpu_ptr(pd->reorder_list, pd->cpu);
329 if (!list_empty(&reorder->list) && padata_find_next(pd, false))
330 queue_work(pinst->serial_wq, &pd->reorder_work);
331 }
332
invoke_padata_reorder(struct work_struct * work)333 static void invoke_padata_reorder(struct work_struct *work)
334 {
335 struct parallel_data *pd;
336
337 local_bh_disable();
338 pd = container_of(work, struct parallel_data, reorder_work);
339 padata_reorder(pd);
340 local_bh_enable();
341 }
342
padata_serial_worker(struct work_struct * serial_work)343 static void padata_serial_worker(struct work_struct *serial_work)
344 {
345 struct padata_serial_queue *squeue;
346 struct parallel_data *pd;
347 LIST_HEAD(local_list);
348 int cnt;
349
350 local_bh_disable();
351 squeue = container_of(serial_work, struct padata_serial_queue, work);
352 pd = squeue->pd;
353
354 spin_lock(&squeue->serial.lock);
355 list_replace_init(&squeue->serial.list, &local_list);
356 spin_unlock(&squeue->serial.lock);
357
358 cnt = 0;
359
360 while (!list_empty(&local_list)) {
361 struct padata_priv *padata;
362
363 padata = list_entry(local_list.next,
364 struct padata_priv, list);
365
366 list_del_init(&padata->list);
367
368 padata->serial(padata);
369 cnt++;
370 }
371 local_bh_enable();
372
373 if (refcount_sub_and_test(cnt, &pd->refcnt))
374 padata_free_pd(pd);
375 }
376
377 /**
378 * padata_do_serial - padata serialization function
379 *
380 * @padata: object to be serialized.
381 *
382 * padata_do_serial must be called for every parallelized object.
383 * The serialization callback function will run with BHs off.
384 */
padata_do_serial(struct padata_priv * padata)385 void padata_do_serial(struct padata_priv *padata)
386 {
387 struct parallel_data *pd = padata->pd;
388 int hashed_cpu = padata_cpu_hash(pd, padata->seq_nr);
389 struct padata_list *reorder = per_cpu_ptr(pd->reorder_list, hashed_cpu);
390 struct padata_priv *cur;
391
392 spin_lock(&reorder->lock);
393 /* Sort in ascending order of sequence number. */
394 list_for_each_entry_reverse(cur, &reorder->list, list)
395 if (cur->seq_nr < padata->seq_nr)
396 break;
397 list_add(&padata->list, &cur->list);
398 spin_unlock(&reorder->lock);
399
400 /*
401 * Ensure the addition to the reorder list is ordered correctly
402 * with the trylock of pd->lock in padata_reorder. Pairs with smp_mb
403 * in padata_reorder.
404 */
405 smp_mb();
406
407 padata_reorder(pd);
408 }
409 EXPORT_SYMBOL(padata_do_serial);
410
padata_setup_cpumasks(struct padata_instance * pinst)411 static int padata_setup_cpumasks(struct padata_instance *pinst)
412 {
413 struct workqueue_attrs *attrs;
414 int err;
415
416 attrs = alloc_workqueue_attrs();
417 if (!attrs)
418 return -ENOMEM;
419
420 /* Restrict parallel_wq workers to pd->cpumask.pcpu. */
421 cpumask_copy(attrs->cpumask, pinst->cpumask.pcpu);
422 err = apply_workqueue_attrs(pinst->parallel_wq, attrs);
423 free_workqueue_attrs(attrs);
424
425 return err;
426 }
427
padata_mt_helper(struct work_struct * w)428 static void __init padata_mt_helper(struct work_struct *w)
429 {
430 struct padata_work *pw = container_of(w, struct padata_work, pw_work);
431 struct padata_mt_job_state *ps = pw->pw_data;
432 struct padata_mt_job *job = ps->job;
433 bool done;
434
435 spin_lock(&ps->lock);
436
437 while (job->size > 0) {
438 unsigned long start, size, end;
439
440 start = job->start;
441 /* So end is chunk size aligned if enough work remains. */
442 size = roundup(start + 1, ps->chunk_size) - start;
443 size = min(size, job->size);
444 end = start + size;
445
446 job->start = end;
447 job->size -= size;
448
449 spin_unlock(&ps->lock);
450 job->thread_fn(start, end, job->fn_arg);
451 spin_lock(&ps->lock);
452 }
453
454 ++ps->nworks_fini;
455 done = (ps->nworks_fini == ps->nworks);
456 spin_unlock(&ps->lock);
457
458 if (done)
459 complete(&ps->completion);
460 }
461
462 /**
463 * padata_do_multithreaded - run a multithreaded job
464 * @job: Description of the job.
465 *
466 * See the definition of struct padata_mt_job for more details.
467 */
padata_do_multithreaded(struct padata_mt_job * job)468 void __init padata_do_multithreaded(struct padata_mt_job *job)
469 {
470 /* In case threads finish at different times. */
471 static const unsigned long load_balance_factor = 4;
472 struct padata_work my_work, *pw;
473 struct padata_mt_job_state ps;
474 LIST_HEAD(works);
475 int nworks;
476
477 if (job->size == 0)
478 return;
479
480 /* Ensure at least one thread when size < min_chunk. */
481 nworks = max(job->size / job->min_chunk, 1ul);
482 nworks = min(nworks, job->max_threads);
483
484 if (nworks == 1) {
485 /* Single thread, no coordination needed, cut to the chase. */
486 job->thread_fn(job->start, job->start + job->size, job->fn_arg);
487 return;
488 }
489
490 spin_lock_init(&ps.lock);
491 init_completion(&ps.completion);
492 ps.job = job;
493 ps.nworks = padata_work_alloc_mt(nworks, &ps, &works);
494 ps.nworks_fini = 0;
495
496 /*
497 * Chunk size is the amount of work a helper does per call to the
498 * thread function. Load balance large jobs between threads by
499 * increasing the number of chunks, guarantee at least the minimum
500 * chunk size from the caller, and honor the caller's alignment.
501 */
502 ps.chunk_size = job->size / (ps.nworks * load_balance_factor);
503 ps.chunk_size = max(ps.chunk_size, job->min_chunk);
504 ps.chunk_size = roundup(ps.chunk_size, job->align);
505
506 list_for_each_entry(pw, &works, pw_list)
507 queue_work(system_unbound_wq, &pw->pw_work);
508
509 /* Use the current thread, which saves starting a workqueue worker. */
510 padata_work_init(&my_work, padata_mt_helper, &ps, PADATA_WORK_ONSTACK);
511 padata_mt_helper(&my_work.pw_work);
512
513 /* Wait for all the helpers to finish. */
514 wait_for_completion(&ps.completion);
515
516 destroy_work_on_stack(&my_work.pw_work);
517 padata_works_free(&works);
518 }
519
__padata_list_init(struct padata_list * pd_list)520 static void __padata_list_init(struct padata_list *pd_list)
521 {
522 INIT_LIST_HEAD(&pd_list->list);
523 spin_lock_init(&pd_list->lock);
524 }
525
526 /* Initialize all percpu queues used by serial workers */
padata_init_squeues(struct parallel_data * pd)527 static void padata_init_squeues(struct parallel_data *pd)
528 {
529 int cpu;
530 struct padata_serial_queue *squeue;
531
532 for_each_cpu(cpu, pd->cpumask.cbcpu) {
533 squeue = per_cpu_ptr(pd->squeue, cpu);
534 squeue->pd = pd;
535 __padata_list_init(&squeue->serial);
536 INIT_WORK(&squeue->work, padata_serial_worker);
537 }
538 }
539
540 /* Initialize per-CPU reorder lists */
padata_init_reorder_list(struct parallel_data * pd)541 static void padata_init_reorder_list(struct parallel_data *pd)
542 {
543 int cpu;
544 struct padata_list *list;
545
546 for_each_cpu(cpu, pd->cpumask.pcpu) {
547 list = per_cpu_ptr(pd->reorder_list, cpu);
548 __padata_list_init(list);
549 }
550 }
551
552 /* Allocate and initialize the internal cpumask dependend resources. */
padata_alloc_pd(struct padata_shell * ps)553 static struct parallel_data *padata_alloc_pd(struct padata_shell *ps)
554 {
555 struct padata_instance *pinst = ps->pinst;
556 struct parallel_data *pd;
557
558 pd = kzalloc(sizeof(struct parallel_data), GFP_KERNEL);
559 if (!pd)
560 goto err;
561
562 pd->reorder_list = alloc_percpu(struct padata_list);
563 if (!pd->reorder_list)
564 goto err_free_pd;
565
566 pd->squeue = alloc_percpu(struct padata_serial_queue);
567 if (!pd->squeue)
568 goto err_free_reorder_list;
569
570 pd->ps = ps;
571
572 if (!alloc_cpumask_var(&pd->cpumask.pcpu, GFP_KERNEL))
573 goto err_free_squeue;
574 if (!alloc_cpumask_var(&pd->cpumask.cbcpu, GFP_KERNEL))
575 goto err_free_pcpu;
576
577 cpumask_and(pd->cpumask.pcpu, pinst->cpumask.pcpu, cpu_online_mask);
578 cpumask_and(pd->cpumask.cbcpu, pinst->cpumask.cbcpu, cpu_online_mask);
579
580 padata_init_reorder_list(pd);
581 padata_init_squeues(pd);
582 pd->seq_nr = -1;
583 refcount_set(&pd->refcnt, 1);
584 spin_lock_init(&pd->lock);
585 pd->cpu = cpumask_first(pd->cpumask.pcpu);
586 INIT_WORK(&pd->reorder_work, invoke_padata_reorder);
587
588 return pd;
589
590 err_free_pcpu:
591 free_cpumask_var(pd->cpumask.pcpu);
592 err_free_squeue:
593 free_percpu(pd->squeue);
594 err_free_reorder_list:
595 free_percpu(pd->reorder_list);
596 err_free_pd:
597 kfree(pd);
598 err:
599 return NULL;
600 }
601
padata_free_pd(struct parallel_data * pd)602 static void padata_free_pd(struct parallel_data *pd)
603 {
604 free_cpumask_var(pd->cpumask.pcpu);
605 free_cpumask_var(pd->cpumask.cbcpu);
606 free_percpu(pd->reorder_list);
607 free_percpu(pd->squeue);
608 kfree(pd);
609 }
610
__padata_start(struct padata_instance * pinst)611 static void __padata_start(struct padata_instance *pinst)
612 {
613 pinst->flags |= PADATA_INIT;
614 }
615
__padata_stop(struct padata_instance * pinst)616 static void __padata_stop(struct padata_instance *pinst)
617 {
618 if (!(pinst->flags & PADATA_INIT))
619 return;
620
621 pinst->flags &= ~PADATA_INIT;
622
623 synchronize_rcu();
624 }
625
626 /* Replace the internal control structure with a new one. */
padata_replace_one(struct padata_shell * ps)627 static int padata_replace_one(struct padata_shell *ps)
628 {
629 struct parallel_data *pd_new;
630
631 pd_new = padata_alloc_pd(ps);
632 if (!pd_new)
633 return -ENOMEM;
634
635 ps->opd = rcu_dereference_protected(ps->pd, 1);
636 rcu_assign_pointer(ps->pd, pd_new);
637
638 return 0;
639 }
640
padata_replace(struct padata_instance * pinst)641 static int padata_replace(struct padata_instance *pinst)
642 {
643 struct padata_shell *ps;
644 int err = 0;
645
646 pinst->flags |= PADATA_RESET;
647
648 list_for_each_entry(ps, &pinst->pslist, list) {
649 err = padata_replace_one(ps);
650 if (err)
651 break;
652 }
653
654 synchronize_rcu();
655
656 list_for_each_entry_continue_reverse(ps, &pinst->pslist, list)
657 if (refcount_dec_and_test(&ps->opd->refcnt))
658 padata_free_pd(ps->opd);
659
660 pinst->flags &= ~PADATA_RESET;
661
662 return err;
663 }
664
665 /* If cpumask contains no active cpu, we mark the instance as invalid. */
padata_validate_cpumask(struct padata_instance * pinst,const struct cpumask * cpumask)666 static bool padata_validate_cpumask(struct padata_instance *pinst,
667 const struct cpumask *cpumask)
668 {
669 if (!cpumask_intersects(cpumask, cpu_online_mask)) {
670 pinst->flags |= PADATA_INVALID;
671 return false;
672 }
673
674 pinst->flags &= ~PADATA_INVALID;
675 return true;
676 }
677
__padata_set_cpumasks(struct padata_instance * pinst,cpumask_var_t pcpumask,cpumask_var_t cbcpumask)678 static int __padata_set_cpumasks(struct padata_instance *pinst,
679 cpumask_var_t pcpumask,
680 cpumask_var_t cbcpumask)
681 {
682 int valid;
683 int err;
684
685 valid = padata_validate_cpumask(pinst, pcpumask);
686 if (!valid) {
687 __padata_stop(pinst);
688 goto out_replace;
689 }
690
691 valid = padata_validate_cpumask(pinst, cbcpumask);
692 if (!valid)
693 __padata_stop(pinst);
694
695 out_replace:
696 cpumask_copy(pinst->cpumask.pcpu, pcpumask);
697 cpumask_copy(pinst->cpumask.cbcpu, cbcpumask);
698
699 err = padata_setup_cpumasks(pinst) ?: padata_replace(pinst);
700
701 if (valid)
702 __padata_start(pinst);
703
704 return err;
705 }
706
707 /**
708 * padata_set_cpumask - Sets specified by @cpumask_type cpumask to the value
709 * equivalent to @cpumask.
710 * @pinst: padata instance
711 * @cpumask_type: PADATA_CPU_SERIAL or PADATA_CPU_PARALLEL corresponding
712 * to parallel and serial cpumasks respectively.
713 * @cpumask: the cpumask to use
714 *
715 * Return: 0 on success or negative error code
716 */
padata_set_cpumask(struct padata_instance * pinst,int cpumask_type,cpumask_var_t cpumask)717 int padata_set_cpumask(struct padata_instance *pinst, int cpumask_type,
718 cpumask_var_t cpumask)
719 {
720 struct cpumask *serial_mask, *parallel_mask;
721 int err = -EINVAL;
722
723 cpus_read_lock();
724 mutex_lock(&pinst->lock);
725
726 switch (cpumask_type) {
727 case PADATA_CPU_PARALLEL:
728 serial_mask = pinst->cpumask.cbcpu;
729 parallel_mask = cpumask;
730 break;
731 case PADATA_CPU_SERIAL:
732 parallel_mask = pinst->cpumask.pcpu;
733 serial_mask = cpumask;
734 break;
735 default:
736 goto out;
737 }
738
739 err = __padata_set_cpumasks(pinst, parallel_mask, serial_mask);
740
741 out:
742 mutex_unlock(&pinst->lock);
743 cpus_read_unlock();
744
745 return err;
746 }
747 EXPORT_SYMBOL(padata_set_cpumask);
748
749 #ifdef CONFIG_HOTPLUG_CPU
750
__padata_add_cpu(struct padata_instance * pinst,int cpu)751 static int __padata_add_cpu(struct padata_instance *pinst, int cpu)
752 {
753 int err = 0;
754
755 if (cpumask_test_cpu(cpu, cpu_online_mask)) {
756 err = padata_replace(pinst);
757
758 if (padata_validate_cpumask(pinst, pinst->cpumask.pcpu) &&
759 padata_validate_cpumask(pinst, pinst->cpumask.cbcpu))
760 __padata_start(pinst);
761 }
762
763 return err;
764 }
765
__padata_remove_cpu(struct padata_instance * pinst,int cpu)766 static int __padata_remove_cpu(struct padata_instance *pinst, int cpu)
767 {
768 int err = 0;
769
770 if (!cpumask_test_cpu(cpu, cpu_online_mask)) {
771 if (!padata_validate_cpumask(pinst, pinst->cpumask.pcpu) ||
772 !padata_validate_cpumask(pinst, pinst->cpumask.cbcpu))
773 __padata_stop(pinst);
774
775 err = padata_replace(pinst);
776 }
777
778 return err;
779 }
780
pinst_has_cpu(struct padata_instance * pinst,int cpu)781 static inline int pinst_has_cpu(struct padata_instance *pinst, int cpu)
782 {
783 return cpumask_test_cpu(cpu, pinst->cpumask.pcpu) ||
784 cpumask_test_cpu(cpu, pinst->cpumask.cbcpu);
785 }
786
padata_cpu_online(unsigned int cpu,struct hlist_node * node)787 static int padata_cpu_online(unsigned int cpu, struct hlist_node *node)
788 {
789 struct padata_instance *pinst;
790 int ret;
791
792 pinst = hlist_entry_safe(node, struct padata_instance, cpu_online_node);
793 if (!pinst_has_cpu(pinst, cpu))
794 return 0;
795
796 mutex_lock(&pinst->lock);
797 ret = __padata_add_cpu(pinst, cpu);
798 mutex_unlock(&pinst->lock);
799 return ret;
800 }
801
padata_cpu_dead(unsigned int cpu,struct hlist_node * node)802 static int padata_cpu_dead(unsigned int cpu, struct hlist_node *node)
803 {
804 struct padata_instance *pinst;
805 int ret;
806
807 pinst = hlist_entry_safe(node, struct padata_instance, cpu_dead_node);
808 if (!pinst_has_cpu(pinst, cpu))
809 return 0;
810
811 mutex_lock(&pinst->lock);
812 ret = __padata_remove_cpu(pinst, cpu);
813 mutex_unlock(&pinst->lock);
814 return ret;
815 }
816
817 static enum cpuhp_state hp_online;
818 #endif
819
__padata_free(struct padata_instance * pinst)820 static void __padata_free(struct padata_instance *pinst)
821 {
822 #ifdef CONFIG_HOTPLUG_CPU
823 cpuhp_state_remove_instance_nocalls(CPUHP_PADATA_DEAD,
824 &pinst->cpu_dead_node);
825 cpuhp_state_remove_instance_nocalls(hp_online, &pinst->cpu_online_node);
826 #endif
827
828 WARN_ON(!list_empty(&pinst->pslist));
829
830 free_cpumask_var(pinst->cpumask.pcpu);
831 free_cpumask_var(pinst->cpumask.cbcpu);
832 destroy_workqueue(pinst->serial_wq);
833 destroy_workqueue(pinst->parallel_wq);
834 kfree(pinst);
835 }
836
837 #define kobj2pinst(_kobj) \
838 container_of(_kobj, struct padata_instance, kobj)
839 #define attr2pentry(_attr) \
840 container_of(_attr, struct padata_sysfs_entry, attr)
841
padata_sysfs_release(struct kobject * kobj)842 static void padata_sysfs_release(struct kobject *kobj)
843 {
844 struct padata_instance *pinst = kobj2pinst(kobj);
845 __padata_free(pinst);
846 }
847
848 struct padata_sysfs_entry {
849 struct attribute attr;
850 ssize_t (*show)(struct padata_instance *, struct attribute *, char *);
851 ssize_t (*store)(struct padata_instance *, struct attribute *,
852 const char *, size_t);
853 };
854
show_cpumask(struct padata_instance * pinst,struct attribute * attr,char * buf)855 static ssize_t show_cpumask(struct padata_instance *pinst,
856 struct attribute *attr, char *buf)
857 {
858 struct cpumask *cpumask;
859 ssize_t len;
860
861 mutex_lock(&pinst->lock);
862 if (!strcmp(attr->name, "serial_cpumask"))
863 cpumask = pinst->cpumask.cbcpu;
864 else
865 cpumask = pinst->cpumask.pcpu;
866
867 len = snprintf(buf, PAGE_SIZE, "%*pb\n",
868 nr_cpu_ids, cpumask_bits(cpumask));
869 mutex_unlock(&pinst->lock);
870 return len < PAGE_SIZE ? len : -EINVAL;
871 }
872
store_cpumask(struct padata_instance * pinst,struct attribute * attr,const char * buf,size_t count)873 static ssize_t store_cpumask(struct padata_instance *pinst,
874 struct attribute *attr,
875 const char *buf, size_t count)
876 {
877 cpumask_var_t new_cpumask;
878 ssize_t ret;
879 int mask_type;
880
881 if (!alloc_cpumask_var(&new_cpumask, GFP_KERNEL))
882 return -ENOMEM;
883
884 ret = bitmap_parse(buf, count, cpumask_bits(new_cpumask),
885 nr_cpumask_bits);
886 if (ret < 0)
887 goto out;
888
889 mask_type = !strcmp(attr->name, "serial_cpumask") ?
890 PADATA_CPU_SERIAL : PADATA_CPU_PARALLEL;
891 ret = padata_set_cpumask(pinst, mask_type, new_cpumask);
892 if (!ret)
893 ret = count;
894
895 out:
896 free_cpumask_var(new_cpumask);
897 return ret;
898 }
899
900 #define PADATA_ATTR_RW(_name, _show_name, _store_name) \
901 static struct padata_sysfs_entry _name##_attr = \
902 __ATTR(_name, 0644, _show_name, _store_name)
903 #define PADATA_ATTR_RO(_name, _show_name) \
904 static struct padata_sysfs_entry _name##_attr = \
905 __ATTR(_name, 0400, _show_name, NULL)
906
907 PADATA_ATTR_RW(serial_cpumask, show_cpumask, store_cpumask);
908 PADATA_ATTR_RW(parallel_cpumask, show_cpumask, store_cpumask);
909
910 /*
911 * Padata sysfs provides the following objects:
912 * serial_cpumask [RW] - cpumask for serial workers
913 * parallel_cpumask [RW] - cpumask for parallel workers
914 */
915 static struct attribute *padata_default_attrs[] = {
916 &serial_cpumask_attr.attr,
917 ¶llel_cpumask_attr.attr,
918 NULL,
919 };
920 ATTRIBUTE_GROUPS(padata_default);
921
padata_sysfs_show(struct kobject * kobj,struct attribute * attr,char * buf)922 static ssize_t padata_sysfs_show(struct kobject *kobj,
923 struct attribute *attr, char *buf)
924 {
925 struct padata_instance *pinst;
926 struct padata_sysfs_entry *pentry;
927 ssize_t ret = -EIO;
928
929 pinst = kobj2pinst(kobj);
930 pentry = attr2pentry(attr);
931 if (pentry->show)
932 ret = pentry->show(pinst, attr, buf);
933
934 return ret;
935 }
936
padata_sysfs_store(struct kobject * kobj,struct attribute * attr,const char * buf,size_t count)937 static ssize_t padata_sysfs_store(struct kobject *kobj, struct attribute *attr,
938 const char *buf, size_t count)
939 {
940 struct padata_instance *pinst;
941 struct padata_sysfs_entry *pentry;
942 ssize_t ret = -EIO;
943
944 pinst = kobj2pinst(kobj);
945 pentry = attr2pentry(attr);
946 if (pentry->show)
947 ret = pentry->store(pinst, attr, buf, count);
948
949 return ret;
950 }
951
952 static const struct sysfs_ops padata_sysfs_ops = {
953 .show = padata_sysfs_show,
954 .store = padata_sysfs_store,
955 };
956
957 static struct kobj_type padata_attr_type = {
958 .sysfs_ops = &padata_sysfs_ops,
959 .default_groups = padata_default_groups,
960 .release = padata_sysfs_release,
961 };
962
963 /**
964 * padata_alloc - allocate and initialize a padata instance
965 * @name: used to identify the instance
966 *
967 * Return: new instance on success, NULL on error
968 */
padata_alloc(const char * name)969 struct padata_instance *padata_alloc(const char *name)
970 {
971 struct padata_instance *pinst;
972
973 pinst = kzalloc(sizeof(struct padata_instance), GFP_KERNEL);
974 if (!pinst)
975 goto err;
976
977 pinst->parallel_wq = alloc_workqueue("%s_parallel", WQ_UNBOUND, 0,
978 name);
979 if (!pinst->parallel_wq)
980 goto err_free_inst;
981
982 cpus_read_lock();
983
984 pinst->serial_wq = alloc_workqueue("%s_serial", WQ_MEM_RECLAIM |
985 WQ_CPU_INTENSIVE, 1, name);
986 if (!pinst->serial_wq)
987 goto err_put_cpus;
988
989 if (!alloc_cpumask_var(&pinst->cpumask.pcpu, GFP_KERNEL))
990 goto err_free_serial_wq;
991 if (!alloc_cpumask_var(&pinst->cpumask.cbcpu, GFP_KERNEL)) {
992 free_cpumask_var(pinst->cpumask.pcpu);
993 goto err_free_serial_wq;
994 }
995
996 INIT_LIST_HEAD(&pinst->pslist);
997
998 cpumask_copy(pinst->cpumask.pcpu, cpu_possible_mask);
999 cpumask_copy(pinst->cpumask.cbcpu, cpu_possible_mask);
1000
1001 if (padata_setup_cpumasks(pinst))
1002 goto err_free_masks;
1003
1004 __padata_start(pinst);
1005
1006 kobject_init(&pinst->kobj, &padata_attr_type);
1007 mutex_init(&pinst->lock);
1008
1009 #ifdef CONFIG_HOTPLUG_CPU
1010 cpuhp_state_add_instance_nocalls_cpuslocked(hp_online,
1011 &pinst->cpu_online_node);
1012 cpuhp_state_add_instance_nocalls_cpuslocked(CPUHP_PADATA_DEAD,
1013 &pinst->cpu_dead_node);
1014 #endif
1015
1016 cpus_read_unlock();
1017
1018 return pinst;
1019
1020 err_free_masks:
1021 free_cpumask_var(pinst->cpumask.pcpu);
1022 free_cpumask_var(pinst->cpumask.cbcpu);
1023 err_free_serial_wq:
1024 destroy_workqueue(pinst->serial_wq);
1025 err_put_cpus:
1026 cpus_read_unlock();
1027 destroy_workqueue(pinst->parallel_wq);
1028 err_free_inst:
1029 kfree(pinst);
1030 err:
1031 return NULL;
1032 }
1033 EXPORT_SYMBOL(padata_alloc);
1034
1035 /**
1036 * padata_free - free a padata instance
1037 *
1038 * @pinst: padata instance to free
1039 */
padata_free(struct padata_instance * pinst)1040 void padata_free(struct padata_instance *pinst)
1041 {
1042 kobject_put(&pinst->kobj);
1043 }
1044 EXPORT_SYMBOL(padata_free);
1045
1046 /**
1047 * padata_alloc_shell - Allocate and initialize padata shell.
1048 *
1049 * @pinst: Parent padata_instance object.
1050 *
1051 * Return: new shell on success, NULL on error
1052 */
padata_alloc_shell(struct padata_instance * pinst)1053 struct padata_shell *padata_alloc_shell(struct padata_instance *pinst)
1054 {
1055 struct parallel_data *pd;
1056 struct padata_shell *ps;
1057
1058 ps = kzalloc(sizeof(*ps), GFP_KERNEL);
1059 if (!ps)
1060 goto out;
1061
1062 ps->pinst = pinst;
1063
1064 cpus_read_lock();
1065 pd = padata_alloc_pd(ps);
1066 cpus_read_unlock();
1067
1068 if (!pd)
1069 goto out_free_ps;
1070
1071 mutex_lock(&pinst->lock);
1072 RCU_INIT_POINTER(ps->pd, pd);
1073 list_add(&ps->list, &pinst->pslist);
1074 mutex_unlock(&pinst->lock);
1075
1076 return ps;
1077
1078 out_free_ps:
1079 kfree(ps);
1080 out:
1081 return NULL;
1082 }
1083 EXPORT_SYMBOL(padata_alloc_shell);
1084
1085 /**
1086 * padata_free_shell - free a padata shell
1087 *
1088 * @ps: padata shell to free
1089 */
padata_free_shell(struct padata_shell * ps)1090 void padata_free_shell(struct padata_shell *ps)
1091 {
1092 if (!ps)
1093 return;
1094
1095 mutex_lock(&ps->pinst->lock);
1096 list_del(&ps->list);
1097 padata_free_pd(rcu_dereference_protected(ps->pd, 1));
1098 mutex_unlock(&ps->pinst->lock);
1099
1100 kfree(ps);
1101 }
1102 EXPORT_SYMBOL(padata_free_shell);
1103
padata_init(void)1104 void __init padata_init(void)
1105 {
1106 unsigned int i, possible_cpus;
1107 #ifdef CONFIG_HOTPLUG_CPU
1108 int ret;
1109
1110 ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "padata:online",
1111 padata_cpu_online, NULL);
1112 if (ret < 0)
1113 goto err;
1114 hp_online = ret;
1115
1116 ret = cpuhp_setup_state_multi(CPUHP_PADATA_DEAD, "padata:dead",
1117 NULL, padata_cpu_dead);
1118 if (ret < 0)
1119 goto remove_online_state;
1120 #endif
1121
1122 possible_cpus = num_possible_cpus();
1123 padata_works = kmalloc_array(possible_cpus, sizeof(struct padata_work),
1124 GFP_KERNEL);
1125 if (!padata_works)
1126 goto remove_dead_state;
1127
1128 for (i = 0; i < possible_cpus; ++i)
1129 list_add(&padata_works[i].pw_list, &padata_free_works);
1130
1131 return;
1132
1133 remove_dead_state:
1134 #ifdef CONFIG_HOTPLUG_CPU
1135 cpuhp_remove_multi_state(CPUHP_PADATA_DEAD);
1136 remove_online_state:
1137 cpuhp_remove_multi_state(hp_online);
1138 err:
1139 #endif
1140 pr_warn("padata: initialization failed\n");
1141 }
1142