xref: /linux/kernel/rcu/rcutorture.c (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
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@joshtriplett.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@joshtriplett.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_cond, false, "Use conditional/async GP wait primitives");
62 torture_param(bool, gp_exp, false, "Use expedited GP wait primitives");
63 torture_param(bool, gp_normal, false,
64 	     "Use normal (non-expedited) GP wait primitives");
65 torture_param(bool, gp_sync, false, "Use synchronous GP wait primitives");
66 torture_param(int, irqreader, 1, "Allow RCU readers from irq handlers");
67 torture_param(int, n_barrier_cbs, 0,
68 	     "# of callbacks/kthreads for barrier testing");
69 torture_param(int, nfakewriters, 4, "Number of RCU fake writer threads");
70 torture_param(int, nreaders, -1, "Number of RCU reader threads");
71 torture_param(int, object_debug, 0,
72 	     "Enable debug-object double call_rcu() testing");
73 torture_param(int, onoff_holdoff, 0, "Time after boot before CPU hotplugs (s)");
74 torture_param(int, onoff_interval, 0,
75 	     "Time between CPU hotplugs (s), 0=disable");
76 torture_param(int, shuffle_interval, 3, "Number of seconds between shuffles");
77 torture_param(int, shutdown_secs, 0, "Shutdown time (s), <= zero to disable.");
78 torture_param(int, stall_cpu, 0, "Stall duration (s), zero to disable.");
79 torture_param(int, stall_cpu_holdoff, 10,
80 	     "Time to wait before starting stall (s).");
81 torture_param(int, stat_interval, 60,
82 	     "Number of seconds between stats printk()s");
83 torture_param(int, stutter, 5, "Number of seconds to run/halt test");
84 torture_param(int, test_boost, 1, "Test RCU prio boost: 0=no, 1=maybe, 2=yes.");
85 torture_param(int, test_boost_duration, 4,
86 	     "Duration of each boost test, seconds.");
87 torture_param(int, test_boost_interval, 7,
88 	     "Interval between boost tests, seconds.");
89 torture_param(bool, test_no_idle_hz, true,
90 	     "Test support for tickless idle CPUs");
91 torture_param(bool, verbose, true,
92 	     "Enable verbose debugging printk()s");
93 
94 static char *torture_type = "rcu";
95 module_param(torture_type, charp, 0444);
96 MODULE_PARM_DESC(torture_type, "Type of RCU to torture (rcu, rcu_bh, ...)");
97 
98 static int nrealreaders;
99 static struct task_struct *writer_task;
100 static struct task_struct **fakewriter_tasks;
101 static struct task_struct **reader_tasks;
102 static struct task_struct *stats_task;
103 static struct task_struct *fqs_task;
104 static struct task_struct *boost_tasks[NR_CPUS];
105 static struct task_struct *stall_task;
106 static struct task_struct **barrier_cbs_tasks;
107 static struct task_struct *barrier_task;
108 
109 #define RCU_TORTURE_PIPE_LEN 10
110 
111 struct rcu_torture {
112 	struct rcu_head rtort_rcu;
113 	int rtort_pipe_count;
114 	struct list_head rtort_free;
115 	int rtort_mbtest;
116 };
117 
118 static LIST_HEAD(rcu_torture_freelist);
119 static struct rcu_torture __rcu *rcu_torture_current;
120 static unsigned long rcu_torture_current_version;
121 static struct rcu_torture rcu_tortures[10 * RCU_TORTURE_PIPE_LEN];
122 static DEFINE_SPINLOCK(rcu_torture_lock);
123 static DEFINE_PER_CPU(long [RCU_TORTURE_PIPE_LEN + 1],
124 		      rcu_torture_count) = { 0 };
125 static DEFINE_PER_CPU(long [RCU_TORTURE_PIPE_LEN + 1],
126 		      rcu_torture_batch) = { 0 };
127 static atomic_t rcu_torture_wcount[RCU_TORTURE_PIPE_LEN + 1];
128 static atomic_t n_rcu_torture_alloc;
129 static atomic_t n_rcu_torture_alloc_fail;
130 static atomic_t n_rcu_torture_free;
131 static atomic_t n_rcu_torture_mberror;
132 static atomic_t n_rcu_torture_error;
133 static long n_rcu_torture_barrier_error;
134 static long n_rcu_torture_boost_ktrerror;
135 static long n_rcu_torture_boost_rterror;
136 static long n_rcu_torture_boost_failure;
137 static long n_rcu_torture_boosts;
138 static long n_rcu_torture_timers;
139 static long n_barrier_attempts;
140 static long n_barrier_successes;
141 static struct list_head rcu_torture_removed;
142 
143 static int rcu_torture_writer_state;
144 #define RTWS_FIXED_DELAY	0
145 #define RTWS_DELAY		1
146 #define RTWS_REPLACE		2
147 #define RTWS_DEF_FREE		3
148 #define RTWS_EXP_SYNC		4
149 #define RTWS_COND_GET		5
150 #define RTWS_COND_SYNC		6
151 #define RTWS_SYNC		7
152 #define RTWS_STUTTER		8
153 #define RTWS_STOPPING		9
154 
155 #if defined(MODULE) || defined(CONFIG_RCU_TORTURE_TEST_RUNNABLE)
156 #define RCUTORTURE_RUNNABLE_INIT 1
157 #else
158 #define RCUTORTURE_RUNNABLE_INIT 0
159 #endif
160 int rcutorture_runnable = RCUTORTURE_RUNNABLE_INIT;
161 module_param(rcutorture_runnable, int, 0444);
162 MODULE_PARM_DESC(rcutorture_runnable, "Start rcutorture at boot");
163 
164 #if defined(CONFIG_RCU_BOOST) && !defined(CONFIG_HOTPLUG_CPU)
165 #define rcu_can_boost() 1
166 #else /* #if defined(CONFIG_RCU_BOOST) && !defined(CONFIG_HOTPLUG_CPU) */
167 #define rcu_can_boost() 0
168 #endif /* #else #if defined(CONFIG_RCU_BOOST) && !defined(CONFIG_HOTPLUG_CPU) */
169 
170 #ifdef CONFIG_RCU_TRACE
171 static u64 notrace rcu_trace_clock_local(void)
172 {
173 	u64 ts = trace_clock_local();
174 	unsigned long __maybe_unused ts_rem = do_div(ts, NSEC_PER_USEC);
175 	return ts;
176 }
177 #else /* #ifdef CONFIG_RCU_TRACE */
178 static u64 notrace rcu_trace_clock_local(void)
179 {
180 	return 0ULL;
181 }
182 #endif /* #else #ifdef CONFIG_RCU_TRACE */
183 
184 static unsigned long boost_starttime;	/* jiffies of next boost test start. */
185 DEFINE_MUTEX(boost_mutex);		/* protect setting boost_starttime */
186 					/*  and boost task create/destroy. */
187 static atomic_t barrier_cbs_count;	/* Barrier callbacks registered. */
188 static bool barrier_phase;		/* Test phase. */
189 static atomic_t barrier_cbs_invoked;	/* Barrier callbacks invoked. */
190 static wait_queue_head_t *barrier_cbs_wq; /* Coordinate barrier testing. */
191 static DECLARE_WAIT_QUEUE_HEAD(barrier_wq);
192 
193 /*
194  * Allocate an element from the rcu_tortures pool.
195  */
196 static struct rcu_torture *
197 rcu_torture_alloc(void)
198 {
199 	struct list_head *p;
200 
201 	spin_lock_bh(&rcu_torture_lock);
202 	if (list_empty(&rcu_torture_freelist)) {
203 		atomic_inc(&n_rcu_torture_alloc_fail);
204 		spin_unlock_bh(&rcu_torture_lock);
205 		return NULL;
206 	}
207 	atomic_inc(&n_rcu_torture_alloc);
208 	p = rcu_torture_freelist.next;
209 	list_del_init(p);
210 	spin_unlock_bh(&rcu_torture_lock);
211 	return container_of(p, struct rcu_torture, rtort_free);
212 }
213 
214 /*
215  * Free an element to the rcu_tortures pool.
216  */
217 static void
218 rcu_torture_free(struct rcu_torture *p)
219 {
220 	atomic_inc(&n_rcu_torture_free);
221 	spin_lock_bh(&rcu_torture_lock);
222 	list_add_tail(&p->rtort_free, &rcu_torture_freelist);
223 	spin_unlock_bh(&rcu_torture_lock);
224 }
225 
226 /*
227  * Operations vector for selecting different types of tests.
228  */
229 
230 struct rcu_torture_ops {
231 	int ttype;
232 	void (*init)(void);
233 	int (*readlock)(void);
234 	void (*read_delay)(struct torture_random_state *rrsp);
235 	void (*readunlock)(int idx);
236 	int (*completed)(void);
237 	void (*deferred_free)(struct rcu_torture *p);
238 	void (*sync)(void);
239 	void (*exp_sync)(void);
240 	unsigned long (*get_state)(void);
241 	void (*cond_sync)(unsigned long oldstate);
242 	void (*call)(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
243 	void (*cb_barrier)(void);
244 	void (*fqs)(void);
245 	void (*stats)(char *page);
246 	int irq_capable;
247 	int can_boost;
248 	const char *name;
249 };
250 
251 static struct rcu_torture_ops *cur_ops;
252 
253 /*
254  * Definitions for rcu torture testing.
255  */
256 
257 static int rcu_torture_read_lock(void) __acquires(RCU)
258 {
259 	rcu_read_lock();
260 	return 0;
261 }
262 
263 static void rcu_read_delay(struct torture_random_state *rrsp)
264 {
265 	const unsigned long shortdelay_us = 200;
266 	const unsigned long longdelay_ms = 50;
267 
268 	/* We want a short delay sometimes to make a reader delay the grace
269 	 * period, and we want a long delay occasionally to trigger
270 	 * force_quiescent_state. */
271 
272 	if (!(torture_random(rrsp) % (nrealreaders * 2000 * longdelay_ms)))
273 		mdelay(longdelay_ms);
274 	if (!(torture_random(rrsp) % (nrealreaders * 2 * shortdelay_us)))
275 		udelay(shortdelay_us);
276 #ifdef CONFIG_PREEMPT
277 	if (!preempt_count() &&
278 	    !(torture_random(rrsp) % (nrealreaders * 20000)))
279 		preempt_schedule();  /* No QS if preempt_disable() in effect */
280 #endif
281 }
282 
283 static void rcu_torture_read_unlock(int idx) __releases(RCU)
284 {
285 	rcu_read_unlock();
286 }
287 
288 static int rcu_torture_completed(void)
289 {
290 	return rcu_batches_completed();
291 }
292 
293 /*
294  * Update callback in the pipe.  This should be invoked after a grace period.
295  */
296 static bool
297 rcu_torture_pipe_update_one(struct rcu_torture *rp)
298 {
299 	int i;
300 
301 	i = rp->rtort_pipe_count;
302 	if (i > RCU_TORTURE_PIPE_LEN)
303 		i = RCU_TORTURE_PIPE_LEN;
304 	atomic_inc(&rcu_torture_wcount[i]);
305 	if (++rp->rtort_pipe_count >= RCU_TORTURE_PIPE_LEN) {
306 		rp->rtort_mbtest = 0;
307 		return true;
308 	}
309 	return false;
310 }
311 
312 /*
313  * Update all callbacks in the pipe.  Suitable for synchronous grace-period
314  * primitives.
315  */
316 static void
317 rcu_torture_pipe_update(struct rcu_torture *old_rp)
318 {
319 	struct rcu_torture *rp;
320 	struct rcu_torture *rp1;
321 
322 	if (old_rp)
323 		list_add(&old_rp->rtort_free, &rcu_torture_removed);
324 	list_for_each_entry_safe(rp, rp1, &rcu_torture_removed, rtort_free) {
325 		if (rcu_torture_pipe_update_one(rp)) {
326 			list_del(&rp->rtort_free);
327 			rcu_torture_free(rp);
328 		}
329 	}
330 }
331 
332 static void
333 rcu_torture_cb(struct rcu_head *p)
334 {
335 	struct rcu_torture *rp = container_of(p, struct rcu_torture, rtort_rcu);
336 
337 	if (torture_must_stop_irq()) {
338 		/* Test is ending, just drop callbacks on the floor. */
339 		/* The next initialization will pick up the pieces. */
340 		return;
341 	}
342 	if (rcu_torture_pipe_update_one(rp))
343 		rcu_torture_free(rp);
344 	else
345 		cur_ops->deferred_free(rp);
346 }
347 
348 static int rcu_no_completed(void)
349 {
350 	return 0;
351 }
352 
353 static void rcu_torture_deferred_free(struct rcu_torture *p)
354 {
355 	call_rcu(&p->rtort_rcu, rcu_torture_cb);
356 }
357 
358 static void rcu_sync_torture_init(void)
359 {
360 	INIT_LIST_HEAD(&rcu_torture_removed);
361 }
362 
363 static struct rcu_torture_ops rcu_ops = {
364 	.ttype		= RCU_FLAVOR,
365 	.init		= rcu_sync_torture_init,
366 	.readlock	= rcu_torture_read_lock,
367 	.read_delay	= rcu_read_delay,
368 	.readunlock	= rcu_torture_read_unlock,
369 	.completed	= rcu_torture_completed,
370 	.deferred_free	= rcu_torture_deferred_free,
371 	.sync		= synchronize_rcu,
372 	.exp_sync	= synchronize_rcu_expedited,
373 	.get_state	= get_state_synchronize_rcu,
374 	.cond_sync	= cond_synchronize_rcu,
375 	.call		= call_rcu,
376 	.cb_barrier	= rcu_barrier,
377 	.fqs		= rcu_force_quiescent_state,
378 	.stats		= NULL,
379 	.irq_capable	= 1,
380 	.can_boost	= rcu_can_boost(),
381 	.name		= "rcu"
382 };
383 
384 /*
385  * Definitions for rcu_bh torture testing.
386  */
387 
388 static int rcu_bh_torture_read_lock(void) __acquires(RCU_BH)
389 {
390 	rcu_read_lock_bh();
391 	return 0;
392 }
393 
394 static void rcu_bh_torture_read_unlock(int idx) __releases(RCU_BH)
395 {
396 	rcu_read_unlock_bh();
397 }
398 
399 static int rcu_bh_torture_completed(void)
400 {
401 	return rcu_batches_completed_bh();
402 }
403 
404 static void rcu_bh_torture_deferred_free(struct rcu_torture *p)
405 {
406 	call_rcu_bh(&p->rtort_rcu, rcu_torture_cb);
407 }
408 
409 static struct rcu_torture_ops rcu_bh_ops = {
410 	.ttype		= RCU_BH_FLAVOR,
411 	.init		= rcu_sync_torture_init,
412 	.readlock	= rcu_bh_torture_read_lock,
413 	.read_delay	= rcu_read_delay,  /* just reuse rcu's version. */
414 	.readunlock	= rcu_bh_torture_read_unlock,
415 	.completed	= rcu_bh_torture_completed,
416 	.deferred_free	= rcu_bh_torture_deferred_free,
417 	.sync		= synchronize_rcu_bh,
418 	.exp_sync	= synchronize_rcu_bh_expedited,
419 	.call		= call_rcu_bh,
420 	.cb_barrier	= rcu_barrier_bh,
421 	.fqs		= rcu_bh_force_quiescent_state,
422 	.stats		= NULL,
423 	.irq_capable	= 1,
424 	.name		= "rcu_bh"
425 };
426 
427 /*
428  * Don't even think about trying any of these in real life!!!
429  * The names includes "busted", and they really means it!
430  * The only purpose of these functions is to provide a buggy RCU
431  * implementation to make sure that rcutorture correctly emits
432  * buggy-RCU error messages.
433  */
434 static void rcu_busted_torture_deferred_free(struct rcu_torture *p)
435 {
436 	/* This is a deliberate bug for testing purposes only! */
437 	rcu_torture_cb(&p->rtort_rcu);
438 }
439 
440 static void synchronize_rcu_busted(void)
441 {
442 	/* This is a deliberate bug for testing purposes only! */
443 }
444 
445 static void
446 call_rcu_busted(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
447 {
448 	/* This is a deliberate bug for testing purposes only! */
449 	func(head);
450 }
451 
452 static struct rcu_torture_ops rcu_busted_ops = {
453 	.ttype		= INVALID_RCU_FLAVOR,
454 	.init		= rcu_sync_torture_init,
455 	.readlock	= rcu_torture_read_lock,
456 	.read_delay	= rcu_read_delay,  /* just reuse rcu's version. */
457 	.readunlock	= rcu_torture_read_unlock,
458 	.completed	= rcu_no_completed,
459 	.deferred_free	= rcu_busted_torture_deferred_free,
460 	.sync		= synchronize_rcu_busted,
461 	.exp_sync	= synchronize_rcu_busted,
462 	.call		= call_rcu_busted,
463 	.cb_barrier	= NULL,
464 	.fqs		= NULL,
465 	.stats		= NULL,
466 	.irq_capable	= 1,
467 	.name		= "rcu_busted"
468 };
469 
470 /*
471  * Definitions for srcu torture testing.
472  */
473 
474 DEFINE_STATIC_SRCU(srcu_ctl);
475 
476 static int srcu_torture_read_lock(void) __acquires(&srcu_ctl)
477 {
478 	return srcu_read_lock(&srcu_ctl);
479 }
480 
481 static void srcu_read_delay(struct torture_random_state *rrsp)
482 {
483 	long delay;
484 	const long uspertick = 1000000 / HZ;
485 	const long longdelay = 10;
486 
487 	/* We want there to be long-running readers, but not all the time. */
488 
489 	delay = torture_random(rrsp) %
490 		(nrealreaders * 2 * longdelay * uspertick);
491 	if (!delay)
492 		schedule_timeout_interruptible(longdelay);
493 	else
494 		rcu_read_delay(rrsp);
495 }
496 
497 static void srcu_torture_read_unlock(int idx) __releases(&srcu_ctl)
498 {
499 	srcu_read_unlock(&srcu_ctl, idx);
500 }
501 
502 static int srcu_torture_completed(void)
503 {
504 	return srcu_batches_completed(&srcu_ctl);
505 }
506 
507 static void srcu_torture_deferred_free(struct rcu_torture *rp)
508 {
509 	call_srcu(&srcu_ctl, &rp->rtort_rcu, rcu_torture_cb);
510 }
511 
512 static void srcu_torture_synchronize(void)
513 {
514 	synchronize_srcu(&srcu_ctl);
515 }
516 
517 static void srcu_torture_call(struct rcu_head *head,
518 			      void (*func)(struct rcu_head *head))
519 {
520 	call_srcu(&srcu_ctl, head, func);
521 }
522 
523 static void srcu_torture_barrier(void)
524 {
525 	srcu_barrier(&srcu_ctl);
526 }
527 
528 static void srcu_torture_stats(char *page)
529 {
530 	int cpu;
531 	int idx = srcu_ctl.completed & 0x1;
532 
533 	page += sprintf(page, "%s%s per-CPU(idx=%d):",
534 		       torture_type, TORTURE_FLAG, idx);
535 	for_each_possible_cpu(cpu) {
536 		long c0, c1;
537 
538 		c0 = (long)per_cpu_ptr(srcu_ctl.per_cpu_ref, cpu)->c[!idx];
539 		c1 = (long)per_cpu_ptr(srcu_ctl.per_cpu_ref, cpu)->c[idx];
540 		page += sprintf(page, " %d(%ld,%ld)", cpu, c0, c1);
541 	}
542 	sprintf(page, "\n");
543 }
544 
545 static void srcu_torture_synchronize_expedited(void)
546 {
547 	synchronize_srcu_expedited(&srcu_ctl);
548 }
549 
550 static struct rcu_torture_ops srcu_ops = {
551 	.ttype		= SRCU_FLAVOR,
552 	.init		= rcu_sync_torture_init,
553 	.readlock	= srcu_torture_read_lock,
554 	.read_delay	= srcu_read_delay,
555 	.readunlock	= srcu_torture_read_unlock,
556 	.completed	= srcu_torture_completed,
557 	.deferred_free	= srcu_torture_deferred_free,
558 	.sync		= srcu_torture_synchronize,
559 	.exp_sync	= srcu_torture_synchronize_expedited,
560 	.call		= srcu_torture_call,
561 	.cb_barrier	= srcu_torture_barrier,
562 	.stats		= srcu_torture_stats,
563 	.name		= "srcu"
564 };
565 
566 /*
567  * Definitions for sched torture testing.
568  */
569 
570 static int sched_torture_read_lock(void)
571 {
572 	preempt_disable();
573 	return 0;
574 }
575 
576 static void sched_torture_read_unlock(int idx)
577 {
578 	preempt_enable();
579 }
580 
581 static void rcu_sched_torture_deferred_free(struct rcu_torture *p)
582 {
583 	call_rcu_sched(&p->rtort_rcu, rcu_torture_cb);
584 }
585 
586 static struct rcu_torture_ops sched_ops = {
587 	.ttype		= RCU_SCHED_FLAVOR,
588 	.init		= rcu_sync_torture_init,
589 	.readlock	= sched_torture_read_lock,
590 	.read_delay	= rcu_read_delay,  /* just reuse rcu's version. */
591 	.readunlock	= sched_torture_read_unlock,
592 	.completed	= rcu_no_completed,
593 	.deferred_free	= rcu_sched_torture_deferred_free,
594 	.sync		= synchronize_sched,
595 	.exp_sync	= synchronize_sched_expedited,
596 	.call		= call_rcu_sched,
597 	.cb_barrier	= rcu_barrier_sched,
598 	.fqs		= rcu_sched_force_quiescent_state,
599 	.stats		= NULL,
600 	.irq_capable	= 1,
601 	.name		= "sched"
602 };
603 
604 /*
605  * RCU torture priority-boost testing.  Runs one real-time thread per
606  * CPU for moderate bursts, repeatedly registering RCU callbacks and
607  * spinning waiting for them to be invoked.  If a given callback takes
608  * too long to be invoked, we assume that priority inversion has occurred.
609  */
610 
611 struct rcu_boost_inflight {
612 	struct rcu_head rcu;
613 	int inflight;
614 };
615 
616 static void rcu_torture_boost_cb(struct rcu_head *head)
617 {
618 	struct rcu_boost_inflight *rbip =
619 		container_of(head, struct rcu_boost_inflight, rcu);
620 
621 	smp_mb(); /* Ensure RCU-core accesses precede clearing ->inflight */
622 	rbip->inflight = 0;
623 }
624 
625 static int rcu_torture_boost(void *arg)
626 {
627 	unsigned long call_rcu_time;
628 	unsigned long endtime;
629 	unsigned long oldstarttime;
630 	struct rcu_boost_inflight rbi = { .inflight = 0 };
631 	struct sched_param sp;
632 
633 	VERBOSE_TOROUT_STRING("rcu_torture_boost started");
634 
635 	/* Set real-time priority. */
636 	sp.sched_priority = 1;
637 	if (sched_setscheduler(current, SCHED_FIFO, &sp) < 0) {
638 		VERBOSE_TOROUT_STRING("rcu_torture_boost RT prio failed!");
639 		n_rcu_torture_boost_rterror++;
640 	}
641 
642 	init_rcu_head_on_stack(&rbi.rcu);
643 	/* Each pass through the following loop does one boost-test cycle. */
644 	do {
645 		/* Wait for the next test interval. */
646 		oldstarttime = boost_starttime;
647 		while (ULONG_CMP_LT(jiffies, oldstarttime)) {
648 			schedule_timeout_interruptible(oldstarttime - jiffies);
649 			stutter_wait("rcu_torture_boost");
650 			if (torture_must_stop())
651 				goto checkwait;
652 		}
653 
654 		/* Do one boost-test interval. */
655 		endtime = oldstarttime + test_boost_duration * HZ;
656 		call_rcu_time = jiffies;
657 		while (ULONG_CMP_LT(jiffies, endtime)) {
658 			/* If we don't have a callback in flight, post one. */
659 			if (!rbi.inflight) {
660 				smp_mb(); /* RCU core before ->inflight = 1. */
661 				rbi.inflight = 1;
662 				call_rcu(&rbi.rcu, rcu_torture_boost_cb);
663 				if (jiffies - call_rcu_time >
664 					 test_boost_duration * HZ - HZ / 2) {
665 					VERBOSE_TOROUT_STRING("rcu_torture_boost boosting failed");
666 					n_rcu_torture_boost_failure++;
667 				}
668 				call_rcu_time = jiffies;
669 			}
670 			cond_resched();
671 			stutter_wait("rcu_torture_boost");
672 			if (torture_must_stop())
673 				goto checkwait;
674 		}
675 
676 		/*
677 		 * Set the start time of the next test interval.
678 		 * Yes, this is vulnerable to long delays, but such
679 		 * delays simply cause a false negative for the next
680 		 * interval.  Besides, we are running at RT priority,
681 		 * so delays should be relatively rare.
682 		 */
683 		while (oldstarttime == boost_starttime &&
684 		       !kthread_should_stop()) {
685 			if (mutex_trylock(&boost_mutex)) {
686 				boost_starttime = jiffies +
687 						  test_boost_interval * HZ;
688 				n_rcu_torture_boosts++;
689 				mutex_unlock(&boost_mutex);
690 				break;
691 			}
692 			schedule_timeout_uninterruptible(1);
693 		}
694 
695 		/* Go do the stutter. */
696 checkwait:	stutter_wait("rcu_torture_boost");
697 	} while (!torture_must_stop());
698 
699 	/* Clean up and exit. */
700 	while (!kthread_should_stop() || rbi.inflight) {
701 		torture_shutdown_absorb("rcu_torture_boost");
702 		schedule_timeout_uninterruptible(1);
703 	}
704 	smp_mb(); /* order accesses to ->inflight before stack-frame death. */
705 	destroy_rcu_head_on_stack(&rbi.rcu);
706 	torture_kthread_stopping("rcu_torture_boost");
707 	return 0;
708 }
709 
710 /*
711  * RCU torture force-quiescent-state kthread.  Repeatedly induces
712  * bursts of calls to force_quiescent_state(), increasing the probability
713  * of occurrence of some important types of race conditions.
714  */
715 static int
716 rcu_torture_fqs(void *arg)
717 {
718 	unsigned long fqs_resume_time;
719 	int fqs_burst_remaining;
720 
721 	VERBOSE_TOROUT_STRING("rcu_torture_fqs task started");
722 	do {
723 		fqs_resume_time = jiffies + fqs_stutter * HZ;
724 		while (ULONG_CMP_LT(jiffies, fqs_resume_time) &&
725 		       !kthread_should_stop()) {
726 			schedule_timeout_interruptible(1);
727 		}
728 		fqs_burst_remaining = fqs_duration;
729 		while (fqs_burst_remaining > 0 &&
730 		       !kthread_should_stop()) {
731 			cur_ops->fqs();
732 			udelay(fqs_holdoff);
733 			fqs_burst_remaining -= fqs_holdoff;
734 		}
735 		stutter_wait("rcu_torture_fqs");
736 	} while (!torture_must_stop());
737 	torture_kthread_stopping("rcu_torture_fqs");
738 	return 0;
739 }
740 
741 /*
742  * RCU torture writer kthread.  Repeatedly substitutes a new structure
743  * for that pointed to by rcu_torture_current, freeing the old structure
744  * after a series of grace periods (the "pipeline").
745  */
746 static int
747 rcu_torture_writer(void *arg)
748 {
749 	unsigned long gp_snap;
750 	bool gp_cond1 = gp_cond, gp_exp1 = gp_exp, gp_normal1 = gp_normal;
751 	bool gp_sync1 = gp_sync;
752 	int i;
753 	struct rcu_torture *rp;
754 	struct rcu_torture *old_rp;
755 	static DEFINE_TORTURE_RANDOM(rand);
756 	int synctype[] = { RTWS_DEF_FREE, RTWS_EXP_SYNC,
757 			   RTWS_COND_GET, RTWS_SYNC };
758 	int nsynctypes = 0;
759 
760 	VERBOSE_TOROUT_STRING("rcu_torture_writer task started");
761 
762 	/* Initialize synctype[] array.  If none set, take default. */
763 	if (!gp_cond1 && !gp_exp1 && !gp_normal1 && !gp_sync)
764 		gp_cond1 = gp_exp1 = gp_normal1 = gp_sync1 = true;
765 	if (gp_cond1 && cur_ops->get_state && cur_ops->cond_sync)
766 		synctype[nsynctypes++] = RTWS_COND_GET;
767 	else if (gp_cond && (!cur_ops->get_state || !cur_ops->cond_sync))
768 		pr_alert("rcu_torture_writer: gp_cond without primitives.\n");
769 	if (gp_exp1 && cur_ops->exp_sync)
770 		synctype[nsynctypes++] = RTWS_EXP_SYNC;
771 	else if (gp_exp && !cur_ops->exp_sync)
772 		pr_alert("rcu_torture_writer: gp_exp without primitives.\n");
773 	if (gp_normal1 && cur_ops->deferred_free)
774 		synctype[nsynctypes++] = RTWS_DEF_FREE;
775 	else if (gp_normal && !cur_ops->deferred_free)
776 		pr_alert("rcu_torture_writer: gp_normal without primitives.\n");
777 	if (gp_sync1 && cur_ops->sync)
778 		synctype[nsynctypes++] = RTWS_SYNC;
779 	else if (gp_sync && !cur_ops->sync)
780 		pr_alert("rcu_torture_writer: gp_sync without primitives.\n");
781 	if (WARN_ONCE(nsynctypes == 0,
782 		      "rcu_torture_writer: No update-side primitives.\n")) {
783 		/*
784 		 * No updates primitives, so don't try updating.
785 		 * The resulting test won't be testing much, hence the
786 		 * above WARN_ONCE().
787 		 */
788 		rcu_torture_writer_state = RTWS_STOPPING;
789 		torture_kthread_stopping("rcu_torture_writer");
790 	}
791 
792 	do {
793 		rcu_torture_writer_state = RTWS_FIXED_DELAY;
794 		schedule_timeout_uninterruptible(1);
795 		rp = rcu_torture_alloc();
796 		if (rp == NULL)
797 			continue;
798 		rp->rtort_pipe_count = 0;
799 		rcu_torture_writer_state = RTWS_DELAY;
800 		udelay(torture_random(&rand) & 0x3ff);
801 		rcu_torture_writer_state = RTWS_REPLACE;
802 		old_rp = rcu_dereference_check(rcu_torture_current,
803 					       current == writer_task);
804 		rp->rtort_mbtest = 1;
805 		rcu_assign_pointer(rcu_torture_current, rp);
806 		smp_wmb(); /* Mods to old_rp must follow rcu_assign_pointer() */
807 		if (old_rp) {
808 			i = old_rp->rtort_pipe_count;
809 			if (i > RCU_TORTURE_PIPE_LEN)
810 				i = RCU_TORTURE_PIPE_LEN;
811 			atomic_inc(&rcu_torture_wcount[i]);
812 			old_rp->rtort_pipe_count++;
813 			switch (synctype[torture_random(&rand) % nsynctypes]) {
814 			case RTWS_DEF_FREE:
815 				rcu_torture_writer_state = RTWS_DEF_FREE;
816 				cur_ops->deferred_free(old_rp);
817 				break;
818 			case RTWS_EXP_SYNC:
819 				rcu_torture_writer_state = RTWS_EXP_SYNC;
820 				cur_ops->exp_sync();
821 				rcu_torture_pipe_update(old_rp);
822 				break;
823 			case RTWS_COND_GET:
824 				rcu_torture_writer_state = RTWS_COND_GET;
825 				gp_snap = cur_ops->get_state();
826 				i = torture_random(&rand) % 16;
827 				if (i != 0)
828 					schedule_timeout_interruptible(i);
829 				udelay(torture_random(&rand) % 1000);
830 				rcu_torture_writer_state = RTWS_COND_SYNC;
831 				cur_ops->cond_sync(gp_snap);
832 				rcu_torture_pipe_update(old_rp);
833 				break;
834 			case RTWS_SYNC:
835 				rcu_torture_writer_state = RTWS_SYNC;
836 				cur_ops->sync();
837 				rcu_torture_pipe_update(old_rp);
838 				break;
839 			default:
840 				WARN_ON_ONCE(1);
841 				break;
842 			}
843 		}
844 		rcutorture_record_progress(++rcu_torture_current_version);
845 		rcu_torture_writer_state = RTWS_STUTTER;
846 		stutter_wait("rcu_torture_writer");
847 	} while (!torture_must_stop());
848 	rcu_torture_writer_state = RTWS_STOPPING;
849 	torture_kthread_stopping("rcu_torture_writer");
850 	return 0;
851 }
852 
853 /*
854  * RCU torture fake writer kthread.  Repeatedly calls sync, with a random
855  * delay between calls.
856  */
857 static int
858 rcu_torture_fakewriter(void *arg)
859 {
860 	DEFINE_TORTURE_RANDOM(rand);
861 
862 	VERBOSE_TOROUT_STRING("rcu_torture_fakewriter task started");
863 	set_user_nice(current, MAX_NICE);
864 
865 	do {
866 		schedule_timeout_uninterruptible(1 + torture_random(&rand)%10);
867 		udelay(torture_random(&rand) & 0x3ff);
868 		if (cur_ops->cb_barrier != NULL &&
869 		    torture_random(&rand) % (nfakewriters * 8) == 0) {
870 			cur_ops->cb_barrier();
871 		} else if (gp_normal == gp_exp) {
872 			if (torture_random(&rand) & 0x80)
873 				cur_ops->sync();
874 			else
875 				cur_ops->exp_sync();
876 		} else if (gp_normal) {
877 			cur_ops->sync();
878 		} else {
879 			cur_ops->exp_sync();
880 		}
881 		stutter_wait("rcu_torture_fakewriter");
882 	} while (!torture_must_stop());
883 
884 	torture_kthread_stopping("rcu_torture_fakewriter");
885 	return 0;
886 }
887 
888 static void rcutorture_trace_dump(void)
889 {
890 	static atomic_t beenhere = ATOMIC_INIT(0);
891 
892 	if (atomic_read(&beenhere))
893 		return;
894 	if (atomic_xchg(&beenhere, 1) != 0)
895 		return;
896 	ftrace_dump(DUMP_ALL);
897 }
898 
899 /*
900  * RCU torture reader from timer handler.  Dereferences rcu_torture_current,
901  * incrementing the corresponding element of the pipeline array.  The
902  * counter in the element should never be greater than 1, otherwise, the
903  * RCU implementation is broken.
904  */
905 static void rcu_torture_timer(unsigned long unused)
906 {
907 	int idx;
908 	int completed;
909 	int completed_end;
910 	static DEFINE_TORTURE_RANDOM(rand);
911 	static DEFINE_SPINLOCK(rand_lock);
912 	struct rcu_torture *p;
913 	int pipe_count;
914 	unsigned long long ts;
915 
916 	idx = cur_ops->readlock();
917 	completed = cur_ops->completed();
918 	ts = rcu_trace_clock_local();
919 	p = rcu_dereference_check(rcu_torture_current,
920 				  rcu_read_lock_bh_held() ||
921 				  rcu_read_lock_sched_held() ||
922 				  srcu_read_lock_held(&srcu_ctl));
923 	if (p == NULL) {
924 		/* Leave because rcu_torture_writer is not yet underway */
925 		cur_ops->readunlock(idx);
926 		return;
927 	}
928 	if (p->rtort_mbtest == 0)
929 		atomic_inc(&n_rcu_torture_mberror);
930 	spin_lock(&rand_lock);
931 	cur_ops->read_delay(&rand);
932 	n_rcu_torture_timers++;
933 	spin_unlock(&rand_lock);
934 	preempt_disable();
935 	pipe_count = p->rtort_pipe_count;
936 	if (pipe_count > RCU_TORTURE_PIPE_LEN) {
937 		/* Should not happen, but... */
938 		pipe_count = RCU_TORTURE_PIPE_LEN;
939 	}
940 	completed_end = cur_ops->completed();
941 	if (pipe_count > 1) {
942 		do_trace_rcu_torture_read(cur_ops->name, &p->rtort_rcu, ts,
943 					  completed, completed_end);
944 		rcutorture_trace_dump();
945 	}
946 	__this_cpu_inc(rcu_torture_count[pipe_count]);
947 	completed = completed_end - completed;
948 	if (completed > RCU_TORTURE_PIPE_LEN) {
949 		/* Should not happen, but... */
950 		completed = RCU_TORTURE_PIPE_LEN;
951 	}
952 	__this_cpu_inc(rcu_torture_batch[completed]);
953 	preempt_enable();
954 	cur_ops->readunlock(idx);
955 }
956 
957 /*
958  * RCU torture reader kthread.  Repeatedly dereferences rcu_torture_current,
959  * incrementing the corresponding element of the pipeline array.  The
960  * counter in the element should never be greater than 1, otherwise, the
961  * RCU implementation is broken.
962  */
963 static int
964 rcu_torture_reader(void *arg)
965 {
966 	int completed;
967 	int completed_end;
968 	int idx;
969 	DEFINE_TORTURE_RANDOM(rand);
970 	struct rcu_torture *p;
971 	int pipe_count;
972 	struct timer_list t;
973 	unsigned long long ts;
974 
975 	VERBOSE_TOROUT_STRING("rcu_torture_reader task started");
976 	set_user_nice(current, MAX_NICE);
977 	if (irqreader && cur_ops->irq_capable)
978 		setup_timer_on_stack(&t, rcu_torture_timer, 0);
979 
980 	do {
981 		if (irqreader && cur_ops->irq_capable) {
982 			if (!timer_pending(&t))
983 				mod_timer(&t, jiffies + 1);
984 		}
985 		idx = cur_ops->readlock();
986 		completed = cur_ops->completed();
987 		ts = rcu_trace_clock_local();
988 		p = rcu_dereference_check(rcu_torture_current,
989 					  rcu_read_lock_bh_held() ||
990 					  rcu_read_lock_sched_held() ||
991 					  srcu_read_lock_held(&srcu_ctl));
992 		if (p == NULL) {
993 			/* Wait for rcu_torture_writer to get underway */
994 			cur_ops->readunlock(idx);
995 			schedule_timeout_interruptible(HZ);
996 			continue;
997 		}
998 		if (p->rtort_mbtest == 0)
999 			atomic_inc(&n_rcu_torture_mberror);
1000 		cur_ops->read_delay(&rand);
1001 		preempt_disable();
1002 		pipe_count = p->rtort_pipe_count;
1003 		if (pipe_count > RCU_TORTURE_PIPE_LEN) {
1004 			/* Should not happen, but... */
1005 			pipe_count = RCU_TORTURE_PIPE_LEN;
1006 		}
1007 		completed_end = cur_ops->completed();
1008 		if (pipe_count > 1) {
1009 			do_trace_rcu_torture_read(cur_ops->name, &p->rtort_rcu,
1010 						  ts, completed, completed_end);
1011 			rcutorture_trace_dump();
1012 		}
1013 		__this_cpu_inc(rcu_torture_count[pipe_count]);
1014 		completed = completed_end - completed;
1015 		if (completed > RCU_TORTURE_PIPE_LEN) {
1016 			/* Should not happen, but... */
1017 			completed = RCU_TORTURE_PIPE_LEN;
1018 		}
1019 		__this_cpu_inc(rcu_torture_batch[completed]);
1020 		preempt_enable();
1021 		cur_ops->readunlock(idx);
1022 		cond_resched();
1023 		stutter_wait("rcu_torture_reader");
1024 	} while (!torture_must_stop());
1025 	if (irqreader && cur_ops->irq_capable) {
1026 		del_timer_sync(&t);
1027 		destroy_timer_on_stack(&t);
1028 	}
1029 	torture_kthread_stopping("rcu_torture_reader");
1030 	return 0;
1031 }
1032 
1033 /*
1034  * Create an RCU-torture statistics message in the specified buffer.
1035  */
1036 static void
1037 rcu_torture_printk(char *page)
1038 {
1039 	int cpu;
1040 	int i;
1041 	long pipesummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
1042 	long batchsummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
1043 	static unsigned long rtcv_snap = ULONG_MAX;
1044 
1045 	for_each_possible_cpu(cpu) {
1046 		for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
1047 			pipesummary[i] += per_cpu(rcu_torture_count, cpu)[i];
1048 			batchsummary[i] += per_cpu(rcu_torture_batch, cpu)[i];
1049 		}
1050 	}
1051 	for (i = RCU_TORTURE_PIPE_LEN - 1; i >= 0; i--) {
1052 		if (pipesummary[i] != 0)
1053 			break;
1054 	}
1055 	page += sprintf(page, "%s%s ", torture_type, TORTURE_FLAG);
1056 	page += sprintf(page,
1057 		       "rtc: %p ver: %lu tfle: %d rta: %d rtaf: %d rtf: %d ",
1058 		       rcu_torture_current,
1059 		       rcu_torture_current_version,
1060 		       list_empty(&rcu_torture_freelist),
1061 		       atomic_read(&n_rcu_torture_alloc),
1062 		       atomic_read(&n_rcu_torture_alloc_fail),
1063 		       atomic_read(&n_rcu_torture_free));
1064 	page += sprintf(page, "rtmbe: %d rtbke: %ld rtbre: %ld ",
1065 		       atomic_read(&n_rcu_torture_mberror),
1066 		       n_rcu_torture_boost_ktrerror,
1067 		       n_rcu_torture_boost_rterror);
1068 	page += sprintf(page, "rtbf: %ld rtb: %ld nt: %ld ",
1069 		       n_rcu_torture_boost_failure,
1070 		       n_rcu_torture_boosts,
1071 		       n_rcu_torture_timers);
1072 	page = torture_onoff_stats(page);
1073 	page += sprintf(page, "barrier: %ld/%ld:%ld",
1074 		       n_barrier_successes,
1075 		       n_barrier_attempts,
1076 		       n_rcu_torture_barrier_error);
1077 	page += sprintf(page, "\n%s%s ", torture_type, TORTURE_FLAG);
1078 	if (atomic_read(&n_rcu_torture_mberror) != 0 ||
1079 	    n_rcu_torture_barrier_error != 0 ||
1080 	    n_rcu_torture_boost_ktrerror != 0 ||
1081 	    n_rcu_torture_boost_rterror != 0 ||
1082 	    n_rcu_torture_boost_failure != 0 ||
1083 	    i > 1) {
1084 		page += sprintf(page, "!!! ");
1085 		atomic_inc(&n_rcu_torture_error);
1086 		WARN_ON_ONCE(1);
1087 	}
1088 	page += sprintf(page, "Reader Pipe: ");
1089 	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
1090 		page += sprintf(page, " %ld", pipesummary[i]);
1091 	page += sprintf(page, "\n%s%s ", torture_type, TORTURE_FLAG);
1092 	page += sprintf(page, "Reader Batch: ");
1093 	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
1094 		page += sprintf(page, " %ld", batchsummary[i]);
1095 	page += sprintf(page, "\n%s%s ", torture_type, TORTURE_FLAG);
1096 	page += sprintf(page, "Free-Block Circulation: ");
1097 	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
1098 		page += sprintf(page, " %d",
1099 			       atomic_read(&rcu_torture_wcount[i]));
1100 	}
1101 	page += sprintf(page, "\n");
1102 	if (cur_ops->stats)
1103 		cur_ops->stats(page);
1104 	if (rtcv_snap == rcu_torture_current_version &&
1105 	    rcu_torture_current != NULL) {
1106 		int __maybe_unused flags;
1107 		unsigned long __maybe_unused gpnum;
1108 		unsigned long __maybe_unused completed;
1109 
1110 		rcutorture_get_gp_data(cur_ops->ttype,
1111 				       &flags, &gpnum, &completed);
1112 		page += sprintf(page,
1113 				"??? Writer stall state %d g%lu c%lu f%#x\n",
1114 				rcu_torture_writer_state,
1115 				gpnum, completed, flags);
1116 		show_rcu_gp_kthreads();
1117 		rcutorture_trace_dump();
1118 	}
1119 	rtcv_snap = rcu_torture_current_version;
1120 }
1121 
1122 /*
1123  * Print torture statistics.  Caller must ensure that there is only
1124  * one call to this function at a given time!!!  This is normally
1125  * accomplished by relying on the module system to only have one copy
1126  * of the module loaded, and then by giving the rcu_torture_stats
1127  * kthread full control (or the init/cleanup functions when rcu_torture_stats
1128  * thread is not running).
1129  */
1130 static void
1131 rcu_torture_stats_print(void)
1132 {
1133 	int size = nr_cpu_ids * 200 + 8192;
1134 	char *buf;
1135 
1136 	buf = kmalloc(size, GFP_KERNEL);
1137 	if (!buf) {
1138 		pr_err("rcu-torture: Out of memory, need: %d", size);
1139 		return;
1140 	}
1141 	rcu_torture_printk(buf);
1142 	pr_alert("%s", buf);
1143 	kfree(buf);
1144 }
1145 
1146 /*
1147  * Periodically prints torture statistics, if periodic statistics printing
1148  * was specified via the stat_interval module parameter.
1149  */
1150 static int
1151 rcu_torture_stats(void *arg)
1152 {
1153 	VERBOSE_TOROUT_STRING("rcu_torture_stats task started");
1154 	do {
1155 		schedule_timeout_interruptible(stat_interval * HZ);
1156 		rcu_torture_stats_print();
1157 		torture_shutdown_absorb("rcu_torture_stats");
1158 	} while (!torture_must_stop());
1159 	torture_kthread_stopping("rcu_torture_stats");
1160 	return 0;
1161 }
1162 
1163 static inline void
1164 rcu_torture_print_module_parms(struct rcu_torture_ops *cur_ops, const char *tag)
1165 {
1166 	pr_alert("%s" TORTURE_FLAG
1167 		 "--- %s: nreaders=%d nfakewriters=%d "
1168 		 "stat_interval=%d verbose=%d test_no_idle_hz=%d "
1169 		 "shuffle_interval=%d stutter=%d irqreader=%d "
1170 		 "fqs_duration=%d fqs_holdoff=%d fqs_stutter=%d "
1171 		 "test_boost=%d/%d test_boost_interval=%d "
1172 		 "test_boost_duration=%d shutdown_secs=%d "
1173 		 "stall_cpu=%d stall_cpu_holdoff=%d "
1174 		 "n_barrier_cbs=%d "
1175 		 "onoff_interval=%d onoff_holdoff=%d\n",
1176 		 torture_type, tag, nrealreaders, nfakewriters,
1177 		 stat_interval, verbose, test_no_idle_hz, shuffle_interval,
1178 		 stutter, irqreader, fqs_duration, fqs_holdoff, fqs_stutter,
1179 		 test_boost, cur_ops->can_boost,
1180 		 test_boost_interval, test_boost_duration, shutdown_secs,
1181 		 stall_cpu, stall_cpu_holdoff,
1182 		 n_barrier_cbs,
1183 		 onoff_interval, onoff_holdoff);
1184 }
1185 
1186 static void rcutorture_booster_cleanup(int cpu)
1187 {
1188 	struct task_struct *t;
1189 
1190 	if (boost_tasks[cpu] == NULL)
1191 		return;
1192 	mutex_lock(&boost_mutex);
1193 	t = boost_tasks[cpu];
1194 	boost_tasks[cpu] = NULL;
1195 	mutex_unlock(&boost_mutex);
1196 
1197 	/* This must be outside of the mutex, otherwise deadlock! */
1198 	torture_stop_kthread(rcu_torture_boost, t);
1199 }
1200 
1201 static int rcutorture_booster_init(int cpu)
1202 {
1203 	int retval;
1204 
1205 	if (boost_tasks[cpu] != NULL)
1206 		return 0;  /* Already created, nothing more to do. */
1207 
1208 	/* Don't allow time recalculation while creating a new task. */
1209 	mutex_lock(&boost_mutex);
1210 	VERBOSE_TOROUT_STRING("Creating rcu_torture_boost task");
1211 	boost_tasks[cpu] = kthread_create_on_node(rcu_torture_boost, NULL,
1212 						  cpu_to_node(cpu),
1213 						  "rcu_torture_boost");
1214 	if (IS_ERR(boost_tasks[cpu])) {
1215 		retval = PTR_ERR(boost_tasks[cpu]);
1216 		VERBOSE_TOROUT_STRING("rcu_torture_boost task create failed");
1217 		n_rcu_torture_boost_ktrerror++;
1218 		boost_tasks[cpu] = NULL;
1219 		mutex_unlock(&boost_mutex);
1220 		return retval;
1221 	}
1222 	kthread_bind(boost_tasks[cpu], cpu);
1223 	wake_up_process(boost_tasks[cpu]);
1224 	mutex_unlock(&boost_mutex);
1225 	return 0;
1226 }
1227 
1228 /*
1229  * CPU-stall kthread.  It waits as specified by stall_cpu_holdoff, then
1230  * induces a CPU stall for the time specified by stall_cpu.
1231  */
1232 static int rcu_torture_stall(void *args)
1233 {
1234 	unsigned long stop_at;
1235 
1236 	VERBOSE_TOROUT_STRING("rcu_torture_stall task started");
1237 	if (stall_cpu_holdoff > 0) {
1238 		VERBOSE_TOROUT_STRING("rcu_torture_stall begin holdoff");
1239 		schedule_timeout_interruptible(stall_cpu_holdoff * HZ);
1240 		VERBOSE_TOROUT_STRING("rcu_torture_stall end holdoff");
1241 	}
1242 	if (!kthread_should_stop()) {
1243 		stop_at = get_seconds() + stall_cpu;
1244 		/* RCU CPU stall is expected behavior in following code. */
1245 		pr_alert("rcu_torture_stall start.\n");
1246 		rcu_read_lock();
1247 		preempt_disable();
1248 		while (ULONG_CMP_LT(get_seconds(), stop_at))
1249 			continue;  /* Induce RCU CPU stall warning. */
1250 		preempt_enable();
1251 		rcu_read_unlock();
1252 		pr_alert("rcu_torture_stall end.\n");
1253 	}
1254 	torture_shutdown_absorb("rcu_torture_stall");
1255 	while (!kthread_should_stop())
1256 		schedule_timeout_interruptible(10 * HZ);
1257 	return 0;
1258 }
1259 
1260 /* Spawn CPU-stall kthread, if stall_cpu specified. */
1261 static int __init rcu_torture_stall_init(void)
1262 {
1263 	if (stall_cpu <= 0)
1264 		return 0;
1265 	return torture_create_kthread(rcu_torture_stall, NULL, stall_task);
1266 }
1267 
1268 /* Callback function for RCU barrier testing. */
1269 static void rcu_torture_barrier_cbf(struct rcu_head *rcu)
1270 {
1271 	atomic_inc(&barrier_cbs_invoked);
1272 }
1273 
1274 /* kthread function to register callbacks used to test RCU barriers. */
1275 static int rcu_torture_barrier_cbs(void *arg)
1276 {
1277 	long myid = (long)arg;
1278 	bool lastphase = 0;
1279 	bool newphase;
1280 	struct rcu_head rcu;
1281 
1282 	init_rcu_head_on_stack(&rcu);
1283 	VERBOSE_TOROUT_STRING("rcu_torture_barrier_cbs task started");
1284 	set_user_nice(current, MAX_NICE);
1285 	do {
1286 		wait_event(barrier_cbs_wq[myid],
1287 			   (newphase =
1288 			    ACCESS_ONCE(barrier_phase)) != lastphase ||
1289 			   torture_must_stop());
1290 		lastphase = newphase;
1291 		smp_mb(); /* ensure barrier_phase load before ->call(). */
1292 		if (torture_must_stop())
1293 			break;
1294 		cur_ops->call(&rcu, rcu_torture_barrier_cbf);
1295 		if (atomic_dec_and_test(&barrier_cbs_count))
1296 			wake_up(&barrier_wq);
1297 	} while (!torture_must_stop());
1298 	cur_ops->cb_barrier();
1299 	destroy_rcu_head_on_stack(&rcu);
1300 	torture_kthread_stopping("rcu_torture_barrier_cbs");
1301 	return 0;
1302 }
1303 
1304 /* kthread function to drive and coordinate RCU barrier testing. */
1305 static int rcu_torture_barrier(void *arg)
1306 {
1307 	int i;
1308 
1309 	VERBOSE_TOROUT_STRING("rcu_torture_barrier task starting");
1310 	do {
1311 		atomic_set(&barrier_cbs_invoked, 0);
1312 		atomic_set(&barrier_cbs_count, n_barrier_cbs);
1313 		smp_mb(); /* Ensure barrier_phase after prior assignments. */
1314 		barrier_phase = !barrier_phase;
1315 		for (i = 0; i < n_barrier_cbs; i++)
1316 			wake_up(&barrier_cbs_wq[i]);
1317 		wait_event(barrier_wq,
1318 			   atomic_read(&barrier_cbs_count) == 0 ||
1319 			   torture_must_stop());
1320 		if (torture_must_stop())
1321 			break;
1322 		n_barrier_attempts++;
1323 		cur_ops->cb_barrier(); /* Implies smp_mb() for wait_event(). */
1324 		if (atomic_read(&barrier_cbs_invoked) != n_barrier_cbs) {
1325 			n_rcu_torture_barrier_error++;
1326 			WARN_ON_ONCE(1);
1327 		}
1328 		n_barrier_successes++;
1329 		schedule_timeout_interruptible(HZ / 10);
1330 	} while (!torture_must_stop());
1331 	torture_kthread_stopping("rcu_torture_barrier");
1332 	return 0;
1333 }
1334 
1335 /* Initialize RCU barrier testing. */
1336 static int rcu_torture_barrier_init(void)
1337 {
1338 	int i;
1339 	int ret;
1340 
1341 	if (n_barrier_cbs == 0)
1342 		return 0;
1343 	if (cur_ops->call == NULL || cur_ops->cb_barrier == NULL) {
1344 		pr_alert("%s" TORTURE_FLAG
1345 			 " Call or barrier ops missing for %s,\n",
1346 			 torture_type, cur_ops->name);
1347 		pr_alert("%s" TORTURE_FLAG
1348 			 " RCU barrier testing omitted from run.\n",
1349 			 torture_type);
1350 		return 0;
1351 	}
1352 	atomic_set(&barrier_cbs_count, 0);
1353 	atomic_set(&barrier_cbs_invoked, 0);
1354 	barrier_cbs_tasks =
1355 		kzalloc(n_barrier_cbs * sizeof(barrier_cbs_tasks[0]),
1356 			GFP_KERNEL);
1357 	barrier_cbs_wq =
1358 		kzalloc(n_barrier_cbs * sizeof(barrier_cbs_wq[0]),
1359 			GFP_KERNEL);
1360 	if (barrier_cbs_tasks == NULL || !barrier_cbs_wq)
1361 		return -ENOMEM;
1362 	for (i = 0; i < n_barrier_cbs; i++) {
1363 		init_waitqueue_head(&barrier_cbs_wq[i]);
1364 		ret = torture_create_kthread(rcu_torture_barrier_cbs,
1365 					     (void *)(long)i,
1366 					     barrier_cbs_tasks[i]);
1367 		if (ret)
1368 			return ret;
1369 	}
1370 	return torture_create_kthread(rcu_torture_barrier, NULL, barrier_task);
1371 }
1372 
1373 /* Clean up after RCU barrier testing. */
1374 static void rcu_torture_barrier_cleanup(void)
1375 {
1376 	int i;
1377 
1378 	torture_stop_kthread(rcu_torture_barrier, barrier_task);
1379 	if (barrier_cbs_tasks != NULL) {
1380 		for (i = 0; i < n_barrier_cbs; i++)
1381 			torture_stop_kthread(rcu_torture_barrier_cbs,
1382 					     barrier_cbs_tasks[i]);
1383 		kfree(barrier_cbs_tasks);
1384 		barrier_cbs_tasks = NULL;
1385 	}
1386 	if (barrier_cbs_wq != NULL) {
1387 		kfree(barrier_cbs_wq);
1388 		barrier_cbs_wq = NULL;
1389 	}
1390 }
1391 
1392 static int rcutorture_cpu_notify(struct notifier_block *self,
1393 				 unsigned long action, void *hcpu)
1394 {
1395 	long cpu = (long)hcpu;
1396 
1397 	switch (action) {
1398 	case CPU_ONLINE:
1399 	case CPU_DOWN_FAILED:
1400 		(void)rcutorture_booster_init(cpu);
1401 		break;
1402 	case CPU_DOWN_PREPARE:
1403 		rcutorture_booster_cleanup(cpu);
1404 		break;
1405 	default:
1406 		break;
1407 	}
1408 	return NOTIFY_OK;
1409 }
1410 
1411 static struct notifier_block rcutorture_cpu_nb = {
1412 	.notifier_call = rcutorture_cpu_notify,
1413 };
1414 
1415 static void
1416 rcu_torture_cleanup(void)
1417 {
1418 	int i;
1419 
1420 	rcutorture_record_test_transition();
1421 	if (torture_cleanup()) {
1422 		if (cur_ops->cb_barrier != NULL)
1423 			cur_ops->cb_barrier();
1424 		return;
1425 	}
1426 
1427 	rcu_torture_barrier_cleanup();
1428 	torture_stop_kthread(rcu_torture_stall, stall_task);
1429 	torture_stop_kthread(rcu_torture_writer, writer_task);
1430 
1431 	if (reader_tasks) {
1432 		for (i = 0; i < nrealreaders; i++)
1433 			torture_stop_kthread(rcu_torture_reader,
1434 					     reader_tasks[i]);
1435 		kfree(reader_tasks);
1436 	}
1437 	rcu_torture_current = NULL;
1438 
1439 	if (fakewriter_tasks) {
1440 		for (i = 0; i < nfakewriters; i++) {
1441 			torture_stop_kthread(rcu_torture_fakewriter,
1442 					     fakewriter_tasks[i]);
1443 		}
1444 		kfree(fakewriter_tasks);
1445 		fakewriter_tasks = NULL;
1446 	}
1447 
1448 	torture_stop_kthread(rcu_torture_stats, stats_task);
1449 	torture_stop_kthread(rcu_torture_fqs, fqs_task);
1450 	if ((test_boost == 1 && cur_ops->can_boost) ||
1451 	    test_boost == 2) {
1452 		unregister_cpu_notifier(&rcutorture_cpu_nb);
1453 		for_each_possible_cpu(i)
1454 			rcutorture_booster_cleanup(i);
1455 	}
1456 
1457 	/* Wait for all RCU callbacks to fire.  */
1458 
1459 	if (cur_ops->cb_barrier != NULL)
1460 		cur_ops->cb_barrier();
1461 
1462 	rcu_torture_stats_print();  /* -After- the stats thread is stopped! */
1463 
1464 	if (atomic_read(&n_rcu_torture_error) || n_rcu_torture_barrier_error)
1465 		rcu_torture_print_module_parms(cur_ops, "End of test: FAILURE");
1466 	else if (torture_onoff_failures())
1467 		rcu_torture_print_module_parms(cur_ops,
1468 					       "End of test: RCU_HOTPLUG");
1469 	else
1470 		rcu_torture_print_module_parms(cur_ops, "End of test: SUCCESS");
1471 }
1472 
1473 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
1474 static void rcu_torture_leak_cb(struct rcu_head *rhp)
1475 {
1476 }
1477 
1478 static void rcu_torture_err_cb(struct rcu_head *rhp)
1479 {
1480 	/*
1481 	 * This -might- happen due to race conditions, but is unlikely.
1482 	 * The scenario that leads to this happening is that the
1483 	 * first of the pair of duplicate callbacks is queued,
1484 	 * someone else starts a grace period that includes that
1485 	 * callback, then the second of the pair must wait for the
1486 	 * next grace period.  Unlikely, but can happen.  If it
1487 	 * does happen, the debug-objects subsystem won't have splatted.
1488 	 */
1489 	pr_alert("rcutorture: duplicated callback was invoked.\n");
1490 }
1491 #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
1492 
1493 /*
1494  * Verify that double-free causes debug-objects to complain, but only
1495  * if CONFIG_DEBUG_OBJECTS_RCU_HEAD=y.  Otherwise, say that the test
1496  * cannot be carried out.
1497  */
1498 static void rcu_test_debug_objects(void)
1499 {
1500 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
1501 	struct rcu_head rh1;
1502 	struct rcu_head rh2;
1503 
1504 	init_rcu_head_on_stack(&rh1);
1505 	init_rcu_head_on_stack(&rh2);
1506 	pr_alert("rcutorture: WARN: Duplicate call_rcu() test starting.\n");
1507 
1508 	/* Try to queue the rh2 pair of callbacks for the same grace period. */
1509 	preempt_disable(); /* Prevent preemption from interrupting test. */
1510 	rcu_read_lock(); /* Make it impossible to finish a grace period. */
1511 	call_rcu(&rh1, rcu_torture_leak_cb); /* Start grace period. */
1512 	local_irq_disable(); /* Make it harder to start a new grace period. */
1513 	call_rcu(&rh2, rcu_torture_leak_cb);
1514 	call_rcu(&rh2, rcu_torture_err_cb); /* Duplicate callback. */
1515 	local_irq_enable();
1516 	rcu_read_unlock();
1517 	preempt_enable();
1518 
1519 	/* Wait for them all to get done so we can safely return. */
1520 	rcu_barrier();
1521 	pr_alert("rcutorture: WARN: Duplicate call_rcu() test complete.\n");
1522 	destroy_rcu_head_on_stack(&rh1);
1523 	destroy_rcu_head_on_stack(&rh2);
1524 #else /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
1525 	pr_alert("rcutorture: !CONFIG_DEBUG_OBJECTS_RCU_HEAD, not testing duplicate call_rcu()\n");
1526 #endif /* #else #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
1527 }
1528 
1529 static int __init
1530 rcu_torture_init(void)
1531 {
1532 	int i;
1533 	int cpu;
1534 	int firsterr = 0;
1535 	static struct rcu_torture_ops *torture_ops[] = {
1536 		&rcu_ops, &rcu_bh_ops, &rcu_busted_ops, &srcu_ops, &sched_ops,
1537 	};
1538 
1539 	if (!torture_init_begin(torture_type, verbose, &rcutorture_runnable))
1540 		return -EBUSY;
1541 
1542 	/* Process args and tell the world that the torturer is on the job. */
1543 	for (i = 0; i < ARRAY_SIZE(torture_ops); i++) {
1544 		cur_ops = torture_ops[i];
1545 		if (strcmp(torture_type, cur_ops->name) == 0)
1546 			break;
1547 	}
1548 	if (i == ARRAY_SIZE(torture_ops)) {
1549 		pr_alert("rcu-torture: invalid torture type: \"%s\"\n",
1550 			 torture_type);
1551 		pr_alert("rcu-torture types:");
1552 		for (i = 0; i < ARRAY_SIZE(torture_ops); i++)
1553 			pr_alert(" %s", torture_ops[i]->name);
1554 		pr_alert("\n");
1555 		torture_init_end();
1556 		return -EINVAL;
1557 	}
1558 	if (cur_ops->fqs == NULL && fqs_duration != 0) {
1559 		pr_alert("rcu-torture: ->fqs NULL and non-zero fqs_duration, fqs disabled.\n");
1560 		fqs_duration = 0;
1561 	}
1562 	if (cur_ops->init)
1563 		cur_ops->init(); /* no "goto unwind" prior to this point!!! */
1564 
1565 	if (nreaders >= 0) {
1566 		nrealreaders = nreaders;
1567 	} else {
1568 		nrealreaders = num_online_cpus() - 1;
1569 		if (nrealreaders <= 0)
1570 			nrealreaders = 1;
1571 	}
1572 	rcu_torture_print_module_parms(cur_ops, "Start of test");
1573 
1574 	/* Set up the freelist. */
1575 
1576 	INIT_LIST_HEAD(&rcu_torture_freelist);
1577 	for (i = 0; i < ARRAY_SIZE(rcu_tortures); i++) {
1578 		rcu_tortures[i].rtort_mbtest = 0;
1579 		list_add_tail(&rcu_tortures[i].rtort_free,
1580 			      &rcu_torture_freelist);
1581 	}
1582 
1583 	/* Initialize the statistics so that each run gets its own numbers. */
1584 
1585 	rcu_torture_current = NULL;
1586 	rcu_torture_current_version = 0;
1587 	atomic_set(&n_rcu_torture_alloc, 0);
1588 	atomic_set(&n_rcu_torture_alloc_fail, 0);
1589 	atomic_set(&n_rcu_torture_free, 0);
1590 	atomic_set(&n_rcu_torture_mberror, 0);
1591 	atomic_set(&n_rcu_torture_error, 0);
1592 	n_rcu_torture_barrier_error = 0;
1593 	n_rcu_torture_boost_ktrerror = 0;
1594 	n_rcu_torture_boost_rterror = 0;
1595 	n_rcu_torture_boost_failure = 0;
1596 	n_rcu_torture_boosts = 0;
1597 	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
1598 		atomic_set(&rcu_torture_wcount[i], 0);
1599 	for_each_possible_cpu(cpu) {
1600 		for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
1601 			per_cpu(rcu_torture_count, cpu)[i] = 0;
1602 			per_cpu(rcu_torture_batch, cpu)[i] = 0;
1603 		}
1604 	}
1605 
1606 	/* Start up the kthreads. */
1607 
1608 	firsterr = torture_create_kthread(rcu_torture_writer, NULL,
1609 					  writer_task);
1610 	if (firsterr)
1611 		goto unwind;
1612 	fakewriter_tasks = kzalloc(nfakewriters * sizeof(fakewriter_tasks[0]),
1613 				   GFP_KERNEL);
1614 	if (fakewriter_tasks == NULL) {
1615 		VERBOSE_TOROUT_ERRSTRING("out of memory");
1616 		firsterr = -ENOMEM;
1617 		goto unwind;
1618 	}
1619 	for (i = 0; i < nfakewriters; i++) {
1620 		firsterr = torture_create_kthread(rcu_torture_fakewriter,
1621 						  NULL, fakewriter_tasks[i]);
1622 		if (firsterr)
1623 			goto unwind;
1624 	}
1625 	reader_tasks = kzalloc(nrealreaders * sizeof(reader_tasks[0]),
1626 			       GFP_KERNEL);
1627 	if (reader_tasks == NULL) {
1628 		VERBOSE_TOROUT_ERRSTRING("out of memory");
1629 		firsterr = -ENOMEM;
1630 		goto unwind;
1631 	}
1632 	for (i = 0; i < nrealreaders; i++) {
1633 		firsterr = torture_create_kthread(rcu_torture_reader, NULL,
1634 						  reader_tasks[i]);
1635 		if (firsterr)
1636 			goto unwind;
1637 	}
1638 	if (stat_interval > 0) {
1639 		firsterr = torture_create_kthread(rcu_torture_stats, NULL,
1640 						  stats_task);
1641 		if (firsterr)
1642 			goto unwind;
1643 	}
1644 	if (test_no_idle_hz) {
1645 		firsterr = torture_shuffle_init(shuffle_interval * HZ);
1646 		if (firsterr)
1647 			goto unwind;
1648 	}
1649 	if (stutter < 0)
1650 		stutter = 0;
1651 	if (stutter) {
1652 		firsterr = torture_stutter_init(stutter * HZ);
1653 		if (firsterr)
1654 			goto unwind;
1655 	}
1656 	if (fqs_duration < 0)
1657 		fqs_duration = 0;
1658 	if (fqs_duration) {
1659 		/* Create the fqs thread */
1660 		firsterr = torture_create_kthread(rcu_torture_fqs, NULL,
1661 						  fqs_task);
1662 		if (firsterr)
1663 			goto unwind;
1664 	}
1665 	if (test_boost_interval < 1)
1666 		test_boost_interval = 1;
1667 	if (test_boost_duration < 2)
1668 		test_boost_duration = 2;
1669 	if ((test_boost == 1 && cur_ops->can_boost) ||
1670 	    test_boost == 2) {
1671 
1672 		boost_starttime = jiffies + test_boost_interval * HZ;
1673 		register_cpu_notifier(&rcutorture_cpu_nb);
1674 		for_each_possible_cpu(i) {
1675 			if (cpu_is_offline(i))
1676 				continue;  /* Heuristic: CPU can go offline. */
1677 			firsterr = rcutorture_booster_init(i);
1678 			if (firsterr)
1679 				goto unwind;
1680 		}
1681 	}
1682 	firsterr = torture_shutdown_init(shutdown_secs, rcu_torture_cleanup);
1683 	if (firsterr)
1684 		goto unwind;
1685 	firsterr = torture_onoff_init(onoff_holdoff * HZ, onoff_interval * HZ);
1686 	if (firsterr)
1687 		goto unwind;
1688 	firsterr = rcu_torture_stall_init();
1689 	if (firsterr)
1690 		goto unwind;
1691 	firsterr = rcu_torture_barrier_init();
1692 	if (firsterr)
1693 		goto unwind;
1694 	if (object_debug)
1695 		rcu_test_debug_objects();
1696 	rcutorture_record_test_transition();
1697 	torture_init_end();
1698 	return 0;
1699 
1700 unwind:
1701 	torture_init_end();
1702 	rcu_torture_cleanup();
1703 	return firsterr;
1704 }
1705 
1706 module_init(rcu_torture_init);
1707 module_exit(rcu_torture_cleanup);
1708