1 // SPDX-License-Identifier: GPL-2.0
3 * padata.c - generic interface to process data streams in parallel
5 * See Documentation/core-api/padata.rst for more information.
7 * Copyright (C) 2008, 2009 secunet Security Networks AG
10 * Copyright (c) 2020 Oracle and/or its affiliates.
13 * This program is free software; you can redistribute it and/or modify it
14 * under the terms and conditions of the GNU General Public License,
15 * version 2, as published by the Free Software Foundation.
17 * This program is distributed in the hope it will be useful, but WITHOUT
18 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
19 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
22 * You should have received a copy of the GNU General Public License along with
23 * this program; if not, write to the Free Software Foundation, Inc.,
24 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
27 #include <linux/completion.h>
28 #include <linux/export.h>
29 #include <linux/cpumask.h>
30 #include <linux/err.h>
31 #include <linux/cpu.h>
32 #include <linux/padata.h>
33 #include <linux/mutex.h>
34 #include <linux/sched.h>
35 #include <linux/slab.h>
36 #include <linux/sysfs.h>
37 #include <linux/rcupdate.h>
39 #define PADATA_WORK_ONSTACK 1 /* Work's memory is on stack */
42 struct work_struct pw_work;
43 struct list_head pw_list; /* padata_free_works linkage */
47 static DEFINE_SPINLOCK(padata_works_lock);
48 static struct padata_work *padata_works;
49 static LIST_HEAD(padata_free_works);
51 struct padata_mt_job_state {
53 struct completion completion;
54 struct padata_mt_job *job;
57 unsigned long chunk_size;
60 static void padata_free_pd(struct parallel_data *pd);
61 static void __init padata_mt_helper(struct work_struct *work);
63 static int padata_index_to_cpu(struct parallel_data *pd, int cpu_index)
67 target_cpu = cpumask_first(pd->cpumask.pcpu);
68 for (cpu = 0; cpu < cpu_index; cpu++)
69 target_cpu = cpumask_next(target_cpu, pd->cpumask.pcpu);
74 static int padata_cpu_hash(struct parallel_data *pd, unsigned int seq_nr)
77 * Hash the sequence numbers to the cpus by taking
78 * seq_nr mod. number of cpus in use.
80 int cpu_index = seq_nr % cpumask_weight(pd->cpumask.pcpu);
82 return padata_index_to_cpu(pd, cpu_index);
85 static struct padata_work *padata_work_alloc(void)
87 struct padata_work *pw;
89 lockdep_assert_held(&padata_works_lock);
91 if (list_empty(&padata_free_works))
92 return NULL; /* No more work items allowed to be queued. */
94 pw = list_first_entry(&padata_free_works, struct padata_work, pw_list);
95 list_del(&pw->pw_list);
99 static void padata_work_init(struct padata_work *pw, work_func_t work_fn,
100 void *data, int flags)
102 if (flags & PADATA_WORK_ONSTACK)
103 INIT_WORK_ONSTACK(&pw->pw_work, work_fn);
105 INIT_WORK(&pw->pw_work, work_fn);
109 static int __init padata_work_alloc_mt(int nworks, void *data,
110 struct list_head *head)
114 spin_lock(&padata_works_lock);
115 /* Start at 1 because the current task participates in the job. */
116 for (i = 1; i < nworks; ++i) {
117 struct padata_work *pw = padata_work_alloc();
121 padata_work_init(pw, padata_mt_helper, data, 0);
122 list_add(&pw->pw_list, head);
124 spin_unlock(&padata_works_lock);
129 static void padata_work_free(struct padata_work *pw)
131 lockdep_assert_held(&padata_works_lock);
132 list_add(&pw->pw_list, &padata_free_works);
135 static void __init padata_works_free(struct list_head *works)
137 struct padata_work *cur, *next;
139 if (list_empty(works))
142 spin_lock(&padata_works_lock);
143 list_for_each_entry_safe(cur, next, works, pw_list) {
144 list_del(&cur->pw_list);
145 padata_work_free(cur);
147 spin_unlock(&padata_works_lock);
150 static void padata_parallel_worker(struct work_struct *parallel_work)
152 struct padata_work *pw = container_of(parallel_work, struct padata_work,
154 struct padata_priv *padata = pw->pw_data;
157 padata->parallel(padata);
158 spin_lock(&padata_works_lock);
159 padata_work_free(pw);
160 spin_unlock(&padata_works_lock);
165 * padata_do_parallel - padata parallelization function
168 * @padata: object to be parallelized
169 * @cb_cpu: pointer to the CPU that the serialization callback function should
170 * run on. If it's not in the serial cpumask of @pinst
171 * (i.e. cpumask.cbcpu), this function selects a fallback CPU and if
172 * none found, returns -EINVAL.
174 * The parallelization callback function will run with BHs off.
175 * Note: Every object which is parallelized by padata_do_parallel
176 * must be seen by padata_do_serial.
178 * Return: 0 on success or else negative error code.
180 int padata_do_parallel(struct padata_shell *ps,
181 struct padata_priv *padata, int *cb_cpu)
183 struct padata_instance *pinst = ps->pinst;
184 int i, cpu, cpu_index, err;
185 struct parallel_data *pd;
186 struct padata_work *pw;
190 pd = rcu_dereference_bh(ps->pd);
193 if (!(pinst->flags & PADATA_INIT) || pinst->flags & PADATA_INVALID)
196 if (!cpumask_test_cpu(*cb_cpu, pd->cpumask.cbcpu)) {
197 if (!cpumask_weight(pd->cpumask.cbcpu))
200 /* Select an alternate fallback CPU and notify the caller. */
201 cpu_index = *cb_cpu % cpumask_weight(pd->cpumask.cbcpu);
203 cpu = cpumask_first(pd->cpumask.cbcpu);
204 for (i = 0; i < cpu_index; i++)
205 cpu = cpumask_next(cpu, pd->cpumask.cbcpu);
211 if ((pinst->flags & PADATA_RESET))
214 atomic_inc(&pd->refcnt);
216 padata->cb_cpu = *cb_cpu;
218 rcu_read_unlock_bh();
220 spin_lock(&padata_works_lock);
221 padata->seq_nr = ++pd->seq_nr;
222 pw = padata_work_alloc();
223 spin_unlock(&padata_works_lock);
225 padata_work_init(pw, padata_parallel_worker, padata, 0);
226 queue_work(pinst->parallel_wq, &pw->pw_work);
228 /* Maximum works limit exceeded, run in the current task. */
229 padata->parallel(padata);
234 rcu_read_unlock_bh();
238 EXPORT_SYMBOL(padata_do_parallel);
241 * padata_find_next - Find the next object that needs serialization.
244 * * A pointer to the control struct of the next object that needs
245 * serialization, if present in one of the percpu reorder queues.
246 * * NULL, if the next object that needs serialization will
247 * be parallel processed by another cpu and is not yet present in
248 * the cpu's reorder queue.
250 static struct padata_priv *padata_find_next(struct parallel_data *pd,
253 struct padata_parallel_queue *next_queue;
254 struct padata_priv *padata;
255 struct padata_list *reorder;
258 next_queue = per_cpu_ptr(pd->pqueue, cpu);
259 reorder = &next_queue->reorder;
261 spin_lock(&reorder->lock);
262 if (list_empty(&reorder->list)) {
263 spin_unlock(&reorder->lock);
267 padata = list_entry(reorder->list.next, struct padata_priv, list);
270 * Checks the rare case where two or more parallel jobs have hashed to
271 * the same CPU and one of the later ones finishes first.
273 if (padata->seq_nr != pd->processed) {
274 spin_unlock(&reorder->lock);
279 list_del_init(&padata->list);
281 pd->cpu = cpumask_next_wrap(cpu, pd->cpumask.pcpu, -1, false);
284 spin_unlock(&reorder->lock);
288 static void padata_reorder(struct parallel_data *pd)
290 struct padata_instance *pinst = pd->ps->pinst;
292 struct padata_priv *padata;
293 struct padata_serial_queue *squeue;
294 struct padata_parallel_queue *next_queue;
297 * We need to ensure that only one cpu can work on dequeueing of
298 * the reorder queue the time. Calculating in which percpu reorder
299 * queue the next object will arrive takes some time. A spinlock
300 * would be highly contended. Also it is not clear in which order
301 * the objects arrive to the reorder queues. So a cpu could wait to
302 * get the lock just to notice that there is nothing to do at the
303 * moment. Therefore we use a trylock and let the holder of the lock
304 * care for all the objects enqueued during the holdtime of the lock.
306 if (!spin_trylock_bh(&pd->lock))
310 padata = padata_find_next(pd, true);
313 * If the next object that needs serialization is parallel
314 * processed by another cpu and is still on it's way to the
315 * cpu's reorder queue, nothing to do for now.
320 cb_cpu = padata->cb_cpu;
321 squeue = per_cpu_ptr(pd->squeue, cb_cpu);
323 spin_lock(&squeue->serial.lock);
324 list_add_tail(&padata->list, &squeue->serial.list);
325 spin_unlock(&squeue->serial.lock);
327 queue_work_on(cb_cpu, pinst->serial_wq, &squeue->work);
330 spin_unlock_bh(&pd->lock);
333 * The next object that needs serialization might have arrived to
334 * the reorder queues in the meantime.
336 * Ensure reorder queue is read after pd->lock is dropped so we see
337 * new objects from another task in padata_do_serial. Pairs with
338 * smp_mb__after_atomic in padata_do_serial.
342 next_queue = per_cpu_ptr(pd->pqueue, pd->cpu);
343 if (!list_empty(&next_queue->reorder.list) &&
344 padata_find_next(pd, false))
345 queue_work(pinst->serial_wq, &pd->reorder_work);
348 static void invoke_padata_reorder(struct work_struct *work)
350 struct parallel_data *pd;
353 pd = container_of(work, struct parallel_data, reorder_work);
358 static void padata_serial_worker(struct work_struct *serial_work)
360 struct padata_serial_queue *squeue;
361 struct parallel_data *pd;
362 LIST_HEAD(local_list);
366 squeue = container_of(serial_work, struct padata_serial_queue, work);
369 spin_lock(&squeue->serial.lock);
370 list_replace_init(&squeue->serial.list, &local_list);
371 spin_unlock(&squeue->serial.lock);
375 while (!list_empty(&local_list)) {
376 struct padata_priv *padata;
378 padata = list_entry(local_list.next,
379 struct padata_priv, list);
381 list_del_init(&padata->list);
383 padata->serial(padata);
388 if (atomic_sub_and_test(cnt, &pd->refcnt))
393 * padata_do_serial - padata serialization function
395 * @padata: object to be serialized.
397 * padata_do_serial must be called for every parallelized object.
398 * The serialization callback function will run with BHs off.
400 void padata_do_serial(struct padata_priv *padata)
402 struct parallel_data *pd = padata->pd;
403 int hashed_cpu = padata_cpu_hash(pd, padata->seq_nr);
404 struct padata_parallel_queue *pqueue = per_cpu_ptr(pd->pqueue,
406 struct padata_priv *cur;
408 spin_lock(&pqueue->reorder.lock);
409 /* Sort in ascending order of sequence number. */
410 list_for_each_entry_reverse(cur, &pqueue->reorder.list, list)
411 if (cur->seq_nr < padata->seq_nr)
413 list_add(&padata->list, &cur->list);
414 spin_unlock(&pqueue->reorder.lock);
417 * Ensure the addition to the reorder list is ordered correctly
418 * with the trylock of pd->lock in padata_reorder. Pairs with smp_mb
421 smp_mb__after_atomic();
425 EXPORT_SYMBOL(padata_do_serial);
427 static int padata_setup_cpumasks(struct padata_instance *pinst)
429 struct workqueue_attrs *attrs;
432 attrs = alloc_workqueue_attrs();
436 /* Restrict parallel_wq workers to pd->cpumask.pcpu. */
437 cpumask_copy(attrs->cpumask, pinst->cpumask.pcpu);
438 err = apply_workqueue_attrs(pinst->parallel_wq, attrs);
439 free_workqueue_attrs(attrs);
444 static int pd_setup_cpumasks(struct parallel_data *pd,
445 const struct cpumask *pcpumask,
446 const struct cpumask *cbcpumask)
450 if (!alloc_cpumask_var(&pd->cpumask.pcpu, GFP_KERNEL))
452 if (!alloc_cpumask_var(&pd->cpumask.cbcpu, GFP_KERNEL))
455 cpumask_copy(pd->cpumask.pcpu, pcpumask);
456 cpumask_copy(pd->cpumask.cbcpu, cbcpumask);
461 free_cpumask_var(pd->cpumask.pcpu);
466 static void __init padata_mt_helper(struct work_struct *w)
468 struct padata_work *pw = container_of(w, struct padata_work, pw_work);
469 struct padata_mt_job_state *ps = pw->pw_data;
470 struct padata_mt_job *job = ps->job;
473 spin_lock(&ps->lock);
475 while (job->size > 0) {
476 unsigned long start, size, end;
479 /* So end is chunk size aligned if enough work remains. */
480 size = roundup(start + 1, ps->chunk_size) - start;
481 size = min(size, job->size);
487 spin_unlock(&ps->lock);
488 job->thread_fn(start, end, job->fn_arg);
489 spin_lock(&ps->lock);
493 done = (ps->nworks_fini == ps->nworks);
494 spin_unlock(&ps->lock);
497 complete(&ps->completion);
501 * padata_do_multithreaded - run a multithreaded job
502 * @job: Description of the job.
504 * See the definition of struct padata_mt_job for more details.
506 void __init padata_do_multithreaded(struct padata_mt_job *job)
508 /* In case threads finish at different times. */
509 static const unsigned long load_balance_factor = 4;
510 struct padata_work my_work, *pw;
511 struct padata_mt_job_state ps;
518 /* Ensure at least one thread when size < min_chunk. */
519 nworks = max(job->size / job->min_chunk, 1ul);
520 nworks = min(nworks, job->max_threads);
523 /* Single thread, no coordination needed, cut to the chase. */
524 job->thread_fn(job->start, job->start + job->size, job->fn_arg);
528 spin_lock_init(&ps.lock);
529 init_completion(&ps.completion);
531 ps.nworks = padata_work_alloc_mt(nworks, &ps, &works);
535 * Chunk size is the amount of work a helper does per call to the
536 * thread function. Load balance large jobs between threads by
537 * increasing the number of chunks, guarantee at least the minimum
538 * chunk size from the caller, and honor the caller's alignment.
540 ps.chunk_size = job->size / (ps.nworks * load_balance_factor);
541 ps.chunk_size = max(ps.chunk_size, job->min_chunk);
542 ps.chunk_size = roundup(ps.chunk_size, job->align);
544 list_for_each_entry(pw, &works, pw_list)
545 queue_work(system_unbound_wq, &pw->pw_work);
547 /* Use the current thread, which saves starting a workqueue worker. */
548 padata_work_init(&my_work, padata_mt_helper, &ps, PADATA_WORK_ONSTACK);
549 padata_mt_helper(&my_work.pw_work);
551 /* Wait for all the helpers to finish. */
552 wait_for_completion(&ps.completion);
554 destroy_work_on_stack(&my_work.pw_work);
555 padata_works_free(&works);
558 static void __padata_list_init(struct padata_list *pd_list)
560 INIT_LIST_HEAD(&pd_list->list);
561 spin_lock_init(&pd_list->lock);
564 /* Initialize all percpu queues used by serial workers */
565 static void padata_init_squeues(struct parallel_data *pd)
568 struct padata_serial_queue *squeue;
570 for_each_cpu(cpu, pd->cpumask.cbcpu) {
571 squeue = per_cpu_ptr(pd->squeue, cpu);
573 __padata_list_init(&squeue->serial);
574 INIT_WORK(&squeue->work, padata_serial_worker);
578 /* Initialize all percpu queues used by parallel workers */
579 static void padata_init_pqueues(struct parallel_data *pd)
582 struct padata_parallel_queue *pqueue;
584 for_each_cpu(cpu, pd->cpumask.pcpu) {
585 pqueue = per_cpu_ptr(pd->pqueue, cpu);
587 __padata_list_init(&pqueue->reorder);
588 atomic_set(&pqueue->num_obj, 0);
592 /* Allocate and initialize the internal cpumask dependend resources. */
593 static struct parallel_data *padata_alloc_pd(struct padata_shell *ps)
595 struct padata_instance *pinst = ps->pinst;
596 const struct cpumask *cbcpumask;
597 const struct cpumask *pcpumask;
598 struct parallel_data *pd;
600 cbcpumask = pinst->rcpumask.cbcpu;
601 pcpumask = pinst->rcpumask.pcpu;
603 pd = kzalloc(sizeof(struct parallel_data), GFP_KERNEL);
607 pd->pqueue = alloc_percpu(struct padata_parallel_queue);
611 pd->squeue = alloc_percpu(struct padata_serial_queue);
613 goto err_free_pqueue;
616 if (pd_setup_cpumasks(pd, pcpumask, cbcpumask))
617 goto err_free_squeue;
619 padata_init_pqueues(pd);
620 padata_init_squeues(pd);
622 atomic_set(&pd->refcnt, 1);
623 spin_lock_init(&pd->lock);
624 pd->cpu = cpumask_first(pd->cpumask.pcpu);
625 INIT_WORK(&pd->reorder_work, invoke_padata_reorder);
630 free_percpu(pd->squeue);
632 free_percpu(pd->pqueue);
639 static void padata_free_pd(struct parallel_data *pd)
641 free_cpumask_var(pd->cpumask.pcpu);
642 free_cpumask_var(pd->cpumask.cbcpu);
643 free_percpu(pd->pqueue);
644 free_percpu(pd->squeue);
648 static void __padata_start(struct padata_instance *pinst)
650 pinst->flags |= PADATA_INIT;
653 static void __padata_stop(struct padata_instance *pinst)
655 if (!(pinst->flags & PADATA_INIT))
658 pinst->flags &= ~PADATA_INIT;
663 /* Replace the internal control structure with a new one. */
664 static int padata_replace_one(struct padata_shell *ps)
666 struct parallel_data *pd_new;
668 pd_new = padata_alloc_pd(ps);
672 ps->opd = rcu_dereference_protected(ps->pd, 1);
673 rcu_assign_pointer(ps->pd, pd_new);
678 static int padata_replace(struct padata_instance *pinst)
680 struct padata_shell *ps;
683 pinst->flags |= PADATA_RESET;
685 cpumask_and(pinst->rcpumask.pcpu, pinst->cpumask.pcpu,
688 cpumask_and(pinst->rcpumask.cbcpu, pinst->cpumask.cbcpu,
691 list_for_each_entry(ps, &pinst->pslist, list) {
692 err = padata_replace_one(ps);
699 list_for_each_entry_continue_reverse(ps, &pinst->pslist, list)
700 if (atomic_dec_and_test(&ps->opd->refcnt))
701 padata_free_pd(ps->opd);
703 pinst->flags &= ~PADATA_RESET;
708 /* If cpumask contains no active cpu, we mark the instance as invalid. */
709 static bool padata_validate_cpumask(struct padata_instance *pinst,
710 const struct cpumask *cpumask)
712 if (!cpumask_intersects(cpumask, cpu_online_mask)) {
713 pinst->flags |= PADATA_INVALID;
717 pinst->flags &= ~PADATA_INVALID;
721 static int __padata_set_cpumasks(struct padata_instance *pinst,
722 cpumask_var_t pcpumask,
723 cpumask_var_t cbcpumask)
728 valid = padata_validate_cpumask(pinst, pcpumask);
730 __padata_stop(pinst);
734 valid = padata_validate_cpumask(pinst, cbcpumask);
736 __padata_stop(pinst);
739 cpumask_copy(pinst->cpumask.pcpu, pcpumask);
740 cpumask_copy(pinst->cpumask.cbcpu, cbcpumask);
742 err = padata_setup_cpumasks(pinst) ?: padata_replace(pinst);
745 __padata_start(pinst);
751 * padata_set_cpumask - Sets specified by @cpumask_type cpumask to the value
752 * equivalent to @cpumask.
753 * @pinst: padata instance
754 * @cpumask_type: PADATA_CPU_SERIAL or PADATA_CPU_PARALLEL corresponding
755 * to parallel and serial cpumasks respectively.
756 * @cpumask: the cpumask to use
758 * Return: 0 on success or negative error code
760 int padata_set_cpumask(struct padata_instance *pinst, int cpumask_type,
761 cpumask_var_t cpumask)
763 struct cpumask *serial_mask, *parallel_mask;
767 mutex_lock(&pinst->lock);
769 switch (cpumask_type) {
770 case PADATA_CPU_PARALLEL:
771 serial_mask = pinst->cpumask.cbcpu;
772 parallel_mask = cpumask;
774 case PADATA_CPU_SERIAL:
775 parallel_mask = pinst->cpumask.pcpu;
776 serial_mask = cpumask;
782 err = __padata_set_cpumasks(pinst, parallel_mask, serial_mask);
785 mutex_unlock(&pinst->lock);
790 EXPORT_SYMBOL(padata_set_cpumask);
793 * padata_start - start the parallel processing
795 * @pinst: padata instance to start
797 * Return: 0 on success or negative error code
799 int padata_start(struct padata_instance *pinst)
803 mutex_lock(&pinst->lock);
805 if (pinst->flags & PADATA_INVALID)
808 __padata_start(pinst);
810 mutex_unlock(&pinst->lock);
814 EXPORT_SYMBOL(padata_start);
817 * padata_stop - stop the parallel processing
819 * @pinst: padata instance to stop
821 void padata_stop(struct padata_instance *pinst)
823 mutex_lock(&pinst->lock);
824 __padata_stop(pinst);
825 mutex_unlock(&pinst->lock);
827 EXPORT_SYMBOL(padata_stop);
829 #ifdef CONFIG_HOTPLUG_CPU
831 static int __padata_add_cpu(struct padata_instance *pinst, int cpu)
835 if (cpumask_test_cpu(cpu, cpu_online_mask)) {
836 err = padata_replace(pinst);
838 if (padata_validate_cpumask(pinst, pinst->cpumask.pcpu) &&
839 padata_validate_cpumask(pinst, pinst->cpumask.cbcpu))
840 __padata_start(pinst);
846 static int __padata_remove_cpu(struct padata_instance *pinst, int cpu)
850 if (!cpumask_test_cpu(cpu, cpu_online_mask)) {
851 if (!padata_validate_cpumask(pinst, pinst->cpumask.pcpu) ||
852 !padata_validate_cpumask(pinst, pinst->cpumask.cbcpu))
853 __padata_stop(pinst);
855 err = padata_replace(pinst);
861 static inline int pinst_has_cpu(struct padata_instance *pinst, int cpu)
863 return cpumask_test_cpu(cpu, pinst->cpumask.pcpu) ||
864 cpumask_test_cpu(cpu, pinst->cpumask.cbcpu);
867 static int padata_cpu_online(unsigned int cpu, struct hlist_node *node)
869 struct padata_instance *pinst;
872 pinst = hlist_entry_safe(node, struct padata_instance, cpu_online_node);
873 if (!pinst_has_cpu(pinst, cpu))
876 mutex_lock(&pinst->lock);
877 ret = __padata_add_cpu(pinst, cpu);
878 mutex_unlock(&pinst->lock);
882 static int padata_cpu_dead(unsigned int cpu, struct hlist_node *node)
884 struct padata_instance *pinst;
887 pinst = hlist_entry_safe(node, struct padata_instance, cpu_dead_node);
888 if (!pinst_has_cpu(pinst, cpu))
891 mutex_lock(&pinst->lock);
892 ret = __padata_remove_cpu(pinst, cpu);
893 mutex_unlock(&pinst->lock);
897 static enum cpuhp_state hp_online;
900 static void __padata_free(struct padata_instance *pinst)
902 #ifdef CONFIG_HOTPLUG_CPU
903 cpuhp_state_remove_instance_nocalls(CPUHP_PADATA_DEAD,
904 &pinst->cpu_dead_node);
905 cpuhp_state_remove_instance_nocalls(hp_online, &pinst->cpu_online_node);
908 WARN_ON(!list_empty(&pinst->pslist));
911 free_cpumask_var(pinst->rcpumask.cbcpu);
912 free_cpumask_var(pinst->rcpumask.pcpu);
913 free_cpumask_var(pinst->cpumask.pcpu);
914 free_cpumask_var(pinst->cpumask.cbcpu);
915 destroy_workqueue(pinst->serial_wq);
916 destroy_workqueue(pinst->parallel_wq);
920 #define kobj2pinst(_kobj) \
921 container_of(_kobj, struct padata_instance, kobj)
922 #define attr2pentry(_attr) \
923 container_of(_attr, struct padata_sysfs_entry, attr)
925 static void padata_sysfs_release(struct kobject *kobj)
927 struct padata_instance *pinst = kobj2pinst(kobj);
928 __padata_free(pinst);
931 struct padata_sysfs_entry {
932 struct attribute attr;
933 ssize_t (*show)(struct padata_instance *, struct attribute *, char *);
934 ssize_t (*store)(struct padata_instance *, struct attribute *,
935 const char *, size_t);
938 static ssize_t show_cpumask(struct padata_instance *pinst,
939 struct attribute *attr, char *buf)
941 struct cpumask *cpumask;
944 mutex_lock(&pinst->lock);
945 if (!strcmp(attr->name, "serial_cpumask"))
946 cpumask = pinst->cpumask.cbcpu;
948 cpumask = pinst->cpumask.pcpu;
950 len = snprintf(buf, PAGE_SIZE, "%*pb\n",
951 nr_cpu_ids, cpumask_bits(cpumask));
952 mutex_unlock(&pinst->lock);
953 return len < PAGE_SIZE ? len : -EINVAL;
956 static ssize_t store_cpumask(struct padata_instance *pinst,
957 struct attribute *attr,
958 const char *buf, size_t count)
960 cpumask_var_t new_cpumask;
964 if (!alloc_cpumask_var(&new_cpumask, GFP_KERNEL))
967 ret = bitmap_parse(buf, count, cpumask_bits(new_cpumask),
972 mask_type = !strcmp(attr->name, "serial_cpumask") ?
973 PADATA_CPU_SERIAL : PADATA_CPU_PARALLEL;
974 ret = padata_set_cpumask(pinst, mask_type, new_cpumask);
979 free_cpumask_var(new_cpumask);
983 #define PADATA_ATTR_RW(_name, _show_name, _store_name) \
984 static struct padata_sysfs_entry _name##_attr = \
985 __ATTR(_name, 0644, _show_name, _store_name)
986 #define PADATA_ATTR_RO(_name, _show_name) \
987 static struct padata_sysfs_entry _name##_attr = \
988 __ATTR(_name, 0400, _show_name, NULL)
990 PADATA_ATTR_RW(serial_cpumask, show_cpumask, store_cpumask);
991 PADATA_ATTR_RW(parallel_cpumask, show_cpumask, store_cpumask);
994 * Padata sysfs provides the following objects:
995 * serial_cpumask [RW] - cpumask for serial workers
996 * parallel_cpumask [RW] - cpumask for parallel workers
998 static struct attribute *padata_default_attrs[] = {
999 &serial_cpumask_attr.attr,
1000 ¶llel_cpumask_attr.attr,
1003 ATTRIBUTE_GROUPS(padata_default);
1005 static ssize_t padata_sysfs_show(struct kobject *kobj,
1006 struct attribute *attr, char *buf)
1008 struct padata_instance *pinst;
1009 struct padata_sysfs_entry *pentry;
1012 pinst = kobj2pinst(kobj);
1013 pentry = attr2pentry(attr);
1015 ret = pentry->show(pinst, attr, buf);
1020 static ssize_t padata_sysfs_store(struct kobject *kobj, struct attribute *attr,
1021 const char *buf, size_t count)
1023 struct padata_instance *pinst;
1024 struct padata_sysfs_entry *pentry;
1027 pinst = kobj2pinst(kobj);
1028 pentry = attr2pentry(attr);
1030 ret = pentry->store(pinst, attr, buf, count);
1035 static const struct sysfs_ops padata_sysfs_ops = {
1036 .show = padata_sysfs_show,
1037 .store = padata_sysfs_store,
1040 static struct kobj_type padata_attr_type = {
1041 .sysfs_ops = &padata_sysfs_ops,
1042 .default_groups = padata_default_groups,
1043 .release = padata_sysfs_release,
1047 * padata_alloc - allocate and initialize a padata instance and specify
1048 * cpumasks for serial and parallel workers.
1050 * @name: used to identify the instance
1051 * @pcpumask: cpumask that will be used for padata parallelization
1052 * @cbcpumask: cpumask that will be used for padata serialization
1054 * Return: new instance on success, NULL on error
1056 static struct padata_instance *padata_alloc(const char *name,
1057 const struct cpumask *pcpumask,
1058 const struct cpumask *cbcpumask)
1060 struct padata_instance *pinst;
1062 pinst = kzalloc(sizeof(struct padata_instance), GFP_KERNEL);
1066 pinst->parallel_wq = alloc_workqueue("%s_parallel", WQ_UNBOUND, 0,
1068 if (!pinst->parallel_wq)
1073 pinst->serial_wq = alloc_workqueue("%s_serial", WQ_MEM_RECLAIM |
1074 WQ_CPU_INTENSIVE, 1, name);
1075 if (!pinst->serial_wq)
1078 if (!alloc_cpumask_var(&pinst->cpumask.pcpu, GFP_KERNEL))
1079 goto err_free_serial_wq;
1080 if (!alloc_cpumask_var(&pinst->cpumask.cbcpu, GFP_KERNEL)) {
1081 free_cpumask_var(pinst->cpumask.pcpu);
1082 goto err_free_serial_wq;
1084 if (!padata_validate_cpumask(pinst, pcpumask) ||
1085 !padata_validate_cpumask(pinst, cbcpumask))
1086 goto err_free_masks;
1088 if (!alloc_cpumask_var(&pinst->rcpumask.pcpu, GFP_KERNEL))
1089 goto err_free_masks;
1090 if (!alloc_cpumask_var(&pinst->rcpumask.cbcpu, GFP_KERNEL))
1091 goto err_free_rcpumask_pcpu;
1093 INIT_LIST_HEAD(&pinst->pslist);
1095 cpumask_copy(pinst->cpumask.pcpu, pcpumask);
1096 cpumask_copy(pinst->cpumask.cbcpu, cbcpumask);
1097 cpumask_and(pinst->rcpumask.pcpu, pcpumask, cpu_online_mask);
1098 cpumask_and(pinst->rcpumask.cbcpu, cbcpumask, cpu_online_mask);
1100 if (padata_setup_cpumasks(pinst))
1101 goto err_free_rcpumask_cbcpu;
1105 kobject_init(&pinst->kobj, &padata_attr_type);
1106 mutex_init(&pinst->lock);
1108 #ifdef CONFIG_HOTPLUG_CPU
1109 cpuhp_state_add_instance_nocalls_cpuslocked(hp_online,
1110 &pinst->cpu_online_node);
1111 cpuhp_state_add_instance_nocalls_cpuslocked(CPUHP_PADATA_DEAD,
1112 &pinst->cpu_dead_node);
1119 err_free_rcpumask_cbcpu:
1120 free_cpumask_var(pinst->rcpumask.cbcpu);
1121 err_free_rcpumask_pcpu:
1122 free_cpumask_var(pinst->rcpumask.pcpu);
1124 free_cpumask_var(pinst->cpumask.pcpu);
1125 free_cpumask_var(pinst->cpumask.cbcpu);
1127 destroy_workqueue(pinst->serial_wq);
1130 destroy_workqueue(pinst->parallel_wq);
1138 * padata_alloc_possible - Allocate and initialize padata instance.
1139 * Use the cpu_possible_mask for serial and
1142 * @name: used to identify the instance
1144 * Return: new instance on success, NULL on error
1146 struct padata_instance *padata_alloc_possible(const char *name)
1148 return padata_alloc(name, cpu_possible_mask, cpu_possible_mask);
1150 EXPORT_SYMBOL(padata_alloc_possible);
1153 * padata_free - free a padata instance
1155 * @pinst: padata instance to free
1157 void padata_free(struct padata_instance *pinst)
1159 kobject_put(&pinst->kobj);
1161 EXPORT_SYMBOL(padata_free);
1164 * padata_alloc_shell - Allocate and initialize padata shell.
1166 * @pinst: Parent padata_instance object.
1168 * Return: new shell on success, NULL on error
1170 struct padata_shell *padata_alloc_shell(struct padata_instance *pinst)
1172 struct parallel_data *pd;
1173 struct padata_shell *ps;
1175 ps = kzalloc(sizeof(*ps), GFP_KERNEL);
1182 pd = padata_alloc_pd(ps);
1188 mutex_lock(&pinst->lock);
1189 RCU_INIT_POINTER(ps->pd, pd);
1190 list_add(&ps->list, &pinst->pslist);
1191 mutex_unlock(&pinst->lock);
1200 EXPORT_SYMBOL(padata_alloc_shell);
1203 * padata_free_shell - free a padata shell
1205 * @ps: padata shell to free
1207 void padata_free_shell(struct padata_shell *ps)
1212 mutex_lock(&ps->pinst->lock);
1213 list_del(&ps->list);
1214 padata_free_pd(rcu_dereference_protected(ps->pd, 1));
1215 mutex_unlock(&ps->pinst->lock);
1219 EXPORT_SYMBOL(padata_free_shell);
1221 void __init padata_init(void)
1223 unsigned int i, possible_cpus;
1224 #ifdef CONFIG_HOTPLUG_CPU
1227 ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "padata:online",
1228 padata_cpu_online, NULL);
1233 ret = cpuhp_setup_state_multi(CPUHP_PADATA_DEAD, "padata:dead",
1234 NULL, padata_cpu_dead);
1236 goto remove_online_state;
1239 possible_cpus = num_possible_cpus();
1240 padata_works = kmalloc_array(possible_cpus, sizeof(struct padata_work),
1243 goto remove_dead_state;
1245 for (i = 0; i < possible_cpus; ++i)
1246 list_add(&padata_works[i].pw_list, &padata_free_works);
1251 #ifdef CONFIG_HOTPLUG_CPU
1252 cpuhp_remove_multi_state(CPUHP_PADATA_DEAD);
1253 remove_online_state:
1254 cpuhp_remove_multi_state(hp_online);
1257 pr_warn("padata: initialization failed\n");