thp: close race between split and zap huge pages
[linux.git] / kernel / rcu / rcutorture.c
1 /*
2  * Read-Copy Update module-based torture test facility
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License as published by
6  * the Free Software Foundation; either version 2 of the License, or
7  * (at your option) any later version.
8  *
9  * This program is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, you can access it online at
16  * http://www.gnu.org/licenses/gpl-2.0.html.
17  *
18  * Copyright (C) IBM Corporation, 2005, 2006
19  *
20  * Authors: Paul E. McKenney <paulmck@us.ibm.com>
21  *        Josh Triplett <josh@freedesktop.org>
22  *
23  * See also:  Documentation/RCU/torture.txt
24  */
25 #include <linux/types.h>
26 #include <linux/kernel.h>
27 #include <linux/init.h>
28 #include <linux/module.h>
29 #include <linux/kthread.h>
30 #include <linux/err.h>
31 #include <linux/spinlock.h>
32 #include <linux/smp.h>
33 #include <linux/rcupdate.h>
34 #include <linux/interrupt.h>
35 #include <linux/sched.h>
36 #include <linux/atomic.h>
37 #include <linux/bitops.h>
38 #include <linux/completion.h>
39 #include <linux/moduleparam.h>
40 #include <linux/percpu.h>
41 #include <linux/notifier.h>
42 #include <linux/reboot.h>
43 #include <linux/freezer.h>
44 #include <linux/cpu.h>
45 #include <linux/delay.h>
46 #include <linux/stat.h>
47 #include <linux/srcu.h>
48 #include <linux/slab.h>
49 #include <linux/trace_clock.h>
50 #include <asm/byteorder.h>
51 #include <linux/torture.h>
52
53 MODULE_LICENSE("GPL");
54 MODULE_AUTHOR("Paul E. McKenney <paulmck@us.ibm.com> and Josh Triplett <josh@freedesktop.org>");
55
56
57 torture_param(int, fqs_duration, 0,
58               "Duration of fqs bursts (us), 0 to disable");
59 torture_param(int, fqs_holdoff, 0, "Holdoff time within fqs bursts (us)");
60 torture_param(int, fqs_stutter, 3, "Wait time between fqs bursts (s)");
61 torture_param(bool, gp_exp, false, "Use expedited GP wait primitives");
62 torture_param(bool, gp_normal, false,
63              "Use normal (non-expedited) GP wait primitives");
64 torture_param(int, irqreader, 1, "Allow RCU readers from irq handlers");
65 torture_param(int, n_barrier_cbs, 0,
66              "# of callbacks/kthreads for barrier testing");
67 torture_param(int, nfakewriters, 4, "Number of RCU fake writer threads");
68 torture_param(int, nreaders, -1, "Number of RCU reader threads");
69 torture_param(int, object_debug, 0,
70              "Enable debug-object double call_rcu() testing");
71 torture_param(int, onoff_holdoff, 0, "Time after boot before CPU hotplugs (s)");
72 torture_param(int, onoff_interval, 0,
73              "Time between CPU hotplugs (s), 0=disable");
74 torture_param(int, shuffle_interval, 3, "Number of seconds between shuffles");
75 torture_param(int, shutdown_secs, 0, "Shutdown time (s), <= zero to disable.");
76 torture_param(int, stall_cpu, 0, "Stall duration (s), zero to disable.");
77 torture_param(int, stall_cpu_holdoff, 10,
78              "Time to wait before starting stall (s).");
79 torture_param(int, stat_interval, 60,
80              "Number of seconds between stats printk()s");
81 torture_param(int, stutter, 5, "Number of seconds to run/halt test");
82 torture_param(int, test_boost, 1, "Test RCU prio boost: 0=no, 1=maybe, 2=yes.");
83 torture_param(int, test_boost_duration, 4,
84              "Duration of each boost test, seconds.");
85 torture_param(int, test_boost_interval, 7,
86              "Interval between boost tests, seconds.");
87 torture_param(bool, test_no_idle_hz, true,
88              "Test support for tickless idle CPUs");
89 torture_param(bool, verbose, true,
90              "Enable verbose debugging printk()s");
91
92 static char *torture_type = "rcu";
93 module_param(torture_type, charp, 0444);
94 MODULE_PARM_DESC(torture_type, "Type of RCU to torture (rcu, rcu_bh, ...)");
95
96 static int nrealreaders;
97 static struct task_struct *writer_task;
98 static struct task_struct **fakewriter_tasks;
99 static struct task_struct **reader_tasks;
100 static struct task_struct *stats_task;
101 static struct task_struct *fqs_task;
102 static struct task_struct *boost_tasks[NR_CPUS];
103 static struct task_struct *stall_task;
104 static struct task_struct **barrier_cbs_tasks;
105 static struct task_struct *barrier_task;
106
107 #define RCU_TORTURE_PIPE_LEN 10
108
109 struct rcu_torture {
110         struct rcu_head rtort_rcu;
111         int rtort_pipe_count;
112         struct list_head rtort_free;
113         int rtort_mbtest;
114 };
115
116 static LIST_HEAD(rcu_torture_freelist);
117 static struct rcu_torture __rcu *rcu_torture_current;
118 static unsigned long rcu_torture_current_version;
119 static struct rcu_torture rcu_tortures[10 * RCU_TORTURE_PIPE_LEN];
120 static DEFINE_SPINLOCK(rcu_torture_lock);
121 static DEFINE_PER_CPU(long [RCU_TORTURE_PIPE_LEN + 1],
122                       rcu_torture_count) = { 0 };
123 static DEFINE_PER_CPU(long [RCU_TORTURE_PIPE_LEN + 1],
124                       rcu_torture_batch) = { 0 };
125 static atomic_t rcu_torture_wcount[RCU_TORTURE_PIPE_LEN + 1];
126 static atomic_t n_rcu_torture_alloc;
127 static atomic_t n_rcu_torture_alloc_fail;
128 static atomic_t n_rcu_torture_free;
129 static atomic_t n_rcu_torture_mberror;
130 static atomic_t n_rcu_torture_error;
131 static long n_rcu_torture_barrier_error;
132 static long n_rcu_torture_boost_ktrerror;
133 static long n_rcu_torture_boost_rterror;
134 static long n_rcu_torture_boost_failure;
135 static long n_rcu_torture_boosts;
136 static long n_rcu_torture_timers;
137 static long n_barrier_attempts;
138 static long n_barrier_successes;
139 static struct list_head rcu_torture_removed;
140
141 #if defined(MODULE) || defined(CONFIG_RCU_TORTURE_TEST_RUNNABLE)
142 #define RCUTORTURE_RUNNABLE_INIT 1
143 #else
144 #define RCUTORTURE_RUNNABLE_INIT 0
145 #endif
146 int rcutorture_runnable = RCUTORTURE_RUNNABLE_INIT;
147 module_param(rcutorture_runnable, int, 0444);
148 MODULE_PARM_DESC(rcutorture_runnable, "Start rcutorture at boot");
149
150 #if defined(CONFIG_RCU_BOOST) && !defined(CONFIG_HOTPLUG_CPU)
151 #define rcu_can_boost() 1
152 #else /* #if defined(CONFIG_RCU_BOOST) && !defined(CONFIG_HOTPLUG_CPU) */
153 #define rcu_can_boost() 0
154 #endif /* #else #if defined(CONFIG_RCU_BOOST) && !defined(CONFIG_HOTPLUG_CPU) */
155
156 #ifdef CONFIG_RCU_TRACE
157 static u64 notrace rcu_trace_clock_local(void)
158 {
159         u64 ts = trace_clock_local();
160         unsigned long __maybe_unused ts_rem = do_div(ts, NSEC_PER_USEC);
161         return ts;
162 }
163 #else /* #ifdef CONFIG_RCU_TRACE */
164 static u64 notrace rcu_trace_clock_local(void)
165 {
166         return 0ULL;
167 }
168 #endif /* #else #ifdef CONFIG_RCU_TRACE */
169
170 static unsigned long boost_starttime;   /* jiffies of next boost test start. */
171 DEFINE_MUTEX(boost_mutex);              /* protect setting boost_starttime */
172                                         /*  and boost task create/destroy. */
173 static atomic_t barrier_cbs_count;      /* Barrier callbacks registered. */
174 static bool barrier_phase;              /* Test phase. */
175 static atomic_t barrier_cbs_invoked;    /* Barrier callbacks invoked. */
176 static wait_queue_head_t *barrier_cbs_wq; /* Coordinate barrier testing. */
177 static DECLARE_WAIT_QUEUE_HEAD(barrier_wq);
178
179 /*
180  * Allocate an element from the rcu_tortures pool.
181  */
182 static struct rcu_torture *
183 rcu_torture_alloc(void)
184 {
185         struct list_head *p;
186
187         spin_lock_bh(&rcu_torture_lock);
188         if (list_empty(&rcu_torture_freelist)) {
189                 atomic_inc(&n_rcu_torture_alloc_fail);
190                 spin_unlock_bh(&rcu_torture_lock);
191                 return NULL;
192         }
193         atomic_inc(&n_rcu_torture_alloc);
194         p = rcu_torture_freelist.next;
195         list_del_init(p);
196         spin_unlock_bh(&rcu_torture_lock);
197         return container_of(p, struct rcu_torture, rtort_free);
198 }
199
200 /*
201  * Free an element to the rcu_tortures pool.
202  */
203 static void
204 rcu_torture_free(struct rcu_torture *p)
205 {
206         atomic_inc(&n_rcu_torture_free);
207         spin_lock_bh(&rcu_torture_lock);
208         list_add_tail(&p->rtort_free, &rcu_torture_freelist);
209         spin_unlock_bh(&rcu_torture_lock);
210 }
211
212 /*
213  * Operations vector for selecting different types of tests.
214  */
215
216 struct rcu_torture_ops {
217         void (*init)(void);
218         int (*readlock)(void);
219         void (*read_delay)(struct torture_random_state *rrsp);
220         void (*readunlock)(int idx);
221         int (*completed)(void);
222         void (*deferred_free)(struct rcu_torture *p);
223         void (*sync)(void);
224         void (*exp_sync)(void);
225         void (*call)(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
226         void (*cb_barrier)(void);
227         void (*fqs)(void);
228         void (*stats)(char *page);
229         int irq_capable;
230         int can_boost;
231         const char *name;
232 };
233
234 static struct rcu_torture_ops *cur_ops;
235
236 /*
237  * Definitions for rcu torture testing.
238  */
239
240 static int rcu_torture_read_lock(void) __acquires(RCU)
241 {
242         rcu_read_lock();
243         return 0;
244 }
245
246 static void rcu_read_delay(struct torture_random_state *rrsp)
247 {
248         const unsigned long shortdelay_us = 200;
249         const unsigned long longdelay_ms = 50;
250
251         /* We want a short delay sometimes to make a reader delay the grace
252          * period, and we want a long delay occasionally to trigger
253          * force_quiescent_state. */
254
255         if (!(torture_random(rrsp) % (nrealreaders * 2000 * longdelay_ms)))
256                 mdelay(longdelay_ms);
257         if (!(torture_random(rrsp) % (nrealreaders * 2 * shortdelay_us)))
258                 udelay(shortdelay_us);
259 #ifdef CONFIG_PREEMPT
260         if (!preempt_count() &&
261             !(torture_random(rrsp) % (nrealreaders * 20000)))
262                 preempt_schedule();  /* No QS if preempt_disable() in effect */
263 #endif
264 }
265
266 static void rcu_torture_read_unlock(int idx) __releases(RCU)
267 {
268         rcu_read_unlock();
269 }
270
271 static int rcu_torture_completed(void)
272 {
273         return rcu_batches_completed();
274 }
275
276 static void
277 rcu_torture_cb(struct rcu_head *p)
278 {
279         int i;
280         struct rcu_torture *rp = container_of(p, struct rcu_torture, rtort_rcu);
281
282         if (torture_must_stop_irq()) {
283                 /* Test is ending, just drop callbacks on the floor. */
284                 /* The next initialization will pick up the pieces. */
285                 return;
286         }
287         i = rp->rtort_pipe_count;
288         if (i > RCU_TORTURE_PIPE_LEN)
289                 i = RCU_TORTURE_PIPE_LEN;
290         atomic_inc(&rcu_torture_wcount[i]);
291         if (++rp->rtort_pipe_count >= RCU_TORTURE_PIPE_LEN) {
292                 rp->rtort_mbtest = 0;
293                 rcu_torture_free(rp);
294         } else {
295                 cur_ops->deferred_free(rp);
296         }
297 }
298
299 static int rcu_no_completed(void)
300 {
301         return 0;
302 }
303
304 static void rcu_torture_deferred_free(struct rcu_torture *p)
305 {
306         call_rcu(&p->rtort_rcu, rcu_torture_cb);
307 }
308
309 static void rcu_sync_torture_init(void)
310 {
311         INIT_LIST_HEAD(&rcu_torture_removed);
312 }
313
314 static struct rcu_torture_ops rcu_ops = {
315         .init           = rcu_sync_torture_init,
316         .readlock       = rcu_torture_read_lock,
317         .read_delay     = rcu_read_delay,
318         .readunlock     = rcu_torture_read_unlock,
319         .completed      = rcu_torture_completed,
320         .deferred_free  = rcu_torture_deferred_free,
321         .sync           = synchronize_rcu,
322         .exp_sync       = synchronize_rcu_expedited,
323         .call           = call_rcu,
324         .cb_barrier     = rcu_barrier,
325         .fqs            = rcu_force_quiescent_state,
326         .stats          = NULL,
327         .irq_capable    = 1,
328         .can_boost      = rcu_can_boost(),
329         .name           = "rcu"
330 };
331
332 /*
333  * Definitions for rcu_bh torture testing.
334  */
335
336 static int rcu_bh_torture_read_lock(void) __acquires(RCU_BH)
337 {
338         rcu_read_lock_bh();
339         return 0;
340 }
341
342 static void rcu_bh_torture_read_unlock(int idx) __releases(RCU_BH)
343 {
344         rcu_read_unlock_bh();
345 }
346
347 static int rcu_bh_torture_completed(void)
348 {
349         return rcu_batches_completed_bh();
350 }
351
352 static void rcu_bh_torture_deferred_free(struct rcu_torture *p)
353 {
354         call_rcu_bh(&p->rtort_rcu, rcu_torture_cb);
355 }
356
357 static struct rcu_torture_ops rcu_bh_ops = {
358         .init           = rcu_sync_torture_init,
359         .readlock       = rcu_bh_torture_read_lock,
360         .read_delay     = rcu_read_delay,  /* just reuse rcu's version. */
361         .readunlock     = rcu_bh_torture_read_unlock,
362         .completed      = rcu_bh_torture_completed,
363         .deferred_free  = rcu_bh_torture_deferred_free,
364         .sync           = synchronize_rcu_bh,
365         .exp_sync       = synchronize_rcu_bh_expedited,
366         .call           = call_rcu_bh,
367         .cb_barrier     = rcu_barrier_bh,
368         .fqs            = rcu_bh_force_quiescent_state,
369         .stats          = NULL,
370         .irq_capable    = 1,
371         .name           = "rcu_bh"
372 };
373
374 /*
375  * Don't even think about trying any of these in real life!!!
376  * The names includes "busted", and they really means it!
377  * The only purpose of these functions is to provide a buggy RCU
378  * implementation to make sure that rcutorture correctly emits
379  * buggy-RCU error messages.
380  */
381 static void rcu_busted_torture_deferred_free(struct rcu_torture *p)
382 {
383         /* This is a deliberate bug for testing purposes only! */
384         rcu_torture_cb(&p->rtort_rcu);
385 }
386
387 static void synchronize_rcu_busted(void)
388 {
389         /* This is a deliberate bug for testing purposes only! */
390 }
391
392 static void
393 call_rcu_busted(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
394 {
395         /* This is a deliberate bug for testing purposes only! */
396         func(head);
397 }
398
399 static struct rcu_torture_ops rcu_busted_ops = {
400         .init           = rcu_sync_torture_init,
401         .readlock       = rcu_torture_read_lock,
402         .read_delay     = rcu_read_delay,  /* just reuse rcu's version. */
403         .readunlock     = rcu_torture_read_unlock,
404         .completed      = rcu_no_completed,
405         .deferred_free  = rcu_busted_torture_deferred_free,
406         .sync           = synchronize_rcu_busted,
407         .exp_sync       = synchronize_rcu_busted,
408         .call           = call_rcu_busted,
409         .cb_barrier     = NULL,
410         .fqs            = NULL,
411         .stats          = NULL,
412         .irq_capable    = 1,
413         .name           = "rcu_busted"
414 };
415
416 /*
417  * Definitions for srcu torture testing.
418  */
419
420 DEFINE_STATIC_SRCU(srcu_ctl);
421
422 static int srcu_torture_read_lock(void) __acquires(&srcu_ctl)
423 {
424         return srcu_read_lock(&srcu_ctl);
425 }
426
427 static void srcu_read_delay(struct torture_random_state *rrsp)
428 {
429         long delay;
430         const long uspertick = 1000000 / HZ;
431         const long longdelay = 10;
432
433         /* We want there to be long-running readers, but not all the time. */
434
435         delay = torture_random(rrsp) %
436                 (nrealreaders * 2 * longdelay * uspertick);
437         if (!delay)
438                 schedule_timeout_interruptible(longdelay);
439         else
440                 rcu_read_delay(rrsp);
441 }
442
443 static void srcu_torture_read_unlock(int idx) __releases(&srcu_ctl)
444 {
445         srcu_read_unlock(&srcu_ctl, idx);
446 }
447
448 static int srcu_torture_completed(void)
449 {
450         return srcu_batches_completed(&srcu_ctl);
451 }
452
453 static void srcu_torture_deferred_free(struct rcu_torture *rp)
454 {
455         call_srcu(&srcu_ctl, &rp->rtort_rcu, rcu_torture_cb);
456 }
457
458 static void srcu_torture_synchronize(void)
459 {
460         synchronize_srcu(&srcu_ctl);
461 }
462
463 static void srcu_torture_call(struct rcu_head *head,
464                               void (*func)(struct rcu_head *head))
465 {
466         call_srcu(&srcu_ctl, head, func);
467 }
468
469 static void srcu_torture_barrier(void)
470 {
471         srcu_barrier(&srcu_ctl);
472 }
473
474 static void srcu_torture_stats(char *page)
475 {
476         int cpu;
477         int idx = srcu_ctl.completed & 0x1;
478
479         page += sprintf(page, "%s%s per-CPU(idx=%d):",
480                        torture_type, TORTURE_FLAG, idx);
481         for_each_possible_cpu(cpu) {
482                 page += sprintf(page, " %d(%lu,%lu)", cpu,
483                                per_cpu_ptr(srcu_ctl.per_cpu_ref, cpu)->c[!idx],
484                                per_cpu_ptr(srcu_ctl.per_cpu_ref, cpu)->c[idx]);
485         }
486         sprintf(page, "\n");
487 }
488
489 static void srcu_torture_synchronize_expedited(void)
490 {
491         synchronize_srcu_expedited(&srcu_ctl);
492 }
493
494 static struct rcu_torture_ops srcu_ops = {
495         .init           = rcu_sync_torture_init,
496         .readlock       = srcu_torture_read_lock,
497         .read_delay     = srcu_read_delay,
498         .readunlock     = srcu_torture_read_unlock,
499         .completed      = srcu_torture_completed,
500         .deferred_free  = srcu_torture_deferred_free,
501         .sync           = srcu_torture_synchronize,
502         .exp_sync       = srcu_torture_synchronize_expedited,
503         .call           = srcu_torture_call,
504         .cb_barrier     = srcu_torture_barrier,
505         .stats          = srcu_torture_stats,
506         .name           = "srcu"
507 };
508
509 /*
510  * Definitions for sched torture testing.
511  */
512
513 static int sched_torture_read_lock(void)
514 {
515         preempt_disable();
516         return 0;
517 }
518
519 static void sched_torture_read_unlock(int idx)
520 {
521         preempt_enable();
522 }
523
524 static void rcu_sched_torture_deferred_free(struct rcu_torture *p)
525 {
526         call_rcu_sched(&p->rtort_rcu, rcu_torture_cb);
527 }
528
529 static struct rcu_torture_ops sched_ops = {
530         .init           = rcu_sync_torture_init,
531         .readlock       = sched_torture_read_lock,
532         .read_delay     = rcu_read_delay,  /* just reuse rcu's version. */
533         .readunlock     = sched_torture_read_unlock,
534         .completed      = rcu_no_completed,
535         .deferred_free  = rcu_sched_torture_deferred_free,
536         .sync           = synchronize_sched,
537         .exp_sync       = synchronize_sched_expedited,
538         .call           = call_rcu_sched,
539         .cb_barrier     = rcu_barrier_sched,
540         .fqs            = rcu_sched_force_quiescent_state,
541         .stats          = NULL,
542         .irq_capable    = 1,
543         .name           = "sched"
544 };
545
546 /*
547  * RCU torture priority-boost testing.  Runs one real-time thread per
548  * CPU for moderate bursts, repeatedly registering RCU callbacks and
549  * spinning waiting for them to be invoked.  If a given callback takes
550  * too long to be invoked, we assume that priority inversion has occurred.
551  */
552
553 struct rcu_boost_inflight {
554         struct rcu_head rcu;
555         int inflight;
556 };
557
558 static void rcu_torture_boost_cb(struct rcu_head *head)
559 {
560         struct rcu_boost_inflight *rbip =
561                 container_of(head, struct rcu_boost_inflight, rcu);
562
563         smp_mb(); /* Ensure RCU-core accesses precede clearing ->inflight */
564         rbip->inflight = 0;
565 }
566
567 static int rcu_torture_boost(void *arg)
568 {
569         unsigned long call_rcu_time;
570         unsigned long endtime;
571         unsigned long oldstarttime;
572         struct rcu_boost_inflight rbi = { .inflight = 0 };
573         struct sched_param sp;
574
575         VERBOSE_TOROUT_STRING("rcu_torture_boost started");
576
577         /* Set real-time priority. */
578         sp.sched_priority = 1;
579         if (sched_setscheduler(current, SCHED_FIFO, &sp) < 0) {
580                 VERBOSE_TOROUT_STRING("rcu_torture_boost RT prio failed!");
581                 n_rcu_torture_boost_rterror++;
582         }
583
584         init_rcu_head_on_stack(&rbi.rcu);
585         /* Each pass through the following loop does one boost-test cycle. */
586         do {
587                 /* Wait for the next test interval. */
588                 oldstarttime = boost_starttime;
589                 while (ULONG_CMP_LT(jiffies, oldstarttime)) {
590                         schedule_timeout_interruptible(oldstarttime - jiffies);
591                         stutter_wait("rcu_torture_boost");
592                         if (torture_must_stop())
593                                 goto checkwait;
594                 }
595
596                 /* Do one boost-test interval. */
597                 endtime = oldstarttime + test_boost_duration * HZ;
598                 call_rcu_time = jiffies;
599                 while (ULONG_CMP_LT(jiffies, endtime)) {
600                         /* If we don't have a callback in flight, post one. */
601                         if (!rbi.inflight) {
602                                 smp_mb(); /* RCU core before ->inflight = 1. */
603                                 rbi.inflight = 1;
604                                 call_rcu(&rbi.rcu, rcu_torture_boost_cb);
605                                 if (jiffies - call_rcu_time >
606                                          test_boost_duration * HZ - HZ / 2) {
607                                         VERBOSE_TOROUT_STRING("rcu_torture_boost boosting failed");
608                                         n_rcu_torture_boost_failure++;
609                                 }
610                                 call_rcu_time = jiffies;
611                         }
612                         cond_resched();
613                         stutter_wait("rcu_torture_boost");
614                         if (torture_must_stop())
615                                 goto checkwait;
616                 }
617
618                 /*
619                  * Set the start time of the next test interval.
620                  * Yes, this is vulnerable to long delays, but such
621                  * delays simply cause a false negative for the next
622                  * interval.  Besides, we are running at RT priority,
623                  * so delays should be relatively rare.
624                  */
625                 while (oldstarttime == boost_starttime &&
626                        !kthread_should_stop()) {
627                         if (mutex_trylock(&boost_mutex)) {
628                                 boost_starttime = jiffies +
629                                                   test_boost_interval * HZ;
630                                 n_rcu_torture_boosts++;
631                                 mutex_unlock(&boost_mutex);
632                                 break;
633                         }
634                         schedule_timeout_uninterruptible(1);
635                 }
636
637                 /* Go do the stutter. */
638 checkwait:      stutter_wait("rcu_torture_boost");
639         } while (!torture_must_stop());
640
641         /* Clean up and exit. */
642         while (!kthread_should_stop() || rbi.inflight) {
643                 torture_shutdown_absorb("rcu_torture_boost");
644                 schedule_timeout_uninterruptible(1);
645         }
646         smp_mb(); /* order accesses to ->inflight before stack-frame death. */
647         destroy_rcu_head_on_stack(&rbi.rcu);
648         torture_kthread_stopping("rcu_torture_boost");
649         return 0;
650 }
651
652 /*
653  * RCU torture force-quiescent-state kthread.  Repeatedly induces
654  * bursts of calls to force_quiescent_state(), increasing the probability
655  * of occurrence of some important types of race conditions.
656  */
657 static int
658 rcu_torture_fqs(void *arg)
659 {
660         unsigned long fqs_resume_time;
661         int fqs_burst_remaining;
662
663         VERBOSE_TOROUT_STRING("rcu_torture_fqs task started");
664         do {
665                 fqs_resume_time = jiffies + fqs_stutter * HZ;
666                 while (ULONG_CMP_LT(jiffies, fqs_resume_time) &&
667                        !kthread_should_stop()) {
668                         schedule_timeout_interruptible(1);
669                 }
670                 fqs_burst_remaining = fqs_duration;
671                 while (fqs_burst_remaining > 0 &&
672                        !kthread_should_stop()) {
673                         cur_ops->fqs();
674                         udelay(fqs_holdoff);
675                         fqs_burst_remaining -= fqs_holdoff;
676                 }
677                 stutter_wait("rcu_torture_fqs");
678         } while (!torture_must_stop());
679         torture_kthread_stopping("rcu_torture_fqs");
680         return 0;
681 }
682
683 /*
684  * RCU torture writer kthread.  Repeatedly substitutes a new structure
685  * for that pointed to by rcu_torture_current, freeing the old structure
686  * after a series of grace periods (the "pipeline").
687  */
688 static int
689 rcu_torture_writer(void *arg)
690 {
691         bool exp;
692         int i;
693         struct rcu_torture *rp;
694         struct rcu_torture *rp1;
695         struct rcu_torture *old_rp;
696         static DEFINE_TORTURE_RANDOM(rand);
697
698         VERBOSE_TOROUT_STRING("rcu_torture_writer task started");
699         set_user_nice(current, MAX_NICE);
700
701         do {
702                 schedule_timeout_uninterruptible(1);
703                 rp = rcu_torture_alloc();
704                 if (rp == NULL)
705                         continue;
706                 rp->rtort_pipe_count = 0;
707                 udelay(torture_random(&rand) & 0x3ff);
708                 old_rp = rcu_dereference_check(rcu_torture_current,
709                                                current == writer_task);
710                 rp->rtort_mbtest = 1;
711                 rcu_assign_pointer(rcu_torture_current, rp);
712                 smp_wmb(); /* Mods to old_rp must follow rcu_assign_pointer() */
713                 if (old_rp) {
714                         i = old_rp->rtort_pipe_count;
715                         if (i > RCU_TORTURE_PIPE_LEN)
716                                 i = RCU_TORTURE_PIPE_LEN;
717                         atomic_inc(&rcu_torture_wcount[i]);
718                         old_rp->rtort_pipe_count++;
719                         if (gp_normal == gp_exp)
720                                 exp = !!(torture_random(&rand) & 0x80);
721                         else
722                                 exp = gp_exp;
723                         if (!exp) {
724                                 cur_ops->deferred_free(old_rp);
725                         } else {
726                                 cur_ops->exp_sync();
727                                 list_add(&old_rp->rtort_free,
728                                          &rcu_torture_removed);
729                                 list_for_each_entry_safe(rp, rp1,
730                                                          &rcu_torture_removed,
731                                                          rtort_free) {
732                                         i = rp->rtort_pipe_count;
733                                         if (i > RCU_TORTURE_PIPE_LEN)
734                                                 i = RCU_TORTURE_PIPE_LEN;
735                                         atomic_inc(&rcu_torture_wcount[i]);
736                                         if (++rp->rtort_pipe_count >=
737                                             RCU_TORTURE_PIPE_LEN) {
738                                                 rp->rtort_mbtest = 0;
739                                                 list_del(&rp->rtort_free);
740                                                 rcu_torture_free(rp);
741                                         }
742                                  }
743                         }
744                 }
745                 rcutorture_record_progress(++rcu_torture_current_version);
746                 stutter_wait("rcu_torture_writer");
747         } while (!torture_must_stop());
748         torture_kthread_stopping("rcu_torture_writer");
749         return 0;
750 }
751
752 /*
753  * RCU torture fake writer kthread.  Repeatedly calls sync, with a random
754  * delay between calls.
755  */
756 static int
757 rcu_torture_fakewriter(void *arg)
758 {
759         DEFINE_TORTURE_RANDOM(rand);
760
761         VERBOSE_TOROUT_STRING("rcu_torture_fakewriter task started");
762         set_user_nice(current, MAX_NICE);
763
764         do {
765                 schedule_timeout_uninterruptible(1 + torture_random(&rand)%10);
766                 udelay(torture_random(&rand) & 0x3ff);
767                 if (cur_ops->cb_barrier != NULL &&
768                     torture_random(&rand) % (nfakewriters * 8) == 0) {
769                         cur_ops->cb_barrier();
770                 } else if (gp_normal == gp_exp) {
771                         if (torture_random(&rand) & 0x80)
772                                 cur_ops->sync();
773                         else
774                                 cur_ops->exp_sync();
775                 } else if (gp_normal) {
776                         cur_ops->sync();
777                 } else {
778                         cur_ops->exp_sync();
779                 }
780                 stutter_wait("rcu_torture_fakewriter");
781         } while (!torture_must_stop());
782
783         torture_kthread_stopping("rcu_torture_fakewriter");
784         return 0;
785 }
786
787 void rcutorture_trace_dump(void)
788 {
789         static atomic_t beenhere = ATOMIC_INIT(0);
790
791         if (atomic_read(&beenhere))
792                 return;
793         if (atomic_xchg(&beenhere, 1) != 0)
794                 return;
795         ftrace_dump(DUMP_ALL);
796 }
797
798 /*
799  * RCU torture reader from timer handler.  Dereferences rcu_torture_current,
800  * incrementing the corresponding element of the pipeline array.  The
801  * counter in the element should never be greater than 1, otherwise, the
802  * RCU implementation is broken.
803  */
804 static void rcu_torture_timer(unsigned long unused)
805 {
806         int idx;
807         int completed;
808         int completed_end;
809         static DEFINE_TORTURE_RANDOM(rand);
810         static DEFINE_SPINLOCK(rand_lock);
811         struct rcu_torture *p;
812         int pipe_count;
813         unsigned long long ts;
814
815         idx = cur_ops->readlock();
816         completed = cur_ops->completed();
817         ts = rcu_trace_clock_local();
818         p = rcu_dereference_check(rcu_torture_current,
819                                   rcu_read_lock_bh_held() ||
820                                   rcu_read_lock_sched_held() ||
821                                   srcu_read_lock_held(&srcu_ctl));
822         if (p == NULL) {
823                 /* Leave because rcu_torture_writer is not yet underway */
824                 cur_ops->readunlock(idx);
825                 return;
826         }
827         if (p->rtort_mbtest == 0)
828                 atomic_inc(&n_rcu_torture_mberror);
829         spin_lock(&rand_lock);
830         cur_ops->read_delay(&rand);
831         n_rcu_torture_timers++;
832         spin_unlock(&rand_lock);
833         preempt_disable();
834         pipe_count = p->rtort_pipe_count;
835         if (pipe_count > RCU_TORTURE_PIPE_LEN) {
836                 /* Should not happen, but... */
837                 pipe_count = RCU_TORTURE_PIPE_LEN;
838         }
839         completed_end = cur_ops->completed();
840         if (pipe_count > 1) {
841                 do_trace_rcu_torture_read(cur_ops->name, &p->rtort_rcu, ts,
842                                           completed, completed_end);
843                 rcutorture_trace_dump();
844         }
845         __this_cpu_inc(rcu_torture_count[pipe_count]);
846         completed = completed_end - completed;
847         if (completed > RCU_TORTURE_PIPE_LEN) {
848                 /* Should not happen, but... */
849                 completed = RCU_TORTURE_PIPE_LEN;
850         }
851         __this_cpu_inc(rcu_torture_batch[completed]);
852         preempt_enable();
853         cur_ops->readunlock(idx);
854 }
855
856 /*
857  * RCU torture reader kthread.  Repeatedly dereferences rcu_torture_current,
858  * incrementing the corresponding element of the pipeline array.  The
859  * counter in the element should never be greater than 1, otherwise, the
860  * RCU implementation is broken.
861  */
862 static int
863 rcu_torture_reader(void *arg)
864 {
865         int completed;
866         int completed_end;
867         int idx;
868         DEFINE_TORTURE_RANDOM(rand);
869         struct rcu_torture *p;
870         int pipe_count;
871         struct timer_list t;
872         unsigned long long ts;
873
874         VERBOSE_TOROUT_STRING("rcu_torture_reader task started");
875         set_user_nice(current, MAX_NICE);
876         if (irqreader && cur_ops->irq_capable)
877                 setup_timer_on_stack(&t, rcu_torture_timer, 0);
878
879         do {
880                 if (irqreader && cur_ops->irq_capable) {
881                         if (!timer_pending(&t))
882                                 mod_timer(&t, jiffies + 1);
883                 }
884                 idx = cur_ops->readlock();
885                 completed = cur_ops->completed();
886                 ts = rcu_trace_clock_local();
887                 p = rcu_dereference_check(rcu_torture_current,
888                                           rcu_read_lock_bh_held() ||
889                                           rcu_read_lock_sched_held() ||
890                                           srcu_read_lock_held(&srcu_ctl));
891                 if (p == NULL) {
892                         /* Wait for rcu_torture_writer to get underway */
893                         cur_ops->readunlock(idx);
894                         schedule_timeout_interruptible(HZ);
895                         continue;
896                 }
897                 if (p->rtort_mbtest == 0)
898                         atomic_inc(&n_rcu_torture_mberror);
899                 cur_ops->read_delay(&rand);
900                 preempt_disable();
901                 pipe_count = p->rtort_pipe_count;
902                 if (pipe_count > RCU_TORTURE_PIPE_LEN) {
903                         /* Should not happen, but... */
904                         pipe_count = RCU_TORTURE_PIPE_LEN;
905                 }
906                 completed_end = cur_ops->completed();
907                 if (pipe_count > 1) {
908                         do_trace_rcu_torture_read(cur_ops->name, &p->rtort_rcu,
909                                                   ts, completed, completed_end);
910                         rcutorture_trace_dump();
911                 }
912                 __this_cpu_inc(rcu_torture_count[pipe_count]);
913                 completed = completed_end - completed;
914                 if (completed > RCU_TORTURE_PIPE_LEN) {
915                         /* Should not happen, but... */
916                         completed = RCU_TORTURE_PIPE_LEN;
917                 }
918                 __this_cpu_inc(rcu_torture_batch[completed]);
919                 preempt_enable();
920                 cur_ops->readunlock(idx);
921                 schedule();
922                 stutter_wait("rcu_torture_reader");
923         } while (!torture_must_stop());
924         if (irqreader && cur_ops->irq_capable)
925                 del_timer_sync(&t);
926         torture_kthread_stopping("rcu_torture_reader");
927         return 0;
928 }
929
930 /*
931  * Create an RCU-torture statistics message in the specified buffer.
932  */
933 static void
934 rcu_torture_printk(char *page)
935 {
936         int cpu;
937         int i;
938         long pipesummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
939         long batchsummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
940
941         for_each_possible_cpu(cpu) {
942                 for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
943                         pipesummary[i] += per_cpu(rcu_torture_count, cpu)[i];
944                         batchsummary[i] += per_cpu(rcu_torture_batch, cpu)[i];
945                 }
946         }
947         for (i = RCU_TORTURE_PIPE_LEN - 1; i >= 0; i--) {
948                 if (pipesummary[i] != 0)
949                         break;
950         }
951         page += sprintf(page, "%s%s ", torture_type, TORTURE_FLAG);
952         page += sprintf(page,
953                        "rtc: %p ver: %lu tfle: %d rta: %d rtaf: %d rtf: %d ",
954                        rcu_torture_current,
955                        rcu_torture_current_version,
956                        list_empty(&rcu_torture_freelist),
957                        atomic_read(&n_rcu_torture_alloc),
958                        atomic_read(&n_rcu_torture_alloc_fail),
959                        atomic_read(&n_rcu_torture_free));
960         page += sprintf(page, "rtmbe: %d rtbke: %ld rtbre: %ld ",
961                        atomic_read(&n_rcu_torture_mberror),
962                        n_rcu_torture_boost_ktrerror,
963                        n_rcu_torture_boost_rterror);
964         page += sprintf(page, "rtbf: %ld rtb: %ld nt: %ld ",
965                        n_rcu_torture_boost_failure,
966                        n_rcu_torture_boosts,
967                        n_rcu_torture_timers);
968         page = torture_onoff_stats(page);
969         page += sprintf(page, "barrier: %ld/%ld:%ld",
970                        n_barrier_successes,
971                        n_barrier_attempts,
972                        n_rcu_torture_barrier_error);
973         page += sprintf(page, "\n%s%s ", torture_type, TORTURE_FLAG);
974         if (atomic_read(&n_rcu_torture_mberror) != 0 ||
975             n_rcu_torture_barrier_error != 0 ||
976             n_rcu_torture_boost_ktrerror != 0 ||
977             n_rcu_torture_boost_rterror != 0 ||
978             n_rcu_torture_boost_failure != 0 ||
979             i > 1) {
980                 page += sprintf(page, "!!! ");
981                 atomic_inc(&n_rcu_torture_error);
982                 WARN_ON_ONCE(1);
983         }
984         page += sprintf(page, "Reader Pipe: ");
985         for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
986                 page += sprintf(page, " %ld", pipesummary[i]);
987         page += sprintf(page, "\n%s%s ", torture_type, TORTURE_FLAG);
988         page += sprintf(page, "Reader Batch: ");
989         for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
990                 page += sprintf(page, " %ld", batchsummary[i]);
991         page += sprintf(page, "\n%s%s ", torture_type, TORTURE_FLAG);
992         page += sprintf(page, "Free-Block Circulation: ");
993         for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
994                 page += sprintf(page, " %d",
995                                atomic_read(&rcu_torture_wcount[i]));
996         }
997         page += sprintf(page, "\n");
998         if (cur_ops->stats)
999                 cur_ops->stats(page);
1000 }
1001
1002 /*
1003  * Print torture statistics.  Caller must ensure that there is only
1004  * one call to this function at a given time!!!  This is normally
1005  * accomplished by relying on the module system to only have one copy
1006  * of the module loaded, and then by giving the rcu_torture_stats
1007  * kthread full control (or the init/cleanup functions when rcu_torture_stats
1008  * thread is not running).
1009  */
1010 static void
1011 rcu_torture_stats_print(void)
1012 {
1013         int size = nr_cpu_ids * 200 + 8192;
1014         char *buf;
1015
1016         buf = kmalloc(size, GFP_KERNEL);
1017         if (!buf) {
1018                 pr_err("rcu-torture: Out of memory, need: %d", size);
1019                 return;
1020         }
1021         rcu_torture_printk(buf);
1022         pr_alert("%s", buf);
1023         kfree(buf);
1024 }
1025
1026 /*
1027  * Periodically prints torture statistics, if periodic statistics printing
1028  * was specified via the stat_interval module parameter.
1029  */
1030 static int
1031 rcu_torture_stats(void *arg)
1032 {
1033         VERBOSE_TOROUT_STRING("rcu_torture_stats task started");
1034         do {
1035                 schedule_timeout_interruptible(stat_interval * HZ);
1036                 rcu_torture_stats_print();
1037                 torture_shutdown_absorb("rcu_torture_stats");
1038         } while (!torture_must_stop());
1039         torture_kthread_stopping("rcu_torture_stats");
1040         return 0;
1041 }
1042
1043 static inline void
1044 rcu_torture_print_module_parms(struct rcu_torture_ops *cur_ops, const char *tag)
1045 {
1046         pr_alert("%s" TORTURE_FLAG
1047                  "--- %s: nreaders=%d nfakewriters=%d "
1048                  "stat_interval=%d verbose=%d test_no_idle_hz=%d "
1049                  "shuffle_interval=%d stutter=%d irqreader=%d "
1050                  "fqs_duration=%d fqs_holdoff=%d fqs_stutter=%d "
1051                  "test_boost=%d/%d test_boost_interval=%d "
1052                  "test_boost_duration=%d shutdown_secs=%d "
1053                  "stall_cpu=%d stall_cpu_holdoff=%d "
1054                  "n_barrier_cbs=%d "
1055                  "onoff_interval=%d onoff_holdoff=%d\n",
1056                  torture_type, tag, nrealreaders, nfakewriters,
1057                  stat_interval, verbose, test_no_idle_hz, shuffle_interval,
1058                  stutter, irqreader, fqs_duration, fqs_holdoff, fqs_stutter,
1059                  test_boost, cur_ops->can_boost,
1060                  test_boost_interval, test_boost_duration, shutdown_secs,
1061                  stall_cpu, stall_cpu_holdoff,
1062                  n_barrier_cbs,
1063                  onoff_interval, onoff_holdoff);
1064 }
1065
1066 static void rcutorture_booster_cleanup(int cpu)
1067 {
1068         struct task_struct *t;
1069
1070         if (boost_tasks[cpu] == NULL)
1071                 return;
1072         mutex_lock(&boost_mutex);
1073         t = boost_tasks[cpu];
1074         boost_tasks[cpu] = NULL;
1075         mutex_unlock(&boost_mutex);
1076
1077         /* This must be outside of the mutex, otherwise deadlock! */
1078         torture_stop_kthread(rcu_torture_boost, t);
1079 }
1080
1081 static int rcutorture_booster_init(int cpu)
1082 {
1083         int retval;
1084
1085         if (boost_tasks[cpu] != NULL)
1086                 return 0;  /* Already created, nothing more to do. */
1087
1088         /* Don't allow time recalculation while creating a new task. */
1089         mutex_lock(&boost_mutex);
1090         VERBOSE_TOROUT_STRING("Creating rcu_torture_boost task");
1091         boost_tasks[cpu] = kthread_create_on_node(rcu_torture_boost, NULL,
1092                                                   cpu_to_node(cpu),
1093                                                   "rcu_torture_boost");
1094         if (IS_ERR(boost_tasks[cpu])) {
1095                 retval = PTR_ERR(boost_tasks[cpu]);
1096                 VERBOSE_TOROUT_STRING("rcu_torture_boost task create failed");
1097                 n_rcu_torture_boost_ktrerror++;
1098                 boost_tasks[cpu] = NULL;
1099                 mutex_unlock(&boost_mutex);
1100                 return retval;
1101         }
1102         kthread_bind(boost_tasks[cpu], cpu);
1103         wake_up_process(boost_tasks[cpu]);
1104         mutex_unlock(&boost_mutex);
1105         return 0;
1106 }
1107
1108 /*
1109  * CPU-stall kthread.  It waits as specified by stall_cpu_holdoff, then
1110  * induces a CPU stall for the time specified by stall_cpu.
1111  */
1112 static int rcu_torture_stall(void *args)
1113 {
1114         unsigned long stop_at;
1115
1116         VERBOSE_TOROUT_STRING("rcu_torture_stall task started");
1117         if (stall_cpu_holdoff > 0) {
1118                 VERBOSE_TOROUT_STRING("rcu_torture_stall begin holdoff");
1119                 schedule_timeout_interruptible(stall_cpu_holdoff * HZ);
1120                 VERBOSE_TOROUT_STRING("rcu_torture_stall end holdoff");
1121         }
1122         if (!kthread_should_stop()) {
1123                 stop_at = get_seconds() + stall_cpu;
1124                 /* RCU CPU stall is expected behavior in following code. */
1125                 pr_alert("rcu_torture_stall start.\n");
1126                 rcu_read_lock();
1127                 preempt_disable();
1128                 while (ULONG_CMP_LT(get_seconds(), stop_at))
1129                         continue;  /* Induce RCU CPU stall warning. */
1130                 preempt_enable();
1131                 rcu_read_unlock();
1132                 pr_alert("rcu_torture_stall end.\n");
1133         }
1134         torture_shutdown_absorb("rcu_torture_stall");
1135         while (!kthread_should_stop())
1136                 schedule_timeout_interruptible(10 * HZ);
1137         return 0;
1138 }
1139
1140 /* Spawn CPU-stall kthread, if stall_cpu specified. */
1141 static int __init rcu_torture_stall_init(void)
1142 {
1143         if (stall_cpu <= 0)
1144                 return 0;
1145         return torture_create_kthread(rcu_torture_stall, NULL, stall_task);
1146 }
1147
1148 /* Callback function for RCU barrier testing. */
1149 void rcu_torture_barrier_cbf(struct rcu_head *rcu)
1150 {
1151         atomic_inc(&barrier_cbs_invoked);
1152 }
1153
1154 /* kthread function to register callbacks used to test RCU barriers. */
1155 static int rcu_torture_barrier_cbs(void *arg)
1156 {
1157         long myid = (long)arg;
1158         bool lastphase = 0;
1159         bool newphase;
1160         struct rcu_head rcu;
1161
1162         init_rcu_head_on_stack(&rcu);
1163         VERBOSE_TOROUT_STRING("rcu_torture_barrier_cbs task started");
1164         set_user_nice(current, MAX_NICE);
1165         do {
1166                 wait_event(barrier_cbs_wq[myid],
1167                            (newphase =
1168                             ACCESS_ONCE(barrier_phase)) != lastphase ||
1169                            torture_must_stop());
1170                 lastphase = newphase;
1171                 smp_mb(); /* ensure barrier_phase load before ->call(). */
1172                 if (torture_must_stop())
1173                         break;
1174                 cur_ops->call(&rcu, rcu_torture_barrier_cbf);
1175                 if (atomic_dec_and_test(&barrier_cbs_count))
1176                         wake_up(&barrier_wq);
1177         } while (!torture_must_stop());
1178         cur_ops->cb_barrier();
1179         destroy_rcu_head_on_stack(&rcu);
1180         torture_kthread_stopping("rcu_torture_barrier_cbs");
1181         return 0;
1182 }
1183
1184 /* kthread function to drive and coordinate RCU barrier testing. */
1185 static int rcu_torture_barrier(void *arg)
1186 {
1187         int i;
1188
1189         VERBOSE_TOROUT_STRING("rcu_torture_barrier task starting");
1190         do {
1191                 atomic_set(&barrier_cbs_invoked, 0);
1192                 atomic_set(&barrier_cbs_count, n_barrier_cbs);
1193                 smp_mb(); /* Ensure barrier_phase after prior assignments. */
1194                 barrier_phase = !barrier_phase;
1195                 for (i = 0; i < n_barrier_cbs; i++)
1196                         wake_up(&barrier_cbs_wq[i]);
1197                 wait_event(barrier_wq,
1198                            atomic_read(&barrier_cbs_count) == 0 ||
1199                            torture_must_stop());
1200                 if (torture_must_stop())
1201                         break;
1202                 n_barrier_attempts++;
1203                 cur_ops->cb_barrier(); /* Implies smp_mb() for wait_event(). */
1204                 if (atomic_read(&barrier_cbs_invoked) != n_barrier_cbs) {
1205                         n_rcu_torture_barrier_error++;
1206                         WARN_ON_ONCE(1);
1207                 }
1208                 n_barrier_successes++;
1209                 schedule_timeout_interruptible(HZ / 10);
1210         } while (!torture_must_stop());
1211         torture_kthread_stopping("rcu_torture_barrier");
1212         return 0;
1213 }
1214
1215 /* Initialize RCU barrier testing. */
1216 static int rcu_torture_barrier_init(void)
1217 {
1218         int i;
1219         int ret;
1220
1221         if (n_barrier_cbs == 0)
1222                 return 0;
1223         if (cur_ops->call == NULL || cur_ops->cb_barrier == NULL) {
1224                 pr_alert("%s" TORTURE_FLAG
1225                          " Call or barrier ops missing for %s,\n",
1226                          torture_type, cur_ops->name);
1227                 pr_alert("%s" TORTURE_FLAG
1228                          " RCU barrier testing omitted from run.\n",
1229                          torture_type);
1230                 return 0;
1231         }
1232         atomic_set(&barrier_cbs_count, 0);
1233         atomic_set(&barrier_cbs_invoked, 0);
1234         barrier_cbs_tasks =
1235                 kzalloc(n_barrier_cbs * sizeof(barrier_cbs_tasks[0]),
1236                         GFP_KERNEL);
1237         barrier_cbs_wq =
1238                 kzalloc(n_barrier_cbs * sizeof(barrier_cbs_wq[0]),
1239                         GFP_KERNEL);
1240         if (barrier_cbs_tasks == NULL || !barrier_cbs_wq)
1241                 return -ENOMEM;
1242         for (i = 0; i < n_barrier_cbs; i++) {
1243                 init_waitqueue_head(&barrier_cbs_wq[i]);
1244                 ret = torture_create_kthread(rcu_torture_barrier_cbs,
1245                                              (void *)(long)i,
1246                                              barrier_cbs_tasks[i]);
1247                 if (ret)
1248                         return ret;
1249         }
1250         return torture_create_kthread(rcu_torture_barrier, NULL, barrier_task);
1251 }
1252
1253 /* Clean up after RCU barrier testing. */
1254 static void rcu_torture_barrier_cleanup(void)
1255 {
1256         int i;
1257
1258         torture_stop_kthread(rcu_torture_barrier, barrier_task);
1259         if (barrier_cbs_tasks != NULL) {
1260                 for (i = 0; i < n_barrier_cbs; i++)
1261                         torture_stop_kthread(rcu_torture_barrier_cbs,
1262                                              barrier_cbs_tasks[i]);
1263                 kfree(barrier_cbs_tasks);
1264                 barrier_cbs_tasks = NULL;
1265         }
1266         if (barrier_cbs_wq != NULL) {
1267                 kfree(barrier_cbs_wq);
1268                 barrier_cbs_wq = NULL;
1269         }
1270 }
1271
1272 static int rcutorture_cpu_notify(struct notifier_block *self,
1273                                  unsigned long action, void *hcpu)
1274 {
1275         long cpu = (long)hcpu;
1276
1277         switch (action) {
1278         case CPU_ONLINE:
1279         case CPU_DOWN_FAILED:
1280                 (void)rcutorture_booster_init(cpu);
1281                 break;
1282         case CPU_DOWN_PREPARE:
1283                 rcutorture_booster_cleanup(cpu);
1284                 break;
1285         default:
1286                 break;
1287         }
1288         return NOTIFY_OK;
1289 }
1290
1291 static struct notifier_block rcutorture_cpu_nb = {
1292         .notifier_call = rcutorture_cpu_notify,
1293 };
1294
1295 static void
1296 rcu_torture_cleanup(void)
1297 {
1298         int i;
1299
1300         rcutorture_record_test_transition();
1301         if (torture_cleanup()) {
1302                 if (cur_ops->cb_barrier != NULL)
1303                         cur_ops->cb_barrier();
1304                 return;
1305         }
1306
1307         rcu_torture_barrier_cleanup();
1308         torture_stop_kthread(rcu_torture_stall, stall_task);
1309         torture_stop_kthread(rcu_torture_writer, writer_task);
1310
1311         if (reader_tasks) {
1312                 for (i = 0; i < nrealreaders; i++)
1313                         torture_stop_kthread(rcu_torture_reader,
1314                                              reader_tasks[i]);
1315                 kfree(reader_tasks);
1316         }
1317         rcu_torture_current = NULL;
1318
1319         if (fakewriter_tasks) {
1320                 for (i = 0; i < nfakewriters; i++) {
1321                         torture_stop_kthread(rcu_torture_fakewriter,
1322                                              fakewriter_tasks[i]);
1323                 }
1324                 kfree(fakewriter_tasks);
1325                 fakewriter_tasks = NULL;
1326         }
1327
1328         torture_stop_kthread(rcu_torture_stats, stats_task);
1329         torture_stop_kthread(rcu_torture_fqs, fqs_task);
1330         if ((test_boost == 1 && cur_ops->can_boost) ||
1331             test_boost == 2) {
1332                 unregister_cpu_notifier(&rcutorture_cpu_nb);
1333                 for_each_possible_cpu(i)
1334                         rcutorture_booster_cleanup(i);
1335         }
1336
1337         /* Wait for all RCU callbacks to fire.  */
1338
1339         if (cur_ops->cb_barrier != NULL)
1340                 cur_ops->cb_barrier();
1341
1342         rcu_torture_stats_print();  /* -After- the stats thread is stopped! */
1343
1344         if (atomic_read(&n_rcu_torture_error) || n_rcu_torture_barrier_error)
1345                 rcu_torture_print_module_parms(cur_ops, "End of test: FAILURE");
1346         else if (torture_onoff_failures())
1347                 rcu_torture_print_module_parms(cur_ops,
1348                                                "End of test: RCU_HOTPLUG");
1349         else
1350                 rcu_torture_print_module_parms(cur_ops, "End of test: SUCCESS");
1351 }
1352
1353 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
1354 static void rcu_torture_leak_cb(struct rcu_head *rhp)
1355 {
1356 }
1357
1358 static void rcu_torture_err_cb(struct rcu_head *rhp)
1359 {
1360         /*
1361          * This -might- happen due to race conditions, but is unlikely.
1362          * The scenario that leads to this happening is that the
1363          * first of the pair of duplicate callbacks is queued,
1364          * someone else starts a grace period that includes that
1365          * callback, then the second of the pair must wait for the
1366          * next grace period.  Unlikely, but can happen.  If it
1367          * does happen, the debug-objects subsystem won't have splatted.
1368          */
1369         pr_alert("rcutorture: duplicated callback was invoked.\n");
1370 }
1371 #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
1372
1373 /*
1374  * Verify that double-free causes debug-objects to complain, but only
1375  * if CONFIG_DEBUG_OBJECTS_RCU_HEAD=y.  Otherwise, say that the test
1376  * cannot be carried out.
1377  */
1378 static void rcu_test_debug_objects(void)
1379 {
1380 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
1381         struct rcu_head rh1;
1382         struct rcu_head rh2;
1383
1384         init_rcu_head_on_stack(&rh1);
1385         init_rcu_head_on_stack(&rh2);
1386         pr_alert("rcutorture: WARN: Duplicate call_rcu() test starting.\n");
1387
1388         /* Try to queue the rh2 pair of callbacks for the same grace period. */
1389         preempt_disable(); /* Prevent preemption from interrupting test. */
1390         rcu_read_lock(); /* Make it impossible to finish a grace period. */
1391         call_rcu(&rh1, rcu_torture_leak_cb); /* Start grace period. */
1392         local_irq_disable(); /* Make it harder to start a new grace period. */
1393         call_rcu(&rh2, rcu_torture_leak_cb);
1394         call_rcu(&rh2, rcu_torture_err_cb); /* Duplicate callback. */
1395         local_irq_enable();
1396         rcu_read_unlock();
1397         preempt_enable();
1398
1399         /* Wait for them all to get done so we can safely return. */
1400         rcu_barrier();
1401         pr_alert("rcutorture: WARN: Duplicate call_rcu() test complete.\n");
1402         destroy_rcu_head_on_stack(&rh1);
1403         destroy_rcu_head_on_stack(&rh2);
1404 #else /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
1405         pr_alert("rcutorture: !CONFIG_DEBUG_OBJECTS_RCU_HEAD, not testing duplicate call_rcu()\n");
1406 #endif /* #else #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
1407 }
1408
1409 static int __init
1410 rcu_torture_init(void)
1411 {
1412         int i;
1413         int cpu;
1414         int firsterr = 0;
1415         static struct rcu_torture_ops *torture_ops[] = {
1416                 &rcu_ops, &rcu_bh_ops, &rcu_busted_ops, &srcu_ops, &sched_ops,
1417         };
1418
1419         torture_init_begin(torture_type, verbose, &rcutorture_runnable);
1420
1421         /* Process args and tell the world that the torturer is on the job. */
1422         for (i = 0; i < ARRAY_SIZE(torture_ops); i++) {
1423                 cur_ops = torture_ops[i];
1424                 if (strcmp(torture_type, cur_ops->name) == 0)
1425                         break;
1426         }
1427         if (i == ARRAY_SIZE(torture_ops)) {
1428                 pr_alert("rcu-torture: invalid torture type: \"%s\"\n",
1429                          torture_type);
1430                 pr_alert("rcu-torture types:");
1431                 for (i = 0; i < ARRAY_SIZE(torture_ops); i++)
1432                         pr_alert(" %s", torture_ops[i]->name);
1433                 pr_alert("\n");
1434                 torture_init_end();
1435                 return -EINVAL;
1436         }
1437         if (cur_ops->fqs == NULL && fqs_duration != 0) {
1438                 pr_alert("rcu-torture: ->fqs NULL and non-zero fqs_duration, fqs disabled.\n");
1439                 fqs_duration = 0;
1440         }
1441         if (cur_ops->init)
1442                 cur_ops->init(); /* no "goto unwind" prior to this point!!! */
1443
1444         if (nreaders >= 0)
1445                 nrealreaders = nreaders;
1446         else
1447                 nrealreaders = 2 * num_online_cpus();
1448         rcu_torture_print_module_parms(cur_ops, "Start of test");
1449
1450         /* Set up the freelist. */
1451
1452         INIT_LIST_HEAD(&rcu_torture_freelist);
1453         for (i = 0; i < ARRAY_SIZE(rcu_tortures); i++) {
1454                 rcu_tortures[i].rtort_mbtest = 0;
1455                 list_add_tail(&rcu_tortures[i].rtort_free,
1456                               &rcu_torture_freelist);
1457         }
1458
1459         /* Initialize the statistics so that each run gets its own numbers. */
1460
1461         rcu_torture_current = NULL;
1462         rcu_torture_current_version = 0;
1463         atomic_set(&n_rcu_torture_alloc, 0);
1464         atomic_set(&n_rcu_torture_alloc_fail, 0);
1465         atomic_set(&n_rcu_torture_free, 0);
1466         atomic_set(&n_rcu_torture_mberror, 0);
1467         atomic_set(&n_rcu_torture_error, 0);
1468         n_rcu_torture_barrier_error = 0;
1469         n_rcu_torture_boost_ktrerror = 0;
1470         n_rcu_torture_boost_rterror = 0;
1471         n_rcu_torture_boost_failure = 0;
1472         n_rcu_torture_boosts = 0;
1473         for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
1474                 atomic_set(&rcu_torture_wcount[i], 0);
1475         for_each_possible_cpu(cpu) {
1476                 for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
1477                         per_cpu(rcu_torture_count, cpu)[i] = 0;
1478                         per_cpu(rcu_torture_batch, cpu)[i] = 0;
1479                 }
1480         }
1481
1482         /* Start up the kthreads. */
1483
1484         firsterr = torture_create_kthread(rcu_torture_writer, NULL,
1485                                           writer_task);
1486         if (firsterr)
1487                 goto unwind;
1488         fakewriter_tasks = kzalloc(nfakewriters * sizeof(fakewriter_tasks[0]),
1489                                    GFP_KERNEL);
1490         if (fakewriter_tasks == NULL) {
1491                 VERBOSE_TOROUT_ERRSTRING("out of memory");
1492                 firsterr = -ENOMEM;
1493                 goto unwind;
1494         }
1495         for (i = 0; i < nfakewriters; i++) {
1496                 firsterr = torture_create_kthread(rcu_torture_fakewriter,
1497                                                   NULL, fakewriter_tasks[i]);
1498                 if (firsterr)
1499                         goto unwind;
1500         }
1501         reader_tasks = kzalloc(nrealreaders * sizeof(reader_tasks[0]),
1502                                GFP_KERNEL);
1503         if (reader_tasks == NULL) {
1504                 VERBOSE_TOROUT_ERRSTRING("out of memory");
1505                 firsterr = -ENOMEM;
1506                 goto unwind;
1507         }
1508         for (i = 0; i < nrealreaders; i++) {
1509                 firsterr = torture_create_kthread(rcu_torture_reader, NULL,
1510                                                   reader_tasks[i]);
1511                 if (firsterr)
1512                         goto unwind;
1513         }
1514         if (stat_interval > 0) {
1515                 firsterr = torture_create_kthread(rcu_torture_stats, NULL,
1516                                                   stats_task);
1517                 if (firsterr)
1518                         goto unwind;
1519         }
1520         if (test_no_idle_hz) {
1521                 firsterr = torture_shuffle_init(shuffle_interval * HZ);
1522                 if (firsterr)
1523                         goto unwind;
1524         }
1525         if (stutter < 0)
1526                 stutter = 0;
1527         if (stutter) {
1528                 firsterr = torture_stutter_init(stutter * HZ);
1529                 if (firsterr)
1530                         goto unwind;
1531         }
1532         if (fqs_duration < 0)
1533                 fqs_duration = 0;
1534         if (fqs_duration) {
1535                 /* Create the fqs thread */
1536                 torture_create_kthread(rcu_torture_fqs, NULL, fqs_task);
1537                 if (firsterr)
1538                         goto unwind;
1539         }
1540         if (test_boost_interval < 1)
1541                 test_boost_interval = 1;
1542         if (test_boost_duration < 2)
1543                 test_boost_duration = 2;
1544         if ((test_boost == 1 && cur_ops->can_boost) ||
1545             test_boost == 2) {
1546
1547                 boost_starttime = jiffies + test_boost_interval * HZ;
1548                 register_cpu_notifier(&rcutorture_cpu_nb);
1549                 for_each_possible_cpu(i) {
1550                         if (cpu_is_offline(i))
1551                                 continue;  /* Heuristic: CPU can go offline. */
1552                         firsterr = rcutorture_booster_init(i);
1553                         if (firsterr)
1554                                 goto unwind;
1555                 }
1556         }
1557         firsterr = torture_shutdown_init(shutdown_secs, rcu_torture_cleanup);
1558         if (firsterr)
1559                 goto unwind;
1560         firsterr = torture_onoff_init(onoff_holdoff * HZ, onoff_interval * HZ);
1561         if (firsterr)
1562                 goto unwind;
1563         firsterr = rcu_torture_stall_init();
1564         if (firsterr)
1565                 goto unwind;
1566         firsterr = rcu_torture_barrier_init();
1567         if (firsterr)
1568                 goto unwind;
1569         if (object_debug)
1570                 rcu_test_debug_objects();
1571         rcutorture_record_test_transition();
1572         torture_init_end();
1573         return 0;
1574
1575 unwind:
1576         torture_init_end();
1577         rcu_torture_cleanup();
1578         return firsterr;
1579 }
1580
1581 module_init(rcu_torture_init);
1582 module_exit(rcu_torture_cleanup);