kernel/rcu/rcuperf.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * Read-Copy Update module-based performance-test facility
   4  *
   5  * Copyright (C) IBM Corporation, 2015
   6  *
   7  * Authors: Paul E. McKenney <paulmck@linux.ibm.com>
   8  */
   9
  10 #define pr_fmt(fmt) fmt
  11
  12 #include <linux/types.h>
  13 #include <linux/kernel.h>
  14 #include <linux/init.h>
  15 #include <linux/module.h>
  16 #include <linux/kthread.h>
  17 #include <linux/err.h>
  18 #include <linux/spinlock.h>
  19 #include <linux/smp.h>
  20 #include <linux/rcupdate.h>
  21 #include <linux/interrupt.h>
  22 #include <linux/sched.h>
  23 #include <uapi/linux/sched/types.h>
  24 #include <linux/atomic.h>
  25 #include <linux/bitops.h>
  26 #include <linux/completion.h>
  27 #include <linux/moduleparam.h>
  28 #include <linux/percpu.h>
  29 #include <linux/notifier.h>
  30 #include <linux/reboot.h>
  31 #include <linux/freezer.h>
  32 #include <linux/cpu.h>
  33 #include <linux/delay.h>
  34 #include <linux/stat.h>
  35 #include <linux/srcu.h>
  36 #include <linux/slab.h>
  37 #include <asm/byteorder.h>
  38 #include <linux/torture.h>
  39 #include <linux/vmalloc.h>
  40
  41 #include "rcu.h"
  42
  43 MODULE_LICENSE("GPL");
  44 MODULE_AUTHOR("Paul E. McKenney <paulmck@linux.ibm.com>");
  45
  46 #define PERF_FLAG "-perf:"
  47 #define PERFOUT_STRING(s) \
  48         pr_alert("%s" PERF_FLAG " %s\n", perf_type, s)
  49 #define VERBOSE_PERFOUT_STRING(s) \
  50         do { if (verbose) pr_alert("%s" PERF_FLAG " %s\n", perf_type, s); } while (0)
  51 #define VERBOSE_PERFOUT_ERRSTRING(s) \
  52         do { if (verbose) pr_alert("%s" PERF_FLAG "!!! %s\n", perf_type, s); } while (0)
  53
  54 /*
  55  * The intended use cases for the nreaders and nwriters module parameters
  56  * are as follows:
  57  *
  58  * 1.   Specify only the nr_cpus kernel boot parameter.  This will
  59  *      set both nreaders and nwriters to the value specified by
  60  *      nr_cpus for a mixed reader/writer test.
  61  *
  62  * 2.   Specify the nr_cpus kernel boot parameter, but set
  63  *      rcuperf.nreaders to zero.  This will set nwriters to the
  64  *      value specified by nr_cpus for an update-only test.
  65  *
  66  * 3.   Specify the nr_cpus kernel boot parameter, but set
  67  *      rcuperf.nwriters to zero.  This will set nreaders to the
  68  *      value specified by nr_cpus for a read-only test.
  69  *
  70  * Various other use cases may of course be specified.
  71  */
  72
  73 #ifdef MODULE
  74 # define RCUPERF_SHUTDOWN 0
  75 #else
  76 # define RCUPERF_SHUTDOWN 1
  77 #endif
  78
  79 torture_param(bool, gp_async, false, "Use asynchronous GP wait primitives");
  80 torture_param(int, gp_async_max, 1000, "Max # outstanding waits per reader");
  81 torture_param(bool, gp_exp, false, "Use expedited GP wait primitives");
  82 torture_param(int, holdoff, 10, "Holdoff time before test start (s)");
  83 torture_param(int, nreaders, -1, "Number of RCU reader threads");
  84 torture_param(int, nwriters, -1, "Number of RCU updater threads");
  85 torture_param(bool, shutdown, RCUPERF_SHUTDOWN,
  86               "Shutdown at end of performance tests.");
  87 torture_param(int, verbose, 1, "Enable verbose debugging printk()s");
  88 torture_param(int, writer_holdoff, 0, "Holdoff (us) between GPs, zero to disable");
  89 torture_param(int, kfree_rcu_test, 0, "Do we run a kfree_rcu() perf test?");
  90
  91 static char *perf_type = "rcu";
  92 module_param(perf_type, charp, 0444);
  93 MODULE_PARM_DESC(perf_type, "Type of RCU to performance-test (rcu, srcu, ...)");
  94
  95 static int nrealreaders;
  96 static int nrealwriters;
  97 static struct task_struct **writer_tasks;
  98 static struct task_struct **reader_tasks;
  99 static struct task_struct *shutdown_task;
 100
 101 static u64 **writer_durations;
 102 static int *writer_n_durations;
 103 static atomic_t n_rcu_perf_reader_started;
 104 static atomic_t n_rcu_perf_writer_started;
 105 static atomic_t n_rcu_perf_writer_finished;
 106 static wait_queue_head_t shutdown_wq;
 107 static u64 t_rcu_perf_writer_started;
 108 static u64 t_rcu_perf_writer_finished;
 109 static unsigned long b_rcu_gp_test_started;
 110 static unsigned long b_rcu_gp_test_finished;
 111 static DEFINE_PER_CPU(atomic_t, n_async_inflight);
 112
 113 #define MAX_MEAS 10000
 114 #define MIN_MEAS 100
 115
 116 /*
 117  * Operations vector for selecting different types of tests.
 118  */
 119
 120 struct rcu_perf_ops {
 121         int ptype;
 122         void (*init)(void);
 123         void (*cleanup)(void);
 124         int (*readlock)(void);
 125         void (*readunlock)(int idx);
 126         unsigned long (*get_gp_seq)(void);
 127         unsigned long (*gp_diff)(unsigned long new, unsigned long old);
 128         unsigned long (*exp_completed)(void);
 129         void (*async)(struct rcu_head *head, rcu_callback_t func);
 130         void (*gp_barrier)(void);
 131         void (*sync)(void);
 132         void (*exp_sync)(void);
 133         const char *name;
 134 };
 135
 136 static struct rcu_perf_ops *cur_ops;
 137
 138 /*
 139  * Definitions for rcu perf testing.
 140  */
 141
 142 static int rcu_perf_read_lock(void) __acquires(RCU)
 143 {
 144         rcu_read_lock();
 145         return 0;
 146 }
 147
 148 static void rcu_perf_read_unlock(int idx) __releases(RCU)
 149 {
 150         rcu_read_unlock();
 151 }
 152
 153 static unsigned long __maybe_unused rcu_no_completed(void)
 154 {
 155         return 0;
 156 }
 157
 158 static void rcu_sync_perf_init(void)
 159 {
 160 }
 161
 162 static struct rcu_perf_ops rcu_ops = {
 163         .ptype          = RCU_FLAVOR,
 164         .init           = rcu_sync_perf_init,
 165         .readlock       = rcu_perf_read_lock,
 166         .readunlock     = rcu_perf_read_unlock,
 167         .get_gp_seq     = rcu_get_gp_seq,
 168         .gp_diff        = rcu_seq_diff,
 169         .exp_completed  = rcu_exp_batches_completed,
 170         .async          = call_rcu,
 171         .gp_barrier     = rcu_barrier,
 172         .sync           = synchronize_rcu,
 173         .exp_sync       = synchronize_rcu_expedited,
 174         .name           = "rcu"
 175 };
 176
 177 /*
 178  * Definitions for srcu perf testing.
 179  */
 180
 181 DEFINE_STATIC_SRCU(srcu_ctl_perf);
 182 static struct srcu_struct *srcu_ctlp = &srcu_ctl_perf;
 183
 184 static int srcu_perf_read_lock(void) __acquires(srcu_ctlp)
 185 {
 186         return srcu_read_lock(srcu_ctlp);
 187 }
 188
 189 static void srcu_perf_read_unlock(int idx) __releases(srcu_ctlp)
 190 {
 191         srcu_read_unlock(srcu_ctlp, idx);
 192 }
 193
 194 static unsigned long srcu_perf_completed(void)
 195 {
 196         return srcu_batches_completed(srcu_ctlp);
 197 }
 198
 199 static void srcu_call_rcu(struct rcu_head *head, rcu_callback_t func)
 200 {
 201         call_srcu(srcu_ctlp, head, func);
 202 }
 203
 204 static void srcu_rcu_barrier(void)
 205 {
 206         srcu_barrier(srcu_ctlp);
 207 }
 208
 209 static void srcu_perf_synchronize(void)
 210 {
 211         synchronize_srcu(srcu_ctlp);
 212 }
 213
 214 static void srcu_perf_synchronize_expedited(void)
 215 {
 216         synchronize_srcu_expedited(srcu_ctlp);
 217 }
 218
 219 static struct rcu_perf_ops srcu_ops = {
 220         .ptype          = SRCU_FLAVOR,
 221         .init           = rcu_sync_perf_init,
 222         .readlock       = srcu_perf_read_lock,
 223         .readunlock     = srcu_perf_read_unlock,
 224         .get_gp_seq     = srcu_perf_completed,
 225         .gp_diff        = rcu_seq_diff,
 226         .exp_completed  = srcu_perf_completed,
 227         .async          = srcu_call_rcu,
 228         .gp_barrier     = srcu_rcu_barrier,
 229         .sync           = srcu_perf_synchronize,
 230         .exp_sync       = srcu_perf_synchronize_expedited,
 231         .name           = "srcu"
 232 };
 233
 234 static struct srcu_struct srcud;
 235
 236 static void srcu_sync_perf_init(void)
 237 {
 238         srcu_ctlp = &srcud;
 239         init_srcu_struct(srcu_ctlp);
 240 }
 241
 242 static void srcu_sync_perf_cleanup(void)
 243 {
 244         cleanup_srcu_struct(srcu_ctlp);
 245 }
 246
 247 static struct rcu_perf_ops srcud_ops = {
 248         .ptype          = SRCU_FLAVOR,
 249         .init           = srcu_sync_perf_init,
 250         .cleanup        = srcu_sync_perf_cleanup,
 251         .readlock       = srcu_perf_read_lock,
 252         .readunlock     = srcu_perf_read_unlock,
 253         .get_gp_seq     = srcu_perf_completed,
 254         .gp_diff        = rcu_seq_diff,
 255         .exp_completed  = srcu_perf_completed,
 256         .async          = srcu_call_rcu,
 257         .gp_barrier     = srcu_rcu_barrier,
 258         .sync           = srcu_perf_synchronize,
 259         .exp_sync       = srcu_perf_synchronize_expedited,
 260         .name           = "srcud"
 261 };
 262
 263 /*
 264  * Definitions for RCU-tasks perf testing.
 265  */
 266
 267 static int tasks_perf_read_lock(void)
 268 {
 269         return 0;
 270 }
 271
 272 static void tasks_perf_read_unlock(int idx)
 273 {
 274 }
 275
 276 static struct rcu_perf_ops tasks_ops = {
 277         .ptype          = RCU_TASKS_FLAVOR,
 278         .init           = rcu_sync_perf_init,
 279         .readlock       = tasks_perf_read_lock,
 280         .readunlock     = tasks_perf_read_unlock,
 281         .get_gp_seq     = rcu_no_completed,
 282         .gp_diff        = rcu_seq_diff,
 283         .async          = call_rcu_tasks,
 284         .gp_barrier     = rcu_barrier_tasks,
 285         .sync           = synchronize_rcu_tasks,
 286         .exp_sync       = synchronize_rcu_tasks,
 287         .name           = "tasks"
 288 };
 289
 290 static unsigned long rcuperf_seq_diff(unsigned long new, unsigned long old)
 291 {
 292         if (!cur_ops->gp_diff)
 293                 return new - old;
 294         return cur_ops->gp_diff(new, old);
 295 }
 296
 297 /*
 298  * If performance tests complete, wait for shutdown to commence.
 299  */
 300 static void rcu_perf_wait_shutdown(void)
 301 {
 302         cond_resched_tasks_rcu_qs();
 303         if (atomic_read(&n_rcu_perf_writer_finished) < nrealwriters)
 304                 return;
 305         while (!torture_must_stop())
 306                 schedule_timeout_uninterruptible(1);
 307 }
 308
 309 /*
 310  * RCU perf reader kthread.  Repeatedly does empty RCU read-side
 311  * critical section, minimizing update-side interference.
 312  */
 313 static int
 314 rcu_perf_reader(void *arg)
 315 {
 316         unsigned long flags;
 317         int idx;
 318         long me = (long)arg;
 319
 320         VERBOSE_PERFOUT_STRING("rcu_perf_reader task started");
 321         set_cpus_allowed_ptr(current, cpumask_of(me % nr_cpu_ids));
 322         set_user_nice(current, MAX_NICE);
 323         atomic_inc(&n_rcu_perf_reader_started);
 324
 325         do {
 326                 local_irq_save(flags);
 327                 idx = cur_ops->readlock();
 328                 cur_ops->readunlock(idx);
 329                 local_irq_restore(flags);
 330                 rcu_perf_wait_shutdown();
 331         } while (!torture_must_stop());
 332         torture_kthread_stopping("rcu_perf_reader");
 333         return 0;
 334 }
 335
 336 /*
 337  * Callback function for asynchronous grace periods from rcu_perf_writer().
 338  */
 339 static void rcu_perf_async_cb(struct rcu_head *rhp)
 340 {
 341         atomic_dec(this_cpu_ptr(&n_async_inflight));
 342         kfree(rhp);
 343 }
 344
 345 /*
 346  * RCU perf writer kthread.  Repeatedly does a grace period.
 347  */
 348 static int
 349 rcu_perf_writer(void *arg)
 350 {
 351         int i = 0;
 352         int i_max;
 353         long me = (long)arg;
 354         struct rcu_head *rhp = NULL;
 355         struct sched_param sp;
 356         bool started = false, done = false, alldone = false;
 357         u64 t;
 358         u64 *wdp;
 359         u64 *wdpp = writer_durations[me];
 360
 361         VERBOSE_PERFOUT_STRING("rcu_perf_writer task started");
 362         WARN_ON(!wdpp);
 363         set_cpus_allowed_ptr(current, cpumask_of(me % nr_cpu_ids));
 364         sp.sched_priority = 1;
 365         sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);
 366
 367         if (holdoff)
 368                 schedule_timeout_uninterruptible(holdoff * HZ);
 369
 370         /*
 371          * Wait until rcu_end_inkernel_boot() is called for normal GP tests
 372          * so that RCU is not always expedited for normal GP tests.
 373          * The system_state test is approximate, but works well in practice.
 374          */
 375         while (!gp_exp && system_state != SYSTEM_RUNNING)
 376                 schedule_timeout_uninterruptible(1);
 377
 378         t = ktime_get_mono_fast_ns();
 379         if (atomic_inc_return(&n_rcu_perf_writer_started) >= nrealwriters) {
 380                 t_rcu_perf_writer_started = t;
 381                 if (gp_exp) {
 382                         b_rcu_gp_test_started =
 383                                 cur_ops->exp_completed() / 2;
 384                 } else {
 385                         b_rcu_gp_test_started = cur_ops->get_gp_seq();
 386                 }
 387         }
 388
 389         do {
 390                 if (writer_holdoff)
 391                         udelay(writer_holdoff);
 392                 wdp = &wdpp[i];
 393                 *wdp = ktime_get_mono_fast_ns();
 394                 if (gp_async) {
 395 retry:
 396                         if (!rhp)
 397                                 rhp = kmalloc(sizeof(*rhp), GFP_KERNEL);
 398                         if (rhp && atomic_read(this_cpu_ptr(&n_async_inflight)) < gp_async_max) {
 399                                 atomic_inc(this_cpu_ptr(&n_async_inflight));
 400                                 cur_ops->async(rhp, rcu_perf_async_cb);
 401                                 rhp = NULL;
 402                         } else if (!kthread_should_stop()) {
 403                                 cur_ops->gp_barrier();
 404                                 goto retry;
 405                         } else {
 406                                 kfree(rhp); /* Because we are stopping. */
 407                         }
 408                 } else if (gp_exp) {
 409                         cur_ops->exp_sync();
 410                 } else {
 411                         cur_ops->sync();
 412                 }
 413                 t = ktime_get_mono_fast_ns();
 414                 *wdp = t - *wdp;
 415                 i_max = i;
 416                 if (!started &&
 417                     atomic_read(&n_rcu_perf_writer_started) >= nrealwriters)
 418                         started = true;
 419                 if (!done && i >= MIN_MEAS) {
 420                         done = true;
 421                         sp.sched_priority = 0;
 422                         sched_setscheduler_nocheck(current,
 423                                                    SCHED_NORMAL, &sp);
 424                         pr_alert("%s%s rcu_perf_writer %ld has %d measurements\n",
 425                                  perf_type, PERF_FLAG, me, MIN_MEAS);
 426                         if (atomic_inc_return(&n_rcu_perf_writer_finished) >=
 427                             nrealwriters) {
 428                                 schedule_timeout_interruptible(10);
 429                                 rcu_ftrace_dump(DUMP_ALL);
 430                                 PERFOUT_STRING("Test complete");
 431                                 t_rcu_perf_writer_finished = t;
 432                                 if (gp_exp) {
 433                                         b_rcu_gp_test_finished =
 434                                                 cur_ops->exp_completed() / 2;
 435                                 } else {
 436                                         b_rcu_gp_test_finished =
 437                                                 cur_ops->get_gp_seq();
 438                                 }
 439                                 if (shutdown) {
 440                                         smp_mb(); /* Assign before wake. */
 441                                         wake_up(&shutdown_wq);
 442                                 }
 443                         }
 444                 }
 445                 if (done && !alldone &&
 446                     atomic_read(&n_rcu_perf_writer_finished) >= nrealwriters)
 447                         alldone = true;
 448                 if (started && !alldone && i < MAX_MEAS - 1)
 449                         i++;
 450                 rcu_perf_wait_shutdown();
 451         } while (!torture_must_stop());
 452         if (gp_async) {
 453                 cur_ops->gp_barrier();
 454         }
 455         writer_n_durations[me] = i_max;
 456         torture_kthread_stopping("rcu_perf_writer");
 457         return 0;
 458 }
 459
 460 static void
 461 rcu_perf_print_module_parms(struct rcu_perf_ops *cur_ops, const char *tag)
 462 {
 463         pr_alert("%s" PERF_FLAG
 464                  "--- %s: nreaders=%d nwriters=%d verbose=%d shutdown=%d\n",
 465                  perf_type, tag, nrealreaders, nrealwriters, verbose, shutdown);
 466 }
 467
 468 static void
 469 rcu_perf_cleanup(void)
 470 {
 471         int i;
 472         int j;
 473         int ngps = 0;
 474         u64 *wdp;
 475         u64 *wdpp;
 476
 477         /*
 478          * Would like warning at start, but everything is expedited
 479          * during the mid-boot phase, so have to wait till the end.
 480          */
 481         if (rcu_gp_is_expedited() && !rcu_gp_is_normal() && !gp_exp)
 482                 VERBOSE_PERFOUT_ERRSTRING("All grace periods expedited, no normal ones to measure!");
 483         if (rcu_gp_is_normal() && gp_exp)
 484                 VERBOSE_PERFOUT_ERRSTRING("All grace periods normal, no expedited ones to measure!");
 485         if (gp_exp && gp_async)
 486                 VERBOSE_PERFOUT_ERRSTRING("No expedited async GPs, so went with async!");
 487
 488         if (torture_cleanup_begin())
 489                 return;
 490         if (!cur_ops) {
 491                 torture_cleanup_end();
 492                 return;
 493         }
 494
 495         if (reader_tasks) {
 496                 for (i = 0; i < nrealreaders; i++)
 497                         torture_stop_kthread(rcu_perf_reader,
 498                                              reader_tasks[i]);
 499                 kfree(reader_tasks);
 500         }
 501
 502         if (writer_tasks) {
 503                 for (i = 0; i < nrealwriters; i++) {
 504                         torture_stop_kthread(rcu_perf_writer,
 505                                              writer_tasks[i]);
 506                         if (!writer_n_durations)
 507                                 continue;
 508                         j = writer_n_durations[i];
 509                         pr_alert("%s%s writer %d gps: %d\n",
 510                                  perf_type, PERF_FLAG, i, j);
 511                         ngps += j;
 512                 }
 513                 pr_alert("%s%s start: %llu end: %llu duration: %llu gps: %d batches: %ld\n",
 514                          perf_type, PERF_FLAG,
 515                          t_rcu_perf_writer_started, t_rcu_perf_writer_finished,
 516                          t_rcu_perf_writer_finished -
 517                          t_rcu_perf_writer_started,
 518                          ngps,
 519                          rcuperf_seq_diff(b_rcu_gp_test_finished,
 520                                           b_rcu_gp_test_started));
 521                 for (i = 0; i < nrealwriters; i++) {
 522                         if (!writer_durations)
 523                                 break;
 524                         if (!writer_n_durations)
 525                                 continue;
 526                         wdpp = writer_durations[i];
 527                         if (!wdpp)
 528                                 continue;
 529                         for (j = 0; j <= writer_n_durations[i]; j++) {
 530                                 wdp = &wdpp[j];
 531                                 pr_alert("%s%s %4d writer-duration: %5d %llu\n",
 532                                         perf_type, PERF_FLAG,
 533                                         i, j, *wdp);
 534                                 if (j % 100 == 0)
 535                                         schedule_timeout_uninterruptible(1);
 536                         }
 537                         kfree(writer_durations[i]);
 538                 }
 539                 kfree(writer_tasks);
 540                 kfree(writer_durations);
 541                 kfree(writer_n_durations);
 542         }
 543
 544         /* Do torture-type-specific cleanup operations.  */
 545         if (cur_ops->cleanup != NULL)
 546                 cur_ops->cleanup();
 547
 548         torture_cleanup_end();
 549 }
 550
 551 /*
 552  * Return the number if non-negative.  If -1, the number of CPUs.
 553  * If less than -1, that much less than the number of CPUs, but
 554  * at least one.
 555  */
 556 static int compute_real(int n)
 557 {
 558         int nr;
 559
 560         if (n >= 0) {
 561                 nr = n;
 562         } else {
 563                 nr = num_online_cpus() + 1 + n;
 564                 if (nr <= 0)
 565                         nr = 1;
 566         }
 567         return nr;
 568 }
 569
 570 /*
 571  * RCU perf shutdown kthread.  Just waits to be awakened, then shuts
 572  * down system.
 573  */
 574 static int
 575 rcu_perf_shutdown(void *arg)
 576 {
 577         do {
 578                 wait_event(shutdown_wq,
 579                            atomic_read(&n_rcu_perf_writer_finished) >=
 580                            nrealwriters);
 581         } while (atomic_read(&n_rcu_perf_writer_finished) < nrealwriters);
 582         smp_mb(); /* Wake before output. */
 583         rcu_perf_cleanup();
 584         kernel_power_off();
 585         return -EINVAL;
 586 }
 587
 588 /*
 589  * kfree_rcu() performance tests: Start a kfree_rcu() loop on all CPUs for number
 590  * of iterations and measure total time and number of GP for all iterations to complete.
 591  */
 592
 593 torture_param(int, kfree_nthreads, -1, "Number of threads running loops of kfree_rcu().");
 594 torture_param(int, kfree_alloc_num, 8000, "Number of allocations and frees done in an iteration.");
 595 torture_param(int, kfree_loops, 10, "Number of loops doing kfree_alloc_num allocations and frees.");
 596
 597 static struct task_struct **kfree_reader_tasks;
 598 static int kfree_nrealthreads;
 599 static atomic_t n_kfree_perf_thread_started;
 600 static atomic_t n_kfree_perf_thread_ended;
 601
 602 struct kfree_obj {
 603         char kfree_obj[8];
 604         struct rcu_head rh;
 605 };
 606
 607 static int
 608 kfree_perf_thread(void *arg)
 609 {
 610         int i, loop = 0;
 611         long me = (long)arg;
 612         struct kfree_obj *alloc_ptr;
 613         u64 start_time, end_time;
 614
 615         VERBOSE_PERFOUT_STRING("kfree_perf_thread task started");
 616         set_cpus_allowed_ptr(current, cpumask_of(me % nr_cpu_ids));
 617         set_user_nice(current, MAX_NICE);
 618
 619         start_time = ktime_get_mono_fast_ns();
 620
 621         if (atomic_inc_return(&n_kfree_perf_thread_started) >= kfree_nrealthreads) {
 622                 if (gp_exp)
 623                         b_rcu_gp_test_started = cur_ops->exp_completed() / 2;
 624                 else
 625                         b_rcu_gp_test_started = cur_ops->get_gp_seq();
 626         }
 627
 628         do {
 629                 for (i = 0; i < kfree_alloc_num; i++) {
 630                         alloc_ptr = kmalloc(sizeof(struct kfree_obj), GFP_KERNEL);
 631                         if (!alloc_ptr)
 632                                 return -ENOMEM;
 633
 634                         kfree_rcu(alloc_ptr, rh);
 635                 }
 636
 637                 cond_resched();
 638         } while (!torture_must_stop() && ++loop < kfree_loops);
 639
 640         if (atomic_inc_return(&n_kfree_perf_thread_ended) >= kfree_nrealthreads) {
 641                 end_time = ktime_get_mono_fast_ns();
 642
 643                 if (gp_exp)
 644                         b_rcu_gp_test_finished = cur_ops->exp_completed() / 2;
 645                 else
 646                         b_rcu_gp_test_finished = cur_ops->get_gp_seq();
 647
 648                 pr_alert("Total time taken by all kfree'ers: %llu ns, loops: %d, batches: %ld\n",
 649                        (unsigned long long)(end_time - start_time), kfree_loops,
 650                        rcuperf_seq_diff(b_rcu_gp_test_finished, b_rcu_gp_test_started));
 651                 if (shutdown) {
 652                         smp_mb(); /* Assign before wake. */
 653                         wake_up(&shutdown_wq);
 654                 }
 655         }
 656
 657         torture_kthread_stopping("kfree_perf_thread");
 658         return 0;
 659 }
 660
 661 static void
 662 kfree_perf_cleanup(void)
 663 {
 664         int i;
 665
 666         if (torture_cleanup_begin())
 667                 return;
 668
 669         if (kfree_reader_tasks) {
 670                 for (i = 0; i < kfree_nrealthreads; i++)
 671                         torture_stop_kthread(kfree_perf_thread,
 672                                              kfree_reader_tasks[i]);
 673                 kfree(kfree_reader_tasks);
 674         }
 675
 676         torture_cleanup_end();
 677 }
 678
 679 /*
 680  * shutdown kthread.  Just waits to be awakened, then shuts down system.
 681  */
 682 static int
 683 kfree_perf_shutdown(void *arg)
 684 {
 685         do {
 686                 wait_event(shutdown_wq,
 687                            atomic_read(&n_kfree_perf_thread_ended) >=
 688                            kfree_nrealthreads);
 689         } while (atomic_read(&n_kfree_perf_thread_ended) < kfree_nrealthreads);
 690
 691         smp_mb(); /* Wake before output. */
 692
 693         kfree_perf_cleanup();
 694         kernel_power_off();
 695         return -EINVAL;
 696 }
 697
 698 static int __init
 699 kfree_perf_init(void)
 700 {
 701         long i;
 702         int firsterr = 0;
 703
 704         kfree_nrealthreads = compute_real(kfree_nthreads);
 705         /* Start up the kthreads. */
 706         if (shutdown) {
 707                 init_waitqueue_head(&shutdown_wq);
 708                 firsterr = torture_create_kthread(kfree_perf_shutdown, NULL,
 709                                                   shutdown_task);
 710                 if (firsterr)
 711                         goto unwind;
 712                 schedule_timeout_uninterruptible(1);
 713         }
 714
 715         kfree_reader_tasks = kcalloc(kfree_nrealthreads, sizeof(kfree_reader_tasks[0]),
 716                                GFP_KERNEL);
 717         if (kfree_reader_tasks == NULL) {
 718                 firsterr = -ENOMEM;
 719                 goto unwind;
 720         }
 721
 722         for (i = 0; i < kfree_nrealthreads; i++) {
 723                 firsterr = torture_create_kthread(kfree_perf_thread, (void *)i,
 724                                                   kfree_reader_tasks[i]);
 725                 if (firsterr)
 726                         goto unwind;
 727         }
 728
 729         while (atomic_read(&n_kfree_perf_thread_started) < kfree_nrealthreads)
 730                 schedule_timeout_uninterruptible(1);
 731
 732         torture_init_end();
 733         return 0;
 734
 735 unwind:
 736         torture_init_end();
 737         kfree_perf_cleanup();
 738         return firsterr;
 739 }
 740
 741 static int __init
 742 rcu_perf_init(void)
 743 {
 744         long i;
 745         int firsterr = 0;
 746         static struct rcu_perf_ops *perf_ops[] = {
 747                 &rcu_ops, &srcu_ops, &srcud_ops, &tasks_ops,
 748         };
 749
 750         if (!torture_init_begin(perf_type, verbose))
 751                 return -EBUSY;
 752
 753         /* Process args and tell the world that the perf'er is on the job. */
 754         for (i = 0; i < ARRAY_SIZE(perf_ops); i++) {
 755                 cur_ops = perf_ops[i];
 756                 if (strcmp(perf_type, cur_ops->name) == 0)
 757                         break;
 758         }
 759         if (i == ARRAY_SIZE(perf_ops)) {
 760                 pr_alert("rcu-perf: invalid perf type: \"%s\"\n", perf_type);
 761                 pr_alert("rcu-perf types:");
 762                 for (i = 0; i < ARRAY_SIZE(perf_ops); i++)
 763                         pr_cont(" %s", perf_ops[i]->name);
 764                 pr_cont("\n");
 765                 WARN_ON(!IS_MODULE(CONFIG_RCU_PERF_TEST));
 766                 firsterr = -EINVAL;
 767                 cur_ops = NULL;
 768                 goto unwind;
 769         }
 770         if (cur_ops->init)
 771                 cur_ops->init();
 772
 773         if (kfree_rcu_test)
 774                 return kfree_perf_init();
 775
 776         nrealwriters = compute_real(nwriters);
 777         nrealreaders = compute_real(nreaders);
 778         atomic_set(&n_rcu_perf_reader_started, 0);
 779         atomic_set(&n_rcu_perf_writer_started, 0);
 780         atomic_set(&n_rcu_perf_writer_finished, 0);
 781         rcu_perf_print_module_parms(cur_ops, "Start of test");
 782
 783         /* Start up the kthreads. */
 784
 785         if (shutdown) {
 786                 init_waitqueue_head(&shutdown_wq);
 787                 firsterr = torture_create_kthread(rcu_perf_shutdown, NULL,
 788                                                   shutdown_task);
 789                 if (firsterr)
 790                         goto unwind;
 791                 schedule_timeout_uninterruptible(1);
 792         }
 793         reader_tasks = kcalloc(nrealreaders, sizeof(reader_tasks[0]),
 794                                GFP_KERNEL);
 795         if (reader_tasks == NULL) {
 796                 VERBOSE_PERFOUT_ERRSTRING("out of memory");
 797                 firsterr = -ENOMEM;
 798                 goto unwind;
 799         }
 800         for (i = 0; i < nrealreaders; i++) {
 801                 firsterr = torture_create_kthread(rcu_perf_reader, (void *)i,
 802                                                   reader_tasks[i]);
 803                 if (firsterr)
 804                         goto unwind;
 805         }
 806         while (atomic_read(&n_rcu_perf_reader_started) < nrealreaders)
 807                 schedule_timeout_uninterruptible(1);
 808         writer_tasks = kcalloc(nrealwriters, sizeof(reader_tasks[0]),
 809                                GFP_KERNEL);
 810         writer_durations = kcalloc(nrealwriters, sizeof(*writer_durations),
 811                                    GFP_KERNEL);
 812         writer_n_durations =
 813                 kcalloc(nrealwriters, sizeof(*writer_n_durations),
 814                         GFP_KERNEL);
 815         if (!writer_tasks || !writer_durations || !writer_n_durations) {
 816                 VERBOSE_PERFOUT_ERRSTRING("out of memory");
 817                 firsterr = -ENOMEM;
 818                 goto unwind;
 819         }
 820         for (i = 0; i < nrealwriters; i++) {
 821                 writer_durations[i] =
 822                         kcalloc(MAX_MEAS, sizeof(*writer_durations[i]),
 823                                 GFP_KERNEL);
 824                 if (!writer_durations[i]) {
 825                         firsterr = -ENOMEM;
 826                         goto unwind;
 827                 }
 828                 firsterr = torture_create_kthread(rcu_perf_writer, (void *)i,
 829                                                   writer_tasks[i]);
 830                 if (firsterr)
 831                         goto unwind;
 832         }
 833         torture_init_end();
 834         return 0;
 835
 836 unwind:
 837         torture_init_end();
 838         rcu_perf_cleanup();
 839         return firsterr;
 840 }
 841
 842 module_init(rcu_perf_init);
 843 module_exit(rcu_perf_cleanup);