~hc/RK356X_SDK_RELEASE.git

..	..	@@ -2,10 +2,13 @@
2	2	/*
3	3	* padata.c - generic interface to process data streams in parallel
4	4	*
5		- * See Documentation/padata.txt for an api documentation.
	5	+ * See Documentation/core-api/padata.rst for more information.
6	6	*
7	7	* Copyright (C) 2008, 2009 secunet Security Networks AG
8	8	* Copyright (C) 2008, 2009 Steffen Klassert <steffen.klassert@secunet.com>
	9	+ *
	10	+ * Copyright (c) 2020 Oracle and/or its affiliates.
	11	+ * Author: Daniel Jordan <daniel.m.jordan@oracle.com>
9	12	*
10	13	* This program is free software; you can redistribute it and/or modify it
11	14	* under the terms and conditions of the GNU General Public License,
..	..	@@ -21,6 +24,7 @@
21	24	* 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
22	25	*/
23	26
	27	+#include <linux/completion.h>
24	28	#include <linux/export.h>
25	29	#include <linux/cpumask.h>
26	30	#include <linux/err.h>
..	..	@@ -31,11 +35,30 @@
31	35	#include <linux/slab.h>
32	36	#include <linux/sysfs.h>
33	37	#include <linux/rcupdate.h>
34		-#include <linux/module.h>
35	38
36		-#define MAX_OBJ_NUM 1000
	39	+#define PADATA_WORK_ONSTACK 1 /* Work's memory is on stack */
	40	+
	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);
	50	+
	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	+};
37	59
38	60	static void padata_free_pd(struct parallel_data *pd);
	61	+static void __init padata_mt_helper(struct work_struct *work);
39	62
40	63	static int padata_index_to_cpu(struct parallel_data *pd, int cpu_index)
41	64	{
..	..	@@ -48,101 +71,168 @@
48	71	return target_cpu;
49	72	}
50	73
51		-static int padata_cpu_hash(struct parallel_data *pd)
	74	+static int padata_cpu_hash(struct parallel_data *pd, unsigned int seq_nr)
52	75	{
53		- unsigned int seq_nr;
54		- int cpu_index;
55		-
56	76	/*
57	77	* Hash the sequence numbers to the cpus by taking
58	78	* seq_nr mod. number of cpus in use.
59	79	*/
60		-
61		- seq_nr = atomic_inc_return(&pd->seq_nr);
62		- cpu_index = seq_nr % cpumask_weight(pd->cpumask.pcpu);
	80	+ int cpu_index = seq_nr % cpumask_weight(pd->cpumask.pcpu);
63	81
64	82	return padata_index_to_cpu(pd, cpu_index);
65	83	}
66	84
	85	+static struct padata_work *padata_work_alloc(void)
	86	+{
	87	+ struct padata_work *pw;
	88	+
	89	+ lockdep_assert_held(&padata_works_lock);
	90	+
	91	+ if (list_empty(&padata_free_works))
	92	+ return NULL; /* No more work items allowed to be queued. */
	93	+
	94	+ pw = list_first_entry(&padata_free_works, struct padata_work, pw_list);
	95	+ list_del(&pw->pw_list);
	96	+ return pw;
	97	+}
	98	+
	99	+static void padata_work_init(struct padata_work *pw, work_func_t work_fn,
	100	+ void *data, int flags)
	101	+{
	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);
	106	+ pw->pw_data = data;
	107	+}
	108	+
	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	+
	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	+}
	134	+
	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	+
67	150	static void padata_parallel_worker(struct work_struct *parallel_work)
68	151	{
69		- struct padata_parallel_queue *pqueue;
70		- LIST_HEAD(local_list);
	152	+ struct padata_work *pw = container_of(parallel_work, struct padata_work,
	153	+ pw_work);
	154	+ struct padata_priv *padata = pw->pw_data;
71	155
72	156	local_bh_disable();
73		- pqueue = container_of(parallel_work,
74		- struct padata_parallel_queue, work);
75		-
76		- spin_lock(&pqueue->parallel.lock);
77		- list_replace_init(&pqueue->parallel.list, &local_list);
78		- spin_unlock(&pqueue->parallel.lock);
79		-
80		- while (!list_empty(&local_list)) {
81		- struct padata_priv *padata;
82		-
83		- padata = list_entry(local_list.next,
84		- struct padata_priv, list);
85		-
86		- list_del_init(&padata->list);
87		-
88		- padata->parallel(padata);
89		- }
90		-
	157	+ padata->parallel(padata);
	158	+ spin_lock(&padata_works_lock);
	159	+ padata_work_free(pw);
	160	+ spin_unlock(&padata_works_lock);
91	161	local_bh_enable();
92	162	}
93	163
94	164	/**
95	165	* padata_do_parallel - padata parallelization function
96	166	*
97		- * @pinst: padata instance
	167	+ * @ps: padatashell
98	168	* @padata: object to be parallelized
99		- * @cb_cpu: cpu the serialization callback function will run on,
100		- * must be in the serial cpumask of padata(i.e. cpumask.cbcpu).
	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.
101	173	*
102	174	* The parallelization callback function will run with BHs off.
103	175	* Note: Every object which is parallelized by padata_do_parallel
104	176	* must be seen by padata_do_serial.
	177	+ *
	178	+ * Return: 0 on success or else negative error code.
105	179	*/
106		-int padata_do_parallel(struct padata_instance *pinst,
107		- struct padata_priv *padata, int cb_cpu)
	180	+int padata_do_parallel(struct padata_shell *ps,
	181	+ struct padata_priv padata, int cb_cpu)
108	182	{
109		- int target_cpu, err;
110		- struct padata_parallel_queue *queue;
	183	+ struct padata_instance *pinst = ps->pinst;
	184	+ int i, cpu, cpu_index, err;
111	185	struct parallel_data *pd;
	186	+ struct padata_work *pw;
112	187
113	188	rcu_read_lock_bh();
114	189
115		- pd = rcu_dereference_bh(pinst->pd);
	190	+ pd = rcu_dereference_bh(ps->pd);
116	191
117	192	err = -EINVAL;
118	193	if (!(pinst->flags & PADATA_INIT) \|\| pinst->flags & PADATA_INVALID)
119	194	goto out;
120	195
121		- if (!cpumask_test_cpu(cb_cpu, pd->cpumask.cbcpu))
122		- goto out;
	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	+ }
123	209
124	210	err = -EBUSY;
125	211	if ((pinst->flags & PADATA_RESET))
126	212	goto out;
127	213
128		- if (atomic_read(&pd->refcnt) >= MAX_OBJ_NUM)
129		- goto out;
130		-
131		- err = 0;
132	214	atomic_inc(&pd->refcnt);
133	215	padata->pd = pd;
134		- padata->cb_cpu = cb_cpu;
	216	+ padata->cb_cpu = *cb_cpu;
135	217
136		- target_cpu = padata_cpu_hash(pd);
137		- padata->cpu = target_cpu;
138		- queue = per_cpu_ptr(pd->pqueue, target_cpu);
	218	+ spin_lock(&padata_works_lock);
	219	+ padata->seq_nr = ++pd->seq_nr;
	220	+ pw = padata_work_alloc();
	221	+ spin_unlock(&padata_works_lock);
139	222
140		- spin_lock(&queue->parallel.lock);
141		- list_add_tail(&padata->list, &queue->parallel.list);
142		- spin_unlock(&queue->parallel.lock);
	223	+ if (!pw) {
	224	+ /* Maximum works limit exceeded, run in the current task. */
	225	+ padata->parallel(padata);
	226	+ }
143	227
144		- queue_work_on(target_cpu, pinst->wq, &queue->work);
	228	+ rcu_read_unlock_bh();
145	229
	230	+ if (pw) {
	231	+ padata_work_init(pw, padata_parallel_worker, padata, 0);
	232	+ queue_work(pinst->parallel_wq, &pw->pw_work);
	233	+ }
	234	+
	235	+ return 0;
146	236	out:
147	237	rcu_read_unlock_bh();
148	238
..	..	@@ -151,63 +241,58 @@
151	241	EXPORT_SYMBOL(padata_do_parallel);
152	242
153	243	/*
154		- * padata_get_next - Get the next object that needs serialization.
	244	+ * padata_find_next - Find the next object that needs serialization.
155	245	*
156		- * Return values are:
157		- *
158		- * A pointer to the control struct of the next object that needs
159		- * serialization, if present in one of the percpu reorder queues.
160		- *
161		- * -EINPROGRESS, if the next object that needs serialization will
162		- * be parallel processed by another cpu and is not yet present in
163		- * the cpu's reorder queue.
164		- *
165		- * -ENODATA, if this cpu has to do the parallel processing for
166		- * the next object.
	246	+ * Return:
	247	+ * * A pointer to the control struct of the next object that needs
	248	+ * serialization, if present in one of the percpu reorder queues.
	249	+ * * NULL, if the next object that needs serialization will
	250	+ * be parallel processed by another cpu and is not yet present in
	251	+ * the cpu's reorder queue.
167	252	*/
168		-static struct padata_priv padata_get_next(struct parallel_data pd)
	253	+static struct padata_priv padata_find_next(struct parallel_data pd,
	254	+ bool remove_object)
169	255	{
170		- struct padata_parallel_queue *next_queue;
171	256	struct padata_priv *padata;
172	257	struct padata_list *reorder;
173	258	int cpu = pd->cpu;
174	259
175		- next_queue = per_cpu_ptr(pd->pqueue, cpu);
176		- reorder = &next_queue->reorder;
	260	+ reorder = per_cpu_ptr(pd->reorder_list, cpu);
177	261
178	262	spin_lock(&reorder->lock);
179		- if (!list_empty(&reorder->list)) {
180		- padata = list_entry(reorder->list.next,
181		- struct padata_priv, list);
182		-
183		- list_del_init(&padata->list);
184		- atomic_dec(&pd->reorder_objects);
185		-
186		- pd->cpu = cpumask_next_wrap(cpu, pd->cpumask.pcpu, -1,
187		- false);
188		-
	263	+ if (list_empty(&reorder->list)) {
189	264	spin_unlock(&reorder->lock);
190		- goto out;
	265	+ return NULL;
191	266	}
	267	+
	268	+ padata = list_entry(reorder->list.next, struct padata_priv, list);
	269	+
	270	+ /*
	271	+ * Checks the rare case where two or more parallel jobs have hashed to
	272	+ * the same CPU and one of the later ones finishes first.
	273	+ */
	274	+ if (padata->seq_nr != pd->processed) {
	275	+ spin_unlock(&reorder->lock);
	276	+ return NULL;
	277	+ }
	278	+
	279	+ if (remove_object) {
	280	+ list_del_init(&padata->list);
	281	+ ++pd->processed;
	282	+ pd->cpu = cpumask_next_wrap(cpu, pd->cpumask.pcpu, -1, false);
	283	+ }
	284	+
192	285	spin_unlock(&reorder->lock);
193		-
194		- if (__this_cpu_read(pd->pqueue->cpu_index) == next_queue->cpu_index) {
195		- padata = ERR_PTR(-ENODATA);
196		- goto out;
197		- }
198		-
199		- padata = ERR_PTR(-EINPROGRESS);
200		-out:
201	286	return padata;
202	287	}
203	288
204	289	static void padata_reorder(struct parallel_data *pd)
205	290	{
	291	+ struct padata_instance *pinst = pd->ps->pinst;
206	292	int cb_cpu;
207	293	struct padata_priv *padata;
208	294	struct padata_serial_queue *squeue;
209		- struct padata_instance *pinst = pd->pinst;
210		- struct padata_parallel_queue *next_queue;
	295	+ struct padata_list *reorder;
211	296
212	297	/*
213	298	* We need to ensure that only one cpu can work on dequeueing of
..	..	@@ -223,25 +308,15 @@
223	308	return;
224	309
225	310	while (1) {
226		- padata = padata_get_next(pd);
	311	+ padata = padata_find_next(pd, true);
227	312
228	313	/*
229	314	* If the next object that needs serialization is parallel
230	315	* processed by another cpu and is still on it's way to the
231	316	* cpu's reorder queue, nothing to do for now.
232	317	*/
233		- if (PTR_ERR(padata) == -EINPROGRESS)
	318	+ if (!padata)
234	319	break;
235		-
236		- /*
237		- * This cpu has to do the parallel processing of the next
238		- * object. It's waiting in the cpu's parallelization queue,
239		- * so exit immediately.
240		- */
241		- if (PTR_ERR(padata) == -ENODATA) {
242		- spin_unlock_bh(&pd->lock);
243		- return;
244		- }
245	320
246	321	cb_cpu = padata->cb_cpu;
247	322	squeue = per_cpu_ptr(pd->squeue, cb_cpu);
..	..	@@ -250,7 +325,7 @@
250	325	list_add_tail(&padata->list, &squeue->serial.list);
251	326	spin_unlock(&squeue->serial.lock);
252	327
253		- queue_work_on(cb_cpu, pinst->wq, &squeue->work);
	328	+ queue_work_on(cb_cpu, pinst->serial_wq, &squeue->work);
254	329	}
255	330
256	331	spin_unlock_bh(&pd->lock);
..	..	@@ -261,13 +336,13 @@
261	336	*
262	337	* Ensure reorder queue is read after pd->lock is dropped so we see
263	338	* new objects from another task in padata_do_serial. Pairs with
264		- * smp_mb__after_atomic in padata_do_serial.
	339	+ * smp_mb in padata_do_serial.
265	340	*/
266	341	smp_mb();
267	342
268		- next_queue = per_cpu_ptr(pd->pqueue, pd->cpu);
269		- if (!list_empty(&next_queue->reorder.list))
270		- queue_work(pinst->wq, &pd->reorder_work);
	343	+ reorder = per_cpu_ptr(pd->reorder_list, pd->cpu);
	344	+ if (!list_empty(&reorder->list) && padata_find_next(pd, false))
	345	+ queue_work(pinst->serial_wq, &pd->reorder_work);
271	346	}
272	347
273	348	static void invoke_padata_reorder(struct work_struct *work)
..	..	@@ -325,40 +400,139 @@
325	400	void padata_do_serial(struct padata_priv *padata)
326	401	{
327	402	struct parallel_data *pd = padata->pd;
328		- struct padata_parallel_queue *pqueue = per_cpu_ptr(pd->pqueue,
329		- padata->cpu);
	403	+ int hashed_cpu = padata_cpu_hash(pd, padata->seq_nr);
	404	+ struct padata_list *reorder = per_cpu_ptr(pd->reorder_list, hashed_cpu);
	405	+ struct padata_priv *cur;
	406	+ struct list_head *pos;
330	407
331		- spin_lock(&pqueue->reorder.lock);
332		- list_add_tail(&padata->list, &pqueue->reorder.list);
333		- atomic_inc(&pd->reorder_objects);
334		- spin_unlock(&pqueue->reorder.lock);
	408	+ spin_lock(&reorder->lock);
	409	+ /* Sort in ascending order of sequence number. */
	410	+ list_for_each_prev(pos, &reorder->list) {
	411	+ cur = list_entry(pos, struct padata_priv, list);
	412	+ if (cur->seq_nr < padata->seq_nr)
	413	+ break;
	414	+ }
	415	+ list_add(&padata->list, pos);
	416	+ spin_unlock(&reorder->lock);
335	417
336	418	/*
337	419	* Ensure the addition to the reorder list is ordered correctly
338	420	* with the trylock of pd->lock in padata_reorder. Pairs with smp_mb
339	421	* in padata_reorder.
340	422	*/
341		- smp_mb__after_atomic();
	423	+ smp_mb();
342	424
343	425	padata_reorder(pd);
344	426	}
345	427	EXPORT_SYMBOL(padata_do_serial);
346	428
347		-static int padata_setup_cpumasks(struct parallel_data *pd,
348		- const struct cpumask *pcpumask,
349		- const struct cpumask *cbcpumask)
	429	+static int padata_setup_cpumasks(struct padata_instance *pinst)
350	430	{
351		- if (!alloc_cpumask_var(&pd->cpumask.pcpu, GFP_KERNEL))
	431	+ struct workqueue_attrs *attrs;
	432	+ int err;
	433	+
	434	+ attrs = alloc_workqueue_attrs();
	435	+ if (!attrs)
352	436	return -ENOMEM;
353	437
354		- cpumask_and(pd->cpumask.pcpu, pcpumask, cpu_online_mask);
355		- if (!alloc_cpumask_var(&pd->cpumask.cbcpu, GFP_KERNEL)) {
356		- free_cpumask_var(pd->cpumask.pcpu);
357		- return -ENOMEM;
	438	+ /* Restrict parallel_wq workers to pd->cpumask.pcpu. */
	439	+ cpumask_copy(attrs->cpumask, pinst->cpumask.pcpu);
	440	+ err = apply_workqueue_attrs(pinst->parallel_wq, attrs);
	441	+ free_workqueue_attrs(attrs);
	442	+
	443	+ return err;
	444	+}
	445	+
	446	+static void __init padata_mt_helper(struct work_struct *w)
	447	+{
	448	+ struct padata_work *pw = container_of(w, struct padata_work, pw_work);
	449	+ struct padata_mt_job_state *ps = pw->pw_data;
	450	+ struct padata_mt_job *job = ps->job;
	451	+ bool done;
	452	+
	453	+ spin_lock(&ps->lock);
	454	+
	455	+ while (job->size > 0) {
	456	+ unsigned long start, size, end;
	457	+
	458	+ start = job->start;
	459	+ /* So end is chunk size aligned if enough work remains. */
	460	+ size = roundup(start + 1, ps->chunk_size) - start;
	461	+ size = min(size, job->size);
	462	+ end = start + size;
	463	+
	464	+ job->start = end;
	465	+ job->size -= size;
	466	+
	467	+ spin_unlock(&ps->lock);
	468	+ job->thread_fn(start, end, job->fn_arg);
	469	+ spin_lock(&ps->lock);
358	470	}
359	471
360		- cpumask_and(pd->cpumask.cbcpu, cbcpumask, cpu_online_mask);
361		- return 0;
	472	+ ++ps->nworks_fini;
	473	+ done = (ps->nworks_fini == ps->nworks);
	474	+ spin_unlock(&ps->lock);
	475	+
	476	+ if (done)
	477	+ complete(&ps->completion);
	478	+}
	479	+
	480	+/**
	481	+ * padata_do_multithreaded - run a multithreaded job
	482	+ * @job: Description of the job.
	483	+ *
	484	+ * See the definition of struct padata_mt_job for more details.
	485	+ */
	486	+void __init padata_do_multithreaded(struct padata_mt_job *job)
	487	+{
	488	+ /* In case threads finish at different times. */
	489	+ static const unsigned long load_balance_factor = 4;
	490	+ struct padata_work my_work, *pw;
	491	+ struct padata_mt_job_state ps;
	492	+ LIST_HEAD(works);
	493	+ int nworks;
	494	+
	495	+ if (job->size == 0)
	496	+ return;
	497	+
	498	+ /* Ensure at least one thread when size < min_chunk. */
	499	+ nworks = max(job->size / job->min_chunk, 1ul);
	500	+ nworks = min(nworks, job->max_threads);
	501	+
	502	+ if (nworks == 1) {
	503	+ /* Single thread, no coordination needed, cut to the chase. */
	504	+ job->thread_fn(job->start, job->start + job->size, job->fn_arg);
	505	+ return;
	506	+ }
	507	+
	508	+ spin_lock_init(&ps.lock);
	509	+ init_completion(&ps.completion);
	510	+ ps.job = job;
	511	+ ps.nworks = padata_work_alloc_mt(nworks, &ps, &works);
	512	+ ps.nworks_fini = 0;
	513	+
	514	+ /*
	515	+ * Chunk size is the amount of work a helper does per call to the
	516	+ * thread function. Load balance large jobs between threads by
	517	+ * increasing the number of chunks, guarantee at least the minimum
	518	+ * chunk size from the caller, and honor the caller's alignment.
	519	+ */
	520	+ ps.chunk_size = job->size / (ps.nworks * load_balance_factor);
	521	+ ps.chunk_size = max(ps.chunk_size, job->min_chunk);
	522	+ ps.chunk_size = roundup(ps.chunk_size, job->align);
	523	+
	524	+ list_for_each_entry(pw, &works, pw_list)
	525	+ queue_work(system_unbound_wq, &pw->pw_work);
	526	+
	527	+ /* Use the current thread, which saves starting a workqueue worker. */
	528	+ padata_work_init(&my_work, padata_mt_helper, &ps, PADATA_WORK_ONSTACK);
	529	+ padata_mt_helper(&my_work.pw_work);
	530	+
	531	+ /* Wait for all the helpers to finish. */
	532	+ wait_for_completion(&ps.completion);
	533	+
	534	+ destroy_work_on_stack(&my_work.pw_work);
	535	+ padata_works_free(&works);
362	536	}
363	537
364	538	static void __padata_list_init(struct padata_list *pd_list)
..	..	@@ -381,68 +555,62 @@
381	555	}
382	556	}
383	557
384		-/* Initialize all percpu queues used by parallel workers */
385		-static void padata_init_pqueues(struct parallel_data *pd)
	558	+/* Initialize per-CPU reorder lists */
	559	+static void padata_init_reorder_list(struct parallel_data *pd)
386	560	{
387		- int cpu_index, cpu;
388		- struct padata_parallel_queue *pqueue;
	561	+ int cpu;
	562	+ struct padata_list *list;
389	563
390		- cpu_index = 0;
391		- for_each_possible_cpu(cpu) {
392		- pqueue = per_cpu_ptr(pd->pqueue, cpu);
393		-
394		- if (!cpumask_test_cpu(cpu, pd->cpumask.pcpu)) {
395		- pqueue->cpu_index = -1;
396		- continue;
397		- }
398		-
399		- pqueue->cpu_index = cpu_index;
400		- cpu_index++;
401		-
402		- __padata_list_init(&pqueue->reorder);
403		- __padata_list_init(&pqueue->parallel);
404		- INIT_WORK(&pqueue->work, padata_parallel_worker);
405		- atomic_set(&pqueue->num_obj, 0);
	564	+ for_each_cpu(cpu, pd->cpumask.pcpu) {
	565	+ list = per_cpu_ptr(pd->reorder_list, cpu);
	566	+ __padata_list_init(list);
406	567	}
407	568	}
408	569
409	570	/* Allocate and initialize the internal cpumask dependend resources. */
410		-static struct parallel_data padata_alloc_pd(struct padata_instance pinst,
411		- const struct cpumask *pcpumask,
412		- const struct cpumask *cbcpumask)
	571	+static struct parallel_data padata_alloc_pd(struct padata_shell ps)
413	572	{
	573	+ struct padata_instance *pinst = ps->pinst;
414	574	struct parallel_data *pd;
415	575
416	576	pd = kzalloc(sizeof(struct parallel_data), GFP_KERNEL);
417	577	if (!pd)
418	578	goto err;
419	579
420		- pd->pqueue = alloc_percpu(struct padata_parallel_queue);
421		- if (!pd->pqueue)
	580	+ pd->reorder_list = alloc_percpu(struct padata_list);
	581	+ if (!pd->reorder_list)
422	582	goto err_free_pd;
423	583
424	584	pd->squeue = alloc_percpu(struct padata_serial_queue);
425	585	if (!pd->squeue)
426		- goto err_free_pqueue;
427		- if (padata_setup_cpumasks(pd, pcpumask, cbcpumask) < 0)
428		- goto err_free_squeue;
	586	+ goto err_free_reorder_list;
429	587
430		- padata_init_pqueues(pd);
	588	+ pd->ps = ps;
	589	+
	590	+ if (!alloc_cpumask_var(&pd->cpumask.pcpu, GFP_KERNEL))
	591	+ goto err_free_squeue;
	592	+ if (!alloc_cpumask_var(&pd->cpumask.cbcpu, GFP_KERNEL))
	593	+ goto err_free_pcpu;
	594	+
	595	+ cpumask_and(pd->cpumask.pcpu, pinst->cpumask.pcpu, cpu_online_mask);
	596	+ cpumask_and(pd->cpumask.cbcpu, pinst->cpumask.cbcpu, cpu_online_mask);
	597	+
	598	+ padata_init_reorder_list(pd);
431	599	padata_init_squeues(pd);
432		- atomic_set(&pd->seq_nr, -1);
433		- atomic_set(&pd->reorder_objects, 0);
	600	+ pd->seq_nr = -1;
434	601	atomic_set(&pd->refcnt, 1);
435		- pd->pinst = pinst;
436	602	spin_lock_init(&pd->lock);
437	603	pd->cpu = cpumask_first(pd->cpumask.pcpu);
438	604	INIT_WORK(&pd->reorder_work, invoke_padata_reorder);
439	605
440	606	return pd;
441	607
	608	+err_free_pcpu:
	609	+ free_cpumask_var(pd->cpumask.pcpu);
442	610	err_free_squeue:
443	611	free_percpu(pd->squeue);
444		-err_free_pqueue:
445		- free_percpu(pd->pqueue);
	612	+err_free_reorder_list:
	613	+ free_percpu(pd->reorder_list);
446	614	err_free_pd:
447	615	kfree(pd);
448	616	err:
..	..	@@ -453,7 +621,7 @@
453	621	{
454	622	free_cpumask_var(pd->cpumask.pcpu);
455	623	free_cpumask_var(pd->cpumask.cbcpu);
456		- free_percpu(pd->pqueue);
	624	+ free_percpu(pd->reorder_list);
457	625	free_percpu(pd->squeue);
458	626	kfree(pd);
459	627	}
..	..	@@ -474,65 +642,43 @@
474	642	}
475	643
476	644	/* Replace the internal control structure with a new one. */
477		-static void padata_replace(struct padata_instance *pinst,
478		- struct parallel_data *pd_new)
	645	+static int padata_replace_one(struct padata_shell *ps)
479	646	{
480		- struct parallel_data *pd_old = pinst->pd;
481		- int notification_mask = 0;
	647	+ struct parallel_data *pd_new;
	648	+
	649	+ pd_new = padata_alloc_pd(ps);
	650	+ if (!pd_new)
	651	+ return -ENOMEM;
	652	+
	653	+ ps->opd = rcu_dereference_protected(ps->pd, 1);
	654	+ rcu_assign_pointer(ps->pd, pd_new);
	655	+
	656	+ return 0;
	657	+}
	658	+
	659	+static int padata_replace(struct padata_instance *pinst)
	660	+{
	661	+ struct padata_shell *ps;
	662	+ int err = 0;
482	663
483	664	pinst->flags \|= PADATA_RESET;
484	665
485		- rcu_assign_pointer(pinst->pd, pd_new);
	666	+ list_for_each_entry(ps, &pinst->pslist, list) {
	667	+ err = padata_replace_one(ps);
	668	+ if (err)
	669	+ break;
	670	+ }
486	671
487	672	synchronize_rcu();
488	673
489		- if (!cpumask_equal(pd_old->cpumask.pcpu, pd_new->cpumask.pcpu))
490		- notification_mask \|= PADATA_CPU_PARALLEL;
491		- if (!cpumask_equal(pd_old->cpumask.cbcpu, pd_new->cpumask.cbcpu))
492		- notification_mask \|= PADATA_CPU_SERIAL;
493		-
494		- if (atomic_dec_and_test(&pd_old->refcnt))
495		- padata_free_pd(pd_old);
496		-
497		- if (notification_mask)
498		- blocking_notifier_call_chain(&pinst->cpumask_change_notifier,
499		- notification_mask,
500		- &pd_new->cpumask);
	674	+ list_for_each_entry_continue_reverse(ps, &pinst->pslist, list)
	675	+ if (atomic_dec_and_test(&ps->opd->refcnt))
	676	+ padata_free_pd(ps->opd);
501	677
502	678	pinst->flags &= ~PADATA_RESET;
503		-}
504	679
505		-/**
506		- * padata_register_cpumask_notifier - Registers a notifier that will be called
507		- * if either pcpu or cbcpu or both cpumasks change.
508		- *
509		- * @pinst: A poineter to padata instance
510		- * @nblock: A pointer to notifier block.
511		- */
512		-int padata_register_cpumask_notifier(struct padata_instance *pinst,
513		- struct notifier_block *nblock)
514		-{
515		- return blocking_notifier_chain_register(&pinst->cpumask_change_notifier,
516		- nblock);
	680	+ return err;
517	681	}
518		-EXPORT_SYMBOL(padata_register_cpumask_notifier);
519		-
520		-/**
521		- * padata_unregister_cpumask_notifier - Unregisters cpumask notifier
522		- * registered earlier using padata_register_cpumask_notifier
523		- *
524		- * @pinst: A pointer to data instance.
525		- * @nlock: A pointer to notifier block.
526		- */
527		-int padata_unregister_cpumask_notifier(struct padata_instance *pinst,
528		- struct notifier_block *nblock)
529		-{
530		- return blocking_notifier_chain_unregister(
531		- &pinst->cpumask_change_notifier,
532		- nblock);
533		-}
534		-EXPORT_SYMBOL(padata_unregister_cpumask_notifier);
535		-
536	682
537	683	/* If cpumask contains no active cpu, we mark the instance as invalid. */
538	684	static bool padata_validate_cpumask(struct padata_instance *pinst,
..	..	@@ -552,7 +698,7 @@
552	698	cpumask_var_t cbcpumask)
553	699	{
554	700	int valid;
555		- struct parallel_data *pd;
	701	+ int err;
556	702
557	703	valid = padata_validate_cpumask(pinst, pcpumask);
558	704	if (!valid) {
..	..	@@ -565,29 +711,26 @@
565	711	__padata_stop(pinst);
566	712
567	713	out_replace:
568		- pd = padata_alloc_pd(pinst, pcpumask, cbcpumask);
569		- if (!pd)
570		- return -ENOMEM;
571		-
572	714	cpumask_copy(pinst->cpumask.pcpu, pcpumask);
573	715	cpumask_copy(pinst->cpumask.cbcpu, cbcpumask);
574	716
575		- padata_replace(pinst, pd);
	717	+ err = padata_setup_cpumasks(pinst) ?: padata_replace(pinst);
576	718
577	719	if (valid)
578	720	__padata_start(pinst);
579	721
580		- return 0;
	722	+ return err;
581	723	}
582	724
583	725	/**
584		- * padata_set_cpumask: Sets specified by @cpumask_type cpumask to the value
585		- * equivalent to @cpumask.
586		- *
	726	+ * padata_set_cpumask - Sets specified by @cpumask_type cpumask to the value
	727	+ * equivalent to @cpumask.
587	728	* @pinst: padata instance
588	729	* @cpumask_type: PADATA_CPU_SERIAL or PADATA_CPU_PARALLEL corresponding
589	730	* to parallel and serial cpumasks respectively.
590	731	* @cpumask: the cpumask to use
	732	+ *
	733	+ * Return: 0 on success or negative error code
591	734	*/
592	735	int padata_set_cpumask(struct padata_instance *pinst, int cpumask_type,
593	736	cpumask_var_t cpumask)
..	..	@@ -621,121 +764,37 @@
621	764	}
622	765	EXPORT_SYMBOL(padata_set_cpumask);
623	766
624		-/**
625		- * padata_start - start the parallel processing
626		- *
627		- * @pinst: padata instance to start
628		- */
629		-int padata_start(struct padata_instance *pinst)
630		-{
631		- int err = 0;
632		-
633		- mutex_lock(&pinst->lock);
634		-
635		- if (pinst->flags & PADATA_INVALID)
636		- err = -EINVAL;
637		-
638		- __padata_start(pinst);
639		-
640		- mutex_unlock(&pinst->lock);
641		-
642		- return err;
643		-}
644		-EXPORT_SYMBOL(padata_start);
645		-
646		-/**
647		- * padata_stop - stop the parallel processing
648		- *
649		- * @pinst: padata instance to stop
650		- */
651		-void padata_stop(struct padata_instance *pinst)
652		-{
653		- mutex_lock(&pinst->lock);
654		- __padata_stop(pinst);
655		- mutex_unlock(&pinst->lock);
656		-}
657		-EXPORT_SYMBOL(padata_stop);
658		-
659	767	#ifdef CONFIG_HOTPLUG_CPU
660	768
661	769	static int __padata_add_cpu(struct padata_instance *pinst, int cpu)
662	770	{
663		- struct parallel_data *pd;
	771	+ int err = 0;
664	772
665	773	if (cpumask_test_cpu(cpu, cpu_online_mask)) {
666		- pd = padata_alloc_pd(pinst, pinst->cpumask.pcpu,
667		- pinst->cpumask.cbcpu);
668		- if (!pd)
669		- return -ENOMEM;
670		-
671		- padata_replace(pinst, pd);
	774	+ err = padata_replace(pinst);
672	775
673	776	if (padata_validate_cpumask(pinst, pinst->cpumask.pcpu) &&
674	777	padata_validate_cpumask(pinst, pinst->cpumask.cbcpu))
675	778	__padata_start(pinst);
676	779	}
677	780
678		- return 0;
	781	+ return err;
679	782	}
680	783
681	784	static int __padata_remove_cpu(struct padata_instance *pinst, int cpu)
682	785	{
683		- struct parallel_data *pd = NULL;
	786	+ int err = 0;
684	787
685		- if (cpumask_test_cpu(cpu, cpu_online_mask)) {
686		-
	788	+ if (!cpumask_test_cpu(cpu, cpu_online_mask)) {
687	789	if (!padata_validate_cpumask(pinst, pinst->cpumask.pcpu) \|\|
688	790	!padata_validate_cpumask(pinst, pinst->cpumask.cbcpu))
689	791	__padata_stop(pinst);
690	792
691		- pd = padata_alloc_pd(pinst, pinst->cpumask.pcpu,
692		- pinst->cpumask.cbcpu);
693		- if (!pd)
694		- return -ENOMEM;
695		-
696		- padata_replace(pinst, pd);
697		-
698		- cpumask_clear_cpu(cpu, pd->cpumask.cbcpu);
699		- cpumask_clear_cpu(cpu, pd->cpumask.pcpu);
	793	+ err = padata_replace(pinst);
700	794	}
701		-
702		- return 0;
703		-}
704		-
705		- /**
706		- * padata_remove_cpu - remove a cpu from the one or both(serial and parallel)
707		- * padata cpumasks.
708		- *
709		- * @pinst: padata instance
710		- * @cpu: cpu to remove
711		- * @mask: bitmask specifying from which cpumask @cpu should be removed
712		- * The @mask may be any combination of the following flags:
713		- * PADATA_CPU_SERIAL - serial cpumask
714		- * PADATA_CPU_PARALLEL - parallel cpumask
715		- */
716		-int padata_remove_cpu(struct padata_instance *pinst, int cpu, int mask)
717		-{
718		- int err;
719		-
720		- if (!(mask & (PADATA_CPU_SERIAL \| PADATA_CPU_PARALLEL)))
721		- return -EINVAL;
722		-
723		- mutex_lock(&pinst->lock);
724		-
725		- get_online_cpus();
726		- if (mask & PADATA_CPU_SERIAL)
727		- cpumask_clear_cpu(cpu, pinst->cpumask.cbcpu);
728		- if (mask & PADATA_CPU_PARALLEL)
729		- cpumask_clear_cpu(cpu, pinst->cpumask.pcpu);
730		-
731		- err = __padata_remove_cpu(pinst, cpu);
732		- put_online_cpus();
733		-
734		- mutex_unlock(&pinst->lock);
735	795
736	796	return err;
737	797	}
738		-EXPORT_SYMBOL(padata_remove_cpu);
739	798
740	799	static inline int pinst_has_cpu(struct padata_instance *pinst, int cpu)
741	800	{
..	..	@@ -748,7 +807,7 @@
748	807	struct padata_instance *pinst;
749	808	int ret;
750	809
751		- pinst = hlist_entry_safe(node, struct padata_instance, node);
	810	+ pinst = hlist_entry_safe(node, struct padata_instance, cpu_online_node);
752	811	if (!pinst_has_cpu(pinst, cpu))
753	812	return 0;
754	813
..	..	@@ -758,12 +817,12 @@
758	817	return ret;
759	818	}
760	819
761		-static int padata_cpu_prep_down(unsigned int cpu, struct hlist_node *node)
	820	+static int padata_cpu_dead(unsigned int cpu, struct hlist_node *node)
762	821	{
763	822	struct padata_instance *pinst;
764	823	int ret;
765	824
766		- pinst = hlist_entry_safe(node, struct padata_instance, node);
	825	+ pinst = hlist_entry_safe(node, struct padata_instance, cpu_dead_node);
767	826	if (!pinst_has_cpu(pinst, cpu))
768	827	return 0;
769	828
..	..	@@ -779,13 +838,17 @@
779	838	static void __padata_free(struct padata_instance *pinst)
780	839	{
781	840	#ifdef CONFIG_HOTPLUG_CPU
782		- cpuhp_state_remove_instance_nocalls(hp_online, &pinst->node);
	841	+ cpuhp_state_remove_instance_nocalls(CPUHP_PADATA_DEAD,
	842	+ &pinst->cpu_dead_node);
	843	+ cpuhp_state_remove_instance_nocalls(hp_online, &pinst->cpu_online_node);
783	844	#endif
784	845
785		- padata_stop(pinst);
786		- padata_free_pd(pinst->pd);
	846	+ WARN_ON(!list_empty(&pinst->pslist));
	847	+
787	848	free_cpumask_var(pinst->cpumask.pcpu);
788	849	free_cpumask_var(pinst->cpumask.cbcpu);
	850	+ destroy_workqueue(pinst->serial_wq);
	851	+ destroy_workqueue(pinst->parallel_wq);
789	852	kfree(pinst);
790	853	}
791	854
..	..	@@ -872,6 +935,7 @@
872	935	&parallel_cpumask_attr.attr,
873	936	NULL,
874	937	};
	938	+ATTRIBUTE_GROUPS(padata_default);
875	939
876	940	static ssize_t padata_sysfs_show(struct kobject *kobj,
877	941	struct attribute attr, char buf)
..	..	@@ -910,92 +974,86 @@
910	974
911	975	static struct kobj_type padata_attr_type = {
912	976	.sysfs_ops = &padata_sysfs_ops,
913		- .default_attrs = padata_default_attrs,
	977	+ .default_groups = padata_default_groups,
914	978	.release = padata_sysfs_release,
915	979	};
916	980
917	981	/**
918		- * padata_alloc - allocate and initialize a padata instance and specify
919		- * cpumasks for serial and parallel workers.
	982	+ * padata_alloc - allocate and initialize a padata instance
	983	+ * @name: used to identify the instance
920	984	*
921		- * @wq: workqueue to use for the allocated padata instance
922		- * @pcpumask: cpumask that will be used for padata parallelization
923		- * @cbcpumask: cpumask that will be used for padata serialization
924		- *
925		- * Must be called from a cpus_read_lock() protected region
	985	+ * Return: new instance on success, NULL on error
926	986	*/
927		-static struct padata_instance padata_alloc(struct workqueue_struct wq,
928		- const struct cpumask *pcpumask,
929		- const struct cpumask *cbcpumask)
	987	+struct padata_instance padata_alloc(const char name)
930	988	{
931	989	struct padata_instance *pinst;
932		- struct parallel_data *pd = NULL;
933	990
934	991	pinst = kzalloc(sizeof(struct padata_instance), GFP_KERNEL);
935	992	if (!pinst)
936	993	goto err;
937	994
938		- if (!alloc_cpumask_var(&pinst->cpumask.pcpu, GFP_KERNEL))
	995	+ pinst->parallel_wq = alloc_workqueue("%s_parallel", WQ_UNBOUND, 0,
	996	+ name);
	997	+ if (!pinst->parallel_wq)
939	998	goto err_free_inst;
	999	+
	1000	+ get_online_cpus();
	1001	+
	1002	+ pinst->serial_wq = alloc_workqueue("%s_serial", WQ_MEM_RECLAIM \|
	1003	+ WQ_CPU_INTENSIVE, 1, name);
	1004	+ if (!pinst->serial_wq)
	1005	+ goto err_put_cpus;
	1006	+
	1007	+ if (!alloc_cpumask_var(&pinst->cpumask.pcpu, GFP_KERNEL))
	1008	+ goto err_free_serial_wq;
940	1009	if (!alloc_cpumask_var(&pinst->cpumask.cbcpu, GFP_KERNEL)) {
941	1010	free_cpumask_var(pinst->cpumask.pcpu);
942		- goto err_free_inst;
	1011	+ goto err_free_serial_wq;
943	1012	}
944		- if (!padata_validate_cpumask(pinst, pcpumask) \|\|
945		- !padata_validate_cpumask(pinst, cbcpumask))
	1013	+
	1014	+ INIT_LIST_HEAD(&pinst->pslist);
	1015	+
	1016	+ cpumask_copy(pinst->cpumask.pcpu, cpu_possible_mask);
	1017	+ cpumask_copy(pinst->cpumask.cbcpu, cpu_possible_mask);
	1018	+
	1019	+ if (padata_setup_cpumasks(pinst))
946	1020	goto err_free_masks;
947	1021
948		- pd = padata_alloc_pd(pinst, pcpumask, cbcpumask);
949		- if (!pd)
950		- goto err_free_masks;
	1022	+ __padata_start(pinst);
951	1023
952		- rcu_assign_pointer(pinst->pd, pd);
953		-
954		- pinst->wq = wq;
955		-
956		- cpumask_copy(pinst->cpumask.pcpu, pcpumask);
957		- cpumask_copy(pinst->cpumask.cbcpu, cbcpumask);
958		-
959		- pinst->flags = 0;
960		-
961		- BLOCKING_INIT_NOTIFIER_HEAD(&pinst->cpumask_change_notifier);
962	1024	kobject_init(&pinst->kobj, &padata_attr_type);
963	1025	mutex_init(&pinst->lock);
964	1026
965	1027	#ifdef CONFIG_HOTPLUG_CPU
966		- cpuhp_state_add_instance_nocalls_cpuslocked(hp_online, &pinst->node);
	1028	+ cpuhp_state_add_instance_nocalls_cpuslocked(hp_online,
	1029	+ &pinst->cpu_online_node);
	1030	+ cpuhp_state_add_instance_nocalls_cpuslocked(CPUHP_PADATA_DEAD,
	1031	+ &pinst->cpu_dead_node);
967	1032	#endif
	1033	+
	1034	+ put_online_cpus();
	1035	+
968	1036	return pinst;
969	1037
970	1038	err_free_masks:
971	1039	free_cpumask_var(pinst->cpumask.pcpu);
972	1040	free_cpumask_var(pinst->cpumask.cbcpu);
	1041	+err_free_serial_wq:
	1042	+ destroy_workqueue(pinst->serial_wq);
	1043	+err_put_cpus:
	1044	+ put_online_cpus();
	1045	+ destroy_workqueue(pinst->parallel_wq);
973	1046	err_free_inst:
974	1047	kfree(pinst);
975	1048	err:
976	1049	return NULL;
977	1050	}
978		-
979		-/**
980		- * padata_alloc_possible - Allocate and initialize padata instance.
981		- * Use the cpu_possible_mask for serial and
982		- * parallel workers.
983		- *
984		- * @wq: workqueue to use for the allocated padata instance
985		- *
986		- * Must be called from a cpus_read_lock() protected region
987		- */
988		-struct padata_instance padata_alloc_possible(struct workqueue_struct wq)
989		-{
990		- lockdep_assert_cpus_held();
991		- return padata_alloc(wq, cpu_possible_mask, cpu_possible_mask);
992		-}
993		-EXPORT_SYMBOL(padata_alloc_possible);
	1051	+EXPORT_SYMBOL(padata_alloc);
994	1052
995	1053	/**
996	1054	* padata_free - free a padata instance
997	1055	*
998		- * @padata_inst: padata instance to free
	1056	+ * @pinst: padata instance to free
999	1057	*/
1000	1058	void padata_free(struct padata_instance *pinst)
1001	1059	{
..	..	@@ -1003,25 +1061,99 @@
1003	1061	}
1004	1062	EXPORT_SYMBOL(padata_free);
1005	1063
1006		-#ifdef CONFIG_HOTPLUG_CPU
1007		-
1008		-static __init int padata_driver_init(void)
	1064	+/**
	1065	+ * padata_alloc_shell - Allocate and initialize padata shell.
	1066	+ *
	1067	+ * @pinst: Parent padata_instance object.
	1068	+ *
	1069	+ * Return: new shell on success, NULL on error
	1070	+ */
	1071	+struct padata_shell padata_alloc_shell(struct padata_instance pinst)
1009	1072	{
	1073	+ struct parallel_data *pd;
	1074	+ struct padata_shell *ps;
	1075	+
	1076	+ ps = kzalloc(sizeof(*ps), GFP_KERNEL);
	1077	+ if (!ps)
	1078	+ goto out;
	1079	+
	1080	+ ps->pinst = pinst;
	1081	+
	1082	+ get_online_cpus();
	1083	+ pd = padata_alloc_pd(ps);
	1084	+ put_online_cpus();
	1085	+
	1086	+ if (!pd)
	1087	+ goto out_free_ps;
	1088	+
	1089	+ mutex_lock(&pinst->lock);
	1090	+ RCU_INIT_POINTER(ps->pd, pd);
	1091	+ list_add(&ps->list, &pinst->pslist);
	1092	+ mutex_unlock(&pinst->lock);
	1093	+
	1094	+ return ps;
	1095	+
	1096	+out_free_ps:
	1097	+ kfree(ps);
	1098	+out:
	1099	+ return NULL;
	1100	+}
	1101	+EXPORT_SYMBOL(padata_alloc_shell);
	1102	+
	1103	+/**
	1104	+ * padata_free_shell - free a padata shell
	1105	+ *
	1106	+ * @ps: padata shell to free
	1107	+ */
	1108	+void padata_free_shell(struct padata_shell *ps)
	1109	+{
	1110	+ if (!ps)
	1111	+ return;
	1112	+
	1113	+ mutex_lock(&ps->pinst->lock);
	1114	+ list_del(&ps->list);
	1115	+ padata_free_pd(rcu_dereference_protected(ps->pd, 1));
	1116	+ mutex_unlock(&ps->pinst->lock);
	1117	+
	1118	+ kfree(ps);
	1119	+}
	1120	+EXPORT_SYMBOL(padata_free_shell);
	1121	+
	1122	+void __init padata_init(void)
	1123	+{
	1124	+ unsigned int i, possible_cpus;
	1125	+#ifdef CONFIG_HOTPLUG_CPU
1010	1126	int ret;
1011	1127
1012	1128	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "padata:online",
1013		- padata_cpu_online,
1014		- padata_cpu_prep_down);
	1129	+ padata_cpu_online, NULL);
1015	1130	if (ret < 0)
1016		- return ret;
	1131	+ goto err;
1017	1132	hp_online = ret;
1018		- return 0;
1019		-}
1020		-module_init(padata_driver_init);
1021	1133
1022		-static __exit void padata_driver_exit(void)
1023		-{
1024		- cpuhp_remove_multi_state(hp_online);
1025		-}
1026		-module_exit(padata_driver_exit);
	1134	+ ret = cpuhp_setup_state_multi(CPUHP_PADATA_DEAD, "padata:dead",
	1135	+ NULL, padata_cpu_dead);
	1136	+ if (ret < 0)
	1137	+ goto remove_online_state;
1027	1138	#endif
	1139	+
	1140	+ possible_cpus = num_possible_cpus();
	1141	+ padata_works = kmalloc_array(possible_cpus, sizeof(struct padata_work),
	1142	+ GFP_KERNEL);
	1143	+ if (!padata_works)
	1144	+ goto remove_dead_state;
	1145	+
	1146	+ for (i = 0; i < possible_cpus; ++i)
	1147	+ list_add(&padata_works[i].pw_list, &padata_free_works);
	1148	+
	1149	+ return;
	1150	+
	1151	+remove_dead_state:
	1152	+#ifdef CONFIG_HOTPLUG_CPU
	1153	+ cpuhp_remove_multi_state(CPUHP_PADATA_DEAD);
	1154	+remove_online_state:
	1155	+ cpuhp_remove_multi_state(hp_online);
	1156	+err:
	1157	+#endif
	1158	+ pr_warn("padata: initialization failed\n");
	1159	+}