blob: 53f4bc9127127c287bad71da7d870cbb42c51d6a [file] [log] [blame]
Cheah Kok Cheong08b21fb2017-12-21 19:35:30 +08001// SPDX-License-Identifier: GPL-2.0
Steffen Klassert16295be2010-01-06 19:47:10 +11002/*
3 * padata.c - generic interface to process data streams in parallel
4 *
Daniel Jordanbfcdcef82019-12-03 14:31:14 -05005 * See Documentation/core-api/padata.rst for more information.
Steffen Klassert107f8bd2012-03-28 08:42:34 +02006 *
Steffen Klassert16295be2010-01-06 19:47:10 +11007 * Copyright (C) 2008, 2009 secunet Security Networks AG
8 * Copyright (C) 2008, 2009 Steffen Klassert <steffen.klassert@secunet.com>
9 *
Daniel Jordan004ed422020-06-03 15:59:43 -070010 * Copyright (c) 2020 Oracle and/or its affiliates.
11 * Author: Daniel Jordan <daniel.m.jordan@oracle.com>
Steffen Klassert16295be2010-01-06 19:47:10 +110012 */
13
Daniel Jordan004ed422020-06-03 15:59:43 -070014#include <linux/completion.h>
Paul Gortmaker9984de12011-05-23 14:51:41 -040015#include <linux/export.h>
Steffen Klassert16295be2010-01-06 19:47:10 +110016#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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090022#include <linux/slab.h>
Dan Kruchinin5e017dc2010-07-14 14:33:08 +040023#include <linux/sysfs.h>
Steffen Klassert16295be2010-01-06 19:47:10 +110024#include <linux/rcupdate.h>
25
Daniel Jordan004ed422020-06-03 15:59:43 -070026#define PADATA_WORK_ONSTACK 1 /* Work's memory is on stack */
27
Daniel Jordan4611ce22020-06-03 15:59:39 -070028struct padata_work {
29 struct work_struct pw_work;
30 struct list_head pw_list; /* padata_free_works linkage */
31 void *pw_data;
32};
33
34static DEFINE_SPINLOCK(padata_works_lock);
35static struct padata_work *padata_works;
36static LIST_HEAD(padata_free_works);
Steffen Klassert16295be2010-01-06 19:47:10 +110037
Daniel Jordan004ed422020-06-03 15:59:43 -070038struct 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
Herbert Xu07928d92019-11-19 13:17:31 +080047static void padata_free_pd(struct parallel_data *pd);
Daniel Jordan004ed422020-06-03 15:59:43 -070048static void __init padata_mt_helper(struct work_struct *work);
Herbert Xu07928d92019-11-19 13:17:31 +080049
Steffen Klassert16295be2010-01-06 19:47:10 +110050static int padata_index_to_cpu(struct parallel_data *pd, int cpu_index)
51{
52 int cpu, target_cpu;
53
Dan Kruchinine15bacb2010-07-14 14:31:57 +040054 target_cpu = cpumask_first(pd->cpumask.pcpu);
Steffen Klassert16295be2010-01-06 19:47:10 +110055 for (cpu = 0; cpu < cpu_index; cpu++)
Dan Kruchinine15bacb2010-07-14 14:31:57 +040056 target_cpu = cpumask_next(target_cpu, pd->cpumask.pcpu);
Steffen Klassert16295be2010-01-06 19:47:10 +110057
58 return target_cpu;
59}
60
Daniel Jordanbfde23c2019-09-05 21:40:28 -040061static int padata_cpu_hash(struct parallel_data *pd, unsigned int seq_nr)
Steffen Klassert16295be2010-01-06 19:47:10 +110062{
Steffen Klassert16295be2010-01-06 19:47:10 +110063 /*
64 * Hash the sequence numbers to the cpus by taking
65 * seq_nr mod. number of cpus in use.
66 */
Daniel Jordanbfde23c2019-09-05 21:40:28 -040067 int cpu_index = seq_nr % cpumask_weight(pd->cpumask.pcpu);
Steffen Klassert16295be2010-01-06 19:47:10 +110068
69 return padata_index_to_cpu(pd, cpu_index);
70}
71
Daniel Jordan4611ce22020-06-03 15:59:39 -070072static 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
Nathan Chancellor0d24f1b2022-12-13 11:35:28 -070086/*
87 * This function is marked __ref because this function may be optimized in such
88 * a way that it directly refers to work_fn's address, which causes modpost to
89 * complain when work_fn is marked __init. This scenario was observed with clang
90 * LTO, where padata_work_init() was optimized to refer directly to
91 * padata_mt_helper() because the calls to padata_work_init() with other work_fn
92 * values were eliminated or inlined.
93 */
94static void __ref padata_work_init(struct padata_work *pw, work_func_t work_fn,
95 void *data, int flags)
Daniel Jordan4611ce22020-06-03 15:59:39 -070096{
Daniel Jordan004ed422020-06-03 15:59:43 -070097 if (flags & PADATA_WORK_ONSTACK)
98 INIT_WORK_ONSTACK(&pw->pw_work, work_fn);
99 else
100 INIT_WORK(&pw->pw_work, work_fn);
Daniel Jordan4611ce22020-06-03 15:59:39 -0700101 pw->pw_data = data;
102}
103
Daniel Jordan004ed422020-06-03 15:59:43 -0700104static int __init padata_work_alloc_mt(int nworks, void *data,
105 struct list_head *head)
106{
107 int i;
108
Herbert Xu58329c42024-04-03 17:36:18 +0800109 spin_lock_bh(&padata_works_lock);
Daniel Jordan004ed422020-06-03 15:59:43 -0700110 /* Start at 1 because the current task participates in the job. */
111 for (i = 1; i < nworks; ++i) {
112 struct padata_work *pw = padata_work_alloc();
113
114 if (!pw)
115 break;
116 padata_work_init(pw, padata_mt_helper, data, 0);
117 list_add(&pw->pw_list, head);
118 }
Herbert Xu58329c42024-04-03 17:36:18 +0800119 spin_unlock_bh(&padata_works_lock);
Daniel Jordan004ed422020-06-03 15:59:43 -0700120
121 return i;
122}
123
Daniel Jordan4611ce22020-06-03 15:59:39 -0700124static void padata_work_free(struct padata_work *pw)
125{
126 lockdep_assert_held(&padata_works_lock);
127 list_add(&pw->pw_list, &padata_free_works);
128}
129
Daniel Jordan004ed422020-06-03 15:59:43 -0700130static void __init padata_works_free(struct list_head *works)
131{
132 struct padata_work *cur, *next;
133
134 if (list_empty(works))
135 return;
136
Herbert Xu58329c42024-04-03 17:36:18 +0800137 spin_lock_bh(&padata_works_lock);
Daniel Jordan004ed422020-06-03 15:59:43 -0700138 list_for_each_entry_safe(cur, next, works, pw_list) {
139 list_del(&cur->pw_list);
140 padata_work_free(cur);
141 }
Herbert Xu58329c42024-04-03 17:36:18 +0800142 spin_unlock_bh(&padata_works_lock);
Daniel Jordan004ed422020-06-03 15:59:43 -0700143}
144
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400145static void padata_parallel_worker(struct work_struct *parallel_work)
Steffen Klassert16295be2010-01-06 19:47:10 +1100146{
Daniel Jordan4611ce22020-06-03 15:59:39 -0700147 struct padata_work *pw = container_of(parallel_work, struct padata_work,
148 pw_work);
149 struct padata_priv *padata = pw->pw_data;
Steffen Klassert16295be2010-01-06 19:47:10 +1100150
151 local_bh_disable();
Daniel Jordan4611ce22020-06-03 15:59:39 -0700152 padata->parallel(padata);
153 spin_lock(&padata_works_lock);
154 padata_work_free(pw);
155 spin_unlock(&padata_works_lock);
Steffen Klassert16295be2010-01-06 19:47:10 +1100156 local_bh_enable();
157}
158
Steffen Klassert0198ffd2010-05-19 13:44:27 +1000159/**
Steffen Klassert16295be2010-01-06 19:47:10 +1100160 * padata_do_parallel - padata parallelization function
161 *
Herbert Xubbefa1d2019-11-26 15:58:45 +0800162 * @ps: padatashell
Steffen Klassert16295be2010-01-06 19:47:10 +1100163 * @padata: object to be parallelized
Daniel Jordane6ce0e02019-09-05 21:40:24 -0400164 * @cb_cpu: pointer to the CPU that the serialization callback function should
165 * run on. If it's not in the serial cpumask of @pinst
166 * (i.e. cpumask.cbcpu), this function selects a fallback CPU and if
167 * none found, returns -EINVAL.
Steffen Klassert16295be2010-01-06 19:47:10 +1100168 *
169 * The parallelization callback function will run with BHs off.
170 * Note: Every object which is parallelized by padata_do_parallel
171 * must be seen by padata_do_serial.
Daniel Jordanbfcdcef82019-12-03 14:31:14 -0500172 *
173 * Return: 0 on success or else negative error code.
Steffen Klassert16295be2010-01-06 19:47:10 +1100174 */
Herbert Xubbefa1d2019-11-26 15:58:45 +0800175int padata_do_parallel(struct padata_shell *ps,
Daniel Jordane6ce0e02019-09-05 21:40:24 -0400176 struct padata_priv *padata, int *cb_cpu)
Steffen Klassert16295be2010-01-06 19:47:10 +1100177{
Herbert Xubbefa1d2019-11-26 15:58:45 +0800178 struct padata_instance *pinst = ps->pinst;
Daniel Jordan4611ce22020-06-03 15:59:39 -0700179 int i, cpu, cpu_index, err;
Steffen Klassert16295be2010-01-06 19:47:10 +1100180 struct parallel_data *pd;
Daniel Jordan4611ce22020-06-03 15:59:39 -0700181 struct padata_work *pw;
Steffen Klassert16295be2010-01-06 19:47:10 +1100182
183 rcu_read_lock_bh();
184
Herbert Xubbefa1d2019-11-26 15:58:45 +0800185 pd = rcu_dereference_bh(ps->pd);
Steffen Klassert16295be2010-01-06 19:47:10 +1100186
Steffen Klassert83f619f2010-07-07 15:32:02 +0200187 err = -EINVAL;
Steffen Klassert74247132010-07-20 08:51:25 +0200188 if (!(pinst->flags & PADATA_INIT) || pinst->flags & PADATA_INVALID)
Steffen Klassert16295be2010-01-06 19:47:10 +1100189 goto out;
190
Daniel Jordane6ce0e02019-09-05 21:40:24 -0400191 if (!cpumask_test_cpu(*cb_cpu, pd->cpumask.cbcpu)) {
Yury Norov1c4cafd2022-01-23 10:38:52 -0800192 if (cpumask_empty(pd->cpumask.cbcpu))
Daniel Jordane6ce0e02019-09-05 21:40:24 -0400193 goto out;
194
195 /* Select an alternate fallback CPU and notify the caller. */
196 cpu_index = *cb_cpu % cpumask_weight(pd->cpumask.cbcpu);
197
198 cpu = cpumask_first(pd->cpumask.cbcpu);
199 for (i = 0; i < cpu_index; i++)
200 cpu = cpumask_next(cpu, pd->cpumask.cbcpu);
201
202 *cb_cpu = cpu;
203 }
Steffen Klassert16295be2010-01-06 19:47:10 +1100204
Lu Jialin8f4f68e2023-09-04 13:33:41 +0000205 err = -EBUSY;
Steffen Klassert16295be2010-01-06 19:47:10 +1100206 if ((pinst->flags & PADATA_RESET))
207 goto out;
208
Xiyu Yangd5ee8e72021-07-20 11:05:11 -0400209 refcount_inc(&pd->refcnt);
Steffen Klassert16295be2010-01-06 19:47:10 +1100210 padata->pd = pd;
Daniel Jordane6ce0e02019-09-05 21:40:24 -0400211 padata->cb_cpu = *cb_cpu;
Steffen Klassert16295be2010-01-06 19:47:10 +1100212
Daniel Jordan4611ce22020-06-03 15:59:39 -0700213 spin_lock(&padata_works_lock);
214 padata->seq_nr = ++pd->seq_nr;
215 pw = padata_work_alloc();
216 spin_unlock(&padata_works_lock);
Daniel Jordan1b0df112020-09-02 13:07:56 -0400217
Daniel Jordan34c3a472022-11-16 20:28:02 -0500218 if (!pw) {
219 /* Maximum works limit exceeded, run in the current task. */
220 padata->parallel(padata);
221 }
222
Daniel Jordan1b0df112020-09-02 13:07:56 -0400223 rcu_read_unlock_bh();
224
Daniel Jordan4611ce22020-06-03 15:59:39 -0700225 if (pw) {
Daniel Jordan004ed422020-06-03 15:59:43 -0700226 padata_work_init(pw, padata_parallel_worker, padata, 0);
Daniel Jordan4611ce22020-06-03 15:59:39 -0700227 queue_work(pinst->parallel_wq, &pw->pw_work);
Daniel Jordan4611ce22020-06-03 15:59:39 -0700228 }
Steffen Klassert16295be2010-01-06 19:47:10 +1100229
Daniel Jordan4611ce22020-06-03 15:59:39 -0700230 return 0;
Steffen Klassert16295be2010-01-06 19:47:10 +1100231out:
232 rcu_read_unlock_bh();
233
234 return err;
235}
236EXPORT_SYMBOL(padata_do_parallel);
237
Steffen Klassert0198ffd2010-05-19 13:44:27 +1000238/*
Daniel Jordanbfde23c2019-09-05 21:40:28 -0400239 * padata_find_next - Find the next object that needs serialization.
Steffen Klassert0198ffd2010-05-19 13:44:27 +1000240 *
Daniel Jordanbfcdcef82019-12-03 14:31:14 -0500241 * Return:
242 * * A pointer to the control struct of the next object that needs
243 * serialization, if present in one of the percpu reorder queues.
244 * * NULL, if the next object that needs serialization will
245 * be parallel processed by another cpu and is not yet present in
246 * the cpu's reorder queue.
Steffen Klassert0198ffd2010-05-19 13:44:27 +1000247 */
Daniel Jordanbfde23c2019-09-05 21:40:28 -0400248static struct padata_priv *padata_find_next(struct parallel_data *pd,
249 bool remove_object)
Steffen Klassert16295be2010-01-06 19:47:10 +1100250{
Steffen Klassert16295be2010-01-06 19:47:10 +1100251 struct padata_priv *padata;
252 struct padata_list *reorder;
Herbert Xu6fc4dbc2019-07-18 23:01:46 +0800253 int cpu = pd->cpu;
Steffen Klassert16295be2010-01-06 19:47:10 +1100254
Daniel Jordanf601c722020-07-14 16:13:56 -0400255 reorder = per_cpu_ptr(pd->reorder_list, cpu);
Steffen Klassert16295be2010-01-06 19:47:10 +1100256
Jason A. Donenfeldde5540d2017-03-23 12:24:43 +0100257 spin_lock(&reorder->lock);
Daniel Jordanbfde23c2019-09-05 21:40:28 -0400258 if (list_empty(&reorder->list)) {
259 spin_unlock(&reorder->lock);
260 return NULL;
261 }
Steffen Klassert16295be2010-01-06 19:47:10 +1100262
Daniel Jordanbfde23c2019-09-05 21:40:28 -0400263 padata = list_entry(reorder->list.next, struct padata_priv, list);
264
265 /*
266 * Checks the rare case where two or more parallel jobs have hashed to
267 * the same CPU and one of the later ones finishes first.
268 */
269 if (padata->seq_nr != pd->processed) {
270 spin_unlock(&reorder->lock);
271 return NULL;
272 }
273
274 if (remove_object) {
Steffen Klassert16295be2010-01-06 19:47:10 +1100275 list_del_init(&padata->list);
Daniel Jordanbfde23c2019-09-05 21:40:28 -0400276 ++pd->processed;
277 pd->cpu = cpumask_next_wrap(cpu, pd->cpumask.pcpu, -1, false);
Steffen Klassert16295be2010-01-06 19:47:10 +1100278 }
Daniel Jordanbfde23c2019-09-05 21:40:28 -0400279
Jason A. Donenfeldde5540d2017-03-23 12:24:43 +0100280 spin_unlock(&reorder->lock);
Steffen Klassert16295be2010-01-06 19:47:10 +1100281 return padata;
282}
283
284static void padata_reorder(struct parallel_data *pd)
285{
Herbert Xubbefa1d2019-11-26 15:58:45 +0800286 struct padata_instance *pinst = pd->ps->pinst;
Steffen Klassert30478172012-03-09 07:20:12 +0100287 int cb_cpu;
Steffen Klassert16295be2010-01-06 19:47:10 +1100288 struct padata_priv *padata;
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400289 struct padata_serial_queue *squeue;
Daniel Jordanf601c722020-07-14 16:13:56 -0400290 struct padata_list *reorder;
Steffen Klassert16295be2010-01-06 19:47:10 +1100291
Steffen Klassert0198ffd2010-05-19 13:44:27 +1000292 /*
293 * We need to ensure that only one cpu can work on dequeueing of
294 * the reorder queue the time. Calculating in which percpu reorder
295 * queue the next object will arrive takes some time. A spinlock
296 * would be highly contended. Also it is not clear in which order
297 * the objects arrive to the reorder queues. So a cpu could wait to
298 * get the lock just to notice that there is nothing to do at the
299 * moment. Therefore we use a trylock and let the holder of the lock
300 * care for all the objects enqueued during the holdtime of the lock.
301 */
Steffen Klassert16295be2010-01-06 19:47:10 +1100302 if (!spin_trylock_bh(&pd->lock))
Steffen Klassertd46a5ac2010-05-19 13:43:14 +1000303 return;
Steffen Klassert16295be2010-01-06 19:47:10 +1100304
305 while (1) {
Daniel Jordanbfde23c2019-09-05 21:40:28 -0400306 padata = padata_find_next(pd, true);
Steffen Klassert16295be2010-01-06 19:47:10 +1100307
Steffen Klassert0198ffd2010-05-19 13:44:27 +1000308 /*
Jason A. Donenfeld69b34842017-04-12 10:40:19 +0200309 * If the next object that needs serialization is parallel
310 * processed by another cpu and is still on it's way to the
311 * cpu's reorder queue, nothing to do for now.
Steffen Klassert0198ffd2010-05-19 13:44:27 +1000312 */
Daniel Jordanbfde23c2019-09-05 21:40:28 -0400313 if (!padata)
Steffen Klassert16295be2010-01-06 19:47:10 +1100314 break;
315
Steffen Klassert30478172012-03-09 07:20:12 +0100316 cb_cpu = padata->cb_cpu;
317 squeue = per_cpu_ptr(pd->squeue, cb_cpu);
Steffen Klassert16295be2010-01-06 19:47:10 +1100318
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400319 spin_lock(&squeue->serial.lock);
320 list_add_tail(&padata->list, &squeue->serial.list);
321 spin_unlock(&squeue->serial.lock);
Steffen Klassert16295be2010-01-06 19:47:10 +1100322
Daniel Jordan45d153c2019-09-05 21:40:27 -0400323 queue_work_on(cb_cpu, pinst->serial_wq, &squeue->work);
Steffen Klassert16295be2010-01-06 19:47:10 +1100324 }
325
326 spin_unlock_bh(&pd->lock);
327
Steffen Klassert0198ffd2010-05-19 13:44:27 +1000328 /*
329 * The next object that needs serialization might have arrived to
Herbert Xu6fc4dbc2019-07-18 23:01:46 +0800330 * the reorder queues in the meantime.
Daniel Jordancf144f82019-07-16 12:32:53 -0400331 *
Herbert Xu6fc4dbc2019-07-18 23:01:46 +0800332 * Ensure reorder queue is read after pd->lock is dropped so we see
333 * new objects from another task in padata_do_serial. Pairs with
Daniel Jordane04ec0d2020-06-08 17:26:52 -0400334 * smp_mb in padata_do_serial.
Steffen Klassert0198ffd2010-05-19 13:44:27 +1000335 */
Daniel Jordancf144f82019-07-16 12:32:53 -0400336 smp_mb();
Steffen Klassert16295be2010-01-06 19:47:10 +1100337
Daniel Jordanf601c722020-07-14 16:13:56 -0400338 reorder = per_cpu_ptr(pd->reorder_list, pd->cpu);
339 if (!list_empty(&reorder->list) && padata_find_next(pd, false))
Daniel Jordan45d153c2019-09-05 21:40:27 -0400340 queue_work(pinst->serial_wq, &pd->reorder_work);
Steffen Klassert16295be2010-01-06 19:47:10 +1100341}
342
Mathias Krausecf5868c2017-09-08 20:57:10 +0200343static void invoke_padata_reorder(struct work_struct *work)
344{
Mathias Krausecf5868c2017-09-08 20:57:10 +0200345 struct parallel_data *pd;
346
347 local_bh_disable();
Herbert Xu6fc4dbc2019-07-18 23:01:46 +0800348 pd = container_of(work, struct parallel_data, reorder_work);
Mathias Krausecf5868c2017-09-08 20:57:10 +0200349 padata_reorder(pd);
350 local_bh_enable();
351}
352
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400353static void padata_serial_worker(struct work_struct *serial_work)
Steffen Klassert16295be2010-01-06 19:47:10 +1100354{
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400355 struct padata_serial_queue *squeue;
Steffen Klassert16295be2010-01-06 19:47:10 +1100356 struct parallel_data *pd;
357 LIST_HEAD(local_list);
Herbert Xu07928d92019-11-19 13:17:31 +0800358 int cnt;
Steffen Klassert16295be2010-01-06 19:47:10 +1100359
360 local_bh_disable();
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400361 squeue = container_of(serial_work, struct padata_serial_queue, work);
362 pd = squeue->pd;
Steffen Klassert16295be2010-01-06 19:47:10 +1100363
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400364 spin_lock(&squeue->serial.lock);
365 list_replace_init(&squeue->serial.list, &local_list);
366 spin_unlock(&squeue->serial.lock);
Steffen Klassert16295be2010-01-06 19:47:10 +1100367
Herbert Xu07928d92019-11-19 13:17:31 +0800368 cnt = 0;
369
Steffen Klassert16295be2010-01-06 19:47:10 +1100370 while (!list_empty(&local_list)) {
371 struct padata_priv *padata;
372
373 padata = list_entry(local_list.next,
374 struct padata_priv, list);
375
376 list_del_init(&padata->list);
377
378 padata->serial(padata);
Herbert Xu07928d92019-11-19 13:17:31 +0800379 cnt++;
Steffen Klassert16295be2010-01-06 19:47:10 +1100380 }
381 local_bh_enable();
Herbert Xu07928d92019-11-19 13:17:31 +0800382
Xiyu Yangd5ee8e72021-07-20 11:05:11 -0400383 if (refcount_sub_and_test(cnt, &pd->refcnt))
Herbert Xu07928d92019-11-19 13:17:31 +0800384 padata_free_pd(pd);
Steffen Klassert16295be2010-01-06 19:47:10 +1100385}
386
Steffen Klassert0198ffd2010-05-19 13:44:27 +1000387/**
Steffen Klassert16295be2010-01-06 19:47:10 +1100388 * padata_do_serial - padata serialization function
389 *
390 * @padata: object to be serialized.
391 *
392 * padata_do_serial must be called for every parallelized object.
393 * The serialization callback function will run with BHs off.
394 */
395void padata_do_serial(struct padata_priv *padata)
396{
Daniel Jordan065cf572019-07-19 15:04:44 -0400397 struct parallel_data *pd = padata->pd;
Daniel Jordan4611ce22020-06-03 15:59:39 -0700398 int hashed_cpu = padata_cpu_hash(pd, padata->seq_nr);
Daniel Jordanf601c722020-07-14 16:13:56 -0400399 struct padata_list *reorder = per_cpu_ptr(pd->reorder_list, hashed_cpu);
Daniel Jordanbfde23c2019-09-05 21:40:28 -0400400 struct padata_priv *cur;
Daniel Jordan57ddfec2022-11-16 20:28:04 -0500401 struct list_head *pos;
Steffen Klassert16295be2010-01-06 19:47:10 +1100402
Daniel Jordanf601c722020-07-14 16:13:56 -0400403 spin_lock(&reorder->lock);
Daniel Jordanbfde23c2019-09-05 21:40:28 -0400404 /* Sort in ascending order of sequence number. */
Daniel Jordan57ddfec2022-11-16 20:28:04 -0500405 list_for_each_prev(pos, &reorder->list) {
406 cur = list_entry(pos, struct padata_priv, list);
Daniel Jordanbfde23c2019-09-05 21:40:28 -0400407 if (cur->seq_nr < padata->seq_nr)
408 break;
Daniel Jordan57ddfec2022-11-16 20:28:04 -0500409 }
410 list_add(&padata->list, pos);
Daniel Jordanf601c722020-07-14 16:13:56 -0400411 spin_unlock(&reorder->lock);
Steffen Klassert16295be2010-01-06 19:47:10 +1100412
Daniel Jordancf144f82019-07-16 12:32:53 -0400413 /*
Herbert Xu6fc4dbc2019-07-18 23:01:46 +0800414 * Ensure the addition to the reorder list is ordered correctly
Daniel Jordancf144f82019-07-16 12:32:53 -0400415 * with the trylock of pd->lock in padata_reorder. Pairs with smp_mb
416 * in padata_reorder.
417 */
Daniel Jordane04ec0d2020-06-08 17:26:52 -0400418 smp_mb();
Daniel Jordancf144f82019-07-16 12:32:53 -0400419
Herbert Xu6fc4dbc2019-07-18 23:01:46 +0800420 padata_reorder(pd);
Steffen Klassert16295be2010-01-06 19:47:10 +1100421}
422EXPORT_SYMBOL(padata_do_serial);
423
Herbert Xubbefa1d2019-11-26 15:58:45 +0800424static int padata_setup_cpumasks(struct padata_instance *pinst)
Steffen Klassert16295be2010-01-06 19:47:10 +1100425{
Daniel Jordanbfde23c2019-09-05 21:40:28 -0400426 struct workqueue_attrs *attrs;
Herbert Xubbefa1d2019-11-26 15:58:45 +0800427 int err;
428
429 attrs = alloc_workqueue_attrs();
430 if (!attrs)
431 return -ENOMEM;
432
433 /* Restrict parallel_wq workers to pd->cpumask.pcpu. */
434 cpumask_copy(attrs->cpumask, pinst->cpumask.pcpu);
435 err = apply_workqueue_attrs(pinst->parallel_wq, attrs);
436 free_workqueue_attrs(attrs);
437
438 return err;
439}
440
Daniel Jordan004ed422020-06-03 15:59:43 -0700441static void __init padata_mt_helper(struct work_struct *w)
442{
443 struct padata_work *pw = container_of(w, struct padata_work, pw_work);
444 struct padata_mt_job_state *ps = pw->pw_data;
445 struct padata_mt_job *job = ps->job;
446 bool done;
447
448 spin_lock(&ps->lock);
449
450 while (job->size > 0) {
451 unsigned long start, size, end;
452
453 start = job->start;
454 /* So end is chunk size aligned if enough work remains. */
455 size = roundup(start + 1, ps->chunk_size) - start;
456 size = min(size, job->size);
457 end = start + size;
458
459 job->start = end;
460 job->size -= size;
461
462 spin_unlock(&ps->lock);
463 job->thread_fn(start, end, job->fn_arg);
464 spin_lock(&ps->lock);
465 }
466
467 ++ps->nworks_fini;
468 done = (ps->nworks_fini == ps->nworks);
469 spin_unlock(&ps->lock);
470
471 if (done)
472 complete(&ps->completion);
473}
474
475/**
476 * padata_do_multithreaded - run a multithreaded job
477 * @job: Description of the job.
478 *
479 * See the definition of struct padata_mt_job for more details.
480 */
481void __init padata_do_multithreaded(struct padata_mt_job *job)
482{
483 /* In case threads finish at different times. */
484 static const unsigned long load_balance_factor = 4;
485 struct padata_work my_work, *pw;
486 struct padata_mt_job_state ps;
487 LIST_HEAD(works);
Gang Li Subject: padata: dispatch works oneb522862024-03-06 13:04:17 -0800488 int nworks, nid;
489 static atomic_t last_used_nid __initdata;
Daniel Jordan004ed422020-06-03 15:59:43 -0700490
491 if (job->size == 0)
492 return;
493
494 /* Ensure at least one thread when size < min_chunk. */
Anthony Yznagaf84155c2023-02-22 16:33:12 -0800495 nworks = max(job->size / max(job->min_chunk, job->align), 1ul);
Daniel Jordan004ed422020-06-03 15:59:43 -0700496 nworks = min(nworks, job->max_threads);
497
498 if (nworks == 1) {
499 /* Single thread, no coordination needed, cut to the chase. */
500 job->thread_fn(job->start, job->start + job->size, job->fn_arg);
501 return;
502 }
503
504 spin_lock_init(&ps.lock);
505 init_completion(&ps.completion);
506 ps.job = job;
507 ps.nworks = padata_work_alloc_mt(nworks, &ps, &works);
508 ps.nworks_fini = 0;
509
510 /*
511 * Chunk size is the amount of work a helper does per call to the
512 * thread function. Load balance large jobs between threads by
513 * increasing the number of chunks, guarantee at least the minimum
514 * chunk size from the caller, and honor the caller's alignment.
515 */
516 ps.chunk_size = job->size / (ps.nworks * load_balance_factor);
517 ps.chunk_size = max(ps.chunk_size, job->min_chunk);
518 ps.chunk_size = roundup(ps.chunk_size, job->align);
519
520 list_for_each_entry(pw, &works, pw_list)
Gang Li Subject: padata: dispatch works oneb522862024-03-06 13:04:17 -0800521 if (job->numa_aware) {
522 int old_node = atomic_read(&last_used_nid);
523
524 do {
525 nid = next_node_in(old_node, node_states[N_CPU]);
526 } while (!atomic_try_cmpxchg(&last_used_nid, &old_node, nid));
527 queue_work_node(nid, system_unbound_wq, &pw->pw_work);
528 } else {
529 queue_work(system_unbound_wq, &pw->pw_work);
530 }
Daniel Jordan004ed422020-06-03 15:59:43 -0700531
532 /* Use the current thread, which saves starting a workqueue worker. */
533 padata_work_init(&my_work, padata_mt_helper, &ps, PADATA_WORK_ONSTACK);
534 padata_mt_helper(&my_work.pw_work);
535
536 /* Wait for all the helpers to finish. */
537 wait_for_completion(&ps.completion);
538
539 destroy_work_on_stack(&my_work.pw_work);
540 padata_works_free(&works);
541}
542
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400543static void __padata_list_init(struct padata_list *pd_list)
544{
545 INIT_LIST_HEAD(&pd_list->list);
546 spin_lock_init(&pd_list->lock);
547}
548
549/* Initialize all percpu queues used by serial workers */
550static void padata_init_squeues(struct parallel_data *pd)
551{
552 int cpu;
553 struct padata_serial_queue *squeue;
554
555 for_each_cpu(cpu, pd->cpumask.cbcpu) {
556 squeue = per_cpu_ptr(pd->squeue, cpu);
557 squeue->pd = pd;
558 __padata_list_init(&squeue->serial);
559 INIT_WORK(&squeue->work, padata_serial_worker);
560 }
561}
562
Daniel Jordanf601c722020-07-14 16:13:56 -0400563/* Initialize per-CPU reorder lists */
564static void padata_init_reorder_list(struct parallel_data *pd)
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400565{
Daniel Jordanc51636a2019-09-05 21:40:29 -0400566 int cpu;
Daniel Jordanf601c722020-07-14 16:13:56 -0400567 struct padata_list *list;
Steffen Klassert16295be2010-01-06 19:47:10 +1100568
Daniel Jordanc51636a2019-09-05 21:40:29 -0400569 for_each_cpu(cpu, pd->cpumask.pcpu) {
Daniel Jordanf601c722020-07-14 16:13:56 -0400570 list = per_cpu_ptr(pd->reorder_list, cpu);
571 __padata_list_init(list);
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400572 }
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400573}
574
575/* Allocate and initialize the internal cpumask dependend resources. */
Herbert Xubbefa1d2019-11-26 15:58:45 +0800576static struct parallel_data *padata_alloc_pd(struct padata_shell *ps)
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400577{
Herbert Xubbefa1d2019-11-26 15:58:45 +0800578 struct padata_instance *pinst = ps->pinst;
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400579 struct parallel_data *pd;
Steffen Klassert16295be2010-01-06 19:47:10 +1100580
581 pd = kzalloc(sizeof(struct parallel_data), GFP_KERNEL);
582 if (!pd)
583 goto err;
584
Daniel Jordanf601c722020-07-14 16:13:56 -0400585 pd->reorder_list = alloc_percpu(struct padata_list);
586 if (!pd->reorder_list)
Steffen Klassert16295be2010-01-06 19:47:10 +1100587 goto err_free_pd;
588
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400589 pd->squeue = alloc_percpu(struct padata_serial_queue);
590 if (!pd->squeue)
Daniel Jordanf601c722020-07-14 16:13:56 -0400591 goto err_free_reorder_list;
Daniel Jordanbfde23c2019-09-05 21:40:28 -0400592
Herbert Xubbefa1d2019-11-26 15:58:45 +0800593 pd->ps = ps;
Daniel Jordancec00e6e2020-07-14 16:13:53 -0400594
595 if (!alloc_cpumask_var(&pd->cpumask.pcpu, GFP_KERNEL))
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400596 goto err_free_squeue;
Daniel Jordancec00e6e2020-07-14 16:13:53 -0400597 if (!alloc_cpumask_var(&pd->cpumask.cbcpu, GFP_KERNEL))
598 goto err_free_pcpu;
599
Daniel Jordand69e0372020-07-14 16:13:54 -0400600 cpumask_and(pd->cpumask.pcpu, pinst->cpumask.pcpu, cpu_online_mask);
601 cpumask_and(pd->cpumask.cbcpu, pinst->cpumask.cbcpu, cpu_online_mask);
Steffen Klassert16295be2010-01-06 19:47:10 +1100602
Daniel Jordanf601c722020-07-14 16:13:56 -0400603 padata_init_reorder_list(pd);
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400604 padata_init_squeues(pd);
Daniel Jordan4611ce22020-06-03 15:59:39 -0700605 pd->seq_nr = -1;
Xiyu Yangd5ee8e72021-07-20 11:05:11 -0400606 refcount_set(&pd->refcnt, 1);
Steffen Klassert16295be2010-01-06 19:47:10 +1100607 spin_lock_init(&pd->lock);
Daniel Jordanec9c7d12019-08-08 12:05:35 -0400608 pd->cpu = cpumask_first(pd->cpumask.pcpu);
Herbert Xu6fc4dbc2019-07-18 23:01:46 +0800609 INIT_WORK(&pd->reorder_work, invoke_padata_reorder);
Steffen Klassert16295be2010-01-06 19:47:10 +1100610
611 return pd;
612
Daniel Jordancec00e6e2020-07-14 16:13:53 -0400613err_free_pcpu:
614 free_cpumask_var(pd->cpumask.pcpu);
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400615err_free_squeue:
616 free_percpu(pd->squeue);
Daniel Jordanf601c722020-07-14 16:13:56 -0400617err_free_reorder_list:
618 free_percpu(pd->reorder_list);
Steffen Klassert16295be2010-01-06 19:47:10 +1100619err_free_pd:
620 kfree(pd);
621err:
622 return NULL;
623}
624
625static void padata_free_pd(struct parallel_data *pd)
626{
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400627 free_cpumask_var(pd->cpumask.pcpu);
628 free_cpumask_var(pd->cpumask.cbcpu);
Daniel Jordanf601c722020-07-14 16:13:56 -0400629 free_percpu(pd->reorder_list);
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400630 free_percpu(pd->squeue);
Steffen Klassert16295be2010-01-06 19:47:10 +1100631 kfree(pd);
632}
633
Steffen Klassert4c879172010-07-07 15:30:10 +0200634static void __padata_start(struct padata_instance *pinst)
635{
636 pinst->flags |= PADATA_INIT;
637}
638
Steffen Klassertee836552010-07-07 15:30:47 +0200639static void __padata_stop(struct padata_instance *pinst)
640{
641 if (!(pinst->flags & PADATA_INIT))
642 return;
643
644 pinst->flags &= ~PADATA_INIT;
645
646 synchronize_rcu();
Steffen Klassertee836552010-07-07 15:30:47 +0200647}
648
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300649/* Replace the internal control structure with a new one. */
Herbert Xubbefa1d2019-11-26 15:58:45 +0800650static int padata_replace_one(struct padata_shell *ps)
Steffen Klassert16295be2010-01-06 19:47:10 +1100651{
Herbert Xubbefa1d2019-11-26 15:58:45 +0800652 struct parallel_data *pd_new;
653
654 pd_new = padata_alloc_pd(ps);
655 if (!pd_new)
656 return -ENOMEM;
657
658 ps->opd = rcu_dereference_protected(ps->pd, 1);
659 rcu_assign_pointer(ps->pd, pd_new);
660
661 return 0;
662}
663
Daniel Jordan894c9ef2019-12-03 14:31:10 -0500664static int padata_replace(struct padata_instance *pinst)
Herbert Xubbefa1d2019-11-26 15:58:45 +0800665{
Herbert Xubbefa1d2019-11-26 15:58:45 +0800666 struct padata_shell *ps;
Daniel Jordan41ccdbf2020-02-10 13:11:00 -0500667 int err = 0;
Steffen Klassert16295be2010-01-06 19:47:10 +1100668
669 pinst->flags |= PADATA_RESET;
670
Herbert Xubbefa1d2019-11-26 15:58:45 +0800671 list_for_each_entry(ps, &pinst->pslist, list) {
672 err = padata_replace_one(ps);
673 if (err)
674 break;
675 }
Steffen Klassert16295be2010-01-06 19:47:10 +1100676
677 synchronize_rcu();
678
Herbert Xubbefa1d2019-11-26 15:58:45 +0800679 list_for_each_entry_continue_reverse(ps, &pinst->pslist, list)
Xiyu Yangd5ee8e72021-07-20 11:05:11 -0400680 if (refcount_dec_and_test(&ps->opd->refcnt))
Herbert Xubbefa1d2019-11-26 15:58:45 +0800681 padata_free_pd(ps->opd);
Steffen Klassert16295be2010-01-06 19:47:10 +1100682
683 pinst->flags &= ~PADATA_RESET;
Herbert Xubbefa1d2019-11-26 15:58:45 +0800684
685 return err;
Steffen Klassert16295be2010-01-06 19:47:10 +1100686}
687
Steffen Klassert33e54452010-07-07 15:31:26 +0200688/* If cpumask contains no active cpu, we mark the instance as invalid. */
689static bool padata_validate_cpumask(struct padata_instance *pinst,
690 const struct cpumask *cpumask)
691{
Steffen Klassert13614e02012-03-28 08:43:21 +0200692 if (!cpumask_intersects(cpumask, cpu_online_mask)) {
Steffen Klassert33e54452010-07-07 15:31:26 +0200693 pinst->flags |= PADATA_INVALID;
694 return false;
Steffen Klassert16295be2010-01-06 19:47:10 +1100695 }
696
Steffen Klassert33e54452010-07-07 15:31:26 +0200697 pinst->flags &= ~PADATA_INVALID;
698 return true;
699}
700
Steffen Klassert65ff5772010-07-27 07:15:06 +0200701static int __padata_set_cpumasks(struct padata_instance *pinst,
702 cpumask_var_t pcpumask,
703 cpumask_var_t cbcpumask)
Steffen Klassert16295be2010-01-06 19:47:10 +1100704{
Steffen Klassert33e54452010-07-07 15:31:26 +0200705 int valid;
Herbert Xubbefa1d2019-11-26 15:58:45 +0800706 int err;
Steffen Klassert16295be2010-01-06 19:47:10 +1100707
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400708 valid = padata_validate_cpumask(pinst, pcpumask);
709 if (!valid) {
710 __padata_stop(pinst);
711 goto out_replace;
712 }
713
714 valid = padata_validate_cpumask(pinst, cbcpumask);
Steffen Klassertb89661d2010-07-20 08:49:20 +0200715 if (!valid)
Steffen Klassert33e54452010-07-07 15:31:26 +0200716 __padata_stop(pinst);
Steffen Klassert33e54452010-07-07 15:31:26 +0200717
Steffen Klassertb89661d2010-07-20 08:49:20 +0200718out_replace:
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400719 cpumask_copy(pinst->cpumask.pcpu, pcpumask);
720 cpumask_copy(pinst->cpumask.cbcpu, cbcpumask);
Steffen Klassert16295be2010-01-06 19:47:10 +1100721
Daniel Jordan894c9ef2019-12-03 14:31:10 -0500722 err = padata_setup_cpumasks(pinst) ?: padata_replace(pinst);
Steffen Klassert16295be2010-01-06 19:47:10 +1100723
Steffen Klassert33e54452010-07-07 15:31:26 +0200724 if (valid)
725 __padata_start(pinst);
726
Herbert Xubbefa1d2019-11-26 15:58:45 +0800727 return err;
Steffen Klassert65ff5772010-07-27 07:15:06 +0200728}
729
730/**
Daniel Jordanbfcdcef82019-12-03 14:31:14 -0500731 * padata_set_cpumask - Sets specified by @cpumask_type cpumask to the value
732 * equivalent to @cpumask.
Steffen Klassert65ff5772010-07-27 07:15:06 +0200733 * @pinst: padata instance
734 * @cpumask_type: PADATA_CPU_SERIAL or PADATA_CPU_PARALLEL corresponding
735 * to parallel and serial cpumasks respectively.
736 * @cpumask: the cpumask to use
Daniel Jordanbfcdcef82019-12-03 14:31:14 -0500737 *
738 * Return: 0 on success or negative error code
Steffen Klassert65ff5772010-07-27 07:15:06 +0200739 */
740int padata_set_cpumask(struct padata_instance *pinst, int cpumask_type,
741 cpumask_var_t cpumask)
742{
743 struct cpumask *serial_mask, *parallel_mask;
744 int err = -EINVAL;
745
Sebastian Andrzej Siewior80771c82021-08-03 16:16:10 +0200746 cpus_read_lock();
Daniel Jordan38228e82019-12-03 14:31:11 -0500747 mutex_lock(&pinst->lock);
Steffen Klassert65ff5772010-07-27 07:15:06 +0200748
749 switch (cpumask_type) {
750 case PADATA_CPU_PARALLEL:
751 serial_mask = pinst->cpumask.cbcpu;
752 parallel_mask = cpumask;
753 break;
754 case PADATA_CPU_SERIAL:
755 parallel_mask = pinst->cpumask.pcpu;
756 serial_mask = cpumask;
757 break;
758 default:
759 goto out;
760 }
761
762 err = __padata_set_cpumasks(pinst, parallel_mask, serial_mask);
763
Steffen Klassert16295be2010-01-06 19:47:10 +1100764out:
765 mutex_unlock(&pinst->lock);
Sebastian Andrzej Siewior80771c82021-08-03 16:16:10 +0200766 cpus_read_unlock();
Steffen Klassert16295be2010-01-06 19:47:10 +1100767
768 return err;
769}
770EXPORT_SYMBOL(padata_set_cpumask);
771
Arnd Bergmann19d795b2016-05-19 17:09:59 -0700772#ifdef CONFIG_HOTPLUG_CPU
773
Steffen Klassert16295be2010-01-06 19:47:10 +1100774static int __padata_add_cpu(struct padata_instance *pinst, int cpu)
775{
Herbert Xubbefa1d2019-11-26 15:58:45 +0800776 int err = 0;
Steffen Klassert16295be2010-01-06 19:47:10 +1100777
Steffen Klassert13614e02012-03-28 08:43:21 +0200778 if (cpumask_test_cpu(cpu, cpu_online_mask)) {
Daniel Jordan894c9ef2019-12-03 14:31:10 -0500779 err = padata_replace(pinst);
Steffen Klassert33e54452010-07-07 15:31:26 +0200780
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400781 if (padata_validate_cpumask(pinst, pinst->cpumask.pcpu) &&
782 padata_validate_cpumask(pinst, pinst->cpumask.cbcpu))
Steffen Klassert33e54452010-07-07 15:31:26 +0200783 __padata_start(pinst);
Steffen Klassert16295be2010-01-06 19:47:10 +1100784 }
785
Herbert Xubbefa1d2019-11-26 15:58:45 +0800786 return err;
Steffen Klassert16295be2010-01-06 19:47:10 +1100787}
788
Steffen Klassert16295be2010-01-06 19:47:10 +1100789static int __padata_remove_cpu(struct padata_instance *pinst, int cpu)
790{
Herbert Xubbefa1d2019-11-26 15:58:45 +0800791 int err = 0;
Steffen Klassert16295be2010-01-06 19:47:10 +1100792
Daniel Jordan894c9ef2019-12-03 14:31:10 -0500793 if (!cpumask_test_cpu(cpu, cpu_online_mask)) {
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400794 if (!padata_validate_cpumask(pinst, pinst->cpumask.pcpu) ||
Steffen Klassertb89661d2010-07-20 08:49:20 +0200795 !padata_validate_cpumask(pinst, pinst->cpumask.cbcpu))
Steffen Klassert33e54452010-07-07 15:31:26 +0200796 __padata_stop(pinst);
Steffen Klassert33e54452010-07-07 15:31:26 +0200797
Daniel Jordan894c9ef2019-12-03 14:31:10 -0500798 err = padata_replace(pinst);
Steffen Klassert16295be2010-01-06 19:47:10 +1100799 }
800
Herbert Xubbefa1d2019-11-26 15:58:45 +0800801 return err;
Steffen Klassert16295be2010-01-06 19:47:10 +1100802}
803
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400804static inline int pinst_has_cpu(struct padata_instance *pinst, int cpu)
805{
806 return cpumask_test_cpu(cpu, pinst->cpumask.pcpu) ||
807 cpumask_test_cpu(cpu, pinst->cpumask.cbcpu);
808}
809
Sebastian Andrzej Siewior30e92152016-09-06 19:04:49 +0200810static int padata_cpu_online(unsigned int cpu, struct hlist_node *node)
Steffen Klassert16295be2010-01-06 19:47:10 +1100811{
Steffen Klassert16295be2010-01-06 19:47:10 +1100812 struct padata_instance *pinst;
Sebastian Andrzej Siewior30e92152016-09-06 19:04:49 +0200813 int ret;
Steffen Klassert16295be2010-01-06 19:47:10 +1100814
Daniel Jordan3c2214b2020-04-21 12:34:55 -0400815 pinst = hlist_entry_safe(node, struct padata_instance, cpu_online_node);
Sebastian Andrzej Siewior30e92152016-09-06 19:04:49 +0200816 if (!pinst_has_cpu(pinst, cpu))
817 return 0;
Steffen Klassert16295be2010-01-06 19:47:10 +1100818
Sebastian Andrzej Siewior30e92152016-09-06 19:04:49 +0200819 mutex_lock(&pinst->lock);
820 ret = __padata_add_cpu(pinst, cpu);
821 mutex_unlock(&pinst->lock);
822 return ret;
Steffen Klassert16295be2010-01-06 19:47:10 +1100823}
Sebastian Andrzej Siewior30e92152016-09-06 19:04:49 +0200824
Daniel Jordan894c9ef2019-12-03 14:31:10 -0500825static int padata_cpu_dead(unsigned int cpu, struct hlist_node *node)
Sebastian Andrzej Siewior30e92152016-09-06 19:04:49 +0200826{
827 struct padata_instance *pinst;
828 int ret;
829
Daniel Jordan3c2214b2020-04-21 12:34:55 -0400830 pinst = hlist_entry_safe(node, struct padata_instance, cpu_dead_node);
Sebastian Andrzej Siewior30e92152016-09-06 19:04:49 +0200831 if (!pinst_has_cpu(pinst, cpu))
832 return 0;
833
834 mutex_lock(&pinst->lock);
835 ret = __padata_remove_cpu(pinst, cpu);
836 mutex_unlock(&pinst->lock);
837 return ret;
838}
839
840static enum cpuhp_state hp_online;
Steffen Klasserte2cb2f12010-04-29 14:40:10 +0200841#endif
Steffen Klassert16295be2010-01-06 19:47:10 +1100842
Dan Kruchinin5e017dc2010-07-14 14:33:08 +0400843static void __padata_free(struct padata_instance *pinst)
844{
845#ifdef CONFIG_HOTPLUG_CPU
Daniel Jordan3c2214b2020-04-21 12:34:55 -0400846 cpuhp_state_remove_instance_nocalls(CPUHP_PADATA_DEAD,
847 &pinst->cpu_dead_node);
848 cpuhp_state_remove_instance_nocalls(hp_online, &pinst->cpu_online_node);
Dan Kruchinin5e017dc2010-07-14 14:33:08 +0400849#endif
850
Herbert Xubbefa1d2019-11-26 15:58:45 +0800851 WARN_ON(!list_empty(&pinst->pslist));
852
Dan Kruchinin5e017dc2010-07-14 14:33:08 +0400853 free_cpumask_var(pinst->cpumask.pcpu);
854 free_cpumask_var(pinst->cpumask.cbcpu);
Daniel Jordan45d153c2019-09-05 21:40:27 -0400855 destroy_workqueue(pinst->serial_wq);
856 destroy_workqueue(pinst->parallel_wq);
Dan Kruchinin5e017dc2010-07-14 14:33:08 +0400857 kfree(pinst);
858}
859
860#define kobj2pinst(_kobj) \
861 container_of(_kobj, struct padata_instance, kobj)
862#define attr2pentry(_attr) \
863 container_of(_attr, struct padata_sysfs_entry, attr)
864
865static void padata_sysfs_release(struct kobject *kobj)
866{
867 struct padata_instance *pinst = kobj2pinst(kobj);
868 __padata_free(pinst);
869}
870
871struct padata_sysfs_entry {
872 struct attribute attr;
873 ssize_t (*show)(struct padata_instance *, struct attribute *, char *);
874 ssize_t (*store)(struct padata_instance *, struct attribute *,
875 const char *, size_t);
876};
877
878static ssize_t show_cpumask(struct padata_instance *pinst,
879 struct attribute *attr, char *buf)
880{
881 struct cpumask *cpumask;
882 ssize_t len;
883
884 mutex_lock(&pinst->lock);
885 if (!strcmp(attr->name, "serial_cpumask"))
886 cpumask = pinst->cpumask.cbcpu;
887 else
888 cpumask = pinst->cpumask.pcpu;
889
Tejun Heo4497da62015-02-13 14:38:05 -0800890 len = snprintf(buf, PAGE_SIZE, "%*pb\n",
891 nr_cpu_ids, cpumask_bits(cpumask));
Dan Kruchinin5e017dc2010-07-14 14:33:08 +0400892 mutex_unlock(&pinst->lock);
Tejun Heo4497da62015-02-13 14:38:05 -0800893 return len < PAGE_SIZE ? len : -EINVAL;
Dan Kruchinin5e017dc2010-07-14 14:33:08 +0400894}
895
896static ssize_t store_cpumask(struct padata_instance *pinst,
897 struct attribute *attr,
898 const char *buf, size_t count)
899{
900 cpumask_var_t new_cpumask;
901 ssize_t ret;
902 int mask_type;
903
904 if (!alloc_cpumask_var(&new_cpumask, GFP_KERNEL))
905 return -ENOMEM;
906
907 ret = bitmap_parse(buf, count, cpumask_bits(new_cpumask),
908 nr_cpumask_bits);
909 if (ret < 0)
910 goto out;
911
912 mask_type = !strcmp(attr->name, "serial_cpumask") ?
913 PADATA_CPU_SERIAL : PADATA_CPU_PARALLEL;
914 ret = padata_set_cpumask(pinst, mask_type, new_cpumask);
915 if (!ret)
916 ret = count;
917
918out:
919 free_cpumask_var(new_cpumask);
920 return ret;
921}
922
923#define PADATA_ATTR_RW(_name, _show_name, _store_name) \
924 static struct padata_sysfs_entry _name##_attr = \
925 __ATTR(_name, 0644, _show_name, _store_name)
926#define PADATA_ATTR_RO(_name, _show_name) \
927 static struct padata_sysfs_entry _name##_attr = \
928 __ATTR(_name, 0400, _show_name, NULL)
929
930PADATA_ATTR_RW(serial_cpumask, show_cpumask, store_cpumask);
931PADATA_ATTR_RW(parallel_cpumask, show_cpumask, store_cpumask);
932
933/*
934 * Padata sysfs provides the following objects:
935 * serial_cpumask [RW] - cpumask for serial workers
936 * parallel_cpumask [RW] - cpumask for parallel workers
Steffen Klassert16295be2010-01-06 19:47:10 +1100937 */
Dan Kruchinin5e017dc2010-07-14 14:33:08 +0400938static struct attribute *padata_default_attrs[] = {
939 &serial_cpumask_attr.attr,
940 &parallel_cpumask_attr.attr,
941 NULL,
942};
Kimberly Brown2064fbc2019-04-01 22:51:47 -0400943ATTRIBUTE_GROUPS(padata_default);
Dan Kruchinin5e017dc2010-07-14 14:33:08 +0400944
945static ssize_t padata_sysfs_show(struct kobject *kobj,
946 struct attribute *attr, char *buf)
Steffen Klassert16295be2010-01-06 19:47:10 +1100947{
Steffen Klassert16295be2010-01-06 19:47:10 +1100948 struct padata_instance *pinst;
Dan Kruchinin5e017dc2010-07-14 14:33:08 +0400949 struct padata_sysfs_entry *pentry;
950 ssize_t ret = -EIO;
951
952 pinst = kobj2pinst(kobj);
953 pentry = attr2pentry(attr);
954 if (pentry->show)
955 ret = pentry->show(pinst, attr, buf);
956
957 return ret;
958}
959
960static ssize_t padata_sysfs_store(struct kobject *kobj, struct attribute *attr,
961 const char *buf, size_t count)
962{
963 struct padata_instance *pinst;
964 struct padata_sysfs_entry *pentry;
965 ssize_t ret = -EIO;
966
967 pinst = kobj2pinst(kobj);
968 pentry = attr2pentry(attr);
969 if (pentry->show)
970 ret = pentry->store(pinst, attr, buf, count);
971
972 return ret;
973}
974
975static const struct sysfs_ops padata_sysfs_ops = {
976 .show = padata_sysfs_show,
977 .store = padata_sysfs_store,
978};
979
Thomas Weißschuh0bedc992023-02-17 03:17:49 +0000980static const struct kobj_type padata_attr_type = {
Dan Kruchinin5e017dc2010-07-14 14:33:08 +0400981 .sysfs_ops = &padata_sysfs_ops,
Kimberly Brown2064fbc2019-04-01 22:51:47 -0400982 .default_groups = padata_default_groups,
Dan Kruchinin5e017dc2010-07-14 14:33:08 +0400983 .release = padata_sysfs_release,
984};
985
Steffen Klassert16295be2010-01-06 19:47:10 +1100986/**
Daniel Jordan3f257192020-07-14 16:13:55 -0400987 * padata_alloc - allocate and initialize a padata instance
Daniel Jordanb128a302019-09-05 21:40:21 -0400988 * @name: used to identify the instance
Daniel Jordanbfcdcef82019-12-03 14:31:14 -0500989 *
990 * Return: new instance on success, NULL on error
Dan Kruchinine15bacb2010-07-14 14:31:57 +0400991 */
Daniel Jordan3f257192020-07-14 16:13:55 -0400992struct padata_instance *padata_alloc(const char *name)
Steffen Klassert16295be2010-01-06 19:47:10 +1100993{
994 struct padata_instance *pinst;
Steffen Klassert16295be2010-01-06 19:47:10 +1100995
996 pinst = kzalloc(sizeof(struct padata_instance), GFP_KERNEL);
997 if (!pinst)
998 goto err;
999
Daniel Jordanbfde23c2019-09-05 21:40:28 -04001000 pinst->parallel_wq = alloc_workqueue("%s_parallel", WQ_UNBOUND, 0,
1001 name);
Daniel Jordan45d153c2019-09-05 21:40:27 -04001002 if (!pinst->parallel_wq)
Steffen Klassert16295be2010-01-06 19:47:10 +11001003 goto err_free_inst;
Daniel Jordanb128a302019-09-05 21:40:21 -04001004
Sebastian Andrzej Siewior80771c82021-08-03 16:16:10 +02001005 cpus_read_lock();
Daniel Jordancc491d82019-09-05 21:40:26 -04001006
Daniel Jordan45d153c2019-09-05 21:40:27 -04001007 pinst->serial_wq = alloc_workqueue("%s_serial", WQ_MEM_RECLAIM |
1008 WQ_CPU_INTENSIVE, 1, name);
1009 if (!pinst->serial_wq)
Daniel Jordancc491d82019-09-05 21:40:26 -04001010 goto err_put_cpus;
Daniel Jordan45d153c2019-09-05 21:40:27 -04001011
1012 if (!alloc_cpumask_var(&pinst->cpumask.pcpu, GFP_KERNEL))
1013 goto err_free_serial_wq;
Dan Kruchinine15bacb2010-07-14 14:31:57 +04001014 if (!alloc_cpumask_var(&pinst->cpumask.cbcpu, GFP_KERNEL)) {
1015 free_cpumask_var(pinst->cpumask.pcpu);
Daniel Jordan45d153c2019-09-05 21:40:27 -04001016 goto err_free_serial_wq;
Steffen Klassert33e54452010-07-07 15:31:26 +02001017 }
Steffen Klassert16295be2010-01-06 19:47:10 +11001018
Herbert Xubbefa1d2019-11-26 15:58:45 +08001019 INIT_LIST_HEAD(&pinst->pslist);
Steffen Klassert16295be2010-01-06 19:47:10 +11001020
Daniel Jordan3f257192020-07-14 16:13:55 -04001021 cpumask_copy(pinst->cpumask.pcpu, cpu_possible_mask);
1022 cpumask_copy(pinst->cpumask.cbcpu, cpu_possible_mask);
Herbert Xubbefa1d2019-11-26 15:58:45 +08001023
1024 if (padata_setup_cpumasks(pinst))
Daniel Jordand69e0372020-07-14 16:13:54 -04001025 goto err_free_masks;
Steffen Klassert16295be2010-01-06 19:47:10 +11001026
Daniel Jordanbd25b4882020-07-14 16:13:51 -04001027 __padata_start(pinst);
Steffen Klassert16295be2010-01-06 19:47:10 +11001028
Dan Kruchinin5e017dc2010-07-14 14:33:08 +04001029 kobject_init(&pinst->kobj, &padata_attr_type);
Steffen Klassert16295be2010-01-06 19:47:10 +11001030 mutex_init(&pinst->lock);
1031
Richard Weinbergerb8b4a412013-08-23 13:12:33 +02001032#ifdef CONFIG_HOTPLUG_CPU
Daniel Jordan3c2214b2020-04-21 12:34:55 -04001033 cpuhp_state_add_instance_nocalls_cpuslocked(hp_online,
1034 &pinst->cpu_online_node);
Daniel Jordan894c9ef2019-12-03 14:31:10 -05001035 cpuhp_state_add_instance_nocalls_cpuslocked(CPUHP_PADATA_DEAD,
Daniel Jordan3c2214b2020-04-21 12:34:55 -04001036 &pinst->cpu_dead_node);
Richard Weinbergerb8b4a412013-08-23 13:12:33 +02001037#endif
Daniel Jordancc491d82019-09-05 21:40:26 -04001038
Sebastian Andrzej Siewior80771c82021-08-03 16:16:10 +02001039 cpus_read_unlock();
Daniel Jordancc491d82019-09-05 21:40:26 -04001040
Steffen Klassert16295be2010-01-06 19:47:10 +11001041 return pinst;
1042
Dan Kruchinine15bacb2010-07-14 14:31:57 +04001043err_free_masks:
1044 free_cpumask_var(pinst->cpumask.pcpu);
1045 free_cpumask_var(pinst->cpumask.cbcpu);
Daniel Jordan45d153c2019-09-05 21:40:27 -04001046err_free_serial_wq:
1047 destroy_workqueue(pinst->serial_wq);
Daniel Jordancc491d82019-09-05 21:40:26 -04001048err_put_cpus:
Sebastian Andrzej Siewior80771c82021-08-03 16:16:10 +02001049 cpus_read_unlock();
Daniel Jordan45d153c2019-09-05 21:40:27 -04001050 destroy_workqueue(pinst->parallel_wq);
Steffen Klassert16295be2010-01-06 19:47:10 +11001051err_free_inst:
1052 kfree(pinst);
1053err:
1054 return NULL;
1055}
Daniel Jordan3f257192020-07-14 16:13:55 -04001056EXPORT_SYMBOL(padata_alloc);
Thomas Gleixner95966952017-05-24 10:15:17 +02001057
1058/**
Steffen Klassert16295be2010-01-06 19:47:10 +11001059 * padata_free - free a padata instance
1060 *
Daniel Jordanbfcdcef82019-12-03 14:31:14 -05001061 * @pinst: padata instance to free
Steffen Klassert16295be2010-01-06 19:47:10 +11001062 */
1063void padata_free(struct padata_instance *pinst)
1064{
Dan Kruchinin5e017dc2010-07-14 14:33:08 +04001065 kobject_put(&pinst->kobj);
Steffen Klassert16295be2010-01-06 19:47:10 +11001066}
1067EXPORT_SYMBOL(padata_free);
Sebastian Andrzej Siewior30e92152016-09-06 19:04:49 +02001068
Herbert Xubbefa1d2019-11-26 15:58:45 +08001069/**
1070 * padata_alloc_shell - Allocate and initialize padata shell.
1071 *
1072 * @pinst: Parent padata_instance object.
Daniel Jordanbfcdcef82019-12-03 14:31:14 -05001073 *
1074 * Return: new shell on success, NULL on error
Herbert Xubbefa1d2019-11-26 15:58:45 +08001075 */
1076struct padata_shell *padata_alloc_shell(struct padata_instance *pinst)
1077{
1078 struct parallel_data *pd;
1079 struct padata_shell *ps;
1080
1081 ps = kzalloc(sizeof(*ps), GFP_KERNEL);
1082 if (!ps)
1083 goto out;
1084
1085 ps->pinst = pinst;
1086
Sebastian Andrzej Siewior80771c82021-08-03 16:16:10 +02001087 cpus_read_lock();
Herbert Xubbefa1d2019-11-26 15:58:45 +08001088 pd = padata_alloc_pd(ps);
Sebastian Andrzej Siewior80771c82021-08-03 16:16:10 +02001089 cpus_read_unlock();
Herbert Xubbefa1d2019-11-26 15:58:45 +08001090
1091 if (!pd)
1092 goto out_free_ps;
1093
1094 mutex_lock(&pinst->lock);
1095 RCU_INIT_POINTER(ps->pd, pd);
1096 list_add(&ps->list, &pinst->pslist);
1097 mutex_unlock(&pinst->lock);
1098
1099 return ps;
1100
1101out_free_ps:
1102 kfree(ps);
1103out:
1104 return NULL;
1105}
1106EXPORT_SYMBOL(padata_alloc_shell);
1107
1108/**
1109 * padata_free_shell - free a padata shell
1110 *
1111 * @ps: padata shell to free
1112 */
1113void padata_free_shell(struct padata_shell *ps)
1114{
WangJinchao7ddc21e2023-10-16 09:15:21 +08001115 struct parallel_data *pd;
1116
Eric Biggers07b24c72020-02-25 20:59:22 -08001117 if (!ps)
1118 return;
Herbert Xubbefa1d2019-11-26 15:58:45 +08001119
Eric Biggers07b24c72020-02-25 20:59:22 -08001120 mutex_lock(&ps->pinst->lock);
Herbert Xubbefa1d2019-11-26 15:58:45 +08001121 list_del(&ps->list);
WangJinchao7ddc21e2023-10-16 09:15:21 +08001122 pd = rcu_dereference_protected(ps->pd, 1);
1123 if (refcount_dec_and_test(&pd->refcnt))
1124 padata_free_pd(pd);
Eric Biggers07b24c72020-02-25 20:59:22 -08001125 mutex_unlock(&ps->pinst->lock);
Herbert Xubbefa1d2019-11-26 15:58:45 +08001126
1127 kfree(ps);
1128}
1129EXPORT_SYMBOL(padata_free_shell);
1130
Daniel Jordanf1b192b2020-06-03 15:59:35 -07001131void __init padata_init(void)
Sebastian Andrzej Siewior30e92152016-09-06 19:04:49 +02001132{
Daniel Jordan4611ce22020-06-03 15:59:39 -07001133 unsigned int i, possible_cpus;
Daniel Jordanf1b192b2020-06-03 15:59:35 -07001134#ifdef CONFIG_HOTPLUG_CPU
Sebastian Andrzej Siewior30e92152016-09-06 19:04:49 +02001135 int ret;
1136
1137 ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "padata:online",
Daniel Jordan894c9ef2019-12-03 14:31:10 -05001138 padata_cpu_online, NULL);
Sebastian Andrzej Siewior30e92152016-09-06 19:04:49 +02001139 if (ret < 0)
Daniel Jordanf1b192b2020-06-03 15:59:35 -07001140 goto err;
Sebastian Andrzej Siewior30e92152016-09-06 19:04:49 +02001141 hp_online = ret;
Daniel Jordan894c9ef2019-12-03 14:31:10 -05001142
1143 ret = cpuhp_setup_state_multi(CPUHP_PADATA_DEAD, "padata:dead",
1144 NULL, padata_cpu_dead);
Daniel Jordan4611ce22020-06-03 15:59:39 -07001145 if (ret < 0)
1146 goto remove_online_state;
1147#endif
1148
1149 possible_cpus = num_possible_cpus();
1150 padata_works = kmalloc_array(possible_cpus, sizeof(struct padata_work),
1151 GFP_KERNEL);
1152 if (!padata_works)
1153 goto remove_dead_state;
1154
1155 for (i = 0; i < possible_cpus; ++i)
1156 list_add(&padata_works[i].pw_list, &padata_free_works);
Sebastian Andrzej Siewior30e92152016-09-06 19:04:49 +02001157
Daniel Jordanf1b192b2020-06-03 15:59:35 -07001158 return;
Daniel Jordan4611ce22020-06-03 15:59:39 -07001159
1160remove_dead_state:
1161#ifdef CONFIG_HOTPLUG_CPU
1162 cpuhp_remove_multi_state(CPUHP_PADATA_DEAD);
1163remove_online_state:
1164 cpuhp_remove_multi_state(hp_online);
Daniel Jordanf1b192b2020-06-03 15:59:35 -07001165err:
Sebastian Andrzej Siewior30e92152016-09-06 19:04:49 +02001166#endif
Daniel Jordan4611ce22020-06-03 15:59:39 -07001167 pr_warn("padata: initialization failed\n");
Daniel Jordanf1b192b2020-06-03 15:59:35 -07001168}