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         struct parallel_data *pd;
  68         struct padata_instance *pinst;
  69         LIST_HEAD(local_list);
  70
  71         local_bh_disable();
  72         pqueue = container_of(parallel_work,
  73                               struct padata_parallel_queue, work);
  74         pd = pqueue->pd;
  75         pinst = pd->pinst;
  76
  77         spin_lock(&pqueue->parallel.lock);
  78         list_replace_init(&pqueue->parallel.list, &local_list);
  79         spin_unlock(&pqueue->parallel.lock);
  80
  81         while (!list_empty(&local_list)) {
  82                 struct padata_priv *padata;
  83
  84                 padata = list_entry(local_list.next,
  85                                     struct padata_priv, list);
  86
  87                 list_del_init(&padata->list);
  88
  89                 padata->parallel(padata);
  90         }
  91
  92         local_bh_enable();
  93 }
  94
  95 /**
  96  * padata_do_parallel - padata parallelization function
  97  *
  98  * @pinst: padata instance
  99  * @padata: object to be parallelized
 100  * @cb_cpu: cpu the serialization callback function will run on,
 101  *          must be in the serial cpumask of padata(i.e. cpumask.cbcpu).
 102  *
 103  * The parallelization callback function will run with BHs off.
 104  * Note: Every object which is parallelized by padata_do_parallel
 105  * must be seen by padata_do_serial.
 106  */
 107 int padata_do_parallel(struct padata_instance *pinst,
 108                        struct padata_priv *padata, int cb_cpu)
 109 {
 110         int target_cpu, err;
 111         struct padata_parallel_queue *queue;
 112         struct parallel_data *pd;
 113
 114         rcu_read_lock_bh();
 115
 116         pd = rcu_dereference_bh(pinst->pd);
 117
 118         err = -EINVAL;
 119         if (!(pinst->flags & PADATA_INIT) || pinst->flags & PADATA_INVALID)
 120                 goto out;
 121
 122         if (!cpumask_test_cpu(cb_cpu, pd->cpumask.cbcpu))
 123                 goto out;
 124
 125         err =  -EBUSY;
 126         if ((pinst->flags & PADATA_RESET))
 127                 goto out;
 128
 129         if (atomic_read(&pd->refcnt) >= MAX_OBJ_NUM)
 130                 goto out;
 131
 132         err = 0;
 133         atomic_inc(&pd->refcnt);
 134         padata->pd = pd;
 135         padata->cb_cpu = cb_cpu;
 136
 137         target_cpu = padata_cpu_hash(pd);
 138         queue = per_cpu_ptr(pd->pqueue, target_cpu);
 139
 140         spin_lock(&queue->parallel.lock);
 141         list_add_tail(&padata->list, &queue->parallel.list);
 142         spin_unlock(&queue->parallel.lock);
 143
 144         queue_work_on(target_cpu, pinst->wq, &queue->work);
 145
 146 out:
 147         rcu_read_unlock_bh();
 148
 149         return err;
 150 }
 151 EXPORT_SYMBOL(padata_do_parallel);
 152
 153 /*
 154  * padata_get_next - Get the next object that needs serialization.
 155  *
 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  * NULL, if all percpu reorder queues are empty.
 162  *
 163  * -EINPROGRESS, if the next object that needs serialization will
 164  *  be parallel processed by another cpu and is not yet present in
 165  *  the cpu's reorder queue.
 166  *
 167  * -ENODATA, if this cpu has to do the parallel processing for
 168  *  the next object.
 169  */
 170 static struct padata_priv *padata_get_next(struct parallel_data *pd)
 171 {
 172         int cpu, num_cpus;
 173         unsigned int next_nr, next_index;
 174         struct padata_parallel_queue *next_queue;
 175         struct padata_priv *padata;
 176         struct padata_list *reorder;
 177
 178         num_cpus = cpumask_weight(pd->cpumask.pcpu);
 179
 180         /*
 181          * Calculate the percpu reorder queue and the sequence
 182          * number of the next object.
 183          */
 184         next_nr = pd->processed;
 185         next_index = next_nr % num_cpus;
 186         cpu = padata_index_to_cpu(pd, next_index);
 187         next_queue = per_cpu_ptr(pd->pqueue, cpu);
 188
 189         padata = NULL;
 190
 191         reorder = &next_queue->reorder;
 192
 193         spin_lock(&reorder->lock);
 194         if (!list_empty(&reorder->list)) {
 195                 padata = list_entry(reorder->list.next,
 196                                     struct padata_priv, list);
 197
 198                 list_del_init(&padata->list);
 199                 atomic_dec(&pd->reorder_objects);
 200
 201                 pd->processed++;
 202
 203                 spin_unlock(&reorder->lock);
 204                 goto out;
 205         }
 206         spin_unlock(&reorder->lock);
 207
 208         if (__this_cpu_read(pd->pqueue->cpu_index) == next_queue->cpu_index) {
 209                 padata = ERR_PTR(-ENODATA);
 210                 goto out;
 211         }
 212
 213         padata = ERR_PTR(-EINPROGRESS);
 214 out:
 215         return padata;
 216 }
 217
 218 static void padata_reorder(struct parallel_data *pd)
 219 {
 220         int cb_cpu;
 221         struct padata_priv *padata;
 222         struct padata_serial_queue *squeue;
 223         struct padata_instance *pinst = pd->pinst;
 224
 225         /*
 226          * We need to ensure that only one cpu can work on dequeueing of
 227          * the reorder queue the time. Calculating in which percpu reorder
 228          * queue the next object will arrive takes some time. A spinlock
 229          * would be highly contended. Also it is not clear in which order
 230          * the objects arrive to the reorder queues. So a cpu could wait to
 231          * get the lock just to notice that there is nothing to do at the
 232          * moment. Therefore we use a trylock and let the holder of the lock
 233          * care for all the objects enqueued during the holdtime of the lock.
 234          */
 235         if (!spin_trylock_bh(&pd->lock))
 236                 return;
 237
 238         while (1) {
 239                 padata = padata_get_next(pd);
 240
 241                 /*
 242                  * All reorder queues are empty, or the next object that needs
 243                  * serialization is parallel processed by another cpu and is
 244                  * still on it's way to the cpu's reorder queue, nothing to
 245                  * do for now.
 246                  */
 247                 if (!padata || PTR_ERR(padata) == -EINPROGRESS)
 248                         break;
 249
 250                 /*
 251                  * This cpu has to do the parallel processing of the next
 252                  * object. It's waiting in the cpu's parallelization queue,
 253                  * so exit immediately.
 254                  */
 255                 if (PTR_ERR(padata) == -ENODATA) {
 256                         del_timer(&pd->timer);
 257                         spin_unlock_bh(&pd->lock);
 258                         return;
 259                 }
 260
 261                 cb_cpu = padata->cb_cpu;
 262                 squeue = per_cpu_ptr(pd->squeue, cb_cpu);
 263
 264                 spin_lock(&squeue->serial.lock);
 265                 list_add_tail(&padata->list, &squeue->serial.list);
 266                 spin_unlock(&squeue->serial.lock);
 267
 268                 queue_work_on(cb_cpu, pinst->wq, &squeue->work);
 269         }
 270
 271         spin_unlock_bh(&pd->lock);
 272
 273         /*
 274          * The next object that needs serialization might have arrived to
 275          * the reorder queues in the meantime, we will be called again
 276          * from the timer function if no one else cares for it.
 277          *
 278          * Ensure reorder_objects is read after pd->lock is dropped so we see
 279          * an increment from another task in padata_do_serial.  Pairs with
 280          * smp_mb__after_atomic in padata_do_serial.
 281          */
 282         smp_mb();
 283         if (atomic_read(&pd->reorder_objects)
 284                         && !(pinst->flags & PADATA_RESET))
 285                 mod_timer(&pd->timer, jiffies + HZ);
 286         else
 287                 del_timer(&pd->timer);
 288
 289         return;
 290 }
 291
 292 static void padata_reorder_timer(unsigned long arg)
 293 {
 294         struct parallel_data *pd = (struct parallel_data *)arg;
 295
 296         padata_reorder(pd);
 297 }
 298
 299 static void padata_serial_worker(struct work_struct *serial_work)
 300 {
 301         struct padata_serial_queue *squeue;
 302         struct parallel_data *pd;
 303         LIST_HEAD(local_list);
 304
 305         local_bh_disable();
 306         squeue = container_of(serial_work, struct padata_serial_queue, work);
 307         pd = squeue->pd;
 308
 309         spin_lock(&squeue->serial.lock);
 310         list_replace_init(&squeue->serial.list, &local_list);
 311         spin_unlock(&squeue->serial.lock);
 312
 313         while (!list_empty(&local_list)) {
 314                 struct padata_priv *padata;
 315
 316                 padata = list_entry(local_list.next,
 317                                     struct padata_priv, list);
 318
 319                 list_del_init(&padata->list);
 320
 321                 padata->serial(padata);
 322                 atomic_dec(&pd->refcnt);
 323         }
 324         local_bh_enable();
 325 }
 326
 327 /**
 328  * padata_do_serial - padata serialization function
 329  *
 330  * @padata: object to be serialized.
 331  *
 332  * padata_do_serial must be called for every parallelized object.
 333  * The serialization callback function will run with BHs off.
 334  */
 335 void padata_do_serial(struct padata_priv *padata)
 336 {
 337         int cpu;
 338         struct padata_parallel_queue *pqueue;
 339         struct parallel_data *pd;
 340
 341         pd = padata->pd;
 342
 343         cpu = get_cpu();
 344         pqueue = per_cpu_ptr(pd->pqueue, cpu);
 345
 346         spin_lock(&pqueue->reorder.lock);
 347         atomic_inc(&pd->reorder_objects);
 348         list_add_tail(&padata->list, &pqueue->reorder.list);
 349         spin_unlock(&pqueue->reorder.lock);
 350
 351         /*
 352          * Ensure the atomic_inc of reorder_objects above is ordered correctly
 353          * with the trylock of pd->lock in padata_reorder.  Pairs with smp_mb
 354          * in padata_reorder.
 355          */
 356         smp_mb__after_atomic();
 357
 358         put_cpu();
 359
 360         padata_reorder(pd);
 361 }
 362 EXPORT_SYMBOL(padata_do_serial);
 363
 364 static int padata_setup_cpumasks(struct parallel_data *pd,
 365                                  const struct cpumask *pcpumask,
 366                                  const struct cpumask *cbcpumask)
 367 {
 368         if (!alloc_cpumask_var(&pd->cpumask.pcpu, GFP_KERNEL))
 369                 return -ENOMEM;
 370
 371         cpumask_and(pd->cpumask.pcpu, pcpumask, cpu_online_mask);
 372         if (!alloc_cpumask_var(&pd->cpumask.cbcpu, GFP_KERNEL)) {
 373                 free_cpumask_var(pd->cpumask.pcpu);
 374                 return -ENOMEM;
 375         }
 376
 377         cpumask_and(pd->cpumask.cbcpu, cbcpumask, cpu_online_mask);
 378         return 0;
 379 }
 380
 381 static void __padata_list_init(struct padata_list *pd_list)
 382 {
 383         INIT_LIST_HEAD(&pd_list->list);
 384         spin_lock_init(&pd_list->lock);
 385 }
 386
 387 /* Initialize all percpu queues used by serial workers */
 388 static void padata_init_squeues(struct parallel_data *pd)
 389 {
 390         int cpu;
 391         struct padata_serial_queue *squeue;
 392
 393         for_each_cpu(cpu, pd->cpumask.cbcpu) {
 394                 squeue = per_cpu_ptr(pd->squeue, cpu);
 395                 squeue->pd = pd;
 396                 __padata_list_init(&squeue->serial);
 397                 INIT_WORK(&squeue->work, padata_serial_worker);
 398         }
 399 }
 400
 401 /* Initialize all percpu queues used by parallel workers */
 402 static void padata_init_pqueues(struct parallel_data *pd)
 403 {
 404         int cpu_index, cpu;
 405         struct padata_parallel_queue *pqueue;
 406
 407         cpu_index = 0;
 408         for_each_cpu(cpu, pd->cpumask.pcpu) {
 409                 pqueue = per_cpu_ptr(pd->pqueue, cpu);
 410                 pqueue->pd = pd;
 411                 pqueue->cpu_index = cpu_index;
 412                 cpu_index++;
 413
 414                 __padata_list_init(&pqueue->reorder);
 415                 __padata_list_init(&pqueue->parallel);
 416                 INIT_WORK(&pqueue->work, padata_parallel_worker);
 417                 atomic_set(&pqueue->num_obj, 0);
 418         }
 419 }
 420
 421 /* Allocate and initialize the internal cpumask dependend resources. */
 422 static struct parallel_data *padata_alloc_pd(struct padata_instance *pinst,
 423                                              const struct cpumask *pcpumask,
 424                                              const struct cpumask *cbcpumask)
 425 {
 426         struct parallel_data *pd;
 427
 428         pd = kzalloc(sizeof(struct parallel_data), GFP_KERNEL);
 429         if (!pd)
 430                 goto err;
 431
 432         pd->pqueue = alloc_percpu(struct padata_parallel_queue);
 433         if (!pd->pqueue)
 434                 goto err_free_pd;
 435
 436         pd->squeue = alloc_percpu(struct padata_serial_queue);
 437         if (!pd->squeue)
 438                 goto err_free_pqueue;
 439         if (padata_setup_cpumasks(pd, pcpumask, cbcpumask) < 0)
 440                 goto err_free_squeue;
 441
 442         padata_init_pqueues(pd);
 443         padata_init_squeues(pd);
 444         setup_timer(&pd->timer, padata_reorder_timer, (unsigned long)pd);
 445         atomic_set(&pd->seq_nr, -1);
 446         atomic_set(&pd->reorder_objects, 0);
 447         atomic_set(&pd->refcnt, 0);
 448         pd->pinst = pinst;
 449         spin_lock_init(&pd->lock);
 450
 451         return pd;
 452
 453 err_free_squeue:
 454         free_percpu(pd->squeue);
 455 err_free_pqueue:
 456         free_percpu(pd->pqueue);
 457 err_free_pd:
 458         kfree(pd);
 459 err:
 460         return NULL;
 461 }
 462
 463 static void padata_free_pd(struct parallel_data *pd)
 464 {
 465         free_cpumask_var(pd->cpumask.pcpu);
 466         free_cpumask_var(pd->cpumask.cbcpu);
 467         free_percpu(pd->pqueue);
 468         free_percpu(pd->squeue);
 469         kfree(pd);
 470 }
 471
 472 /* Flush all objects out of the padata queues. */
 473 static void padata_flush_queues(struct parallel_data *pd)
 474 {
 475         int cpu;
 476         struct padata_parallel_queue *pqueue;
 477         struct padata_serial_queue *squeue;
 478
 479         for_each_cpu(cpu, pd->cpumask.pcpu) {
 480                 pqueue = per_cpu_ptr(pd->pqueue, cpu);
 481                 flush_work(&pqueue->work);
 482         }
 483
 484         del_timer_sync(&pd->timer);
 485
 486         if (atomic_read(&pd->reorder_objects))
 487                 padata_reorder(pd);
 488
 489         for_each_cpu(cpu, pd->cpumask.cbcpu) {
 490                 squeue = per_cpu_ptr(pd->squeue, cpu);
 491                 flush_work(&squeue->work);
 492         }
 493
 494         BUG_ON(atomic_read(&pd->refcnt) != 0);
 495 }
 496
 497 static void __padata_start(struct padata_instance *pinst)
 498 {
 499         pinst->flags |= PADATA_INIT;
 500 }
 501
 502 static void __padata_stop(struct padata_instance *pinst)
 503 {
 504         if (!(pinst->flags & PADATA_INIT))
 505                 return;
 506
 507         pinst->flags &= ~PADATA_INIT;
 508
 509         synchronize_rcu();
 510
 511         get_online_cpus();
 512         padata_flush_queues(pinst->pd);
 513         put_online_cpus();
 514 }
 515
 516 /* Replace the internal control structure with a new one. */
 517 static void padata_replace(struct padata_instance *pinst,
 518                            struct parallel_data *pd_new)
 519 {
 520         struct parallel_data *pd_old = pinst->pd;
 521         int notification_mask = 0;
 522
 523         pinst->flags |= PADATA_RESET;
 524
 525         rcu_assign_pointer(pinst->pd, pd_new);
 526
 527         synchronize_rcu();
 528
 529         if (!cpumask_equal(pd_old->cpumask.pcpu, pd_new->cpumask.pcpu))
 530                 notification_mask |= PADATA_CPU_PARALLEL;
 531         if (!cpumask_equal(pd_old->cpumask.cbcpu, pd_new->cpumask.cbcpu))
 532                 notification_mask |= PADATA_CPU_SERIAL;
 533
 534         padata_flush_queues(pd_old);
 535         padata_free_pd(pd_old);
 536
 537         if (notification_mask)
 538                 blocking_notifier_call_chain(&pinst->cpumask_change_notifier,
 539                                              notification_mask,
 540                                              &pd_new->cpumask);
 541
 542         pinst->flags &= ~PADATA_RESET;
 543 }
 544
 545 /**
 546  * padata_register_cpumask_notifier - Registers a notifier that will be called
 547  *                             if either pcpu or cbcpu or both cpumasks change.
 548  *
 549  * @pinst: A poineter to padata instance
 550  * @nblock: A pointer to notifier block.
 551  */
 552 int padata_register_cpumask_notifier(struct padata_instance *pinst,
 553                                      struct notifier_block *nblock)
 554 {
 555         return blocking_notifier_chain_register(&pinst->cpumask_change_notifier,
 556                                                 nblock);
 557 }
 558 EXPORT_SYMBOL(padata_register_cpumask_notifier);
 559
 560 /**
 561  * padata_unregister_cpumask_notifier - Unregisters cpumask notifier
 562  *        registered earlier  using padata_register_cpumask_notifier
 563  *
 564  * @pinst: A pointer to data instance.
 565  * @nlock: A pointer to notifier block.
 566  */
 567 int padata_unregister_cpumask_notifier(struct padata_instance *pinst,
 568                                        struct notifier_block *nblock)
 569 {
 570         return blocking_notifier_chain_unregister(
 571                 &pinst->cpumask_change_notifier,
 572                 nblock);
 573 }
 574 EXPORT_SYMBOL(padata_unregister_cpumask_notifier);
 575
 576
 577 /* If cpumask contains no active cpu, we mark the instance as invalid. */
 578 static bool padata_validate_cpumask(struct padata_instance *pinst,
 579                                     const struct cpumask *cpumask)
 580 {
 581         if (!cpumask_intersects(cpumask, cpu_online_mask)) {
 582                 pinst->flags |= PADATA_INVALID;
 583                 return false;
 584         }
 585
 586         pinst->flags &= ~PADATA_INVALID;
 587         return true;
 588 }
 589
 590 static int __padata_set_cpumasks(struct padata_instance *pinst,
 591                                  cpumask_var_t pcpumask,
 592                                  cpumask_var_t cbcpumask)
 593 {
 594         int valid;
 595         struct parallel_data *pd;
 596
 597         valid = padata_validate_cpumask(pinst, pcpumask);
 598         if (!valid) {
 599                 __padata_stop(pinst);
 600                 goto out_replace;
 601         }
 602
 603         valid = padata_validate_cpumask(pinst, cbcpumask);
 604         if (!valid)
 605                 __padata_stop(pinst);
 606
 607 out_replace:
 608         pd = padata_alloc_pd(pinst, pcpumask, cbcpumask);
 609         if (!pd)
 610                 return -ENOMEM;
 611
 612         cpumask_copy(pinst->cpumask.pcpu, pcpumask);
 613         cpumask_copy(pinst->cpumask.cbcpu, cbcpumask);
 614
 615         padata_replace(pinst, pd);
 616
 617         if (valid)
 618                 __padata_start(pinst);
 619
 620         return 0;
 621 }
 622
 623 /**
 624  * padata_set_cpumask: Sets specified by @cpumask_type cpumask to the value
 625  *                     equivalent to @cpumask.
 626  *
 627  * @pinst: padata instance
 628  * @cpumask_type: PADATA_CPU_SERIAL or PADATA_CPU_PARALLEL corresponding
 629  *                to parallel and serial cpumasks respectively.
 630  * @cpumask: the cpumask to use
 631  */
 632 int padata_set_cpumask(struct padata_instance *pinst, int cpumask_type,
 633                        cpumask_var_t cpumask)
 634 {
 635         struct cpumask *serial_mask, *parallel_mask;
 636         int err = -EINVAL;
 637
 638         mutex_lock(&pinst->lock);
 639         get_online_cpus();
 640
 641         switch (cpumask_type) {
 642         case PADATA_CPU_PARALLEL:
 643                 serial_mask = pinst->cpumask.cbcpu;
 644                 parallel_mask = cpumask;
 645                 break;
 646         case PADATA_CPU_SERIAL:
 647                 parallel_mask = pinst->cpumask.pcpu;
 648                 serial_mask = cpumask;
 649                 break;
 650         default:
 651                  goto out;
 652         }
 653
 654         err =  __padata_set_cpumasks(pinst, parallel_mask, serial_mask);
 655
 656 out:
 657         put_online_cpus();
 658         mutex_unlock(&pinst->lock);
 659
 660         return err;
 661 }
 662 EXPORT_SYMBOL(padata_set_cpumask);
 663
 664 /**
 665  * padata_start - start the parallel processing
 666  *
 667  * @pinst: padata instance to start
 668  */
 669 int padata_start(struct padata_instance *pinst)
 670 {
 671         int err = 0;
 672
 673         mutex_lock(&pinst->lock);
 674
 675         if (pinst->flags & PADATA_INVALID)
 676                 err = -EINVAL;
 677
 678          __padata_start(pinst);
 679
 680         mutex_unlock(&pinst->lock);
 681
 682         return err;
 683 }
 684 EXPORT_SYMBOL(padata_start);
 685
 686 /**
 687  * padata_stop - stop the parallel processing
 688  *
 689  * @pinst: padata instance to stop
 690  */
 691 void padata_stop(struct padata_instance *pinst)
 692 {
 693         mutex_lock(&pinst->lock);
 694         __padata_stop(pinst);
 695         mutex_unlock(&pinst->lock);
 696 }
 697 EXPORT_SYMBOL(padata_stop);
 698
 699 #ifdef CONFIG_HOTPLUG_CPU
 700
 701 static int __padata_add_cpu(struct padata_instance *pinst, int cpu)
 702 {
 703         struct parallel_data *pd;
 704
 705         if (cpumask_test_cpu(cpu, cpu_online_mask)) {
 706                 pd = padata_alloc_pd(pinst, pinst->cpumask.pcpu,
 707                                      pinst->cpumask.cbcpu);
 708                 if (!pd)
 709                         return -ENOMEM;
 710
 711                 padata_replace(pinst, pd);
 712
 713                 if (padata_validate_cpumask(pinst, pinst->cpumask.pcpu) &&
 714                     padata_validate_cpumask(pinst, pinst->cpumask.cbcpu))
 715                         __padata_start(pinst);
 716         }
 717
 718         return 0;
 719 }
 720
 721 static int __padata_remove_cpu(struct padata_instance *pinst, int cpu)
 722 {
 723         struct parallel_data *pd = NULL;
 724
 725         if (cpumask_test_cpu(cpu, cpu_online_mask)) {
 726
 727                 if (!padata_validate_cpumask(pinst, pinst->cpumask.pcpu) ||
 728                     !padata_validate_cpumask(pinst, pinst->cpumask.cbcpu))
 729                         __padata_stop(pinst);
 730
 731                 pd = padata_alloc_pd(pinst, pinst->cpumask.pcpu,
 732                                      pinst->cpumask.cbcpu);
 733                 if (!pd)
 734                         return -ENOMEM;
 735
 736                 padata_replace(pinst, pd);
 737
 738                 cpumask_clear_cpu(cpu, pd->cpumask.cbcpu);
 739                 cpumask_clear_cpu(cpu, pd->cpumask.pcpu);
 740         }
 741
 742         return 0;
 743 }
 744
 745  /**
 746  * padata_remove_cpu - remove a cpu from the one or both(serial and parallel)
 747  *                     padata cpumasks.
 748  *
 749  * @pinst: padata instance
 750  * @cpu: cpu to remove
 751  * @mask: bitmask specifying from which cpumask @cpu should be removed
 752  *        The @mask may be any combination of the following flags:
 753  *          PADATA_CPU_SERIAL   - serial cpumask
 754  *          PADATA_CPU_PARALLEL - parallel cpumask
 755  */
 756 int padata_remove_cpu(struct padata_instance *pinst, int cpu, int mask)
 757 {
 758         int err;
 759
 760         if (!(mask & (PADATA_CPU_SERIAL | PADATA_CPU_PARALLEL)))
 761                 return -EINVAL;
 762
 763         mutex_lock(&pinst->lock);
 764
 765         get_online_cpus();
 766         if (mask & PADATA_CPU_SERIAL)
 767                 cpumask_clear_cpu(cpu, pinst->cpumask.cbcpu);
 768         if (mask & PADATA_CPU_PARALLEL)
 769                 cpumask_clear_cpu(cpu, pinst->cpumask.pcpu);
 770
 771         err = __padata_remove_cpu(pinst, cpu);
 772         put_online_cpus();
 773
 774         mutex_unlock(&pinst->lock);
 775
 776         return err;
 777 }
 778 EXPORT_SYMBOL(padata_remove_cpu);
 779
 780 static inline int pinst_has_cpu(struct padata_instance *pinst, int cpu)
 781 {
 782         return cpumask_test_cpu(cpu, pinst->cpumask.pcpu) ||
 783                 cpumask_test_cpu(cpu, pinst->cpumask.cbcpu);
 784 }
 785
 786 static int padata_cpu_online(unsigned int cpu, struct hlist_node *node)
 787 {
 788         struct padata_instance *pinst;
 789         int ret;
 790
 791         pinst = hlist_entry_safe(node, struct padata_instance, node);
 792         if (!pinst_has_cpu(pinst, cpu))
 793                 return 0;
 794
 795         mutex_lock(&pinst->lock);
 796         ret = __padata_add_cpu(pinst, cpu);
 797         mutex_unlock(&pinst->lock);
 798         return ret;
 799 }
 800
 801 static int padata_cpu_prep_down(unsigned int cpu, struct hlist_node *node)
 802 {
 803         struct padata_instance *pinst;
 804         int ret;
 805
 806         pinst = hlist_entry_safe(node, struct padata_instance, node);
 807         if (!pinst_has_cpu(pinst, cpu))
 808                 return 0;
 809
 810         mutex_lock(&pinst->lock);
 811         ret = __padata_remove_cpu(pinst, cpu);
 812         mutex_unlock(&pinst->lock);
 813         return ret;
 814 }
 815
 816 static enum cpuhp_state hp_online;
 817 #endif
 818
 819 static void __padata_free(struct padata_instance *pinst)
 820 {
 821 #ifdef CONFIG_HOTPLUG_CPU
 822         cpuhp_state_remove_instance_nocalls(hp_online, &pinst->node);
 823 #endif
 824
 825         padata_stop(pinst);
 826         padata_free_pd(pinst->pd);
 827         free_cpumask_var(pinst->cpumask.pcpu);
 828         free_cpumask_var(pinst->cpumask.cbcpu);
 829         kfree(pinst);
 830 }
 831
 832 #define kobj2pinst(_kobj)                                       \
 833         container_of(_kobj, struct padata_instance, kobj)
 834 #define attr2pentry(_attr)                                      \
 835         container_of(_attr, struct padata_sysfs_entry, attr)
 836
 837 static void padata_sysfs_release(struct kobject *kobj)
 838 {
 839         struct padata_instance *pinst = kobj2pinst(kobj);
 840         __padata_free(pinst);
 841 }
 842
 843 struct padata_sysfs_entry {
 844         struct attribute attr;
 845         ssize_t (*show)(struct padata_instance *, struct attribute *, char *);
 846         ssize_t (*store)(struct padata_instance *, struct attribute *,
 847                          const char *, size_t);
 848 };
 849
 850 static ssize_t show_cpumask(struct padata_instance *pinst,
 851                             struct attribute *attr,  char *buf)
 852 {
 853         struct cpumask *cpumask;
 854         ssize_t len;
 855
 856         mutex_lock(&pinst->lock);
 857         if (!strcmp(attr->name, "serial_cpumask"))
 858                 cpumask = pinst->cpumask.cbcpu;
 859         else
 860                 cpumask = pinst->cpumask.pcpu;
 861
 862         len = snprintf(buf, PAGE_SIZE, "%*pb\n",
 863                        nr_cpu_ids, cpumask_bits(cpumask));
 864         mutex_unlock(&pinst->lock);
 865         return len < PAGE_SIZE ? len : -EINVAL;
 866 }
 867
 868 static ssize_t store_cpumask(struct padata_instance *pinst,
 869                              struct attribute *attr,
 870                              const char *buf, size_t count)
 871 {
 872         cpumask_var_t new_cpumask;
 873         ssize_t ret;
 874         int mask_type;
 875
 876         if (!alloc_cpumask_var(&new_cpumask, GFP_KERNEL))
 877                 return -ENOMEM;
 878
 879         ret = bitmap_parse(buf, count, cpumask_bits(new_cpumask),
 880                            nr_cpumask_bits);
 881         if (ret < 0)
 882                 goto out;
 883
 884         mask_type = !strcmp(attr->name, "serial_cpumask") ?
 885                 PADATA_CPU_SERIAL : PADATA_CPU_PARALLEL;
 886         ret = padata_set_cpumask(pinst, mask_type, new_cpumask);
 887         if (!ret)
 888                 ret = count;
 889
 890 out:
 891         free_cpumask_var(new_cpumask);
 892         return ret;
 893 }
 894
 895 #define PADATA_ATTR_RW(_name, _show_name, _store_name)          \
 896         static struct padata_sysfs_entry _name##_attr =         \
 897                 __ATTR(_name, 0644, _show_name, _store_name)
 898 #define PADATA_ATTR_RO(_name, _show_name)               \
 899         static struct padata_sysfs_entry _name##_attr = \
 900                 __ATTR(_name, 0400, _show_name, NULL)
 901
 902 PADATA_ATTR_RW(serial_cpumask, show_cpumask, store_cpumask);
 903 PADATA_ATTR_RW(parallel_cpumask, show_cpumask, store_cpumask);
 904
 905 /*
 906  * Padata sysfs provides the following objects:
 907  * serial_cpumask   [RW] - cpumask for serial workers
 908  * parallel_cpumask [RW] - cpumask for parallel workers
 909  */
 910 static struct attribute *padata_default_attrs[] = {
 911         &serial_cpumask_attr.attr,
 912         &parallel_cpumask_attr.attr,
 913         NULL,
 914 };
 915
 916 static ssize_t padata_sysfs_show(struct kobject *kobj,
 917                                  struct attribute *attr, char *buf)
 918 {
 919         struct padata_instance *pinst;
 920         struct padata_sysfs_entry *pentry;
 921         ssize_t ret = -EIO;
 922
 923         pinst = kobj2pinst(kobj);
 924         pentry = attr2pentry(attr);
 925         if (pentry->show)
 926                 ret = pentry->show(pinst, attr, buf);
 927
 928         return ret;
 929 }
 930
 931 static ssize_t padata_sysfs_store(struct kobject *kobj, struct attribute *attr,
 932                                   const char *buf, size_t count)
 933 {
 934         struct padata_instance *pinst;
 935         struct padata_sysfs_entry *pentry;
 936         ssize_t ret = -EIO;
 937
 938         pinst = kobj2pinst(kobj);
 939         pentry = attr2pentry(attr);
 940         if (pentry->show)
 941                 ret = pentry->store(pinst, attr, buf, count);
 942
 943         return ret;
 944 }
 945
 946 static const struct sysfs_ops padata_sysfs_ops = {
 947         .show = padata_sysfs_show,
 948         .store = padata_sysfs_store,
 949 };
 950
 951 static struct kobj_type padata_attr_type = {
 952         .sysfs_ops = &padata_sysfs_ops,
 953         .default_attrs = padata_default_attrs,
 954         .release = padata_sysfs_release,
 955 };
 956
 957 /**
 958  * padata_alloc_possible - Allocate and initialize padata instance.
 959  *                         Use the cpu_possible_mask for serial and
 960  *                         parallel workers.
 961  *
 962  * @wq: workqueue to use for the allocated padata instance
 963  */
 964 struct padata_instance *padata_alloc_possible(struct workqueue_struct *wq)
 965 {
 966         return padata_alloc(wq, cpu_possible_mask, cpu_possible_mask);
 967 }
 968 EXPORT_SYMBOL(padata_alloc_possible);
 969
 970 /**
 971  * padata_alloc - allocate and initialize a padata instance and specify
 972  *                cpumasks for serial and parallel workers.
 973  *
 974  * @wq: workqueue to use for the allocated padata instance
 975  * @pcpumask: cpumask that will be used for padata parallelization
 976  * @cbcpumask: cpumask that will be used for padata serialization
 977  */
 978 struct padata_instance *padata_alloc(struct workqueue_struct *wq,
 979                                      const struct cpumask *pcpumask,
 980                                      const struct cpumask *cbcpumask)
 981 {
 982         struct padata_instance *pinst;
 983         struct parallel_data *pd = NULL;
 984
 985         pinst = kzalloc(sizeof(struct padata_instance), GFP_KERNEL);
 986         if (!pinst)
 987                 goto err;
 988
 989         get_online_cpus();
 990         if (!alloc_cpumask_var(&pinst->cpumask.pcpu, GFP_KERNEL))
 991                 goto err_free_inst;
 992         if (!alloc_cpumask_var(&pinst->cpumask.cbcpu, GFP_KERNEL)) {
 993                 free_cpumask_var(pinst->cpumask.pcpu);
 994                 goto err_free_inst;
 995         }
 996         if (!padata_validate_cpumask(pinst, pcpumask) ||
 997             !padata_validate_cpumask(pinst, cbcpumask))
 998                 goto err_free_masks;
 999
1000         pd = padata_alloc_pd(pinst, pcpumask, cbcpumask);
1001         if (!pd)
1002                 goto err_free_masks;
1003
1004         rcu_assign_pointer(pinst->pd, pd);
1005
1006         pinst->wq = wq;
1007
1008         cpumask_copy(pinst->cpumask.pcpu, pcpumask);
1009         cpumask_copy(pinst->cpumask.cbcpu, cbcpumask);
1010
1011         pinst->flags = 0;
1012
1013         put_online_cpus();
1014
1015         BLOCKING_INIT_NOTIFIER_HEAD(&pinst->cpumask_change_notifier);
1016         kobject_init(&pinst->kobj, &padata_attr_type);
1017         mutex_init(&pinst->lock);
1018
1019 #ifdef CONFIG_HOTPLUG_CPU
1020         cpuhp_state_add_instance_nocalls(hp_online, &pinst->node);
1021 #endif
1022         return pinst;
1023
1024 err_free_masks:
1025         free_cpumask_var(pinst->cpumask.pcpu);
1026         free_cpumask_var(pinst->cpumask.cbcpu);
1027 err_free_inst:
1028         kfree(pinst);
1029         put_online_cpus();
1030 err:
1031         return NULL;
1032 }
1033
1034 /**
1035  * padata_free - free a padata instance
1036  *
1037  * @padata_inst: padata instance to free
1038  */
1039 void padata_free(struct padata_instance *pinst)
1040 {
1041         kobject_put(&pinst->kobj);
1042 }
1043 EXPORT_SYMBOL(padata_free);
1044
1045 #ifdef CONFIG_HOTPLUG_CPU
1046
1047 static __init int padata_driver_init(void)
1048 {
1049         int ret;
1050
1051         ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "padata:online",
1052                                       padata_cpu_online,
1053                                       padata_cpu_prep_down);
1054         if (ret < 0)
1055                 return ret;
1056         hp_online = ret;
1057         return 0;
1058 }
1059 module_init(padata_driver_init);
1060
1061 static __exit void padata_driver_exit(void)
1062 {
1063         cpuhp_remove_multi_state(hp_online);
1064 }
1065 module_exit(padata_driver_exit);
1066 #endif