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 int padata_index_to_cpu(struct parallel_data *pd, int cpu_index)
  38 {
  39         int cpu, target_cpu;
  40
  41         target_cpu = cpumask_first(pd->cpumask.pcpu);
  42         for (cpu = 0; cpu < cpu_index; cpu++)
  43                 target_cpu = cpumask_next(target_cpu, pd->cpumask.pcpu);
  44
  45         return target_cpu;
  46 }
  47
  48 static int padata_cpu_hash(struct parallel_data *pd)
  49 {
  50         unsigned int seq_nr;
  51         int cpu_index;
  52
  53         /*
  54          * Hash the sequence numbers to the cpus by taking
  55          * seq_nr mod. number of cpus in use.
  56          */
  57
  58         seq_nr = atomic_inc_return(&pd->seq_nr);
  59         cpu_index = seq_nr % cpumask_weight(pd->cpumask.pcpu);
  60
  61         return padata_index_to_cpu(pd, cpu_index);
  62 }
  63
  64 static void padata_parallel_worker(struct work_struct *parallel_work)
  65 {
  66         struct padata_parallel_queue *pqueue;
  67         LIST_HEAD(local_list);
  68
  69         local_bh_disable();
  70         pqueue = container_of(parallel_work,
  71                               struct padata_parallel_queue, work);
  72
  73         spin_lock(&pqueue->parallel.lock);
  74         list_replace_init(&pqueue->parallel.list, &local_list);
  75         spin_unlock(&pqueue->parallel.lock);
  76
  77         while (!list_empty(&local_list)) {
  78                 struct padata_priv *padata;
  79
  80                 padata = list_entry(local_list.next,
  81                                     struct padata_priv, list);
  82
  83                 list_del_init(&padata->list);
  84
  85                 padata->parallel(padata);
  86         }
  87
  88         local_bh_enable();
  89 }
  90
  91 /**
  92  * padata_do_parallel - padata parallelization function
  93  *
  94  * @pinst: padata instance
  95  * @padata: object to be parallelized
  96  * @cb_cpu: cpu the serialization callback function will run on,
  97  *          must be in the serial cpumask of padata(i.e. cpumask.cbcpu).
  98  *
  99  * The parallelization callback function will run with BHs off.
 100  * Note: Every object which is parallelized by padata_do_parallel
 101  * must be seen by padata_do_serial.
 102  */
 103 int padata_do_parallel(struct padata_instance *pinst,
 104                        struct padata_priv *padata, int cb_cpu)
 105 {
 106         int target_cpu, err;
 107         struct padata_parallel_queue *queue;
 108         struct parallel_data *pd;
 109
 110         rcu_read_lock_bh();
 111
 112         pd = rcu_dereference_bh(pinst->pd);
 113
 114         err = -EINVAL;
 115         if (!(pinst->flags & PADATA_INIT) || pinst->flags & PADATA_INVALID)
 116                 goto out;
 117
 118         if (!cpumask_test_cpu(cb_cpu, pd->cpumask.cbcpu))
 119                 goto out;
 120
 121         err =  -EBUSY;
 122         if ((pinst->flags & PADATA_RESET))
 123                 goto out;
 124
 125         if (atomic_read(&pd->refcnt) >= MAX_OBJ_NUM)
 126                 goto out;
 127
 128         err = 0;
 129         atomic_inc(&pd->refcnt);
 130         padata->pd = pd;
 131         padata->cb_cpu = cb_cpu;
 132
 133         target_cpu = padata_cpu_hash(pd);
 134         queue = per_cpu_ptr(pd->pqueue, target_cpu);
 135
 136         spin_lock(&queue->parallel.lock);
 137         list_add_tail(&padata->list, &queue->parallel.list);
 138         spin_unlock(&queue->parallel.lock);
 139
 140         queue_work_on(target_cpu, pinst->wq, &queue->work);
 141
 142 out:
 143         rcu_read_unlock_bh();
 144
 145         return err;
 146 }
 147 EXPORT_SYMBOL(padata_do_parallel);
 148
 149 /*
 150  * padata_get_next - Get the next object that needs serialization.
 151  *
 152  * Return values are:
 153  *
 154  * A pointer to the control struct of the next object that needs
 155  * serialization, if present in one of the percpu reorder queues.
 156  *
 157  * NULL, if all percpu reorder queues are empty.
 158  *
 159  * -EINPROGRESS, if the next object that needs serialization will
 160  *  be parallel processed by another cpu and is not yet present in
 161  *  the cpu's reorder queue.
 162  *
 163  * -ENODATA, if this cpu has to do the parallel processing for
 164  *  the next object.
 165  */
 166 static struct padata_priv *padata_get_next(struct parallel_data *pd)
 167 {
 168         int cpu, num_cpus;
 169         unsigned int next_nr, next_index;
 170         struct padata_parallel_queue *next_queue;
 171         struct padata_priv *padata;
 172         struct padata_list *reorder;
 173
 174         num_cpus = cpumask_weight(pd->cpumask.pcpu);
 175
 176         /*
 177          * Calculate the percpu reorder queue and the sequence
 178          * number of the next object.
 179          */
 180         next_nr = pd->processed;
 181         next_index = next_nr % num_cpus;
 182         cpu = padata_index_to_cpu(pd, next_index);
 183         next_queue = per_cpu_ptr(pd->pqueue, cpu);
 184
 185         padata = NULL;
 186
 187         reorder = &next_queue->reorder;
 188
 189         spin_lock(&reorder->lock);
 190         if (!list_empty(&reorder->list)) {
 191                 padata = list_entry(reorder->list.next,
 192                                     struct padata_priv, list);
 193
 194                 list_del_init(&padata->list);
 195                 atomic_dec(&pd->reorder_objects);
 196
 197                 pd->processed++;
 198
 199                 spin_unlock(&reorder->lock);
 200                 goto out;
 201         }
 202         spin_unlock(&reorder->lock);
 203
 204         if (__this_cpu_read(pd->pqueue->cpu_index) == next_queue->cpu_index) {
 205                 padata = ERR_PTR(-ENODATA);
 206                 goto out;
 207         }
 208
 209         padata = ERR_PTR(-EINPROGRESS);
 210 out:
 211         return padata;
 212 }
 213
 214 static void padata_reorder(struct parallel_data *pd)
 215 {
 216         int cb_cpu;
 217         struct padata_priv *padata;
 218         struct padata_serial_queue *squeue;
 219         struct padata_instance *pinst = pd->pinst;
 220
 221         /*
 222          * We need to ensure that only one cpu can work on dequeueing of
 223          * the reorder queue the time. Calculating in which percpu reorder
 224          * queue the next object will arrive takes some time. A spinlock
 225          * would be highly contended. Also it is not clear in which order
 226          * the objects arrive to the reorder queues. So a cpu could wait to
 227          * get the lock just to notice that there is nothing to do at the
 228          * moment. Therefore we use a trylock and let the holder of the lock
 229          * care for all the objects enqueued during the holdtime of the lock.
 230          */
 231         if (!spin_trylock_bh(&pd->lock))
 232                 return;
 233
 234         while (1) {
 235                 padata = padata_get_next(pd);
 236
 237                 /*
 238                  * All reorder queues are empty, or the next object that needs
 239                  * serialization is parallel processed by another cpu and is
 240                  * still on it's way to the cpu's reorder queue, nothing to
 241                  * do for now.
 242                  */
 243                 if (!padata || PTR_ERR(padata) == -EINPROGRESS)
 244                         break;
 245
 246                 /*
 247                  * This cpu has to do the parallel processing of the next
 248                  * object. It's waiting in the cpu's parallelization queue,
 249                  * so exit immediately.
 250                  */
 251                 if (PTR_ERR(padata) == -ENODATA) {
 252                         del_timer(&pd->timer);
 253                         spin_unlock_bh(&pd->lock);
 254                         return;
 255                 }
 256
 257                 cb_cpu = padata->cb_cpu;
 258                 squeue = per_cpu_ptr(pd->squeue, cb_cpu);
 259
 260                 spin_lock(&squeue->serial.lock);
 261                 list_add_tail(&padata->list, &squeue->serial.list);
 262                 spin_unlock(&squeue->serial.lock);
 263
 264                 queue_work_on(cb_cpu, pinst->wq, &squeue->work);
 265         }
 266
 267         spin_unlock_bh(&pd->lock);
 268
 269         /*
 270          * The next object that needs serialization might have arrived to
 271          * the reorder queues in the meantime, we will be called again
 272          * from the timer function if no one else cares for it.
 273          */
 274         if (atomic_read(&pd->reorder_objects)
 275                         && !(pinst->flags & PADATA_RESET))
 276                 mod_timer(&pd->timer, jiffies + HZ);
 277         else
 278                 del_timer(&pd->timer);
 279
 280         return;
 281 }
 282
 283 static void padata_reorder_timer(unsigned long arg)
 284 {
 285         struct parallel_data *pd = (struct parallel_data *)arg;
 286
 287         padata_reorder(pd);
 288 }
 289
 290 static void padata_serial_worker(struct work_struct *serial_work)
 291 {
 292         struct padata_serial_queue *squeue;
 293         struct parallel_data *pd;
 294         LIST_HEAD(local_list);
 295
 296         local_bh_disable();
 297         squeue = container_of(serial_work, struct padata_serial_queue, work);
 298         pd = squeue->pd;
 299
 300         spin_lock(&squeue->serial.lock);
 301         list_replace_init(&squeue->serial.list, &local_list);
 302         spin_unlock(&squeue->serial.lock);
 303
 304         while (!list_empty(&local_list)) {
 305                 struct padata_priv *padata;
 306
 307                 padata = list_entry(local_list.next,
 308                                     struct padata_priv, list);
 309
 310                 list_del_init(&padata->list);
 311
 312                 padata->serial(padata);
 313                 atomic_dec(&pd->refcnt);
 314         }
 315         local_bh_enable();
 316 }
 317
 318 /**
 319  * padata_do_serial - padata serialization function
 320  *
 321  * @padata: object to be serialized.
 322  *
 323  * padata_do_serial must be called for every parallelized object.
 324  * The serialization callback function will run with BHs off.
 325  */
 326 void padata_do_serial(struct padata_priv *padata)
 327 {
 328         int cpu;
 329         struct padata_parallel_queue *pqueue;
 330         struct parallel_data *pd;
 331
 332         pd = padata->pd;
 333
 334         cpu = get_cpu();
 335         pqueue = per_cpu_ptr(pd->pqueue, cpu);
 336
 337         spin_lock(&pqueue->reorder.lock);
 338         atomic_inc(&pd->reorder_objects);
 339         list_add_tail(&padata->list, &pqueue->reorder.list);
 340         spin_unlock(&pqueue->reorder.lock);
 341
 342         put_cpu();
 343
 344         padata_reorder(pd);
 345 }
 346 EXPORT_SYMBOL(padata_do_serial);
 347
 348 static int padata_setup_cpumasks(struct parallel_data *pd,
 349                                  const struct cpumask *pcpumask,
 350                                  const struct cpumask *cbcpumask)
 351 {
 352         if (!alloc_cpumask_var(&pd->cpumask.pcpu, GFP_KERNEL))
 353                 return -ENOMEM;
 354
 355         cpumask_and(pd->cpumask.pcpu, pcpumask, cpu_online_mask);
 356         if (!alloc_cpumask_var(&pd->cpumask.cbcpu, GFP_KERNEL)) {
 357                 free_cpumask_var(pd->cpumask.cbcpu);
 358                 return -ENOMEM;
 359         }
 360
 361         cpumask_and(pd->cpumask.cbcpu, cbcpumask, cpu_online_mask);
 362         return 0;
 363 }
 364
 365 static void __padata_list_init(struct padata_list *pd_list)
 366 {
 367         INIT_LIST_HEAD(&pd_list->list);
 368         spin_lock_init(&pd_list->lock);
 369 }
 370
 371 /* Initialize all percpu queues used by serial workers */
 372 static void padata_init_squeues(struct parallel_data *pd)
 373 {
 374         int cpu;
 375         struct padata_serial_queue *squeue;
 376
 377         for_each_cpu(cpu, pd->cpumask.cbcpu) {
 378                 squeue = per_cpu_ptr(pd->squeue, cpu);
 379                 squeue->pd = pd;
 380                 __padata_list_init(&squeue->serial);
 381                 INIT_WORK(&squeue->work, padata_serial_worker);
 382         }
 383 }
 384
 385 /* Initialize all percpu queues used by parallel workers */
 386 static void padata_init_pqueues(struct parallel_data *pd)
 387 {
 388         int cpu_index, cpu;
 389         struct padata_parallel_queue *pqueue;
 390
 391         cpu_index = 0;
 392         for_each_cpu(cpu, pd->cpumask.pcpu) {
 393                 pqueue = per_cpu_ptr(pd->pqueue, cpu);
 394                 pqueue->pd = pd;
 395                 pqueue->cpu_index = cpu_index;
 396                 cpu_index++;
 397
 398                 __padata_list_init(&pqueue->reorder);
 399                 __padata_list_init(&pqueue->parallel);
 400                 INIT_WORK(&pqueue->work, padata_parallel_worker);
 401                 atomic_set(&pqueue->num_obj, 0);
 402         }
 403 }
 404
 405 /* Allocate and initialize the internal cpumask dependend resources. */
 406 static struct parallel_data *padata_alloc_pd(struct padata_instance *pinst,
 407                                              const struct cpumask *pcpumask,
 408                                              const struct cpumask *cbcpumask)
 409 {
 410         struct parallel_data *pd;
 411
 412         pd = kzalloc(sizeof(struct parallel_data), GFP_KERNEL);
 413         if (!pd)
 414                 goto err;
 415
 416         pd->pqueue = alloc_percpu(struct padata_parallel_queue);
 417         if (!pd->pqueue)
 418                 goto err_free_pd;
 419
 420         pd->squeue = alloc_percpu(struct padata_serial_queue);
 421         if (!pd->squeue)
 422                 goto err_free_pqueue;
 423         if (padata_setup_cpumasks(pd, pcpumask, cbcpumask) < 0)
 424                 goto err_free_squeue;
 425
 426         padata_init_pqueues(pd);
 427         padata_init_squeues(pd);
 428         setup_timer(&pd->timer, padata_reorder_timer, (unsigned long)pd);
 429         atomic_set(&pd->seq_nr, -1);
 430         atomic_set(&pd->reorder_objects, 0);
 431         atomic_set(&pd->refcnt, 0);
 432         pd->pinst = pinst;
 433         spin_lock_init(&pd->lock);
 434
 435         return pd;
 436
 437 err_free_squeue:
 438         free_percpu(pd->squeue);
 439 err_free_pqueue:
 440         free_percpu(pd->pqueue);
 441 err_free_pd:
 442         kfree(pd);
 443 err:
 444         return NULL;
 445 }
 446
 447 static void padata_free_pd(struct parallel_data *pd)
 448 {
 449         free_cpumask_var(pd->cpumask.pcpu);
 450         free_cpumask_var(pd->cpumask.cbcpu);
 451         free_percpu(pd->pqueue);
 452         free_percpu(pd->squeue);
 453         kfree(pd);
 454 }
 455
 456 /* Flush all objects out of the padata queues. */
 457 static void padata_flush_queues(struct parallel_data *pd)
 458 {
 459         int cpu;
 460         struct padata_parallel_queue *pqueue;
 461         struct padata_serial_queue *squeue;
 462
 463         for_each_cpu(cpu, pd->cpumask.pcpu) {
 464                 pqueue = per_cpu_ptr(pd->pqueue, cpu);
 465                 flush_work(&pqueue->work);
 466         }
 467
 468         del_timer_sync(&pd->timer);
 469
 470         if (atomic_read(&pd->reorder_objects))
 471                 padata_reorder(pd);
 472
 473         for_each_cpu(cpu, pd->cpumask.cbcpu) {
 474                 squeue = per_cpu_ptr(pd->squeue, cpu);
 475                 flush_work(&squeue->work);
 476         }
 477
 478         BUG_ON(atomic_read(&pd->refcnt) != 0);
 479 }
 480
 481 static void __padata_start(struct padata_instance *pinst)
 482 {
 483         pinst->flags |= PADATA_INIT;
 484 }
 485
 486 static void __padata_stop(struct padata_instance *pinst)
 487 {
 488         if (!(pinst->flags & PADATA_INIT))
 489                 return;
 490
 491         pinst->flags &= ~PADATA_INIT;
 492
 493         synchronize_rcu();
 494
 495         get_online_cpus();
 496         padata_flush_queues(pinst->pd);
 497         put_online_cpus();
 498 }
 499
 500 /* Replace the internal control structure with a new one. */
 501 static void padata_replace(struct padata_instance *pinst,
 502                            struct parallel_data *pd_new)
 503 {
 504         struct parallel_data *pd_old = pinst->pd;
 505         int notification_mask = 0;
 506
 507         pinst->flags |= PADATA_RESET;
 508
 509         rcu_assign_pointer(pinst->pd, pd_new);
 510
 511         synchronize_rcu();
 512
 513         if (!cpumask_equal(pd_old->cpumask.pcpu, pd_new->cpumask.pcpu))
 514                 notification_mask |= PADATA_CPU_PARALLEL;
 515         if (!cpumask_equal(pd_old->cpumask.cbcpu, pd_new->cpumask.cbcpu))
 516                 notification_mask |= PADATA_CPU_SERIAL;
 517
 518         padata_flush_queues(pd_old);
 519         padata_free_pd(pd_old);
 520
 521         if (notification_mask)
 522                 blocking_notifier_call_chain(&pinst->cpumask_change_notifier,
 523                                              notification_mask,
 524                                              &pd_new->cpumask);
 525
 526         pinst->flags &= ~PADATA_RESET;
 527 }
 528
 529 /**
 530  * padata_register_cpumask_notifier - Registers a notifier that will be called
 531  *                             if either pcpu or cbcpu or both cpumasks change.
 532  *
 533  * @pinst: A poineter to padata instance
 534  * @nblock: A pointer to notifier block.
 535  */
 536 int padata_register_cpumask_notifier(struct padata_instance *pinst,
 537                                      struct notifier_block *nblock)
 538 {
 539         return blocking_notifier_chain_register(&pinst->cpumask_change_notifier,
 540                                                 nblock);
 541 }
 542 EXPORT_SYMBOL(padata_register_cpumask_notifier);
 543
 544 /**
 545  * padata_unregister_cpumask_notifier - Unregisters cpumask notifier
 546  *        registered earlier  using padata_register_cpumask_notifier
 547  *
 548  * @pinst: A pointer to data instance.
 549  * @nlock: A pointer to notifier block.
 550  */
 551 int padata_unregister_cpumask_notifier(struct padata_instance *pinst,
 552                                        struct notifier_block *nblock)
 553 {
 554         return blocking_notifier_chain_unregister(
 555                 &pinst->cpumask_change_notifier,
 556                 nblock);
 557 }
 558 EXPORT_SYMBOL(padata_unregister_cpumask_notifier);
 559
 560
 561 /* If cpumask contains no active cpu, we mark the instance as invalid. */
 562 static bool padata_validate_cpumask(struct padata_instance *pinst,
 563                                     const struct cpumask *cpumask)
 564 {
 565         if (!cpumask_intersects(cpumask, cpu_online_mask)) {
 566                 pinst->flags |= PADATA_INVALID;
 567                 return false;
 568         }
 569
 570         pinst->flags &= ~PADATA_INVALID;
 571         return true;
 572 }
 573
 574 static int __padata_set_cpumasks(struct padata_instance *pinst,
 575                                  cpumask_var_t pcpumask,
 576                                  cpumask_var_t cbcpumask)
 577 {
 578         int valid;
 579         struct parallel_data *pd;
 580
 581         valid = padata_validate_cpumask(pinst, pcpumask);
 582         if (!valid) {
 583                 __padata_stop(pinst);
 584                 goto out_replace;
 585         }
 586
 587         valid = padata_validate_cpumask(pinst, cbcpumask);
 588         if (!valid)
 589                 __padata_stop(pinst);
 590
 591 out_replace:
 592         pd = padata_alloc_pd(pinst, pcpumask, cbcpumask);
 593         if (!pd)
 594                 return -ENOMEM;
 595
 596         cpumask_copy(pinst->cpumask.pcpu, pcpumask);
 597         cpumask_copy(pinst->cpumask.cbcpu, cbcpumask);
 598
 599         padata_replace(pinst, pd);
 600
 601         if (valid)
 602                 __padata_start(pinst);
 603
 604         return 0;
 605 }
 606
 607 /**
 608  * padata_set_cpumask: Sets specified by @cpumask_type cpumask to the value
 609  *                     equivalent to @cpumask.
 610  *
 611  * @pinst: padata instance
 612  * @cpumask_type: PADATA_CPU_SERIAL or PADATA_CPU_PARALLEL corresponding
 613  *                to parallel and serial cpumasks respectively.
 614  * @cpumask: the cpumask to use
 615  */
 616 int padata_set_cpumask(struct padata_instance *pinst, int cpumask_type,
 617                        cpumask_var_t cpumask)
 618 {
 619         struct cpumask *serial_mask, *parallel_mask;
 620         int err = -EINVAL;
 621
 622         mutex_lock(&pinst->lock);
 623         get_online_cpus();
 624
 625         switch (cpumask_type) {
 626         case PADATA_CPU_PARALLEL:
 627                 serial_mask = pinst->cpumask.cbcpu;
 628                 parallel_mask = cpumask;
 629                 break;
 630         case PADATA_CPU_SERIAL:
 631                 parallel_mask = pinst->cpumask.pcpu;
 632                 serial_mask = cpumask;
 633                 break;
 634         default:
 635                  goto out;
 636         }
 637
 638         err =  __padata_set_cpumasks(pinst, parallel_mask, serial_mask);
 639
 640 out:
 641         put_online_cpus();
 642         mutex_unlock(&pinst->lock);
 643
 644         return err;
 645 }
 646 EXPORT_SYMBOL(padata_set_cpumask);
 647
 648 /**
 649  * padata_start - start the parallel processing
 650  *
 651  * @pinst: padata instance to start
 652  */
 653 int padata_start(struct padata_instance *pinst)
 654 {
 655         int err = 0;
 656
 657         mutex_lock(&pinst->lock);
 658
 659         if (pinst->flags & PADATA_INVALID)
 660                 err = -EINVAL;
 661
 662          __padata_start(pinst);
 663
 664         mutex_unlock(&pinst->lock);
 665
 666         return err;
 667 }
 668 EXPORT_SYMBOL(padata_start);
 669
 670 /**
 671  * padata_stop - stop the parallel processing
 672  *
 673  * @pinst: padata instance to stop
 674  */
 675 void padata_stop(struct padata_instance *pinst)
 676 {
 677         mutex_lock(&pinst->lock);
 678         __padata_stop(pinst);
 679         mutex_unlock(&pinst->lock);
 680 }
 681 EXPORT_SYMBOL(padata_stop);
 682
 683 #ifdef CONFIG_HOTPLUG_CPU
 684
 685 static int __padata_add_cpu(struct padata_instance *pinst, int cpu)
 686 {
 687         struct parallel_data *pd;
 688
 689         if (cpumask_test_cpu(cpu, cpu_online_mask)) {
 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                 if (padata_validate_cpumask(pinst, pinst->cpumask.pcpu) &&
 698                     padata_validate_cpumask(pinst, pinst->cpumask.cbcpu))
 699                         __padata_start(pinst);
 700         }
 701
 702         return 0;
 703 }
 704
 705 static int __padata_remove_cpu(struct padata_instance *pinst, int cpu)
 706 {
 707         struct parallel_data *pd = NULL;
 708
 709         if (cpumask_test_cpu(cpu, cpu_online_mask)) {
 710
 711                 if (!padata_validate_cpumask(pinst, pinst->cpumask.pcpu) ||
 712                     !padata_validate_cpumask(pinst, pinst->cpumask.cbcpu))
 713                         __padata_stop(pinst);
 714
 715                 pd = padata_alloc_pd(pinst, pinst->cpumask.pcpu,
 716                                      pinst->cpumask.cbcpu);
 717                 if (!pd)
 718                         return -ENOMEM;
 719
 720                 padata_replace(pinst, pd);
 721
 722                 cpumask_clear_cpu(cpu, pd->cpumask.cbcpu);
 723                 cpumask_clear_cpu(cpu, pd->cpumask.pcpu);
 724         }
 725
 726         return 0;
 727 }
 728
 729  /**
 730  * padata_remove_cpu - remove a cpu from the one or both(serial and parallel)
 731  *                     padata cpumasks.
 732  *
 733  * @pinst: padata instance
 734  * @cpu: cpu to remove
 735  * @mask: bitmask specifying from which cpumask @cpu should be removed
 736  *        The @mask may be any combination of the following flags:
 737  *          PADATA_CPU_SERIAL   - serial cpumask
 738  *          PADATA_CPU_PARALLEL - parallel cpumask
 739  */
 740 int padata_remove_cpu(struct padata_instance *pinst, int cpu, int mask)
 741 {
 742         int err;
 743
 744         if (!(mask & (PADATA_CPU_SERIAL | PADATA_CPU_PARALLEL)))
 745                 return -EINVAL;
 746
 747         mutex_lock(&pinst->lock);
 748
 749         get_online_cpus();
 750         if (mask & PADATA_CPU_SERIAL)
 751                 cpumask_clear_cpu(cpu, pinst->cpumask.cbcpu);
 752         if (mask & PADATA_CPU_PARALLEL)
 753                 cpumask_clear_cpu(cpu, pinst->cpumask.pcpu);
 754
 755         err = __padata_remove_cpu(pinst, cpu);
 756         put_online_cpus();
 757
 758         mutex_unlock(&pinst->lock);
 759
 760         return err;
 761 }
 762 EXPORT_SYMBOL(padata_remove_cpu);
 763
 764 static inline int pinst_has_cpu(struct padata_instance *pinst, int cpu)
 765 {
 766         return cpumask_test_cpu(cpu, pinst->cpumask.pcpu) ||
 767                 cpumask_test_cpu(cpu, pinst->cpumask.cbcpu);
 768 }
 769
 770 static int padata_cpu_online(unsigned int cpu, struct hlist_node *node)
 771 {
 772         struct padata_instance *pinst;
 773         int ret;
 774
 775         pinst = hlist_entry_safe(node, struct padata_instance, node);
 776         if (!pinst_has_cpu(pinst, cpu))
 777                 return 0;
 778
 779         mutex_lock(&pinst->lock);
 780         ret = __padata_add_cpu(pinst, cpu);
 781         mutex_unlock(&pinst->lock);
 782         return ret;
 783 }
 784
 785 static int padata_cpu_prep_down(unsigned int cpu, struct hlist_node *node)
 786 {
 787         struct padata_instance *pinst;
 788         int ret;
 789
 790         pinst = hlist_entry_safe(node, struct padata_instance, node);
 791         if (!pinst_has_cpu(pinst, cpu))
 792                 return 0;
 793
 794         mutex_lock(&pinst->lock);
 795         ret = __padata_remove_cpu(pinst, cpu);
 796         mutex_unlock(&pinst->lock);
 797         return ret;
 798 }
 799
 800 static enum cpuhp_state hp_online;
 801 #endif
 802
 803 static void __padata_free(struct padata_instance *pinst)
 804 {
 805 #ifdef CONFIG_HOTPLUG_CPU
 806         cpuhp_state_remove_instance_nocalls(hp_online, &pinst->node);
 807 #endif
 808
 809         padata_stop(pinst);
 810         padata_free_pd(pinst->pd);
 811         free_cpumask_var(pinst->cpumask.pcpu);
 812         free_cpumask_var(pinst->cpumask.cbcpu);
 813         kfree(pinst);
 814 }
 815
 816 #define kobj2pinst(_kobj)                                       \
 817         container_of(_kobj, struct padata_instance, kobj)
 818 #define attr2pentry(_attr)                                      \
 819         container_of(_attr, struct padata_sysfs_entry, attr)
 820
 821 static void padata_sysfs_release(struct kobject *kobj)
 822 {
 823         struct padata_instance *pinst = kobj2pinst(kobj);
 824         __padata_free(pinst);
 825 }
 826
 827 struct padata_sysfs_entry {
 828         struct attribute attr;
 829         ssize_t (*show)(struct padata_instance *, struct attribute *, char *);
 830         ssize_t (*store)(struct padata_instance *, struct attribute *,
 831                          const char *, size_t);
 832 };
 833
 834 static ssize_t show_cpumask(struct padata_instance *pinst,
 835                             struct attribute *attr,  char *buf)
 836 {
 837         struct cpumask *cpumask;
 838         ssize_t len;
 839
 840         mutex_lock(&pinst->lock);
 841         if (!strcmp(attr->name, "serial_cpumask"))
 842                 cpumask = pinst->cpumask.cbcpu;
 843         else
 844                 cpumask = pinst->cpumask.pcpu;
 845
 846         len = snprintf(buf, PAGE_SIZE, "%*pb\n",
 847                        nr_cpu_ids, cpumask_bits(cpumask));
 848         mutex_unlock(&pinst->lock);
 849         return len < PAGE_SIZE ? len : -EINVAL;
 850 }
 851
 852 static ssize_t store_cpumask(struct padata_instance *pinst,
 853                              struct attribute *attr,
 854                              const char *buf, size_t count)
 855 {
 856         cpumask_var_t new_cpumask;
 857         ssize_t ret;
 858         int mask_type;
 859
 860         if (!alloc_cpumask_var(&new_cpumask, GFP_KERNEL))
 861                 return -ENOMEM;
 862
 863         ret = bitmap_parse(buf, count, cpumask_bits(new_cpumask),
 864                            nr_cpumask_bits);
 865         if (ret < 0)
 866                 goto out;
 867
 868         mask_type = !strcmp(attr->name, "serial_cpumask") ?
 869                 PADATA_CPU_SERIAL : PADATA_CPU_PARALLEL;
 870         ret = padata_set_cpumask(pinst, mask_type, new_cpumask);
 871         if (!ret)
 872                 ret = count;
 873
 874 out:
 875         free_cpumask_var(new_cpumask);
 876         return ret;
 877 }
 878
 879 #define PADATA_ATTR_RW(_name, _show_name, _store_name)          \
 880         static struct padata_sysfs_entry _name##_attr =         \
 881                 __ATTR(_name, 0644, _show_name, _store_name)
 882 #define PADATA_ATTR_RO(_name, _show_name)               \
 883         static struct padata_sysfs_entry _name##_attr = \
 884                 __ATTR(_name, 0400, _show_name, NULL)
 885
 886 PADATA_ATTR_RW(serial_cpumask, show_cpumask, store_cpumask);
 887 PADATA_ATTR_RW(parallel_cpumask, show_cpumask, store_cpumask);
 888
 889 /*
 890  * Padata sysfs provides the following objects:
 891  * serial_cpumask   [RW] - cpumask for serial workers
 892  * parallel_cpumask [RW] - cpumask for parallel workers
 893  */
 894 static struct attribute *padata_default_attrs[] = {
 895         &serial_cpumask_attr.attr,
 896         &parallel_cpumask_attr.attr,
 897         NULL,
 898 };
 899
 900 static ssize_t padata_sysfs_show(struct kobject *kobj,
 901                                  struct attribute *attr, char *buf)
 902 {
 903         struct padata_instance *pinst;
 904         struct padata_sysfs_entry *pentry;
 905         ssize_t ret = -EIO;
 906
 907         pinst = kobj2pinst(kobj);
 908         pentry = attr2pentry(attr);
 909         if (pentry->show)
 910                 ret = pentry->show(pinst, attr, buf);
 911
 912         return ret;
 913 }
 914
 915 static ssize_t padata_sysfs_store(struct kobject *kobj, struct attribute *attr,
 916                                   const char *buf, size_t count)
 917 {
 918         struct padata_instance *pinst;
 919         struct padata_sysfs_entry *pentry;
 920         ssize_t ret = -EIO;
 921
 922         pinst = kobj2pinst(kobj);
 923         pentry = attr2pentry(attr);
 924         if (pentry->show)
 925                 ret = pentry->store(pinst, attr, buf, count);
 926
 927         return ret;
 928 }
 929
 930 static const struct sysfs_ops padata_sysfs_ops = {
 931         .show = padata_sysfs_show,
 932         .store = padata_sysfs_store,
 933 };
 934
 935 static struct kobj_type padata_attr_type = {
 936         .sysfs_ops = &padata_sysfs_ops,
 937         .default_attrs = padata_default_attrs,
 938         .release = padata_sysfs_release,
 939 };
 940
 941 /**
 942  * padata_alloc_possible - Allocate and initialize padata instance.
 943  *                         Use the cpu_possible_mask for serial and
 944  *                         parallel workers.
 945  *
 946  * @wq: workqueue to use for the allocated padata instance
 947  */
 948 struct padata_instance *padata_alloc_possible(struct workqueue_struct *wq)
 949 {
 950         return padata_alloc(wq, cpu_possible_mask, cpu_possible_mask);
 951 }
 952 EXPORT_SYMBOL(padata_alloc_possible);
 953
 954 /**
 955  * padata_alloc - allocate and initialize a padata instance and specify
 956  *                cpumasks for serial and parallel workers.
 957  *
 958  * @wq: workqueue to use for the allocated padata instance
 959  * @pcpumask: cpumask that will be used for padata parallelization
 960  * @cbcpumask: cpumask that will be used for padata serialization
 961  */
 962 struct padata_instance *padata_alloc(struct workqueue_struct *wq,
 963                                      const struct cpumask *pcpumask,
 964                                      const struct cpumask *cbcpumask)
 965 {
 966         struct padata_instance *pinst;
 967         struct parallel_data *pd = NULL;
 968
 969         pinst = kzalloc(sizeof(struct padata_instance), GFP_KERNEL);
 970         if (!pinst)
 971                 goto err;
 972
 973         get_online_cpus();
 974         if (!alloc_cpumask_var(&pinst->cpumask.pcpu, GFP_KERNEL))
 975                 goto err_free_inst;
 976         if (!alloc_cpumask_var(&pinst->cpumask.cbcpu, GFP_KERNEL)) {
 977                 free_cpumask_var(pinst->cpumask.pcpu);
 978                 goto err_free_inst;
 979         }
 980         if (!padata_validate_cpumask(pinst, pcpumask) ||
 981             !padata_validate_cpumask(pinst, cbcpumask))
 982                 goto err_free_masks;
 983
 984         pd = padata_alloc_pd(pinst, pcpumask, cbcpumask);
 985         if (!pd)
 986                 goto err_free_masks;
 987
 988         rcu_assign_pointer(pinst->pd, pd);
 989
 990         pinst->wq = wq;
 991
 992         cpumask_copy(pinst->cpumask.pcpu, pcpumask);
 993         cpumask_copy(pinst->cpumask.cbcpu, cbcpumask);
 994
 995         pinst->flags = 0;
 996
 997         put_online_cpus();
 998
 999         BLOCKING_INIT_NOTIFIER_HEAD(&pinst->cpumask_change_notifier);
1000         kobject_init(&pinst->kobj, &padata_attr_type);
1001         mutex_init(&pinst->lock);
1002
1003 #ifdef CONFIG_HOTPLUG_CPU
1004         cpuhp_state_add_instance_nocalls(hp_online, &pinst->node);
1005 #endif
1006         return pinst;
1007
1008 err_free_masks:
1009         free_cpumask_var(pinst->cpumask.pcpu);
1010         free_cpumask_var(pinst->cpumask.cbcpu);
1011 err_free_inst:
1012         kfree(pinst);
1013         put_online_cpus();
1014 err:
1015         return NULL;
1016 }
1017
1018 /**
1019  * padata_free - free a padata instance
1020  *
1021  * @padata_inst: padata instance to free
1022  */
1023 void padata_free(struct padata_instance *pinst)
1024 {
1025         kobject_put(&pinst->kobj);
1026 }
1027 EXPORT_SYMBOL(padata_free);
1028
1029 #ifdef CONFIG_HOTPLUG_CPU
1030
1031 static __init int padata_driver_init(void)
1032 {
1033         int ret;
1034
1035         ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "padata:online",
1036                                       padata_cpu_online,
1037                                       padata_cpu_prep_down);
1038         if (ret < 0)
1039                 return ret;
1040         hp_online = ret;
1041         return 0;
1042 }
1043 module_init(padata_driver_init);
1044
1045 static __exit void padata_driver_exit(void)
1046 {
1047         cpuhp_remove_multi_state(hp_online);
1048 }
1049 module_exit(padata_driver_exit);
1050 #endif