xref: /openbmc/linux/kernel/rcu/rcutorture.c (revision b2765275)
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.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 
49 #include "rcu.h"
50 
51 MODULE_LICENSE("GPL");
52 MODULE_AUTHOR("Paul E. McKenney <paulmck@linux.ibm.com> and Josh Triplett <josh@joshtriplett.org>");
53 
54 
55 /* Bits for ->extendables field, extendables param, and related definitions. */
56 #define RCUTORTURE_RDR_SHIFT	 8	/* Put SRCU index in upper bits. */
57 #define RCUTORTURE_RDR_MASK	 ((1 << RCUTORTURE_RDR_SHIFT) - 1)
58 #define RCUTORTURE_RDR_BH	 0x01	/* Extend readers by disabling bh. */
59 #define RCUTORTURE_RDR_IRQ	 0x02	/*  ... disabling interrupts. */
60 #define RCUTORTURE_RDR_PREEMPT	 0x04	/*  ... disabling preemption. */
61 #define RCUTORTURE_RDR_RBH	 0x08	/*  ... rcu_read_lock_bh(). */
62 #define RCUTORTURE_RDR_SCHED	 0x10	/*  ... rcu_read_lock_sched(). */
63 #define RCUTORTURE_RDR_RCU	 0x20	/*  ... entering another RCU reader. */
64 #define RCUTORTURE_RDR_NBITS	 6	/* Number of bits defined above. */
65 #define RCUTORTURE_MAX_EXTEND	 \
66 	(RCUTORTURE_RDR_BH | RCUTORTURE_RDR_IRQ | RCUTORTURE_RDR_PREEMPT | \
67 	 RCUTORTURE_RDR_RBH | RCUTORTURE_RDR_SCHED)
68 #define RCUTORTURE_RDR_MAX_LOOPS 0x7	/* Maximum reader extensions. */
69 					/* Must be power of two minus one. */
70 #define RCUTORTURE_RDR_MAX_SEGS (RCUTORTURE_RDR_MAX_LOOPS + 3)
71 
72 torture_param(int, extendables, RCUTORTURE_MAX_EXTEND,
73 	      "Extend readers by disabling bh (1), irqs (2), or preempt (4)");
74 torture_param(int, fqs_duration, 0,
75 	      "Duration of fqs bursts (us), 0 to disable");
76 torture_param(int, fqs_holdoff, 0, "Holdoff time within fqs bursts (us)");
77 torture_param(int, fqs_stutter, 3, "Wait time between fqs bursts (s)");
78 torture_param(bool, fwd_progress, 1, "Test grace-period forward progress");
79 torture_param(int, fwd_progress_div, 4, "Fraction of CPU stall to wait");
80 torture_param(int, fwd_progress_holdoff, 60,
81 	      "Time between forward-progress tests (s)");
82 torture_param(bool, fwd_progress_need_resched, 1,
83 	      "Hide cond_resched() behind need_resched()");
84 torture_param(bool, gp_cond, false, "Use conditional/async GP wait primitives");
85 torture_param(bool, gp_exp, false, "Use expedited GP wait primitives");
86 torture_param(bool, gp_normal, false,
87 	     "Use normal (non-expedited) GP wait primitives");
88 torture_param(bool, gp_sync, false, "Use synchronous GP wait primitives");
89 torture_param(int, irqreader, 1, "Allow RCU readers from irq handlers");
90 torture_param(int, n_barrier_cbs, 0,
91 	     "# of callbacks/kthreads for barrier testing");
92 torture_param(int, nfakewriters, 4, "Number of RCU fake writer threads");
93 torture_param(int, nreaders, -1, "Number of RCU reader threads");
94 torture_param(int, object_debug, 0,
95 	     "Enable debug-object double call_rcu() testing");
96 torture_param(int, onoff_holdoff, 0, "Time after boot before CPU hotplugs (s)");
97 torture_param(int, onoff_interval, 0,
98 	     "Time between CPU hotplugs (jiffies), 0=disable");
99 torture_param(int, shuffle_interval, 3, "Number of seconds between shuffles");
100 torture_param(int, shutdown_secs, 0, "Shutdown time (s), <= zero to disable.");
101 torture_param(int, stall_cpu, 0, "Stall duration (s), zero to disable.");
102 torture_param(int, stall_cpu_holdoff, 10,
103 	     "Time to wait before starting stall (s).");
104 torture_param(int, stall_cpu_irqsoff, 0, "Disable interrupts while stalling.");
105 torture_param(int, stat_interval, 60,
106 	     "Number of seconds between stats printk()s");
107 torture_param(int, stutter, 5, "Number of seconds to run/halt test");
108 torture_param(int, test_boost, 1, "Test RCU prio boost: 0=no, 1=maybe, 2=yes.");
109 torture_param(int, test_boost_duration, 4,
110 	     "Duration of each boost test, seconds.");
111 torture_param(int, test_boost_interval, 7,
112 	     "Interval between boost tests, seconds.");
113 torture_param(bool, test_no_idle_hz, true,
114 	     "Test support for tickless idle CPUs");
115 torture_param(int, verbose, 1,
116 	     "Enable verbose debugging printk()s");
117 
118 static char *torture_type = "rcu";
119 module_param(torture_type, charp, 0444);
120 MODULE_PARM_DESC(torture_type, "Type of RCU to torture (rcu, srcu, ...)");
121 
122 static int nrealreaders;
123 static struct task_struct *writer_task;
124 static struct task_struct **fakewriter_tasks;
125 static struct task_struct **reader_tasks;
126 static struct task_struct *stats_task;
127 static struct task_struct *fqs_task;
128 static struct task_struct *boost_tasks[NR_CPUS];
129 static struct task_struct *stall_task;
130 static struct task_struct *fwd_prog_task;
131 static struct task_struct **barrier_cbs_tasks;
132 static struct task_struct *barrier_task;
133 
134 #define RCU_TORTURE_PIPE_LEN 10
135 
136 struct rcu_torture {
137 	struct rcu_head rtort_rcu;
138 	int rtort_pipe_count;
139 	struct list_head rtort_free;
140 	int rtort_mbtest;
141 };
142 
143 static LIST_HEAD(rcu_torture_freelist);
144 static struct rcu_torture __rcu *rcu_torture_current;
145 static unsigned long rcu_torture_current_version;
146 static struct rcu_torture rcu_tortures[10 * RCU_TORTURE_PIPE_LEN];
147 static DEFINE_SPINLOCK(rcu_torture_lock);
148 static DEFINE_PER_CPU(long [RCU_TORTURE_PIPE_LEN + 1], rcu_torture_count);
149 static DEFINE_PER_CPU(long [RCU_TORTURE_PIPE_LEN + 1], rcu_torture_batch);
150 static atomic_t rcu_torture_wcount[RCU_TORTURE_PIPE_LEN + 1];
151 static atomic_t n_rcu_torture_alloc;
152 static atomic_t n_rcu_torture_alloc_fail;
153 static atomic_t n_rcu_torture_free;
154 static atomic_t n_rcu_torture_mberror;
155 static atomic_t n_rcu_torture_error;
156 static long n_rcu_torture_barrier_error;
157 static long n_rcu_torture_boost_ktrerror;
158 static long n_rcu_torture_boost_rterror;
159 static long n_rcu_torture_boost_failure;
160 static long n_rcu_torture_boosts;
161 static atomic_long_t n_rcu_torture_timers;
162 static long n_barrier_attempts;
163 static long n_barrier_successes; /* did rcu_barrier test succeed? */
164 static struct list_head rcu_torture_removed;
165 static unsigned long shutdown_jiffies;
166 
167 static int rcu_torture_writer_state;
168 #define RTWS_FIXED_DELAY	0
169 #define RTWS_DELAY		1
170 #define RTWS_REPLACE		2
171 #define RTWS_DEF_FREE		3
172 #define RTWS_EXP_SYNC		4
173 #define RTWS_COND_GET		5
174 #define RTWS_COND_SYNC		6
175 #define RTWS_SYNC		7
176 #define RTWS_STUTTER		8
177 #define RTWS_STOPPING		9
178 static const char * const rcu_torture_writer_state_names[] = {
179 	"RTWS_FIXED_DELAY",
180 	"RTWS_DELAY",
181 	"RTWS_REPLACE",
182 	"RTWS_DEF_FREE",
183 	"RTWS_EXP_SYNC",
184 	"RTWS_COND_GET",
185 	"RTWS_COND_SYNC",
186 	"RTWS_SYNC",
187 	"RTWS_STUTTER",
188 	"RTWS_STOPPING",
189 };
190 
191 /* Record reader segment types and duration for first failing read. */
192 struct rt_read_seg {
193 	int rt_readstate;
194 	unsigned long rt_delay_jiffies;
195 	unsigned long rt_delay_ms;
196 	unsigned long rt_delay_us;
197 	bool rt_preempted;
198 };
199 static int err_segs_recorded;
200 static struct rt_read_seg err_segs[RCUTORTURE_RDR_MAX_SEGS];
201 static int rt_read_nsegs;
202 
203 static const char *rcu_torture_writer_state_getname(void)
204 {
205 	unsigned int i = READ_ONCE(rcu_torture_writer_state);
206 
207 	if (i >= ARRAY_SIZE(rcu_torture_writer_state_names))
208 		return "???";
209 	return rcu_torture_writer_state_names[i];
210 }
211 
212 #if defined(CONFIG_RCU_BOOST) && !defined(CONFIG_HOTPLUG_CPU)
213 #define rcu_can_boost() 1
214 #else /* #if defined(CONFIG_RCU_BOOST) && !defined(CONFIG_HOTPLUG_CPU) */
215 #define rcu_can_boost() 0
216 #endif /* #else #if defined(CONFIG_RCU_BOOST) && !defined(CONFIG_HOTPLUG_CPU) */
217 
218 #ifdef CONFIG_RCU_TRACE
219 static u64 notrace rcu_trace_clock_local(void)
220 {
221 	u64 ts = trace_clock_local();
222 
223 	(void)do_div(ts, NSEC_PER_USEC);
224 	return ts;
225 }
226 #else /* #ifdef CONFIG_RCU_TRACE */
227 static u64 notrace rcu_trace_clock_local(void)
228 {
229 	return 0ULL;
230 }
231 #endif /* #else #ifdef CONFIG_RCU_TRACE */
232 
233 /*
234  * Stop aggressive CPU-hog tests a bit before the end of the test in order
235  * to avoid interfering with test shutdown.
236  */
237 static bool shutdown_time_arrived(void)
238 {
239 	return shutdown_secs && time_after(jiffies, shutdown_jiffies - 30 * HZ);
240 }
241 
242 static unsigned long boost_starttime;	/* jiffies of next boost test start. */
243 static DEFINE_MUTEX(boost_mutex);	/* protect setting boost_starttime */
244 					/*  and boost task create/destroy. */
245 static atomic_t barrier_cbs_count;	/* Barrier callbacks registered. */
246 static bool barrier_phase;		/* Test phase. */
247 static atomic_t barrier_cbs_invoked;	/* Barrier callbacks invoked. */
248 static wait_queue_head_t *barrier_cbs_wq; /* Coordinate barrier testing. */
249 static DECLARE_WAIT_QUEUE_HEAD(barrier_wq);
250 
251 static bool rcu_fwd_cb_nodelay;		/* Short rcu_torture_delay() delays. */
252 
253 /*
254  * Allocate an element from the rcu_tortures pool.
255  */
256 static struct rcu_torture *
257 rcu_torture_alloc(void)
258 {
259 	struct list_head *p;
260 
261 	spin_lock_bh(&rcu_torture_lock);
262 	if (list_empty(&rcu_torture_freelist)) {
263 		atomic_inc(&n_rcu_torture_alloc_fail);
264 		spin_unlock_bh(&rcu_torture_lock);
265 		return NULL;
266 	}
267 	atomic_inc(&n_rcu_torture_alloc);
268 	p = rcu_torture_freelist.next;
269 	list_del_init(p);
270 	spin_unlock_bh(&rcu_torture_lock);
271 	return container_of(p, struct rcu_torture, rtort_free);
272 }
273 
274 /*
275  * Free an element to the rcu_tortures pool.
276  */
277 static void
278 rcu_torture_free(struct rcu_torture *p)
279 {
280 	atomic_inc(&n_rcu_torture_free);
281 	spin_lock_bh(&rcu_torture_lock);
282 	list_add_tail(&p->rtort_free, &rcu_torture_freelist);
283 	spin_unlock_bh(&rcu_torture_lock);
284 }
285 
286 /*
287  * Operations vector for selecting different types of tests.
288  */
289 
290 struct rcu_torture_ops {
291 	int ttype;
292 	void (*init)(void);
293 	void (*cleanup)(void);
294 	int (*readlock)(void);
295 	void (*read_delay)(struct torture_random_state *rrsp,
296 			   struct rt_read_seg *rtrsp);
297 	void (*readunlock)(int idx);
298 	unsigned long (*get_gp_seq)(void);
299 	unsigned long (*gp_diff)(unsigned long new, unsigned long old);
300 	void (*deferred_free)(struct rcu_torture *p);
301 	void (*sync)(void);
302 	void (*exp_sync)(void);
303 	unsigned long (*get_state)(void);
304 	void (*cond_sync)(unsigned long oldstate);
305 	call_rcu_func_t call;
306 	void (*cb_barrier)(void);
307 	void (*fqs)(void);
308 	void (*stats)(void);
309 	int (*stall_dur)(void);
310 	int irq_capable;
311 	int can_boost;
312 	int extendables;
313 	int slow_gps;
314 	const char *name;
315 };
316 
317 static struct rcu_torture_ops *cur_ops;
318 
319 /*
320  * Definitions for rcu torture testing.
321  */
322 
323 static int rcu_torture_read_lock(void) __acquires(RCU)
324 {
325 	rcu_read_lock();
326 	return 0;
327 }
328 
329 static void
330 rcu_read_delay(struct torture_random_state *rrsp, struct rt_read_seg *rtrsp)
331 {
332 	unsigned long started;
333 	unsigned long completed;
334 	const unsigned long shortdelay_us = 200;
335 	unsigned long longdelay_ms = 300;
336 	unsigned long long ts;
337 
338 	/* We want a short delay sometimes to make a reader delay the grace
339 	 * period, and we want a long delay occasionally to trigger
340 	 * force_quiescent_state. */
341 
342 	if (!rcu_fwd_cb_nodelay &&
343 	    !(torture_random(rrsp) % (nrealreaders * 2000 * longdelay_ms))) {
344 		started = cur_ops->get_gp_seq();
345 		ts = rcu_trace_clock_local();
346 		if (preempt_count() & (SOFTIRQ_MASK | HARDIRQ_MASK))
347 			longdelay_ms = 5; /* Avoid triggering BH limits. */
348 		mdelay(longdelay_ms);
349 		rtrsp->rt_delay_ms = longdelay_ms;
350 		completed = cur_ops->get_gp_seq();
351 		do_trace_rcu_torture_read(cur_ops->name, NULL, ts,
352 					  started, completed);
353 	}
354 	if (!(torture_random(rrsp) % (nrealreaders * 2 * shortdelay_us))) {
355 		udelay(shortdelay_us);
356 		rtrsp->rt_delay_us = shortdelay_us;
357 	}
358 	if (!preempt_count() &&
359 	    !(torture_random(rrsp) % (nrealreaders * 500))) {
360 		torture_preempt_schedule();  /* QS only if preemptible. */
361 		rtrsp->rt_preempted = true;
362 	}
363 }
364 
365 static void rcu_torture_read_unlock(int idx) __releases(RCU)
366 {
367 	rcu_read_unlock();
368 }
369 
370 /*
371  * Update callback in the pipe.  This should be invoked after a grace period.
372  */
373 static bool
374 rcu_torture_pipe_update_one(struct rcu_torture *rp)
375 {
376 	int i;
377 
378 	i = rp->rtort_pipe_count;
379 	if (i > RCU_TORTURE_PIPE_LEN)
380 		i = RCU_TORTURE_PIPE_LEN;
381 	atomic_inc(&rcu_torture_wcount[i]);
382 	if (++rp->rtort_pipe_count >= RCU_TORTURE_PIPE_LEN) {
383 		rp->rtort_mbtest = 0;
384 		return true;
385 	}
386 	return false;
387 }
388 
389 /*
390  * Update all callbacks in the pipe.  Suitable for synchronous grace-period
391  * primitives.
392  */
393 static void
394 rcu_torture_pipe_update(struct rcu_torture *old_rp)
395 {
396 	struct rcu_torture *rp;
397 	struct rcu_torture *rp1;
398 
399 	if (old_rp)
400 		list_add(&old_rp->rtort_free, &rcu_torture_removed);
401 	list_for_each_entry_safe(rp, rp1, &rcu_torture_removed, rtort_free) {
402 		if (rcu_torture_pipe_update_one(rp)) {
403 			list_del(&rp->rtort_free);
404 			rcu_torture_free(rp);
405 		}
406 	}
407 }
408 
409 static void
410 rcu_torture_cb(struct rcu_head *p)
411 {
412 	struct rcu_torture *rp = container_of(p, struct rcu_torture, rtort_rcu);
413 
414 	if (torture_must_stop_irq()) {
415 		/* Test is ending, just drop callbacks on the floor. */
416 		/* The next initialization will pick up the pieces. */
417 		return;
418 	}
419 	if (rcu_torture_pipe_update_one(rp))
420 		rcu_torture_free(rp);
421 	else
422 		cur_ops->deferred_free(rp);
423 }
424 
425 static unsigned long rcu_no_completed(void)
426 {
427 	return 0;
428 }
429 
430 static void rcu_torture_deferred_free(struct rcu_torture *p)
431 {
432 	call_rcu(&p->rtort_rcu, rcu_torture_cb);
433 }
434 
435 static void rcu_sync_torture_init(void)
436 {
437 	INIT_LIST_HEAD(&rcu_torture_removed);
438 }
439 
440 static struct rcu_torture_ops rcu_ops = {
441 	.ttype		= RCU_FLAVOR,
442 	.init		= rcu_sync_torture_init,
443 	.readlock	= rcu_torture_read_lock,
444 	.read_delay	= rcu_read_delay,
445 	.readunlock	= rcu_torture_read_unlock,
446 	.get_gp_seq	= rcu_get_gp_seq,
447 	.gp_diff	= rcu_seq_diff,
448 	.deferred_free	= rcu_torture_deferred_free,
449 	.sync		= synchronize_rcu,
450 	.exp_sync	= synchronize_rcu_expedited,
451 	.get_state	= get_state_synchronize_rcu,
452 	.cond_sync	= cond_synchronize_rcu,
453 	.call		= call_rcu,
454 	.cb_barrier	= rcu_barrier,
455 	.fqs		= rcu_force_quiescent_state,
456 	.stats		= NULL,
457 	.stall_dur	= rcu_jiffies_till_stall_check,
458 	.irq_capable	= 1,
459 	.can_boost	= rcu_can_boost(),
460 	.extendables	= RCUTORTURE_MAX_EXTEND,
461 	.name		= "rcu"
462 };
463 
464 /*
465  * Don't even think about trying any of these in real life!!!
466  * The names includes "busted", and they really means it!
467  * The only purpose of these functions is to provide a buggy RCU
468  * implementation to make sure that rcutorture correctly emits
469  * buggy-RCU error messages.
470  */
471 static void rcu_busted_torture_deferred_free(struct rcu_torture *p)
472 {
473 	/* This is a deliberate bug for testing purposes only! */
474 	rcu_torture_cb(&p->rtort_rcu);
475 }
476 
477 static void synchronize_rcu_busted(void)
478 {
479 	/* This is a deliberate bug for testing purposes only! */
480 }
481 
482 static void
483 call_rcu_busted(struct rcu_head *head, rcu_callback_t func)
484 {
485 	/* This is a deliberate bug for testing purposes only! */
486 	func(head);
487 }
488 
489 static struct rcu_torture_ops rcu_busted_ops = {
490 	.ttype		= INVALID_RCU_FLAVOR,
491 	.init		= rcu_sync_torture_init,
492 	.readlock	= rcu_torture_read_lock,
493 	.read_delay	= rcu_read_delay,  /* just reuse rcu's version. */
494 	.readunlock	= rcu_torture_read_unlock,
495 	.get_gp_seq	= rcu_no_completed,
496 	.deferred_free	= rcu_busted_torture_deferred_free,
497 	.sync		= synchronize_rcu_busted,
498 	.exp_sync	= synchronize_rcu_busted,
499 	.call		= call_rcu_busted,
500 	.cb_barrier	= NULL,
501 	.fqs		= NULL,
502 	.stats		= NULL,
503 	.irq_capable	= 1,
504 	.name		= "busted"
505 };
506 
507 /*
508  * Definitions for srcu torture testing.
509  */
510 
511 DEFINE_STATIC_SRCU(srcu_ctl);
512 static struct srcu_struct srcu_ctld;
513 static struct srcu_struct *srcu_ctlp = &srcu_ctl;
514 
515 static int srcu_torture_read_lock(void) __acquires(srcu_ctlp)
516 {
517 	return srcu_read_lock(srcu_ctlp);
518 }
519 
520 static void
521 srcu_read_delay(struct torture_random_state *rrsp, struct rt_read_seg *rtrsp)
522 {
523 	long delay;
524 	const long uspertick = 1000000 / HZ;
525 	const long longdelay = 10;
526 
527 	/* We want there to be long-running readers, but not all the time. */
528 
529 	delay = torture_random(rrsp) %
530 		(nrealreaders * 2 * longdelay * uspertick);
531 	if (!delay && in_task()) {
532 		schedule_timeout_interruptible(longdelay);
533 		rtrsp->rt_delay_jiffies = longdelay;
534 	} else {
535 		rcu_read_delay(rrsp, rtrsp);
536 	}
537 }
538 
539 static void srcu_torture_read_unlock(int idx) __releases(srcu_ctlp)
540 {
541 	srcu_read_unlock(srcu_ctlp, idx);
542 }
543 
544 static unsigned long srcu_torture_completed(void)
545 {
546 	return srcu_batches_completed(srcu_ctlp);
547 }
548 
549 static void srcu_torture_deferred_free(struct rcu_torture *rp)
550 {
551 	call_srcu(srcu_ctlp, &rp->rtort_rcu, rcu_torture_cb);
552 }
553 
554 static void srcu_torture_synchronize(void)
555 {
556 	synchronize_srcu(srcu_ctlp);
557 }
558 
559 static void srcu_torture_call(struct rcu_head *head,
560 			      rcu_callback_t func)
561 {
562 	call_srcu(srcu_ctlp, head, func);
563 }
564 
565 static void srcu_torture_barrier(void)
566 {
567 	srcu_barrier(srcu_ctlp);
568 }
569 
570 static void srcu_torture_stats(void)
571 {
572 	srcu_torture_stats_print(srcu_ctlp, torture_type, TORTURE_FLAG);
573 }
574 
575 static void srcu_torture_synchronize_expedited(void)
576 {
577 	synchronize_srcu_expedited(srcu_ctlp);
578 }
579 
580 static struct rcu_torture_ops srcu_ops = {
581 	.ttype		= SRCU_FLAVOR,
582 	.init		= rcu_sync_torture_init,
583 	.readlock	= srcu_torture_read_lock,
584 	.read_delay	= srcu_read_delay,
585 	.readunlock	= srcu_torture_read_unlock,
586 	.get_gp_seq	= srcu_torture_completed,
587 	.deferred_free	= srcu_torture_deferred_free,
588 	.sync		= srcu_torture_synchronize,
589 	.exp_sync	= srcu_torture_synchronize_expedited,
590 	.call		= srcu_torture_call,
591 	.cb_barrier	= srcu_torture_barrier,
592 	.stats		= srcu_torture_stats,
593 	.irq_capable	= 1,
594 	.name		= "srcu"
595 };
596 
597 static void srcu_torture_init(void)
598 {
599 	rcu_sync_torture_init();
600 	WARN_ON(init_srcu_struct(&srcu_ctld));
601 	srcu_ctlp = &srcu_ctld;
602 }
603 
604 static void srcu_torture_cleanup(void)
605 {
606 	cleanup_srcu_struct(&srcu_ctld);
607 	srcu_ctlp = &srcu_ctl; /* In case of a later rcutorture run. */
608 }
609 
610 /* As above, but dynamically allocated. */
611 static struct rcu_torture_ops srcud_ops = {
612 	.ttype		= SRCU_FLAVOR,
613 	.init		= srcu_torture_init,
614 	.cleanup	= srcu_torture_cleanup,
615 	.readlock	= srcu_torture_read_lock,
616 	.read_delay	= srcu_read_delay,
617 	.readunlock	= srcu_torture_read_unlock,
618 	.get_gp_seq	= srcu_torture_completed,
619 	.deferred_free	= srcu_torture_deferred_free,
620 	.sync		= srcu_torture_synchronize,
621 	.exp_sync	= srcu_torture_synchronize_expedited,
622 	.call		= srcu_torture_call,
623 	.cb_barrier	= srcu_torture_barrier,
624 	.stats		= srcu_torture_stats,
625 	.irq_capable	= 1,
626 	.name		= "srcud"
627 };
628 
629 /* As above, but broken due to inappropriate reader extension. */
630 static struct rcu_torture_ops busted_srcud_ops = {
631 	.ttype		= SRCU_FLAVOR,
632 	.init		= srcu_torture_init,
633 	.cleanup	= srcu_torture_cleanup,
634 	.readlock	= srcu_torture_read_lock,
635 	.read_delay	= rcu_read_delay,
636 	.readunlock	= srcu_torture_read_unlock,
637 	.get_gp_seq	= srcu_torture_completed,
638 	.deferred_free	= srcu_torture_deferred_free,
639 	.sync		= srcu_torture_synchronize,
640 	.exp_sync	= srcu_torture_synchronize_expedited,
641 	.call		= srcu_torture_call,
642 	.cb_barrier	= srcu_torture_barrier,
643 	.stats		= srcu_torture_stats,
644 	.irq_capable	= 1,
645 	.extendables	= RCUTORTURE_MAX_EXTEND,
646 	.name		= "busted_srcud"
647 };
648 
649 /*
650  * Definitions for RCU-tasks torture testing.
651  */
652 
653 static int tasks_torture_read_lock(void)
654 {
655 	return 0;
656 }
657 
658 static void tasks_torture_read_unlock(int idx)
659 {
660 }
661 
662 static void rcu_tasks_torture_deferred_free(struct rcu_torture *p)
663 {
664 	call_rcu_tasks(&p->rtort_rcu, rcu_torture_cb);
665 }
666 
667 static struct rcu_torture_ops tasks_ops = {
668 	.ttype		= RCU_TASKS_FLAVOR,
669 	.init		= rcu_sync_torture_init,
670 	.readlock	= tasks_torture_read_lock,
671 	.read_delay	= rcu_read_delay,  /* just reuse rcu's version. */
672 	.readunlock	= tasks_torture_read_unlock,
673 	.get_gp_seq	= rcu_no_completed,
674 	.deferred_free	= rcu_tasks_torture_deferred_free,
675 	.sync		= synchronize_rcu_tasks,
676 	.exp_sync	= synchronize_rcu_tasks,
677 	.call		= call_rcu_tasks,
678 	.cb_barrier	= rcu_barrier_tasks,
679 	.fqs		= NULL,
680 	.stats		= NULL,
681 	.irq_capable	= 1,
682 	.slow_gps	= 1,
683 	.name		= "tasks"
684 };
685 
686 /*
687  * Definitions for trivial CONFIG_PREEMPT=n-only torture testing.
688  * This implementation does not necessarily work well with CPU hotplug.
689  */
690 
691 static void synchronize_rcu_trivial(void)
692 {
693 	int cpu;
694 
695 	for_each_online_cpu(cpu) {
696 		rcutorture_sched_setaffinity(current->pid, cpumask_of(cpu));
697 		WARN_ON_ONCE(raw_smp_processor_id() != cpu);
698 	}
699 }
700 
701 static int rcu_torture_read_lock_trivial(void) __acquires(RCU)
702 {
703 	preempt_disable();
704 	return 0;
705 }
706 
707 static void rcu_torture_read_unlock_trivial(int idx) __releases(RCU)
708 {
709 	preempt_enable();
710 }
711 
712 static struct rcu_torture_ops trivial_ops = {
713 	.ttype		= RCU_TRIVIAL_FLAVOR,
714 	.init		= rcu_sync_torture_init,
715 	.readlock	= rcu_torture_read_lock_trivial,
716 	.read_delay	= rcu_read_delay,  /* just reuse rcu's version. */
717 	.readunlock	= rcu_torture_read_unlock_trivial,
718 	.get_gp_seq	= rcu_no_completed,
719 	.sync		= synchronize_rcu_trivial,
720 	.exp_sync	= synchronize_rcu_trivial,
721 	.fqs		= NULL,
722 	.stats		= NULL,
723 	.irq_capable	= 1,
724 	.name		= "trivial"
725 };
726 
727 static unsigned long rcutorture_seq_diff(unsigned long new, unsigned long old)
728 {
729 	if (!cur_ops->gp_diff)
730 		return new - old;
731 	return cur_ops->gp_diff(new, old);
732 }
733 
734 static bool __maybe_unused torturing_tasks(void)
735 {
736 	return cur_ops == &tasks_ops;
737 }
738 
739 /*
740  * RCU torture priority-boost testing.  Runs one real-time thread per
741  * CPU for moderate bursts, repeatedly registering RCU callbacks and
742  * spinning waiting for them to be invoked.  If a given callback takes
743  * too long to be invoked, we assume that priority inversion has occurred.
744  */
745 
746 struct rcu_boost_inflight {
747 	struct rcu_head rcu;
748 	int inflight;
749 };
750 
751 static void rcu_torture_boost_cb(struct rcu_head *head)
752 {
753 	struct rcu_boost_inflight *rbip =
754 		container_of(head, struct rcu_boost_inflight, rcu);
755 
756 	/* Ensure RCU-core accesses precede clearing ->inflight */
757 	smp_store_release(&rbip->inflight, 0);
758 }
759 
760 static int old_rt_runtime = -1;
761 
762 static void rcu_torture_disable_rt_throttle(void)
763 {
764 	/*
765 	 * Disable RT throttling so that rcutorture's boost threads don't get
766 	 * throttled. Only possible if rcutorture is built-in otherwise the
767 	 * user should manually do this by setting the sched_rt_period_us and
768 	 * sched_rt_runtime sysctls.
769 	 */
770 	if (!IS_BUILTIN(CONFIG_RCU_TORTURE_TEST) || old_rt_runtime != -1)
771 		return;
772 
773 	old_rt_runtime = sysctl_sched_rt_runtime;
774 	sysctl_sched_rt_runtime = -1;
775 }
776 
777 static void rcu_torture_enable_rt_throttle(void)
778 {
779 	if (!IS_BUILTIN(CONFIG_RCU_TORTURE_TEST) || old_rt_runtime == -1)
780 		return;
781 
782 	sysctl_sched_rt_runtime = old_rt_runtime;
783 	old_rt_runtime = -1;
784 }
785 
786 static bool rcu_torture_boost_failed(unsigned long start, unsigned long end)
787 {
788 	if (end - start > test_boost_duration * HZ - HZ / 2) {
789 		VERBOSE_TOROUT_STRING("rcu_torture_boost boosting failed");
790 		n_rcu_torture_boost_failure++;
791 
792 		return true; /* failed */
793 	}
794 
795 	return false; /* passed */
796 }
797 
798 static int rcu_torture_boost(void *arg)
799 {
800 	unsigned long call_rcu_time;
801 	unsigned long endtime;
802 	unsigned long oldstarttime;
803 	struct rcu_boost_inflight rbi = { .inflight = 0 };
804 	struct sched_param sp;
805 
806 	VERBOSE_TOROUT_STRING("rcu_torture_boost started");
807 
808 	/* Set real-time priority. */
809 	sp.sched_priority = 1;
810 	if (sched_setscheduler(current, SCHED_FIFO, &sp) < 0) {
811 		VERBOSE_TOROUT_STRING("rcu_torture_boost RT prio failed!");
812 		n_rcu_torture_boost_rterror++;
813 	}
814 
815 	init_rcu_head_on_stack(&rbi.rcu);
816 	/* Each pass through the following loop does one boost-test cycle. */
817 	do {
818 		/* Track if the test failed already in this test interval? */
819 		bool failed = false;
820 
821 		/* Increment n_rcu_torture_boosts once per boost-test */
822 		while (!kthread_should_stop()) {
823 			if (mutex_trylock(&boost_mutex)) {
824 				n_rcu_torture_boosts++;
825 				mutex_unlock(&boost_mutex);
826 				break;
827 			}
828 			schedule_timeout_uninterruptible(1);
829 		}
830 		if (kthread_should_stop())
831 			goto checkwait;
832 
833 		/* Wait for the next test interval. */
834 		oldstarttime = boost_starttime;
835 		while (ULONG_CMP_LT(jiffies, oldstarttime)) {
836 			schedule_timeout_interruptible(oldstarttime - jiffies);
837 			stutter_wait("rcu_torture_boost");
838 			if (torture_must_stop())
839 				goto checkwait;
840 		}
841 
842 		/* Do one boost-test interval. */
843 		endtime = oldstarttime + test_boost_duration * HZ;
844 		call_rcu_time = jiffies;
845 		while (ULONG_CMP_LT(jiffies, endtime)) {
846 			/* If we don't have a callback in flight, post one. */
847 			if (!smp_load_acquire(&rbi.inflight)) {
848 				/* RCU core before ->inflight = 1. */
849 				smp_store_release(&rbi.inflight, 1);
850 				call_rcu(&rbi.rcu, rcu_torture_boost_cb);
851 				/* Check if the boost test failed */
852 				failed = failed ||
853 					 rcu_torture_boost_failed(call_rcu_time,
854 								 jiffies);
855 				call_rcu_time = jiffies;
856 			}
857 			stutter_wait("rcu_torture_boost");
858 			if (torture_must_stop())
859 				goto checkwait;
860 		}
861 
862 		/*
863 		 * If boost never happened, then inflight will always be 1, in
864 		 * this case the boost check would never happen in the above
865 		 * loop so do another one here.
866 		 */
867 		if (!failed && smp_load_acquire(&rbi.inflight))
868 			rcu_torture_boost_failed(call_rcu_time, jiffies);
869 
870 		/*
871 		 * Set the start time of the next test interval.
872 		 * Yes, this is vulnerable to long delays, but such
873 		 * delays simply cause a false negative for the next
874 		 * interval.  Besides, we are running at RT priority,
875 		 * so delays should be relatively rare.
876 		 */
877 		while (oldstarttime == boost_starttime &&
878 		       !kthread_should_stop()) {
879 			if (mutex_trylock(&boost_mutex)) {
880 				boost_starttime = jiffies +
881 						  test_boost_interval * HZ;
882 				mutex_unlock(&boost_mutex);
883 				break;
884 			}
885 			schedule_timeout_uninterruptible(1);
886 		}
887 
888 		/* Go do the stutter. */
889 checkwait:	stutter_wait("rcu_torture_boost");
890 	} while (!torture_must_stop());
891 
892 	/* Clean up and exit. */
893 	while (!kthread_should_stop() || smp_load_acquire(&rbi.inflight)) {
894 		torture_shutdown_absorb("rcu_torture_boost");
895 		schedule_timeout_uninterruptible(1);
896 	}
897 	destroy_rcu_head_on_stack(&rbi.rcu);
898 	torture_kthread_stopping("rcu_torture_boost");
899 	return 0;
900 }
901 
902 /*
903  * RCU torture force-quiescent-state kthread.  Repeatedly induces
904  * bursts of calls to force_quiescent_state(), increasing the probability
905  * of occurrence of some important types of race conditions.
906  */
907 static int
908 rcu_torture_fqs(void *arg)
909 {
910 	unsigned long fqs_resume_time;
911 	int fqs_burst_remaining;
912 
913 	VERBOSE_TOROUT_STRING("rcu_torture_fqs task started");
914 	do {
915 		fqs_resume_time = jiffies + fqs_stutter * HZ;
916 		while (ULONG_CMP_LT(jiffies, fqs_resume_time) &&
917 		       !kthread_should_stop()) {
918 			schedule_timeout_interruptible(1);
919 		}
920 		fqs_burst_remaining = fqs_duration;
921 		while (fqs_burst_remaining > 0 &&
922 		       !kthread_should_stop()) {
923 			cur_ops->fqs();
924 			udelay(fqs_holdoff);
925 			fqs_burst_remaining -= fqs_holdoff;
926 		}
927 		stutter_wait("rcu_torture_fqs");
928 	} while (!torture_must_stop());
929 	torture_kthread_stopping("rcu_torture_fqs");
930 	return 0;
931 }
932 
933 /*
934  * RCU torture writer kthread.  Repeatedly substitutes a new structure
935  * for that pointed to by rcu_torture_current, freeing the old structure
936  * after a series of grace periods (the "pipeline").
937  */
938 static int
939 rcu_torture_writer(void *arg)
940 {
941 	bool can_expedite = !rcu_gp_is_expedited() && !rcu_gp_is_normal();
942 	int expediting = 0;
943 	unsigned long gp_snap;
944 	bool gp_cond1 = gp_cond, gp_exp1 = gp_exp, gp_normal1 = gp_normal;
945 	bool gp_sync1 = gp_sync;
946 	int i;
947 	struct rcu_torture *rp;
948 	struct rcu_torture *old_rp;
949 	static DEFINE_TORTURE_RANDOM(rand);
950 	int synctype[] = { RTWS_DEF_FREE, RTWS_EXP_SYNC,
951 			   RTWS_COND_GET, RTWS_SYNC };
952 	int nsynctypes = 0;
953 
954 	VERBOSE_TOROUT_STRING("rcu_torture_writer task started");
955 	if (!can_expedite)
956 		pr_alert("%s" TORTURE_FLAG
957 			 " GP expediting controlled from boot/sysfs for %s.\n",
958 			 torture_type, cur_ops->name);
959 
960 	/* Initialize synctype[] array.  If none set, take default. */
961 	if (!gp_cond1 && !gp_exp1 && !gp_normal1 && !gp_sync1)
962 		gp_cond1 = gp_exp1 = gp_normal1 = gp_sync1 = true;
963 	if (gp_cond1 && cur_ops->get_state && cur_ops->cond_sync) {
964 		synctype[nsynctypes++] = RTWS_COND_GET;
965 		pr_info("%s: Testing conditional GPs.\n", __func__);
966 	} else if (gp_cond && (!cur_ops->get_state || !cur_ops->cond_sync)) {
967 		pr_alert("%s: gp_cond without primitives.\n", __func__);
968 	}
969 	if (gp_exp1 && cur_ops->exp_sync) {
970 		synctype[nsynctypes++] = RTWS_EXP_SYNC;
971 		pr_info("%s: Testing expedited GPs.\n", __func__);
972 	} else if (gp_exp && !cur_ops->exp_sync) {
973 		pr_alert("%s: gp_exp without primitives.\n", __func__);
974 	}
975 	if (gp_normal1 && cur_ops->deferred_free) {
976 		synctype[nsynctypes++] = RTWS_DEF_FREE;
977 		pr_info("%s: Testing asynchronous GPs.\n", __func__);
978 	} else if (gp_normal && !cur_ops->deferred_free) {
979 		pr_alert("%s: gp_normal without primitives.\n", __func__);
980 	}
981 	if (gp_sync1 && cur_ops->sync) {
982 		synctype[nsynctypes++] = RTWS_SYNC;
983 		pr_info("%s: Testing normal GPs.\n", __func__);
984 	} else if (gp_sync && !cur_ops->sync) {
985 		pr_alert("%s: gp_sync without primitives.\n", __func__);
986 	}
987 	if (WARN_ONCE(nsynctypes == 0,
988 		      "rcu_torture_writer: No update-side primitives.\n")) {
989 		/*
990 		 * No updates primitives, so don't try updating.
991 		 * The resulting test won't be testing much, hence the
992 		 * above WARN_ONCE().
993 		 */
994 		rcu_torture_writer_state = RTWS_STOPPING;
995 		torture_kthread_stopping("rcu_torture_writer");
996 	}
997 
998 	do {
999 		rcu_torture_writer_state = RTWS_FIXED_DELAY;
1000 		schedule_timeout_uninterruptible(1);
1001 		rp = rcu_torture_alloc();
1002 		if (rp == NULL)
1003 			continue;
1004 		rp->rtort_pipe_count = 0;
1005 		rcu_torture_writer_state = RTWS_DELAY;
1006 		udelay(torture_random(&rand) & 0x3ff);
1007 		rcu_torture_writer_state = RTWS_REPLACE;
1008 		old_rp = rcu_dereference_check(rcu_torture_current,
1009 					       current == writer_task);
1010 		rp->rtort_mbtest = 1;
1011 		rcu_assign_pointer(rcu_torture_current, rp);
1012 		smp_wmb(); /* Mods to old_rp must follow rcu_assign_pointer() */
1013 		if (old_rp) {
1014 			i = old_rp->rtort_pipe_count;
1015 			if (i > RCU_TORTURE_PIPE_LEN)
1016 				i = RCU_TORTURE_PIPE_LEN;
1017 			atomic_inc(&rcu_torture_wcount[i]);
1018 			old_rp->rtort_pipe_count++;
1019 			switch (synctype[torture_random(&rand) % nsynctypes]) {
1020 			case RTWS_DEF_FREE:
1021 				rcu_torture_writer_state = RTWS_DEF_FREE;
1022 				cur_ops->deferred_free(old_rp);
1023 				break;
1024 			case RTWS_EXP_SYNC:
1025 				rcu_torture_writer_state = RTWS_EXP_SYNC;
1026 				cur_ops->exp_sync();
1027 				rcu_torture_pipe_update(old_rp);
1028 				break;
1029 			case RTWS_COND_GET:
1030 				rcu_torture_writer_state = RTWS_COND_GET;
1031 				gp_snap = cur_ops->get_state();
1032 				i = torture_random(&rand) % 16;
1033 				if (i != 0)
1034 					schedule_timeout_interruptible(i);
1035 				udelay(torture_random(&rand) % 1000);
1036 				rcu_torture_writer_state = RTWS_COND_SYNC;
1037 				cur_ops->cond_sync(gp_snap);
1038 				rcu_torture_pipe_update(old_rp);
1039 				break;
1040 			case RTWS_SYNC:
1041 				rcu_torture_writer_state = RTWS_SYNC;
1042 				cur_ops->sync();
1043 				rcu_torture_pipe_update(old_rp);
1044 				break;
1045 			default:
1046 				WARN_ON_ONCE(1);
1047 				break;
1048 			}
1049 		}
1050 		WRITE_ONCE(rcu_torture_current_version,
1051 			   rcu_torture_current_version + 1);
1052 		/* Cycle through nesting levels of rcu_expedite_gp() calls. */
1053 		if (can_expedite &&
1054 		    !(torture_random(&rand) & 0xff & (!!expediting - 1))) {
1055 			WARN_ON_ONCE(expediting == 0 && rcu_gp_is_expedited());
1056 			if (expediting >= 0)
1057 				rcu_expedite_gp();
1058 			else
1059 				rcu_unexpedite_gp();
1060 			if (++expediting > 3)
1061 				expediting = -expediting;
1062 		} else if (!can_expedite) { /* Disabled during boot, recheck. */
1063 			can_expedite = !rcu_gp_is_expedited() &&
1064 				       !rcu_gp_is_normal();
1065 		}
1066 		rcu_torture_writer_state = RTWS_STUTTER;
1067 		if (stutter_wait("rcu_torture_writer") &&
1068 		    !READ_ONCE(rcu_fwd_cb_nodelay) &&
1069 		    !cur_ops->slow_gps &&
1070 		    !torture_must_stop())
1071 			for (i = 0; i < ARRAY_SIZE(rcu_tortures); i++)
1072 				if (list_empty(&rcu_tortures[i].rtort_free) &&
1073 				    rcu_access_pointer(rcu_torture_current) !=
1074 				    &rcu_tortures[i]) {
1075 					rcu_ftrace_dump(DUMP_ALL);
1076 					WARN(1, "%s: rtort_pipe_count: %d\n", __func__, rcu_tortures[i].rtort_pipe_count);
1077 				}
1078 	} while (!torture_must_stop());
1079 	/* Reset expediting back to unexpedited. */
1080 	if (expediting > 0)
1081 		expediting = -expediting;
1082 	while (can_expedite && expediting++ < 0)
1083 		rcu_unexpedite_gp();
1084 	WARN_ON_ONCE(can_expedite && rcu_gp_is_expedited());
1085 	if (!can_expedite)
1086 		pr_alert("%s" TORTURE_FLAG
1087 			 " Dynamic grace-period expediting was disabled.\n",
1088 			 torture_type);
1089 	rcu_torture_writer_state = RTWS_STOPPING;
1090 	torture_kthread_stopping("rcu_torture_writer");
1091 	return 0;
1092 }
1093 
1094 /*
1095  * RCU torture fake writer kthread.  Repeatedly calls sync, with a random
1096  * delay between calls.
1097  */
1098 static int
1099 rcu_torture_fakewriter(void *arg)
1100 {
1101 	DEFINE_TORTURE_RANDOM(rand);
1102 
1103 	VERBOSE_TOROUT_STRING("rcu_torture_fakewriter task started");
1104 	set_user_nice(current, MAX_NICE);
1105 
1106 	do {
1107 		schedule_timeout_uninterruptible(1 + torture_random(&rand)%10);
1108 		udelay(torture_random(&rand) & 0x3ff);
1109 		if (cur_ops->cb_barrier != NULL &&
1110 		    torture_random(&rand) % (nfakewriters * 8) == 0) {
1111 			cur_ops->cb_barrier();
1112 		} else if (gp_normal == gp_exp) {
1113 			if (cur_ops->sync && torture_random(&rand) & 0x80)
1114 				cur_ops->sync();
1115 			else if (cur_ops->exp_sync)
1116 				cur_ops->exp_sync();
1117 		} else if (gp_normal && cur_ops->sync) {
1118 			cur_ops->sync();
1119 		} else if (cur_ops->exp_sync) {
1120 			cur_ops->exp_sync();
1121 		}
1122 		stutter_wait("rcu_torture_fakewriter");
1123 	} while (!torture_must_stop());
1124 
1125 	torture_kthread_stopping("rcu_torture_fakewriter");
1126 	return 0;
1127 }
1128 
1129 static void rcu_torture_timer_cb(struct rcu_head *rhp)
1130 {
1131 	kfree(rhp);
1132 }
1133 
1134 /*
1135  * Do one extension of an RCU read-side critical section using the
1136  * current reader state in readstate (set to zero for initial entry
1137  * to extended critical section), set the new state as specified by
1138  * newstate (set to zero for final exit from extended critical section),
1139  * and random-number-generator state in trsp.  If this is neither the
1140  * beginning or end of the critical section and if there was actually a
1141  * change, do a ->read_delay().
1142  */
1143 static void rcutorture_one_extend(int *readstate, int newstate,
1144 				  struct torture_random_state *trsp,
1145 				  struct rt_read_seg *rtrsp)
1146 {
1147 	int idxnew = -1;
1148 	int idxold = *readstate;
1149 	int statesnew = ~*readstate & newstate;
1150 	int statesold = *readstate & ~newstate;
1151 
1152 	WARN_ON_ONCE(idxold < 0);
1153 	WARN_ON_ONCE((idxold >> RCUTORTURE_RDR_SHIFT) > 1);
1154 	rtrsp->rt_readstate = newstate;
1155 
1156 	/* First, put new protection in place to avoid critical-section gap. */
1157 	if (statesnew & RCUTORTURE_RDR_BH)
1158 		local_bh_disable();
1159 	if (statesnew & RCUTORTURE_RDR_IRQ)
1160 		local_irq_disable();
1161 	if (statesnew & RCUTORTURE_RDR_PREEMPT)
1162 		preempt_disable();
1163 	if (statesnew & RCUTORTURE_RDR_RBH)
1164 		rcu_read_lock_bh();
1165 	if (statesnew & RCUTORTURE_RDR_SCHED)
1166 		rcu_read_lock_sched();
1167 	if (statesnew & RCUTORTURE_RDR_RCU)
1168 		idxnew = cur_ops->readlock() << RCUTORTURE_RDR_SHIFT;
1169 
1170 	/* Next, remove old protection, irq first due to bh conflict. */
1171 	if (statesold & RCUTORTURE_RDR_IRQ)
1172 		local_irq_enable();
1173 	if (statesold & RCUTORTURE_RDR_BH)
1174 		local_bh_enable();
1175 	if (statesold & RCUTORTURE_RDR_PREEMPT)
1176 		preempt_enable();
1177 	if (statesold & RCUTORTURE_RDR_RBH)
1178 		rcu_read_unlock_bh();
1179 	if (statesold & RCUTORTURE_RDR_SCHED)
1180 		rcu_read_unlock_sched();
1181 	if (statesold & RCUTORTURE_RDR_RCU)
1182 		cur_ops->readunlock(idxold >> RCUTORTURE_RDR_SHIFT);
1183 
1184 	/* Delay if neither beginning nor end and there was a change. */
1185 	if ((statesnew || statesold) && *readstate && newstate)
1186 		cur_ops->read_delay(trsp, rtrsp);
1187 
1188 	/* Update the reader state. */
1189 	if (idxnew == -1)
1190 		idxnew = idxold & ~RCUTORTURE_RDR_MASK;
1191 	WARN_ON_ONCE(idxnew < 0);
1192 	WARN_ON_ONCE((idxnew >> RCUTORTURE_RDR_SHIFT) > 1);
1193 	*readstate = idxnew | newstate;
1194 	WARN_ON_ONCE((*readstate >> RCUTORTURE_RDR_SHIFT) < 0);
1195 	WARN_ON_ONCE((*readstate >> RCUTORTURE_RDR_SHIFT) > 1);
1196 }
1197 
1198 /* Return the biggest extendables mask given current RCU and boot parameters. */
1199 static int rcutorture_extend_mask_max(void)
1200 {
1201 	int mask;
1202 
1203 	WARN_ON_ONCE(extendables & ~RCUTORTURE_MAX_EXTEND);
1204 	mask = extendables & RCUTORTURE_MAX_EXTEND & cur_ops->extendables;
1205 	mask = mask | RCUTORTURE_RDR_RCU;
1206 	return mask;
1207 }
1208 
1209 /* Return a random protection state mask, but with at least one bit set. */
1210 static int
1211 rcutorture_extend_mask(int oldmask, struct torture_random_state *trsp)
1212 {
1213 	int mask = rcutorture_extend_mask_max();
1214 	unsigned long randmask1 = torture_random(trsp) >> 8;
1215 	unsigned long randmask2 = randmask1 >> 3;
1216 
1217 	WARN_ON_ONCE(mask >> RCUTORTURE_RDR_SHIFT);
1218 	/* Mostly only one bit (need preemption!), sometimes lots of bits. */
1219 	if (!(randmask1 & 0x7))
1220 		mask = mask & randmask2;
1221 	else
1222 		mask = mask & (1 << (randmask2 % RCUTORTURE_RDR_NBITS));
1223 	/* Can't enable bh w/irq disabled. */
1224 	if ((mask & RCUTORTURE_RDR_IRQ) &&
1225 	    ((!(mask & RCUTORTURE_RDR_BH) && (oldmask & RCUTORTURE_RDR_BH)) ||
1226 	     (!(mask & RCUTORTURE_RDR_RBH) && (oldmask & RCUTORTURE_RDR_RBH))))
1227 		mask |= RCUTORTURE_RDR_BH | RCUTORTURE_RDR_RBH;
1228 	return mask ?: RCUTORTURE_RDR_RCU;
1229 }
1230 
1231 /*
1232  * Do a randomly selected number of extensions of an existing RCU read-side
1233  * critical section.
1234  */
1235 static struct rt_read_seg *
1236 rcutorture_loop_extend(int *readstate, struct torture_random_state *trsp,
1237 		       struct rt_read_seg *rtrsp)
1238 {
1239 	int i;
1240 	int j;
1241 	int mask = rcutorture_extend_mask_max();
1242 
1243 	WARN_ON_ONCE(!*readstate); /* -Existing- RCU read-side critsect! */
1244 	if (!((mask - 1) & mask))
1245 		return rtrsp;  /* Current RCU reader not extendable. */
1246 	/* Bias towards larger numbers of loops. */
1247 	i = (torture_random(trsp) >> 3);
1248 	i = ((i | (i >> 3)) & RCUTORTURE_RDR_MAX_LOOPS) + 1;
1249 	for (j = 0; j < i; j++) {
1250 		mask = rcutorture_extend_mask(*readstate, trsp);
1251 		rcutorture_one_extend(readstate, mask, trsp, &rtrsp[j]);
1252 	}
1253 	return &rtrsp[j];
1254 }
1255 
1256 /*
1257  * Do one read-side critical section, returning false if there was
1258  * no data to read.  Can be invoked both from process context and
1259  * from a timer handler.
1260  */
1261 static bool rcu_torture_one_read(struct torture_random_state *trsp)
1262 {
1263 	int i;
1264 	unsigned long started;
1265 	unsigned long completed;
1266 	int newstate;
1267 	struct rcu_torture *p;
1268 	int pipe_count;
1269 	int readstate = 0;
1270 	struct rt_read_seg rtseg[RCUTORTURE_RDR_MAX_SEGS] = { { 0 } };
1271 	struct rt_read_seg *rtrsp = &rtseg[0];
1272 	struct rt_read_seg *rtrsp1;
1273 	unsigned long long ts;
1274 
1275 	newstate = rcutorture_extend_mask(readstate, trsp);
1276 	rcutorture_one_extend(&readstate, newstate, trsp, rtrsp++);
1277 	started = cur_ops->get_gp_seq();
1278 	ts = rcu_trace_clock_local();
1279 	p = rcu_dereference_check(rcu_torture_current,
1280 				  rcu_read_lock_bh_held() ||
1281 				  rcu_read_lock_sched_held() ||
1282 				  srcu_read_lock_held(srcu_ctlp) ||
1283 				  torturing_tasks());
1284 	if (p == NULL) {
1285 		/* Wait for rcu_torture_writer to get underway */
1286 		rcutorture_one_extend(&readstate, 0, trsp, rtrsp);
1287 		return false;
1288 	}
1289 	if (p->rtort_mbtest == 0)
1290 		atomic_inc(&n_rcu_torture_mberror);
1291 	rtrsp = rcutorture_loop_extend(&readstate, trsp, rtrsp);
1292 	preempt_disable();
1293 	pipe_count = p->rtort_pipe_count;
1294 	if (pipe_count > RCU_TORTURE_PIPE_LEN) {
1295 		/* Should not happen, but... */
1296 		pipe_count = RCU_TORTURE_PIPE_LEN;
1297 	}
1298 	completed = cur_ops->get_gp_seq();
1299 	if (pipe_count > 1) {
1300 		do_trace_rcu_torture_read(cur_ops->name, &p->rtort_rcu,
1301 					  ts, started, completed);
1302 		rcu_ftrace_dump(DUMP_ALL);
1303 	}
1304 	__this_cpu_inc(rcu_torture_count[pipe_count]);
1305 	completed = rcutorture_seq_diff(completed, started);
1306 	if (completed > RCU_TORTURE_PIPE_LEN) {
1307 		/* Should not happen, but... */
1308 		completed = RCU_TORTURE_PIPE_LEN;
1309 	}
1310 	__this_cpu_inc(rcu_torture_batch[completed]);
1311 	preempt_enable();
1312 	rcutorture_one_extend(&readstate, 0, trsp, rtrsp);
1313 	WARN_ON_ONCE(readstate & RCUTORTURE_RDR_MASK);
1314 
1315 	/* If error or close call, record the sequence of reader protections. */
1316 	if ((pipe_count > 1 || completed > 1) && !xchg(&err_segs_recorded, 1)) {
1317 		i = 0;
1318 		for (rtrsp1 = &rtseg[0]; rtrsp1 < rtrsp; rtrsp1++)
1319 			err_segs[i++] = *rtrsp1;
1320 		rt_read_nsegs = i;
1321 	}
1322 
1323 	return true;
1324 }
1325 
1326 static DEFINE_TORTURE_RANDOM_PERCPU(rcu_torture_timer_rand);
1327 
1328 /*
1329  * RCU torture reader from timer handler.  Dereferences rcu_torture_current,
1330  * incrementing the corresponding element of the pipeline array.  The
1331  * counter in the element should never be greater than 1, otherwise, the
1332  * RCU implementation is broken.
1333  */
1334 static void rcu_torture_timer(struct timer_list *unused)
1335 {
1336 	atomic_long_inc(&n_rcu_torture_timers);
1337 	(void)rcu_torture_one_read(this_cpu_ptr(&rcu_torture_timer_rand));
1338 
1339 	/* Test call_rcu() invocation from interrupt handler. */
1340 	if (cur_ops->call) {
1341 		struct rcu_head *rhp = kmalloc(sizeof(*rhp), GFP_NOWAIT);
1342 
1343 		if (rhp)
1344 			cur_ops->call(rhp, rcu_torture_timer_cb);
1345 	}
1346 }
1347 
1348 /*
1349  * RCU torture reader kthread.  Repeatedly dereferences rcu_torture_current,
1350  * incrementing the corresponding element of the pipeline array.  The
1351  * counter in the element should never be greater than 1, otherwise, the
1352  * RCU implementation is broken.
1353  */
1354 static int
1355 rcu_torture_reader(void *arg)
1356 {
1357 	unsigned long lastsleep = jiffies;
1358 	long myid = (long)arg;
1359 	int mynumonline = myid;
1360 	DEFINE_TORTURE_RANDOM(rand);
1361 	struct timer_list t;
1362 
1363 	VERBOSE_TOROUT_STRING("rcu_torture_reader task started");
1364 	set_user_nice(current, MAX_NICE);
1365 	if (irqreader && cur_ops->irq_capable)
1366 		timer_setup_on_stack(&t, rcu_torture_timer, 0);
1367 	tick_dep_set_task(current, TICK_DEP_BIT_RCU);
1368 	do {
1369 		if (irqreader && cur_ops->irq_capable) {
1370 			if (!timer_pending(&t))
1371 				mod_timer(&t, jiffies + 1);
1372 		}
1373 		if (!rcu_torture_one_read(&rand) && !torture_must_stop())
1374 			schedule_timeout_interruptible(HZ);
1375 		if (time_after(jiffies, lastsleep) && !torture_must_stop()) {
1376 			schedule_timeout_interruptible(1);
1377 			lastsleep = jiffies + 10;
1378 		}
1379 		while (num_online_cpus() < mynumonline && !torture_must_stop())
1380 			schedule_timeout_interruptible(HZ / 5);
1381 		stutter_wait("rcu_torture_reader");
1382 	} while (!torture_must_stop());
1383 	if (irqreader && cur_ops->irq_capable) {
1384 		del_timer_sync(&t);
1385 		destroy_timer_on_stack(&t);
1386 	}
1387 	tick_dep_clear_task(current, TICK_DEP_BIT_RCU);
1388 	torture_kthread_stopping("rcu_torture_reader");
1389 	return 0;
1390 }
1391 
1392 /*
1393  * Print torture statistics.  Caller must ensure that there is only
1394  * one call to this function at a given time!!!  This is normally
1395  * accomplished by relying on the module system to only have one copy
1396  * of the module loaded, and then by giving the rcu_torture_stats
1397  * kthread full control (or the init/cleanup functions when rcu_torture_stats
1398  * thread is not running).
1399  */
1400 static void
1401 rcu_torture_stats_print(void)
1402 {
1403 	int cpu;
1404 	int i;
1405 	long pipesummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
1406 	long batchsummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
1407 	static unsigned long rtcv_snap = ULONG_MAX;
1408 	static bool splatted;
1409 	struct task_struct *wtp;
1410 
1411 	for_each_possible_cpu(cpu) {
1412 		for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
1413 			pipesummary[i] += per_cpu(rcu_torture_count, cpu)[i];
1414 			batchsummary[i] += per_cpu(rcu_torture_batch, cpu)[i];
1415 		}
1416 	}
1417 	for (i = RCU_TORTURE_PIPE_LEN - 1; i >= 0; i--) {
1418 		if (pipesummary[i] != 0)
1419 			break;
1420 	}
1421 
1422 	pr_alert("%s%s ", torture_type, TORTURE_FLAG);
1423 	pr_cont("rtc: %p %s: %lu tfle: %d rta: %d rtaf: %d rtf: %d ",
1424 		rcu_torture_current,
1425 		rcu_torture_current ? "ver" : "VER",
1426 		rcu_torture_current_version,
1427 		list_empty(&rcu_torture_freelist),
1428 		atomic_read(&n_rcu_torture_alloc),
1429 		atomic_read(&n_rcu_torture_alloc_fail),
1430 		atomic_read(&n_rcu_torture_free));
1431 	pr_cont("rtmbe: %d rtbe: %ld rtbke: %ld rtbre: %ld ",
1432 		atomic_read(&n_rcu_torture_mberror),
1433 		n_rcu_torture_barrier_error,
1434 		n_rcu_torture_boost_ktrerror,
1435 		n_rcu_torture_boost_rterror);
1436 	pr_cont("rtbf: %ld rtb: %ld nt: %ld ",
1437 		n_rcu_torture_boost_failure,
1438 		n_rcu_torture_boosts,
1439 		atomic_long_read(&n_rcu_torture_timers));
1440 	torture_onoff_stats();
1441 	pr_cont("barrier: %ld/%ld:%ld\n",
1442 		n_barrier_successes,
1443 		n_barrier_attempts,
1444 		n_rcu_torture_barrier_error);
1445 
1446 	pr_alert("%s%s ", torture_type, TORTURE_FLAG);
1447 	if (atomic_read(&n_rcu_torture_mberror) ||
1448 	    n_rcu_torture_barrier_error || n_rcu_torture_boost_ktrerror ||
1449 	    n_rcu_torture_boost_rterror || n_rcu_torture_boost_failure ||
1450 	    i > 1) {
1451 		pr_cont("%s", "!!! ");
1452 		atomic_inc(&n_rcu_torture_error);
1453 		WARN_ON_ONCE(atomic_read(&n_rcu_torture_mberror));
1454 		WARN_ON_ONCE(n_rcu_torture_barrier_error);  // rcu_barrier()
1455 		WARN_ON_ONCE(n_rcu_torture_boost_ktrerror); // no boost kthread
1456 		WARN_ON_ONCE(n_rcu_torture_boost_rterror); // can't set RT prio
1457 		WARN_ON_ONCE(n_rcu_torture_boost_failure); // RCU boost failed
1458 		WARN_ON_ONCE(i > 1); // Too-short grace period
1459 	}
1460 	pr_cont("Reader Pipe: ");
1461 	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
1462 		pr_cont(" %ld", pipesummary[i]);
1463 	pr_cont("\n");
1464 
1465 	pr_alert("%s%s ", torture_type, TORTURE_FLAG);
1466 	pr_cont("Reader Batch: ");
1467 	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
1468 		pr_cont(" %ld", batchsummary[i]);
1469 	pr_cont("\n");
1470 
1471 	pr_alert("%s%s ", torture_type, TORTURE_FLAG);
1472 	pr_cont("Free-Block Circulation: ");
1473 	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
1474 		pr_cont(" %d", atomic_read(&rcu_torture_wcount[i]));
1475 	}
1476 	pr_cont("\n");
1477 
1478 	if (cur_ops->stats)
1479 		cur_ops->stats();
1480 	if (rtcv_snap == rcu_torture_current_version &&
1481 	    rcu_torture_current != NULL) {
1482 		int __maybe_unused flags = 0;
1483 		unsigned long __maybe_unused gp_seq = 0;
1484 
1485 		rcutorture_get_gp_data(cur_ops->ttype,
1486 				       &flags, &gp_seq);
1487 		srcutorture_get_gp_data(cur_ops->ttype, srcu_ctlp,
1488 					&flags, &gp_seq);
1489 		wtp = READ_ONCE(writer_task);
1490 		pr_alert("??? Writer stall state %s(%d) g%lu f%#x ->state %#lx cpu %d\n",
1491 			 rcu_torture_writer_state_getname(),
1492 			 rcu_torture_writer_state, gp_seq, flags,
1493 			 wtp == NULL ? ~0UL : wtp->state,
1494 			 wtp == NULL ? -1 : (int)task_cpu(wtp));
1495 		if (!splatted && wtp) {
1496 			sched_show_task(wtp);
1497 			splatted = true;
1498 		}
1499 		show_rcu_gp_kthreads();
1500 		rcu_ftrace_dump(DUMP_ALL);
1501 	}
1502 	rtcv_snap = rcu_torture_current_version;
1503 }
1504 
1505 /*
1506  * Periodically prints torture statistics, if periodic statistics printing
1507  * was specified via the stat_interval module parameter.
1508  */
1509 static int
1510 rcu_torture_stats(void *arg)
1511 {
1512 	VERBOSE_TOROUT_STRING("rcu_torture_stats task started");
1513 	do {
1514 		schedule_timeout_interruptible(stat_interval * HZ);
1515 		rcu_torture_stats_print();
1516 		torture_shutdown_absorb("rcu_torture_stats");
1517 	} while (!torture_must_stop());
1518 	torture_kthread_stopping("rcu_torture_stats");
1519 	return 0;
1520 }
1521 
1522 static void
1523 rcu_torture_print_module_parms(struct rcu_torture_ops *cur_ops, const char *tag)
1524 {
1525 	pr_alert("%s" TORTURE_FLAG
1526 		 "--- %s: nreaders=%d nfakewriters=%d "
1527 		 "stat_interval=%d verbose=%d test_no_idle_hz=%d "
1528 		 "shuffle_interval=%d stutter=%d irqreader=%d "
1529 		 "fqs_duration=%d fqs_holdoff=%d fqs_stutter=%d "
1530 		 "test_boost=%d/%d test_boost_interval=%d "
1531 		 "test_boost_duration=%d shutdown_secs=%d "
1532 		 "stall_cpu=%d stall_cpu_holdoff=%d stall_cpu_irqsoff=%d "
1533 		 "n_barrier_cbs=%d "
1534 		 "onoff_interval=%d onoff_holdoff=%d\n",
1535 		 torture_type, tag, nrealreaders, nfakewriters,
1536 		 stat_interval, verbose, test_no_idle_hz, shuffle_interval,
1537 		 stutter, irqreader, fqs_duration, fqs_holdoff, fqs_stutter,
1538 		 test_boost, cur_ops->can_boost,
1539 		 test_boost_interval, test_boost_duration, shutdown_secs,
1540 		 stall_cpu, stall_cpu_holdoff, stall_cpu_irqsoff,
1541 		 n_barrier_cbs,
1542 		 onoff_interval, onoff_holdoff);
1543 }
1544 
1545 static int rcutorture_booster_cleanup(unsigned int cpu)
1546 {
1547 	struct task_struct *t;
1548 
1549 	if (boost_tasks[cpu] == NULL)
1550 		return 0;
1551 	mutex_lock(&boost_mutex);
1552 	t = boost_tasks[cpu];
1553 	boost_tasks[cpu] = NULL;
1554 	rcu_torture_enable_rt_throttle();
1555 	mutex_unlock(&boost_mutex);
1556 
1557 	/* This must be outside of the mutex, otherwise deadlock! */
1558 	torture_stop_kthread(rcu_torture_boost, t);
1559 	return 0;
1560 }
1561 
1562 static int rcutorture_booster_init(unsigned int cpu)
1563 {
1564 	int retval;
1565 
1566 	if (boost_tasks[cpu] != NULL)
1567 		return 0;  /* Already created, nothing more to do. */
1568 
1569 	/* Don't allow time recalculation while creating a new task. */
1570 	mutex_lock(&boost_mutex);
1571 	rcu_torture_disable_rt_throttle();
1572 	VERBOSE_TOROUT_STRING("Creating rcu_torture_boost task");
1573 	boost_tasks[cpu] = kthread_create_on_node(rcu_torture_boost, NULL,
1574 						  cpu_to_node(cpu),
1575 						  "rcu_torture_boost");
1576 	if (IS_ERR(boost_tasks[cpu])) {
1577 		retval = PTR_ERR(boost_tasks[cpu]);
1578 		VERBOSE_TOROUT_STRING("rcu_torture_boost task create failed");
1579 		n_rcu_torture_boost_ktrerror++;
1580 		boost_tasks[cpu] = NULL;
1581 		mutex_unlock(&boost_mutex);
1582 		return retval;
1583 	}
1584 	kthread_bind(boost_tasks[cpu], cpu);
1585 	wake_up_process(boost_tasks[cpu]);
1586 	mutex_unlock(&boost_mutex);
1587 	return 0;
1588 }
1589 
1590 /*
1591  * CPU-stall kthread.  It waits as specified by stall_cpu_holdoff, then
1592  * induces a CPU stall for the time specified by stall_cpu.
1593  */
1594 static int rcu_torture_stall(void *args)
1595 {
1596 	unsigned long stop_at;
1597 
1598 	VERBOSE_TOROUT_STRING("rcu_torture_stall task started");
1599 	if (stall_cpu_holdoff > 0) {
1600 		VERBOSE_TOROUT_STRING("rcu_torture_stall begin holdoff");
1601 		schedule_timeout_interruptible(stall_cpu_holdoff * HZ);
1602 		VERBOSE_TOROUT_STRING("rcu_torture_stall end holdoff");
1603 	}
1604 	if (!kthread_should_stop()) {
1605 		stop_at = ktime_get_seconds() + stall_cpu;
1606 		/* RCU CPU stall is expected behavior in following code. */
1607 		rcu_read_lock();
1608 		if (stall_cpu_irqsoff)
1609 			local_irq_disable();
1610 		else
1611 			preempt_disable();
1612 		pr_alert("rcu_torture_stall start on CPU %d.\n",
1613 			 smp_processor_id());
1614 		while (ULONG_CMP_LT((unsigned long)ktime_get_seconds(),
1615 				    stop_at))
1616 			continue;  /* Induce RCU CPU stall warning. */
1617 		if (stall_cpu_irqsoff)
1618 			local_irq_enable();
1619 		else
1620 			preempt_enable();
1621 		rcu_read_unlock();
1622 		pr_alert("rcu_torture_stall end.\n");
1623 	}
1624 	torture_shutdown_absorb("rcu_torture_stall");
1625 	while (!kthread_should_stop())
1626 		schedule_timeout_interruptible(10 * HZ);
1627 	return 0;
1628 }
1629 
1630 /* Spawn CPU-stall kthread, if stall_cpu specified. */
1631 static int __init rcu_torture_stall_init(void)
1632 {
1633 	if (stall_cpu <= 0)
1634 		return 0;
1635 	return torture_create_kthread(rcu_torture_stall, NULL, stall_task);
1636 }
1637 
1638 /* State structure for forward-progress self-propagating RCU callback. */
1639 struct fwd_cb_state {
1640 	struct rcu_head rh;
1641 	int stop;
1642 };
1643 
1644 /*
1645  * Forward-progress self-propagating RCU callback function.  Because
1646  * callbacks run from softirq, this function is an implicit RCU read-side
1647  * critical section.
1648  */
1649 static void rcu_torture_fwd_prog_cb(struct rcu_head *rhp)
1650 {
1651 	struct fwd_cb_state *fcsp = container_of(rhp, struct fwd_cb_state, rh);
1652 
1653 	if (READ_ONCE(fcsp->stop)) {
1654 		WRITE_ONCE(fcsp->stop, 2);
1655 		return;
1656 	}
1657 	cur_ops->call(&fcsp->rh, rcu_torture_fwd_prog_cb);
1658 }
1659 
1660 /* State for continuous-flood RCU callbacks. */
1661 struct rcu_fwd_cb {
1662 	struct rcu_head rh;
1663 	struct rcu_fwd_cb *rfc_next;
1664 	struct rcu_fwd *rfc_rfp;
1665 	int rfc_gps;
1666 };
1667 
1668 #define MAX_FWD_CB_JIFFIES	(8 * HZ) /* Maximum CB test duration. */
1669 #define MIN_FWD_CB_LAUNDERS	3	/* This many CB invocations to count. */
1670 #define MIN_FWD_CBS_LAUNDERED	100	/* Number of counted CBs. */
1671 #define FWD_CBS_HIST_DIV	10	/* Histogram buckets/second. */
1672 #define N_LAUNDERS_HIST (2 * MAX_FWD_CB_JIFFIES / (HZ / FWD_CBS_HIST_DIV))
1673 
1674 struct rcu_launder_hist {
1675 	long n_launders;
1676 	unsigned long launder_gp_seq;
1677 };
1678 
1679 struct rcu_fwd {
1680 	spinlock_t rcu_fwd_lock;
1681 	struct rcu_fwd_cb *rcu_fwd_cb_head;
1682 	struct rcu_fwd_cb **rcu_fwd_cb_tail;
1683 	long n_launders_cb;
1684 	unsigned long rcu_fwd_startat;
1685 	struct rcu_launder_hist n_launders_hist[N_LAUNDERS_HIST];
1686 	unsigned long rcu_launder_gp_seq_start;
1687 };
1688 
1689 struct rcu_fwd *rcu_fwds;
1690 bool rcu_fwd_emergency_stop;
1691 
1692 static void rcu_torture_fwd_cb_hist(struct rcu_fwd *rfp)
1693 {
1694 	unsigned long gps;
1695 	unsigned long gps_old;
1696 	int i;
1697 	int j;
1698 
1699 	for (i = ARRAY_SIZE(rfp->n_launders_hist) - 1; i > 0; i--)
1700 		if (rfp->n_launders_hist[i].n_launders > 0)
1701 			break;
1702 	pr_alert("%s: Callback-invocation histogram (duration %lu jiffies):",
1703 		 __func__, jiffies - rfp->rcu_fwd_startat);
1704 	gps_old = rfp->rcu_launder_gp_seq_start;
1705 	for (j = 0; j <= i; j++) {
1706 		gps = rfp->n_launders_hist[j].launder_gp_seq;
1707 		pr_cont(" %ds/%d: %ld:%ld",
1708 			j + 1, FWD_CBS_HIST_DIV,
1709 			rfp->n_launders_hist[j].n_launders,
1710 			rcutorture_seq_diff(gps, gps_old));
1711 		gps_old = gps;
1712 	}
1713 	pr_cont("\n");
1714 }
1715 
1716 /* Callback function for continuous-flood RCU callbacks. */
1717 static void rcu_torture_fwd_cb_cr(struct rcu_head *rhp)
1718 {
1719 	unsigned long flags;
1720 	int i;
1721 	struct rcu_fwd_cb *rfcp = container_of(rhp, struct rcu_fwd_cb, rh);
1722 	struct rcu_fwd_cb **rfcpp;
1723 	struct rcu_fwd *rfp = rfcp->rfc_rfp;
1724 
1725 	rfcp->rfc_next = NULL;
1726 	rfcp->rfc_gps++;
1727 	spin_lock_irqsave(&rfp->rcu_fwd_lock, flags);
1728 	rfcpp = rfp->rcu_fwd_cb_tail;
1729 	rfp->rcu_fwd_cb_tail = &rfcp->rfc_next;
1730 	WRITE_ONCE(*rfcpp, rfcp);
1731 	WRITE_ONCE(rfp->n_launders_cb, rfp->n_launders_cb + 1);
1732 	i = ((jiffies - rfp->rcu_fwd_startat) / (HZ / FWD_CBS_HIST_DIV));
1733 	if (i >= ARRAY_SIZE(rfp->n_launders_hist))
1734 		i = ARRAY_SIZE(rfp->n_launders_hist) - 1;
1735 	rfp->n_launders_hist[i].n_launders++;
1736 	rfp->n_launders_hist[i].launder_gp_seq = cur_ops->get_gp_seq();
1737 	spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
1738 }
1739 
1740 // Give the scheduler a chance, even on nohz_full CPUs.
1741 static void rcu_torture_fwd_prog_cond_resched(unsigned long iter)
1742 {
1743 	if (IS_ENABLED(CONFIG_PREEMPTION) && IS_ENABLED(CONFIG_NO_HZ_FULL)) {
1744 		// Real call_rcu() floods hit userspace, so emulate that.
1745 		if (need_resched() || (iter & 0xfff))
1746 			schedule();
1747 		return;
1748 	}
1749 	// No userspace emulation: CB invocation throttles call_rcu()
1750 	cond_resched();
1751 }
1752 
1753 /*
1754  * Free all callbacks on the rcu_fwd_cb_head list, either because the
1755  * test is over or because we hit an OOM event.
1756  */
1757 static unsigned long rcu_torture_fwd_prog_cbfree(struct rcu_fwd *rfp)
1758 {
1759 	unsigned long flags;
1760 	unsigned long freed = 0;
1761 	struct rcu_fwd_cb *rfcp;
1762 
1763 	for (;;) {
1764 		spin_lock_irqsave(&rfp->rcu_fwd_lock, flags);
1765 		rfcp = rfp->rcu_fwd_cb_head;
1766 		if (!rfcp) {
1767 			spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
1768 			break;
1769 		}
1770 		rfp->rcu_fwd_cb_head = rfcp->rfc_next;
1771 		if (!rfp->rcu_fwd_cb_head)
1772 			rfp->rcu_fwd_cb_tail = &rfp->rcu_fwd_cb_head;
1773 		spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
1774 		kfree(rfcp);
1775 		freed++;
1776 		rcu_torture_fwd_prog_cond_resched(freed);
1777 		if (tick_nohz_full_enabled()) {
1778 			local_irq_save(flags);
1779 			rcu_momentary_dyntick_idle();
1780 			local_irq_restore(flags);
1781 		}
1782 	}
1783 	return freed;
1784 }
1785 
1786 /* Carry out need_resched()/cond_resched() forward-progress testing. */
1787 static void rcu_torture_fwd_prog_nr(struct rcu_fwd *rfp,
1788 				    int *tested, int *tested_tries)
1789 {
1790 	unsigned long cver;
1791 	unsigned long dur;
1792 	struct fwd_cb_state fcs;
1793 	unsigned long gps;
1794 	int idx;
1795 	int sd;
1796 	int sd4;
1797 	bool selfpropcb = false;
1798 	unsigned long stopat;
1799 	static DEFINE_TORTURE_RANDOM(trs);
1800 
1801 	if  (cur_ops->call && cur_ops->sync && cur_ops->cb_barrier) {
1802 		init_rcu_head_on_stack(&fcs.rh);
1803 		selfpropcb = true;
1804 	}
1805 
1806 	/* Tight loop containing cond_resched(). */
1807 	WRITE_ONCE(rcu_fwd_cb_nodelay, true);
1808 	cur_ops->sync(); /* Later readers see above write. */
1809 	if  (selfpropcb) {
1810 		WRITE_ONCE(fcs.stop, 0);
1811 		cur_ops->call(&fcs.rh, rcu_torture_fwd_prog_cb);
1812 	}
1813 	cver = READ_ONCE(rcu_torture_current_version);
1814 	gps = cur_ops->get_gp_seq();
1815 	sd = cur_ops->stall_dur() + 1;
1816 	sd4 = (sd + fwd_progress_div - 1) / fwd_progress_div;
1817 	dur = sd4 + torture_random(&trs) % (sd - sd4);
1818 	WRITE_ONCE(rfp->rcu_fwd_startat, jiffies);
1819 	stopat = rfp->rcu_fwd_startat + dur;
1820 	while (time_before(jiffies, stopat) &&
1821 	       !shutdown_time_arrived() &&
1822 	       !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
1823 		idx = cur_ops->readlock();
1824 		udelay(10);
1825 		cur_ops->readunlock(idx);
1826 		if (!fwd_progress_need_resched || need_resched())
1827 			cond_resched();
1828 	}
1829 	(*tested_tries)++;
1830 	if (!time_before(jiffies, stopat) &&
1831 	    !shutdown_time_arrived() &&
1832 	    !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
1833 		(*tested)++;
1834 		cver = READ_ONCE(rcu_torture_current_version) - cver;
1835 		gps = rcutorture_seq_diff(cur_ops->get_gp_seq(), gps);
1836 		WARN_ON(!cver && gps < 2);
1837 		pr_alert("%s: Duration %ld cver %ld gps %ld\n", __func__, dur, cver, gps);
1838 	}
1839 	if (selfpropcb) {
1840 		WRITE_ONCE(fcs.stop, 1);
1841 		cur_ops->sync(); /* Wait for running CB to complete. */
1842 		cur_ops->cb_barrier(); /* Wait for queued callbacks. */
1843 	}
1844 
1845 	if (selfpropcb) {
1846 		WARN_ON(READ_ONCE(fcs.stop) != 2);
1847 		destroy_rcu_head_on_stack(&fcs.rh);
1848 	}
1849 	schedule_timeout_uninterruptible(HZ / 10); /* Let kthreads recover. */
1850 	WRITE_ONCE(rcu_fwd_cb_nodelay, false);
1851 }
1852 
1853 /* Carry out call_rcu() forward-progress testing. */
1854 static void rcu_torture_fwd_prog_cr(struct rcu_fwd *rfp)
1855 {
1856 	unsigned long cver;
1857 	unsigned long flags;
1858 	unsigned long gps;
1859 	int i;
1860 	long n_launders;
1861 	long n_launders_cb_snap;
1862 	long n_launders_sa;
1863 	long n_max_cbs;
1864 	long n_max_gps;
1865 	struct rcu_fwd_cb *rfcp;
1866 	struct rcu_fwd_cb *rfcpn;
1867 	unsigned long stopat;
1868 	unsigned long stoppedat;
1869 
1870 	if (READ_ONCE(rcu_fwd_emergency_stop))
1871 		return; /* Get out of the way quickly, no GP wait! */
1872 	if (!cur_ops->call)
1873 		return; /* Can't do call_rcu() fwd prog without ->call. */
1874 
1875 	/* Loop continuously posting RCU callbacks. */
1876 	WRITE_ONCE(rcu_fwd_cb_nodelay, true);
1877 	cur_ops->sync(); /* Later readers see above write. */
1878 	WRITE_ONCE(rfp->rcu_fwd_startat, jiffies);
1879 	stopat = rfp->rcu_fwd_startat + MAX_FWD_CB_JIFFIES;
1880 	n_launders = 0;
1881 	rfp->n_launders_cb = 0; // Hoist initialization for multi-kthread
1882 	n_launders_sa = 0;
1883 	n_max_cbs = 0;
1884 	n_max_gps = 0;
1885 	for (i = 0; i < ARRAY_SIZE(rfp->n_launders_hist); i++)
1886 		rfp->n_launders_hist[i].n_launders = 0;
1887 	cver = READ_ONCE(rcu_torture_current_version);
1888 	gps = cur_ops->get_gp_seq();
1889 	rfp->rcu_launder_gp_seq_start = gps;
1890 	tick_dep_set_task(current, TICK_DEP_BIT_RCU);
1891 	while (time_before(jiffies, stopat) &&
1892 	       !shutdown_time_arrived() &&
1893 	       !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
1894 		rfcp = READ_ONCE(rfp->rcu_fwd_cb_head);
1895 		rfcpn = NULL;
1896 		if (rfcp)
1897 			rfcpn = READ_ONCE(rfcp->rfc_next);
1898 		if (rfcpn) {
1899 			if (rfcp->rfc_gps >= MIN_FWD_CB_LAUNDERS &&
1900 			    ++n_max_gps >= MIN_FWD_CBS_LAUNDERED)
1901 				break;
1902 			rfp->rcu_fwd_cb_head = rfcpn;
1903 			n_launders++;
1904 			n_launders_sa++;
1905 		} else {
1906 			rfcp = kmalloc(sizeof(*rfcp), GFP_KERNEL);
1907 			if (WARN_ON_ONCE(!rfcp)) {
1908 				schedule_timeout_interruptible(1);
1909 				continue;
1910 			}
1911 			n_max_cbs++;
1912 			n_launders_sa = 0;
1913 			rfcp->rfc_gps = 0;
1914 			rfcp->rfc_rfp = rfp;
1915 		}
1916 		cur_ops->call(&rfcp->rh, rcu_torture_fwd_cb_cr);
1917 		rcu_torture_fwd_prog_cond_resched(n_launders + n_max_cbs);
1918 		if (tick_nohz_full_enabled()) {
1919 			local_irq_save(flags);
1920 			rcu_momentary_dyntick_idle();
1921 			local_irq_restore(flags);
1922 		}
1923 	}
1924 	stoppedat = jiffies;
1925 	n_launders_cb_snap = READ_ONCE(rfp->n_launders_cb);
1926 	cver = READ_ONCE(rcu_torture_current_version) - cver;
1927 	gps = rcutorture_seq_diff(cur_ops->get_gp_seq(), gps);
1928 	cur_ops->cb_barrier(); /* Wait for callbacks to be invoked. */
1929 	(void)rcu_torture_fwd_prog_cbfree(rfp);
1930 
1931 	if (!torture_must_stop() && !READ_ONCE(rcu_fwd_emergency_stop) &&
1932 	    !shutdown_time_arrived()) {
1933 		WARN_ON(n_max_gps < MIN_FWD_CBS_LAUNDERED);
1934 		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",
1935 			 __func__,
1936 			 stoppedat - rfp->rcu_fwd_startat, jiffies - stoppedat,
1937 			 n_launders + n_max_cbs - n_launders_cb_snap,
1938 			 n_launders, n_launders_sa,
1939 			 n_max_gps, n_max_cbs, cver, gps);
1940 		rcu_torture_fwd_cb_hist(rfp);
1941 	}
1942 	schedule_timeout_uninterruptible(HZ); /* Let CBs drain. */
1943 	tick_dep_clear_task(current, TICK_DEP_BIT_RCU);
1944 	WRITE_ONCE(rcu_fwd_cb_nodelay, false);
1945 }
1946 
1947 
1948 /*
1949  * OOM notifier, but this only prints diagnostic information for the
1950  * current forward-progress test.
1951  */
1952 static int rcutorture_oom_notify(struct notifier_block *self,
1953 				 unsigned long notused, void *nfreed)
1954 {
1955 	struct rcu_fwd *rfp = rcu_fwds;
1956 
1957 	WARN(1, "%s invoked upon OOM during forward-progress testing.\n",
1958 	     __func__);
1959 	rcu_torture_fwd_cb_hist(rfp);
1960 	rcu_fwd_progress_check(1 + (jiffies - READ_ONCE(rfp->rcu_fwd_startat)) / 2);
1961 	WRITE_ONCE(rcu_fwd_emergency_stop, true);
1962 	smp_mb(); /* Emergency stop before free and wait to avoid hangs. */
1963 	pr_info("%s: Freed %lu RCU callbacks.\n",
1964 		__func__, rcu_torture_fwd_prog_cbfree(rfp));
1965 	rcu_barrier();
1966 	pr_info("%s: Freed %lu RCU callbacks.\n",
1967 		__func__, rcu_torture_fwd_prog_cbfree(rfp));
1968 	rcu_barrier();
1969 	pr_info("%s: Freed %lu RCU callbacks.\n",
1970 		__func__, rcu_torture_fwd_prog_cbfree(rfp));
1971 	smp_mb(); /* Frees before return to avoid redoing OOM. */
1972 	(*(unsigned long *)nfreed)++; /* Forward progress CBs freed! */
1973 	pr_info("%s returning after OOM processing.\n", __func__);
1974 	return NOTIFY_OK;
1975 }
1976 
1977 static struct notifier_block rcutorture_oom_nb = {
1978 	.notifier_call = rcutorture_oom_notify
1979 };
1980 
1981 /* Carry out grace-period forward-progress testing. */
1982 static int rcu_torture_fwd_prog(void *args)
1983 {
1984 	struct rcu_fwd *rfp = args;
1985 	int tested = 0;
1986 	int tested_tries = 0;
1987 
1988 	VERBOSE_TOROUT_STRING("rcu_torture_fwd_progress task started");
1989 	rcu_bind_current_to_nocb();
1990 	if (!IS_ENABLED(CONFIG_SMP) || !IS_ENABLED(CONFIG_RCU_BOOST))
1991 		set_user_nice(current, MAX_NICE);
1992 	do {
1993 		schedule_timeout_interruptible(fwd_progress_holdoff * HZ);
1994 		WRITE_ONCE(rcu_fwd_emergency_stop, false);
1995 		register_oom_notifier(&rcutorture_oom_nb);
1996 		rcu_torture_fwd_prog_nr(rfp, &tested, &tested_tries);
1997 		rcu_torture_fwd_prog_cr(rfp);
1998 		unregister_oom_notifier(&rcutorture_oom_nb);
1999 
2000 		/* Avoid slow periods, better to test when busy. */
2001 		stutter_wait("rcu_torture_fwd_prog");
2002 	} while (!torture_must_stop());
2003 	/* Short runs might not contain a valid forward-progress attempt. */
2004 	WARN_ON(!tested && tested_tries >= 5);
2005 	pr_alert("%s: tested %d tested_tries %d\n", __func__, tested, tested_tries);
2006 	torture_kthread_stopping("rcu_torture_fwd_prog");
2007 	return 0;
2008 }
2009 
2010 /* If forward-progress checking is requested and feasible, spawn the thread. */
2011 static int __init rcu_torture_fwd_prog_init(void)
2012 {
2013 	struct rcu_fwd *rfp;
2014 
2015 	if (!fwd_progress)
2016 		return 0; /* Not requested, so don't do it. */
2017 	if (!cur_ops->stall_dur || cur_ops->stall_dur() <= 0 ||
2018 	    cur_ops == &rcu_busted_ops) {
2019 		VERBOSE_TOROUT_STRING("rcu_torture_fwd_prog_init: Disabled, unsupported by RCU flavor under test");
2020 		return 0;
2021 	}
2022 	if (stall_cpu > 0) {
2023 		VERBOSE_TOROUT_STRING("rcu_torture_fwd_prog_init: Disabled, conflicts with CPU-stall testing");
2024 		if (IS_MODULE(CONFIG_RCU_TORTURE_TESTS))
2025 			return -EINVAL; /* In module, can fail back to user. */
2026 		WARN_ON(1); /* Make sure rcutorture notices conflict. */
2027 		return 0;
2028 	}
2029 	if (fwd_progress_holdoff <= 0)
2030 		fwd_progress_holdoff = 1;
2031 	if (fwd_progress_div <= 0)
2032 		fwd_progress_div = 4;
2033 	rfp = kzalloc(sizeof(*rfp), GFP_KERNEL);
2034 	if (!rfp)
2035 		return -ENOMEM;
2036 	spin_lock_init(&rfp->rcu_fwd_lock);
2037 	rfp->rcu_fwd_cb_tail = &rfp->rcu_fwd_cb_head;
2038 	return torture_create_kthread(rcu_torture_fwd_prog, rfp, fwd_prog_task);
2039 }
2040 
2041 /* Callback function for RCU barrier testing. */
2042 static void rcu_torture_barrier_cbf(struct rcu_head *rcu)
2043 {
2044 	atomic_inc(&barrier_cbs_invoked);
2045 }
2046 
2047 /* kthread function to register callbacks used to test RCU barriers. */
2048 static int rcu_torture_barrier_cbs(void *arg)
2049 {
2050 	long myid = (long)arg;
2051 	bool lastphase = 0;
2052 	bool newphase;
2053 	struct rcu_head rcu;
2054 
2055 	init_rcu_head_on_stack(&rcu);
2056 	VERBOSE_TOROUT_STRING("rcu_torture_barrier_cbs task started");
2057 	set_user_nice(current, MAX_NICE);
2058 	do {
2059 		wait_event(barrier_cbs_wq[myid],
2060 			   (newphase =
2061 			    smp_load_acquire(&barrier_phase)) != lastphase ||
2062 			   torture_must_stop());
2063 		lastphase = newphase;
2064 		if (torture_must_stop())
2065 			break;
2066 		/*
2067 		 * The above smp_load_acquire() ensures barrier_phase load
2068 		 * is ordered before the following ->call().
2069 		 */
2070 		local_irq_disable(); /* Just to test no-irq call_rcu(). */
2071 		cur_ops->call(&rcu, rcu_torture_barrier_cbf);
2072 		local_irq_enable();
2073 		if (atomic_dec_and_test(&barrier_cbs_count))
2074 			wake_up(&barrier_wq);
2075 	} while (!torture_must_stop());
2076 	if (cur_ops->cb_barrier != NULL)
2077 		cur_ops->cb_barrier();
2078 	destroy_rcu_head_on_stack(&rcu);
2079 	torture_kthread_stopping("rcu_torture_barrier_cbs");
2080 	return 0;
2081 }
2082 
2083 /* kthread function to drive and coordinate RCU barrier testing. */
2084 static int rcu_torture_barrier(void *arg)
2085 {
2086 	int i;
2087 
2088 	VERBOSE_TOROUT_STRING("rcu_torture_barrier task starting");
2089 	do {
2090 		atomic_set(&barrier_cbs_invoked, 0);
2091 		atomic_set(&barrier_cbs_count, n_barrier_cbs);
2092 		/* Ensure barrier_phase ordered after prior assignments. */
2093 		smp_store_release(&barrier_phase, !barrier_phase);
2094 		for (i = 0; i < n_barrier_cbs; i++)
2095 			wake_up(&barrier_cbs_wq[i]);
2096 		wait_event(barrier_wq,
2097 			   atomic_read(&barrier_cbs_count) == 0 ||
2098 			   torture_must_stop());
2099 		if (torture_must_stop())
2100 			break;
2101 		n_barrier_attempts++;
2102 		cur_ops->cb_barrier(); /* Implies smp_mb() for wait_event(). */
2103 		if (atomic_read(&barrier_cbs_invoked) != n_barrier_cbs) {
2104 			n_rcu_torture_barrier_error++;
2105 			pr_err("barrier_cbs_invoked = %d, n_barrier_cbs = %d\n",
2106 			       atomic_read(&barrier_cbs_invoked),
2107 			       n_barrier_cbs);
2108 			WARN_ON_ONCE(1);
2109 		} else {
2110 			n_barrier_successes++;
2111 		}
2112 		schedule_timeout_interruptible(HZ / 10);
2113 	} while (!torture_must_stop());
2114 	torture_kthread_stopping("rcu_torture_barrier");
2115 	return 0;
2116 }
2117 
2118 /* Initialize RCU barrier testing. */
2119 static int rcu_torture_barrier_init(void)
2120 {
2121 	int i;
2122 	int ret;
2123 
2124 	if (n_barrier_cbs <= 0)
2125 		return 0;
2126 	if (cur_ops->call == NULL || cur_ops->cb_barrier == NULL) {
2127 		pr_alert("%s" TORTURE_FLAG
2128 			 " Call or barrier ops missing for %s,\n",
2129 			 torture_type, cur_ops->name);
2130 		pr_alert("%s" TORTURE_FLAG
2131 			 " RCU barrier testing omitted from run.\n",
2132 			 torture_type);
2133 		return 0;
2134 	}
2135 	atomic_set(&barrier_cbs_count, 0);
2136 	atomic_set(&barrier_cbs_invoked, 0);
2137 	barrier_cbs_tasks =
2138 		kcalloc(n_barrier_cbs, sizeof(barrier_cbs_tasks[0]),
2139 			GFP_KERNEL);
2140 	barrier_cbs_wq =
2141 		kcalloc(n_barrier_cbs, sizeof(barrier_cbs_wq[0]), GFP_KERNEL);
2142 	if (barrier_cbs_tasks == NULL || !barrier_cbs_wq)
2143 		return -ENOMEM;
2144 	for (i = 0; i < n_barrier_cbs; i++) {
2145 		init_waitqueue_head(&barrier_cbs_wq[i]);
2146 		ret = torture_create_kthread(rcu_torture_barrier_cbs,
2147 					     (void *)(long)i,
2148 					     barrier_cbs_tasks[i]);
2149 		if (ret)
2150 			return ret;
2151 	}
2152 	return torture_create_kthread(rcu_torture_barrier, NULL, barrier_task);
2153 }
2154 
2155 /* Clean up after RCU barrier testing. */
2156 static void rcu_torture_barrier_cleanup(void)
2157 {
2158 	int i;
2159 
2160 	torture_stop_kthread(rcu_torture_barrier, barrier_task);
2161 	if (barrier_cbs_tasks != NULL) {
2162 		for (i = 0; i < n_barrier_cbs; i++)
2163 			torture_stop_kthread(rcu_torture_barrier_cbs,
2164 					     barrier_cbs_tasks[i]);
2165 		kfree(barrier_cbs_tasks);
2166 		barrier_cbs_tasks = NULL;
2167 	}
2168 	if (barrier_cbs_wq != NULL) {
2169 		kfree(barrier_cbs_wq);
2170 		barrier_cbs_wq = NULL;
2171 	}
2172 }
2173 
2174 static bool rcu_torture_can_boost(void)
2175 {
2176 	static int boost_warn_once;
2177 	int prio;
2178 
2179 	if (!(test_boost == 1 && cur_ops->can_boost) && test_boost != 2)
2180 		return false;
2181 
2182 	prio = rcu_get_gp_kthreads_prio();
2183 	if (!prio)
2184 		return false;
2185 
2186 	if (prio < 2) {
2187 		if (boost_warn_once  == 1)
2188 			return false;
2189 
2190 		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);
2191 		boost_warn_once = 1;
2192 		return false;
2193 	}
2194 
2195 	return true;
2196 }
2197 
2198 static enum cpuhp_state rcutor_hp;
2199 
2200 static void
2201 rcu_torture_cleanup(void)
2202 {
2203 	int firsttime;
2204 	int flags = 0;
2205 	unsigned long gp_seq = 0;
2206 	int i;
2207 
2208 	if (torture_cleanup_begin()) {
2209 		if (cur_ops->cb_barrier != NULL)
2210 			cur_ops->cb_barrier();
2211 		return;
2212 	}
2213 	if (!cur_ops) {
2214 		torture_cleanup_end();
2215 		return;
2216 	}
2217 
2218 	show_rcu_gp_kthreads();
2219 	rcu_torture_barrier_cleanup();
2220 	torture_stop_kthread(rcu_torture_fwd_prog, fwd_prog_task);
2221 	torture_stop_kthread(rcu_torture_stall, stall_task);
2222 	torture_stop_kthread(rcu_torture_writer, writer_task);
2223 
2224 	if (reader_tasks) {
2225 		for (i = 0; i < nrealreaders; i++)
2226 			torture_stop_kthread(rcu_torture_reader,
2227 					     reader_tasks[i]);
2228 		kfree(reader_tasks);
2229 	}
2230 	rcu_torture_current = NULL;
2231 
2232 	if (fakewriter_tasks) {
2233 		for (i = 0; i < nfakewriters; i++) {
2234 			torture_stop_kthread(rcu_torture_fakewriter,
2235 					     fakewriter_tasks[i]);
2236 		}
2237 		kfree(fakewriter_tasks);
2238 		fakewriter_tasks = NULL;
2239 	}
2240 
2241 	rcutorture_get_gp_data(cur_ops->ttype, &flags, &gp_seq);
2242 	srcutorture_get_gp_data(cur_ops->ttype, srcu_ctlp, &flags, &gp_seq);
2243 	pr_alert("%s:  End-test grace-period state: g%lu f%#x\n",
2244 		 cur_ops->name, gp_seq, flags);
2245 	torture_stop_kthread(rcu_torture_stats, stats_task);
2246 	torture_stop_kthread(rcu_torture_fqs, fqs_task);
2247 	if (rcu_torture_can_boost())
2248 		cpuhp_remove_state(rcutor_hp);
2249 
2250 	/*
2251 	 * Wait for all RCU callbacks to fire, then do torture-type-specific
2252 	 * cleanup operations.
2253 	 */
2254 	if (cur_ops->cb_barrier != NULL)
2255 		cur_ops->cb_barrier();
2256 	if (cur_ops->cleanup != NULL)
2257 		cur_ops->cleanup();
2258 
2259 	rcu_torture_stats_print();  /* -After- the stats thread is stopped! */
2260 
2261 	if (err_segs_recorded) {
2262 		pr_alert("Failure/close-call rcutorture reader segments:\n");
2263 		if (rt_read_nsegs == 0)
2264 			pr_alert("\t: No segments recorded!!!\n");
2265 		firsttime = 1;
2266 		for (i = 0; i < rt_read_nsegs; i++) {
2267 			pr_alert("\t%d: %#x ", i, err_segs[i].rt_readstate);
2268 			if (err_segs[i].rt_delay_jiffies != 0) {
2269 				pr_cont("%s%ldjiffies", firsttime ? "" : "+",
2270 					err_segs[i].rt_delay_jiffies);
2271 				firsttime = 0;
2272 			}
2273 			if (err_segs[i].rt_delay_ms != 0) {
2274 				pr_cont("%s%ldms", firsttime ? "" : "+",
2275 					err_segs[i].rt_delay_ms);
2276 				firsttime = 0;
2277 			}
2278 			if (err_segs[i].rt_delay_us != 0) {
2279 				pr_cont("%s%ldus", firsttime ? "" : "+",
2280 					err_segs[i].rt_delay_us);
2281 				firsttime = 0;
2282 			}
2283 			pr_cont("%s\n",
2284 				err_segs[i].rt_preempted ? "preempted" : "");
2285 
2286 		}
2287 	}
2288 	if (atomic_read(&n_rcu_torture_error) || n_rcu_torture_barrier_error)
2289 		rcu_torture_print_module_parms(cur_ops, "End of test: FAILURE");
2290 	else if (torture_onoff_failures())
2291 		rcu_torture_print_module_parms(cur_ops,
2292 					       "End of test: RCU_HOTPLUG");
2293 	else
2294 		rcu_torture_print_module_parms(cur_ops, "End of test: SUCCESS");
2295 	torture_cleanup_end();
2296 }
2297 
2298 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
2299 static void rcu_torture_leak_cb(struct rcu_head *rhp)
2300 {
2301 }
2302 
2303 static void rcu_torture_err_cb(struct rcu_head *rhp)
2304 {
2305 	/*
2306 	 * This -might- happen due to race conditions, but is unlikely.
2307 	 * The scenario that leads to this happening is that the
2308 	 * first of the pair of duplicate callbacks is queued,
2309 	 * someone else starts a grace period that includes that
2310 	 * callback, then the second of the pair must wait for the
2311 	 * next grace period.  Unlikely, but can happen.  If it
2312 	 * does happen, the debug-objects subsystem won't have splatted.
2313 	 */
2314 	pr_alert("%s: duplicated callback was invoked.\n", KBUILD_MODNAME);
2315 }
2316 #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
2317 
2318 /*
2319  * Verify that double-free causes debug-objects to complain, but only
2320  * if CONFIG_DEBUG_OBJECTS_RCU_HEAD=y.  Otherwise, say that the test
2321  * cannot be carried out.
2322  */
2323 static void rcu_test_debug_objects(void)
2324 {
2325 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
2326 	struct rcu_head rh1;
2327 	struct rcu_head rh2;
2328 
2329 	init_rcu_head_on_stack(&rh1);
2330 	init_rcu_head_on_stack(&rh2);
2331 	pr_alert("%s: WARN: Duplicate call_rcu() test starting.\n", KBUILD_MODNAME);
2332 
2333 	/* Try to queue the rh2 pair of callbacks for the same grace period. */
2334 	preempt_disable(); /* Prevent preemption from interrupting test. */
2335 	rcu_read_lock(); /* Make it impossible to finish a grace period. */
2336 	call_rcu(&rh1, rcu_torture_leak_cb); /* Start grace period. */
2337 	local_irq_disable(); /* Make it harder to start a new grace period. */
2338 	call_rcu(&rh2, rcu_torture_leak_cb);
2339 	call_rcu(&rh2, rcu_torture_err_cb); /* Duplicate callback. */
2340 	local_irq_enable();
2341 	rcu_read_unlock();
2342 	preempt_enable();
2343 
2344 	/* Wait for them all to get done so we can safely return. */
2345 	rcu_barrier();
2346 	pr_alert("%s: WARN: Duplicate call_rcu() test complete.\n", KBUILD_MODNAME);
2347 	destroy_rcu_head_on_stack(&rh1);
2348 	destroy_rcu_head_on_stack(&rh2);
2349 #else /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
2350 	pr_alert("%s: !CONFIG_DEBUG_OBJECTS_RCU_HEAD, not testing duplicate call_rcu()\n", KBUILD_MODNAME);
2351 #endif /* #else #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
2352 }
2353 
2354 static void rcutorture_sync(void)
2355 {
2356 	static unsigned long n;
2357 
2358 	if (cur_ops->sync && !(++n & 0xfff))
2359 		cur_ops->sync();
2360 }
2361 
2362 static int __init
2363 rcu_torture_init(void)
2364 {
2365 	long i;
2366 	int cpu;
2367 	int firsterr = 0;
2368 	static struct rcu_torture_ops *torture_ops[] = {
2369 		&rcu_ops, &rcu_busted_ops, &srcu_ops, &srcud_ops,
2370 		&busted_srcud_ops, &tasks_ops, &trivial_ops,
2371 	};
2372 
2373 	if (!torture_init_begin(torture_type, verbose))
2374 		return -EBUSY;
2375 
2376 	/* Process args and tell the world that the torturer is on the job. */
2377 	for (i = 0; i < ARRAY_SIZE(torture_ops); i++) {
2378 		cur_ops = torture_ops[i];
2379 		if (strcmp(torture_type, cur_ops->name) == 0)
2380 			break;
2381 	}
2382 	if (i == ARRAY_SIZE(torture_ops)) {
2383 		pr_alert("rcu-torture: invalid torture type: \"%s\"\n",
2384 			 torture_type);
2385 		pr_alert("rcu-torture types:");
2386 		for (i = 0; i < ARRAY_SIZE(torture_ops); i++)
2387 			pr_cont(" %s", torture_ops[i]->name);
2388 		pr_cont("\n");
2389 		WARN_ON(!IS_MODULE(CONFIG_RCU_TORTURE_TEST));
2390 		firsterr = -EINVAL;
2391 		cur_ops = NULL;
2392 		goto unwind;
2393 	}
2394 	if (cur_ops->fqs == NULL && fqs_duration != 0) {
2395 		pr_alert("rcu-torture: ->fqs NULL and non-zero fqs_duration, fqs disabled.\n");
2396 		fqs_duration = 0;
2397 	}
2398 	if (cur_ops->init)
2399 		cur_ops->init();
2400 
2401 	if (nreaders >= 0) {
2402 		nrealreaders = nreaders;
2403 	} else {
2404 		nrealreaders = num_online_cpus() - 2 - nreaders;
2405 		if (nrealreaders <= 0)
2406 			nrealreaders = 1;
2407 	}
2408 	rcu_torture_print_module_parms(cur_ops, "Start of test");
2409 
2410 	/* Set up the freelist. */
2411 
2412 	INIT_LIST_HEAD(&rcu_torture_freelist);
2413 	for (i = 0; i < ARRAY_SIZE(rcu_tortures); i++) {
2414 		rcu_tortures[i].rtort_mbtest = 0;
2415 		list_add_tail(&rcu_tortures[i].rtort_free,
2416 			      &rcu_torture_freelist);
2417 	}
2418 
2419 	/* Initialize the statistics so that each run gets its own numbers. */
2420 
2421 	rcu_torture_current = NULL;
2422 	rcu_torture_current_version = 0;
2423 	atomic_set(&n_rcu_torture_alloc, 0);
2424 	atomic_set(&n_rcu_torture_alloc_fail, 0);
2425 	atomic_set(&n_rcu_torture_free, 0);
2426 	atomic_set(&n_rcu_torture_mberror, 0);
2427 	atomic_set(&n_rcu_torture_error, 0);
2428 	n_rcu_torture_barrier_error = 0;
2429 	n_rcu_torture_boost_ktrerror = 0;
2430 	n_rcu_torture_boost_rterror = 0;
2431 	n_rcu_torture_boost_failure = 0;
2432 	n_rcu_torture_boosts = 0;
2433 	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
2434 		atomic_set(&rcu_torture_wcount[i], 0);
2435 	for_each_possible_cpu(cpu) {
2436 		for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
2437 			per_cpu(rcu_torture_count, cpu)[i] = 0;
2438 			per_cpu(rcu_torture_batch, cpu)[i] = 0;
2439 		}
2440 	}
2441 	err_segs_recorded = 0;
2442 	rt_read_nsegs = 0;
2443 
2444 	/* Start up the kthreads. */
2445 
2446 	firsterr = torture_create_kthread(rcu_torture_writer, NULL,
2447 					  writer_task);
2448 	if (firsterr)
2449 		goto unwind;
2450 	if (nfakewriters > 0) {
2451 		fakewriter_tasks = kcalloc(nfakewriters,
2452 					   sizeof(fakewriter_tasks[0]),
2453 					   GFP_KERNEL);
2454 		if (fakewriter_tasks == NULL) {
2455 			VERBOSE_TOROUT_ERRSTRING("out of memory");
2456 			firsterr = -ENOMEM;
2457 			goto unwind;
2458 		}
2459 	}
2460 	for (i = 0; i < nfakewriters; i++) {
2461 		firsterr = torture_create_kthread(rcu_torture_fakewriter,
2462 						  NULL, fakewriter_tasks[i]);
2463 		if (firsterr)
2464 			goto unwind;
2465 	}
2466 	reader_tasks = kcalloc(nrealreaders, sizeof(reader_tasks[0]),
2467 			       GFP_KERNEL);
2468 	if (reader_tasks == NULL) {
2469 		VERBOSE_TOROUT_ERRSTRING("out of memory");
2470 		firsterr = -ENOMEM;
2471 		goto unwind;
2472 	}
2473 	for (i = 0; i < nrealreaders; i++) {
2474 		firsterr = torture_create_kthread(rcu_torture_reader, (void *)i,
2475 						  reader_tasks[i]);
2476 		if (firsterr)
2477 			goto unwind;
2478 	}
2479 	if (stat_interval > 0) {
2480 		firsterr = torture_create_kthread(rcu_torture_stats, NULL,
2481 						  stats_task);
2482 		if (firsterr)
2483 			goto unwind;
2484 	}
2485 	if (test_no_idle_hz && shuffle_interval > 0) {
2486 		firsterr = torture_shuffle_init(shuffle_interval * HZ);
2487 		if (firsterr)
2488 			goto unwind;
2489 	}
2490 	if (stutter < 0)
2491 		stutter = 0;
2492 	if (stutter) {
2493 		int t;
2494 
2495 		t = cur_ops->stall_dur ? cur_ops->stall_dur() : stutter * HZ;
2496 		firsterr = torture_stutter_init(stutter * HZ, t);
2497 		if (firsterr)
2498 			goto unwind;
2499 	}
2500 	if (fqs_duration < 0)
2501 		fqs_duration = 0;
2502 	if (fqs_duration) {
2503 		/* Create the fqs thread */
2504 		firsterr = torture_create_kthread(rcu_torture_fqs, NULL,
2505 						  fqs_task);
2506 		if (firsterr)
2507 			goto unwind;
2508 	}
2509 	if (test_boost_interval < 1)
2510 		test_boost_interval = 1;
2511 	if (test_boost_duration < 2)
2512 		test_boost_duration = 2;
2513 	if (rcu_torture_can_boost()) {
2514 
2515 		boost_starttime = jiffies + test_boost_interval * HZ;
2516 
2517 		firsterr = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "RCU_TORTURE",
2518 					     rcutorture_booster_init,
2519 					     rcutorture_booster_cleanup);
2520 		if (firsterr < 0)
2521 			goto unwind;
2522 		rcutor_hp = firsterr;
2523 	}
2524 	shutdown_jiffies = jiffies + shutdown_secs * HZ;
2525 	firsterr = torture_shutdown_init(shutdown_secs, rcu_torture_cleanup);
2526 	if (firsterr)
2527 		goto unwind;
2528 	firsterr = torture_onoff_init(onoff_holdoff * HZ, onoff_interval,
2529 				      rcutorture_sync);
2530 	if (firsterr)
2531 		goto unwind;
2532 	firsterr = rcu_torture_stall_init();
2533 	if (firsterr)
2534 		goto unwind;
2535 	firsterr = rcu_torture_fwd_prog_init();
2536 	if (firsterr)
2537 		goto unwind;
2538 	firsterr = rcu_torture_barrier_init();
2539 	if (firsterr)
2540 		goto unwind;
2541 	if (object_debug)
2542 		rcu_test_debug_objects();
2543 	torture_init_end();
2544 	return 0;
2545 
2546 unwind:
2547 	torture_init_end();
2548 	rcu_torture_cleanup();
2549 	return firsterr;
2550 }
2551 
2552 module_init(rcu_torture_init);
2553 module_exit(rcu_torture_cleanup);
2554