xref: /linux/kernel/rcu/rcutorture.c (revision a7f7f6248d9740d710fd6bd190293fe5e16410ac)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * Read-Copy Update module-based torture test facility
4  *
5  * Copyright (C) IBM Corporation, 2005, 2006
6  *
7  * Authors: Paul E. McKenney <paulmck@linux.ibm.com>
8  *	  Josh Triplett <josh@joshtriplett.org>
9  *
10  * See also:  Documentation/RCU/torture.txt
11  */
12 
13 #define pr_fmt(fmt) fmt
14 
15 #include <linux/types.h>
16 #include <linux/kernel.h>
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/kthread.h>
20 #include <linux/err.h>
21 #include <linux/spinlock.h>
22 #include <linux/smp.h>
23 #include <linux/rcupdate_wait.h>
24 #include <linux/interrupt.h>
25 #include <linux/sched/signal.h>
26 #include <uapi/linux/sched/types.h>
27 #include <linux/atomic.h>
28 #include <linux/bitops.h>
29 #include <linux/completion.h>
30 #include <linux/moduleparam.h>
31 #include <linux/percpu.h>
32 #include <linux/notifier.h>
33 #include <linux/reboot.h>
34 #include <linux/freezer.h>
35 #include <linux/cpu.h>
36 #include <linux/delay.h>
37 #include <linux/stat.h>
38 #include <linux/srcu.h>
39 #include <linux/slab.h>
40 #include <linux/trace_clock.h>
41 #include <asm/byteorder.h>
42 #include <linux/torture.h>
43 #include <linux/vmalloc.h>
44 #include <linux/sched/debug.h>
45 #include <linux/sched/sysctl.h>
46 #include <linux/oom.h>
47 #include <linux/tick.h>
48 #include <linux/rcupdate_trace.h>
49 
50 #include "rcu.h"
51 
52 MODULE_LICENSE("GPL");
53 MODULE_AUTHOR("Paul E. McKenney <paulmck@linux.ibm.com> and Josh Triplett <josh@joshtriplett.org>");
54 
55 #ifndef data_race
56 #define data_race(expr)							\
57 	({								\
58 		expr;							\
59 	})
60 #endif
61 #ifndef ASSERT_EXCLUSIVE_WRITER
62 #define ASSERT_EXCLUSIVE_WRITER(var) do { } while (0)
63 #endif
64 #ifndef ASSERT_EXCLUSIVE_ACCESS
65 #define ASSERT_EXCLUSIVE_ACCESS(var) do { } while (0)
66 #endif
67 
68 /* Bits for ->extendables field, extendables param, and related definitions. */
69 #define RCUTORTURE_RDR_SHIFT	 8	/* Put SRCU index in upper bits. */
70 #define RCUTORTURE_RDR_MASK	 ((1 << RCUTORTURE_RDR_SHIFT) - 1)
71 #define RCUTORTURE_RDR_BH	 0x01	/* Extend readers by disabling bh. */
72 #define RCUTORTURE_RDR_IRQ	 0x02	/*  ... disabling interrupts. */
73 #define RCUTORTURE_RDR_PREEMPT	 0x04	/*  ... disabling preemption. */
74 #define RCUTORTURE_RDR_RBH	 0x08	/*  ... rcu_read_lock_bh(). */
75 #define RCUTORTURE_RDR_SCHED	 0x10	/*  ... rcu_read_lock_sched(). */
76 #define RCUTORTURE_RDR_RCU	 0x20	/*  ... entering another RCU reader. */
77 #define RCUTORTURE_RDR_NBITS	 6	/* Number of bits defined above. */
78 #define RCUTORTURE_MAX_EXTEND	 \
79 	(RCUTORTURE_RDR_BH | RCUTORTURE_RDR_IRQ | RCUTORTURE_RDR_PREEMPT | \
80 	 RCUTORTURE_RDR_RBH | RCUTORTURE_RDR_SCHED)
81 #define RCUTORTURE_RDR_MAX_LOOPS 0x7	/* Maximum reader extensions. */
82 					/* Must be power of two minus one. */
83 #define RCUTORTURE_RDR_MAX_SEGS (RCUTORTURE_RDR_MAX_LOOPS + 3)
84 
85 torture_param(int, extendables, RCUTORTURE_MAX_EXTEND,
86 	      "Extend readers by disabling bh (1), irqs (2), or preempt (4)");
87 torture_param(int, fqs_duration, 0,
88 	      "Duration of fqs bursts (us), 0 to disable");
89 torture_param(int, fqs_holdoff, 0, "Holdoff time within fqs bursts (us)");
90 torture_param(int, fqs_stutter, 3, "Wait time between fqs bursts (s)");
91 torture_param(bool, fwd_progress, 1, "Test grace-period forward progress");
92 torture_param(int, fwd_progress_div, 4, "Fraction of CPU stall to wait");
93 torture_param(int, fwd_progress_holdoff, 60,
94 	      "Time between forward-progress tests (s)");
95 torture_param(bool, fwd_progress_need_resched, 1,
96 	      "Hide cond_resched() behind need_resched()");
97 torture_param(bool, gp_cond, false, "Use conditional/async GP wait primitives");
98 torture_param(bool, gp_exp, false, "Use expedited GP wait primitives");
99 torture_param(bool, gp_normal, false,
100 	     "Use normal (non-expedited) GP wait primitives");
101 torture_param(bool, gp_sync, false, "Use synchronous GP wait primitives");
102 torture_param(int, irqreader, 1, "Allow RCU readers from irq handlers");
103 torture_param(int, n_barrier_cbs, 0,
104 	     "# of callbacks/kthreads for barrier testing");
105 torture_param(int, nfakewriters, 4, "Number of RCU fake writer threads");
106 torture_param(int, nreaders, -1, "Number of RCU reader threads");
107 torture_param(int, object_debug, 0,
108 	     "Enable debug-object double call_rcu() testing");
109 torture_param(int, onoff_holdoff, 0, "Time after boot before CPU hotplugs (s)");
110 torture_param(int, onoff_interval, 0,
111 	     "Time between CPU hotplugs (jiffies), 0=disable");
112 torture_param(int, shuffle_interval, 3, "Number of seconds between shuffles");
113 torture_param(int, shutdown_secs, 0, "Shutdown time (s), <= zero to disable.");
114 torture_param(int, stall_cpu, 0, "Stall duration (s), zero to disable.");
115 torture_param(int, stall_cpu_holdoff, 10,
116 	     "Time to wait before starting stall (s).");
117 torture_param(int, stall_cpu_irqsoff, 0, "Disable interrupts while stalling.");
118 torture_param(int, stall_cpu_block, 0, "Sleep while stalling.");
119 torture_param(int, stall_gp_kthread, 0,
120 	      "Grace-period kthread stall duration (s).");
121 torture_param(int, stat_interval, 60,
122 	     "Number of seconds between stats printk()s");
123 torture_param(int, stutter, 5, "Number of seconds to run/halt test");
124 torture_param(int, test_boost, 1, "Test RCU prio boost: 0=no, 1=maybe, 2=yes.");
125 torture_param(int, test_boost_duration, 4,
126 	     "Duration of each boost test, seconds.");
127 torture_param(int, test_boost_interval, 7,
128 	     "Interval between boost tests, seconds.");
129 torture_param(bool, test_no_idle_hz, true,
130 	     "Test support for tickless idle CPUs");
131 torture_param(int, verbose, 1,
132 	     "Enable verbose debugging printk()s");
133 
134 static char *torture_type = "rcu";
135 module_param(torture_type, charp, 0444);
136 MODULE_PARM_DESC(torture_type, "Type of RCU to torture (rcu, srcu, ...)");
137 
138 static int nrealreaders;
139 static struct task_struct *writer_task;
140 static struct task_struct **fakewriter_tasks;
141 static struct task_struct **reader_tasks;
142 static struct task_struct *stats_task;
143 static struct task_struct *fqs_task;
144 static struct task_struct *boost_tasks[NR_CPUS];
145 static struct task_struct *stall_task;
146 static struct task_struct *fwd_prog_task;
147 static struct task_struct **barrier_cbs_tasks;
148 static struct task_struct *barrier_task;
149 
150 #define RCU_TORTURE_PIPE_LEN 10
151 
152 struct rcu_torture {
153 	struct rcu_head rtort_rcu;
154 	int rtort_pipe_count;
155 	struct list_head rtort_free;
156 	int rtort_mbtest;
157 };
158 
159 static LIST_HEAD(rcu_torture_freelist);
160 static struct rcu_torture __rcu *rcu_torture_current;
161 static unsigned long rcu_torture_current_version;
162 static struct rcu_torture rcu_tortures[10 * RCU_TORTURE_PIPE_LEN];
163 static DEFINE_SPINLOCK(rcu_torture_lock);
164 static DEFINE_PER_CPU(long [RCU_TORTURE_PIPE_LEN + 1], rcu_torture_count);
165 static DEFINE_PER_CPU(long [RCU_TORTURE_PIPE_LEN + 1], rcu_torture_batch);
166 static atomic_t rcu_torture_wcount[RCU_TORTURE_PIPE_LEN + 1];
167 static atomic_t n_rcu_torture_alloc;
168 static atomic_t n_rcu_torture_alloc_fail;
169 static atomic_t n_rcu_torture_free;
170 static atomic_t n_rcu_torture_mberror;
171 static atomic_t n_rcu_torture_error;
172 static long n_rcu_torture_barrier_error;
173 static long n_rcu_torture_boost_ktrerror;
174 static long n_rcu_torture_boost_rterror;
175 static long n_rcu_torture_boost_failure;
176 static long n_rcu_torture_boosts;
177 static atomic_long_t n_rcu_torture_timers;
178 static long n_barrier_attempts;
179 static long n_barrier_successes; /* did rcu_barrier test succeed? */
180 static struct list_head rcu_torture_removed;
181 static unsigned long shutdown_jiffies;
182 
183 static int rcu_torture_writer_state;
184 #define RTWS_FIXED_DELAY	0
185 #define RTWS_DELAY		1
186 #define RTWS_REPLACE		2
187 #define RTWS_DEF_FREE		3
188 #define RTWS_EXP_SYNC		4
189 #define RTWS_COND_GET		5
190 #define RTWS_COND_SYNC		6
191 #define RTWS_SYNC		7
192 #define RTWS_STUTTER		8
193 #define RTWS_STOPPING		9
194 static const char * const rcu_torture_writer_state_names[] = {
195 	"RTWS_FIXED_DELAY",
196 	"RTWS_DELAY",
197 	"RTWS_REPLACE",
198 	"RTWS_DEF_FREE",
199 	"RTWS_EXP_SYNC",
200 	"RTWS_COND_GET",
201 	"RTWS_COND_SYNC",
202 	"RTWS_SYNC",
203 	"RTWS_STUTTER",
204 	"RTWS_STOPPING",
205 };
206 
207 /* Record reader segment types and duration for first failing read. */
208 struct rt_read_seg {
209 	int rt_readstate;
210 	unsigned long rt_delay_jiffies;
211 	unsigned long rt_delay_ms;
212 	unsigned long rt_delay_us;
213 	bool rt_preempted;
214 };
215 static int err_segs_recorded;
216 static struct rt_read_seg err_segs[RCUTORTURE_RDR_MAX_SEGS];
217 static int rt_read_nsegs;
218 
219 static const char *rcu_torture_writer_state_getname(void)
220 {
221 	unsigned int i = READ_ONCE(rcu_torture_writer_state);
222 
223 	if (i >= ARRAY_SIZE(rcu_torture_writer_state_names))
224 		return "???";
225 	return rcu_torture_writer_state_names[i];
226 }
227 
228 #if defined(CONFIG_RCU_BOOST) && !defined(CONFIG_HOTPLUG_CPU)
229 #define rcu_can_boost() 1
230 #else /* #if defined(CONFIG_RCU_BOOST) && !defined(CONFIG_HOTPLUG_CPU) */
231 #define rcu_can_boost() 0
232 #endif /* #else #if defined(CONFIG_RCU_BOOST) && !defined(CONFIG_HOTPLUG_CPU) */
233 
234 #ifdef CONFIG_RCU_TRACE
235 static u64 notrace rcu_trace_clock_local(void)
236 {
237 	u64 ts = trace_clock_local();
238 
239 	(void)do_div(ts, NSEC_PER_USEC);
240 	return ts;
241 }
242 #else /* #ifdef CONFIG_RCU_TRACE */
243 static u64 notrace rcu_trace_clock_local(void)
244 {
245 	return 0ULL;
246 }
247 #endif /* #else #ifdef CONFIG_RCU_TRACE */
248 
249 /*
250  * Stop aggressive CPU-hog tests a bit before the end of the test in order
251  * to avoid interfering with test shutdown.
252  */
253 static bool shutdown_time_arrived(void)
254 {
255 	return shutdown_secs && time_after(jiffies, shutdown_jiffies - 30 * HZ);
256 }
257 
258 static unsigned long boost_starttime;	/* jiffies of next boost test start. */
259 static DEFINE_MUTEX(boost_mutex);	/* protect setting boost_starttime */
260 					/*  and boost task create/destroy. */
261 static atomic_t barrier_cbs_count;	/* Barrier callbacks registered. */
262 static bool barrier_phase;		/* Test phase. */
263 static atomic_t barrier_cbs_invoked;	/* Barrier callbacks invoked. */
264 static wait_queue_head_t *barrier_cbs_wq; /* Coordinate barrier testing. */
265 static DECLARE_WAIT_QUEUE_HEAD(barrier_wq);
266 
267 static bool rcu_fwd_cb_nodelay;		/* Short rcu_torture_delay() delays. */
268 
269 /*
270  * Allocate an element from the rcu_tortures pool.
271  */
272 static struct rcu_torture *
273 rcu_torture_alloc(void)
274 {
275 	struct list_head *p;
276 
277 	spin_lock_bh(&rcu_torture_lock);
278 	if (list_empty(&rcu_torture_freelist)) {
279 		atomic_inc(&n_rcu_torture_alloc_fail);
280 		spin_unlock_bh(&rcu_torture_lock);
281 		return NULL;
282 	}
283 	atomic_inc(&n_rcu_torture_alloc);
284 	p = rcu_torture_freelist.next;
285 	list_del_init(p);
286 	spin_unlock_bh(&rcu_torture_lock);
287 	return container_of(p, struct rcu_torture, rtort_free);
288 }
289 
290 /*
291  * Free an element to the rcu_tortures pool.
292  */
293 static void
294 rcu_torture_free(struct rcu_torture *p)
295 {
296 	atomic_inc(&n_rcu_torture_free);
297 	spin_lock_bh(&rcu_torture_lock);
298 	list_add_tail(&p->rtort_free, &rcu_torture_freelist);
299 	spin_unlock_bh(&rcu_torture_lock);
300 }
301 
302 /*
303  * Operations vector for selecting different types of tests.
304  */
305 
306 struct rcu_torture_ops {
307 	int ttype;
308 	void (*init)(void);
309 	void (*cleanup)(void);
310 	int (*readlock)(void);
311 	void (*read_delay)(struct torture_random_state *rrsp,
312 			   struct rt_read_seg *rtrsp);
313 	void (*readunlock)(int idx);
314 	unsigned long (*get_gp_seq)(void);
315 	unsigned long (*gp_diff)(unsigned long new, unsigned long old);
316 	void (*deferred_free)(struct rcu_torture *p);
317 	void (*sync)(void);
318 	void (*exp_sync)(void);
319 	unsigned long (*get_state)(void);
320 	void (*cond_sync)(unsigned long oldstate);
321 	call_rcu_func_t call;
322 	void (*cb_barrier)(void);
323 	void (*fqs)(void);
324 	void (*stats)(void);
325 	int (*stall_dur)(void);
326 	int irq_capable;
327 	int can_boost;
328 	int extendables;
329 	int slow_gps;
330 	const char *name;
331 };
332 
333 static struct rcu_torture_ops *cur_ops;
334 
335 /*
336  * Definitions for rcu torture testing.
337  */
338 
339 static int rcu_torture_read_lock(void) __acquires(RCU)
340 {
341 	rcu_read_lock();
342 	return 0;
343 }
344 
345 static void
346 rcu_read_delay(struct torture_random_state *rrsp, struct rt_read_seg *rtrsp)
347 {
348 	unsigned long started;
349 	unsigned long completed;
350 	const unsigned long shortdelay_us = 200;
351 	unsigned long longdelay_ms = 300;
352 	unsigned long long ts;
353 
354 	/* We want a short delay sometimes to make a reader delay the grace
355 	 * period, and we want a long delay occasionally to trigger
356 	 * force_quiescent_state. */
357 
358 	if (!READ_ONCE(rcu_fwd_cb_nodelay) &&
359 	    !(torture_random(rrsp) % (nrealreaders * 2000 * longdelay_ms))) {
360 		started = cur_ops->get_gp_seq();
361 		ts = rcu_trace_clock_local();
362 		if (preempt_count() & (SOFTIRQ_MASK | HARDIRQ_MASK))
363 			longdelay_ms = 5; /* Avoid triggering BH limits. */
364 		mdelay(longdelay_ms);
365 		rtrsp->rt_delay_ms = longdelay_ms;
366 		completed = cur_ops->get_gp_seq();
367 		do_trace_rcu_torture_read(cur_ops->name, NULL, ts,
368 					  started, completed);
369 	}
370 	if (!(torture_random(rrsp) % (nrealreaders * 2 * shortdelay_us))) {
371 		udelay(shortdelay_us);
372 		rtrsp->rt_delay_us = shortdelay_us;
373 	}
374 	if (!preempt_count() &&
375 	    !(torture_random(rrsp) % (nrealreaders * 500))) {
376 		torture_preempt_schedule();  /* QS only if preemptible. */
377 		rtrsp->rt_preempted = true;
378 	}
379 }
380 
381 static void rcu_torture_read_unlock(int idx) __releases(RCU)
382 {
383 	rcu_read_unlock();
384 }
385 
386 /*
387  * Update callback in the pipe.  This should be invoked after a grace period.
388  */
389 static bool
390 rcu_torture_pipe_update_one(struct rcu_torture *rp)
391 {
392 	int i;
393 
394 	i = READ_ONCE(rp->rtort_pipe_count);
395 	if (i > RCU_TORTURE_PIPE_LEN)
396 		i = RCU_TORTURE_PIPE_LEN;
397 	atomic_inc(&rcu_torture_wcount[i]);
398 	WRITE_ONCE(rp->rtort_pipe_count, i + 1);
399 	if (rp->rtort_pipe_count >= RCU_TORTURE_PIPE_LEN) {
400 		rp->rtort_mbtest = 0;
401 		return true;
402 	}
403 	return false;
404 }
405 
406 /*
407  * Update all callbacks in the pipe.  Suitable for synchronous grace-period
408  * primitives.
409  */
410 static void
411 rcu_torture_pipe_update(struct rcu_torture *old_rp)
412 {
413 	struct rcu_torture *rp;
414 	struct rcu_torture *rp1;
415 
416 	if (old_rp)
417 		list_add(&old_rp->rtort_free, &rcu_torture_removed);
418 	list_for_each_entry_safe(rp, rp1, &rcu_torture_removed, rtort_free) {
419 		if (rcu_torture_pipe_update_one(rp)) {
420 			list_del(&rp->rtort_free);
421 			rcu_torture_free(rp);
422 		}
423 	}
424 }
425 
426 static void
427 rcu_torture_cb(struct rcu_head *p)
428 {
429 	struct rcu_torture *rp = container_of(p, struct rcu_torture, rtort_rcu);
430 
431 	if (torture_must_stop_irq()) {
432 		/* Test is ending, just drop callbacks on the floor. */
433 		/* The next initialization will pick up the pieces. */
434 		return;
435 	}
436 	if (rcu_torture_pipe_update_one(rp))
437 		rcu_torture_free(rp);
438 	else
439 		cur_ops->deferred_free(rp);
440 }
441 
442 static unsigned long rcu_no_completed(void)
443 {
444 	return 0;
445 }
446 
447 static void rcu_torture_deferred_free(struct rcu_torture *p)
448 {
449 	call_rcu(&p->rtort_rcu, rcu_torture_cb);
450 }
451 
452 static void rcu_sync_torture_init(void)
453 {
454 	INIT_LIST_HEAD(&rcu_torture_removed);
455 }
456 
457 static struct rcu_torture_ops rcu_ops = {
458 	.ttype		= RCU_FLAVOR,
459 	.init		= rcu_sync_torture_init,
460 	.readlock	= rcu_torture_read_lock,
461 	.read_delay	= rcu_read_delay,
462 	.readunlock	= rcu_torture_read_unlock,
463 	.get_gp_seq	= rcu_get_gp_seq,
464 	.gp_diff	= rcu_seq_diff,
465 	.deferred_free	= rcu_torture_deferred_free,
466 	.sync		= synchronize_rcu,
467 	.exp_sync	= synchronize_rcu_expedited,
468 	.get_state	= get_state_synchronize_rcu,
469 	.cond_sync	= cond_synchronize_rcu,
470 	.call		= call_rcu,
471 	.cb_barrier	= rcu_barrier,
472 	.fqs		= rcu_force_quiescent_state,
473 	.stats		= NULL,
474 	.stall_dur	= rcu_jiffies_till_stall_check,
475 	.irq_capable	= 1,
476 	.can_boost	= rcu_can_boost(),
477 	.extendables	= RCUTORTURE_MAX_EXTEND,
478 	.name		= "rcu"
479 };
480 
481 /*
482  * Don't even think about trying any of these in real life!!!
483  * The names includes "busted", and they really means it!
484  * The only purpose of these functions is to provide a buggy RCU
485  * implementation to make sure that rcutorture correctly emits
486  * buggy-RCU error messages.
487  */
488 static void rcu_busted_torture_deferred_free(struct rcu_torture *p)
489 {
490 	/* This is a deliberate bug for testing purposes only! */
491 	rcu_torture_cb(&p->rtort_rcu);
492 }
493 
494 static void synchronize_rcu_busted(void)
495 {
496 	/* This is a deliberate bug for testing purposes only! */
497 }
498 
499 static void
500 call_rcu_busted(struct rcu_head *head, rcu_callback_t func)
501 {
502 	/* This is a deliberate bug for testing purposes only! */
503 	func(head);
504 }
505 
506 static struct rcu_torture_ops rcu_busted_ops = {
507 	.ttype		= INVALID_RCU_FLAVOR,
508 	.init		= rcu_sync_torture_init,
509 	.readlock	= rcu_torture_read_lock,
510 	.read_delay	= rcu_read_delay,  /* just reuse rcu's version. */
511 	.readunlock	= rcu_torture_read_unlock,
512 	.get_gp_seq	= rcu_no_completed,
513 	.deferred_free	= rcu_busted_torture_deferred_free,
514 	.sync		= synchronize_rcu_busted,
515 	.exp_sync	= synchronize_rcu_busted,
516 	.call		= call_rcu_busted,
517 	.cb_barrier	= NULL,
518 	.fqs		= NULL,
519 	.stats		= NULL,
520 	.irq_capable	= 1,
521 	.name		= "busted"
522 };
523 
524 /*
525  * Definitions for srcu torture testing.
526  */
527 
528 DEFINE_STATIC_SRCU(srcu_ctl);
529 static struct srcu_struct srcu_ctld;
530 static struct srcu_struct *srcu_ctlp = &srcu_ctl;
531 
532 static int srcu_torture_read_lock(void) __acquires(srcu_ctlp)
533 {
534 	return srcu_read_lock(srcu_ctlp);
535 }
536 
537 static void
538 srcu_read_delay(struct torture_random_state *rrsp, struct rt_read_seg *rtrsp)
539 {
540 	long delay;
541 	const long uspertick = 1000000 / HZ;
542 	const long longdelay = 10;
543 
544 	/* We want there to be long-running readers, but not all the time. */
545 
546 	delay = torture_random(rrsp) %
547 		(nrealreaders * 2 * longdelay * uspertick);
548 	if (!delay && in_task()) {
549 		schedule_timeout_interruptible(longdelay);
550 		rtrsp->rt_delay_jiffies = longdelay;
551 	} else {
552 		rcu_read_delay(rrsp, rtrsp);
553 	}
554 }
555 
556 static void srcu_torture_read_unlock(int idx) __releases(srcu_ctlp)
557 {
558 	srcu_read_unlock(srcu_ctlp, idx);
559 }
560 
561 static unsigned long srcu_torture_completed(void)
562 {
563 	return srcu_batches_completed(srcu_ctlp);
564 }
565 
566 static void srcu_torture_deferred_free(struct rcu_torture *rp)
567 {
568 	call_srcu(srcu_ctlp, &rp->rtort_rcu, rcu_torture_cb);
569 }
570 
571 static void srcu_torture_synchronize(void)
572 {
573 	synchronize_srcu(srcu_ctlp);
574 }
575 
576 static void srcu_torture_call(struct rcu_head *head,
577 			      rcu_callback_t func)
578 {
579 	call_srcu(srcu_ctlp, head, func);
580 }
581 
582 static void srcu_torture_barrier(void)
583 {
584 	srcu_barrier(srcu_ctlp);
585 }
586 
587 static void srcu_torture_stats(void)
588 {
589 	srcu_torture_stats_print(srcu_ctlp, torture_type, TORTURE_FLAG);
590 }
591 
592 static void srcu_torture_synchronize_expedited(void)
593 {
594 	synchronize_srcu_expedited(srcu_ctlp);
595 }
596 
597 static struct rcu_torture_ops srcu_ops = {
598 	.ttype		= SRCU_FLAVOR,
599 	.init		= rcu_sync_torture_init,
600 	.readlock	= srcu_torture_read_lock,
601 	.read_delay	= srcu_read_delay,
602 	.readunlock	= srcu_torture_read_unlock,
603 	.get_gp_seq	= srcu_torture_completed,
604 	.deferred_free	= srcu_torture_deferred_free,
605 	.sync		= srcu_torture_synchronize,
606 	.exp_sync	= srcu_torture_synchronize_expedited,
607 	.call		= srcu_torture_call,
608 	.cb_barrier	= srcu_torture_barrier,
609 	.stats		= srcu_torture_stats,
610 	.irq_capable	= 1,
611 	.name		= "srcu"
612 };
613 
614 static void srcu_torture_init(void)
615 {
616 	rcu_sync_torture_init();
617 	WARN_ON(init_srcu_struct(&srcu_ctld));
618 	srcu_ctlp = &srcu_ctld;
619 }
620 
621 static void srcu_torture_cleanup(void)
622 {
623 	cleanup_srcu_struct(&srcu_ctld);
624 	srcu_ctlp = &srcu_ctl; /* In case of a later rcutorture run. */
625 }
626 
627 /* As above, but dynamically allocated. */
628 static struct rcu_torture_ops srcud_ops = {
629 	.ttype		= SRCU_FLAVOR,
630 	.init		= srcu_torture_init,
631 	.cleanup	= srcu_torture_cleanup,
632 	.readlock	= srcu_torture_read_lock,
633 	.read_delay	= srcu_read_delay,
634 	.readunlock	= srcu_torture_read_unlock,
635 	.get_gp_seq	= srcu_torture_completed,
636 	.deferred_free	= srcu_torture_deferred_free,
637 	.sync		= srcu_torture_synchronize,
638 	.exp_sync	= srcu_torture_synchronize_expedited,
639 	.call		= srcu_torture_call,
640 	.cb_barrier	= srcu_torture_barrier,
641 	.stats		= srcu_torture_stats,
642 	.irq_capable	= 1,
643 	.name		= "srcud"
644 };
645 
646 /* As above, but broken due to inappropriate reader extension. */
647 static struct rcu_torture_ops busted_srcud_ops = {
648 	.ttype		= SRCU_FLAVOR,
649 	.init		= srcu_torture_init,
650 	.cleanup	= srcu_torture_cleanup,
651 	.readlock	= srcu_torture_read_lock,
652 	.read_delay	= rcu_read_delay,
653 	.readunlock	= srcu_torture_read_unlock,
654 	.get_gp_seq	= srcu_torture_completed,
655 	.deferred_free	= srcu_torture_deferred_free,
656 	.sync		= srcu_torture_synchronize,
657 	.exp_sync	= srcu_torture_synchronize_expedited,
658 	.call		= srcu_torture_call,
659 	.cb_barrier	= srcu_torture_barrier,
660 	.stats		= srcu_torture_stats,
661 	.irq_capable	= 1,
662 	.extendables	= RCUTORTURE_MAX_EXTEND,
663 	.name		= "busted_srcud"
664 };
665 
666 /*
667  * Definitions for RCU-tasks torture testing.
668  */
669 
670 static int tasks_torture_read_lock(void)
671 {
672 	return 0;
673 }
674 
675 static void tasks_torture_read_unlock(int idx)
676 {
677 }
678 
679 static void rcu_tasks_torture_deferred_free(struct rcu_torture *p)
680 {
681 	call_rcu_tasks(&p->rtort_rcu, rcu_torture_cb);
682 }
683 
684 static void synchronize_rcu_mult_test(void)
685 {
686 	synchronize_rcu_mult(call_rcu_tasks, call_rcu);
687 }
688 
689 static struct rcu_torture_ops tasks_ops = {
690 	.ttype		= RCU_TASKS_FLAVOR,
691 	.init		= rcu_sync_torture_init,
692 	.readlock	= tasks_torture_read_lock,
693 	.read_delay	= rcu_read_delay,  /* just reuse rcu's version. */
694 	.readunlock	= tasks_torture_read_unlock,
695 	.get_gp_seq	= rcu_no_completed,
696 	.deferred_free	= rcu_tasks_torture_deferred_free,
697 	.sync		= synchronize_rcu_tasks,
698 	.exp_sync	= synchronize_rcu_mult_test,
699 	.call		= call_rcu_tasks,
700 	.cb_barrier	= rcu_barrier_tasks,
701 	.fqs		= NULL,
702 	.stats		= NULL,
703 	.irq_capable	= 1,
704 	.slow_gps	= 1,
705 	.name		= "tasks"
706 };
707 
708 /*
709  * Definitions for trivial CONFIG_PREEMPT=n-only torture testing.
710  * This implementation does not necessarily work well with CPU hotplug.
711  */
712 
713 static void synchronize_rcu_trivial(void)
714 {
715 	int cpu;
716 
717 	for_each_online_cpu(cpu) {
718 		rcutorture_sched_setaffinity(current->pid, cpumask_of(cpu));
719 		WARN_ON_ONCE(raw_smp_processor_id() != cpu);
720 	}
721 }
722 
723 static int rcu_torture_read_lock_trivial(void) __acquires(RCU)
724 {
725 	preempt_disable();
726 	return 0;
727 }
728 
729 static void rcu_torture_read_unlock_trivial(int idx) __releases(RCU)
730 {
731 	preempt_enable();
732 }
733 
734 static struct rcu_torture_ops trivial_ops = {
735 	.ttype		= RCU_TRIVIAL_FLAVOR,
736 	.init		= rcu_sync_torture_init,
737 	.readlock	= rcu_torture_read_lock_trivial,
738 	.read_delay	= rcu_read_delay,  /* just reuse rcu's version. */
739 	.readunlock	= rcu_torture_read_unlock_trivial,
740 	.get_gp_seq	= rcu_no_completed,
741 	.sync		= synchronize_rcu_trivial,
742 	.exp_sync	= synchronize_rcu_trivial,
743 	.fqs		= NULL,
744 	.stats		= NULL,
745 	.irq_capable	= 1,
746 	.name		= "trivial"
747 };
748 
749 /*
750  * Definitions for rude RCU-tasks torture testing.
751  */
752 
753 static void rcu_tasks_rude_torture_deferred_free(struct rcu_torture *p)
754 {
755 	call_rcu_tasks_rude(&p->rtort_rcu, rcu_torture_cb);
756 }
757 
758 static struct rcu_torture_ops tasks_rude_ops = {
759 	.ttype		= RCU_TASKS_RUDE_FLAVOR,
760 	.init		= rcu_sync_torture_init,
761 	.readlock	= rcu_torture_read_lock_trivial,
762 	.read_delay	= rcu_read_delay,  /* just reuse rcu's version. */
763 	.readunlock	= rcu_torture_read_unlock_trivial,
764 	.get_gp_seq	= rcu_no_completed,
765 	.deferred_free	= rcu_tasks_rude_torture_deferred_free,
766 	.sync		= synchronize_rcu_tasks_rude,
767 	.exp_sync	= synchronize_rcu_tasks_rude,
768 	.call		= call_rcu_tasks_rude,
769 	.cb_barrier	= rcu_barrier_tasks_rude,
770 	.fqs		= NULL,
771 	.stats		= NULL,
772 	.irq_capable	= 1,
773 	.name		= "tasks-rude"
774 };
775 
776 /*
777  * Definitions for tracing RCU-tasks torture testing.
778  */
779 
780 static int tasks_tracing_torture_read_lock(void)
781 {
782 	rcu_read_lock_trace();
783 	return 0;
784 }
785 
786 static void tasks_tracing_torture_read_unlock(int idx)
787 {
788 	rcu_read_unlock_trace();
789 }
790 
791 static void rcu_tasks_tracing_torture_deferred_free(struct rcu_torture *p)
792 {
793 	call_rcu_tasks_trace(&p->rtort_rcu, rcu_torture_cb);
794 }
795 
796 static struct rcu_torture_ops tasks_tracing_ops = {
797 	.ttype		= RCU_TASKS_TRACING_FLAVOR,
798 	.init		= rcu_sync_torture_init,
799 	.readlock	= tasks_tracing_torture_read_lock,
800 	.read_delay	= srcu_read_delay,  /* just reuse srcu's version. */
801 	.readunlock	= tasks_tracing_torture_read_unlock,
802 	.get_gp_seq	= rcu_no_completed,
803 	.deferred_free	= rcu_tasks_tracing_torture_deferred_free,
804 	.sync		= synchronize_rcu_tasks_trace,
805 	.exp_sync	= synchronize_rcu_tasks_trace,
806 	.call		= call_rcu_tasks_trace,
807 	.cb_barrier	= rcu_barrier_tasks_trace,
808 	.fqs		= NULL,
809 	.stats		= NULL,
810 	.irq_capable	= 1,
811 	.slow_gps	= 1,
812 	.name		= "tasks-tracing"
813 };
814 
815 static unsigned long rcutorture_seq_diff(unsigned long new, unsigned long old)
816 {
817 	if (!cur_ops->gp_diff)
818 		return new - old;
819 	return cur_ops->gp_diff(new, old);
820 }
821 
822 static bool __maybe_unused torturing_tasks(void)
823 {
824 	return cur_ops == &tasks_ops || cur_ops == &tasks_rude_ops;
825 }
826 
827 /*
828  * RCU torture priority-boost testing.  Runs one real-time thread per
829  * CPU for moderate bursts, repeatedly registering RCU callbacks and
830  * spinning waiting for them to be invoked.  If a given callback takes
831  * too long to be invoked, we assume that priority inversion has occurred.
832  */
833 
834 struct rcu_boost_inflight {
835 	struct rcu_head rcu;
836 	int inflight;
837 };
838 
839 static void rcu_torture_boost_cb(struct rcu_head *head)
840 {
841 	struct rcu_boost_inflight *rbip =
842 		container_of(head, struct rcu_boost_inflight, rcu);
843 
844 	/* Ensure RCU-core accesses precede clearing ->inflight */
845 	smp_store_release(&rbip->inflight, 0);
846 }
847 
848 static int old_rt_runtime = -1;
849 
850 static void rcu_torture_disable_rt_throttle(void)
851 {
852 	/*
853 	 * Disable RT throttling so that rcutorture's boost threads don't get
854 	 * throttled. Only possible if rcutorture is built-in otherwise the
855 	 * user should manually do this by setting the sched_rt_period_us and
856 	 * sched_rt_runtime sysctls.
857 	 */
858 	if (!IS_BUILTIN(CONFIG_RCU_TORTURE_TEST) || old_rt_runtime != -1)
859 		return;
860 
861 	old_rt_runtime = sysctl_sched_rt_runtime;
862 	sysctl_sched_rt_runtime = -1;
863 }
864 
865 static void rcu_torture_enable_rt_throttle(void)
866 {
867 	if (!IS_BUILTIN(CONFIG_RCU_TORTURE_TEST) || old_rt_runtime == -1)
868 		return;
869 
870 	sysctl_sched_rt_runtime = old_rt_runtime;
871 	old_rt_runtime = -1;
872 }
873 
874 static bool rcu_torture_boost_failed(unsigned long start, unsigned long end)
875 {
876 	if (end - start > test_boost_duration * HZ - HZ / 2) {
877 		VERBOSE_TOROUT_STRING("rcu_torture_boost boosting failed");
878 		n_rcu_torture_boost_failure++;
879 
880 		return true; /* failed */
881 	}
882 
883 	return false; /* passed */
884 }
885 
886 static int rcu_torture_boost(void *arg)
887 {
888 	unsigned long call_rcu_time;
889 	unsigned long endtime;
890 	unsigned long oldstarttime;
891 	struct rcu_boost_inflight rbi = { .inflight = 0 };
892 	struct sched_param sp;
893 
894 	VERBOSE_TOROUT_STRING("rcu_torture_boost started");
895 
896 	/* Set real-time priority. */
897 	sp.sched_priority = 1;
898 	if (sched_setscheduler(current, SCHED_FIFO, &sp) < 0) {
899 		VERBOSE_TOROUT_STRING("rcu_torture_boost RT prio failed!");
900 		n_rcu_torture_boost_rterror++;
901 	}
902 
903 	init_rcu_head_on_stack(&rbi.rcu);
904 	/* Each pass through the following loop does one boost-test cycle. */
905 	do {
906 		/* Track if the test failed already in this test interval? */
907 		bool failed = false;
908 
909 		/* Increment n_rcu_torture_boosts once per boost-test */
910 		while (!kthread_should_stop()) {
911 			if (mutex_trylock(&boost_mutex)) {
912 				n_rcu_torture_boosts++;
913 				mutex_unlock(&boost_mutex);
914 				break;
915 			}
916 			schedule_timeout_uninterruptible(1);
917 		}
918 		if (kthread_should_stop())
919 			goto checkwait;
920 
921 		/* Wait for the next test interval. */
922 		oldstarttime = boost_starttime;
923 		while (time_before(jiffies, oldstarttime)) {
924 			schedule_timeout_interruptible(oldstarttime - jiffies);
925 			stutter_wait("rcu_torture_boost");
926 			if (torture_must_stop())
927 				goto checkwait;
928 		}
929 
930 		/* Do one boost-test interval. */
931 		endtime = oldstarttime + test_boost_duration * HZ;
932 		call_rcu_time = jiffies;
933 		while (time_before(jiffies, endtime)) {
934 			/* If we don't have a callback in flight, post one. */
935 			if (!smp_load_acquire(&rbi.inflight)) {
936 				/* RCU core before ->inflight = 1. */
937 				smp_store_release(&rbi.inflight, 1);
938 				call_rcu(&rbi.rcu, rcu_torture_boost_cb);
939 				/* Check if the boost test failed */
940 				failed = failed ||
941 					 rcu_torture_boost_failed(call_rcu_time,
942 								 jiffies);
943 				call_rcu_time = jiffies;
944 			}
945 			stutter_wait("rcu_torture_boost");
946 			if (torture_must_stop())
947 				goto checkwait;
948 		}
949 
950 		/*
951 		 * If boost never happened, then inflight will always be 1, in
952 		 * this case the boost check would never happen in the above
953 		 * loop so do another one here.
954 		 */
955 		if (!failed && smp_load_acquire(&rbi.inflight))
956 			rcu_torture_boost_failed(call_rcu_time, jiffies);
957 
958 		/*
959 		 * Set the start time of the next test interval.
960 		 * Yes, this is vulnerable to long delays, but such
961 		 * delays simply cause a false negative for the next
962 		 * interval.  Besides, we are running at RT priority,
963 		 * so delays should be relatively rare.
964 		 */
965 		while (oldstarttime == boost_starttime &&
966 		       !kthread_should_stop()) {
967 			if (mutex_trylock(&boost_mutex)) {
968 				boost_starttime = jiffies +
969 						  test_boost_interval * HZ;
970 				mutex_unlock(&boost_mutex);
971 				break;
972 			}
973 			schedule_timeout_uninterruptible(1);
974 		}
975 
976 		/* Go do the stutter. */
977 checkwait:	stutter_wait("rcu_torture_boost");
978 	} while (!torture_must_stop());
979 
980 	/* Clean up and exit. */
981 	while (!kthread_should_stop() || smp_load_acquire(&rbi.inflight)) {
982 		torture_shutdown_absorb("rcu_torture_boost");
983 		schedule_timeout_uninterruptible(1);
984 	}
985 	destroy_rcu_head_on_stack(&rbi.rcu);
986 	torture_kthread_stopping("rcu_torture_boost");
987 	return 0;
988 }
989 
990 /*
991  * RCU torture force-quiescent-state kthread.  Repeatedly induces
992  * bursts of calls to force_quiescent_state(), increasing the probability
993  * of occurrence of some important types of race conditions.
994  */
995 static int
996 rcu_torture_fqs(void *arg)
997 {
998 	unsigned long fqs_resume_time;
999 	int fqs_burst_remaining;
1000 
1001 	VERBOSE_TOROUT_STRING("rcu_torture_fqs task started");
1002 	do {
1003 		fqs_resume_time = jiffies + fqs_stutter * HZ;
1004 		while (time_before(jiffies, fqs_resume_time) &&
1005 		       !kthread_should_stop()) {
1006 			schedule_timeout_interruptible(1);
1007 		}
1008 		fqs_burst_remaining = fqs_duration;
1009 		while (fqs_burst_remaining > 0 &&
1010 		       !kthread_should_stop()) {
1011 			cur_ops->fqs();
1012 			udelay(fqs_holdoff);
1013 			fqs_burst_remaining -= fqs_holdoff;
1014 		}
1015 		stutter_wait("rcu_torture_fqs");
1016 	} while (!torture_must_stop());
1017 	torture_kthread_stopping("rcu_torture_fqs");
1018 	return 0;
1019 }
1020 
1021 /*
1022  * RCU torture writer kthread.  Repeatedly substitutes a new structure
1023  * for that pointed to by rcu_torture_current, freeing the old structure
1024  * after a series of grace periods (the "pipeline").
1025  */
1026 static int
1027 rcu_torture_writer(void *arg)
1028 {
1029 	bool can_expedite = !rcu_gp_is_expedited() && !rcu_gp_is_normal();
1030 	int expediting = 0;
1031 	unsigned long gp_snap;
1032 	bool gp_cond1 = gp_cond, gp_exp1 = gp_exp, gp_normal1 = gp_normal;
1033 	bool gp_sync1 = gp_sync;
1034 	int i;
1035 	struct rcu_torture *rp;
1036 	struct rcu_torture *old_rp;
1037 	static DEFINE_TORTURE_RANDOM(rand);
1038 	int synctype[] = { RTWS_DEF_FREE, RTWS_EXP_SYNC,
1039 			   RTWS_COND_GET, RTWS_SYNC };
1040 	int nsynctypes = 0;
1041 
1042 	VERBOSE_TOROUT_STRING("rcu_torture_writer task started");
1043 	if (!can_expedite)
1044 		pr_alert("%s" TORTURE_FLAG
1045 			 " GP expediting controlled from boot/sysfs for %s.\n",
1046 			 torture_type, cur_ops->name);
1047 
1048 	/* Initialize synctype[] array.  If none set, take default. */
1049 	if (!gp_cond1 && !gp_exp1 && !gp_normal1 && !gp_sync1)
1050 		gp_cond1 = gp_exp1 = gp_normal1 = gp_sync1 = true;
1051 	if (gp_cond1 && cur_ops->get_state && cur_ops->cond_sync) {
1052 		synctype[nsynctypes++] = RTWS_COND_GET;
1053 		pr_info("%s: Testing conditional GPs.\n", __func__);
1054 	} else if (gp_cond && (!cur_ops->get_state || !cur_ops->cond_sync)) {
1055 		pr_alert("%s: gp_cond without primitives.\n", __func__);
1056 	}
1057 	if (gp_exp1 && cur_ops->exp_sync) {
1058 		synctype[nsynctypes++] = RTWS_EXP_SYNC;
1059 		pr_info("%s: Testing expedited GPs.\n", __func__);
1060 	} else if (gp_exp && !cur_ops->exp_sync) {
1061 		pr_alert("%s: gp_exp without primitives.\n", __func__);
1062 	}
1063 	if (gp_normal1 && cur_ops->deferred_free) {
1064 		synctype[nsynctypes++] = RTWS_DEF_FREE;
1065 		pr_info("%s: Testing asynchronous GPs.\n", __func__);
1066 	} else if (gp_normal && !cur_ops->deferred_free) {
1067 		pr_alert("%s: gp_normal without primitives.\n", __func__);
1068 	}
1069 	if (gp_sync1 && cur_ops->sync) {
1070 		synctype[nsynctypes++] = RTWS_SYNC;
1071 		pr_info("%s: Testing normal GPs.\n", __func__);
1072 	} else if (gp_sync && !cur_ops->sync) {
1073 		pr_alert("%s: gp_sync without primitives.\n", __func__);
1074 	}
1075 	if (WARN_ONCE(nsynctypes == 0,
1076 		      "rcu_torture_writer: No update-side primitives.\n")) {
1077 		/*
1078 		 * No updates primitives, so don't try updating.
1079 		 * The resulting test won't be testing much, hence the
1080 		 * above WARN_ONCE().
1081 		 */
1082 		rcu_torture_writer_state = RTWS_STOPPING;
1083 		torture_kthread_stopping("rcu_torture_writer");
1084 	}
1085 
1086 	do {
1087 		rcu_torture_writer_state = RTWS_FIXED_DELAY;
1088 		schedule_timeout_uninterruptible(1);
1089 		rp = rcu_torture_alloc();
1090 		if (rp == NULL)
1091 			continue;
1092 		rp->rtort_pipe_count = 0;
1093 		rcu_torture_writer_state = RTWS_DELAY;
1094 		udelay(torture_random(&rand) & 0x3ff);
1095 		rcu_torture_writer_state = RTWS_REPLACE;
1096 		old_rp = rcu_dereference_check(rcu_torture_current,
1097 					       current == writer_task);
1098 		rp->rtort_mbtest = 1;
1099 		rcu_assign_pointer(rcu_torture_current, rp);
1100 		smp_wmb(); /* Mods to old_rp must follow rcu_assign_pointer() */
1101 		if (old_rp) {
1102 			i = old_rp->rtort_pipe_count;
1103 			if (i > RCU_TORTURE_PIPE_LEN)
1104 				i = RCU_TORTURE_PIPE_LEN;
1105 			atomic_inc(&rcu_torture_wcount[i]);
1106 			WRITE_ONCE(old_rp->rtort_pipe_count,
1107 				   old_rp->rtort_pipe_count + 1);
1108 			switch (synctype[torture_random(&rand) % nsynctypes]) {
1109 			case RTWS_DEF_FREE:
1110 				rcu_torture_writer_state = RTWS_DEF_FREE;
1111 				cur_ops->deferred_free(old_rp);
1112 				break;
1113 			case RTWS_EXP_SYNC:
1114 				rcu_torture_writer_state = RTWS_EXP_SYNC;
1115 				cur_ops->exp_sync();
1116 				rcu_torture_pipe_update(old_rp);
1117 				break;
1118 			case RTWS_COND_GET:
1119 				rcu_torture_writer_state = RTWS_COND_GET;
1120 				gp_snap = cur_ops->get_state();
1121 				i = torture_random(&rand) % 16;
1122 				if (i != 0)
1123 					schedule_timeout_interruptible(i);
1124 				udelay(torture_random(&rand) % 1000);
1125 				rcu_torture_writer_state = RTWS_COND_SYNC;
1126 				cur_ops->cond_sync(gp_snap);
1127 				rcu_torture_pipe_update(old_rp);
1128 				break;
1129 			case RTWS_SYNC:
1130 				rcu_torture_writer_state = RTWS_SYNC;
1131 				cur_ops->sync();
1132 				rcu_torture_pipe_update(old_rp);
1133 				break;
1134 			default:
1135 				WARN_ON_ONCE(1);
1136 				break;
1137 			}
1138 		}
1139 		WRITE_ONCE(rcu_torture_current_version,
1140 			   rcu_torture_current_version + 1);
1141 		/* Cycle through nesting levels of rcu_expedite_gp() calls. */
1142 		if (can_expedite &&
1143 		    !(torture_random(&rand) & 0xff & (!!expediting - 1))) {
1144 			WARN_ON_ONCE(expediting == 0 && rcu_gp_is_expedited());
1145 			if (expediting >= 0)
1146 				rcu_expedite_gp();
1147 			else
1148 				rcu_unexpedite_gp();
1149 			if (++expediting > 3)
1150 				expediting = -expediting;
1151 		} else if (!can_expedite) { /* Disabled during boot, recheck. */
1152 			can_expedite = !rcu_gp_is_expedited() &&
1153 				       !rcu_gp_is_normal();
1154 		}
1155 		rcu_torture_writer_state = RTWS_STUTTER;
1156 		if (stutter_wait("rcu_torture_writer") &&
1157 		    !READ_ONCE(rcu_fwd_cb_nodelay) &&
1158 		    !cur_ops->slow_gps &&
1159 		    !torture_must_stop() &&
1160 		    rcu_inkernel_boot_has_ended())
1161 			for (i = 0; i < ARRAY_SIZE(rcu_tortures); i++)
1162 				if (list_empty(&rcu_tortures[i].rtort_free) &&
1163 				    rcu_access_pointer(rcu_torture_current) !=
1164 				    &rcu_tortures[i]) {
1165 					rcu_ftrace_dump(DUMP_ALL);
1166 					WARN(1, "%s: rtort_pipe_count: %d\n", __func__, rcu_tortures[i].rtort_pipe_count);
1167 				}
1168 	} while (!torture_must_stop());
1169 	/* Reset expediting back to unexpedited. */
1170 	if (expediting > 0)
1171 		expediting = -expediting;
1172 	while (can_expedite && expediting++ < 0)
1173 		rcu_unexpedite_gp();
1174 	WARN_ON_ONCE(can_expedite && rcu_gp_is_expedited());
1175 	if (!can_expedite)
1176 		pr_alert("%s" TORTURE_FLAG
1177 			 " Dynamic grace-period expediting was disabled.\n",
1178 			 torture_type);
1179 	rcu_torture_writer_state = RTWS_STOPPING;
1180 	torture_kthread_stopping("rcu_torture_writer");
1181 	return 0;
1182 }
1183 
1184 /*
1185  * RCU torture fake writer kthread.  Repeatedly calls sync, with a random
1186  * delay between calls.
1187  */
1188 static int
1189 rcu_torture_fakewriter(void *arg)
1190 {
1191 	DEFINE_TORTURE_RANDOM(rand);
1192 
1193 	VERBOSE_TOROUT_STRING("rcu_torture_fakewriter task started");
1194 	set_user_nice(current, MAX_NICE);
1195 
1196 	do {
1197 		schedule_timeout_uninterruptible(1 + torture_random(&rand)%10);
1198 		udelay(torture_random(&rand) & 0x3ff);
1199 		if (cur_ops->cb_barrier != NULL &&
1200 		    torture_random(&rand) % (nfakewriters * 8) == 0) {
1201 			cur_ops->cb_barrier();
1202 		} else if (gp_normal == gp_exp) {
1203 			if (cur_ops->sync && torture_random(&rand) & 0x80)
1204 				cur_ops->sync();
1205 			else if (cur_ops->exp_sync)
1206 				cur_ops->exp_sync();
1207 		} else if (gp_normal && cur_ops->sync) {
1208 			cur_ops->sync();
1209 		} else if (cur_ops->exp_sync) {
1210 			cur_ops->exp_sync();
1211 		}
1212 		stutter_wait("rcu_torture_fakewriter");
1213 	} while (!torture_must_stop());
1214 
1215 	torture_kthread_stopping("rcu_torture_fakewriter");
1216 	return 0;
1217 }
1218 
1219 static void rcu_torture_timer_cb(struct rcu_head *rhp)
1220 {
1221 	kfree(rhp);
1222 }
1223 
1224 /*
1225  * Do one extension of an RCU read-side critical section using the
1226  * current reader state in readstate (set to zero for initial entry
1227  * to extended critical section), set the new state as specified by
1228  * newstate (set to zero for final exit from extended critical section),
1229  * and random-number-generator state in trsp.  If this is neither the
1230  * beginning or end of the critical section and if there was actually a
1231  * change, do a ->read_delay().
1232  */
1233 static void rcutorture_one_extend(int *readstate, int newstate,
1234 				  struct torture_random_state *trsp,
1235 				  struct rt_read_seg *rtrsp)
1236 {
1237 	unsigned long flags;
1238 	int idxnew = -1;
1239 	int idxold = *readstate;
1240 	int statesnew = ~*readstate & newstate;
1241 	int statesold = *readstate & ~newstate;
1242 
1243 	WARN_ON_ONCE(idxold < 0);
1244 	WARN_ON_ONCE((idxold >> RCUTORTURE_RDR_SHIFT) > 1);
1245 	rtrsp->rt_readstate = newstate;
1246 
1247 	/* First, put new protection in place to avoid critical-section gap. */
1248 	if (statesnew & RCUTORTURE_RDR_BH)
1249 		local_bh_disable();
1250 	if (statesnew & RCUTORTURE_RDR_IRQ)
1251 		local_irq_disable();
1252 	if (statesnew & RCUTORTURE_RDR_PREEMPT)
1253 		preempt_disable();
1254 	if (statesnew & RCUTORTURE_RDR_RBH)
1255 		rcu_read_lock_bh();
1256 	if (statesnew & RCUTORTURE_RDR_SCHED)
1257 		rcu_read_lock_sched();
1258 	if (statesnew & RCUTORTURE_RDR_RCU)
1259 		idxnew = cur_ops->readlock() << RCUTORTURE_RDR_SHIFT;
1260 
1261 	/* Next, remove old protection, irq first due to bh conflict. */
1262 	if (statesold & RCUTORTURE_RDR_IRQ)
1263 		local_irq_enable();
1264 	if (statesold & RCUTORTURE_RDR_BH)
1265 		local_bh_enable();
1266 	if (statesold & RCUTORTURE_RDR_PREEMPT)
1267 		preempt_enable();
1268 	if (statesold & RCUTORTURE_RDR_RBH)
1269 		rcu_read_unlock_bh();
1270 	if (statesold & RCUTORTURE_RDR_SCHED)
1271 		rcu_read_unlock_sched();
1272 	if (statesold & RCUTORTURE_RDR_RCU) {
1273 		bool lockit = !statesnew && !(torture_random(trsp) & 0xffff);
1274 
1275 		if (lockit)
1276 			raw_spin_lock_irqsave(&current->pi_lock, flags);
1277 		cur_ops->readunlock(idxold >> RCUTORTURE_RDR_SHIFT);
1278 		if (lockit)
1279 			raw_spin_unlock_irqrestore(&current->pi_lock, flags);
1280 	}
1281 
1282 	/* Delay if neither beginning nor end and there was a change. */
1283 	if ((statesnew || statesold) && *readstate && newstate)
1284 		cur_ops->read_delay(trsp, rtrsp);
1285 
1286 	/* Update the reader state. */
1287 	if (idxnew == -1)
1288 		idxnew = idxold & ~RCUTORTURE_RDR_MASK;
1289 	WARN_ON_ONCE(idxnew < 0);
1290 	WARN_ON_ONCE((idxnew >> RCUTORTURE_RDR_SHIFT) > 1);
1291 	*readstate = idxnew | newstate;
1292 	WARN_ON_ONCE((*readstate >> RCUTORTURE_RDR_SHIFT) < 0);
1293 	WARN_ON_ONCE((*readstate >> RCUTORTURE_RDR_SHIFT) > 1);
1294 }
1295 
1296 /* Return the biggest extendables mask given current RCU and boot parameters. */
1297 static int rcutorture_extend_mask_max(void)
1298 {
1299 	int mask;
1300 
1301 	WARN_ON_ONCE(extendables & ~RCUTORTURE_MAX_EXTEND);
1302 	mask = extendables & RCUTORTURE_MAX_EXTEND & cur_ops->extendables;
1303 	mask = mask | RCUTORTURE_RDR_RCU;
1304 	return mask;
1305 }
1306 
1307 /* Return a random protection state mask, but with at least one bit set. */
1308 static int
1309 rcutorture_extend_mask(int oldmask, struct torture_random_state *trsp)
1310 {
1311 	int mask = rcutorture_extend_mask_max();
1312 	unsigned long randmask1 = torture_random(trsp) >> 8;
1313 	unsigned long randmask2 = randmask1 >> 3;
1314 
1315 	WARN_ON_ONCE(mask >> RCUTORTURE_RDR_SHIFT);
1316 	/* Mostly only one bit (need preemption!), sometimes lots of bits. */
1317 	if (!(randmask1 & 0x7))
1318 		mask = mask & randmask2;
1319 	else
1320 		mask = mask & (1 << (randmask2 % RCUTORTURE_RDR_NBITS));
1321 	/* Can't enable bh w/irq disabled. */
1322 	if ((mask & RCUTORTURE_RDR_IRQ) &&
1323 	    ((!(mask & RCUTORTURE_RDR_BH) && (oldmask & RCUTORTURE_RDR_BH)) ||
1324 	     (!(mask & RCUTORTURE_RDR_RBH) && (oldmask & RCUTORTURE_RDR_RBH))))
1325 		mask |= RCUTORTURE_RDR_BH | RCUTORTURE_RDR_RBH;
1326 	return mask ?: RCUTORTURE_RDR_RCU;
1327 }
1328 
1329 /*
1330  * Do a randomly selected number of extensions of an existing RCU read-side
1331  * critical section.
1332  */
1333 static struct rt_read_seg *
1334 rcutorture_loop_extend(int *readstate, struct torture_random_state *trsp,
1335 		       struct rt_read_seg *rtrsp)
1336 {
1337 	int i;
1338 	int j;
1339 	int mask = rcutorture_extend_mask_max();
1340 
1341 	WARN_ON_ONCE(!*readstate); /* -Existing- RCU read-side critsect! */
1342 	if (!((mask - 1) & mask))
1343 		return rtrsp;  /* Current RCU reader not extendable. */
1344 	/* Bias towards larger numbers of loops. */
1345 	i = (torture_random(trsp) >> 3);
1346 	i = ((i | (i >> 3)) & RCUTORTURE_RDR_MAX_LOOPS) + 1;
1347 	for (j = 0; j < i; j++) {
1348 		mask = rcutorture_extend_mask(*readstate, trsp);
1349 		rcutorture_one_extend(readstate, mask, trsp, &rtrsp[j]);
1350 	}
1351 	return &rtrsp[j];
1352 }
1353 
1354 /*
1355  * Do one read-side critical section, returning false if there was
1356  * no data to read.  Can be invoked both from process context and
1357  * from a timer handler.
1358  */
1359 static bool rcu_torture_one_read(struct torture_random_state *trsp)
1360 {
1361 	int i;
1362 	unsigned long started;
1363 	unsigned long completed;
1364 	int newstate;
1365 	struct rcu_torture *p;
1366 	int pipe_count;
1367 	int readstate = 0;
1368 	struct rt_read_seg rtseg[RCUTORTURE_RDR_MAX_SEGS] = { { 0 } };
1369 	struct rt_read_seg *rtrsp = &rtseg[0];
1370 	struct rt_read_seg *rtrsp1;
1371 	unsigned long long ts;
1372 
1373 	newstate = rcutorture_extend_mask(readstate, trsp);
1374 	rcutorture_one_extend(&readstate, newstate, trsp, rtrsp++);
1375 	started = cur_ops->get_gp_seq();
1376 	ts = rcu_trace_clock_local();
1377 	p = rcu_dereference_check(rcu_torture_current,
1378 				  rcu_read_lock_bh_held() ||
1379 				  rcu_read_lock_sched_held() ||
1380 				  srcu_read_lock_held(srcu_ctlp) ||
1381 				  rcu_read_lock_trace_held() ||
1382 				  torturing_tasks());
1383 	if (p == NULL) {
1384 		/* Wait for rcu_torture_writer to get underway */
1385 		rcutorture_one_extend(&readstate, 0, trsp, rtrsp);
1386 		return false;
1387 	}
1388 	if (p->rtort_mbtest == 0)
1389 		atomic_inc(&n_rcu_torture_mberror);
1390 	rtrsp = rcutorture_loop_extend(&readstate, trsp, rtrsp);
1391 	preempt_disable();
1392 	pipe_count = READ_ONCE(p->rtort_pipe_count);
1393 	if (pipe_count > RCU_TORTURE_PIPE_LEN) {
1394 		/* Should not happen, but... */
1395 		pipe_count = RCU_TORTURE_PIPE_LEN;
1396 	}
1397 	completed = cur_ops->get_gp_seq();
1398 	if (pipe_count > 1) {
1399 		do_trace_rcu_torture_read(cur_ops->name, &p->rtort_rcu,
1400 					  ts, started, completed);
1401 		rcu_ftrace_dump(DUMP_ALL);
1402 	}
1403 	__this_cpu_inc(rcu_torture_count[pipe_count]);
1404 	completed = rcutorture_seq_diff(completed, started);
1405 	if (completed > RCU_TORTURE_PIPE_LEN) {
1406 		/* Should not happen, but... */
1407 		completed = RCU_TORTURE_PIPE_LEN;
1408 	}
1409 	__this_cpu_inc(rcu_torture_batch[completed]);
1410 	preempt_enable();
1411 	rcutorture_one_extend(&readstate, 0, trsp, rtrsp);
1412 	WARN_ON_ONCE(readstate & RCUTORTURE_RDR_MASK);
1413 
1414 	/* If error or close call, record the sequence of reader protections. */
1415 	if ((pipe_count > 1 || completed > 1) && !xchg(&err_segs_recorded, 1)) {
1416 		i = 0;
1417 		for (rtrsp1 = &rtseg[0]; rtrsp1 < rtrsp; rtrsp1++)
1418 			err_segs[i++] = *rtrsp1;
1419 		rt_read_nsegs = i;
1420 	}
1421 
1422 	return true;
1423 }
1424 
1425 static DEFINE_TORTURE_RANDOM_PERCPU(rcu_torture_timer_rand);
1426 
1427 /*
1428  * RCU torture reader from timer handler.  Dereferences rcu_torture_current,
1429  * incrementing the corresponding element of the pipeline array.  The
1430  * counter in the element should never be greater than 1, otherwise, the
1431  * RCU implementation is broken.
1432  */
1433 static void rcu_torture_timer(struct timer_list *unused)
1434 {
1435 	atomic_long_inc(&n_rcu_torture_timers);
1436 	(void)rcu_torture_one_read(this_cpu_ptr(&rcu_torture_timer_rand));
1437 
1438 	/* Test call_rcu() invocation from interrupt handler. */
1439 	if (cur_ops->call) {
1440 		struct rcu_head *rhp = kmalloc(sizeof(*rhp), GFP_NOWAIT);
1441 
1442 		if (rhp)
1443 			cur_ops->call(rhp, rcu_torture_timer_cb);
1444 	}
1445 }
1446 
1447 /*
1448  * RCU torture reader kthread.  Repeatedly dereferences rcu_torture_current,
1449  * incrementing the corresponding element of the pipeline array.  The
1450  * counter in the element should never be greater than 1, otherwise, the
1451  * RCU implementation is broken.
1452  */
1453 static int
1454 rcu_torture_reader(void *arg)
1455 {
1456 	unsigned long lastsleep = jiffies;
1457 	long myid = (long)arg;
1458 	int mynumonline = myid;
1459 	DEFINE_TORTURE_RANDOM(rand);
1460 	struct timer_list t;
1461 
1462 	VERBOSE_TOROUT_STRING("rcu_torture_reader task started");
1463 	set_user_nice(current, MAX_NICE);
1464 	if (irqreader && cur_ops->irq_capable)
1465 		timer_setup_on_stack(&t, rcu_torture_timer, 0);
1466 	tick_dep_set_task(current, TICK_DEP_BIT_RCU);
1467 	do {
1468 		if (irqreader && cur_ops->irq_capable) {
1469 			if (!timer_pending(&t))
1470 				mod_timer(&t, jiffies + 1);
1471 		}
1472 		if (!rcu_torture_one_read(&rand) && !torture_must_stop())
1473 			schedule_timeout_interruptible(HZ);
1474 		if (time_after(jiffies, lastsleep) && !torture_must_stop()) {
1475 			schedule_timeout_interruptible(1);
1476 			lastsleep = jiffies + 10;
1477 		}
1478 		while (num_online_cpus() < mynumonline && !torture_must_stop())
1479 			schedule_timeout_interruptible(HZ / 5);
1480 		stutter_wait("rcu_torture_reader");
1481 	} while (!torture_must_stop());
1482 	if (irqreader && cur_ops->irq_capable) {
1483 		del_timer_sync(&t);
1484 		destroy_timer_on_stack(&t);
1485 	}
1486 	tick_dep_clear_task(current, TICK_DEP_BIT_RCU);
1487 	torture_kthread_stopping("rcu_torture_reader");
1488 	return 0;
1489 }
1490 
1491 /*
1492  * Print torture statistics.  Caller must ensure that there is only
1493  * one call to this function at a given time!!!  This is normally
1494  * accomplished by relying on the module system to only have one copy
1495  * of the module loaded, and then by giving the rcu_torture_stats
1496  * kthread full control (or the init/cleanup functions when rcu_torture_stats
1497  * thread is not running).
1498  */
1499 static void
1500 rcu_torture_stats_print(void)
1501 {
1502 	int cpu;
1503 	int i;
1504 	long pipesummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
1505 	long batchsummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
1506 	struct rcu_torture *rtcp;
1507 	static unsigned long rtcv_snap = ULONG_MAX;
1508 	static bool splatted;
1509 	struct task_struct *wtp;
1510 
1511 	for_each_possible_cpu(cpu) {
1512 		for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
1513 			pipesummary[i] += READ_ONCE(per_cpu(rcu_torture_count, cpu)[i]);
1514 			batchsummary[i] += READ_ONCE(per_cpu(rcu_torture_batch, cpu)[i]);
1515 		}
1516 	}
1517 	for (i = RCU_TORTURE_PIPE_LEN - 1; i >= 0; i--) {
1518 		if (pipesummary[i] != 0)
1519 			break;
1520 	}
1521 
1522 	pr_alert("%s%s ", torture_type, TORTURE_FLAG);
1523 	rtcp = rcu_access_pointer(rcu_torture_current);
1524 	pr_cont("rtc: %p %s: %lu tfle: %d rta: %d rtaf: %d rtf: %d ",
1525 		rtcp,
1526 		rtcp && !rcu_stall_is_suppressed_at_boot() ? "ver" : "VER",
1527 		rcu_torture_current_version,
1528 		list_empty(&rcu_torture_freelist),
1529 		atomic_read(&n_rcu_torture_alloc),
1530 		atomic_read(&n_rcu_torture_alloc_fail),
1531 		atomic_read(&n_rcu_torture_free));
1532 	pr_cont("rtmbe: %d rtbe: %ld rtbke: %ld rtbre: %ld ",
1533 		atomic_read(&n_rcu_torture_mberror),
1534 		n_rcu_torture_barrier_error,
1535 		n_rcu_torture_boost_ktrerror,
1536 		n_rcu_torture_boost_rterror);
1537 	pr_cont("rtbf: %ld rtb: %ld nt: %ld ",
1538 		n_rcu_torture_boost_failure,
1539 		n_rcu_torture_boosts,
1540 		atomic_long_read(&n_rcu_torture_timers));
1541 	torture_onoff_stats();
1542 	pr_cont("barrier: %ld/%ld:%ld\n",
1543 		data_race(n_barrier_successes),
1544 		data_race(n_barrier_attempts),
1545 		data_race(n_rcu_torture_barrier_error));
1546 
1547 	pr_alert("%s%s ", torture_type, TORTURE_FLAG);
1548 	if (atomic_read(&n_rcu_torture_mberror) ||
1549 	    n_rcu_torture_barrier_error || n_rcu_torture_boost_ktrerror ||
1550 	    n_rcu_torture_boost_rterror || n_rcu_torture_boost_failure ||
1551 	    i > 1) {
1552 		pr_cont("%s", "!!! ");
1553 		atomic_inc(&n_rcu_torture_error);
1554 		WARN_ON_ONCE(atomic_read(&n_rcu_torture_mberror));
1555 		WARN_ON_ONCE(n_rcu_torture_barrier_error);  // rcu_barrier()
1556 		WARN_ON_ONCE(n_rcu_torture_boost_ktrerror); // no boost kthread
1557 		WARN_ON_ONCE(n_rcu_torture_boost_rterror); // can't set RT prio
1558 		WARN_ON_ONCE(n_rcu_torture_boost_failure); // RCU boost failed
1559 		WARN_ON_ONCE(i > 1); // Too-short grace period
1560 	}
1561 	pr_cont("Reader Pipe: ");
1562 	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
1563 		pr_cont(" %ld", pipesummary[i]);
1564 	pr_cont("\n");
1565 
1566 	pr_alert("%s%s ", torture_type, TORTURE_FLAG);
1567 	pr_cont("Reader Batch: ");
1568 	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
1569 		pr_cont(" %ld", batchsummary[i]);
1570 	pr_cont("\n");
1571 
1572 	pr_alert("%s%s ", torture_type, TORTURE_FLAG);
1573 	pr_cont("Free-Block Circulation: ");
1574 	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
1575 		pr_cont(" %d", atomic_read(&rcu_torture_wcount[i]));
1576 	}
1577 	pr_cont("\n");
1578 
1579 	if (cur_ops->stats)
1580 		cur_ops->stats();
1581 	if (rtcv_snap == rcu_torture_current_version &&
1582 	    rcu_access_pointer(rcu_torture_current) &&
1583 	    !rcu_stall_is_suppressed()) {
1584 		int __maybe_unused flags = 0;
1585 		unsigned long __maybe_unused gp_seq = 0;
1586 
1587 		rcutorture_get_gp_data(cur_ops->ttype,
1588 				       &flags, &gp_seq);
1589 		srcutorture_get_gp_data(cur_ops->ttype, srcu_ctlp,
1590 					&flags, &gp_seq);
1591 		wtp = READ_ONCE(writer_task);
1592 		pr_alert("??? Writer stall state %s(%d) g%lu f%#x ->state %#lx cpu %d\n",
1593 			 rcu_torture_writer_state_getname(),
1594 			 rcu_torture_writer_state, gp_seq, flags,
1595 			 wtp == NULL ? ~0UL : wtp->state,
1596 			 wtp == NULL ? -1 : (int)task_cpu(wtp));
1597 		if (!splatted && wtp) {
1598 			sched_show_task(wtp);
1599 			splatted = true;
1600 		}
1601 		show_rcu_gp_kthreads();
1602 		rcu_ftrace_dump(DUMP_ALL);
1603 	}
1604 	rtcv_snap = rcu_torture_current_version;
1605 }
1606 
1607 /*
1608  * Periodically prints torture statistics, if periodic statistics printing
1609  * was specified via the stat_interval module parameter.
1610  */
1611 static int
1612 rcu_torture_stats(void *arg)
1613 {
1614 	VERBOSE_TOROUT_STRING("rcu_torture_stats task started");
1615 	do {
1616 		schedule_timeout_interruptible(stat_interval * HZ);
1617 		rcu_torture_stats_print();
1618 		torture_shutdown_absorb("rcu_torture_stats");
1619 	} while (!torture_must_stop());
1620 	torture_kthread_stopping("rcu_torture_stats");
1621 	return 0;
1622 }
1623 
1624 static void
1625 rcu_torture_print_module_parms(struct rcu_torture_ops *cur_ops, const char *tag)
1626 {
1627 	pr_alert("%s" TORTURE_FLAG
1628 		 "--- %s: nreaders=%d nfakewriters=%d "
1629 		 "stat_interval=%d verbose=%d test_no_idle_hz=%d "
1630 		 "shuffle_interval=%d stutter=%d irqreader=%d "
1631 		 "fqs_duration=%d fqs_holdoff=%d fqs_stutter=%d "
1632 		 "test_boost=%d/%d test_boost_interval=%d "
1633 		 "test_boost_duration=%d shutdown_secs=%d "
1634 		 "stall_cpu=%d stall_cpu_holdoff=%d stall_cpu_irqsoff=%d "
1635 		 "stall_cpu_block=%d "
1636 		 "n_barrier_cbs=%d "
1637 		 "onoff_interval=%d onoff_holdoff=%d\n",
1638 		 torture_type, tag, nrealreaders, nfakewriters,
1639 		 stat_interval, verbose, test_no_idle_hz, shuffle_interval,
1640 		 stutter, irqreader, fqs_duration, fqs_holdoff, fqs_stutter,
1641 		 test_boost, cur_ops->can_boost,
1642 		 test_boost_interval, test_boost_duration, shutdown_secs,
1643 		 stall_cpu, stall_cpu_holdoff, stall_cpu_irqsoff,
1644 		 stall_cpu_block,
1645 		 n_barrier_cbs,
1646 		 onoff_interval, onoff_holdoff);
1647 }
1648 
1649 static int rcutorture_booster_cleanup(unsigned int cpu)
1650 {
1651 	struct task_struct *t;
1652 
1653 	if (boost_tasks[cpu] == NULL)
1654 		return 0;
1655 	mutex_lock(&boost_mutex);
1656 	t = boost_tasks[cpu];
1657 	boost_tasks[cpu] = NULL;
1658 	rcu_torture_enable_rt_throttle();
1659 	mutex_unlock(&boost_mutex);
1660 
1661 	/* This must be outside of the mutex, otherwise deadlock! */
1662 	torture_stop_kthread(rcu_torture_boost, t);
1663 	return 0;
1664 }
1665 
1666 static int rcutorture_booster_init(unsigned int cpu)
1667 {
1668 	int retval;
1669 
1670 	if (boost_tasks[cpu] != NULL)
1671 		return 0;  /* Already created, nothing more to do. */
1672 
1673 	/* Don't allow time recalculation while creating a new task. */
1674 	mutex_lock(&boost_mutex);
1675 	rcu_torture_disable_rt_throttle();
1676 	VERBOSE_TOROUT_STRING("Creating rcu_torture_boost task");
1677 	boost_tasks[cpu] = kthread_create_on_node(rcu_torture_boost, NULL,
1678 						  cpu_to_node(cpu),
1679 						  "rcu_torture_boost");
1680 	if (IS_ERR(boost_tasks[cpu])) {
1681 		retval = PTR_ERR(boost_tasks[cpu]);
1682 		VERBOSE_TOROUT_STRING("rcu_torture_boost task create failed");
1683 		n_rcu_torture_boost_ktrerror++;
1684 		boost_tasks[cpu] = NULL;
1685 		mutex_unlock(&boost_mutex);
1686 		return retval;
1687 	}
1688 	kthread_bind(boost_tasks[cpu], cpu);
1689 	wake_up_process(boost_tasks[cpu]);
1690 	mutex_unlock(&boost_mutex);
1691 	return 0;
1692 }
1693 
1694 /*
1695  * CPU-stall kthread.  It waits as specified by stall_cpu_holdoff, then
1696  * induces a CPU stall for the time specified by stall_cpu.
1697  */
1698 static int rcu_torture_stall(void *args)
1699 {
1700 	int idx;
1701 	unsigned long stop_at;
1702 
1703 	VERBOSE_TOROUT_STRING("rcu_torture_stall task started");
1704 	if (stall_cpu_holdoff > 0) {
1705 		VERBOSE_TOROUT_STRING("rcu_torture_stall begin holdoff");
1706 		schedule_timeout_interruptible(stall_cpu_holdoff * HZ);
1707 		VERBOSE_TOROUT_STRING("rcu_torture_stall end holdoff");
1708 	}
1709 	if (!kthread_should_stop() && stall_gp_kthread > 0) {
1710 		VERBOSE_TOROUT_STRING("rcu_torture_stall begin GP stall");
1711 		rcu_gp_set_torture_wait(stall_gp_kthread * HZ);
1712 		for (idx = 0; idx < stall_gp_kthread + 2; idx++) {
1713 			if (kthread_should_stop())
1714 				break;
1715 			schedule_timeout_uninterruptible(HZ);
1716 		}
1717 	}
1718 	if (!kthread_should_stop() && stall_cpu > 0) {
1719 		VERBOSE_TOROUT_STRING("rcu_torture_stall begin CPU stall");
1720 		stop_at = ktime_get_seconds() + stall_cpu;
1721 		/* RCU CPU stall is expected behavior in following code. */
1722 		idx = cur_ops->readlock();
1723 		if (stall_cpu_irqsoff)
1724 			local_irq_disable();
1725 		else if (!stall_cpu_block)
1726 			preempt_disable();
1727 		pr_alert("rcu_torture_stall start on CPU %d.\n",
1728 			 raw_smp_processor_id());
1729 		while (ULONG_CMP_LT((unsigned long)ktime_get_seconds(),
1730 				    stop_at))
1731 			if (stall_cpu_block)
1732 				schedule_timeout_uninterruptible(HZ);
1733 		if (stall_cpu_irqsoff)
1734 			local_irq_enable();
1735 		else if (!stall_cpu_block)
1736 			preempt_enable();
1737 		cur_ops->readunlock(idx);
1738 	}
1739 	pr_alert("rcu_torture_stall end.\n");
1740 	torture_shutdown_absorb("rcu_torture_stall");
1741 	while (!kthread_should_stop())
1742 		schedule_timeout_interruptible(10 * HZ);
1743 	return 0;
1744 }
1745 
1746 /* Spawn CPU-stall kthread, if stall_cpu specified. */
1747 static int __init rcu_torture_stall_init(void)
1748 {
1749 	if (stall_cpu <= 0 && stall_gp_kthread <= 0)
1750 		return 0;
1751 	return torture_create_kthread(rcu_torture_stall, NULL, stall_task);
1752 }
1753 
1754 /* State structure for forward-progress self-propagating RCU callback. */
1755 struct fwd_cb_state {
1756 	struct rcu_head rh;
1757 	int stop;
1758 };
1759 
1760 /*
1761  * Forward-progress self-propagating RCU callback function.  Because
1762  * callbacks run from softirq, this function is an implicit RCU read-side
1763  * critical section.
1764  */
1765 static void rcu_torture_fwd_prog_cb(struct rcu_head *rhp)
1766 {
1767 	struct fwd_cb_state *fcsp = container_of(rhp, struct fwd_cb_state, rh);
1768 
1769 	if (READ_ONCE(fcsp->stop)) {
1770 		WRITE_ONCE(fcsp->stop, 2);
1771 		return;
1772 	}
1773 	cur_ops->call(&fcsp->rh, rcu_torture_fwd_prog_cb);
1774 }
1775 
1776 /* State for continuous-flood RCU callbacks. */
1777 struct rcu_fwd_cb {
1778 	struct rcu_head rh;
1779 	struct rcu_fwd_cb *rfc_next;
1780 	struct rcu_fwd *rfc_rfp;
1781 	int rfc_gps;
1782 };
1783 
1784 #define MAX_FWD_CB_JIFFIES	(8 * HZ) /* Maximum CB test duration. */
1785 #define MIN_FWD_CB_LAUNDERS	3	/* This many CB invocations to count. */
1786 #define MIN_FWD_CBS_LAUNDERED	100	/* Number of counted CBs. */
1787 #define FWD_CBS_HIST_DIV	10	/* Histogram buckets/second. */
1788 #define N_LAUNDERS_HIST (2 * MAX_FWD_CB_JIFFIES / (HZ / FWD_CBS_HIST_DIV))
1789 
1790 struct rcu_launder_hist {
1791 	long n_launders;
1792 	unsigned long launder_gp_seq;
1793 };
1794 
1795 struct rcu_fwd {
1796 	spinlock_t rcu_fwd_lock;
1797 	struct rcu_fwd_cb *rcu_fwd_cb_head;
1798 	struct rcu_fwd_cb **rcu_fwd_cb_tail;
1799 	long n_launders_cb;
1800 	unsigned long rcu_fwd_startat;
1801 	struct rcu_launder_hist n_launders_hist[N_LAUNDERS_HIST];
1802 	unsigned long rcu_launder_gp_seq_start;
1803 };
1804 
1805 static struct rcu_fwd *rcu_fwds;
1806 static bool rcu_fwd_emergency_stop;
1807 
1808 static void rcu_torture_fwd_cb_hist(struct rcu_fwd *rfp)
1809 {
1810 	unsigned long gps;
1811 	unsigned long gps_old;
1812 	int i;
1813 	int j;
1814 
1815 	for (i = ARRAY_SIZE(rfp->n_launders_hist) - 1; i > 0; i--)
1816 		if (rfp->n_launders_hist[i].n_launders > 0)
1817 			break;
1818 	pr_alert("%s: Callback-invocation histogram (duration %lu jiffies):",
1819 		 __func__, jiffies - rfp->rcu_fwd_startat);
1820 	gps_old = rfp->rcu_launder_gp_seq_start;
1821 	for (j = 0; j <= i; j++) {
1822 		gps = rfp->n_launders_hist[j].launder_gp_seq;
1823 		pr_cont(" %ds/%d: %ld:%ld",
1824 			j + 1, FWD_CBS_HIST_DIV,
1825 			rfp->n_launders_hist[j].n_launders,
1826 			rcutorture_seq_diff(gps, gps_old));
1827 		gps_old = gps;
1828 	}
1829 	pr_cont("\n");
1830 }
1831 
1832 /* Callback function for continuous-flood RCU callbacks. */
1833 static void rcu_torture_fwd_cb_cr(struct rcu_head *rhp)
1834 {
1835 	unsigned long flags;
1836 	int i;
1837 	struct rcu_fwd_cb *rfcp = container_of(rhp, struct rcu_fwd_cb, rh);
1838 	struct rcu_fwd_cb **rfcpp;
1839 	struct rcu_fwd *rfp = rfcp->rfc_rfp;
1840 
1841 	rfcp->rfc_next = NULL;
1842 	rfcp->rfc_gps++;
1843 	spin_lock_irqsave(&rfp->rcu_fwd_lock, flags);
1844 	rfcpp = rfp->rcu_fwd_cb_tail;
1845 	rfp->rcu_fwd_cb_tail = &rfcp->rfc_next;
1846 	WRITE_ONCE(*rfcpp, rfcp);
1847 	WRITE_ONCE(rfp->n_launders_cb, rfp->n_launders_cb + 1);
1848 	i = ((jiffies - rfp->rcu_fwd_startat) / (HZ / FWD_CBS_HIST_DIV));
1849 	if (i >= ARRAY_SIZE(rfp->n_launders_hist))
1850 		i = ARRAY_SIZE(rfp->n_launders_hist) - 1;
1851 	rfp->n_launders_hist[i].n_launders++;
1852 	rfp->n_launders_hist[i].launder_gp_seq = cur_ops->get_gp_seq();
1853 	spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
1854 }
1855 
1856 // Give the scheduler a chance, even on nohz_full CPUs.
1857 static void rcu_torture_fwd_prog_cond_resched(unsigned long iter)
1858 {
1859 	if (IS_ENABLED(CONFIG_PREEMPTION) && IS_ENABLED(CONFIG_NO_HZ_FULL)) {
1860 		// Real call_rcu() floods hit userspace, so emulate that.
1861 		if (need_resched() || (iter & 0xfff))
1862 			schedule();
1863 		return;
1864 	}
1865 	// No userspace emulation: CB invocation throttles call_rcu()
1866 	cond_resched();
1867 }
1868 
1869 /*
1870  * Free all callbacks on the rcu_fwd_cb_head list, either because the
1871  * test is over or because we hit an OOM event.
1872  */
1873 static unsigned long rcu_torture_fwd_prog_cbfree(struct rcu_fwd *rfp)
1874 {
1875 	unsigned long flags;
1876 	unsigned long freed = 0;
1877 	struct rcu_fwd_cb *rfcp;
1878 
1879 	for (;;) {
1880 		spin_lock_irqsave(&rfp->rcu_fwd_lock, flags);
1881 		rfcp = rfp->rcu_fwd_cb_head;
1882 		if (!rfcp) {
1883 			spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
1884 			break;
1885 		}
1886 		rfp->rcu_fwd_cb_head = rfcp->rfc_next;
1887 		if (!rfp->rcu_fwd_cb_head)
1888 			rfp->rcu_fwd_cb_tail = &rfp->rcu_fwd_cb_head;
1889 		spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
1890 		kfree(rfcp);
1891 		freed++;
1892 		rcu_torture_fwd_prog_cond_resched(freed);
1893 		if (tick_nohz_full_enabled()) {
1894 			local_irq_save(flags);
1895 			rcu_momentary_dyntick_idle();
1896 			local_irq_restore(flags);
1897 		}
1898 	}
1899 	return freed;
1900 }
1901 
1902 /* Carry out need_resched()/cond_resched() forward-progress testing. */
1903 static void rcu_torture_fwd_prog_nr(struct rcu_fwd *rfp,
1904 				    int *tested, int *tested_tries)
1905 {
1906 	unsigned long cver;
1907 	unsigned long dur;
1908 	struct fwd_cb_state fcs;
1909 	unsigned long gps;
1910 	int idx;
1911 	int sd;
1912 	int sd4;
1913 	bool selfpropcb = false;
1914 	unsigned long stopat;
1915 	static DEFINE_TORTURE_RANDOM(trs);
1916 
1917 	if  (cur_ops->call && cur_ops->sync && cur_ops->cb_barrier) {
1918 		init_rcu_head_on_stack(&fcs.rh);
1919 		selfpropcb = true;
1920 	}
1921 
1922 	/* Tight loop containing cond_resched(). */
1923 	WRITE_ONCE(rcu_fwd_cb_nodelay, true);
1924 	cur_ops->sync(); /* Later readers see above write. */
1925 	if  (selfpropcb) {
1926 		WRITE_ONCE(fcs.stop, 0);
1927 		cur_ops->call(&fcs.rh, rcu_torture_fwd_prog_cb);
1928 	}
1929 	cver = READ_ONCE(rcu_torture_current_version);
1930 	gps = cur_ops->get_gp_seq();
1931 	sd = cur_ops->stall_dur() + 1;
1932 	sd4 = (sd + fwd_progress_div - 1) / fwd_progress_div;
1933 	dur = sd4 + torture_random(&trs) % (sd - sd4);
1934 	WRITE_ONCE(rfp->rcu_fwd_startat, jiffies);
1935 	stopat = rfp->rcu_fwd_startat + dur;
1936 	while (time_before(jiffies, stopat) &&
1937 	       !shutdown_time_arrived() &&
1938 	       !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
1939 		idx = cur_ops->readlock();
1940 		udelay(10);
1941 		cur_ops->readunlock(idx);
1942 		if (!fwd_progress_need_resched || need_resched())
1943 			cond_resched();
1944 	}
1945 	(*tested_tries)++;
1946 	if (!time_before(jiffies, stopat) &&
1947 	    !shutdown_time_arrived() &&
1948 	    !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
1949 		(*tested)++;
1950 		cver = READ_ONCE(rcu_torture_current_version) - cver;
1951 		gps = rcutorture_seq_diff(cur_ops->get_gp_seq(), gps);
1952 		WARN_ON(!cver && gps < 2);
1953 		pr_alert("%s: Duration %ld cver %ld gps %ld\n", __func__, dur, cver, gps);
1954 	}
1955 	if (selfpropcb) {
1956 		WRITE_ONCE(fcs.stop, 1);
1957 		cur_ops->sync(); /* Wait for running CB to complete. */
1958 		cur_ops->cb_barrier(); /* Wait for queued callbacks. */
1959 	}
1960 
1961 	if (selfpropcb) {
1962 		WARN_ON(READ_ONCE(fcs.stop) != 2);
1963 		destroy_rcu_head_on_stack(&fcs.rh);
1964 	}
1965 	schedule_timeout_uninterruptible(HZ / 10); /* Let kthreads recover. */
1966 	WRITE_ONCE(rcu_fwd_cb_nodelay, false);
1967 }
1968 
1969 /* Carry out call_rcu() forward-progress testing. */
1970 static void rcu_torture_fwd_prog_cr(struct rcu_fwd *rfp)
1971 {
1972 	unsigned long cver;
1973 	unsigned long flags;
1974 	unsigned long gps;
1975 	int i;
1976 	long n_launders;
1977 	long n_launders_cb_snap;
1978 	long n_launders_sa;
1979 	long n_max_cbs;
1980 	long n_max_gps;
1981 	struct rcu_fwd_cb *rfcp;
1982 	struct rcu_fwd_cb *rfcpn;
1983 	unsigned long stopat;
1984 	unsigned long stoppedat;
1985 
1986 	if (READ_ONCE(rcu_fwd_emergency_stop))
1987 		return; /* Get out of the way quickly, no GP wait! */
1988 	if (!cur_ops->call)
1989 		return; /* Can't do call_rcu() fwd prog without ->call. */
1990 
1991 	/* Loop continuously posting RCU callbacks. */
1992 	WRITE_ONCE(rcu_fwd_cb_nodelay, true);
1993 	cur_ops->sync(); /* Later readers see above write. */
1994 	WRITE_ONCE(rfp->rcu_fwd_startat, jiffies);
1995 	stopat = rfp->rcu_fwd_startat + MAX_FWD_CB_JIFFIES;
1996 	n_launders = 0;
1997 	rfp->n_launders_cb = 0; // Hoist initialization for multi-kthread
1998 	n_launders_sa = 0;
1999 	n_max_cbs = 0;
2000 	n_max_gps = 0;
2001 	for (i = 0; i < ARRAY_SIZE(rfp->n_launders_hist); i++)
2002 		rfp->n_launders_hist[i].n_launders = 0;
2003 	cver = READ_ONCE(rcu_torture_current_version);
2004 	gps = cur_ops->get_gp_seq();
2005 	rfp->rcu_launder_gp_seq_start = gps;
2006 	tick_dep_set_task(current, TICK_DEP_BIT_RCU);
2007 	while (time_before(jiffies, stopat) &&
2008 	       !shutdown_time_arrived() &&
2009 	       !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
2010 		rfcp = READ_ONCE(rfp->rcu_fwd_cb_head);
2011 		rfcpn = NULL;
2012 		if (rfcp)
2013 			rfcpn = READ_ONCE(rfcp->rfc_next);
2014 		if (rfcpn) {
2015 			if (rfcp->rfc_gps >= MIN_FWD_CB_LAUNDERS &&
2016 			    ++n_max_gps >= MIN_FWD_CBS_LAUNDERED)
2017 				break;
2018 			rfp->rcu_fwd_cb_head = rfcpn;
2019 			n_launders++;
2020 			n_launders_sa++;
2021 		} else {
2022 			rfcp = kmalloc(sizeof(*rfcp), GFP_KERNEL);
2023 			if (WARN_ON_ONCE(!rfcp)) {
2024 				schedule_timeout_interruptible(1);
2025 				continue;
2026 			}
2027 			n_max_cbs++;
2028 			n_launders_sa = 0;
2029 			rfcp->rfc_gps = 0;
2030 			rfcp->rfc_rfp = rfp;
2031 		}
2032 		cur_ops->call(&rfcp->rh, rcu_torture_fwd_cb_cr);
2033 		rcu_torture_fwd_prog_cond_resched(n_launders + n_max_cbs);
2034 		if (tick_nohz_full_enabled()) {
2035 			local_irq_save(flags);
2036 			rcu_momentary_dyntick_idle();
2037 			local_irq_restore(flags);
2038 		}
2039 	}
2040 	stoppedat = jiffies;
2041 	n_launders_cb_snap = READ_ONCE(rfp->n_launders_cb);
2042 	cver = READ_ONCE(rcu_torture_current_version) - cver;
2043 	gps = rcutorture_seq_diff(cur_ops->get_gp_seq(), gps);
2044 	cur_ops->cb_barrier(); /* Wait for callbacks to be invoked. */
2045 	(void)rcu_torture_fwd_prog_cbfree(rfp);
2046 
2047 	if (!torture_must_stop() && !READ_ONCE(rcu_fwd_emergency_stop) &&
2048 	    !shutdown_time_arrived()) {
2049 		WARN_ON(n_max_gps < MIN_FWD_CBS_LAUNDERED);
2050 		pr_alert("%s Duration %lu barrier: %lu pending %ld n_launders: %ld n_launders_sa: %ld n_max_gps: %ld n_max_cbs: %ld cver %ld gps %ld\n",
2051 			 __func__,
2052 			 stoppedat - rfp->rcu_fwd_startat, jiffies - stoppedat,
2053 			 n_launders + n_max_cbs - n_launders_cb_snap,
2054 			 n_launders, n_launders_sa,
2055 			 n_max_gps, n_max_cbs, cver, gps);
2056 		rcu_torture_fwd_cb_hist(rfp);
2057 	}
2058 	schedule_timeout_uninterruptible(HZ); /* Let CBs drain. */
2059 	tick_dep_clear_task(current, TICK_DEP_BIT_RCU);
2060 	WRITE_ONCE(rcu_fwd_cb_nodelay, false);
2061 }
2062 
2063 
2064 /*
2065  * OOM notifier, but this only prints diagnostic information for the
2066  * current forward-progress test.
2067  */
2068 static int rcutorture_oom_notify(struct notifier_block *self,
2069 				 unsigned long notused, void *nfreed)
2070 {
2071 	struct rcu_fwd *rfp = rcu_fwds;
2072 
2073 	WARN(1, "%s invoked upon OOM during forward-progress testing.\n",
2074 	     __func__);
2075 	rcu_torture_fwd_cb_hist(rfp);
2076 	rcu_fwd_progress_check(1 + (jiffies - READ_ONCE(rfp->rcu_fwd_startat)) / 2);
2077 	WRITE_ONCE(rcu_fwd_emergency_stop, true);
2078 	smp_mb(); /* Emergency stop before free and wait to avoid hangs. */
2079 	pr_info("%s: Freed %lu RCU callbacks.\n",
2080 		__func__, rcu_torture_fwd_prog_cbfree(rfp));
2081 	rcu_barrier();
2082 	pr_info("%s: Freed %lu RCU callbacks.\n",
2083 		__func__, rcu_torture_fwd_prog_cbfree(rfp));
2084 	rcu_barrier();
2085 	pr_info("%s: Freed %lu RCU callbacks.\n",
2086 		__func__, rcu_torture_fwd_prog_cbfree(rfp));
2087 	smp_mb(); /* Frees before return to avoid redoing OOM. */
2088 	(*(unsigned long *)nfreed)++; /* Forward progress CBs freed! */
2089 	pr_info("%s returning after OOM processing.\n", __func__);
2090 	return NOTIFY_OK;
2091 }
2092 
2093 static struct notifier_block rcutorture_oom_nb = {
2094 	.notifier_call = rcutorture_oom_notify
2095 };
2096 
2097 /* Carry out grace-period forward-progress testing. */
2098 static int rcu_torture_fwd_prog(void *args)
2099 {
2100 	struct rcu_fwd *rfp = args;
2101 	int tested = 0;
2102 	int tested_tries = 0;
2103 
2104 	VERBOSE_TOROUT_STRING("rcu_torture_fwd_progress task started");
2105 	rcu_bind_current_to_nocb();
2106 	if (!IS_ENABLED(CONFIG_SMP) || !IS_ENABLED(CONFIG_RCU_BOOST))
2107 		set_user_nice(current, MAX_NICE);
2108 	do {
2109 		schedule_timeout_interruptible(fwd_progress_holdoff * HZ);
2110 		WRITE_ONCE(rcu_fwd_emergency_stop, false);
2111 		register_oom_notifier(&rcutorture_oom_nb);
2112 		if (!IS_ENABLED(CONFIG_TINY_RCU) ||
2113 		    rcu_inkernel_boot_has_ended())
2114 			rcu_torture_fwd_prog_nr(rfp, &tested, &tested_tries);
2115 		if (rcu_inkernel_boot_has_ended())
2116 			rcu_torture_fwd_prog_cr(rfp);
2117 		unregister_oom_notifier(&rcutorture_oom_nb);
2118 
2119 		/* Avoid slow periods, better to test when busy. */
2120 		stutter_wait("rcu_torture_fwd_prog");
2121 	} while (!torture_must_stop());
2122 	/* Short runs might not contain a valid forward-progress attempt. */
2123 	WARN_ON(!tested && tested_tries >= 5);
2124 	pr_alert("%s: tested %d tested_tries %d\n", __func__, tested, tested_tries);
2125 	torture_kthread_stopping("rcu_torture_fwd_prog");
2126 	return 0;
2127 }
2128 
2129 /* If forward-progress checking is requested and feasible, spawn the thread. */
2130 static int __init rcu_torture_fwd_prog_init(void)
2131 {
2132 	struct rcu_fwd *rfp;
2133 
2134 	if (!fwd_progress)
2135 		return 0; /* Not requested, so don't do it. */
2136 	if (!cur_ops->stall_dur || cur_ops->stall_dur() <= 0 ||
2137 	    cur_ops == &rcu_busted_ops) {
2138 		VERBOSE_TOROUT_STRING("rcu_torture_fwd_prog_init: Disabled, unsupported by RCU flavor under test");
2139 		return 0;
2140 	}
2141 	if (stall_cpu > 0) {
2142 		VERBOSE_TOROUT_STRING("rcu_torture_fwd_prog_init: Disabled, conflicts with CPU-stall testing");
2143 		if (IS_MODULE(CONFIG_RCU_TORTURE_TESTS))
2144 			return -EINVAL; /* In module, can fail back to user. */
2145 		WARN_ON(1); /* Make sure rcutorture notices conflict. */
2146 		return 0;
2147 	}
2148 	if (fwd_progress_holdoff <= 0)
2149 		fwd_progress_holdoff = 1;
2150 	if (fwd_progress_div <= 0)
2151 		fwd_progress_div = 4;
2152 	rfp = kzalloc(sizeof(*rfp), GFP_KERNEL);
2153 	if (!rfp)
2154 		return -ENOMEM;
2155 	spin_lock_init(&rfp->rcu_fwd_lock);
2156 	rfp->rcu_fwd_cb_tail = &rfp->rcu_fwd_cb_head;
2157 	return torture_create_kthread(rcu_torture_fwd_prog, rfp, fwd_prog_task);
2158 }
2159 
2160 /* Callback function for RCU barrier testing. */
2161 static void rcu_torture_barrier_cbf(struct rcu_head *rcu)
2162 {
2163 	atomic_inc(&barrier_cbs_invoked);
2164 }
2165 
2166 /* IPI handler to get callback posted on desired CPU, if online. */
2167 static void rcu_torture_barrier1cb(void *rcu_void)
2168 {
2169 	struct rcu_head *rhp = rcu_void;
2170 
2171 	cur_ops->call(rhp, rcu_torture_barrier_cbf);
2172 }
2173 
2174 /* kthread function to register callbacks used to test RCU barriers. */
2175 static int rcu_torture_barrier_cbs(void *arg)
2176 {
2177 	long myid = (long)arg;
2178 	bool lastphase = 0;
2179 	bool newphase;
2180 	struct rcu_head rcu;
2181 
2182 	init_rcu_head_on_stack(&rcu);
2183 	VERBOSE_TOROUT_STRING("rcu_torture_barrier_cbs task started");
2184 	set_user_nice(current, MAX_NICE);
2185 	do {
2186 		wait_event(barrier_cbs_wq[myid],
2187 			   (newphase =
2188 			    smp_load_acquire(&barrier_phase)) != lastphase ||
2189 			   torture_must_stop());
2190 		lastphase = newphase;
2191 		if (torture_must_stop())
2192 			break;
2193 		/*
2194 		 * The above smp_load_acquire() ensures barrier_phase load
2195 		 * is ordered before the following ->call().
2196 		 */
2197 		if (smp_call_function_single(myid, rcu_torture_barrier1cb,
2198 					     &rcu, 1)) {
2199 			// IPI failed, so use direct call from current CPU.
2200 			cur_ops->call(&rcu, rcu_torture_barrier_cbf);
2201 		}
2202 		if (atomic_dec_and_test(&barrier_cbs_count))
2203 			wake_up(&barrier_wq);
2204 	} while (!torture_must_stop());
2205 	if (cur_ops->cb_barrier != NULL)
2206 		cur_ops->cb_barrier();
2207 	destroy_rcu_head_on_stack(&rcu);
2208 	torture_kthread_stopping("rcu_torture_barrier_cbs");
2209 	return 0;
2210 }
2211 
2212 /* kthread function to drive and coordinate RCU barrier testing. */
2213 static int rcu_torture_barrier(void *arg)
2214 {
2215 	int i;
2216 
2217 	VERBOSE_TOROUT_STRING("rcu_torture_barrier task starting");
2218 	do {
2219 		atomic_set(&barrier_cbs_invoked, 0);
2220 		atomic_set(&barrier_cbs_count, n_barrier_cbs);
2221 		/* Ensure barrier_phase ordered after prior assignments. */
2222 		smp_store_release(&barrier_phase, !barrier_phase);
2223 		for (i = 0; i < n_barrier_cbs; i++)
2224 			wake_up(&barrier_cbs_wq[i]);
2225 		wait_event(barrier_wq,
2226 			   atomic_read(&barrier_cbs_count) == 0 ||
2227 			   torture_must_stop());
2228 		if (torture_must_stop())
2229 			break;
2230 		n_barrier_attempts++;
2231 		cur_ops->cb_barrier(); /* Implies smp_mb() for wait_event(). */
2232 		if (atomic_read(&barrier_cbs_invoked) != n_barrier_cbs) {
2233 			n_rcu_torture_barrier_error++;
2234 			pr_err("barrier_cbs_invoked = %d, n_barrier_cbs = %d\n",
2235 			       atomic_read(&barrier_cbs_invoked),
2236 			       n_barrier_cbs);
2237 			WARN_ON(1);
2238 			// Wait manually for the remaining callbacks
2239 			i = 0;
2240 			do {
2241 				if (WARN_ON(i++ > HZ))
2242 					i = INT_MIN;
2243 				schedule_timeout_interruptible(1);
2244 				cur_ops->cb_barrier();
2245 			} while (atomic_read(&barrier_cbs_invoked) !=
2246 				 n_barrier_cbs &&
2247 				 !torture_must_stop());
2248 			smp_mb(); // Can't trust ordering if broken.
2249 			if (!torture_must_stop())
2250 				pr_err("Recovered: barrier_cbs_invoked = %d\n",
2251 				       atomic_read(&barrier_cbs_invoked));
2252 		} else {
2253 			n_barrier_successes++;
2254 		}
2255 		schedule_timeout_interruptible(HZ / 10);
2256 	} while (!torture_must_stop());
2257 	torture_kthread_stopping("rcu_torture_barrier");
2258 	return 0;
2259 }
2260 
2261 /* Initialize RCU barrier testing. */
2262 static int rcu_torture_barrier_init(void)
2263 {
2264 	int i;
2265 	int ret;
2266 
2267 	if (n_barrier_cbs <= 0)
2268 		return 0;
2269 	if (cur_ops->call == NULL || cur_ops->cb_barrier == NULL) {
2270 		pr_alert("%s" TORTURE_FLAG
2271 			 " Call or barrier ops missing for %s,\n",
2272 			 torture_type, cur_ops->name);
2273 		pr_alert("%s" TORTURE_FLAG
2274 			 " RCU barrier testing omitted from run.\n",
2275 			 torture_type);
2276 		return 0;
2277 	}
2278 	atomic_set(&barrier_cbs_count, 0);
2279 	atomic_set(&barrier_cbs_invoked, 0);
2280 	barrier_cbs_tasks =
2281 		kcalloc(n_barrier_cbs, sizeof(barrier_cbs_tasks[0]),
2282 			GFP_KERNEL);
2283 	barrier_cbs_wq =
2284 		kcalloc(n_barrier_cbs, sizeof(barrier_cbs_wq[0]), GFP_KERNEL);
2285 	if (barrier_cbs_tasks == NULL || !barrier_cbs_wq)
2286 		return -ENOMEM;
2287 	for (i = 0; i < n_barrier_cbs; i++) {
2288 		init_waitqueue_head(&barrier_cbs_wq[i]);
2289 		ret = torture_create_kthread(rcu_torture_barrier_cbs,
2290 					     (void *)(long)i,
2291 					     barrier_cbs_tasks[i]);
2292 		if (ret)
2293 			return ret;
2294 	}
2295 	return torture_create_kthread(rcu_torture_barrier, NULL, barrier_task);
2296 }
2297 
2298 /* Clean up after RCU barrier testing. */
2299 static void rcu_torture_barrier_cleanup(void)
2300 {
2301 	int i;
2302 
2303 	torture_stop_kthread(rcu_torture_barrier, barrier_task);
2304 	if (barrier_cbs_tasks != NULL) {
2305 		for (i = 0; i < n_barrier_cbs; i++)
2306 			torture_stop_kthread(rcu_torture_barrier_cbs,
2307 					     barrier_cbs_tasks[i]);
2308 		kfree(barrier_cbs_tasks);
2309 		barrier_cbs_tasks = NULL;
2310 	}
2311 	if (barrier_cbs_wq != NULL) {
2312 		kfree(barrier_cbs_wq);
2313 		barrier_cbs_wq = NULL;
2314 	}
2315 }
2316 
2317 static bool rcu_torture_can_boost(void)
2318 {
2319 	static int boost_warn_once;
2320 	int prio;
2321 
2322 	if (!(test_boost == 1 && cur_ops->can_boost) && test_boost != 2)
2323 		return false;
2324 
2325 	prio = rcu_get_gp_kthreads_prio();
2326 	if (!prio)
2327 		return false;
2328 
2329 	if (prio < 2) {
2330 		if (boost_warn_once  == 1)
2331 			return false;
2332 
2333 		pr_alert("%s: WARN: RCU kthread priority too low to test boosting.  Skipping RCU boost test. Try passing rcutree.kthread_prio > 1 on the kernel command line.\n", KBUILD_MODNAME);
2334 		boost_warn_once = 1;
2335 		return false;
2336 	}
2337 
2338 	return true;
2339 }
2340 
2341 static enum cpuhp_state rcutor_hp;
2342 
2343 static void
2344 rcu_torture_cleanup(void)
2345 {
2346 	int firsttime;
2347 	int flags = 0;
2348 	unsigned long gp_seq = 0;
2349 	int i;
2350 
2351 	if (torture_cleanup_begin()) {
2352 		if (cur_ops->cb_barrier != NULL)
2353 			cur_ops->cb_barrier();
2354 		return;
2355 	}
2356 	if (!cur_ops) {
2357 		torture_cleanup_end();
2358 		return;
2359 	}
2360 
2361 	show_rcu_gp_kthreads();
2362 	rcu_torture_barrier_cleanup();
2363 	torture_stop_kthread(rcu_torture_fwd_prog, fwd_prog_task);
2364 	torture_stop_kthread(rcu_torture_stall, stall_task);
2365 	torture_stop_kthread(rcu_torture_writer, writer_task);
2366 
2367 	if (reader_tasks) {
2368 		for (i = 0; i < nrealreaders; i++)
2369 			torture_stop_kthread(rcu_torture_reader,
2370 					     reader_tasks[i]);
2371 		kfree(reader_tasks);
2372 	}
2373 	rcu_torture_current = NULL;
2374 
2375 	if (fakewriter_tasks) {
2376 		for (i = 0; i < nfakewriters; i++) {
2377 			torture_stop_kthread(rcu_torture_fakewriter,
2378 					     fakewriter_tasks[i]);
2379 		}
2380 		kfree(fakewriter_tasks);
2381 		fakewriter_tasks = NULL;
2382 	}
2383 
2384 	rcutorture_get_gp_data(cur_ops->ttype, &flags, &gp_seq);
2385 	srcutorture_get_gp_data(cur_ops->ttype, srcu_ctlp, &flags, &gp_seq);
2386 	pr_alert("%s:  End-test grace-period state: g%lu f%#x\n",
2387 		 cur_ops->name, gp_seq, flags);
2388 	torture_stop_kthread(rcu_torture_stats, stats_task);
2389 	torture_stop_kthread(rcu_torture_fqs, fqs_task);
2390 	if (rcu_torture_can_boost())
2391 		cpuhp_remove_state(rcutor_hp);
2392 
2393 	/*
2394 	 * Wait for all RCU callbacks to fire, then do torture-type-specific
2395 	 * cleanup operations.
2396 	 */
2397 	if (cur_ops->cb_barrier != NULL)
2398 		cur_ops->cb_barrier();
2399 	if (cur_ops->cleanup != NULL)
2400 		cur_ops->cleanup();
2401 
2402 	rcu_torture_stats_print();  /* -After- the stats thread is stopped! */
2403 
2404 	if (err_segs_recorded) {
2405 		pr_alert("Failure/close-call rcutorture reader segments:\n");
2406 		if (rt_read_nsegs == 0)
2407 			pr_alert("\t: No segments recorded!!!\n");
2408 		firsttime = 1;
2409 		for (i = 0; i < rt_read_nsegs; i++) {
2410 			pr_alert("\t%d: %#x ", i, err_segs[i].rt_readstate);
2411 			if (err_segs[i].rt_delay_jiffies != 0) {
2412 				pr_cont("%s%ldjiffies", firsttime ? "" : "+",
2413 					err_segs[i].rt_delay_jiffies);
2414 				firsttime = 0;
2415 			}
2416 			if (err_segs[i].rt_delay_ms != 0) {
2417 				pr_cont("%s%ldms", firsttime ? "" : "+",
2418 					err_segs[i].rt_delay_ms);
2419 				firsttime = 0;
2420 			}
2421 			if (err_segs[i].rt_delay_us != 0) {
2422 				pr_cont("%s%ldus", firsttime ? "" : "+",
2423 					err_segs[i].rt_delay_us);
2424 				firsttime = 0;
2425 			}
2426 			pr_cont("%s\n",
2427 				err_segs[i].rt_preempted ? "preempted" : "");
2428 
2429 		}
2430 	}
2431 	if (atomic_read(&n_rcu_torture_error) || n_rcu_torture_barrier_error)
2432 		rcu_torture_print_module_parms(cur_ops, "End of test: FAILURE");
2433 	else if (torture_onoff_failures())
2434 		rcu_torture_print_module_parms(cur_ops,
2435 					       "End of test: RCU_HOTPLUG");
2436 	else
2437 		rcu_torture_print_module_parms(cur_ops, "End of test: SUCCESS");
2438 	torture_cleanup_end();
2439 }
2440 
2441 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
2442 static void rcu_torture_leak_cb(struct rcu_head *rhp)
2443 {
2444 }
2445 
2446 static void rcu_torture_err_cb(struct rcu_head *rhp)
2447 {
2448 	/*
2449 	 * This -might- happen due to race conditions, but is unlikely.
2450 	 * The scenario that leads to this happening is that the
2451 	 * first of the pair of duplicate callbacks is queued,
2452 	 * someone else starts a grace period that includes that
2453 	 * callback, then the second of the pair must wait for the
2454 	 * next grace period.  Unlikely, but can happen.  If it
2455 	 * does happen, the debug-objects subsystem won't have splatted.
2456 	 */
2457 	pr_alert("%s: duplicated callback was invoked.\n", KBUILD_MODNAME);
2458 }
2459 #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
2460 
2461 /*
2462  * Verify that double-free causes debug-objects to complain, but only
2463  * if CONFIG_DEBUG_OBJECTS_RCU_HEAD=y.  Otherwise, say that the test
2464  * cannot be carried out.
2465  */
2466 static void rcu_test_debug_objects(void)
2467 {
2468 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
2469 	struct rcu_head rh1;
2470 	struct rcu_head rh2;
2471 
2472 	init_rcu_head_on_stack(&rh1);
2473 	init_rcu_head_on_stack(&rh2);
2474 	pr_alert("%s: WARN: Duplicate call_rcu() test starting.\n", KBUILD_MODNAME);
2475 
2476 	/* Try to queue the rh2 pair of callbacks for the same grace period. */
2477 	preempt_disable(); /* Prevent preemption from interrupting test. */
2478 	rcu_read_lock(); /* Make it impossible to finish a grace period. */
2479 	call_rcu(&rh1, rcu_torture_leak_cb); /* Start grace period. */
2480 	local_irq_disable(); /* Make it harder to start a new grace period. */
2481 	call_rcu(&rh2, rcu_torture_leak_cb);
2482 	call_rcu(&rh2, rcu_torture_err_cb); /* Duplicate callback. */
2483 	local_irq_enable();
2484 	rcu_read_unlock();
2485 	preempt_enable();
2486 
2487 	/* Wait for them all to get done so we can safely return. */
2488 	rcu_barrier();
2489 	pr_alert("%s: WARN: Duplicate call_rcu() test complete.\n", KBUILD_MODNAME);
2490 	destroy_rcu_head_on_stack(&rh1);
2491 	destroy_rcu_head_on_stack(&rh2);
2492 #else /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
2493 	pr_alert("%s: !CONFIG_DEBUG_OBJECTS_RCU_HEAD, not testing duplicate call_rcu()\n", KBUILD_MODNAME);
2494 #endif /* #else #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
2495 }
2496 
2497 static void rcutorture_sync(void)
2498 {
2499 	static unsigned long n;
2500 
2501 	if (cur_ops->sync && !(++n & 0xfff))
2502 		cur_ops->sync();
2503 }
2504 
2505 static int __init
2506 rcu_torture_init(void)
2507 {
2508 	long i;
2509 	int cpu;
2510 	int firsterr = 0;
2511 	static struct rcu_torture_ops *torture_ops[] = {
2512 		&rcu_ops, &rcu_busted_ops, &srcu_ops, &srcud_ops,
2513 		&busted_srcud_ops, &tasks_ops, &tasks_rude_ops,
2514 		&tasks_tracing_ops, &trivial_ops,
2515 	};
2516 
2517 	if (!torture_init_begin(torture_type, verbose))
2518 		return -EBUSY;
2519 
2520 	/* Process args and tell the world that the torturer is on the job. */
2521 	for (i = 0; i < ARRAY_SIZE(torture_ops); i++) {
2522 		cur_ops = torture_ops[i];
2523 		if (strcmp(torture_type, cur_ops->name) == 0)
2524 			break;
2525 	}
2526 	if (i == ARRAY_SIZE(torture_ops)) {
2527 		pr_alert("rcu-torture: invalid torture type: \"%s\"\n",
2528 			 torture_type);
2529 		pr_alert("rcu-torture types:");
2530 		for (i = 0; i < ARRAY_SIZE(torture_ops); i++)
2531 			pr_cont(" %s", torture_ops[i]->name);
2532 		pr_cont("\n");
2533 		WARN_ON(!IS_MODULE(CONFIG_RCU_TORTURE_TEST));
2534 		firsterr = -EINVAL;
2535 		cur_ops = NULL;
2536 		goto unwind;
2537 	}
2538 	if (cur_ops->fqs == NULL && fqs_duration != 0) {
2539 		pr_alert("rcu-torture: ->fqs NULL and non-zero fqs_duration, fqs disabled.\n");
2540 		fqs_duration = 0;
2541 	}
2542 	if (cur_ops->init)
2543 		cur_ops->init();
2544 
2545 	if (nreaders >= 0) {
2546 		nrealreaders = nreaders;
2547 	} else {
2548 		nrealreaders = num_online_cpus() - 2 - nreaders;
2549 		if (nrealreaders <= 0)
2550 			nrealreaders = 1;
2551 	}
2552 	rcu_torture_print_module_parms(cur_ops, "Start of test");
2553 
2554 	/* Set up the freelist. */
2555 
2556 	INIT_LIST_HEAD(&rcu_torture_freelist);
2557 	for (i = 0; i < ARRAY_SIZE(rcu_tortures); i++) {
2558 		rcu_tortures[i].rtort_mbtest = 0;
2559 		list_add_tail(&rcu_tortures[i].rtort_free,
2560 			      &rcu_torture_freelist);
2561 	}
2562 
2563 	/* Initialize the statistics so that each run gets its own numbers. */
2564 
2565 	rcu_torture_current = NULL;
2566 	rcu_torture_current_version = 0;
2567 	atomic_set(&n_rcu_torture_alloc, 0);
2568 	atomic_set(&n_rcu_torture_alloc_fail, 0);
2569 	atomic_set(&n_rcu_torture_free, 0);
2570 	atomic_set(&n_rcu_torture_mberror, 0);
2571 	atomic_set(&n_rcu_torture_error, 0);
2572 	n_rcu_torture_barrier_error = 0;
2573 	n_rcu_torture_boost_ktrerror = 0;
2574 	n_rcu_torture_boost_rterror = 0;
2575 	n_rcu_torture_boost_failure = 0;
2576 	n_rcu_torture_boosts = 0;
2577 	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
2578 		atomic_set(&rcu_torture_wcount[i], 0);
2579 	for_each_possible_cpu(cpu) {
2580 		for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
2581 			per_cpu(rcu_torture_count, cpu)[i] = 0;
2582 			per_cpu(rcu_torture_batch, cpu)[i] = 0;
2583 		}
2584 	}
2585 	err_segs_recorded = 0;
2586 	rt_read_nsegs = 0;
2587 
2588 	/* Start up the kthreads. */
2589 
2590 	firsterr = torture_create_kthread(rcu_torture_writer, NULL,
2591 					  writer_task);
2592 	if (firsterr)
2593 		goto unwind;
2594 	if (nfakewriters > 0) {
2595 		fakewriter_tasks = kcalloc(nfakewriters,
2596 					   sizeof(fakewriter_tasks[0]),
2597 					   GFP_KERNEL);
2598 		if (fakewriter_tasks == NULL) {
2599 			VERBOSE_TOROUT_ERRSTRING("out of memory");
2600 			firsterr = -ENOMEM;
2601 			goto unwind;
2602 		}
2603 	}
2604 	for (i = 0; i < nfakewriters; i++) {
2605 		firsterr = torture_create_kthread(rcu_torture_fakewriter,
2606 						  NULL, fakewriter_tasks[i]);
2607 		if (firsterr)
2608 			goto unwind;
2609 	}
2610 	reader_tasks = kcalloc(nrealreaders, sizeof(reader_tasks[0]),
2611 			       GFP_KERNEL);
2612 	if (reader_tasks == NULL) {
2613 		VERBOSE_TOROUT_ERRSTRING("out of memory");
2614 		firsterr = -ENOMEM;
2615 		goto unwind;
2616 	}
2617 	for (i = 0; i < nrealreaders; i++) {
2618 		firsterr = torture_create_kthread(rcu_torture_reader, (void *)i,
2619 						  reader_tasks[i]);
2620 		if (firsterr)
2621 			goto unwind;
2622 	}
2623 	if (stat_interval > 0) {
2624 		firsterr = torture_create_kthread(rcu_torture_stats, NULL,
2625 						  stats_task);
2626 		if (firsterr)
2627 			goto unwind;
2628 	}
2629 	if (test_no_idle_hz && shuffle_interval > 0) {
2630 		firsterr = torture_shuffle_init(shuffle_interval * HZ);
2631 		if (firsterr)
2632 			goto unwind;
2633 	}
2634 	if (stutter < 0)
2635 		stutter = 0;
2636 	if (stutter) {
2637 		int t;
2638 
2639 		t = cur_ops->stall_dur ? cur_ops->stall_dur() : stutter * HZ;
2640 		firsterr = torture_stutter_init(stutter * HZ, t);
2641 		if (firsterr)
2642 			goto unwind;
2643 	}
2644 	if (fqs_duration < 0)
2645 		fqs_duration = 0;
2646 	if (fqs_duration) {
2647 		/* Create the fqs thread */
2648 		firsterr = torture_create_kthread(rcu_torture_fqs, NULL,
2649 						  fqs_task);
2650 		if (firsterr)
2651 			goto unwind;
2652 	}
2653 	if (test_boost_interval < 1)
2654 		test_boost_interval = 1;
2655 	if (test_boost_duration < 2)
2656 		test_boost_duration = 2;
2657 	if (rcu_torture_can_boost()) {
2658 
2659 		boost_starttime = jiffies + test_boost_interval * HZ;
2660 
2661 		firsterr = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "RCU_TORTURE",
2662 					     rcutorture_booster_init,
2663 					     rcutorture_booster_cleanup);
2664 		if (firsterr < 0)
2665 			goto unwind;
2666 		rcutor_hp = firsterr;
2667 	}
2668 	shutdown_jiffies = jiffies + shutdown_secs * HZ;
2669 	firsterr = torture_shutdown_init(shutdown_secs, rcu_torture_cleanup);
2670 	if (firsterr)
2671 		goto unwind;
2672 	firsterr = torture_onoff_init(onoff_holdoff * HZ, onoff_interval,
2673 				      rcutorture_sync);
2674 	if (firsterr)
2675 		goto unwind;
2676 	firsterr = rcu_torture_stall_init();
2677 	if (firsterr)
2678 		goto unwind;
2679 	firsterr = rcu_torture_fwd_prog_init();
2680 	if (firsterr)
2681 		goto unwind;
2682 	firsterr = rcu_torture_barrier_init();
2683 	if (firsterr)
2684 		goto unwind;
2685 	if (object_debug)
2686 		rcu_test_debug_objects();
2687 	torture_init_end();
2688 	return 0;
2689 
2690 unwind:
2691 	torture_init_end();
2692 	rcu_torture_cleanup();
2693 	return firsterr;
2694 }
2695 
2696 module_init(rcu_torture_init);
2697 module_exit(rcu_torture_cleanup);
2698