kernel/padata.c

   1 /*
   2  * padata.c - generic interface to process data streams in parallel
   3  *
   4  * See Documentation/padata.txt for an api documentation.
   5  *
   6  * Copyright (C) 2008, 2009 secunet Security Networks AG
   7  * Copyright (C) 2008, 2009 Steffen Klassert <steffen.klassert@secunet.com>
   8  *
   9  * This program is free software; you can redistribute it and/or modify it
  10  * under the terms and conditions of the GNU General Public License,
  11  * version 2, as published by the Free Software Foundation.
  12  *
  13  * This program is distributed in the hope it will be useful, but WITHOUT
  14  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  15  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  16  * more details.
  17  *
  18  * You should have received a copy of the GNU General Public License along with
  19  * this program; if not, write to the Free Software Foundation, Inc.,
  20  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  21  */
  22
  23 #include <linux/export.h>
  24 #include <linux/cpumask.h>
  25 #include <linux/err.h>
  26 #include <linux/cpu.h>
  27 #include <linux/padata.h>
  28 #include <linux/mutex.h>
  29 #include <linux/sched.h>
  30 #include <linux/slab.h>
  31 #include <linux/sysfs.h>
  32 #include <linux/rcupdate.h>
  33 #include <linux/module.h>
  34
  35 #define MAX_OBJ_NUM 1000
  36
  37 static void padata_free_pd(struct parallel_data *pd);
  38
  39 static int padata_index_to_cpu(struct parallel_data *pd, int cpu_index)
  40 {
  41         int cpu, target_cpu;
  42
  43         target_cpu = cpumask_first(pd->cpumask.pcpu);
  44         for (cpu = 0; cpu < cpu_index; cpu++)
  45                 target_cpu = cpumask_next(target_cpu, pd->cpumask.pcpu);
  46
  47         return target_cpu;
  48 }
  49
  50 static int padata_cpu_hash(struct parallel_data *pd)
  51 {
  52         unsigned int seq_nr;
  53         int cpu_index;
  54
  55         /*
  56          * Hash the sequence numbers to the cpus by taking
  57          * seq_nr mod. number of cpus in use.
  58          */
  59
  60         seq_nr = atomic_inc_return(&pd->seq_nr);
  61         cpu_index = seq_nr % cpumask_weight(pd->cpumask.pcpu);
  62
  63         return padata_index_to_cpu(pd, cpu_index);
  64 }
  65
  66 static void padata_parallel_worker(struct work_struct *parallel_work)
  67 {
  68         struct padata_parallel_queue *pqueue;
  69         LIST_HEAD(local_list);
  70
  71         local_bh_disable();
  72         pqueue = container_of(parallel_work,
  73                               struct padata_parallel_queue, work);
  74
  75         spin_lock(&pqueue->parallel.lock);
  76         list_replace_init(&pqueue->parallel.list, &local_list);
  77         spin_unlock(&pqueue->parallel.lock);
  78
  79         while (!list_empty(&local_list)) {
  80                 struct padata_priv *padata;
  81
  82                 padata = list_entry(local_list.next,
  83                                     struct padata_priv, list);
  84
  85                 list_del_init(&padata->list);
  86
  87                 padata->parallel(padata);
  88         }
  89
  90         local_bh_enable();
  91 }
  92
  93 /**
  94  * padata_do_parallel - padata parallelization function
  95  *
  96  * @pinst: padata instance
  97  * @padata: object to be parallelized
  98  * @cb_cpu: cpu the serialization callback function will run on,
  99  *          must be in the serial cpumask of padata(i.e. cpumask.cbcpu).
 100  *
 101  * The parallelization callback function will run with BHs off.
 102  * Note: Every object which is parallelized by padata_do_parallel
 103  * must be seen by padata_do_serial.
 104  */
 105 int padata_do_parallel(struct padata_instance *pinst,
 106                        struct padata_priv *padata, int cb_cpu)
 107 {
 108         int target_cpu, err;
 109         struct padata_parallel_queue *queue;
 110         struct parallel_data *pd;
 111
 112         rcu_read_lock_bh();
 113
 114         pd = rcu_dereference_bh(pinst->pd);
 115
 116         err = -EINVAL;
 117         if (!(pinst->flags & PADATA_INIT) || pinst->flags & PADATA_INVALID)
 118                 goto out;
 119
 120         if (!cpumask_test_cpu(cb_cpu, pd->cpumask.cbcpu))
 121                 goto out;
 122
 123         err =  -EBUSY;
 124         if ((pinst->flags & PADATA_RESET))
 125                 goto out;
 126
 127         if (atomic_read(&pd->refcnt) >= MAX_OBJ_NUM)
 128                 goto out;
 129
 130         err = 0;
 131         atomic_inc(&pd->refcnt);
 132         padata->pd = pd;
 133         padata->cb_cpu = cb_cpu;
 134
 135         target_cpu = padata_cpu_hash(pd);
 136         padata->cpu = target_cpu;
 137         queue = per_cpu_ptr(pd->pqueue, target_cpu);
 138
 139         spin_lock(&queue->parallel.lock);
 140         list_add_tail(&padata->list, &queue->parallel.list);
 141         spin_unlock(&queue->parallel.lock);
 142
 143         queue_work_on(target_cpu, pinst->wq, &queue->work);
 144
 145 out:
 146         rcu_read_unlock_bh();
 147
 148         return err;
 149 }
 150 EXPORT_SYMBOL(padata_do_parallel);
 151
 152 /*
 153  * padata_get_next - Get the next object that needs serialization.
 154  *
 155  * Return values are:
 156  *
 157  * A pointer to the control struct of the next object that needs
 158  * serialization, if present in one of the percpu reorder queues.
 159  *
 160  * -EINPROGRESS, if the next object that needs serialization will
 161  *  be parallel processed by another cpu and is not yet present in
 162  *  the cpu's reorder queue.
 163  *
 164  * -ENODATA, if this cpu has to do the parallel processing for
 165  *  the next object.
 166  */
 167 static struct padata_priv *padata_get_next(struct parallel_data *pd)
 168 {
 169         struct padata_parallel_queue *next_queue;
 170         struct padata_priv *padata;
 171         struct padata_list *reorder;
 172         int cpu = pd->cpu;
 173
 174         next_queue = per_cpu_ptr(pd->pqueue, cpu);
 175         reorder = &next_queue->reorder;
 176
 177         spin_lock(&reorder->lock);
 178         if (!list_empty(&reorder->list)) {
 179                 padata = list_entry(reorder->list.next,
 180                                     struct padata_priv, list);
 181
 182                 list_del_init(&padata->list);
 183                 atomic_dec(&pd->reorder_objects);
 184
 185                 pd->cpu = cpumask_next_wrap(cpu, pd->cpumask.pcpu, -1,
 186                                             false);
 187
 188                 spin_unlock(&reorder->lock);
 189                 goto out;
 190         }
 191         spin_unlock(&reorder->lock);
 192
 193         if (__this_cpu_read(pd->pqueue->cpu_index) == next_queue->cpu_index) {
 194                 padata = ERR_PTR(-ENODATA);
 195                 goto out;
 196         }
 197
 198         padata = ERR_PTR(-EINPROGRESS);
 199 out:
 200         return padata;
 201 }
 202
 203 static void padata_reorder(struct parallel_data *pd)
 204 {
 205         int cb_cpu;
 206         struct padata_priv *padata;
 207         struct padata_serial_queue *squeue;
 208         struct padata_instance *pinst = pd->pinst;
 209         struct padata_parallel_queue *next_queue;
 210
 211         /*
 212          * We need to ensure that only one cpu can work on dequeueing of
 213          * the reorder queue the time. Calculating in which percpu reorder
 214          * queue the next object will arrive takes some time. A spinlock
 215          * would be highly contended. Also it is not clear in which order
 216          * the objects arrive to the reorder queues. So a cpu could wait to
 217          * get the lock just to notice that there is nothing to do at the
 218          * moment. Therefore we use a trylock and let the holder of the lock
 219          * care for all the objects enqueued during the holdtime of the lock.
 220          */
 221         if (!spin_trylock_bh(&pd->lock))
 222                 return;
 223
 224         while (1) {
 225                 padata = padata_get_next(pd);
 226
 227                 /*
 228                  * If the next object that needs serialization is parallel
 229                  * processed by another cpu and is still on it's way to the
 230                  * cpu's reorder queue, nothing to do for now.
 231                  */
 232                 if (PTR_ERR(padata) == -EINPROGRESS)
 233                         break;
 234
 235                 /*
 236                  * This cpu has to do the parallel processing of the next
 237                  * object. It's waiting in the cpu's parallelization queue,
 238                  * so exit immediately.
 239                  */
 240                 if (PTR_ERR(padata) == -ENODATA) {
 241                         spin_unlock_bh(&pd->lock);
 242                         return;
 243                 }
 244
 245                 cb_cpu = padata->cb_cpu;
 246                 squeue = per_cpu_ptr(pd->squeue, cb_cpu);
 247
 248                 spin_lock(&squeue->serial.lock);
 249                 list_add_tail(&padata->list, &squeue->serial.list);
 250                 spin_unlock(&squeue->serial.lock);
 251
 252                 queue_work_on(cb_cpu, pinst->wq, &squeue->work);
 253         }
 254
 255         spin_unlock_bh(&pd->lock);
 256
 257         /*
 258          * The next object that needs serialization might have arrived to
 259          * the reorder queues in the meantime.
 260          *
 261          * Ensure reorder queue is read after pd->lock is dropped so we see
 262          * new objects from another task in padata_do_serial.  Pairs with
 263          * smp_mb__after_atomic in padata_do_serial.
 264          */
 265         smp_mb();
 266
 267         next_queue = per_cpu_ptr(pd->pqueue, pd->cpu);
 268         if (!list_empty(&next_queue->reorder.list))
 269                 queue_work(pinst->wq, &pd->reorder_work);
 270 }
 271
 272 static void invoke_padata_reorder(struct work_struct *work)
 273 {
 274         struct parallel_data *pd;
 275
 276         local_bh_disable();
 277         pd = container_of(work, struct parallel_data, reorder_work);
 278         padata_reorder(pd);
 279         local_bh_enable();
 280 }
 281
 282 static void padata_serial_worker(struct work_struct *serial_work)
 283 {
 284         struct padata_serial_queue *squeue;
 285         struct parallel_data *pd;
 286         LIST_HEAD(local_list);
 287         int cnt;
 288
 289         local_bh_disable();
 290         squeue = container_of(serial_work, struct padata_serial_queue, work);
 291         pd = squeue->pd;
 292
 293         spin_lock(&squeue->serial.lock);
 294         list_replace_init(&squeue->serial.list, &local_list);
 295         spin_unlock(&squeue->serial.lock);
 296
 297         cnt = 0;
 298
 299         while (!list_empty(&local_list)) {
 300                 struct padata_priv *padata;
 301
 302                 padata = list_entry(local_list.next,
 303                                     struct padata_priv, list);
 304
 305                 list_del_init(&padata->list);
 306
 307                 padata->serial(padata);
 308                 cnt++;
 309         }
 310         local_bh_enable();
 311
 312         if (atomic_sub_and_test(cnt, &pd->refcnt))
 313                 padata_free_pd(pd);
 314 }
 315
 316 /**
 317  * padata_do_serial - padata serialization function
 318  *
 319  * @padata: object to be serialized.
 320  *
 321  * padata_do_serial must be called for every parallelized object.
 322  * The serialization callback function will run with BHs off.
 323  */
 324 void padata_do_serial(struct padata_priv *padata)
 325 {
 326         struct parallel_data *pd = padata->pd;
 327         struct padata_parallel_queue *pqueue = per_cpu_ptr(pd->pqueue,
 328                                                            padata->cpu);
 329
 330         spin_lock(&pqueue->reorder.lock);
 331         list_add_tail(&padata->list, &pqueue->reorder.list);
 332         atomic_inc(&pd->reorder_objects);
 333         spin_unlock(&pqueue->reorder.lock);
 334
 335         /*
 336          * Ensure the addition to the reorder list is ordered correctly
 337          * with the trylock of pd->lock in padata_reorder.  Pairs with smp_mb
 338          * in padata_reorder.
 339          */
 340         smp_mb__after_atomic();
 341
 342         padata_reorder(pd);
 343 }
 344 EXPORT_SYMBOL(padata_do_serial);
 345
 346 static int padata_setup_cpumasks(struct parallel_data *pd,
 347                                  const struct cpumask *pcpumask,
 348                                  const struct cpumask *cbcpumask)
 349 {
 350         if (!alloc_cpumask_var(&pd->cpumask.pcpu, GFP_KERNEL))
 351                 return -ENOMEM;
 352
 353         cpumask_and(pd->cpumask.pcpu, pcpumask, cpu_online_mask);
 354         if (!alloc_cpumask_var(&pd->cpumask.cbcpu, GFP_KERNEL)) {
 355                 free_cpumask_var(pd->cpumask.pcpu);
 356                 return -ENOMEM;
 357         }
 358
 359         cpumask_and(pd->cpumask.cbcpu, cbcpumask, cpu_online_mask);
 360         return 0;
 361 }
 362
 363 static void __padata_list_init(struct padata_list *pd_list)
 364 {
 365         INIT_LIST_HEAD(&pd_list->list);
 366         spin_lock_init(&pd_list->lock);
 367 }
 368
 369 /* Initialize all percpu queues used by serial workers */
 370 static void padata_init_squeues(struct parallel_data *pd)
 371 {
 372         int cpu;
 373         struct padata_serial_queue *squeue;
 374
 375         for_each_cpu(cpu, pd->cpumask.cbcpu) {
 376                 squeue = per_cpu_ptr(pd->squeue, cpu);
 377                 squeue->pd = pd;
 378                 __padata_list_init(&squeue->serial);
 379                 INIT_WORK(&squeue->work, padata_serial_worker);
 380         }
 381 }
 382
 383 /* Initialize all percpu queues used by parallel workers */
 384 static void padata_init_pqueues(struct parallel_data *pd)
 385 {
 386         int cpu_index, cpu;
 387         struct padata_parallel_queue *pqueue;
 388
 389         cpu_index = 0;
 390         for_each_possible_cpu(cpu) {
 391                 pqueue = per_cpu_ptr(pd->pqueue, cpu);
 392
 393                 if (!cpumask_test_cpu(cpu, pd->cpumask.pcpu)) {
 394                         pqueue->cpu_index = -1;
 395                         continue;
 396                 }
 397
 398                 pqueue->cpu_index = cpu_index;
 399                 cpu_index++;
 400
 401                 __padata_list_init(&pqueue->reorder);
 402                 __padata_list_init(&pqueue->parallel);
 403                 INIT_WORK(&pqueue->work, padata_parallel_worker);
 404                 atomic_set(&pqueue->num_obj, 0);
 405         }
 406 }
 407
 408 /* Allocate and initialize the internal cpumask dependend resources. */
 409 static struct parallel_data *padata_alloc_pd(struct padata_instance *pinst,
 410                                              const struct cpumask *pcpumask,
 411                                              const struct cpumask *cbcpumask)
 412 {
 413         struct parallel_data *pd;
 414
 415         pd = kzalloc(sizeof(struct parallel_data), GFP_KERNEL);
 416         if (!pd)
 417                 goto err;
 418
 419         pd->pqueue = alloc_percpu(struct padata_parallel_queue);
 420         if (!pd->pqueue)
 421                 goto err_free_pd;
 422
 423         pd->squeue = alloc_percpu(struct padata_serial_queue);
 424         if (!pd->squeue)
 425                 goto err_free_pqueue;
 426         if (padata_setup_cpumasks(pd, pcpumask, cbcpumask) < 0)
 427                 goto err_free_squeue;
 428
 429         padata_init_pqueues(pd);
 430         padata_init_squeues(pd);
 431         atomic_set(&pd->seq_nr, -1);
 432         atomic_set(&pd->reorder_objects, 0);
 433         atomic_set(&pd->refcnt, 1);
 434         pd->pinst = pinst;
 435         spin_lock_init(&pd->lock);
 436         pd->cpu = cpumask_first(pd->cpumask.pcpu);
 437         INIT_WORK(&pd->reorder_work, invoke_padata_reorder);
 438
 439         return pd;
 440
 441 err_free_squeue:
 442         free_percpu(pd->squeue);
 443 err_free_pqueue:
 444         free_percpu(pd->pqueue);
 445 err_free_pd:
 446         kfree(pd);
 447 err:
 448         return NULL;
 449 }
 450
 451 static void padata_free_pd(struct parallel_data *pd)
 452 {
 453         free_cpumask_var(pd->cpumask.pcpu);
 454         free_cpumask_var(pd->cpumask.cbcpu);
 455         free_percpu(pd->pqueue);
 456         free_percpu(pd->squeue);
 457         kfree(pd);
 458 }
 459
 460 static void __padata_start(struct padata_instance *pinst)
 461 {
 462         pinst->flags |= PADATA_INIT;
 463 }
 464
 465 static void __padata_stop(struct padata_instance *pinst)
 466 {
 467         if (!(pinst->flags & PADATA_INIT))
 468                 return;
 469
 470         pinst->flags &= ~PADATA_INIT;
 471
 472         synchronize_rcu();
 473 }
 474
 475 /* Replace the internal control structure with a new one. */
 476 static void padata_replace(struct padata_instance *pinst,
 477                            struct parallel_data *pd_new)
 478 {
 479         struct parallel_data *pd_old = pinst->pd;
 480         int notification_mask = 0;
 481
 482         pinst->flags |= PADATA_RESET;
 483
 484         rcu_assign_pointer(pinst->pd, pd_new);
 485
 486         synchronize_rcu();
 487
 488         if (!cpumask_equal(pd_old->cpumask.pcpu, pd_new->cpumask.pcpu))
 489                 notification_mask |= PADATA_CPU_PARALLEL;
 490         if (!cpumask_equal(pd_old->cpumask.cbcpu, pd_new->cpumask.cbcpu))
 491                 notification_mask |= PADATA_CPU_SERIAL;
 492
 493         if (atomic_dec_and_test(&pd_old->refcnt))
 494                 padata_free_pd(pd_old);
 495
 496         if (notification_mask)
 497                 blocking_notifier_call_chain(&pinst->cpumask_change_notifier,
 498                                              notification_mask,
 499                                              &pd_new->cpumask);
 500
 501         pinst->flags &= ~PADATA_RESET;
 502 }
 503
 504 /**
 505  * padata_register_cpumask_notifier - Registers a notifier that will be called
 506  *                             if either pcpu or cbcpu or both cpumasks change.
 507  *
 508  * @pinst: A poineter to padata instance
 509  * @nblock: A pointer to notifier block.
 510  */
 511 int padata_register_cpumask_notifier(struct padata_instance *pinst,
 512                                      struct notifier_block *nblock)
 513 {
 514         return blocking_notifier_chain_register(&pinst->cpumask_change_notifier,
 515                                                 nblock);
 516 }
 517 EXPORT_SYMBOL(padata_register_cpumask_notifier);
 518
 519 /**
 520  * padata_unregister_cpumask_notifier - Unregisters cpumask notifier
 521  *        registered earlier  using padata_register_cpumask_notifier
 522  *
 523  * @pinst: A pointer to data instance.
 524  * @nlock: A pointer to notifier block.
 525  */
 526 int padata_unregister_cpumask_notifier(struct padata_instance *pinst,
 527                                        struct notifier_block *nblock)
 528 {
 529         return blocking_notifier_chain_unregister(
 530                 &pinst->cpumask_change_notifier,
 531                 nblock);
 532 }
 533 EXPORT_SYMBOL(padata_unregister_cpumask_notifier);
 534
 535
 536 /* If cpumask contains no active cpu, we mark the instance as invalid. */
 537 static bool padata_validate_cpumask(struct padata_instance *pinst,
 538                                     const struct cpumask *cpumask)
 539 {
 540         if (!cpumask_intersects(cpumask, cpu_online_mask)) {
 541                 pinst->flags |= PADATA_INVALID;
 542                 return false;
 543         }
 544
 545         pinst->flags &= ~PADATA_INVALID;
 546         return true;
 547 }
 548
 549 static int __padata_set_cpumasks(struct padata_instance *pinst,
 550                                  cpumask_var_t pcpumask,
 551                                  cpumask_var_t cbcpumask)
 552 {
 553         int valid;
 554         struct parallel_data *pd;
 555
 556         valid = padata_validate_cpumask(pinst, pcpumask);
 557         if (!valid) {
 558                 __padata_stop(pinst);
 559                 goto out_replace;
 560         }
 561
 562         valid = padata_validate_cpumask(pinst, cbcpumask);
 563         if (!valid)
 564                 __padata_stop(pinst);
 565
 566 out_replace:
 567         pd = padata_alloc_pd(pinst, pcpumask, cbcpumask);
 568         if (!pd)
 569                 return -ENOMEM;
 570
 571         cpumask_copy(pinst->cpumask.pcpu, pcpumask);
 572         cpumask_copy(pinst->cpumask.cbcpu, cbcpumask);
 573
 574         padata_replace(pinst, pd);
 575
 576         if (valid)
 577                 __padata_start(pinst);
 578
 579         return 0;
 580 }
 581
 582 /**
 583  * padata_set_cpumask: Sets specified by @cpumask_type cpumask to the value
 584  *                     equivalent to @cpumask.
 585  *
 586  * @pinst: padata instance
 587  * @cpumask_type: PADATA_CPU_SERIAL or PADATA_CPU_PARALLEL corresponding
 588  *                to parallel and serial cpumasks respectively.
 589  * @cpumask: the cpumask to use
 590  */
 591 int padata_set_cpumask(struct padata_instance *pinst, int cpumask_type,
 592                        cpumask_var_t cpumask)
 593 {
 594         struct cpumask *serial_mask, *parallel_mask;
 595         int err = -EINVAL;
 596
 597         get_online_cpus();
 598         mutex_lock(&pinst->lock);
 599
 600         switch (cpumask_type) {
 601         case PADATA_CPU_PARALLEL:
 602                 serial_mask = pinst->cpumask.cbcpu;
 603                 parallel_mask = cpumask;
 604                 break;
 605         case PADATA_CPU_SERIAL:
 606                 parallel_mask = pinst->cpumask.pcpu;
 607                 serial_mask = cpumask;
 608                 break;
 609         default:
 610                  goto out;
 611         }
 612
 613         err =  __padata_set_cpumasks(pinst, parallel_mask, serial_mask);
 614
 615 out:
 616         mutex_unlock(&pinst->lock);
 617         put_online_cpus();
 618
 619         return err;
 620 }
 621 EXPORT_SYMBOL(padata_set_cpumask);
 622
 623 /**
 624  * padata_start - start the parallel processing
 625  *
 626  * @pinst: padata instance to start
 627  */
 628 int padata_start(struct padata_instance *pinst)
 629 {
 630         int err = 0;
 631
 632         mutex_lock(&pinst->lock);
 633
 634         if (pinst->flags & PADATA_INVALID)
 635                 err = -EINVAL;
 636
 637          __padata_start(pinst);
 638
 639         mutex_unlock(&pinst->lock);
 640
 641         return err;
 642 }
 643 EXPORT_SYMBOL(padata_start);
 644
 645 /**
 646  * padata_stop - stop the parallel processing
 647  *
 648  * @pinst: padata instance to stop
 649  */
 650 void padata_stop(struct padata_instance *pinst)
 651 {
 652         mutex_lock(&pinst->lock);
 653         __padata_stop(pinst);
 654         mutex_unlock(&pinst->lock);
 655 }
 656 EXPORT_SYMBOL(padata_stop);
 657
 658 #ifdef CONFIG_HOTPLUG_CPU
 659
 660 static int __padata_add_cpu(struct padata_instance *pinst, int cpu)
 661 {
 662         struct parallel_data *pd;
 663
 664         if (cpumask_test_cpu(cpu, cpu_online_mask)) {
 665                 pd = padata_alloc_pd(pinst, pinst->cpumask.pcpu,
 666                                      pinst->cpumask.cbcpu);
 667                 if (!pd)
 668                         return -ENOMEM;
 669
 670                 padata_replace(pinst, pd);
 671
 672                 if (padata_validate_cpumask(pinst, pinst->cpumask.pcpu) &&
 673                     padata_validate_cpumask(pinst, pinst->cpumask.cbcpu))
 674                         __padata_start(pinst);
 675         }
 676
 677         return 0;
 678 }
 679
 680 static int __padata_remove_cpu(struct padata_instance *pinst, int cpu)
 681 {
 682         struct parallel_data *pd = NULL;
 683
 684         if (cpumask_test_cpu(cpu, cpu_online_mask)) {
 685
 686                 if (!padata_validate_cpumask(pinst, pinst->cpumask.pcpu) ||
 687                     !padata_validate_cpumask(pinst, pinst->cpumask.cbcpu))
 688                         __padata_stop(pinst);
 689
 690                 pd = padata_alloc_pd(pinst, pinst->cpumask.pcpu,
 691                                      pinst->cpumask.cbcpu);
 692                 if (!pd)
 693                         return -ENOMEM;
 694
 695                 padata_replace(pinst, pd);
 696
 697                 cpumask_clear_cpu(cpu, pd->cpumask.cbcpu);
 698                 cpumask_clear_cpu(cpu, pd->cpumask.pcpu);
 699         }
 700
 701         return 0;
 702 }
 703
 704  /**
 705  * padata_remove_cpu - remove a cpu from the one or both(serial and parallel)
 706  *                     padata cpumasks.
 707  *
 708  * @pinst: padata instance
 709  * @cpu: cpu to remove
 710  * @mask: bitmask specifying from which cpumask @cpu should be removed
 711  *        The @mask may be any combination of the following flags:
 712  *          PADATA_CPU_SERIAL   - serial cpumask
 713  *          PADATA_CPU_PARALLEL - parallel cpumask
 714  */
 715 int padata_remove_cpu(struct padata_instance *pinst, int cpu, int mask)
 716 {
 717         int err;
 718
 719         if (!(mask & (PADATA_CPU_SERIAL | PADATA_CPU_PARALLEL)))
 720                 return -EINVAL;
 721
 722         mutex_lock(&pinst->lock);
 723
 724         get_online_cpus();
 725         if (mask & PADATA_CPU_SERIAL)
 726                 cpumask_clear_cpu(cpu, pinst->cpumask.cbcpu);
 727         if (mask & PADATA_CPU_PARALLEL)
 728                 cpumask_clear_cpu(cpu, pinst->cpumask.pcpu);
 729
 730         err = __padata_remove_cpu(pinst, cpu);
 731         put_online_cpus();
 732
 733         mutex_unlock(&pinst->lock);
 734
 735         return err;
 736 }
 737 EXPORT_SYMBOL(padata_remove_cpu);
 738
 739 static inline int pinst_has_cpu(struct padata_instance *pinst, int cpu)
 740 {
 741         return cpumask_test_cpu(cpu, pinst->cpumask.pcpu) ||
 742                 cpumask_test_cpu(cpu, pinst->cpumask.cbcpu);
 743 }
 744
 745 static int padata_cpu_online(unsigned int cpu, struct hlist_node *node)
 746 {
 747         struct padata_instance *pinst;
 748         int ret;
 749
 750         pinst = hlist_entry_safe(node, struct padata_instance, node);
 751         if (!pinst_has_cpu(pinst, cpu))
 752                 return 0;
 753
 754         mutex_lock(&pinst->lock);
 755         ret = __padata_add_cpu(pinst, cpu);
 756         mutex_unlock(&pinst->lock);
 757         return ret;
 758 }
 759
 760 static int padata_cpu_prep_down(unsigned int cpu, struct hlist_node *node)
 761 {
 762         struct padata_instance *pinst;
 763         int ret;
 764
 765         pinst = hlist_entry_safe(node, struct padata_instance, node);
 766         if (!pinst_has_cpu(pinst, cpu))
 767                 return 0;
 768
 769         mutex_lock(&pinst->lock);
 770         ret = __padata_remove_cpu(pinst, cpu);
 771         mutex_unlock(&pinst->lock);
 772         return ret;
 773 }
 774
 775 static enum cpuhp_state hp_online;
 776 #endif
 777
 778 static void __padata_free(struct padata_instance *pinst)
 779 {
 780 #ifdef CONFIG_HOTPLUG_CPU
 781         cpuhp_state_remove_instance_nocalls(hp_online, &pinst->node);
 782 #endif
 783
 784         padata_stop(pinst);
 785         padata_free_pd(pinst->pd);
 786         free_cpumask_var(pinst->cpumask.pcpu);
 787         free_cpumask_var(pinst->cpumask.cbcpu);
 788         kfree(pinst);
 789 }
 790
 791 #define kobj2pinst(_kobj)                                       \
 792         container_of(_kobj, struct padata_instance, kobj)
 793 #define attr2pentry(_attr)                                      \
 794         container_of(_attr, struct padata_sysfs_entry, attr)
 795
 796 static void padata_sysfs_release(struct kobject *kobj)
 797 {
 798         struct padata_instance *pinst = kobj2pinst(kobj);
 799         __padata_free(pinst);
 800 }
 801
 802 struct padata_sysfs_entry {
 803         struct attribute attr;
 804         ssize_t (*show)(struct padata_instance *, struct attribute *, char *);
 805         ssize_t (*store)(struct padata_instance *, struct attribute *,
 806                          const char *, size_t);
 807 };
 808
 809 static ssize_t show_cpumask(struct padata_instance *pinst,
 810                             struct attribute *attr,  char *buf)
 811 {
 812         struct cpumask *cpumask;
 813         ssize_t len;
 814
 815         mutex_lock(&pinst->lock);
 816         if (!strcmp(attr->name, "serial_cpumask"))
 817                 cpumask = pinst->cpumask.cbcpu;
 818         else
 819                 cpumask = pinst->cpumask.pcpu;
 820
 821         len = snprintf(buf, PAGE_SIZE, "%*pb\n",
 822                        nr_cpu_ids, cpumask_bits(cpumask));
 823         mutex_unlock(&pinst->lock);
 824         return len < PAGE_SIZE ? len : -EINVAL;
 825 }
 826
 827 static ssize_t store_cpumask(struct padata_instance *pinst,
 828                              struct attribute *attr,
 829                              const char *buf, size_t count)
 830 {
 831         cpumask_var_t new_cpumask;
 832         ssize_t ret;
 833         int mask_type;
 834
 835         if (!alloc_cpumask_var(&new_cpumask, GFP_KERNEL))
 836                 return -ENOMEM;
 837
 838         ret = bitmap_parse(buf, count, cpumask_bits(new_cpumask),
 839                            nr_cpumask_bits);
 840         if (ret < 0)
 841                 goto out;
 842
 843         mask_type = !strcmp(attr->name, "serial_cpumask") ?
 844                 PADATA_CPU_SERIAL : PADATA_CPU_PARALLEL;
 845         ret = padata_set_cpumask(pinst, mask_type, new_cpumask);
 846         if (!ret)
 847                 ret = count;
 848
 849 out:
 850         free_cpumask_var(new_cpumask);
 851         return ret;
 852 }
 853
 854 #define PADATA_ATTR_RW(_name, _show_name, _store_name)          \
 855         static struct padata_sysfs_entry _name##_attr =         \
 856                 __ATTR(_name, 0644, _show_name, _store_name)
 857 #define PADATA_ATTR_RO(_name, _show_name)               \
 858         static struct padata_sysfs_entry _name##_attr = \
 859                 __ATTR(_name, 0400, _show_name, NULL)
 860
 861 PADATA_ATTR_RW(serial_cpumask, show_cpumask, store_cpumask);
 862 PADATA_ATTR_RW(parallel_cpumask, show_cpumask, store_cpumask);
 863
 864 /*
 865  * Padata sysfs provides the following objects:
 866  * serial_cpumask   [RW] - cpumask for serial workers
 867  * parallel_cpumask [RW] - cpumask for parallel workers
 868  */
 869 static struct attribute *padata_default_attrs[] = {
 870         &serial_cpumask_attr.attr,
 871         &parallel_cpumask_attr.attr,
 872         NULL,
 873 };
 874
 875 static ssize_t padata_sysfs_show(struct kobject *kobj,
 876                                  struct attribute *attr, char *buf)
 877 {
 878         struct padata_instance *pinst;
 879         struct padata_sysfs_entry *pentry;
 880         ssize_t ret = -EIO;
 881
 882         pinst = kobj2pinst(kobj);
 883         pentry = attr2pentry(attr);
 884         if (pentry->show)
 885                 ret = pentry->show(pinst, attr, buf);
 886
 887         return ret;
 888 }
 889
 890 static ssize_t padata_sysfs_store(struct kobject *kobj, struct attribute *attr,
 891                                   const char *buf, size_t count)
 892 {
 893         struct padata_instance *pinst;
 894         struct padata_sysfs_entry *pentry;
 895         ssize_t ret = -EIO;
 896
 897         pinst = kobj2pinst(kobj);
 898         pentry = attr2pentry(attr);
 899         if (pentry->show)
 900                 ret = pentry->store(pinst, attr, buf, count);
 901
 902         return ret;
 903 }
 904
 905 static const struct sysfs_ops padata_sysfs_ops = {
 906         .show = padata_sysfs_show,
 907         .store = padata_sysfs_store,
 908 };
 909
 910 static struct kobj_type padata_attr_type = {
 911         .sysfs_ops = &padata_sysfs_ops,
 912         .default_attrs = padata_default_attrs,
 913         .release = padata_sysfs_release,
 914 };
 915
 916 /**
 917  * padata_alloc_possible - Allocate and initialize padata instance.
 918  *                         Use the cpu_possible_mask for serial and
 919  *                         parallel workers.
 920  *
 921  * @wq: workqueue to use for the allocated padata instance
 922  */
 923 struct padata_instance *padata_alloc_possible(struct workqueue_struct *wq)
 924 {
 925         return padata_alloc(wq, cpu_possible_mask, cpu_possible_mask);
 926 }
 927 EXPORT_SYMBOL(padata_alloc_possible);
 928
 929 /**
 930  * padata_alloc - allocate and initialize a padata instance and specify
 931  *                cpumasks for serial and parallel workers.
 932  *
 933  * @wq: workqueue to use for the allocated padata instance
 934  * @pcpumask: cpumask that will be used for padata parallelization
 935  * @cbcpumask: cpumask that will be used for padata serialization
 936  */
 937 struct padata_instance *padata_alloc(struct workqueue_struct *wq,
 938                                      const struct cpumask *pcpumask,
 939                                      const struct cpumask *cbcpumask)
 940 {
 941         struct padata_instance *pinst;
 942         struct parallel_data *pd = NULL;
 943
 944         pinst = kzalloc(sizeof(struct padata_instance), GFP_KERNEL);
 945         if (!pinst)
 946                 goto err;
 947
 948         get_online_cpus();
 949         if (!alloc_cpumask_var(&pinst->cpumask.pcpu, GFP_KERNEL))
 950                 goto err_free_inst;
 951         if (!alloc_cpumask_var(&pinst->cpumask.cbcpu, GFP_KERNEL)) {
 952                 free_cpumask_var(pinst->cpumask.pcpu);
 953                 goto err_free_inst;
 954         }
 955         if (!padata_validate_cpumask(pinst, pcpumask) ||
 956             !padata_validate_cpumask(pinst, cbcpumask))
 957                 goto err_free_masks;
 958
 959         pd = padata_alloc_pd(pinst, pcpumask, cbcpumask);
 960         if (!pd)
 961                 goto err_free_masks;
 962
 963         rcu_assign_pointer(pinst->pd, pd);
 964
 965         pinst->wq = wq;
 966
 967         cpumask_copy(pinst->cpumask.pcpu, pcpumask);
 968         cpumask_copy(pinst->cpumask.cbcpu, cbcpumask);
 969
 970         pinst->flags = 0;
 971
 972         put_online_cpus();
 973
 974         BLOCKING_INIT_NOTIFIER_HEAD(&pinst->cpumask_change_notifier);
 975         kobject_init(&pinst->kobj, &padata_attr_type);
 976         mutex_init(&pinst->lock);
 977
 978 #ifdef CONFIG_HOTPLUG_CPU
 979         cpuhp_state_add_instance_nocalls(hp_online, &pinst->node);
 980 #endif
 981         return pinst;
 982
 983 err_free_masks:
 984         free_cpumask_var(pinst->cpumask.pcpu);
 985         free_cpumask_var(pinst->cpumask.cbcpu);
 986 err_free_inst:
 987         kfree(pinst);
 988         put_online_cpus();
 989 err:
 990         return NULL;
 991 }
 992
 993 /**
 994  * padata_free - free a padata instance
 995  *
 996  * @padata_inst: padata instance to free
 997  */
 998 void padata_free(struct padata_instance *pinst)
 999 {
1000         kobject_put(&pinst->kobj);
1001 }
1002 EXPORT_SYMBOL(padata_free);
1003
1004 #ifdef CONFIG_HOTPLUG_CPU
1005
1006 static __init int padata_driver_init(void)
1007 {
1008         int ret;
1009
1010         ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "padata:online",
1011                                       padata_cpu_online,
1012                                       padata_cpu_prep_down);
1013         if (ret < 0)
1014                 return ret;
1015         hp_online = ret;
1016         return 0;
1017 }
1018 module_init(padata_driver_init);
1019
1020 static __exit void padata_driver_exit(void)
1021 {
1022         cpuhp_remove_multi_state(hp_online);
1023 }
1024 module_exit(padata_driver_exit);
1025 #endif