]>
Commit | Line | Data |
---|---|---|
08b21fbf | 1 | // SPDX-License-Identifier: GPL-2.0 |
16295bec SK |
2 | /* |
3 | * padata.c - generic interface to process data streams in parallel | |
4 | * | |
bfcdcef8 | 5 | * See Documentation/core-api/padata.rst for more information. |
107f8bda | 6 | * |
16295bec SK |
7 | * Copyright (C) 2008, 2009 secunet Security Networks AG |
8 | * Copyright (C) 2008, 2009 Steffen Klassert <steffen.klassert@secunet.com> | |
9 | * | |
004ed426 DJ |
10 | * Copyright (c) 2020 Oracle and/or its affiliates. |
11 | * Author: Daniel Jordan <daniel.m.jordan@oracle.com> | |
12 | * | |
16295bec SK |
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. | |
16 | * | |
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 | |
20 | * more details. | |
21 | * | |
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. | |
25 | */ | |
26 | ||
004ed426 | 27 | #include <linux/completion.h> |
9984de1a | 28 | #include <linux/export.h> |
16295bec SK |
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> | |
5a0e3ad6 | 35 | #include <linux/slab.h> |
5e017dc3 | 36 | #include <linux/sysfs.h> |
16295bec SK |
37 | #include <linux/rcupdate.h> |
38 | ||
004ed426 DJ |
39 | #define PADATA_WORK_ONSTACK 1 /* Work's memory is on stack */ |
40 | ||
4611ce22 DJ |
41 | struct padata_work { |
42 | struct work_struct pw_work; | |
43 | struct list_head pw_list; /* padata_free_works linkage */ | |
44 | void *pw_data; | |
45 | }; | |
46 | ||
47 | static DEFINE_SPINLOCK(padata_works_lock); | |
48 | static struct padata_work *padata_works; | |
49 | static LIST_HEAD(padata_free_works); | |
16295bec | 50 | |
004ed426 DJ |
51 | struct padata_mt_job_state { |
52 | spinlock_t lock; | |
53 | struct completion completion; | |
54 | struct padata_mt_job *job; | |
55 | int nworks; | |
56 | int nworks_fini; | |
57 | unsigned long chunk_size; | |
58 | }; | |
59 | ||
07928d9b | 60 | static void padata_free_pd(struct parallel_data *pd); |
004ed426 | 61 | static void __init padata_mt_helper(struct work_struct *work); |
07928d9b | 62 | |
16295bec SK |
63 | static int padata_index_to_cpu(struct parallel_data *pd, int cpu_index) |
64 | { | |
65 | int cpu, target_cpu; | |
66 | ||
e15bacbe | 67 | target_cpu = cpumask_first(pd->cpumask.pcpu); |
16295bec | 68 | for (cpu = 0; cpu < cpu_index; cpu++) |
e15bacbe | 69 | target_cpu = cpumask_next(target_cpu, pd->cpumask.pcpu); |
16295bec SK |
70 | |
71 | return target_cpu; | |
72 | } | |
73 | ||
bfde23ce | 74 | static int padata_cpu_hash(struct parallel_data *pd, unsigned int seq_nr) |
16295bec | 75 | { |
16295bec SK |
76 | /* |
77 | * Hash the sequence numbers to the cpus by taking | |
78 | * seq_nr mod. number of cpus in use. | |
79 | */ | |
bfde23ce | 80 | int cpu_index = seq_nr % cpumask_weight(pd->cpumask.pcpu); |
16295bec SK |
81 | |
82 | return padata_index_to_cpu(pd, cpu_index); | |
83 | } | |
84 | ||
4611ce22 | 85 | static struct padata_work *padata_work_alloc(void) |
16295bec | 86 | { |
4611ce22 | 87 | struct padata_work *pw; |
16295bec | 88 | |
4611ce22 | 89 | lockdep_assert_held(&padata_works_lock); |
16295bec | 90 | |
4611ce22 DJ |
91 | if (list_empty(&padata_free_works)) |
92 | return NULL; /* No more work items allowed to be queued. */ | |
16295bec | 93 | |
4611ce22 DJ |
94 | pw = list_first_entry(&padata_free_works, struct padata_work, pw_list); |
95 | list_del(&pw->pw_list); | |
96 | return pw; | |
97 | } | |
16295bec | 98 | |
4611ce22 | 99 | static void padata_work_init(struct padata_work *pw, work_func_t work_fn, |
004ed426 | 100 | void *data, int flags) |
4611ce22 | 101 | { |
004ed426 DJ |
102 | if (flags & PADATA_WORK_ONSTACK) |
103 | INIT_WORK_ONSTACK(&pw->pw_work, work_fn); | |
104 | else | |
105 | INIT_WORK(&pw->pw_work, work_fn); | |
4611ce22 DJ |
106 | pw->pw_data = data; |
107 | } | |
16295bec | 108 | |
004ed426 DJ |
109 | static int __init padata_work_alloc_mt(int nworks, void *data, |
110 | struct list_head *head) | |
111 | { | |
112 | int i; | |
113 | ||
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(); | |
118 | ||
119 | if (!pw) | |
120 | break; | |
121 | padata_work_init(pw, padata_mt_helper, data, 0); | |
122 | list_add(&pw->pw_list, head); | |
123 | } | |
124 | spin_unlock(&padata_works_lock); | |
125 | ||
126 | return i; | |
127 | } | |
128 | ||
4611ce22 DJ |
129 | static void padata_work_free(struct padata_work *pw) |
130 | { | |
131 | lockdep_assert_held(&padata_works_lock); | |
132 | list_add(&pw->pw_list, &padata_free_works); | |
133 | } | |
16295bec | 134 | |
004ed426 DJ |
135 | static void __init padata_works_free(struct list_head *works) |
136 | { | |
137 | struct padata_work *cur, *next; | |
138 | ||
139 | if (list_empty(works)) | |
140 | return; | |
141 | ||
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); | |
146 | } | |
147 | spin_unlock(&padata_works_lock); | |
148 | } | |
149 | ||
4611ce22 DJ |
150 | static void padata_parallel_worker(struct work_struct *parallel_work) |
151 | { | |
152 | struct padata_work *pw = container_of(parallel_work, struct padata_work, | |
153 | pw_work); | |
154 | struct padata_priv *padata = pw->pw_data; | |
16295bec | 155 | |
4611ce22 DJ |
156 | local_bh_disable(); |
157 | padata->parallel(padata); | |
158 | spin_lock(&padata_works_lock); | |
159 | padata_work_free(pw); | |
160 | spin_unlock(&padata_works_lock); | |
16295bec SK |
161 | local_bh_enable(); |
162 | } | |
163 | ||
0198ffd1 | 164 | /** |
16295bec SK |
165 | * padata_do_parallel - padata parallelization function |
166 | * | |
bbefa1dd | 167 | * @ps: padatashell |
16295bec | 168 | * @padata: object to be parallelized |
e6ce0e08 DJ |
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. | |
16295bec SK |
173 | * |
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. | |
bfcdcef8 DJ |
177 | * |
178 | * Return: 0 on success or else negative error code. | |
16295bec | 179 | */ |
bbefa1dd | 180 | int padata_do_parallel(struct padata_shell *ps, |
e6ce0e08 | 181 | struct padata_priv *padata, int *cb_cpu) |
16295bec | 182 | { |
bbefa1dd | 183 | struct padata_instance *pinst = ps->pinst; |
4611ce22 | 184 | int i, cpu, cpu_index, err; |
16295bec | 185 | struct parallel_data *pd; |
4611ce22 | 186 | struct padata_work *pw; |
16295bec SK |
187 | |
188 | rcu_read_lock_bh(); | |
189 | ||
bbefa1dd | 190 | pd = rcu_dereference_bh(ps->pd); |
16295bec | 191 | |
83f619f3 | 192 | err = -EINVAL; |
7424713b | 193 | if (!(pinst->flags & PADATA_INIT) || pinst->flags & PADATA_INVALID) |
16295bec SK |
194 | goto out; |
195 | ||
e6ce0e08 DJ |
196 | if (!cpumask_test_cpu(*cb_cpu, pd->cpumask.cbcpu)) { |
197 | if (!cpumask_weight(pd->cpumask.cbcpu)) | |
198 | goto out; | |
199 | ||
200 | /* Select an alternate fallback CPU and notify the caller. */ | |
201 | cpu_index = *cb_cpu % cpumask_weight(pd->cpumask.cbcpu); | |
202 | ||
203 | cpu = cpumask_first(pd->cpumask.cbcpu); | |
204 | for (i = 0; i < cpu_index; i++) | |
205 | cpu = cpumask_next(cpu, pd->cpumask.cbcpu); | |
206 | ||
207 | *cb_cpu = cpu; | |
208 | } | |
16295bec SK |
209 | |
210 | err = -EBUSY; | |
211 | if ((pinst->flags & PADATA_RESET)) | |
212 | goto out; | |
213 | ||
16295bec SK |
214 | atomic_inc(&pd->refcnt); |
215 | padata->pd = pd; | |
e6ce0e08 | 216 | padata->cb_cpu = *cb_cpu; |
16295bec | 217 | |
4611ce22 | 218 | rcu_read_unlock_bh(); |
16295bec | 219 | |
4611ce22 DJ |
220 | spin_lock(&padata_works_lock); |
221 | padata->seq_nr = ++pd->seq_nr; | |
222 | pw = padata_work_alloc(); | |
223 | spin_unlock(&padata_works_lock); | |
224 | if (pw) { | |
004ed426 | 225 | padata_work_init(pw, padata_parallel_worker, padata, 0); |
4611ce22 DJ |
226 | queue_work(pinst->parallel_wq, &pw->pw_work); |
227 | } else { | |
228 | /* Maximum works limit exceeded, run in the current task. */ | |
229 | padata->parallel(padata); | |
230 | } | |
16295bec | 231 | |
4611ce22 | 232 | return 0; |
16295bec SK |
233 | out: |
234 | rcu_read_unlock_bh(); | |
235 | ||
236 | return err; | |
237 | } | |
238 | EXPORT_SYMBOL(padata_do_parallel); | |
239 | ||
0198ffd1 | 240 | /* |
bfde23ce | 241 | * padata_find_next - Find the next object that needs serialization. |
0198ffd1 | 242 | * |
bfcdcef8 DJ |
243 | * Return: |
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. | |
0198ffd1 | 249 | */ |
bfde23ce DJ |
250 | static struct padata_priv *padata_find_next(struct parallel_data *pd, |
251 | bool remove_object) | |
16295bec | 252 | { |
f0fcf200 | 253 | struct padata_parallel_queue *next_queue; |
16295bec SK |
254 | struct padata_priv *padata; |
255 | struct padata_list *reorder; | |
6fc4dbcf | 256 | int cpu = pd->cpu; |
16295bec | 257 | |
e15bacbe | 258 | next_queue = per_cpu_ptr(pd->pqueue, cpu); |
16295bec SK |
259 | reorder = &next_queue->reorder; |
260 | ||
de5540d0 | 261 | spin_lock(&reorder->lock); |
bfde23ce DJ |
262 | if (list_empty(&reorder->list)) { |
263 | spin_unlock(&reorder->lock); | |
264 | return NULL; | |
265 | } | |
16295bec | 266 | |
bfde23ce | 267 | padata = list_entry(reorder->list.next, struct padata_priv, list); |
16295bec | 268 | |
bfde23ce DJ |
269 | /* |
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. | |
272 | */ | |
273 | if (padata->seq_nr != pd->processed) { | |
de5540d0 | 274 | spin_unlock(&reorder->lock); |
bfde23ce | 275 | return NULL; |
16295bec SK |
276 | } |
277 | ||
bfde23ce DJ |
278 | if (remove_object) { |
279 | list_del_init(&padata->list); | |
bfde23ce DJ |
280 | ++pd->processed; |
281 | pd->cpu = cpumask_next_wrap(cpu, pd->cpumask.pcpu, -1, false); | |
16295bec SK |
282 | } |
283 | ||
bfde23ce | 284 | spin_unlock(&reorder->lock); |
16295bec SK |
285 | return padata; |
286 | } | |
287 | ||
288 | static void padata_reorder(struct parallel_data *pd) | |
289 | { | |
bbefa1dd | 290 | struct padata_instance *pinst = pd->ps->pinst; |
3047817b | 291 | int cb_cpu; |
16295bec | 292 | struct padata_priv *padata; |
e15bacbe | 293 | struct padata_serial_queue *squeue; |
6fc4dbcf | 294 | struct padata_parallel_queue *next_queue; |
16295bec | 295 | |
0198ffd1 SK |
296 | /* |
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. | |
305 | */ | |
16295bec | 306 | if (!spin_trylock_bh(&pd->lock)) |
d46a5ac7 | 307 | return; |
16295bec SK |
308 | |
309 | while (1) { | |
bfde23ce | 310 | padata = padata_find_next(pd, true); |
16295bec | 311 | |
0198ffd1 | 312 | /* |
69b34844 JD |
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. | |
0198ffd1 | 316 | */ |
bfde23ce | 317 | if (!padata) |
16295bec SK |
318 | break; |
319 | ||
3047817b SK |
320 | cb_cpu = padata->cb_cpu; |
321 | squeue = per_cpu_ptr(pd->squeue, cb_cpu); | |
16295bec | 322 | |
e15bacbe DK |
323 | spin_lock(&squeue->serial.lock); |
324 | list_add_tail(&padata->list, &squeue->serial.list); | |
325 | spin_unlock(&squeue->serial.lock); | |
16295bec | 326 | |
45d153c0 | 327 | queue_work_on(cb_cpu, pinst->serial_wq, &squeue->work); |
16295bec SK |
328 | } |
329 | ||
330 | spin_unlock_bh(&pd->lock); | |
331 | ||
0198ffd1 SK |
332 | /* |
333 | * The next object that needs serialization might have arrived to | |
6fc4dbcf | 334 | * the reorder queues in the meantime. |
cf144f81 | 335 | * |
6fc4dbcf HX |
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 | |
cf144f81 | 338 | * smp_mb__after_atomic in padata_do_serial. |
0198ffd1 | 339 | */ |
cf144f81 | 340 | smp_mb(); |
16295bec | 341 | |
6fc4dbcf | 342 | next_queue = per_cpu_ptr(pd->pqueue, pd->cpu); |
bfde23ce DJ |
343 | if (!list_empty(&next_queue->reorder.list) && |
344 | padata_find_next(pd, false)) | |
45d153c0 | 345 | queue_work(pinst->serial_wq, &pd->reorder_work); |
16295bec SK |
346 | } |
347 | ||
cf5868c8 MK |
348 | static void invoke_padata_reorder(struct work_struct *work) |
349 | { | |
cf5868c8 MK |
350 | struct parallel_data *pd; |
351 | ||
352 | local_bh_disable(); | |
6fc4dbcf | 353 | pd = container_of(work, struct parallel_data, reorder_work); |
cf5868c8 MK |
354 | padata_reorder(pd); |
355 | local_bh_enable(); | |
356 | } | |
357 | ||
e15bacbe | 358 | static void padata_serial_worker(struct work_struct *serial_work) |
16295bec | 359 | { |
e15bacbe | 360 | struct padata_serial_queue *squeue; |
16295bec SK |
361 | struct parallel_data *pd; |
362 | LIST_HEAD(local_list); | |
07928d9b | 363 | int cnt; |
16295bec SK |
364 | |
365 | local_bh_disable(); | |
e15bacbe DK |
366 | squeue = container_of(serial_work, struct padata_serial_queue, work); |
367 | pd = squeue->pd; | |
16295bec | 368 | |
e15bacbe DK |
369 | spin_lock(&squeue->serial.lock); |
370 | list_replace_init(&squeue->serial.list, &local_list); | |
371 | spin_unlock(&squeue->serial.lock); | |
16295bec | 372 | |
07928d9b HX |
373 | cnt = 0; |
374 | ||
16295bec SK |
375 | while (!list_empty(&local_list)) { |
376 | struct padata_priv *padata; | |
377 | ||
378 | padata = list_entry(local_list.next, | |
379 | struct padata_priv, list); | |
380 | ||
381 | list_del_init(&padata->list); | |
382 | ||
383 | padata->serial(padata); | |
07928d9b | 384 | cnt++; |
16295bec SK |
385 | } |
386 | local_bh_enable(); | |
07928d9b HX |
387 | |
388 | if (atomic_sub_and_test(cnt, &pd->refcnt)) | |
389 | padata_free_pd(pd); | |
16295bec SK |
390 | } |
391 | ||
0198ffd1 | 392 | /** |
16295bec SK |
393 | * padata_do_serial - padata serialization function |
394 | * | |
395 | * @padata: object to be serialized. | |
396 | * | |
397 | * padata_do_serial must be called for every parallelized object. | |
398 | * The serialization callback function will run with BHs off. | |
399 | */ | |
400 | void padata_do_serial(struct padata_priv *padata) | |
401 | { | |
065cf577 | 402 | struct parallel_data *pd = padata->pd; |
4611ce22 | 403 | int hashed_cpu = padata_cpu_hash(pd, padata->seq_nr); |
065cf577 | 404 | struct padata_parallel_queue *pqueue = per_cpu_ptr(pd->pqueue, |
4611ce22 | 405 | hashed_cpu); |
bfde23ce | 406 | struct padata_priv *cur; |
16295bec | 407 | |
e15bacbe | 408 | spin_lock(&pqueue->reorder.lock); |
bfde23ce DJ |
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) | |
412 | break; | |
413 | list_add(&padata->list, &cur->list); | |
e15bacbe | 414 | spin_unlock(&pqueue->reorder.lock); |
16295bec | 415 | |
cf144f81 | 416 | /* |
6fc4dbcf | 417 | * Ensure the addition to the reorder list is ordered correctly |
cf144f81 DJ |
418 | * with the trylock of pd->lock in padata_reorder. Pairs with smp_mb |
419 | * in padata_reorder. | |
420 | */ | |
421 | smp_mb__after_atomic(); | |
422 | ||
6fc4dbcf | 423 | padata_reorder(pd); |
16295bec SK |
424 | } |
425 | EXPORT_SYMBOL(padata_do_serial); | |
426 | ||
bbefa1dd | 427 | static int padata_setup_cpumasks(struct padata_instance *pinst) |
16295bec | 428 | { |
bfde23ce | 429 | struct workqueue_attrs *attrs; |
bbefa1dd HX |
430 | int err; |
431 | ||
432 | attrs = alloc_workqueue_attrs(); | |
433 | if (!attrs) | |
434 | return -ENOMEM; | |
435 | ||
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); | |
440 | ||
441 | return err; | |
442 | } | |
443 | ||
444 | static int pd_setup_cpumasks(struct parallel_data *pd, | |
445 | const struct cpumask *pcpumask, | |
446 | const struct cpumask *cbcpumask) | |
447 | { | |
bfde23ce | 448 | int err = -ENOMEM; |
16295bec | 449 | |
bfde23ce DJ |
450 | if (!alloc_cpumask_var(&pd->cpumask.pcpu, GFP_KERNEL)) |
451 | goto out; | |
bfde23ce DJ |
452 | if (!alloc_cpumask_var(&pd->cpumask.cbcpu, GFP_KERNEL)) |
453 | goto free_pcpu_mask; | |
bfde23ce | 454 | |
bbefa1dd HX |
455 | cpumask_copy(pd->cpumask.pcpu, pcpumask); |
456 | cpumask_copy(pd->cpumask.cbcpu, cbcpumask); | |
bfde23ce | 457 | |
e15bacbe | 458 | return 0; |
bfde23ce | 459 | |
bfde23ce DJ |
460 | free_pcpu_mask: |
461 | free_cpumask_var(pd->cpumask.pcpu); | |
462 | out: | |
463 | return err; | |
e15bacbe | 464 | } |
16295bec | 465 | |
004ed426 DJ |
466 | static void __init padata_mt_helper(struct work_struct *w) |
467 | { | |
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; | |
471 | bool done; | |
472 | ||
473 | spin_lock(&ps->lock); | |
474 | ||
475 | while (job->size > 0) { | |
476 | unsigned long start, size, end; | |
477 | ||
478 | start = job->start; | |
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); | |
482 | end = start + size; | |
483 | ||
484 | job->start = end; | |
485 | job->size -= size; | |
486 | ||
487 | spin_unlock(&ps->lock); | |
488 | job->thread_fn(start, end, job->fn_arg); | |
489 | spin_lock(&ps->lock); | |
490 | } | |
491 | ||
492 | ++ps->nworks_fini; | |
493 | done = (ps->nworks_fini == ps->nworks); | |
494 | spin_unlock(&ps->lock); | |
495 | ||
496 | if (done) | |
497 | complete(&ps->completion); | |
498 | } | |
499 | ||
500 | /** | |
501 | * padata_do_multithreaded - run a multithreaded job | |
502 | * @job: Description of the job. | |
503 | * | |
504 | * See the definition of struct padata_mt_job for more details. | |
505 | */ | |
506 | void __init padata_do_multithreaded(struct padata_mt_job *job) | |
507 | { | |
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; | |
512 | LIST_HEAD(works); | |
513 | int nworks; | |
514 | ||
515 | if (job->size == 0) | |
516 | return; | |
517 | ||
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); | |
521 | ||
522 | if (nworks == 1) { | |
523 | /* Single thread, no coordination needed, cut to the chase. */ | |
524 | job->thread_fn(job->start, job->start + job->size, job->fn_arg); | |
525 | return; | |
526 | } | |
527 | ||
528 | spin_lock_init(&ps.lock); | |
529 | init_completion(&ps.completion); | |
530 | ps.job = job; | |
531 | ps.nworks = padata_work_alloc_mt(nworks, &ps, &works); | |
532 | ps.nworks_fini = 0; | |
533 | ||
534 | /* | |
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. | |
539 | */ | |
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); | |
543 | ||
544 | list_for_each_entry(pw, &works, pw_list) | |
545 | queue_work(system_unbound_wq, &pw->pw_work); | |
546 | ||
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); | |
550 | ||
551 | /* Wait for all the helpers to finish. */ | |
552 | wait_for_completion(&ps.completion); | |
553 | ||
554 | destroy_work_on_stack(&my_work.pw_work); | |
555 | padata_works_free(&works); | |
556 | } | |
557 | ||
e15bacbe DK |
558 | static void __padata_list_init(struct padata_list *pd_list) |
559 | { | |
560 | INIT_LIST_HEAD(&pd_list->list); | |
561 | spin_lock_init(&pd_list->lock); | |
562 | } | |
16295bec | 563 | |
e15bacbe DK |
564 | /* Initialize all percpu queues used by serial workers */ |
565 | static void padata_init_squeues(struct parallel_data *pd) | |
566 | { | |
567 | int cpu; | |
568 | struct padata_serial_queue *squeue; | |
7b389b2c | 569 | |
e15bacbe DK |
570 | for_each_cpu(cpu, pd->cpumask.cbcpu) { |
571 | squeue = per_cpu_ptr(pd->squeue, cpu); | |
572 | squeue->pd = pd; | |
573 | __padata_list_init(&squeue->serial); | |
574 | INIT_WORK(&squeue->work, padata_serial_worker); | |
575 | } | |
576 | } | |
16295bec | 577 | |
e15bacbe DK |
578 | /* Initialize all percpu queues used by parallel workers */ |
579 | static void padata_init_pqueues(struct parallel_data *pd) | |
580 | { | |
c51636a3 | 581 | int cpu; |
e15bacbe | 582 | struct padata_parallel_queue *pqueue; |
16295bec | 583 | |
c51636a3 | 584 | for_each_cpu(cpu, pd->cpumask.pcpu) { |
e15bacbe | 585 | pqueue = per_cpu_ptr(pd->pqueue, cpu); |
1bd845bc | 586 | |
e15bacbe | 587 | __padata_list_init(&pqueue->reorder); |
e15bacbe | 588 | atomic_set(&pqueue->num_obj, 0); |
16295bec | 589 | } |
e15bacbe | 590 | } |
16295bec | 591 | |
e15bacbe | 592 | /* Allocate and initialize the internal cpumask dependend resources. */ |
bbefa1dd | 593 | static struct parallel_data *padata_alloc_pd(struct padata_shell *ps) |
e15bacbe | 594 | { |
bbefa1dd HX |
595 | struct padata_instance *pinst = ps->pinst; |
596 | const struct cpumask *cbcpumask; | |
597 | const struct cpumask *pcpumask; | |
e15bacbe | 598 | struct parallel_data *pd; |
16295bec | 599 | |
bbefa1dd HX |
600 | cbcpumask = pinst->rcpumask.cbcpu; |
601 | pcpumask = pinst->rcpumask.pcpu; | |
602 | ||
e15bacbe DK |
603 | pd = kzalloc(sizeof(struct parallel_data), GFP_KERNEL); |
604 | if (!pd) | |
605 | goto err; | |
16295bec | 606 | |
e15bacbe DK |
607 | pd->pqueue = alloc_percpu(struct padata_parallel_queue); |
608 | if (!pd->pqueue) | |
609 | goto err_free_pd; | |
610 | ||
611 | pd->squeue = alloc_percpu(struct padata_serial_queue); | |
612 | if (!pd->squeue) | |
613 | goto err_free_pqueue; | |
bfde23ce | 614 | |
bbefa1dd HX |
615 | pd->ps = ps; |
616 | if (pd_setup_cpumasks(pd, pcpumask, cbcpumask)) | |
e15bacbe | 617 | goto err_free_squeue; |
16295bec | 618 | |
e15bacbe DK |
619 | padata_init_pqueues(pd); |
620 | padata_init_squeues(pd); | |
4611ce22 | 621 | pd->seq_nr = -1; |
07928d9b | 622 | atomic_set(&pd->refcnt, 1); |
16295bec | 623 | spin_lock_init(&pd->lock); |
ec9c7d19 | 624 | pd->cpu = cpumask_first(pd->cpumask.pcpu); |
6fc4dbcf | 625 | INIT_WORK(&pd->reorder_work, invoke_padata_reorder); |
16295bec SK |
626 | |
627 | return pd; | |
628 | ||
e15bacbe DK |
629 | err_free_squeue: |
630 | free_percpu(pd->squeue); | |
631 | err_free_pqueue: | |
632 | free_percpu(pd->pqueue); | |
16295bec SK |
633 | err_free_pd: |
634 | kfree(pd); | |
635 | err: | |
636 | return NULL; | |
637 | } | |
638 | ||
639 | static void padata_free_pd(struct parallel_data *pd) | |
640 | { | |
e15bacbe DK |
641 | free_cpumask_var(pd->cpumask.pcpu); |
642 | free_cpumask_var(pd->cpumask.cbcpu); | |
643 | free_percpu(pd->pqueue); | |
644 | free_percpu(pd->squeue); | |
16295bec SK |
645 | kfree(pd); |
646 | } | |
647 | ||
4c879170 SK |
648 | static void __padata_start(struct padata_instance *pinst) |
649 | { | |
650 | pinst->flags |= PADATA_INIT; | |
651 | } | |
652 | ||
ee836555 SK |
653 | static void __padata_stop(struct padata_instance *pinst) |
654 | { | |
655 | if (!(pinst->flags & PADATA_INIT)) | |
656 | return; | |
657 | ||
658 | pinst->flags &= ~PADATA_INIT; | |
659 | ||
660 | synchronize_rcu(); | |
ee836555 SK |
661 | } |
662 | ||
25985edc | 663 | /* Replace the internal control structure with a new one. */ |
bbefa1dd | 664 | static int padata_replace_one(struct padata_shell *ps) |
16295bec | 665 | { |
bbefa1dd | 666 | struct parallel_data *pd_new; |
16295bec | 667 | |
bbefa1dd HX |
668 | pd_new = padata_alloc_pd(ps); |
669 | if (!pd_new) | |
670 | return -ENOMEM; | |
16295bec | 671 | |
bbefa1dd HX |
672 | ps->opd = rcu_dereference_protected(ps->pd, 1); |
673 | rcu_assign_pointer(ps->pd, pd_new); | |
16295bec | 674 | |
bbefa1dd HX |
675 | return 0; |
676 | } | |
677 | ||
894c9ef9 | 678 | static int padata_replace(struct padata_instance *pinst) |
bbefa1dd | 679 | { |
bbefa1dd | 680 | struct padata_shell *ps; |
41ccdbfd | 681 | int err = 0; |
bbefa1dd HX |
682 | |
683 | pinst->flags |= PADATA_RESET; | |
16295bec | 684 | |
bbefa1dd HX |
685 | cpumask_and(pinst->rcpumask.pcpu, pinst->cpumask.pcpu, |
686 | cpu_online_mask); | |
bbefa1dd | 687 | |
bbefa1dd HX |
688 | cpumask_and(pinst->rcpumask.cbcpu, pinst->cpumask.cbcpu, |
689 | cpu_online_mask); | |
e15bacbe | 690 | |
bbefa1dd HX |
691 | list_for_each_entry(ps, &pinst->pslist, list) { |
692 | err = padata_replace_one(ps); | |
693 | if (err) | |
694 | break; | |
695 | } | |
696 | ||
697 | synchronize_rcu(); | |
698 | ||
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); | |
16295bec SK |
702 | |
703 | pinst->flags &= ~PADATA_RESET; | |
bbefa1dd HX |
704 | |
705 | return err; | |
16295bec SK |
706 | } |
707 | ||
33e54450 SK |
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) | |
711 | { | |
13614e0f | 712 | if (!cpumask_intersects(cpumask, cpu_online_mask)) { |
33e54450 SK |
713 | pinst->flags |= PADATA_INVALID; |
714 | return false; | |
715 | } | |
716 | ||
717 | pinst->flags &= ~PADATA_INVALID; | |
718 | return true; | |
719 | } | |
720 | ||
65ff577e SK |
721 | static int __padata_set_cpumasks(struct padata_instance *pinst, |
722 | cpumask_var_t pcpumask, | |
723 | cpumask_var_t cbcpumask) | |
724 | { | |
725 | int valid; | |
bbefa1dd | 726 | int err; |
65ff577e SK |
727 | |
728 | valid = padata_validate_cpumask(pinst, pcpumask); | |
729 | if (!valid) { | |
730 | __padata_stop(pinst); | |
731 | goto out_replace; | |
732 | } | |
733 | ||
734 | valid = padata_validate_cpumask(pinst, cbcpumask); | |
735 | if (!valid) | |
736 | __padata_stop(pinst); | |
737 | ||
738 | out_replace: | |
65ff577e SK |
739 | cpumask_copy(pinst->cpumask.pcpu, pcpumask); |
740 | cpumask_copy(pinst->cpumask.cbcpu, cbcpumask); | |
741 | ||
894c9ef9 | 742 | err = padata_setup_cpumasks(pinst) ?: padata_replace(pinst); |
65ff577e SK |
743 | |
744 | if (valid) | |
745 | __padata_start(pinst); | |
746 | ||
bbefa1dd | 747 | return err; |
65ff577e SK |
748 | } |
749 | ||
e15bacbe | 750 | /** |
bfcdcef8 DJ |
751 | * padata_set_cpumask - Sets specified by @cpumask_type cpumask to the value |
752 | * equivalent to @cpumask. | |
16295bec | 753 | * @pinst: padata instance |
e15bacbe DK |
754 | * @cpumask_type: PADATA_CPU_SERIAL or PADATA_CPU_PARALLEL corresponding |
755 | * to parallel and serial cpumasks respectively. | |
16295bec | 756 | * @cpumask: the cpumask to use |
bfcdcef8 DJ |
757 | * |
758 | * Return: 0 on success or negative error code | |
16295bec | 759 | */ |
e15bacbe DK |
760 | int padata_set_cpumask(struct padata_instance *pinst, int cpumask_type, |
761 | cpumask_var_t cpumask) | |
762 | { | |
763 | struct cpumask *serial_mask, *parallel_mask; | |
65ff577e SK |
764 | int err = -EINVAL; |
765 | ||
6751fb3c | 766 | get_online_cpus(); |
38228e88 | 767 | mutex_lock(&pinst->lock); |
6751fb3c | 768 | |
e15bacbe DK |
769 | switch (cpumask_type) { |
770 | case PADATA_CPU_PARALLEL: | |
771 | serial_mask = pinst->cpumask.cbcpu; | |
772 | parallel_mask = cpumask; | |
773 | break; | |
774 | case PADATA_CPU_SERIAL: | |
775 | parallel_mask = pinst->cpumask.pcpu; | |
776 | serial_mask = cpumask; | |
777 | break; | |
778 | default: | |
65ff577e | 779 | goto out; |
16295bec SK |
780 | } |
781 | ||
65ff577e | 782 | err = __padata_set_cpumasks(pinst, parallel_mask, serial_mask); |
16295bec SK |
783 | |
784 | out: | |
785 | mutex_unlock(&pinst->lock); | |
38228e88 | 786 | put_online_cpus(); |
16295bec SK |
787 | |
788 | return err; | |
789 | } | |
790 | EXPORT_SYMBOL(padata_set_cpumask); | |
791 | ||
19d795b6 AB |
792 | /** |
793 | * padata_start - start the parallel processing | |
794 | * | |
795 | * @pinst: padata instance to start | |
bfcdcef8 DJ |
796 | * |
797 | * Return: 0 on success or negative error code | |
19d795b6 AB |
798 | */ |
799 | int padata_start(struct padata_instance *pinst) | |
800 | { | |
801 | int err = 0; | |
802 | ||
803 | mutex_lock(&pinst->lock); | |
804 | ||
805 | if (pinst->flags & PADATA_INVALID) | |
806 | err = -EINVAL; | |
807 | ||
8ddab428 | 808 | __padata_start(pinst); |
19d795b6 AB |
809 | |
810 | mutex_unlock(&pinst->lock); | |
811 | ||
812 | return err; | |
813 | } | |
814 | EXPORT_SYMBOL(padata_start); | |
815 | ||
816 | /** | |
817 | * padata_stop - stop the parallel processing | |
818 | * | |
819 | * @pinst: padata instance to stop | |
820 | */ | |
821 | void padata_stop(struct padata_instance *pinst) | |
822 | { | |
823 | mutex_lock(&pinst->lock); | |
824 | __padata_stop(pinst); | |
825 | mutex_unlock(&pinst->lock); | |
826 | } | |
827 | EXPORT_SYMBOL(padata_stop); | |
828 | ||
829 | #ifdef CONFIG_HOTPLUG_CPU | |
830 | ||
16295bec SK |
831 | static int __padata_add_cpu(struct padata_instance *pinst, int cpu) |
832 | { | |
bbefa1dd | 833 | int err = 0; |
16295bec | 834 | |
13614e0f | 835 | if (cpumask_test_cpu(cpu, cpu_online_mask)) { |
894c9ef9 | 836 | err = padata_replace(pinst); |
33e54450 | 837 | |
e15bacbe DK |
838 | if (padata_validate_cpumask(pinst, pinst->cpumask.pcpu) && |
839 | padata_validate_cpumask(pinst, pinst->cpumask.cbcpu)) | |
33e54450 | 840 | __padata_start(pinst); |
16295bec SK |
841 | } |
842 | ||
bbefa1dd | 843 | return err; |
16295bec SK |
844 | } |
845 | ||
16295bec SK |
846 | static int __padata_remove_cpu(struct padata_instance *pinst, int cpu) |
847 | { | |
bbefa1dd | 848 | int err = 0; |
16295bec | 849 | |
894c9ef9 | 850 | if (!cpumask_test_cpu(cpu, cpu_online_mask)) { |
e15bacbe | 851 | if (!padata_validate_cpumask(pinst, pinst->cpumask.pcpu) || |
b89661df | 852 | !padata_validate_cpumask(pinst, pinst->cpumask.cbcpu)) |
33e54450 | 853 | __padata_stop(pinst); |
33e54450 | 854 | |
894c9ef9 | 855 | err = padata_replace(pinst); |
16295bec SK |
856 | } |
857 | ||
bbefa1dd | 858 | return err; |
16295bec SK |
859 | } |
860 | ||
e15bacbe DK |
861 | static inline int pinst_has_cpu(struct padata_instance *pinst, int cpu) |
862 | { | |
863 | return cpumask_test_cpu(cpu, pinst->cpumask.pcpu) || | |
864 | cpumask_test_cpu(cpu, pinst->cpumask.cbcpu); | |
865 | } | |
866 | ||
30e92153 | 867 | static int padata_cpu_online(unsigned int cpu, struct hlist_node *node) |
16295bec | 868 | { |
16295bec | 869 | struct padata_instance *pinst; |
30e92153 | 870 | int ret; |
16295bec | 871 | |
3c2214b6 | 872 | pinst = hlist_entry_safe(node, struct padata_instance, cpu_online_node); |
30e92153 SAS |
873 | if (!pinst_has_cpu(pinst, cpu)) |
874 | return 0; | |
16295bec | 875 | |
30e92153 SAS |
876 | mutex_lock(&pinst->lock); |
877 | ret = __padata_add_cpu(pinst, cpu); | |
878 | mutex_unlock(&pinst->lock); | |
879 | return ret; | |
880 | } | |
16295bec | 881 | |
894c9ef9 | 882 | static int padata_cpu_dead(unsigned int cpu, struct hlist_node *node) |
30e92153 SAS |
883 | { |
884 | struct padata_instance *pinst; | |
885 | int ret; | |
886 | ||
3c2214b6 | 887 | pinst = hlist_entry_safe(node, struct padata_instance, cpu_dead_node); |
30e92153 SAS |
888 | if (!pinst_has_cpu(pinst, cpu)) |
889 | return 0; | |
16295bec | 890 | |
30e92153 SAS |
891 | mutex_lock(&pinst->lock); |
892 | ret = __padata_remove_cpu(pinst, cpu); | |
893 | mutex_unlock(&pinst->lock); | |
894 | return ret; | |
16295bec | 895 | } |
30e92153 SAS |
896 | |
897 | static enum cpuhp_state hp_online; | |
e2cb2f1c | 898 | #endif |
16295bec | 899 | |
5e017dc3 DK |
900 | static void __padata_free(struct padata_instance *pinst) |
901 | { | |
902 | #ifdef CONFIG_HOTPLUG_CPU | |
3c2214b6 DJ |
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); | |
5e017dc3 DK |
906 | #endif |
907 | ||
bbefa1dd HX |
908 | WARN_ON(!list_empty(&pinst->pslist)); |
909 | ||
5e017dc3 | 910 | padata_stop(pinst); |
bbefa1dd HX |
911 | free_cpumask_var(pinst->rcpumask.cbcpu); |
912 | free_cpumask_var(pinst->rcpumask.pcpu); | |
5e017dc3 DK |
913 | free_cpumask_var(pinst->cpumask.pcpu); |
914 | free_cpumask_var(pinst->cpumask.cbcpu); | |
45d153c0 DJ |
915 | destroy_workqueue(pinst->serial_wq); |
916 | destroy_workqueue(pinst->parallel_wq); | |
5e017dc3 DK |
917 | kfree(pinst); |
918 | } | |
919 | ||
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) | |
924 | ||
925 | static void padata_sysfs_release(struct kobject *kobj) | |
926 | { | |
927 | struct padata_instance *pinst = kobj2pinst(kobj); | |
928 | __padata_free(pinst); | |
929 | } | |
930 | ||
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); | |
936 | }; | |
937 | ||
938 | static ssize_t show_cpumask(struct padata_instance *pinst, | |
939 | struct attribute *attr, char *buf) | |
940 | { | |
941 | struct cpumask *cpumask; | |
942 | ssize_t len; | |
943 | ||
944 | mutex_lock(&pinst->lock); | |
945 | if (!strcmp(attr->name, "serial_cpumask")) | |
946 | cpumask = pinst->cpumask.cbcpu; | |
947 | else | |
948 | cpumask = pinst->cpumask.pcpu; | |
949 | ||
4497da6f TH |
950 | len = snprintf(buf, PAGE_SIZE, "%*pb\n", |
951 | nr_cpu_ids, cpumask_bits(cpumask)); | |
5e017dc3 | 952 | mutex_unlock(&pinst->lock); |
4497da6f | 953 | return len < PAGE_SIZE ? len : -EINVAL; |
5e017dc3 DK |
954 | } |
955 | ||
956 | static ssize_t store_cpumask(struct padata_instance *pinst, | |
957 | struct attribute *attr, | |
958 | const char *buf, size_t count) | |
959 | { | |
960 | cpumask_var_t new_cpumask; | |
961 | ssize_t ret; | |
962 | int mask_type; | |
963 | ||
964 | if (!alloc_cpumask_var(&new_cpumask, GFP_KERNEL)) | |
965 | return -ENOMEM; | |
966 | ||
967 | ret = bitmap_parse(buf, count, cpumask_bits(new_cpumask), | |
968 | nr_cpumask_bits); | |
969 | if (ret < 0) | |
970 | goto out; | |
971 | ||
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); | |
975 | if (!ret) | |
976 | ret = count; | |
977 | ||
978 | out: | |
979 | free_cpumask_var(new_cpumask); | |
980 | return ret; | |
981 | } | |
982 | ||
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) | |
989 | ||
990 | PADATA_ATTR_RW(serial_cpumask, show_cpumask, store_cpumask); | |
991 | PADATA_ATTR_RW(parallel_cpumask, show_cpumask, store_cpumask); | |
992 | ||
993 | /* | |
994 | * Padata sysfs provides the following objects: | |
995 | * serial_cpumask [RW] - cpumask for serial workers | |
996 | * parallel_cpumask [RW] - cpumask for parallel workers | |
997 | */ | |
998 | static struct attribute *padata_default_attrs[] = { | |
999 | &serial_cpumask_attr.attr, | |
1000 | ¶llel_cpumask_attr.attr, | |
1001 | NULL, | |
1002 | }; | |
2064fbc7 | 1003 | ATTRIBUTE_GROUPS(padata_default); |
5e017dc3 DK |
1004 | |
1005 | static ssize_t padata_sysfs_show(struct kobject *kobj, | |
1006 | struct attribute *attr, char *buf) | |
1007 | { | |
1008 | struct padata_instance *pinst; | |
1009 | struct padata_sysfs_entry *pentry; | |
1010 | ssize_t ret = -EIO; | |
1011 | ||
1012 | pinst = kobj2pinst(kobj); | |
1013 | pentry = attr2pentry(attr); | |
1014 | if (pentry->show) | |
1015 | ret = pentry->show(pinst, attr, buf); | |
1016 | ||
1017 | return ret; | |
1018 | } | |
1019 | ||
1020 | static ssize_t padata_sysfs_store(struct kobject *kobj, struct attribute *attr, | |
1021 | const char *buf, size_t count) | |
1022 | { | |
1023 | struct padata_instance *pinst; | |
1024 | struct padata_sysfs_entry *pentry; | |
1025 | ssize_t ret = -EIO; | |
1026 | ||
1027 | pinst = kobj2pinst(kobj); | |
1028 | pentry = attr2pentry(attr); | |
1029 | if (pentry->show) | |
1030 | ret = pentry->store(pinst, attr, buf, count); | |
1031 | ||
1032 | return ret; | |
1033 | } | |
1034 | ||
1035 | static const struct sysfs_ops padata_sysfs_ops = { | |
1036 | .show = padata_sysfs_show, | |
1037 | .store = padata_sysfs_store, | |
1038 | }; | |
1039 | ||
1040 | static struct kobj_type padata_attr_type = { | |
1041 | .sysfs_ops = &padata_sysfs_ops, | |
2064fbc7 | 1042 | .default_groups = padata_default_groups, |
5e017dc3 DK |
1043 | .release = padata_sysfs_release, |
1044 | }; | |
1045 | ||
e15bacbe | 1046 | /** |
e6cc1170 SK |
1047 | * padata_alloc - allocate and initialize a padata instance and specify |
1048 | * cpumasks for serial and parallel workers. | |
16295bec | 1049 | * |
b128a304 | 1050 | * @name: used to identify the instance |
e15bacbe DK |
1051 | * @pcpumask: cpumask that will be used for padata parallelization |
1052 | * @cbcpumask: cpumask that will be used for padata serialization | |
bfcdcef8 DJ |
1053 | * |
1054 | * Return: new instance on success, NULL on error | |
16295bec | 1055 | */ |
b128a304 | 1056 | static struct padata_instance *padata_alloc(const char *name, |
9596695e TG |
1057 | const struct cpumask *pcpumask, |
1058 | const struct cpumask *cbcpumask) | |
16295bec | 1059 | { |
16295bec | 1060 | struct padata_instance *pinst; |
16295bec SK |
1061 | |
1062 | pinst = kzalloc(sizeof(struct padata_instance), GFP_KERNEL); | |
1063 | if (!pinst) | |
1064 | goto err; | |
1065 | ||
bfde23ce DJ |
1066 | pinst->parallel_wq = alloc_workqueue("%s_parallel", WQ_UNBOUND, 0, |
1067 | name); | |
45d153c0 | 1068 | if (!pinst->parallel_wq) |
16295bec | 1069 | goto err_free_inst; |
b128a304 | 1070 | |
cc491d8e DJ |
1071 | get_online_cpus(); |
1072 | ||
45d153c0 DJ |
1073 | pinst->serial_wq = alloc_workqueue("%s_serial", WQ_MEM_RECLAIM | |
1074 | WQ_CPU_INTENSIVE, 1, name); | |
1075 | if (!pinst->serial_wq) | |
cc491d8e | 1076 | goto err_put_cpus; |
45d153c0 DJ |
1077 | |
1078 | if (!alloc_cpumask_var(&pinst->cpumask.pcpu, GFP_KERNEL)) | |
1079 | goto err_free_serial_wq; | |
e15bacbe DK |
1080 | if (!alloc_cpumask_var(&pinst->cpumask.cbcpu, GFP_KERNEL)) { |
1081 | free_cpumask_var(pinst->cpumask.pcpu); | |
45d153c0 | 1082 | goto err_free_serial_wq; |
33e54450 | 1083 | } |
e15bacbe DK |
1084 | if (!padata_validate_cpumask(pinst, pcpumask) || |
1085 | !padata_validate_cpumask(pinst, cbcpumask)) | |
1086 | goto err_free_masks; | |
16295bec | 1087 | |
bbefa1dd | 1088 | if (!alloc_cpumask_var(&pinst->rcpumask.pcpu, GFP_KERNEL)) |
e15bacbe | 1089 | goto err_free_masks; |
bbefa1dd HX |
1090 | if (!alloc_cpumask_var(&pinst->rcpumask.cbcpu, GFP_KERNEL)) |
1091 | goto err_free_rcpumask_pcpu; | |
74781387 | 1092 | |
bbefa1dd | 1093 | INIT_LIST_HEAD(&pinst->pslist); |
16295bec | 1094 | |
e15bacbe DK |
1095 | cpumask_copy(pinst->cpumask.pcpu, pcpumask); |
1096 | cpumask_copy(pinst->cpumask.cbcpu, cbcpumask); | |
bbefa1dd HX |
1097 | cpumask_and(pinst->rcpumask.pcpu, pcpumask, cpu_online_mask); |
1098 | cpumask_and(pinst->rcpumask.cbcpu, cbcpumask, cpu_online_mask); | |
1099 | ||
1100 | if (padata_setup_cpumasks(pinst)) | |
91a71d61 | 1101 | goto err_free_rcpumask_cbcpu; |
16295bec SK |
1102 | |
1103 | pinst->flags = 0; | |
1104 | ||
5e017dc3 | 1105 | kobject_init(&pinst->kobj, &padata_attr_type); |
16295bec SK |
1106 | mutex_init(&pinst->lock); |
1107 | ||
b8b4a416 | 1108 | #ifdef CONFIG_HOTPLUG_CPU |
3c2214b6 DJ |
1109 | cpuhp_state_add_instance_nocalls_cpuslocked(hp_online, |
1110 | &pinst->cpu_online_node); | |
894c9ef9 | 1111 | cpuhp_state_add_instance_nocalls_cpuslocked(CPUHP_PADATA_DEAD, |
3c2214b6 | 1112 | &pinst->cpu_dead_node); |
b8b4a416 | 1113 | #endif |
cc491d8e DJ |
1114 | |
1115 | put_online_cpus(); | |
1116 | ||
16295bec SK |
1117 | return pinst; |
1118 | ||
bbefa1dd HX |
1119 | err_free_rcpumask_cbcpu: |
1120 | free_cpumask_var(pinst->rcpumask.cbcpu); | |
1121 | err_free_rcpumask_pcpu: | |
1122 | free_cpumask_var(pinst->rcpumask.pcpu); | |
e15bacbe DK |
1123 | err_free_masks: |
1124 | free_cpumask_var(pinst->cpumask.pcpu); | |
1125 | free_cpumask_var(pinst->cpumask.cbcpu); | |
45d153c0 DJ |
1126 | err_free_serial_wq: |
1127 | destroy_workqueue(pinst->serial_wq); | |
cc491d8e DJ |
1128 | err_put_cpus: |
1129 | put_online_cpus(); | |
45d153c0 | 1130 | destroy_workqueue(pinst->parallel_wq); |
16295bec SK |
1131 | err_free_inst: |
1132 | kfree(pinst); | |
1133 | err: | |
1134 | return NULL; | |
1135 | } | |
16295bec | 1136 | |
9596695e TG |
1137 | /** |
1138 | * padata_alloc_possible - Allocate and initialize padata instance. | |
1139 | * Use the cpu_possible_mask for serial and | |
1140 | * parallel workers. | |
1141 | * | |
b128a304 | 1142 | * @name: used to identify the instance |
bfcdcef8 DJ |
1143 | * |
1144 | * Return: new instance on success, NULL on error | |
9596695e | 1145 | */ |
b128a304 | 1146 | struct padata_instance *padata_alloc_possible(const char *name) |
9596695e | 1147 | { |
b128a304 | 1148 | return padata_alloc(name, cpu_possible_mask, cpu_possible_mask); |
9596695e TG |
1149 | } |
1150 | EXPORT_SYMBOL(padata_alloc_possible); | |
1151 | ||
0198ffd1 | 1152 | /** |
16295bec SK |
1153 | * padata_free - free a padata instance |
1154 | * | |
bfcdcef8 | 1155 | * @pinst: padata instance to free |
16295bec SK |
1156 | */ |
1157 | void padata_free(struct padata_instance *pinst) | |
1158 | { | |
5e017dc3 | 1159 | kobject_put(&pinst->kobj); |
16295bec SK |
1160 | } |
1161 | EXPORT_SYMBOL(padata_free); | |
30e92153 | 1162 | |
bbefa1dd HX |
1163 | /** |
1164 | * padata_alloc_shell - Allocate and initialize padata shell. | |
1165 | * | |
1166 | * @pinst: Parent padata_instance object. | |
bfcdcef8 DJ |
1167 | * |
1168 | * Return: new shell on success, NULL on error | |
bbefa1dd HX |
1169 | */ |
1170 | struct padata_shell *padata_alloc_shell(struct padata_instance *pinst) | |
1171 | { | |
1172 | struct parallel_data *pd; | |
1173 | struct padata_shell *ps; | |
1174 | ||
1175 | ps = kzalloc(sizeof(*ps), GFP_KERNEL); | |
1176 | if (!ps) | |
1177 | goto out; | |
1178 | ||
1179 | ps->pinst = pinst; | |
1180 | ||
1181 | get_online_cpus(); | |
1182 | pd = padata_alloc_pd(ps); | |
1183 | put_online_cpus(); | |
1184 | ||
1185 | if (!pd) | |
1186 | goto out_free_ps; | |
1187 | ||
1188 | mutex_lock(&pinst->lock); | |
1189 | RCU_INIT_POINTER(ps->pd, pd); | |
1190 | list_add(&ps->list, &pinst->pslist); | |
1191 | mutex_unlock(&pinst->lock); | |
1192 | ||
1193 | return ps; | |
1194 | ||
1195 | out_free_ps: | |
1196 | kfree(ps); | |
1197 | out: | |
1198 | return NULL; | |
1199 | } | |
1200 | EXPORT_SYMBOL(padata_alloc_shell); | |
1201 | ||
1202 | /** | |
1203 | * padata_free_shell - free a padata shell | |
1204 | * | |
1205 | * @ps: padata shell to free | |
1206 | */ | |
1207 | void padata_free_shell(struct padata_shell *ps) | |
1208 | { | |
07b24c7c EB |
1209 | if (!ps) |
1210 | return; | |
bbefa1dd | 1211 | |
07b24c7c | 1212 | mutex_lock(&ps->pinst->lock); |
bbefa1dd HX |
1213 | list_del(&ps->list); |
1214 | padata_free_pd(rcu_dereference_protected(ps->pd, 1)); | |
07b24c7c | 1215 | mutex_unlock(&ps->pinst->lock); |
bbefa1dd HX |
1216 | |
1217 | kfree(ps); | |
1218 | } | |
1219 | EXPORT_SYMBOL(padata_free_shell); | |
1220 | ||
f1b192b1 | 1221 | void __init padata_init(void) |
30e92153 | 1222 | { |
4611ce22 | 1223 | unsigned int i, possible_cpus; |
f1b192b1 | 1224 | #ifdef CONFIG_HOTPLUG_CPU |
30e92153 SAS |
1225 | int ret; |
1226 | ||
1227 | ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "padata:online", | |
894c9ef9 | 1228 | padata_cpu_online, NULL); |
30e92153 | 1229 | if (ret < 0) |
f1b192b1 | 1230 | goto err; |
30e92153 | 1231 | hp_online = ret; |
894c9ef9 DJ |
1232 | |
1233 | ret = cpuhp_setup_state_multi(CPUHP_PADATA_DEAD, "padata:dead", | |
1234 | NULL, padata_cpu_dead); | |
4611ce22 DJ |
1235 | if (ret < 0) |
1236 | goto remove_online_state; | |
1237 | #endif | |
1238 | ||
1239 | possible_cpus = num_possible_cpus(); | |
1240 | padata_works = kmalloc_array(possible_cpus, sizeof(struct padata_work), | |
1241 | GFP_KERNEL); | |
1242 | if (!padata_works) | |
1243 | goto remove_dead_state; | |
1244 | ||
1245 | for (i = 0; i < possible_cpus; ++i) | |
1246 | list_add(&padata_works[i].pw_list, &padata_free_works); | |
30e92153 | 1247 | |
f1b192b1 | 1248 | return; |
4611ce22 DJ |
1249 | |
1250 | remove_dead_state: | |
1251 | #ifdef CONFIG_HOTPLUG_CPU | |
1252 | cpuhp_remove_multi_state(CPUHP_PADATA_DEAD); | |
1253 | remove_online_state: | |
1254 | cpuhp_remove_multi_state(hp_online); | |
f1b192b1 | 1255 | err: |
30e92153 | 1256 | #endif |
4611ce22 | 1257 | pr_warn("padata: initialization failed\n"); |
f1b192b1 | 1258 | } |