xref: /openbmc/linux/kernel/rcu/rcutorture.c (revision e149ca29)
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 (!READ_ONCE(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 = READ_ONCE(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 	WRITE_ONCE(rp->rtort_pipe_count, i + 1);
383 	if (rp->rtort_pipe_count >= RCU_TORTURE_PIPE_LEN) {
384 		rp->rtort_mbtest = 0;
385 		return true;
386 	}
387 	return false;
388 }
389 
390 /*
391  * Update all callbacks in the pipe.  Suitable for synchronous grace-period
392  * primitives.
393  */
394 static void
395 rcu_torture_pipe_update(struct rcu_torture *old_rp)
396 {
397 	struct rcu_torture *rp;
398 	struct rcu_torture *rp1;
399 
400 	if (old_rp)
401 		list_add(&old_rp->rtort_free, &rcu_torture_removed);
402 	list_for_each_entry_safe(rp, rp1, &rcu_torture_removed, rtort_free) {
403 		if (rcu_torture_pipe_update_one(rp)) {
404 			list_del(&rp->rtort_free);
405 			rcu_torture_free(rp);
406 		}
407 	}
408 }
409 
410 static void
411 rcu_torture_cb(struct rcu_head *p)
412 {
413 	struct rcu_torture *rp = container_of(p, struct rcu_torture, rtort_rcu);
414 
415 	if (torture_must_stop_irq()) {
416 		/* Test is ending, just drop callbacks on the floor. */
417 		/* The next initialization will pick up the pieces. */
418 		return;
419 	}
420 	if (rcu_torture_pipe_update_one(rp))
421 		rcu_torture_free(rp);
422 	else
423 		cur_ops->deferred_free(rp);
424 }
425 
426 static unsigned long rcu_no_completed(void)
427 {
428 	return 0;
429 }
430 
431 static void rcu_torture_deferred_free(struct rcu_torture *p)
432 {
433 	call_rcu(&p->rtort_rcu, rcu_torture_cb);
434 }
435 
436 static void rcu_sync_torture_init(void)
437 {
438 	INIT_LIST_HEAD(&rcu_torture_removed);
439 }
440 
441 static struct rcu_torture_ops rcu_ops = {
442 	.ttype		= RCU_FLAVOR,
443 	.init		= rcu_sync_torture_init,
444 	.readlock	= rcu_torture_read_lock,
445 	.read_delay	= rcu_read_delay,
446 	.readunlock	= rcu_torture_read_unlock,
447 	.get_gp_seq	= rcu_get_gp_seq,
448 	.gp_diff	= rcu_seq_diff,
449 	.deferred_free	= rcu_torture_deferred_free,
450 	.sync		= synchronize_rcu,
451 	.exp_sync	= synchronize_rcu_expedited,
452 	.get_state	= get_state_synchronize_rcu,
453 	.cond_sync	= cond_synchronize_rcu,
454 	.call		= call_rcu,
455 	.cb_barrier	= rcu_barrier,
456 	.fqs		= rcu_force_quiescent_state,
457 	.stats		= NULL,
458 	.stall_dur	= rcu_jiffies_till_stall_check,
459 	.irq_capable	= 1,
460 	.can_boost	= rcu_can_boost(),
461 	.extendables	= RCUTORTURE_MAX_EXTEND,
462 	.name		= "rcu"
463 };
464 
465 /*
466  * Don't even think about trying any of these in real life!!!
467  * The names includes "busted", and they really means it!
468  * The only purpose of these functions is to provide a buggy RCU
469  * implementation to make sure that rcutorture correctly emits
470  * buggy-RCU error messages.
471  */
472 static void rcu_busted_torture_deferred_free(struct rcu_torture *p)
473 {
474 	/* This is a deliberate bug for testing purposes only! */
475 	rcu_torture_cb(&p->rtort_rcu);
476 }
477 
478 static void synchronize_rcu_busted(void)
479 {
480 	/* This is a deliberate bug for testing purposes only! */
481 }
482 
483 static void
484 call_rcu_busted(struct rcu_head *head, rcu_callback_t func)
485 {
486 	/* This is a deliberate bug for testing purposes only! */
487 	func(head);
488 }
489 
490 static struct rcu_torture_ops rcu_busted_ops = {
491 	.ttype		= INVALID_RCU_FLAVOR,
492 	.init		= rcu_sync_torture_init,
493 	.readlock	= rcu_torture_read_lock,
494 	.read_delay	= rcu_read_delay,  /* just reuse rcu's version. */
495 	.readunlock	= rcu_torture_read_unlock,
496 	.get_gp_seq	= rcu_no_completed,
497 	.deferred_free	= rcu_busted_torture_deferred_free,
498 	.sync		= synchronize_rcu_busted,
499 	.exp_sync	= synchronize_rcu_busted,
500 	.call		= call_rcu_busted,
501 	.cb_barrier	= NULL,
502 	.fqs		= NULL,
503 	.stats		= NULL,
504 	.irq_capable	= 1,
505 	.name		= "busted"
506 };
507 
508 /*
509  * Definitions for srcu torture testing.
510  */
511 
512 DEFINE_STATIC_SRCU(srcu_ctl);
513 static struct srcu_struct srcu_ctld;
514 static struct srcu_struct *srcu_ctlp = &srcu_ctl;
515 
516 static int srcu_torture_read_lock(void) __acquires(srcu_ctlp)
517 {
518 	return srcu_read_lock(srcu_ctlp);
519 }
520 
521 static void
522 srcu_read_delay(struct torture_random_state *rrsp, struct rt_read_seg *rtrsp)
523 {
524 	long delay;
525 	const long uspertick = 1000000 / HZ;
526 	const long longdelay = 10;
527 
528 	/* We want there to be long-running readers, but not all the time. */
529 
530 	delay = torture_random(rrsp) %
531 		(nrealreaders * 2 * longdelay * uspertick);
532 	if (!delay && in_task()) {
533 		schedule_timeout_interruptible(longdelay);
534 		rtrsp->rt_delay_jiffies = longdelay;
535 	} else {
536 		rcu_read_delay(rrsp, rtrsp);
537 	}
538 }
539 
540 static void srcu_torture_read_unlock(int idx) __releases(srcu_ctlp)
541 {
542 	srcu_read_unlock(srcu_ctlp, idx);
543 }
544 
545 static unsigned long srcu_torture_completed(void)
546 {
547 	return srcu_batches_completed(srcu_ctlp);
548 }
549 
550 static void srcu_torture_deferred_free(struct rcu_torture *rp)
551 {
552 	call_srcu(srcu_ctlp, &rp->rtort_rcu, rcu_torture_cb);
553 }
554 
555 static void srcu_torture_synchronize(void)
556 {
557 	synchronize_srcu(srcu_ctlp);
558 }
559 
560 static void srcu_torture_call(struct rcu_head *head,
561 			      rcu_callback_t func)
562 {
563 	call_srcu(srcu_ctlp, head, func);
564 }
565 
566 static void srcu_torture_barrier(void)
567 {
568 	srcu_barrier(srcu_ctlp);
569 }
570 
571 static void srcu_torture_stats(void)
572 {
573 	srcu_torture_stats_print(srcu_ctlp, torture_type, TORTURE_FLAG);
574 }
575 
576 static void srcu_torture_synchronize_expedited(void)
577 {
578 	synchronize_srcu_expedited(srcu_ctlp);
579 }
580 
581 static struct rcu_torture_ops srcu_ops = {
582 	.ttype		= SRCU_FLAVOR,
583 	.init		= rcu_sync_torture_init,
584 	.readlock	= srcu_torture_read_lock,
585 	.read_delay	= srcu_read_delay,
586 	.readunlock	= srcu_torture_read_unlock,
587 	.get_gp_seq	= srcu_torture_completed,
588 	.deferred_free	= srcu_torture_deferred_free,
589 	.sync		= srcu_torture_synchronize,
590 	.exp_sync	= srcu_torture_synchronize_expedited,
591 	.call		= srcu_torture_call,
592 	.cb_barrier	= srcu_torture_barrier,
593 	.stats		= srcu_torture_stats,
594 	.irq_capable	= 1,
595 	.name		= "srcu"
596 };
597 
598 static void srcu_torture_init(void)
599 {
600 	rcu_sync_torture_init();
601 	WARN_ON(init_srcu_struct(&srcu_ctld));
602 	srcu_ctlp = &srcu_ctld;
603 }
604 
605 static void srcu_torture_cleanup(void)
606 {
607 	cleanup_srcu_struct(&srcu_ctld);
608 	srcu_ctlp = &srcu_ctl; /* In case of a later rcutorture run. */
609 }
610 
611 /* As above, but dynamically allocated. */
612 static struct rcu_torture_ops srcud_ops = {
613 	.ttype		= SRCU_FLAVOR,
614 	.init		= srcu_torture_init,
615 	.cleanup	= srcu_torture_cleanup,
616 	.readlock	= srcu_torture_read_lock,
617 	.read_delay	= srcu_read_delay,
618 	.readunlock	= srcu_torture_read_unlock,
619 	.get_gp_seq	= srcu_torture_completed,
620 	.deferred_free	= srcu_torture_deferred_free,
621 	.sync		= srcu_torture_synchronize,
622 	.exp_sync	= srcu_torture_synchronize_expedited,
623 	.call		= srcu_torture_call,
624 	.cb_barrier	= srcu_torture_barrier,
625 	.stats		= srcu_torture_stats,
626 	.irq_capable	= 1,
627 	.name		= "srcud"
628 };
629 
630 /* As above, but broken due to inappropriate reader extension. */
631 static struct rcu_torture_ops busted_srcud_ops = {
632 	.ttype		= SRCU_FLAVOR,
633 	.init		= srcu_torture_init,
634 	.cleanup	= srcu_torture_cleanup,
635 	.readlock	= srcu_torture_read_lock,
636 	.read_delay	= rcu_read_delay,
637 	.readunlock	= srcu_torture_read_unlock,
638 	.get_gp_seq	= srcu_torture_completed,
639 	.deferred_free	= srcu_torture_deferred_free,
640 	.sync		= srcu_torture_synchronize,
641 	.exp_sync	= srcu_torture_synchronize_expedited,
642 	.call		= srcu_torture_call,
643 	.cb_barrier	= srcu_torture_barrier,
644 	.stats		= srcu_torture_stats,
645 	.irq_capable	= 1,
646 	.extendables	= RCUTORTURE_MAX_EXTEND,
647 	.name		= "busted_srcud"
648 };
649 
650 /*
651  * Definitions for RCU-tasks torture testing.
652  */
653 
654 static int tasks_torture_read_lock(void)
655 {
656 	return 0;
657 }
658 
659 static void tasks_torture_read_unlock(int idx)
660 {
661 }
662 
663 static void rcu_tasks_torture_deferred_free(struct rcu_torture *p)
664 {
665 	call_rcu_tasks(&p->rtort_rcu, rcu_torture_cb);
666 }
667 
668 static struct rcu_torture_ops tasks_ops = {
669 	.ttype		= RCU_TASKS_FLAVOR,
670 	.init		= rcu_sync_torture_init,
671 	.readlock	= tasks_torture_read_lock,
672 	.read_delay	= rcu_read_delay,  /* just reuse rcu's version. */
673 	.readunlock	= tasks_torture_read_unlock,
674 	.get_gp_seq	= rcu_no_completed,
675 	.deferred_free	= rcu_tasks_torture_deferred_free,
676 	.sync		= synchronize_rcu_tasks,
677 	.exp_sync	= synchronize_rcu_tasks,
678 	.call		= call_rcu_tasks,
679 	.cb_barrier	= rcu_barrier_tasks,
680 	.fqs		= NULL,
681 	.stats		= NULL,
682 	.irq_capable	= 1,
683 	.slow_gps	= 1,
684 	.name		= "tasks"
685 };
686 
687 /*
688  * Definitions for trivial CONFIG_PREEMPT=n-only torture testing.
689  * This implementation does not necessarily work well with CPU hotplug.
690  */
691 
692 static void synchronize_rcu_trivial(void)
693 {
694 	int cpu;
695 
696 	for_each_online_cpu(cpu) {
697 		rcutorture_sched_setaffinity(current->pid, cpumask_of(cpu));
698 		WARN_ON_ONCE(raw_smp_processor_id() != cpu);
699 	}
700 }
701 
702 static int rcu_torture_read_lock_trivial(void) __acquires(RCU)
703 {
704 	preempt_disable();
705 	return 0;
706 }
707 
708 static void rcu_torture_read_unlock_trivial(int idx) __releases(RCU)
709 {
710 	preempt_enable();
711 }
712 
713 static struct rcu_torture_ops trivial_ops = {
714 	.ttype		= RCU_TRIVIAL_FLAVOR,
715 	.init		= rcu_sync_torture_init,
716 	.readlock	= rcu_torture_read_lock_trivial,
717 	.read_delay	= rcu_read_delay,  /* just reuse rcu's version. */
718 	.readunlock	= rcu_torture_read_unlock_trivial,
719 	.get_gp_seq	= rcu_no_completed,
720 	.sync		= synchronize_rcu_trivial,
721 	.exp_sync	= synchronize_rcu_trivial,
722 	.fqs		= NULL,
723 	.stats		= NULL,
724 	.irq_capable	= 1,
725 	.name		= "trivial"
726 };
727 
728 static unsigned long rcutorture_seq_diff(unsigned long new, unsigned long old)
729 {
730 	if (!cur_ops->gp_diff)
731 		return new - old;
732 	return cur_ops->gp_diff(new, old);
733 }
734 
735 static bool __maybe_unused torturing_tasks(void)
736 {
737 	return cur_ops == &tasks_ops;
738 }
739 
740 /*
741  * RCU torture priority-boost testing.  Runs one real-time thread per
742  * CPU for moderate bursts, repeatedly registering RCU callbacks and
743  * spinning waiting for them to be invoked.  If a given callback takes
744  * too long to be invoked, we assume that priority inversion has occurred.
745  */
746 
747 struct rcu_boost_inflight {
748 	struct rcu_head rcu;
749 	int inflight;
750 };
751 
752 static void rcu_torture_boost_cb(struct rcu_head *head)
753 {
754 	struct rcu_boost_inflight *rbip =
755 		container_of(head, struct rcu_boost_inflight, rcu);
756 
757 	/* Ensure RCU-core accesses precede clearing ->inflight */
758 	smp_store_release(&rbip->inflight, 0);
759 }
760 
761 static int old_rt_runtime = -1;
762 
763 static void rcu_torture_disable_rt_throttle(void)
764 {
765 	/*
766 	 * Disable RT throttling so that rcutorture's boost threads don't get
767 	 * throttled. Only possible if rcutorture is built-in otherwise the
768 	 * user should manually do this by setting the sched_rt_period_us and
769 	 * sched_rt_runtime sysctls.
770 	 */
771 	if (!IS_BUILTIN(CONFIG_RCU_TORTURE_TEST) || old_rt_runtime != -1)
772 		return;
773 
774 	old_rt_runtime = sysctl_sched_rt_runtime;
775 	sysctl_sched_rt_runtime = -1;
776 }
777 
778 static void rcu_torture_enable_rt_throttle(void)
779 {
780 	if (!IS_BUILTIN(CONFIG_RCU_TORTURE_TEST) || old_rt_runtime == -1)
781 		return;
782 
783 	sysctl_sched_rt_runtime = old_rt_runtime;
784 	old_rt_runtime = -1;
785 }
786 
787 static bool rcu_torture_boost_failed(unsigned long start, unsigned long end)
788 {
789 	if (end - start > test_boost_duration * HZ - HZ / 2) {
790 		VERBOSE_TOROUT_STRING("rcu_torture_boost boosting failed");
791 		n_rcu_torture_boost_failure++;
792 
793 		return true; /* failed */
794 	}
795 
796 	return false; /* passed */
797 }
798 
799 static int rcu_torture_boost(void *arg)
800 {
801 	unsigned long call_rcu_time;
802 	unsigned long endtime;
803 	unsigned long oldstarttime;
804 	struct rcu_boost_inflight rbi = { .inflight = 0 };
805 	struct sched_param sp;
806 
807 	VERBOSE_TOROUT_STRING("rcu_torture_boost started");
808 
809 	/* Set real-time priority. */
810 	sp.sched_priority = 1;
811 	if (sched_setscheduler(current, SCHED_FIFO, &sp) < 0) {
812 		VERBOSE_TOROUT_STRING("rcu_torture_boost RT prio failed!");
813 		n_rcu_torture_boost_rterror++;
814 	}
815 
816 	init_rcu_head_on_stack(&rbi.rcu);
817 	/* Each pass through the following loop does one boost-test cycle. */
818 	do {
819 		/* Track if the test failed already in this test interval? */
820 		bool failed = false;
821 
822 		/* Increment n_rcu_torture_boosts once per boost-test */
823 		while (!kthread_should_stop()) {
824 			if (mutex_trylock(&boost_mutex)) {
825 				n_rcu_torture_boosts++;
826 				mutex_unlock(&boost_mutex);
827 				break;
828 			}
829 			schedule_timeout_uninterruptible(1);
830 		}
831 		if (kthread_should_stop())
832 			goto checkwait;
833 
834 		/* Wait for the next test interval. */
835 		oldstarttime = boost_starttime;
836 		while (ULONG_CMP_LT(jiffies, oldstarttime)) {
837 			schedule_timeout_interruptible(oldstarttime - jiffies);
838 			stutter_wait("rcu_torture_boost");
839 			if (torture_must_stop())
840 				goto checkwait;
841 		}
842 
843 		/* Do one boost-test interval. */
844 		endtime = oldstarttime + test_boost_duration * HZ;
845 		call_rcu_time = jiffies;
846 		while (ULONG_CMP_LT(jiffies, endtime)) {
847 			/* If we don't have a callback in flight, post one. */
848 			if (!smp_load_acquire(&rbi.inflight)) {
849 				/* RCU core before ->inflight = 1. */
850 				smp_store_release(&rbi.inflight, 1);
851 				call_rcu(&rbi.rcu, rcu_torture_boost_cb);
852 				/* Check if the boost test failed */
853 				failed = failed ||
854 					 rcu_torture_boost_failed(call_rcu_time,
855 								 jiffies);
856 				call_rcu_time = jiffies;
857 			}
858 			stutter_wait("rcu_torture_boost");
859 			if (torture_must_stop())
860 				goto checkwait;
861 		}
862 
863 		/*
864 		 * If boost never happened, then inflight will always be 1, in
865 		 * this case the boost check would never happen in the above
866 		 * loop so do another one here.
867 		 */
868 		if (!failed && smp_load_acquire(&rbi.inflight))
869 			rcu_torture_boost_failed(call_rcu_time, jiffies);
870 
871 		/*
872 		 * Set the start time of the next test interval.
873 		 * Yes, this is vulnerable to long delays, but such
874 		 * delays simply cause a false negative for the next
875 		 * interval.  Besides, we are running at RT priority,
876 		 * so delays should be relatively rare.
877 		 */
878 		while (oldstarttime == boost_starttime &&
879 		       !kthread_should_stop()) {
880 			if (mutex_trylock(&boost_mutex)) {
881 				boost_starttime = jiffies +
882 						  test_boost_interval * HZ;
883 				mutex_unlock(&boost_mutex);
884 				break;
885 			}
886 			schedule_timeout_uninterruptible(1);
887 		}
888 
889 		/* Go do the stutter. */
890 checkwait:	stutter_wait("rcu_torture_boost");
891 	} while (!torture_must_stop());
892 
893 	/* Clean up and exit. */
894 	while (!kthread_should_stop() || smp_load_acquire(&rbi.inflight)) {
895 		torture_shutdown_absorb("rcu_torture_boost");
896 		schedule_timeout_uninterruptible(1);
897 	}
898 	destroy_rcu_head_on_stack(&rbi.rcu);
899 	torture_kthread_stopping("rcu_torture_boost");
900 	return 0;
901 }
902 
903 /*
904  * RCU torture force-quiescent-state kthread.  Repeatedly induces
905  * bursts of calls to force_quiescent_state(), increasing the probability
906  * of occurrence of some important types of race conditions.
907  */
908 static int
909 rcu_torture_fqs(void *arg)
910 {
911 	unsigned long fqs_resume_time;
912 	int fqs_burst_remaining;
913 
914 	VERBOSE_TOROUT_STRING("rcu_torture_fqs task started");
915 	do {
916 		fqs_resume_time = jiffies + fqs_stutter * HZ;
917 		while (ULONG_CMP_LT(jiffies, fqs_resume_time) &&
918 		       !kthread_should_stop()) {
919 			schedule_timeout_interruptible(1);
920 		}
921 		fqs_burst_remaining = fqs_duration;
922 		while (fqs_burst_remaining > 0 &&
923 		       !kthread_should_stop()) {
924 			cur_ops->fqs();
925 			udelay(fqs_holdoff);
926 			fqs_burst_remaining -= fqs_holdoff;
927 		}
928 		stutter_wait("rcu_torture_fqs");
929 	} while (!torture_must_stop());
930 	torture_kthread_stopping("rcu_torture_fqs");
931 	return 0;
932 }
933 
934 /*
935  * RCU torture writer kthread.  Repeatedly substitutes a new structure
936  * for that pointed to by rcu_torture_current, freeing the old structure
937  * after a series of grace periods (the "pipeline").
938  */
939 static int
940 rcu_torture_writer(void *arg)
941 {
942 	bool can_expedite = !rcu_gp_is_expedited() && !rcu_gp_is_normal();
943 	int expediting = 0;
944 	unsigned long gp_snap;
945 	bool gp_cond1 = gp_cond, gp_exp1 = gp_exp, gp_normal1 = gp_normal;
946 	bool gp_sync1 = gp_sync;
947 	int i;
948 	struct rcu_torture *rp;
949 	struct rcu_torture *old_rp;
950 	static DEFINE_TORTURE_RANDOM(rand);
951 	int synctype[] = { RTWS_DEF_FREE, RTWS_EXP_SYNC,
952 			   RTWS_COND_GET, RTWS_SYNC };
953 	int nsynctypes = 0;
954 
955 	VERBOSE_TOROUT_STRING("rcu_torture_writer task started");
956 	if (!can_expedite)
957 		pr_alert("%s" TORTURE_FLAG
958 			 " GP expediting controlled from boot/sysfs for %s.\n",
959 			 torture_type, cur_ops->name);
960 
961 	/* Initialize synctype[] array.  If none set, take default. */
962 	if (!gp_cond1 && !gp_exp1 && !gp_normal1 && !gp_sync1)
963 		gp_cond1 = gp_exp1 = gp_normal1 = gp_sync1 = true;
964 	if (gp_cond1 && cur_ops->get_state && cur_ops->cond_sync) {
965 		synctype[nsynctypes++] = RTWS_COND_GET;
966 		pr_info("%s: Testing conditional GPs.\n", __func__);
967 	} else if (gp_cond && (!cur_ops->get_state || !cur_ops->cond_sync)) {
968 		pr_alert("%s: gp_cond without primitives.\n", __func__);
969 	}
970 	if (gp_exp1 && cur_ops->exp_sync) {
971 		synctype[nsynctypes++] = RTWS_EXP_SYNC;
972 		pr_info("%s: Testing expedited GPs.\n", __func__);
973 	} else if (gp_exp && !cur_ops->exp_sync) {
974 		pr_alert("%s: gp_exp without primitives.\n", __func__);
975 	}
976 	if (gp_normal1 && cur_ops->deferred_free) {
977 		synctype[nsynctypes++] = RTWS_DEF_FREE;
978 		pr_info("%s: Testing asynchronous GPs.\n", __func__);
979 	} else if (gp_normal && !cur_ops->deferred_free) {
980 		pr_alert("%s: gp_normal without primitives.\n", __func__);
981 	}
982 	if (gp_sync1 && cur_ops->sync) {
983 		synctype[nsynctypes++] = RTWS_SYNC;
984 		pr_info("%s: Testing normal GPs.\n", __func__);
985 	} else if (gp_sync && !cur_ops->sync) {
986 		pr_alert("%s: gp_sync without primitives.\n", __func__);
987 	}
988 	if (WARN_ONCE(nsynctypes == 0,
989 		      "rcu_torture_writer: No update-side primitives.\n")) {
990 		/*
991 		 * No updates primitives, so don't try updating.
992 		 * The resulting test won't be testing much, hence the
993 		 * above WARN_ONCE().
994 		 */
995 		rcu_torture_writer_state = RTWS_STOPPING;
996 		torture_kthread_stopping("rcu_torture_writer");
997 	}
998 
999 	do {
1000 		rcu_torture_writer_state = RTWS_FIXED_DELAY;
1001 		schedule_timeout_uninterruptible(1);
1002 		rp = rcu_torture_alloc();
1003 		if (rp == NULL)
1004 			continue;
1005 		rp->rtort_pipe_count = 0;
1006 		rcu_torture_writer_state = RTWS_DELAY;
1007 		udelay(torture_random(&rand) & 0x3ff);
1008 		rcu_torture_writer_state = RTWS_REPLACE;
1009 		old_rp = rcu_dereference_check(rcu_torture_current,
1010 					       current == writer_task);
1011 		rp->rtort_mbtest = 1;
1012 		rcu_assign_pointer(rcu_torture_current, rp);
1013 		smp_wmb(); /* Mods to old_rp must follow rcu_assign_pointer() */
1014 		if (old_rp) {
1015 			i = old_rp->rtort_pipe_count;
1016 			if (i > RCU_TORTURE_PIPE_LEN)
1017 				i = RCU_TORTURE_PIPE_LEN;
1018 			atomic_inc(&rcu_torture_wcount[i]);
1019 			WRITE_ONCE(old_rp->rtort_pipe_count,
1020 				   old_rp->rtort_pipe_count + 1);
1021 			switch (synctype[torture_random(&rand) % nsynctypes]) {
1022 			case RTWS_DEF_FREE:
1023 				rcu_torture_writer_state = RTWS_DEF_FREE;
1024 				cur_ops->deferred_free(old_rp);
1025 				break;
1026 			case RTWS_EXP_SYNC:
1027 				rcu_torture_writer_state = RTWS_EXP_SYNC;
1028 				cur_ops->exp_sync();
1029 				rcu_torture_pipe_update(old_rp);
1030 				break;
1031 			case RTWS_COND_GET:
1032 				rcu_torture_writer_state = RTWS_COND_GET;
1033 				gp_snap = cur_ops->get_state();
1034 				i = torture_random(&rand) % 16;
1035 				if (i != 0)
1036 					schedule_timeout_interruptible(i);
1037 				udelay(torture_random(&rand) % 1000);
1038 				rcu_torture_writer_state = RTWS_COND_SYNC;
1039 				cur_ops->cond_sync(gp_snap);
1040 				rcu_torture_pipe_update(old_rp);
1041 				break;
1042 			case RTWS_SYNC:
1043 				rcu_torture_writer_state = RTWS_SYNC;
1044 				cur_ops->sync();
1045 				rcu_torture_pipe_update(old_rp);
1046 				break;
1047 			default:
1048 				WARN_ON_ONCE(1);
1049 				break;
1050 			}
1051 		}
1052 		WRITE_ONCE(rcu_torture_current_version,
1053 			   rcu_torture_current_version + 1);
1054 		/* Cycle through nesting levels of rcu_expedite_gp() calls. */
1055 		if (can_expedite &&
1056 		    !(torture_random(&rand) & 0xff & (!!expediting - 1))) {
1057 			WARN_ON_ONCE(expediting == 0 && rcu_gp_is_expedited());
1058 			if (expediting >= 0)
1059 				rcu_expedite_gp();
1060 			else
1061 				rcu_unexpedite_gp();
1062 			if (++expediting > 3)
1063 				expediting = -expediting;
1064 		} else if (!can_expedite) { /* Disabled during boot, recheck. */
1065 			can_expedite = !rcu_gp_is_expedited() &&
1066 				       !rcu_gp_is_normal();
1067 		}
1068 		rcu_torture_writer_state = RTWS_STUTTER;
1069 		if (stutter_wait("rcu_torture_writer") &&
1070 		    !READ_ONCE(rcu_fwd_cb_nodelay) &&
1071 		    !cur_ops->slow_gps &&
1072 		    !torture_must_stop() &&
1073 		    rcu_inkernel_boot_has_ended())
1074 			for (i = 0; i < ARRAY_SIZE(rcu_tortures); i++)
1075 				if (list_empty(&rcu_tortures[i].rtort_free) &&
1076 				    rcu_access_pointer(rcu_torture_current) !=
1077 				    &rcu_tortures[i]) {
1078 					rcu_ftrace_dump(DUMP_ALL);
1079 					WARN(1, "%s: rtort_pipe_count: %d\n", __func__, rcu_tortures[i].rtort_pipe_count);
1080 				}
1081 	} while (!torture_must_stop());
1082 	/* Reset expediting back to unexpedited. */
1083 	if (expediting > 0)
1084 		expediting = -expediting;
1085 	while (can_expedite && expediting++ < 0)
1086 		rcu_unexpedite_gp();
1087 	WARN_ON_ONCE(can_expedite && rcu_gp_is_expedited());
1088 	if (!can_expedite)
1089 		pr_alert("%s" TORTURE_FLAG
1090 			 " Dynamic grace-period expediting was disabled.\n",
1091 			 torture_type);
1092 	rcu_torture_writer_state = RTWS_STOPPING;
1093 	torture_kthread_stopping("rcu_torture_writer");
1094 	return 0;
1095 }
1096 
1097 /*
1098  * RCU torture fake writer kthread.  Repeatedly calls sync, with a random
1099  * delay between calls.
1100  */
1101 static int
1102 rcu_torture_fakewriter(void *arg)
1103 {
1104 	DEFINE_TORTURE_RANDOM(rand);
1105 
1106 	VERBOSE_TOROUT_STRING("rcu_torture_fakewriter task started");
1107 	set_user_nice(current, MAX_NICE);
1108 
1109 	do {
1110 		schedule_timeout_uninterruptible(1 + torture_random(&rand)%10);
1111 		udelay(torture_random(&rand) & 0x3ff);
1112 		if (cur_ops->cb_barrier != NULL &&
1113 		    torture_random(&rand) % (nfakewriters * 8) == 0) {
1114 			cur_ops->cb_barrier();
1115 		} else if (gp_normal == gp_exp) {
1116 			if (cur_ops->sync && torture_random(&rand) & 0x80)
1117 				cur_ops->sync();
1118 			else if (cur_ops->exp_sync)
1119 				cur_ops->exp_sync();
1120 		} else if (gp_normal && cur_ops->sync) {
1121 			cur_ops->sync();
1122 		} else if (cur_ops->exp_sync) {
1123 			cur_ops->exp_sync();
1124 		}
1125 		stutter_wait("rcu_torture_fakewriter");
1126 	} while (!torture_must_stop());
1127 
1128 	torture_kthread_stopping("rcu_torture_fakewriter");
1129 	return 0;
1130 }
1131 
1132 static void rcu_torture_timer_cb(struct rcu_head *rhp)
1133 {
1134 	kfree(rhp);
1135 }
1136 
1137 /*
1138  * Do one extension of an RCU read-side critical section using the
1139  * current reader state in readstate (set to zero for initial entry
1140  * to extended critical section), set the new state as specified by
1141  * newstate (set to zero for final exit from extended critical section),
1142  * and random-number-generator state in trsp.  If this is neither the
1143  * beginning or end of the critical section and if there was actually a
1144  * change, do a ->read_delay().
1145  */
1146 static void rcutorture_one_extend(int *readstate, int newstate,
1147 				  struct torture_random_state *trsp,
1148 				  struct rt_read_seg *rtrsp)
1149 {
1150 	int idxnew = -1;
1151 	int idxold = *readstate;
1152 	int statesnew = ~*readstate & newstate;
1153 	int statesold = *readstate & ~newstate;
1154 
1155 	WARN_ON_ONCE(idxold < 0);
1156 	WARN_ON_ONCE((idxold >> RCUTORTURE_RDR_SHIFT) > 1);
1157 	rtrsp->rt_readstate = newstate;
1158 
1159 	/* First, put new protection in place to avoid critical-section gap. */
1160 	if (statesnew & RCUTORTURE_RDR_BH)
1161 		local_bh_disable();
1162 	if (statesnew & RCUTORTURE_RDR_IRQ)
1163 		local_irq_disable();
1164 	if (statesnew & RCUTORTURE_RDR_PREEMPT)
1165 		preempt_disable();
1166 	if (statesnew & RCUTORTURE_RDR_RBH)
1167 		rcu_read_lock_bh();
1168 	if (statesnew & RCUTORTURE_RDR_SCHED)
1169 		rcu_read_lock_sched();
1170 	if (statesnew & RCUTORTURE_RDR_RCU)
1171 		idxnew = cur_ops->readlock() << RCUTORTURE_RDR_SHIFT;
1172 
1173 	/* Next, remove old protection, irq first due to bh conflict. */
1174 	if (statesold & RCUTORTURE_RDR_IRQ)
1175 		local_irq_enable();
1176 	if (statesold & RCUTORTURE_RDR_BH)
1177 		local_bh_enable();
1178 	if (statesold & RCUTORTURE_RDR_PREEMPT)
1179 		preempt_enable();
1180 	if (statesold & RCUTORTURE_RDR_RBH)
1181 		rcu_read_unlock_bh();
1182 	if (statesold & RCUTORTURE_RDR_SCHED)
1183 		rcu_read_unlock_sched();
1184 	if (statesold & RCUTORTURE_RDR_RCU)
1185 		cur_ops->readunlock(idxold >> RCUTORTURE_RDR_SHIFT);
1186 
1187 	/* Delay if neither beginning nor end and there was a change. */
1188 	if ((statesnew || statesold) && *readstate && newstate)
1189 		cur_ops->read_delay(trsp, rtrsp);
1190 
1191 	/* Update the reader state. */
1192 	if (idxnew == -1)
1193 		idxnew = idxold & ~RCUTORTURE_RDR_MASK;
1194 	WARN_ON_ONCE(idxnew < 0);
1195 	WARN_ON_ONCE((idxnew >> RCUTORTURE_RDR_SHIFT) > 1);
1196 	*readstate = idxnew | newstate;
1197 	WARN_ON_ONCE((*readstate >> RCUTORTURE_RDR_SHIFT) < 0);
1198 	WARN_ON_ONCE((*readstate >> RCUTORTURE_RDR_SHIFT) > 1);
1199 }
1200 
1201 /* Return the biggest extendables mask given current RCU and boot parameters. */
1202 static int rcutorture_extend_mask_max(void)
1203 {
1204 	int mask;
1205 
1206 	WARN_ON_ONCE(extendables & ~RCUTORTURE_MAX_EXTEND);
1207 	mask = extendables & RCUTORTURE_MAX_EXTEND & cur_ops->extendables;
1208 	mask = mask | RCUTORTURE_RDR_RCU;
1209 	return mask;
1210 }
1211 
1212 /* Return a random protection state mask, but with at least one bit set. */
1213 static int
1214 rcutorture_extend_mask(int oldmask, struct torture_random_state *trsp)
1215 {
1216 	int mask = rcutorture_extend_mask_max();
1217 	unsigned long randmask1 = torture_random(trsp) >> 8;
1218 	unsigned long randmask2 = randmask1 >> 3;
1219 
1220 	WARN_ON_ONCE(mask >> RCUTORTURE_RDR_SHIFT);
1221 	/* Mostly only one bit (need preemption!), sometimes lots of bits. */
1222 	if (!(randmask1 & 0x7))
1223 		mask = mask & randmask2;
1224 	else
1225 		mask = mask & (1 << (randmask2 % RCUTORTURE_RDR_NBITS));
1226 	/* Can't enable bh w/irq disabled. */
1227 	if ((mask & RCUTORTURE_RDR_IRQ) &&
1228 	    ((!(mask & RCUTORTURE_RDR_BH) && (oldmask & RCUTORTURE_RDR_BH)) ||
1229 	     (!(mask & RCUTORTURE_RDR_RBH) && (oldmask & RCUTORTURE_RDR_RBH))))
1230 		mask |= RCUTORTURE_RDR_BH | RCUTORTURE_RDR_RBH;
1231 	return mask ?: RCUTORTURE_RDR_RCU;
1232 }
1233 
1234 /*
1235  * Do a randomly selected number of extensions of an existing RCU read-side
1236  * critical section.
1237  */
1238 static struct rt_read_seg *
1239 rcutorture_loop_extend(int *readstate, struct torture_random_state *trsp,
1240 		       struct rt_read_seg *rtrsp)
1241 {
1242 	int i;
1243 	int j;
1244 	int mask = rcutorture_extend_mask_max();
1245 
1246 	WARN_ON_ONCE(!*readstate); /* -Existing- RCU read-side critsect! */
1247 	if (!((mask - 1) & mask))
1248 		return rtrsp;  /* Current RCU reader not extendable. */
1249 	/* Bias towards larger numbers of loops. */
1250 	i = (torture_random(trsp) >> 3);
1251 	i = ((i | (i >> 3)) & RCUTORTURE_RDR_MAX_LOOPS) + 1;
1252 	for (j = 0; j < i; j++) {
1253 		mask = rcutorture_extend_mask(*readstate, trsp);
1254 		rcutorture_one_extend(readstate, mask, trsp, &rtrsp[j]);
1255 	}
1256 	return &rtrsp[j];
1257 }
1258 
1259 /*
1260  * Do one read-side critical section, returning false if there was
1261  * no data to read.  Can be invoked both from process context and
1262  * from a timer handler.
1263  */
1264 static bool rcu_torture_one_read(struct torture_random_state *trsp)
1265 {
1266 	int i;
1267 	unsigned long started;
1268 	unsigned long completed;
1269 	int newstate;
1270 	struct rcu_torture *p;
1271 	int pipe_count;
1272 	int readstate = 0;
1273 	struct rt_read_seg rtseg[RCUTORTURE_RDR_MAX_SEGS] = { { 0 } };
1274 	struct rt_read_seg *rtrsp = &rtseg[0];
1275 	struct rt_read_seg *rtrsp1;
1276 	unsigned long long ts;
1277 
1278 	newstate = rcutorture_extend_mask(readstate, trsp);
1279 	rcutorture_one_extend(&readstate, newstate, trsp, rtrsp++);
1280 	started = cur_ops->get_gp_seq();
1281 	ts = rcu_trace_clock_local();
1282 	p = rcu_dereference_check(rcu_torture_current,
1283 				  rcu_read_lock_bh_held() ||
1284 				  rcu_read_lock_sched_held() ||
1285 				  srcu_read_lock_held(srcu_ctlp) ||
1286 				  torturing_tasks());
1287 	if (p == NULL) {
1288 		/* Wait for rcu_torture_writer to get underway */
1289 		rcutorture_one_extend(&readstate, 0, trsp, rtrsp);
1290 		return false;
1291 	}
1292 	if (p->rtort_mbtest == 0)
1293 		atomic_inc(&n_rcu_torture_mberror);
1294 	rtrsp = rcutorture_loop_extend(&readstate, trsp, rtrsp);
1295 	preempt_disable();
1296 	pipe_count = READ_ONCE(p->rtort_pipe_count);
1297 	if (pipe_count > RCU_TORTURE_PIPE_LEN) {
1298 		/* Should not happen, but... */
1299 		pipe_count = RCU_TORTURE_PIPE_LEN;
1300 	}
1301 	completed = cur_ops->get_gp_seq();
1302 	if (pipe_count > 1) {
1303 		do_trace_rcu_torture_read(cur_ops->name, &p->rtort_rcu,
1304 					  ts, started, completed);
1305 		rcu_ftrace_dump(DUMP_ALL);
1306 	}
1307 	__this_cpu_inc(rcu_torture_count[pipe_count]);
1308 	completed = rcutorture_seq_diff(completed, started);
1309 	if (completed > RCU_TORTURE_PIPE_LEN) {
1310 		/* Should not happen, but... */
1311 		completed = RCU_TORTURE_PIPE_LEN;
1312 	}
1313 	__this_cpu_inc(rcu_torture_batch[completed]);
1314 	preempt_enable();
1315 	rcutorture_one_extend(&readstate, 0, trsp, rtrsp);
1316 	WARN_ON_ONCE(readstate & RCUTORTURE_RDR_MASK);
1317 
1318 	/* If error or close call, record the sequence of reader protections. */
1319 	if ((pipe_count > 1 || completed > 1) && !xchg(&err_segs_recorded, 1)) {
1320 		i = 0;
1321 		for (rtrsp1 = &rtseg[0]; rtrsp1 < rtrsp; rtrsp1++)
1322 			err_segs[i++] = *rtrsp1;
1323 		rt_read_nsegs = i;
1324 	}
1325 
1326 	return true;
1327 }
1328 
1329 static DEFINE_TORTURE_RANDOM_PERCPU(rcu_torture_timer_rand);
1330 
1331 /*
1332  * RCU torture reader from timer handler.  Dereferences rcu_torture_current,
1333  * incrementing the corresponding element of the pipeline array.  The
1334  * counter in the element should never be greater than 1, otherwise, the
1335  * RCU implementation is broken.
1336  */
1337 static void rcu_torture_timer(struct timer_list *unused)
1338 {
1339 	atomic_long_inc(&n_rcu_torture_timers);
1340 	(void)rcu_torture_one_read(this_cpu_ptr(&rcu_torture_timer_rand));
1341 
1342 	/* Test call_rcu() invocation from interrupt handler. */
1343 	if (cur_ops->call) {
1344 		struct rcu_head *rhp = kmalloc(sizeof(*rhp), GFP_NOWAIT);
1345 
1346 		if (rhp)
1347 			cur_ops->call(rhp, rcu_torture_timer_cb);
1348 	}
1349 }
1350 
1351 /*
1352  * RCU torture reader kthread.  Repeatedly dereferences rcu_torture_current,
1353  * incrementing the corresponding element of the pipeline array.  The
1354  * counter in the element should never be greater than 1, otherwise, the
1355  * RCU implementation is broken.
1356  */
1357 static int
1358 rcu_torture_reader(void *arg)
1359 {
1360 	unsigned long lastsleep = jiffies;
1361 	long myid = (long)arg;
1362 	int mynumonline = myid;
1363 	DEFINE_TORTURE_RANDOM(rand);
1364 	struct timer_list t;
1365 
1366 	VERBOSE_TOROUT_STRING("rcu_torture_reader task started");
1367 	set_user_nice(current, MAX_NICE);
1368 	if (irqreader && cur_ops->irq_capable)
1369 		timer_setup_on_stack(&t, rcu_torture_timer, 0);
1370 	tick_dep_set_task(current, TICK_DEP_BIT_RCU);
1371 	do {
1372 		if (irqreader && cur_ops->irq_capable) {
1373 			if (!timer_pending(&t))
1374 				mod_timer(&t, jiffies + 1);
1375 		}
1376 		if (!rcu_torture_one_read(&rand) && !torture_must_stop())
1377 			schedule_timeout_interruptible(HZ);
1378 		if (time_after(jiffies, lastsleep) && !torture_must_stop()) {
1379 			schedule_timeout_interruptible(1);
1380 			lastsleep = jiffies + 10;
1381 		}
1382 		while (num_online_cpus() < mynumonline && !torture_must_stop())
1383 			schedule_timeout_interruptible(HZ / 5);
1384 		stutter_wait("rcu_torture_reader");
1385 	} while (!torture_must_stop());
1386 	if (irqreader && cur_ops->irq_capable) {
1387 		del_timer_sync(&t);
1388 		destroy_timer_on_stack(&t);
1389 	}
1390 	tick_dep_clear_task(current, TICK_DEP_BIT_RCU);
1391 	torture_kthread_stopping("rcu_torture_reader");
1392 	return 0;
1393 }
1394 
1395 /*
1396  * Print torture statistics.  Caller must ensure that there is only
1397  * one call to this function at a given time!!!  This is normally
1398  * accomplished by relying on the module system to only have one copy
1399  * of the module loaded, and then by giving the rcu_torture_stats
1400  * kthread full control (or the init/cleanup functions when rcu_torture_stats
1401  * thread is not running).
1402  */
1403 static void
1404 rcu_torture_stats_print(void)
1405 {
1406 	int cpu;
1407 	int i;
1408 	long pipesummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
1409 	long batchsummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
1410 	struct rcu_torture *rtcp;
1411 	static unsigned long rtcv_snap = ULONG_MAX;
1412 	static bool splatted;
1413 	struct task_struct *wtp;
1414 
1415 	for_each_possible_cpu(cpu) {
1416 		for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
1417 			pipesummary[i] += READ_ONCE(per_cpu(rcu_torture_count, cpu)[i]);
1418 			batchsummary[i] += READ_ONCE(per_cpu(rcu_torture_batch, cpu)[i]);
1419 		}
1420 	}
1421 	for (i = RCU_TORTURE_PIPE_LEN - 1; i >= 0; i--) {
1422 		if (pipesummary[i] != 0)
1423 			break;
1424 	}
1425 
1426 	pr_alert("%s%s ", torture_type, TORTURE_FLAG);
1427 	rtcp = rcu_access_pointer(rcu_torture_current);
1428 	pr_cont("rtc: %p %s: %lu tfle: %d rta: %d rtaf: %d rtf: %d ",
1429 		rtcp,
1430 		rtcp && !rcu_stall_is_suppressed_at_boot() ? "ver" : "VER",
1431 		rcu_torture_current_version,
1432 		list_empty(&rcu_torture_freelist),
1433 		atomic_read(&n_rcu_torture_alloc),
1434 		atomic_read(&n_rcu_torture_alloc_fail),
1435 		atomic_read(&n_rcu_torture_free));
1436 	pr_cont("rtmbe: %d rtbe: %ld rtbke: %ld rtbre: %ld ",
1437 		atomic_read(&n_rcu_torture_mberror),
1438 		n_rcu_torture_barrier_error,
1439 		n_rcu_torture_boost_ktrerror,
1440 		n_rcu_torture_boost_rterror);
1441 	pr_cont("rtbf: %ld rtb: %ld nt: %ld ",
1442 		n_rcu_torture_boost_failure,
1443 		n_rcu_torture_boosts,
1444 		atomic_long_read(&n_rcu_torture_timers));
1445 	torture_onoff_stats();
1446 	pr_cont("barrier: %ld/%ld:%ld\n",
1447 		n_barrier_successes,
1448 		n_barrier_attempts,
1449 		n_rcu_torture_barrier_error);
1450 
1451 	pr_alert("%s%s ", torture_type, TORTURE_FLAG);
1452 	if (atomic_read(&n_rcu_torture_mberror) ||
1453 	    n_rcu_torture_barrier_error || n_rcu_torture_boost_ktrerror ||
1454 	    n_rcu_torture_boost_rterror || n_rcu_torture_boost_failure ||
1455 	    i > 1) {
1456 		pr_cont("%s", "!!! ");
1457 		atomic_inc(&n_rcu_torture_error);
1458 		WARN_ON_ONCE(atomic_read(&n_rcu_torture_mberror));
1459 		WARN_ON_ONCE(n_rcu_torture_barrier_error);  // rcu_barrier()
1460 		WARN_ON_ONCE(n_rcu_torture_boost_ktrerror); // no boost kthread
1461 		WARN_ON_ONCE(n_rcu_torture_boost_rterror); // can't set RT prio
1462 		WARN_ON_ONCE(n_rcu_torture_boost_failure); // RCU boost failed
1463 		WARN_ON_ONCE(i > 1); // Too-short grace period
1464 	}
1465 	pr_cont("Reader Pipe: ");
1466 	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
1467 		pr_cont(" %ld", pipesummary[i]);
1468 	pr_cont("\n");
1469 
1470 	pr_alert("%s%s ", torture_type, TORTURE_FLAG);
1471 	pr_cont("Reader Batch: ");
1472 	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
1473 		pr_cont(" %ld", batchsummary[i]);
1474 	pr_cont("\n");
1475 
1476 	pr_alert("%s%s ", torture_type, TORTURE_FLAG);
1477 	pr_cont("Free-Block Circulation: ");
1478 	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
1479 		pr_cont(" %d", atomic_read(&rcu_torture_wcount[i]));
1480 	}
1481 	pr_cont("\n");
1482 
1483 	if (cur_ops->stats)
1484 		cur_ops->stats();
1485 	if (rtcv_snap == rcu_torture_current_version &&
1486 	    rcu_access_pointer(rcu_torture_current) &&
1487 	    !rcu_stall_is_suppressed()) {
1488 		int __maybe_unused flags = 0;
1489 		unsigned long __maybe_unused gp_seq = 0;
1490 
1491 		rcutorture_get_gp_data(cur_ops->ttype,
1492 				       &flags, &gp_seq);
1493 		srcutorture_get_gp_data(cur_ops->ttype, srcu_ctlp,
1494 					&flags, &gp_seq);
1495 		wtp = READ_ONCE(writer_task);
1496 		pr_alert("??? Writer stall state %s(%d) g%lu f%#x ->state %#lx cpu %d\n",
1497 			 rcu_torture_writer_state_getname(),
1498 			 rcu_torture_writer_state, gp_seq, flags,
1499 			 wtp == NULL ? ~0UL : wtp->state,
1500 			 wtp == NULL ? -1 : (int)task_cpu(wtp));
1501 		if (!splatted && wtp) {
1502 			sched_show_task(wtp);
1503 			splatted = true;
1504 		}
1505 		show_rcu_gp_kthreads();
1506 		rcu_ftrace_dump(DUMP_ALL);
1507 	}
1508 	rtcv_snap = rcu_torture_current_version;
1509 }
1510 
1511 /*
1512  * Periodically prints torture statistics, if periodic statistics printing
1513  * was specified via the stat_interval module parameter.
1514  */
1515 static int
1516 rcu_torture_stats(void *arg)
1517 {
1518 	VERBOSE_TOROUT_STRING("rcu_torture_stats task started");
1519 	do {
1520 		schedule_timeout_interruptible(stat_interval * HZ);
1521 		rcu_torture_stats_print();
1522 		torture_shutdown_absorb("rcu_torture_stats");
1523 	} while (!torture_must_stop());
1524 	torture_kthread_stopping("rcu_torture_stats");
1525 	return 0;
1526 }
1527 
1528 static void
1529 rcu_torture_print_module_parms(struct rcu_torture_ops *cur_ops, const char *tag)
1530 {
1531 	pr_alert("%s" TORTURE_FLAG
1532 		 "--- %s: nreaders=%d nfakewriters=%d "
1533 		 "stat_interval=%d verbose=%d test_no_idle_hz=%d "
1534 		 "shuffle_interval=%d stutter=%d irqreader=%d "
1535 		 "fqs_duration=%d fqs_holdoff=%d fqs_stutter=%d "
1536 		 "test_boost=%d/%d test_boost_interval=%d "
1537 		 "test_boost_duration=%d shutdown_secs=%d "
1538 		 "stall_cpu=%d stall_cpu_holdoff=%d stall_cpu_irqsoff=%d "
1539 		 "n_barrier_cbs=%d "
1540 		 "onoff_interval=%d onoff_holdoff=%d\n",
1541 		 torture_type, tag, nrealreaders, nfakewriters,
1542 		 stat_interval, verbose, test_no_idle_hz, shuffle_interval,
1543 		 stutter, irqreader, fqs_duration, fqs_holdoff, fqs_stutter,
1544 		 test_boost, cur_ops->can_boost,
1545 		 test_boost_interval, test_boost_duration, shutdown_secs,
1546 		 stall_cpu, stall_cpu_holdoff, stall_cpu_irqsoff,
1547 		 n_barrier_cbs,
1548 		 onoff_interval, onoff_holdoff);
1549 }
1550 
1551 static int rcutorture_booster_cleanup(unsigned int cpu)
1552 {
1553 	struct task_struct *t;
1554 
1555 	if (boost_tasks[cpu] == NULL)
1556 		return 0;
1557 	mutex_lock(&boost_mutex);
1558 	t = boost_tasks[cpu];
1559 	boost_tasks[cpu] = NULL;
1560 	rcu_torture_enable_rt_throttle();
1561 	mutex_unlock(&boost_mutex);
1562 
1563 	/* This must be outside of the mutex, otherwise deadlock! */
1564 	torture_stop_kthread(rcu_torture_boost, t);
1565 	return 0;
1566 }
1567 
1568 static int rcutorture_booster_init(unsigned int cpu)
1569 {
1570 	int retval;
1571 
1572 	if (boost_tasks[cpu] != NULL)
1573 		return 0;  /* Already created, nothing more to do. */
1574 
1575 	/* Don't allow time recalculation while creating a new task. */
1576 	mutex_lock(&boost_mutex);
1577 	rcu_torture_disable_rt_throttle();
1578 	VERBOSE_TOROUT_STRING("Creating rcu_torture_boost task");
1579 	boost_tasks[cpu] = kthread_create_on_node(rcu_torture_boost, NULL,
1580 						  cpu_to_node(cpu),
1581 						  "rcu_torture_boost");
1582 	if (IS_ERR(boost_tasks[cpu])) {
1583 		retval = PTR_ERR(boost_tasks[cpu]);
1584 		VERBOSE_TOROUT_STRING("rcu_torture_boost task create failed");
1585 		n_rcu_torture_boost_ktrerror++;
1586 		boost_tasks[cpu] = NULL;
1587 		mutex_unlock(&boost_mutex);
1588 		return retval;
1589 	}
1590 	kthread_bind(boost_tasks[cpu], cpu);
1591 	wake_up_process(boost_tasks[cpu]);
1592 	mutex_unlock(&boost_mutex);
1593 	return 0;
1594 }
1595 
1596 /*
1597  * CPU-stall kthread.  It waits as specified by stall_cpu_holdoff, then
1598  * induces a CPU stall for the time specified by stall_cpu.
1599  */
1600 static int rcu_torture_stall(void *args)
1601 {
1602 	unsigned long stop_at;
1603 
1604 	VERBOSE_TOROUT_STRING("rcu_torture_stall task started");
1605 	if (stall_cpu_holdoff > 0) {
1606 		VERBOSE_TOROUT_STRING("rcu_torture_stall begin holdoff");
1607 		schedule_timeout_interruptible(stall_cpu_holdoff * HZ);
1608 		VERBOSE_TOROUT_STRING("rcu_torture_stall end holdoff");
1609 	}
1610 	if (!kthread_should_stop()) {
1611 		stop_at = ktime_get_seconds() + stall_cpu;
1612 		/* RCU CPU stall is expected behavior in following code. */
1613 		rcu_read_lock();
1614 		if (stall_cpu_irqsoff)
1615 			local_irq_disable();
1616 		else
1617 			preempt_disable();
1618 		pr_alert("rcu_torture_stall start on CPU %d.\n",
1619 			 smp_processor_id());
1620 		while (ULONG_CMP_LT((unsigned long)ktime_get_seconds(),
1621 				    stop_at))
1622 			continue;  /* Induce RCU CPU stall warning. */
1623 		if (stall_cpu_irqsoff)
1624 			local_irq_enable();
1625 		else
1626 			preempt_enable();
1627 		rcu_read_unlock();
1628 		pr_alert("rcu_torture_stall end.\n");
1629 	}
1630 	torture_shutdown_absorb("rcu_torture_stall");
1631 	while (!kthread_should_stop())
1632 		schedule_timeout_interruptible(10 * HZ);
1633 	return 0;
1634 }
1635 
1636 /* Spawn CPU-stall kthread, if stall_cpu specified. */
1637 static int __init rcu_torture_stall_init(void)
1638 {
1639 	if (stall_cpu <= 0)
1640 		return 0;
1641 	return torture_create_kthread(rcu_torture_stall, NULL, stall_task);
1642 }
1643 
1644 /* State structure for forward-progress self-propagating RCU callback. */
1645 struct fwd_cb_state {
1646 	struct rcu_head rh;
1647 	int stop;
1648 };
1649 
1650 /*
1651  * Forward-progress self-propagating RCU callback function.  Because
1652  * callbacks run from softirq, this function is an implicit RCU read-side
1653  * critical section.
1654  */
1655 static void rcu_torture_fwd_prog_cb(struct rcu_head *rhp)
1656 {
1657 	struct fwd_cb_state *fcsp = container_of(rhp, struct fwd_cb_state, rh);
1658 
1659 	if (READ_ONCE(fcsp->stop)) {
1660 		WRITE_ONCE(fcsp->stop, 2);
1661 		return;
1662 	}
1663 	cur_ops->call(&fcsp->rh, rcu_torture_fwd_prog_cb);
1664 }
1665 
1666 /* State for continuous-flood RCU callbacks. */
1667 struct rcu_fwd_cb {
1668 	struct rcu_head rh;
1669 	struct rcu_fwd_cb *rfc_next;
1670 	struct rcu_fwd *rfc_rfp;
1671 	int rfc_gps;
1672 };
1673 
1674 #define MAX_FWD_CB_JIFFIES	(8 * HZ) /* Maximum CB test duration. */
1675 #define MIN_FWD_CB_LAUNDERS	3	/* This many CB invocations to count. */
1676 #define MIN_FWD_CBS_LAUNDERED	100	/* Number of counted CBs. */
1677 #define FWD_CBS_HIST_DIV	10	/* Histogram buckets/second. */
1678 #define N_LAUNDERS_HIST (2 * MAX_FWD_CB_JIFFIES / (HZ / FWD_CBS_HIST_DIV))
1679 
1680 struct rcu_launder_hist {
1681 	long n_launders;
1682 	unsigned long launder_gp_seq;
1683 };
1684 
1685 struct rcu_fwd {
1686 	spinlock_t rcu_fwd_lock;
1687 	struct rcu_fwd_cb *rcu_fwd_cb_head;
1688 	struct rcu_fwd_cb **rcu_fwd_cb_tail;
1689 	long n_launders_cb;
1690 	unsigned long rcu_fwd_startat;
1691 	struct rcu_launder_hist n_launders_hist[N_LAUNDERS_HIST];
1692 	unsigned long rcu_launder_gp_seq_start;
1693 };
1694 
1695 struct rcu_fwd *rcu_fwds;
1696 bool rcu_fwd_emergency_stop;
1697 
1698 static void rcu_torture_fwd_cb_hist(struct rcu_fwd *rfp)
1699 {
1700 	unsigned long gps;
1701 	unsigned long gps_old;
1702 	int i;
1703 	int j;
1704 
1705 	for (i = ARRAY_SIZE(rfp->n_launders_hist) - 1; i > 0; i--)
1706 		if (rfp->n_launders_hist[i].n_launders > 0)
1707 			break;
1708 	pr_alert("%s: Callback-invocation histogram (duration %lu jiffies):",
1709 		 __func__, jiffies - rfp->rcu_fwd_startat);
1710 	gps_old = rfp->rcu_launder_gp_seq_start;
1711 	for (j = 0; j <= i; j++) {
1712 		gps = rfp->n_launders_hist[j].launder_gp_seq;
1713 		pr_cont(" %ds/%d: %ld:%ld",
1714 			j + 1, FWD_CBS_HIST_DIV,
1715 			rfp->n_launders_hist[j].n_launders,
1716 			rcutorture_seq_diff(gps, gps_old));
1717 		gps_old = gps;
1718 	}
1719 	pr_cont("\n");
1720 }
1721 
1722 /* Callback function for continuous-flood RCU callbacks. */
1723 static void rcu_torture_fwd_cb_cr(struct rcu_head *rhp)
1724 {
1725 	unsigned long flags;
1726 	int i;
1727 	struct rcu_fwd_cb *rfcp = container_of(rhp, struct rcu_fwd_cb, rh);
1728 	struct rcu_fwd_cb **rfcpp;
1729 	struct rcu_fwd *rfp = rfcp->rfc_rfp;
1730 
1731 	rfcp->rfc_next = NULL;
1732 	rfcp->rfc_gps++;
1733 	spin_lock_irqsave(&rfp->rcu_fwd_lock, flags);
1734 	rfcpp = rfp->rcu_fwd_cb_tail;
1735 	rfp->rcu_fwd_cb_tail = &rfcp->rfc_next;
1736 	WRITE_ONCE(*rfcpp, rfcp);
1737 	WRITE_ONCE(rfp->n_launders_cb, rfp->n_launders_cb + 1);
1738 	i = ((jiffies - rfp->rcu_fwd_startat) / (HZ / FWD_CBS_HIST_DIV));
1739 	if (i >= ARRAY_SIZE(rfp->n_launders_hist))
1740 		i = ARRAY_SIZE(rfp->n_launders_hist) - 1;
1741 	rfp->n_launders_hist[i].n_launders++;
1742 	rfp->n_launders_hist[i].launder_gp_seq = cur_ops->get_gp_seq();
1743 	spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
1744 }
1745 
1746 // Give the scheduler a chance, even on nohz_full CPUs.
1747 static void rcu_torture_fwd_prog_cond_resched(unsigned long iter)
1748 {
1749 	if (IS_ENABLED(CONFIG_PREEMPTION) && IS_ENABLED(CONFIG_NO_HZ_FULL)) {
1750 		// Real call_rcu() floods hit userspace, so emulate that.
1751 		if (need_resched() || (iter & 0xfff))
1752 			schedule();
1753 		return;
1754 	}
1755 	// No userspace emulation: CB invocation throttles call_rcu()
1756 	cond_resched();
1757 }
1758 
1759 /*
1760  * Free all callbacks on the rcu_fwd_cb_head list, either because the
1761  * test is over or because we hit an OOM event.
1762  */
1763 static unsigned long rcu_torture_fwd_prog_cbfree(struct rcu_fwd *rfp)
1764 {
1765 	unsigned long flags;
1766 	unsigned long freed = 0;
1767 	struct rcu_fwd_cb *rfcp;
1768 
1769 	for (;;) {
1770 		spin_lock_irqsave(&rfp->rcu_fwd_lock, flags);
1771 		rfcp = rfp->rcu_fwd_cb_head;
1772 		if (!rfcp) {
1773 			spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
1774 			break;
1775 		}
1776 		rfp->rcu_fwd_cb_head = rfcp->rfc_next;
1777 		if (!rfp->rcu_fwd_cb_head)
1778 			rfp->rcu_fwd_cb_tail = &rfp->rcu_fwd_cb_head;
1779 		spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
1780 		kfree(rfcp);
1781 		freed++;
1782 		rcu_torture_fwd_prog_cond_resched(freed);
1783 		if (tick_nohz_full_enabled()) {
1784 			local_irq_save(flags);
1785 			rcu_momentary_dyntick_idle();
1786 			local_irq_restore(flags);
1787 		}
1788 	}
1789 	return freed;
1790 }
1791 
1792 /* Carry out need_resched()/cond_resched() forward-progress testing. */
1793 static void rcu_torture_fwd_prog_nr(struct rcu_fwd *rfp,
1794 				    int *tested, int *tested_tries)
1795 {
1796 	unsigned long cver;
1797 	unsigned long dur;
1798 	struct fwd_cb_state fcs;
1799 	unsigned long gps;
1800 	int idx;
1801 	int sd;
1802 	int sd4;
1803 	bool selfpropcb = false;
1804 	unsigned long stopat;
1805 	static DEFINE_TORTURE_RANDOM(trs);
1806 
1807 	if  (cur_ops->call && cur_ops->sync && cur_ops->cb_barrier) {
1808 		init_rcu_head_on_stack(&fcs.rh);
1809 		selfpropcb = true;
1810 	}
1811 
1812 	/* Tight loop containing cond_resched(). */
1813 	WRITE_ONCE(rcu_fwd_cb_nodelay, true);
1814 	cur_ops->sync(); /* Later readers see above write. */
1815 	if  (selfpropcb) {
1816 		WRITE_ONCE(fcs.stop, 0);
1817 		cur_ops->call(&fcs.rh, rcu_torture_fwd_prog_cb);
1818 	}
1819 	cver = READ_ONCE(rcu_torture_current_version);
1820 	gps = cur_ops->get_gp_seq();
1821 	sd = cur_ops->stall_dur() + 1;
1822 	sd4 = (sd + fwd_progress_div - 1) / fwd_progress_div;
1823 	dur = sd4 + torture_random(&trs) % (sd - sd4);
1824 	WRITE_ONCE(rfp->rcu_fwd_startat, jiffies);
1825 	stopat = rfp->rcu_fwd_startat + dur;
1826 	while (time_before(jiffies, stopat) &&
1827 	       !shutdown_time_arrived() &&
1828 	       !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
1829 		idx = cur_ops->readlock();
1830 		udelay(10);
1831 		cur_ops->readunlock(idx);
1832 		if (!fwd_progress_need_resched || need_resched())
1833 			cond_resched();
1834 	}
1835 	(*tested_tries)++;
1836 	if (!time_before(jiffies, stopat) &&
1837 	    !shutdown_time_arrived() &&
1838 	    !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
1839 		(*tested)++;
1840 		cver = READ_ONCE(rcu_torture_current_version) - cver;
1841 		gps = rcutorture_seq_diff(cur_ops->get_gp_seq(), gps);
1842 		WARN_ON(!cver && gps < 2);
1843 		pr_alert("%s: Duration %ld cver %ld gps %ld\n", __func__, dur, cver, gps);
1844 	}
1845 	if (selfpropcb) {
1846 		WRITE_ONCE(fcs.stop, 1);
1847 		cur_ops->sync(); /* Wait for running CB to complete. */
1848 		cur_ops->cb_barrier(); /* Wait for queued callbacks. */
1849 	}
1850 
1851 	if (selfpropcb) {
1852 		WARN_ON(READ_ONCE(fcs.stop) != 2);
1853 		destroy_rcu_head_on_stack(&fcs.rh);
1854 	}
1855 	schedule_timeout_uninterruptible(HZ / 10); /* Let kthreads recover. */
1856 	WRITE_ONCE(rcu_fwd_cb_nodelay, false);
1857 }
1858 
1859 /* Carry out call_rcu() forward-progress testing. */
1860 static void rcu_torture_fwd_prog_cr(struct rcu_fwd *rfp)
1861 {
1862 	unsigned long cver;
1863 	unsigned long flags;
1864 	unsigned long gps;
1865 	int i;
1866 	long n_launders;
1867 	long n_launders_cb_snap;
1868 	long n_launders_sa;
1869 	long n_max_cbs;
1870 	long n_max_gps;
1871 	struct rcu_fwd_cb *rfcp;
1872 	struct rcu_fwd_cb *rfcpn;
1873 	unsigned long stopat;
1874 	unsigned long stoppedat;
1875 
1876 	if (READ_ONCE(rcu_fwd_emergency_stop))
1877 		return; /* Get out of the way quickly, no GP wait! */
1878 	if (!cur_ops->call)
1879 		return; /* Can't do call_rcu() fwd prog without ->call. */
1880 
1881 	/* Loop continuously posting RCU callbacks. */
1882 	WRITE_ONCE(rcu_fwd_cb_nodelay, true);
1883 	cur_ops->sync(); /* Later readers see above write. */
1884 	WRITE_ONCE(rfp->rcu_fwd_startat, jiffies);
1885 	stopat = rfp->rcu_fwd_startat + MAX_FWD_CB_JIFFIES;
1886 	n_launders = 0;
1887 	rfp->n_launders_cb = 0; // Hoist initialization for multi-kthread
1888 	n_launders_sa = 0;
1889 	n_max_cbs = 0;
1890 	n_max_gps = 0;
1891 	for (i = 0; i < ARRAY_SIZE(rfp->n_launders_hist); i++)
1892 		rfp->n_launders_hist[i].n_launders = 0;
1893 	cver = READ_ONCE(rcu_torture_current_version);
1894 	gps = cur_ops->get_gp_seq();
1895 	rfp->rcu_launder_gp_seq_start = gps;
1896 	tick_dep_set_task(current, TICK_DEP_BIT_RCU);
1897 	while (time_before(jiffies, stopat) &&
1898 	       !shutdown_time_arrived() &&
1899 	       !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
1900 		rfcp = READ_ONCE(rfp->rcu_fwd_cb_head);
1901 		rfcpn = NULL;
1902 		if (rfcp)
1903 			rfcpn = READ_ONCE(rfcp->rfc_next);
1904 		if (rfcpn) {
1905 			if (rfcp->rfc_gps >= MIN_FWD_CB_LAUNDERS &&
1906 			    ++n_max_gps >= MIN_FWD_CBS_LAUNDERED)
1907 				break;
1908 			rfp->rcu_fwd_cb_head = rfcpn;
1909 			n_launders++;
1910 			n_launders_sa++;
1911 		} else {
1912 			rfcp = kmalloc(sizeof(*rfcp), GFP_KERNEL);
1913 			if (WARN_ON_ONCE(!rfcp)) {
1914 				schedule_timeout_interruptible(1);
1915 				continue;
1916 			}
1917 			n_max_cbs++;
1918 			n_launders_sa = 0;
1919 			rfcp->rfc_gps = 0;
1920 			rfcp->rfc_rfp = rfp;
1921 		}
1922 		cur_ops->call(&rfcp->rh, rcu_torture_fwd_cb_cr);
1923 		rcu_torture_fwd_prog_cond_resched(n_launders + n_max_cbs);
1924 		if (tick_nohz_full_enabled()) {
1925 			local_irq_save(flags);
1926 			rcu_momentary_dyntick_idle();
1927 			local_irq_restore(flags);
1928 		}
1929 	}
1930 	stoppedat = jiffies;
1931 	n_launders_cb_snap = READ_ONCE(rfp->n_launders_cb);
1932 	cver = READ_ONCE(rcu_torture_current_version) - cver;
1933 	gps = rcutorture_seq_diff(cur_ops->get_gp_seq(), gps);
1934 	cur_ops->cb_barrier(); /* Wait for callbacks to be invoked. */
1935 	(void)rcu_torture_fwd_prog_cbfree(rfp);
1936 
1937 	if (!torture_must_stop() && !READ_ONCE(rcu_fwd_emergency_stop) &&
1938 	    !shutdown_time_arrived()) {
1939 		WARN_ON(n_max_gps < MIN_FWD_CBS_LAUNDERED);
1940 		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",
1941 			 __func__,
1942 			 stoppedat - rfp->rcu_fwd_startat, jiffies - stoppedat,
1943 			 n_launders + n_max_cbs - n_launders_cb_snap,
1944 			 n_launders, n_launders_sa,
1945 			 n_max_gps, n_max_cbs, cver, gps);
1946 		rcu_torture_fwd_cb_hist(rfp);
1947 	}
1948 	schedule_timeout_uninterruptible(HZ); /* Let CBs drain. */
1949 	tick_dep_clear_task(current, TICK_DEP_BIT_RCU);
1950 	WRITE_ONCE(rcu_fwd_cb_nodelay, false);
1951 }
1952 
1953 
1954 /*
1955  * OOM notifier, but this only prints diagnostic information for the
1956  * current forward-progress test.
1957  */
1958 static int rcutorture_oom_notify(struct notifier_block *self,
1959 				 unsigned long notused, void *nfreed)
1960 {
1961 	struct rcu_fwd *rfp = rcu_fwds;
1962 
1963 	WARN(1, "%s invoked upon OOM during forward-progress testing.\n",
1964 	     __func__);
1965 	rcu_torture_fwd_cb_hist(rfp);
1966 	rcu_fwd_progress_check(1 + (jiffies - READ_ONCE(rfp->rcu_fwd_startat)) / 2);
1967 	WRITE_ONCE(rcu_fwd_emergency_stop, true);
1968 	smp_mb(); /* Emergency stop before free and wait to avoid hangs. */
1969 	pr_info("%s: Freed %lu RCU callbacks.\n",
1970 		__func__, rcu_torture_fwd_prog_cbfree(rfp));
1971 	rcu_barrier();
1972 	pr_info("%s: Freed %lu RCU callbacks.\n",
1973 		__func__, rcu_torture_fwd_prog_cbfree(rfp));
1974 	rcu_barrier();
1975 	pr_info("%s: Freed %lu RCU callbacks.\n",
1976 		__func__, rcu_torture_fwd_prog_cbfree(rfp));
1977 	smp_mb(); /* Frees before return to avoid redoing OOM. */
1978 	(*(unsigned long *)nfreed)++; /* Forward progress CBs freed! */
1979 	pr_info("%s returning after OOM processing.\n", __func__);
1980 	return NOTIFY_OK;
1981 }
1982 
1983 static struct notifier_block rcutorture_oom_nb = {
1984 	.notifier_call = rcutorture_oom_notify
1985 };
1986 
1987 /* Carry out grace-period forward-progress testing. */
1988 static int rcu_torture_fwd_prog(void *args)
1989 {
1990 	struct rcu_fwd *rfp = args;
1991 	int tested = 0;
1992 	int tested_tries = 0;
1993 
1994 	VERBOSE_TOROUT_STRING("rcu_torture_fwd_progress task started");
1995 	rcu_bind_current_to_nocb();
1996 	if (!IS_ENABLED(CONFIG_SMP) || !IS_ENABLED(CONFIG_RCU_BOOST))
1997 		set_user_nice(current, MAX_NICE);
1998 	do {
1999 		schedule_timeout_interruptible(fwd_progress_holdoff * HZ);
2000 		WRITE_ONCE(rcu_fwd_emergency_stop, false);
2001 		register_oom_notifier(&rcutorture_oom_nb);
2002 		if (!IS_ENABLED(CONFIG_TINY_RCU) ||
2003 		    rcu_inkernel_boot_has_ended())
2004 			rcu_torture_fwd_prog_nr(rfp, &tested, &tested_tries);
2005 		if (rcu_inkernel_boot_has_ended())
2006 			rcu_torture_fwd_prog_cr(rfp);
2007 		unregister_oom_notifier(&rcutorture_oom_nb);
2008 
2009 		/* Avoid slow periods, better to test when busy. */
2010 		stutter_wait("rcu_torture_fwd_prog");
2011 	} while (!torture_must_stop());
2012 	/* Short runs might not contain a valid forward-progress attempt. */
2013 	WARN_ON(!tested && tested_tries >= 5);
2014 	pr_alert("%s: tested %d tested_tries %d\n", __func__, tested, tested_tries);
2015 	torture_kthread_stopping("rcu_torture_fwd_prog");
2016 	return 0;
2017 }
2018 
2019 /* If forward-progress checking is requested and feasible, spawn the thread. */
2020 static int __init rcu_torture_fwd_prog_init(void)
2021 {
2022 	struct rcu_fwd *rfp;
2023 
2024 	if (!fwd_progress)
2025 		return 0; /* Not requested, so don't do it. */
2026 	if (!cur_ops->stall_dur || cur_ops->stall_dur() <= 0 ||
2027 	    cur_ops == &rcu_busted_ops) {
2028 		VERBOSE_TOROUT_STRING("rcu_torture_fwd_prog_init: Disabled, unsupported by RCU flavor under test");
2029 		return 0;
2030 	}
2031 	if (stall_cpu > 0) {
2032 		VERBOSE_TOROUT_STRING("rcu_torture_fwd_prog_init: Disabled, conflicts with CPU-stall testing");
2033 		if (IS_MODULE(CONFIG_RCU_TORTURE_TESTS))
2034 			return -EINVAL; /* In module, can fail back to user. */
2035 		WARN_ON(1); /* Make sure rcutorture notices conflict. */
2036 		return 0;
2037 	}
2038 	if (fwd_progress_holdoff <= 0)
2039 		fwd_progress_holdoff = 1;
2040 	if (fwd_progress_div <= 0)
2041 		fwd_progress_div = 4;
2042 	rfp = kzalloc(sizeof(*rfp), GFP_KERNEL);
2043 	if (!rfp)
2044 		return -ENOMEM;
2045 	spin_lock_init(&rfp->rcu_fwd_lock);
2046 	rfp->rcu_fwd_cb_tail = &rfp->rcu_fwd_cb_head;
2047 	return torture_create_kthread(rcu_torture_fwd_prog, rfp, fwd_prog_task);
2048 }
2049 
2050 /* Callback function for RCU barrier testing. */
2051 static void rcu_torture_barrier_cbf(struct rcu_head *rcu)
2052 {
2053 	atomic_inc(&barrier_cbs_invoked);
2054 }
2055 
2056 /* IPI handler to get callback posted on desired CPU, if online. */
2057 static void rcu_torture_barrier1cb(void *rcu_void)
2058 {
2059 	struct rcu_head *rhp = rcu_void;
2060 
2061 	cur_ops->call(rhp, rcu_torture_barrier_cbf);
2062 }
2063 
2064 /* kthread function to register callbacks used to test RCU barriers. */
2065 static int rcu_torture_barrier_cbs(void *arg)
2066 {
2067 	long myid = (long)arg;
2068 	bool lastphase = 0;
2069 	bool newphase;
2070 	struct rcu_head rcu;
2071 
2072 	init_rcu_head_on_stack(&rcu);
2073 	VERBOSE_TOROUT_STRING("rcu_torture_barrier_cbs task started");
2074 	set_user_nice(current, MAX_NICE);
2075 	do {
2076 		wait_event(barrier_cbs_wq[myid],
2077 			   (newphase =
2078 			    smp_load_acquire(&barrier_phase)) != lastphase ||
2079 			   torture_must_stop());
2080 		lastphase = newphase;
2081 		if (torture_must_stop())
2082 			break;
2083 		/*
2084 		 * The above smp_load_acquire() ensures barrier_phase load
2085 		 * is ordered before the following ->call().
2086 		 */
2087 		if (smp_call_function_single(myid, rcu_torture_barrier1cb,
2088 					     &rcu, 1)) {
2089 			// IPI failed, so use direct call from current CPU.
2090 			cur_ops->call(&rcu, rcu_torture_barrier_cbf);
2091 		}
2092 		if (atomic_dec_and_test(&barrier_cbs_count))
2093 			wake_up(&barrier_wq);
2094 	} while (!torture_must_stop());
2095 	if (cur_ops->cb_barrier != NULL)
2096 		cur_ops->cb_barrier();
2097 	destroy_rcu_head_on_stack(&rcu);
2098 	torture_kthread_stopping("rcu_torture_barrier_cbs");
2099 	return 0;
2100 }
2101 
2102 /* kthread function to drive and coordinate RCU barrier testing. */
2103 static int rcu_torture_barrier(void *arg)
2104 {
2105 	int i;
2106 
2107 	VERBOSE_TOROUT_STRING("rcu_torture_barrier task starting");
2108 	do {
2109 		atomic_set(&barrier_cbs_invoked, 0);
2110 		atomic_set(&barrier_cbs_count, n_barrier_cbs);
2111 		/* Ensure barrier_phase ordered after prior assignments. */
2112 		smp_store_release(&barrier_phase, !barrier_phase);
2113 		for (i = 0; i < n_barrier_cbs; i++)
2114 			wake_up(&barrier_cbs_wq[i]);
2115 		wait_event(barrier_wq,
2116 			   atomic_read(&barrier_cbs_count) == 0 ||
2117 			   torture_must_stop());
2118 		if (torture_must_stop())
2119 			break;
2120 		n_barrier_attempts++;
2121 		cur_ops->cb_barrier(); /* Implies smp_mb() for wait_event(). */
2122 		if (atomic_read(&barrier_cbs_invoked) != n_barrier_cbs) {
2123 			n_rcu_torture_barrier_error++;
2124 			pr_err("barrier_cbs_invoked = %d, n_barrier_cbs = %d\n",
2125 			       atomic_read(&barrier_cbs_invoked),
2126 			       n_barrier_cbs);
2127 			WARN_ON(1);
2128 			// Wait manually for the remaining callbacks
2129 			i = 0;
2130 			do {
2131 				if (WARN_ON(i++ > HZ))
2132 					i = INT_MIN;
2133 				schedule_timeout_interruptible(1);
2134 				cur_ops->cb_barrier();
2135 			} while (atomic_read(&barrier_cbs_invoked) !=
2136 				 n_barrier_cbs &&
2137 				 !torture_must_stop());
2138 			smp_mb(); // Can't trust ordering if broken.
2139 			if (!torture_must_stop())
2140 				pr_err("Recovered: barrier_cbs_invoked = %d\n",
2141 				       atomic_read(&barrier_cbs_invoked));
2142 		} else {
2143 			n_barrier_successes++;
2144 		}
2145 		schedule_timeout_interruptible(HZ / 10);
2146 	} while (!torture_must_stop());
2147 	torture_kthread_stopping("rcu_torture_barrier");
2148 	return 0;
2149 }
2150 
2151 /* Initialize RCU barrier testing. */
2152 static int rcu_torture_barrier_init(void)
2153 {
2154 	int i;
2155 	int ret;
2156 
2157 	if (n_barrier_cbs <= 0)
2158 		return 0;
2159 	if (cur_ops->call == NULL || cur_ops->cb_barrier == NULL) {
2160 		pr_alert("%s" TORTURE_FLAG
2161 			 " Call or barrier ops missing for %s,\n",
2162 			 torture_type, cur_ops->name);
2163 		pr_alert("%s" TORTURE_FLAG
2164 			 " RCU barrier testing omitted from run.\n",
2165 			 torture_type);
2166 		return 0;
2167 	}
2168 	atomic_set(&barrier_cbs_count, 0);
2169 	atomic_set(&barrier_cbs_invoked, 0);
2170 	barrier_cbs_tasks =
2171 		kcalloc(n_barrier_cbs, sizeof(barrier_cbs_tasks[0]),
2172 			GFP_KERNEL);
2173 	barrier_cbs_wq =
2174 		kcalloc(n_barrier_cbs, sizeof(barrier_cbs_wq[0]), GFP_KERNEL);
2175 	if (barrier_cbs_tasks == NULL || !barrier_cbs_wq)
2176 		return -ENOMEM;
2177 	for (i = 0; i < n_barrier_cbs; i++) {
2178 		init_waitqueue_head(&barrier_cbs_wq[i]);
2179 		ret = torture_create_kthread(rcu_torture_barrier_cbs,
2180 					     (void *)(long)i,
2181 					     barrier_cbs_tasks[i]);
2182 		if (ret)
2183 			return ret;
2184 	}
2185 	return torture_create_kthread(rcu_torture_barrier, NULL, barrier_task);
2186 }
2187 
2188 /* Clean up after RCU barrier testing. */
2189 static void rcu_torture_barrier_cleanup(void)
2190 {
2191 	int i;
2192 
2193 	torture_stop_kthread(rcu_torture_barrier, barrier_task);
2194 	if (barrier_cbs_tasks != NULL) {
2195 		for (i = 0; i < n_barrier_cbs; i++)
2196 			torture_stop_kthread(rcu_torture_barrier_cbs,
2197 					     barrier_cbs_tasks[i]);
2198 		kfree(barrier_cbs_tasks);
2199 		barrier_cbs_tasks = NULL;
2200 	}
2201 	if (barrier_cbs_wq != NULL) {
2202 		kfree(barrier_cbs_wq);
2203 		barrier_cbs_wq = NULL;
2204 	}
2205 }
2206 
2207 static bool rcu_torture_can_boost(void)
2208 {
2209 	static int boost_warn_once;
2210 	int prio;
2211 
2212 	if (!(test_boost == 1 && cur_ops->can_boost) && test_boost != 2)
2213 		return false;
2214 
2215 	prio = rcu_get_gp_kthreads_prio();
2216 	if (!prio)
2217 		return false;
2218 
2219 	if (prio < 2) {
2220 		if (boost_warn_once  == 1)
2221 			return false;
2222 
2223 		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);
2224 		boost_warn_once = 1;
2225 		return false;
2226 	}
2227 
2228 	return true;
2229 }
2230 
2231 static enum cpuhp_state rcutor_hp;
2232 
2233 static void
2234 rcu_torture_cleanup(void)
2235 {
2236 	int firsttime;
2237 	int flags = 0;
2238 	unsigned long gp_seq = 0;
2239 	int i;
2240 
2241 	if (torture_cleanup_begin()) {
2242 		if (cur_ops->cb_barrier != NULL)
2243 			cur_ops->cb_barrier();
2244 		return;
2245 	}
2246 	if (!cur_ops) {
2247 		torture_cleanup_end();
2248 		return;
2249 	}
2250 
2251 	show_rcu_gp_kthreads();
2252 	rcu_torture_barrier_cleanup();
2253 	torture_stop_kthread(rcu_torture_fwd_prog, fwd_prog_task);
2254 	torture_stop_kthread(rcu_torture_stall, stall_task);
2255 	torture_stop_kthread(rcu_torture_writer, writer_task);
2256 
2257 	if (reader_tasks) {
2258 		for (i = 0; i < nrealreaders; i++)
2259 			torture_stop_kthread(rcu_torture_reader,
2260 					     reader_tasks[i]);
2261 		kfree(reader_tasks);
2262 	}
2263 	rcu_torture_current = NULL;
2264 
2265 	if (fakewriter_tasks) {
2266 		for (i = 0; i < nfakewriters; i++) {
2267 			torture_stop_kthread(rcu_torture_fakewriter,
2268 					     fakewriter_tasks[i]);
2269 		}
2270 		kfree(fakewriter_tasks);
2271 		fakewriter_tasks = NULL;
2272 	}
2273 
2274 	rcutorture_get_gp_data(cur_ops->ttype, &flags, &gp_seq);
2275 	srcutorture_get_gp_data(cur_ops->ttype, srcu_ctlp, &flags, &gp_seq);
2276 	pr_alert("%s:  End-test grace-period state: g%lu f%#x\n",
2277 		 cur_ops->name, gp_seq, flags);
2278 	torture_stop_kthread(rcu_torture_stats, stats_task);
2279 	torture_stop_kthread(rcu_torture_fqs, fqs_task);
2280 	if (rcu_torture_can_boost())
2281 		cpuhp_remove_state(rcutor_hp);
2282 
2283 	/*
2284 	 * Wait for all RCU callbacks to fire, then do torture-type-specific
2285 	 * cleanup operations.
2286 	 */
2287 	if (cur_ops->cb_barrier != NULL)
2288 		cur_ops->cb_barrier();
2289 	if (cur_ops->cleanup != NULL)
2290 		cur_ops->cleanup();
2291 
2292 	rcu_torture_stats_print();  /* -After- the stats thread is stopped! */
2293 
2294 	if (err_segs_recorded) {
2295 		pr_alert("Failure/close-call rcutorture reader segments:\n");
2296 		if (rt_read_nsegs == 0)
2297 			pr_alert("\t: No segments recorded!!!\n");
2298 		firsttime = 1;
2299 		for (i = 0; i < rt_read_nsegs; i++) {
2300 			pr_alert("\t%d: %#x ", i, err_segs[i].rt_readstate);
2301 			if (err_segs[i].rt_delay_jiffies != 0) {
2302 				pr_cont("%s%ldjiffies", firsttime ? "" : "+",
2303 					err_segs[i].rt_delay_jiffies);
2304 				firsttime = 0;
2305 			}
2306 			if (err_segs[i].rt_delay_ms != 0) {
2307 				pr_cont("%s%ldms", firsttime ? "" : "+",
2308 					err_segs[i].rt_delay_ms);
2309 				firsttime = 0;
2310 			}
2311 			if (err_segs[i].rt_delay_us != 0) {
2312 				pr_cont("%s%ldus", firsttime ? "" : "+",
2313 					err_segs[i].rt_delay_us);
2314 				firsttime = 0;
2315 			}
2316 			pr_cont("%s\n",
2317 				err_segs[i].rt_preempted ? "preempted" : "");
2318 
2319 		}
2320 	}
2321 	if (atomic_read(&n_rcu_torture_error) || n_rcu_torture_barrier_error)
2322 		rcu_torture_print_module_parms(cur_ops, "End of test: FAILURE");
2323 	else if (torture_onoff_failures())
2324 		rcu_torture_print_module_parms(cur_ops,
2325 					       "End of test: RCU_HOTPLUG");
2326 	else
2327 		rcu_torture_print_module_parms(cur_ops, "End of test: SUCCESS");
2328 	torture_cleanup_end();
2329 }
2330 
2331 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
2332 static void rcu_torture_leak_cb(struct rcu_head *rhp)
2333 {
2334 }
2335 
2336 static void rcu_torture_err_cb(struct rcu_head *rhp)
2337 {
2338 	/*
2339 	 * This -might- happen due to race conditions, but is unlikely.
2340 	 * The scenario that leads to this happening is that the
2341 	 * first of the pair of duplicate callbacks is queued,
2342 	 * someone else starts a grace period that includes that
2343 	 * callback, then the second of the pair must wait for the
2344 	 * next grace period.  Unlikely, but can happen.  If it
2345 	 * does happen, the debug-objects subsystem won't have splatted.
2346 	 */
2347 	pr_alert("%s: duplicated callback was invoked.\n", KBUILD_MODNAME);
2348 }
2349 #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
2350 
2351 /*
2352  * Verify that double-free causes debug-objects to complain, but only
2353  * if CONFIG_DEBUG_OBJECTS_RCU_HEAD=y.  Otherwise, say that the test
2354  * cannot be carried out.
2355  */
2356 static void rcu_test_debug_objects(void)
2357 {
2358 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
2359 	struct rcu_head rh1;
2360 	struct rcu_head rh2;
2361 
2362 	init_rcu_head_on_stack(&rh1);
2363 	init_rcu_head_on_stack(&rh2);
2364 	pr_alert("%s: WARN: Duplicate call_rcu() test starting.\n", KBUILD_MODNAME);
2365 
2366 	/* Try to queue the rh2 pair of callbacks for the same grace period. */
2367 	preempt_disable(); /* Prevent preemption from interrupting test. */
2368 	rcu_read_lock(); /* Make it impossible to finish a grace period. */
2369 	call_rcu(&rh1, rcu_torture_leak_cb); /* Start grace period. */
2370 	local_irq_disable(); /* Make it harder to start a new grace period. */
2371 	call_rcu(&rh2, rcu_torture_leak_cb);
2372 	call_rcu(&rh2, rcu_torture_err_cb); /* Duplicate callback. */
2373 	local_irq_enable();
2374 	rcu_read_unlock();
2375 	preempt_enable();
2376 
2377 	/* Wait for them all to get done so we can safely return. */
2378 	rcu_barrier();
2379 	pr_alert("%s: WARN: Duplicate call_rcu() test complete.\n", KBUILD_MODNAME);
2380 	destroy_rcu_head_on_stack(&rh1);
2381 	destroy_rcu_head_on_stack(&rh2);
2382 #else /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
2383 	pr_alert("%s: !CONFIG_DEBUG_OBJECTS_RCU_HEAD, not testing duplicate call_rcu()\n", KBUILD_MODNAME);
2384 #endif /* #else #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
2385 }
2386 
2387 static void rcutorture_sync(void)
2388 {
2389 	static unsigned long n;
2390 
2391 	if (cur_ops->sync && !(++n & 0xfff))
2392 		cur_ops->sync();
2393 }
2394 
2395 static int __init
2396 rcu_torture_init(void)
2397 {
2398 	long i;
2399 	int cpu;
2400 	int firsterr = 0;
2401 	static struct rcu_torture_ops *torture_ops[] = {
2402 		&rcu_ops, &rcu_busted_ops, &srcu_ops, &srcud_ops,
2403 		&busted_srcud_ops, &tasks_ops, &trivial_ops,
2404 	};
2405 
2406 	if (!torture_init_begin(torture_type, verbose))
2407 		return -EBUSY;
2408 
2409 	/* Process args and tell the world that the torturer is on the job. */
2410 	for (i = 0; i < ARRAY_SIZE(torture_ops); i++) {
2411 		cur_ops = torture_ops[i];
2412 		if (strcmp(torture_type, cur_ops->name) == 0)
2413 			break;
2414 	}
2415 	if (i == ARRAY_SIZE(torture_ops)) {
2416 		pr_alert("rcu-torture: invalid torture type: \"%s\"\n",
2417 			 torture_type);
2418 		pr_alert("rcu-torture types:");
2419 		for (i = 0; i < ARRAY_SIZE(torture_ops); i++)
2420 			pr_cont(" %s", torture_ops[i]->name);
2421 		pr_cont("\n");
2422 		WARN_ON(!IS_MODULE(CONFIG_RCU_TORTURE_TEST));
2423 		firsterr = -EINVAL;
2424 		cur_ops = NULL;
2425 		goto unwind;
2426 	}
2427 	if (cur_ops->fqs == NULL && fqs_duration != 0) {
2428 		pr_alert("rcu-torture: ->fqs NULL and non-zero fqs_duration, fqs disabled.\n");
2429 		fqs_duration = 0;
2430 	}
2431 	if (cur_ops->init)
2432 		cur_ops->init();
2433 
2434 	if (nreaders >= 0) {
2435 		nrealreaders = nreaders;
2436 	} else {
2437 		nrealreaders = num_online_cpus() - 2 - nreaders;
2438 		if (nrealreaders <= 0)
2439 			nrealreaders = 1;
2440 	}
2441 	rcu_torture_print_module_parms(cur_ops, "Start of test");
2442 
2443 	/* Set up the freelist. */
2444 
2445 	INIT_LIST_HEAD(&rcu_torture_freelist);
2446 	for (i = 0; i < ARRAY_SIZE(rcu_tortures); i++) {
2447 		rcu_tortures[i].rtort_mbtest = 0;
2448 		list_add_tail(&rcu_tortures[i].rtort_free,
2449 			      &rcu_torture_freelist);
2450 	}
2451 
2452 	/* Initialize the statistics so that each run gets its own numbers. */
2453 
2454 	rcu_torture_current = NULL;
2455 	rcu_torture_current_version = 0;
2456 	atomic_set(&n_rcu_torture_alloc, 0);
2457 	atomic_set(&n_rcu_torture_alloc_fail, 0);
2458 	atomic_set(&n_rcu_torture_free, 0);
2459 	atomic_set(&n_rcu_torture_mberror, 0);
2460 	atomic_set(&n_rcu_torture_error, 0);
2461 	n_rcu_torture_barrier_error = 0;
2462 	n_rcu_torture_boost_ktrerror = 0;
2463 	n_rcu_torture_boost_rterror = 0;
2464 	n_rcu_torture_boost_failure = 0;
2465 	n_rcu_torture_boosts = 0;
2466 	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
2467 		atomic_set(&rcu_torture_wcount[i], 0);
2468 	for_each_possible_cpu(cpu) {
2469 		for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
2470 			per_cpu(rcu_torture_count, cpu)[i] = 0;
2471 			per_cpu(rcu_torture_batch, cpu)[i] = 0;
2472 		}
2473 	}
2474 	err_segs_recorded = 0;
2475 	rt_read_nsegs = 0;
2476 
2477 	/* Start up the kthreads. */
2478 
2479 	firsterr = torture_create_kthread(rcu_torture_writer, NULL,
2480 					  writer_task);
2481 	if (firsterr)
2482 		goto unwind;
2483 	if (nfakewriters > 0) {
2484 		fakewriter_tasks = kcalloc(nfakewriters,
2485 					   sizeof(fakewriter_tasks[0]),
2486 					   GFP_KERNEL);
2487 		if (fakewriter_tasks == NULL) {
2488 			VERBOSE_TOROUT_ERRSTRING("out of memory");
2489 			firsterr = -ENOMEM;
2490 			goto unwind;
2491 		}
2492 	}
2493 	for (i = 0; i < nfakewriters; i++) {
2494 		firsterr = torture_create_kthread(rcu_torture_fakewriter,
2495 						  NULL, fakewriter_tasks[i]);
2496 		if (firsterr)
2497 			goto unwind;
2498 	}
2499 	reader_tasks = kcalloc(nrealreaders, sizeof(reader_tasks[0]),
2500 			       GFP_KERNEL);
2501 	if (reader_tasks == NULL) {
2502 		VERBOSE_TOROUT_ERRSTRING("out of memory");
2503 		firsterr = -ENOMEM;
2504 		goto unwind;
2505 	}
2506 	for (i = 0; i < nrealreaders; i++) {
2507 		firsterr = torture_create_kthread(rcu_torture_reader, (void *)i,
2508 						  reader_tasks[i]);
2509 		if (firsterr)
2510 			goto unwind;
2511 	}
2512 	if (stat_interval > 0) {
2513 		firsterr = torture_create_kthread(rcu_torture_stats, NULL,
2514 						  stats_task);
2515 		if (firsterr)
2516 			goto unwind;
2517 	}
2518 	if (test_no_idle_hz && shuffle_interval > 0) {
2519 		firsterr = torture_shuffle_init(shuffle_interval * HZ);
2520 		if (firsterr)
2521 			goto unwind;
2522 	}
2523 	if (stutter < 0)
2524 		stutter = 0;
2525 	if (stutter) {
2526 		int t;
2527 
2528 		t = cur_ops->stall_dur ? cur_ops->stall_dur() : stutter * HZ;
2529 		firsterr = torture_stutter_init(stutter * HZ, t);
2530 		if (firsterr)
2531 			goto unwind;
2532 	}
2533 	if (fqs_duration < 0)
2534 		fqs_duration = 0;
2535 	if (fqs_duration) {
2536 		/* Create the fqs thread */
2537 		firsterr = torture_create_kthread(rcu_torture_fqs, NULL,
2538 						  fqs_task);
2539 		if (firsterr)
2540 			goto unwind;
2541 	}
2542 	if (test_boost_interval < 1)
2543 		test_boost_interval = 1;
2544 	if (test_boost_duration < 2)
2545 		test_boost_duration = 2;
2546 	if (rcu_torture_can_boost()) {
2547 
2548 		boost_starttime = jiffies + test_boost_interval * HZ;
2549 
2550 		firsterr = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "RCU_TORTURE",
2551 					     rcutorture_booster_init,
2552 					     rcutorture_booster_cleanup);
2553 		if (firsterr < 0)
2554 			goto unwind;
2555 		rcutor_hp = firsterr;
2556 	}
2557 	shutdown_jiffies = jiffies + shutdown_secs * HZ;
2558 	firsterr = torture_shutdown_init(shutdown_secs, rcu_torture_cleanup);
2559 	if (firsterr)
2560 		goto unwind;
2561 	firsterr = torture_onoff_init(onoff_holdoff * HZ, onoff_interval,
2562 				      rcutorture_sync);
2563 	if (firsterr)
2564 		goto unwind;
2565 	firsterr = rcu_torture_stall_init();
2566 	if (firsterr)
2567 		goto unwind;
2568 	firsterr = rcu_torture_fwd_prog_init();
2569 	if (firsterr)
2570 		goto unwind;
2571 	firsterr = rcu_torture_barrier_init();
2572 	if (firsterr)
2573 		goto unwind;
2574 	if (object_debug)
2575 		rcu_test_debug_objects();
2576 	torture_init_end();
2577 	return 0;
2578 
2579 unwind:
2580 	torture_init_end();
2581 	rcu_torture_cleanup();
2582 	return firsterr;
2583 }
2584 
2585 module_init(rcu_torture_init);
2586 module_exit(rcu_torture_cleanup);
2587