xref: /linux/kernel/locking/locktorture.c (revision 69050f8d6d075dc01af7a5f2f550a8067510366f)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * Module-based torture test facility for locking
4  *
5  * Copyright (C) IBM Corporation, 2014
6  *
7  * Authors: Paul E. McKenney <paulmck@linux.ibm.com>
8  *          Davidlohr Bueso <dave@stgolabs.net>
9  *	Based on kernel/rcu/torture.c.
10  */
11 
12 #define pr_fmt(fmt) fmt
13 
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/kthread.h>
17 #include <linux/sched/rt.h>
18 #include <linux/spinlock.h>
19 #include <linux/mutex.h>
20 #include <linux/rwsem.h>
21 #include <linux/smp.h>
22 #include <linux/interrupt.h>
23 #include <linux/sched.h>
24 #include <uapi/linux/sched/types.h>
25 #include <linux/rtmutex.h>
26 #include <linux/atomic.h>
27 #include <linux/moduleparam.h>
28 #include <linux/delay.h>
29 #include <linux/slab.h>
30 #include <linux/torture.h>
31 #include <linux/reboot.h>
32 
33 MODULE_DESCRIPTION("torture test facility for locking");
34 MODULE_LICENSE("GPL");
35 MODULE_AUTHOR("Paul E. McKenney <paulmck@linux.ibm.com>");
36 
37 torture_param(int, acq_writer_lim, 0, "Write_acquisition time limit (jiffies).");
38 torture_param(int, call_rcu_chains, 0, "Self-propagate call_rcu() chains during test (0=disable).");
39 torture_param(int, long_hold, 100, "Do occasional long hold of lock (ms), 0=disable");
40 torture_param(int, nested_locks, 0, "Number of nested locks (max = 8)");
41 torture_param(int, nreaders_stress, -1, "Number of read-locking stress-test threads");
42 torture_param(int, nwriters_stress, -1, "Number of write-locking stress-test threads");
43 torture_param(int, onoff_holdoff, 0, "Time after boot before CPU hotplugs (s)");
44 torture_param(int, onoff_interval, 0, "Time between CPU hotplugs (s), 0=disable");
45 torture_param(int, rt_boost, 2,
46 		   "Do periodic rt-boost. 0=Disable, 1=Only for rt_mutex, 2=For all lock types.");
47 torture_param(int, rt_boost_factor, 50, "A factor determining how often rt-boost happens.");
48 torture_param(int, shuffle_interval, 3, "Number of jiffies between shuffles, 0=disable");
49 torture_param(int, shutdown_secs, 0, "Shutdown time (j), <= zero to disable.");
50 torture_param(int, stat_interval, 60, "Number of seconds between stats printk()s");
51 torture_param(int, stutter, 5, "Number of jiffies to run/halt test, 0=disable");
52 torture_param(int, verbose, 1, "Enable verbose debugging printk()s");
53 torture_param(int, writer_fifo, 0, "Run writers at sched_set_fifo() priority");
54 /* Going much higher trips "BUG: MAX_LOCKDEP_CHAIN_HLOCKS too low!" errors */
55 #define MAX_NESTED_LOCKS 8
56 
57 static char *torture_type = IS_ENABLED(CONFIG_PREEMPT_RT) ? "raw_spin_lock" : "spin_lock";
58 module_param(torture_type, charp, 0444);
59 MODULE_PARM_DESC(torture_type,
60 		 "Type of lock to torture (spin_lock, spin_lock_irq, mutex_lock, ...)");
61 
62 static cpumask_var_t bind_readers; // Bind the readers to the specified set of CPUs.
63 static cpumask_var_t bind_writers; // Bind the writers to the specified set of CPUs.
64 
65 // Parse a cpumask kernel parameter.  If there are more users later on,
66 // this might need to got to a more central location.
67 static int param_set_cpumask(const char *val, const struct kernel_param *kp)
68 {
69 	cpumask_var_t *cm_bind = kp->arg;
70 	int ret;
71 	char *s;
72 
73 	if (!alloc_cpumask_var(cm_bind, GFP_KERNEL)) {
74 		s = "Out of memory";
75 		ret = -ENOMEM;
76 		goto out_err;
77 	}
78 	ret = cpulist_parse(val, *cm_bind);
79 	if (!ret)
80 		return ret;
81 	s = "Bad CPU range";
82 out_err:
83 	pr_warn("%s: %s, all CPUs set\n", kp->name, s);
84 	cpumask_setall(*cm_bind);
85 	return ret;
86 }
87 
88 // Output a cpumask kernel parameter.
89 static int param_get_cpumask(char *buffer, const struct kernel_param *kp)
90 {
91 	cpumask_var_t *cm_bind = kp->arg;
92 
93 	return sprintf(buffer, "%*pbl", cpumask_pr_args(*cm_bind));
94 }
95 
96 static bool cpumask_nonempty(cpumask_var_t mask)
97 {
98 	return cpumask_available(mask) && !cpumask_empty(mask);
99 }
100 
101 static const struct kernel_param_ops lt_bind_ops = {
102 	.set = param_set_cpumask,
103 	.get = param_get_cpumask,
104 };
105 
106 module_param_cb(bind_readers, &lt_bind_ops, &bind_readers, 0444);
107 module_param_cb(bind_writers, &lt_bind_ops, &bind_writers, 0444);
108 
109 long torture_sched_setaffinity(pid_t pid, const struct cpumask *in_mask, bool dowarn);
110 
111 static struct task_struct *stats_task;
112 static struct task_struct **writer_tasks;
113 static struct task_struct **reader_tasks;
114 
115 static bool lock_is_write_held;
116 static atomic_t lock_is_read_held;
117 static unsigned long last_lock_release;
118 
119 struct lock_stress_stats {
120 	long n_lock_fail;
121 	long n_lock_acquired;
122 };
123 
124 struct call_rcu_chain {
125 	struct rcu_head crc_rh;
126 	bool crc_stop;
127 };
128 struct call_rcu_chain *call_rcu_chain_list;
129 
130 /* Forward reference. */
131 static void lock_torture_cleanup(void);
132 
133 /*
134  * Operations vector for selecting different types of tests.
135  */
136 struct lock_torture_ops {
137 	void (*init)(void);
138 	void (*exit)(void);
139 	int (*nested_lock)(int tid, u32 lockset);
140 	int (*writelock)(int tid);
141 	void (*write_delay)(struct torture_random_state *trsp);
142 	void (*task_boost)(struct torture_random_state *trsp);
143 	void (*writeunlock)(int tid);
144 	void (*nested_unlock)(int tid, u32 lockset);
145 	int (*readlock)(int tid);
146 	void (*read_delay)(struct torture_random_state *trsp);
147 	void (*readunlock)(int tid);
148 
149 	unsigned long flags; /* for irq spinlocks */
150 	const char *name;
151 };
152 
153 struct lock_torture_cxt {
154 	int nrealwriters_stress;
155 	int nrealreaders_stress;
156 	bool debug_lock;
157 	bool init_called;
158 	atomic_t n_lock_torture_errors;
159 	struct lock_torture_ops *cur_ops;
160 	struct lock_stress_stats *lwsa; /* writer statistics */
161 	struct lock_stress_stats *lrsa; /* reader statistics */
162 };
163 static struct lock_torture_cxt cxt = { 0, 0, false, false,
164 				       ATOMIC_INIT(0),
165 				       NULL, NULL};
166 /*
167  * Definitions for lock torture testing.
168  */
169 
170 static int torture_lock_busted_write_lock(int tid __maybe_unused)
171 {
172 	return 0;  /* BUGGY, do not use in real life!!! */
173 }
174 
175 static void torture_lock_busted_write_delay(struct torture_random_state *trsp)
176 {
177 	/* We want a long delay occasionally to force massive contention.  */
178 	if (long_hold && !(torture_random(trsp) % (cxt.nrealwriters_stress * 2000 * long_hold)))
179 		mdelay(long_hold);
180 	if (!(torture_random(trsp) % (cxt.nrealwriters_stress * 20000)))
181 		torture_preempt_schedule();  /* Allow test to be preempted. */
182 }
183 
184 static void torture_lock_busted_write_unlock(int tid __maybe_unused)
185 {
186 	  /* BUGGY, do not use in real life!!! */
187 }
188 
189 static void __torture_rt_boost(struct torture_random_state *trsp)
190 {
191 	const unsigned int factor = rt_boost_factor;
192 
193 	if (!rt_task(current)) {
194 		/*
195 		 * Boost priority once every rt_boost_factor operations. When
196 		 * the task tries to take the lock, the rtmutex it will account
197 		 * for the new priority, and do any corresponding pi-dance.
198 		 */
199 		if (trsp && !(torture_random(trsp) %
200 			      (cxt.nrealwriters_stress * factor))) {
201 			sched_set_fifo(current);
202 		} else /* common case, do nothing */
203 			return;
204 	} else {
205 		/*
206 		 * The task will remain boosted for another 10 * rt_boost_factor
207 		 * operations, then restored back to its original prio, and so
208 		 * forth.
209 		 *
210 		 * When @trsp is nil, we want to force-reset the task for
211 		 * stopping the kthread.
212 		 */
213 		if (!trsp || !(torture_random(trsp) %
214 			       (cxt.nrealwriters_stress * factor * 2))) {
215 			sched_set_normal(current, 0);
216 		} else /* common case, do nothing */
217 			return;
218 	}
219 }
220 
221 static void torture_rt_boost(struct torture_random_state *trsp)
222 {
223 	if (rt_boost != 2)
224 		return;
225 
226 	__torture_rt_boost(trsp);
227 }
228 
229 static struct lock_torture_ops lock_busted_ops = {
230 	.writelock	= torture_lock_busted_write_lock,
231 	.write_delay	= torture_lock_busted_write_delay,
232 	.task_boost     = torture_rt_boost,
233 	.writeunlock	= torture_lock_busted_write_unlock,
234 	.readlock       = NULL,
235 	.read_delay     = NULL,
236 	.readunlock     = NULL,
237 	.name		= "lock_busted"
238 };
239 
240 static DEFINE_SPINLOCK(torture_spinlock);
241 
242 static int torture_spin_lock_write_lock(int tid __maybe_unused)
243 __acquires(torture_spinlock)
244 {
245 	spin_lock(&torture_spinlock);
246 	return 0;
247 }
248 
249 static void torture_spin_lock_write_delay(struct torture_random_state *trsp)
250 {
251 	const unsigned long shortdelay_us = 2;
252 	unsigned long j;
253 
254 	/* We want a short delay mostly to emulate likely code, and
255 	 * we want a long delay occasionally to force massive contention.
256 	 */
257 	if (long_hold && !(torture_random(trsp) % (cxt.nrealwriters_stress * 2000 * long_hold))) {
258 		j = jiffies;
259 		mdelay(long_hold);
260 		pr_alert("%s: delay = %lu jiffies.\n", __func__, jiffies - j);
261 	}
262 	if (!(torture_random(trsp) % (cxt.nrealwriters_stress * 200 * shortdelay_us)))
263 		udelay(shortdelay_us);
264 	if (!(torture_random(trsp) % (cxt.nrealwriters_stress * 20000)))
265 		torture_preempt_schedule();  /* Allow test to be preempted. */
266 }
267 
268 static void torture_spin_lock_write_unlock(int tid __maybe_unused)
269 __releases(torture_spinlock)
270 {
271 	spin_unlock(&torture_spinlock);
272 }
273 
274 static struct lock_torture_ops spin_lock_ops = {
275 	.writelock	= torture_spin_lock_write_lock,
276 	.write_delay	= torture_spin_lock_write_delay,
277 	.task_boost     = torture_rt_boost,
278 	.writeunlock	= torture_spin_lock_write_unlock,
279 	.readlock       = NULL,
280 	.read_delay     = NULL,
281 	.readunlock     = NULL,
282 	.name		= "spin_lock"
283 };
284 
285 static int torture_spin_lock_write_lock_irq(int tid __maybe_unused)
286 __acquires(torture_spinlock)
287 {
288 	unsigned long flags;
289 
290 	spin_lock_irqsave(&torture_spinlock, flags);
291 	cxt.cur_ops->flags = flags;
292 	return 0;
293 }
294 
295 static void torture_lock_spin_write_unlock_irq(int tid __maybe_unused)
296 __releases(torture_spinlock)
297 {
298 	spin_unlock_irqrestore(&torture_spinlock, cxt.cur_ops->flags);
299 }
300 
301 static struct lock_torture_ops spin_lock_irq_ops = {
302 	.writelock	= torture_spin_lock_write_lock_irq,
303 	.write_delay	= torture_spin_lock_write_delay,
304 	.task_boost     = torture_rt_boost,
305 	.writeunlock	= torture_lock_spin_write_unlock_irq,
306 	.readlock       = NULL,
307 	.read_delay     = NULL,
308 	.readunlock     = NULL,
309 	.name		= "spin_lock_irq"
310 };
311 
312 static DEFINE_RAW_SPINLOCK(torture_raw_spinlock);
313 
314 static int torture_raw_spin_lock_write_lock(int tid __maybe_unused)
315 __acquires(torture_raw_spinlock)
316 {
317 	raw_spin_lock(&torture_raw_spinlock);
318 	return 0;
319 }
320 
321 static void torture_raw_spin_lock_write_unlock(int tid __maybe_unused)
322 __releases(torture_raw_spinlock)
323 {
324 	raw_spin_unlock(&torture_raw_spinlock);
325 }
326 
327 static struct lock_torture_ops raw_spin_lock_ops = {
328 	.writelock	= torture_raw_spin_lock_write_lock,
329 	.write_delay	= torture_spin_lock_write_delay,
330 	.task_boost	= torture_rt_boost,
331 	.writeunlock	= torture_raw_spin_lock_write_unlock,
332 	.readlock	= NULL,
333 	.read_delay	= NULL,
334 	.readunlock	= NULL,
335 	.name		= "raw_spin_lock"
336 };
337 
338 static int torture_raw_spin_lock_write_lock_irq(int tid __maybe_unused)
339 __acquires(torture_raw_spinlock)
340 {
341 	unsigned long flags;
342 
343 	raw_spin_lock_irqsave(&torture_raw_spinlock, flags);
344 	cxt.cur_ops->flags = flags;
345 	return 0;
346 }
347 
348 static void torture_raw_spin_lock_write_unlock_irq(int tid __maybe_unused)
349 __releases(torture_raw_spinlock)
350 {
351 	raw_spin_unlock_irqrestore(&torture_raw_spinlock, cxt.cur_ops->flags);
352 }
353 
354 static struct lock_torture_ops raw_spin_lock_irq_ops = {
355 	.writelock	= torture_raw_spin_lock_write_lock_irq,
356 	.write_delay	= torture_spin_lock_write_delay,
357 	.task_boost	= torture_rt_boost,
358 	.writeunlock	= torture_raw_spin_lock_write_unlock_irq,
359 	.readlock	= NULL,
360 	.read_delay	= NULL,
361 	.readunlock	= NULL,
362 	.name		= "raw_spin_lock_irq"
363 };
364 
365 #ifdef CONFIG_BPF_SYSCALL
366 
367 #include <asm/rqspinlock.h>
368 static rqspinlock_t rqspinlock;
369 
370 static int torture_raw_res_spin_write_lock(int tid __maybe_unused)
371 {
372 	raw_res_spin_lock(&rqspinlock);
373 	return 0;
374 }
375 
376 static void torture_raw_res_spin_write_unlock(int tid __maybe_unused)
377 {
378 	raw_res_spin_unlock(&rqspinlock);
379 }
380 
381 static struct lock_torture_ops raw_res_spin_lock_ops = {
382 	.writelock	= torture_raw_res_spin_write_lock,
383 	.write_delay	= torture_spin_lock_write_delay,
384 	.task_boost     = torture_rt_boost,
385 	.writeunlock	= torture_raw_res_spin_write_unlock,
386 	.readlock       = NULL,
387 	.read_delay     = NULL,
388 	.readunlock     = NULL,
389 	.name		= "raw_res_spin_lock"
390 };
391 
392 static int torture_raw_res_spin_write_lock_irq(int tid __maybe_unused)
393 {
394 	unsigned long flags;
395 
396 	raw_res_spin_lock_irqsave(&rqspinlock, flags);
397 	cxt.cur_ops->flags = flags;
398 	return 0;
399 }
400 
401 static void torture_raw_res_spin_write_unlock_irq(int tid __maybe_unused)
402 {
403 	raw_res_spin_unlock_irqrestore(&rqspinlock, cxt.cur_ops->flags);
404 }
405 
406 static struct lock_torture_ops raw_res_spin_lock_irq_ops = {
407 	.writelock	= torture_raw_res_spin_write_lock_irq,
408 	.write_delay	= torture_spin_lock_write_delay,
409 	.task_boost     = torture_rt_boost,
410 	.writeunlock	= torture_raw_res_spin_write_unlock_irq,
411 	.readlock       = NULL,
412 	.read_delay     = NULL,
413 	.readunlock     = NULL,
414 	.name		= "raw_res_spin_lock_irq"
415 };
416 
417 #endif
418 
419 static DEFINE_RWLOCK(torture_rwlock);
420 
421 static int torture_rwlock_write_lock(int tid __maybe_unused)
422 __acquires(torture_rwlock)
423 {
424 	write_lock(&torture_rwlock);
425 	return 0;
426 }
427 
428 static void torture_rwlock_write_delay(struct torture_random_state *trsp)
429 {
430 	const unsigned long shortdelay_us = 2;
431 
432 	/* We want a short delay mostly to emulate likely code, and
433 	 * we want a long delay occasionally to force massive contention.
434 	 */
435 	if (long_hold && !(torture_random(trsp) % (cxt.nrealwriters_stress * 2000 * long_hold)))
436 		mdelay(long_hold);
437 	else
438 		udelay(shortdelay_us);
439 }
440 
441 static void torture_rwlock_write_unlock(int tid __maybe_unused)
442 __releases(torture_rwlock)
443 {
444 	write_unlock(&torture_rwlock);
445 }
446 
447 static int torture_rwlock_read_lock(int tid __maybe_unused)
448 __acquires(torture_rwlock)
449 {
450 	read_lock(&torture_rwlock);
451 	return 0;
452 }
453 
454 static void torture_rwlock_read_delay(struct torture_random_state *trsp)
455 {
456 	const unsigned long shortdelay_us = 10;
457 
458 	/* We want a short delay mostly to emulate likely code, and
459 	 * we want a long delay occasionally to force massive contention.
460 	 */
461 	if (long_hold && !(torture_random(trsp) % (cxt.nrealreaders_stress * 2000 * long_hold)))
462 		mdelay(long_hold);
463 	else
464 		udelay(shortdelay_us);
465 }
466 
467 static void torture_rwlock_read_unlock(int tid __maybe_unused)
468 __releases(torture_rwlock)
469 {
470 	read_unlock(&torture_rwlock);
471 }
472 
473 static struct lock_torture_ops rw_lock_ops = {
474 	.writelock	= torture_rwlock_write_lock,
475 	.write_delay	= torture_rwlock_write_delay,
476 	.task_boost     = torture_rt_boost,
477 	.writeunlock	= torture_rwlock_write_unlock,
478 	.readlock       = torture_rwlock_read_lock,
479 	.read_delay     = torture_rwlock_read_delay,
480 	.readunlock     = torture_rwlock_read_unlock,
481 	.name		= "rw_lock"
482 };
483 
484 static int torture_rwlock_write_lock_irq(int tid __maybe_unused)
485 __acquires(torture_rwlock)
486 {
487 	unsigned long flags;
488 
489 	write_lock_irqsave(&torture_rwlock, flags);
490 	cxt.cur_ops->flags = flags;
491 	return 0;
492 }
493 
494 static void torture_rwlock_write_unlock_irq(int tid __maybe_unused)
495 __releases(torture_rwlock)
496 {
497 	write_unlock_irqrestore(&torture_rwlock, cxt.cur_ops->flags);
498 }
499 
500 static int torture_rwlock_read_lock_irq(int tid __maybe_unused)
501 __acquires(torture_rwlock)
502 {
503 	unsigned long flags;
504 
505 	read_lock_irqsave(&torture_rwlock, flags);
506 	cxt.cur_ops->flags = flags;
507 	return 0;
508 }
509 
510 static void torture_rwlock_read_unlock_irq(int tid __maybe_unused)
511 __releases(torture_rwlock)
512 {
513 	read_unlock_irqrestore(&torture_rwlock, cxt.cur_ops->flags);
514 }
515 
516 static struct lock_torture_ops rw_lock_irq_ops = {
517 	.writelock	= torture_rwlock_write_lock_irq,
518 	.write_delay	= torture_rwlock_write_delay,
519 	.task_boost     = torture_rt_boost,
520 	.writeunlock	= torture_rwlock_write_unlock_irq,
521 	.readlock       = torture_rwlock_read_lock_irq,
522 	.read_delay     = torture_rwlock_read_delay,
523 	.readunlock     = torture_rwlock_read_unlock_irq,
524 	.name		= "rw_lock_irq"
525 };
526 
527 static DEFINE_MUTEX(torture_mutex);
528 static struct mutex torture_nested_mutexes[MAX_NESTED_LOCKS];
529 static struct lock_class_key nested_mutex_keys[MAX_NESTED_LOCKS];
530 
531 static void torture_mutex_init(void)
532 {
533 	int i;
534 
535 	for (i = 0; i < MAX_NESTED_LOCKS; i++)
536 		__mutex_init(&torture_nested_mutexes[i], __func__,
537 			     &nested_mutex_keys[i]);
538 }
539 
540 static int torture_mutex_nested_lock(int tid __maybe_unused,
541 				     u32 lockset)
542 {
543 	int i;
544 
545 	for (i = 0; i < nested_locks; i++)
546 		if (lockset & (1 << i))
547 			mutex_lock(&torture_nested_mutexes[i]);
548 	return 0;
549 }
550 
551 static int torture_mutex_lock(int tid __maybe_unused)
552 __acquires(torture_mutex)
553 {
554 	mutex_lock(&torture_mutex);
555 	return 0;
556 }
557 
558 static void torture_mutex_delay(struct torture_random_state *trsp)
559 {
560 	/* We want a long delay occasionally to force massive contention.  */
561 	if (long_hold && !(torture_random(trsp) % (cxt.nrealwriters_stress * 2000 * long_hold)))
562 		mdelay(long_hold * 5);
563 	if (!(torture_random(trsp) % (cxt.nrealwriters_stress * 20000)))
564 		torture_preempt_schedule();  /* Allow test to be preempted. */
565 }
566 
567 static void torture_mutex_unlock(int tid __maybe_unused)
568 __releases(torture_mutex)
569 {
570 	mutex_unlock(&torture_mutex);
571 }
572 
573 static void torture_mutex_nested_unlock(int tid __maybe_unused,
574 					u32 lockset)
575 {
576 	int i;
577 
578 	for (i = nested_locks - 1; i >= 0; i--)
579 		if (lockset & (1 << i))
580 			mutex_unlock(&torture_nested_mutexes[i]);
581 }
582 
583 static struct lock_torture_ops mutex_lock_ops = {
584 	.init		= torture_mutex_init,
585 	.nested_lock	= torture_mutex_nested_lock,
586 	.writelock	= torture_mutex_lock,
587 	.write_delay	= torture_mutex_delay,
588 	.task_boost     = torture_rt_boost,
589 	.writeunlock	= torture_mutex_unlock,
590 	.nested_unlock	= torture_mutex_nested_unlock,
591 	.readlock       = NULL,
592 	.read_delay     = NULL,
593 	.readunlock     = NULL,
594 	.name		= "mutex_lock"
595 };
596 
597 #include <linux/ww_mutex.h>
598 /*
599  * The torture ww_mutexes should belong to the same lock class as
600  * torture_ww_class to avoid lockdep problem. The ww_mutex_init()
601  * function is called for initialization to ensure that.
602  */
603 static DEFINE_WD_CLASS(torture_ww_class);
604 static struct ww_mutex torture_ww_mutex_0, torture_ww_mutex_1, torture_ww_mutex_2;
605 static struct ww_acquire_ctx *ww_acquire_ctxs;
606 
607 static void torture_ww_mutex_init(void)
608 {
609 	ww_mutex_init(&torture_ww_mutex_0, &torture_ww_class);
610 	ww_mutex_init(&torture_ww_mutex_1, &torture_ww_class);
611 	ww_mutex_init(&torture_ww_mutex_2, &torture_ww_class);
612 
613 	ww_acquire_ctxs = kmalloc_objs(*ww_acquire_ctxs,
614 				       cxt.nrealwriters_stress, GFP_KERNEL);
615 	if (!ww_acquire_ctxs)
616 		VERBOSE_TOROUT_STRING("ww_acquire_ctx: Out of memory");
617 }
618 
619 static void torture_ww_mutex_exit(void)
620 {
621 	kfree(ww_acquire_ctxs);
622 }
623 
624 static int torture_ww_mutex_lock(int tid)
625 __acquires(torture_ww_mutex_0)
626 __acquires(torture_ww_mutex_1)
627 __acquires(torture_ww_mutex_2)
628 {
629 	LIST_HEAD(list);
630 	struct reorder_lock {
631 		struct list_head link;
632 		struct ww_mutex *lock;
633 	} locks[3], *ll, *ln;
634 	struct ww_acquire_ctx *ctx = &ww_acquire_ctxs[tid];
635 
636 	locks[0].lock = &torture_ww_mutex_0;
637 	list_add(&locks[0].link, &list);
638 
639 	locks[1].lock = &torture_ww_mutex_1;
640 	list_add(&locks[1].link, &list);
641 
642 	locks[2].lock = &torture_ww_mutex_2;
643 	list_add(&locks[2].link, &list);
644 
645 	ww_acquire_init(ctx, &torture_ww_class);
646 
647 	list_for_each_entry(ll, &list, link) {
648 		int err;
649 
650 		err = ww_mutex_lock(ll->lock, ctx);
651 		if (!err)
652 			continue;
653 
654 		ln = ll;
655 		list_for_each_entry_continue_reverse(ln, &list, link)
656 			ww_mutex_unlock(ln->lock);
657 
658 		if (err != -EDEADLK)
659 			return err;
660 
661 		ww_mutex_lock_slow(ll->lock, ctx);
662 		list_move(&ll->link, &list);
663 	}
664 
665 	return 0;
666 }
667 
668 static void torture_ww_mutex_unlock(int tid)
669 __releases(torture_ww_mutex_0)
670 __releases(torture_ww_mutex_1)
671 __releases(torture_ww_mutex_2)
672 {
673 	struct ww_acquire_ctx *ctx = &ww_acquire_ctxs[tid];
674 
675 	ww_mutex_unlock(&torture_ww_mutex_0);
676 	ww_mutex_unlock(&torture_ww_mutex_1);
677 	ww_mutex_unlock(&torture_ww_mutex_2);
678 	ww_acquire_fini(ctx);
679 }
680 
681 static struct lock_torture_ops ww_mutex_lock_ops = {
682 	.init		= torture_ww_mutex_init,
683 	.exit		= torture_ww_mutex_exit,
684 	.writelock	= torture_ww_mutex_lock,
685 	.write_delay	= torture_mutex_delay,
686 	.task_boost     = torture_rt_boost,
687 	.writeunlock	= torture_ww_mutex_unlock,
688 	.readlock       = NULL,
689 	.read_delay     = NULL,
690 	.readunlock     = NULL,
691 	.name		= "ww_mutex_lock"
692 };
693 
694 #ifdef CONFIG_RT_MUTEXES
695 static DEFINE_RT_MUTEX(torture_rtmutex);
696 static struct rt_mutex torture_nested_rtmutexes[MAX_NESTED_LOCKS];
697 static struct lock_class_key nested_rtmutex_keys[MAX_NESTED_LOCKS];
698 
699 static void torture_rtmutex_init(void)
700 {
701 	int i;
702 
703 	for (i = 0; i < MAX_NESTED_LOCKS; i++)
704 		__rt_mutex_init(&torture_nested_rtmutexes[i], __func__,
705 				&nested_rtmutex_keys[i]);
706 }
707 
708 static int torture_rtmutex_nested_lock(int tid __maybe_unused,
709 				       u32 lockset)
710 {
711 	int i;
712 
713 	for (i = 0; i < nested_locks; i++)
714 		if (lockset & (1 << i))
715 			rt_mutex_lock(&torture_nested_rtmutexes[i]);
716 	return 0;
717 }
718 
719 static int torture_rtmutex_lock(int tid __maybe_unused)
720 __acquires(torture_rtmutex)
721 {
722 	rt_mutex_lock(&torture_rtmutex);
723 	return 0;
724 }
725 
726 static void torture_rtmutex_delay(struct torture_random_state *trsp)
727 {
728 	const unsigned long shortdelay_us = 2;
729 
730 	/*
731 	 * We want a short delay mostly to emulate likely code, and
732 	 * we want a long delay occasionally to force massive contention.
733 	 */
734 	if (long_hold && !(torture_random(trsp) % (cxt.nrealwriters_stress * 2000 * long_hold)))
735 		mdelay(long_hold);
736 	if (!(torture_random(trsp) %
737 	      (cxt.nrealwriters_stress * 200 * shortdelay_us)))
738 		udelay(shortdelay_us);
739 	if (!(torture_random(trsp) % (cxt.nrealwriters_stress * 20000)))
740 		torture_preempt_schedule();  /* Allow test to be preempted. */
741 }
742 
743 static void torture_rtmutex_unlock(int tid __maybe_unused)
744 __releases(torture_rtmutex)
745 {
746 	rt_mutex_unlock(&torture_rtmutex);
747 }
748 
749 static void torture_rt_boost_rtmutex(struct torture_random_state *trsp)
750 {
751 	if (!rt_boost)
752 		return;
753 
754 	__torture_rt_boost(trsp);
755 }
756 
757 static void torture_rtmutex_nested_unlock(int tid __maybe_unused,
758 					  u32 lockset)
759 {
760 	int i;
761 
762 	for (i = nested_locks - 1; i >= 0; i--)
763 		if (lockset & (1 << i))
764 			rt_mutex_unlock(&torture_nested_rtmutexes[i]);
765 }
766 
767 static struct lock_torture_ops rtmutex_lock_ops = {
768 	.init		= torture_rtmutex_init,
769 	.nested_lock	= torture_rtmutex_nested_lock,
770 	.writelock	= torture_rtmutex_lock,
771 	.write_delay	= torture_rtmutex_delay,
772 	.task_boost     = torture_rt_boost_rtmutex,
773 	.writeunlock	= torture_rtmutex_unlock,
774 	.nested_unlock	= torture_rtmutex_nested_unlock,
775 	.readlock       = NULL,
776 	.read_delay     = NULL,
777 	.readunlock     = NULL,
778 	.name		= "rtmutex_lock"
779 };
780 #endif
781 
782 static DECLARE_RWSEM(torture_rwsem);
783 static int torture_rwsem_down_write(int tid __maybe_unused)
784 __acquires(torture_rwsem)
785 {
786 	down_write(&torture_rwsem);
787 	return 0;
788 }
789 
790 static void torture_rwsem_write_delay(struct torture_random_state *trsp)
791 {
792 	/* We want a long delay occasionally to force massive contention.  */
793 	if (long_hold && !(torture_random(trsp) % (cxt.nrealwriters_stress * 2000 * long_hold)))
794 		mdelay(long_hold * 10);
795 	if (!(torture_random(trsp) % (cxt.nrealwriters_stress * 20000)))
796 		torture_preempt_schedule();  /* Allow test to be preempted. */
797 }
798 
799 static void torture_rwsem_up_write(int tid __maybe_unused)
800 __releases(torture_rwsem)
801 {
802 	up_write(&torture_rwsem);
803 }
804 
805 static int torture_rwsem_down_read(int tid __maybe_unused)
806 __acquires(torture_rwsem)
807 {
808 	down_read(&torture_rwsem);
809 	return 0;
810 }
811 
812 static void torture_rwsem_read_delay(struct torture_random_state *trsp)
813 {
814 	/* We want a long delay occasionally to force massive contention.  */
815 	if (long_hold && !(torture_random(trsp) % (cxt.nrealreaders_stress * 2000 * long_hold)))
816 		mdelay(long_hold * 2);
817 	else
818 		mdelay(long_hold / 2);
819 	if (!(torture_random(trsp) % (cxt.nrealreaders_stress * 20000)))
820 		torture_preempt_schedule();  /* Allow test to be preempted. */
821 }
822 
823 static void torture_rwsem_up_read(int tid __maybe_unused)
824 __releases(torture_rwsem)
825 {
826 	up_read(&torture_rwsem);
827 }
828 
829 static struct lock_torture_ops rwsem_lock_ops = {
830 	.writelock	= torture_rwsem_down_write,
831 	.write_delay	= torture_rwsem_write_delay,
832 	.task_boost     = torture_rt_boost,
833 	.writeunlock	= torture_rwsem_up_write,
834 	.readlock       = torture_rwsem_down_read,
835 	.read_delay     = torture_rwsem_read_delay,
836 	.readunlock     = torture_rwsem_up_read,
837 	.name		= "rwsem_lock"
838 };
839 
840 #include <linux/percpu-rwsem.h>
841 static struct percpu_rw_semaphore pcpu_rwsem;
842 
843 static void torture_percpu_rwsem_init(void)
844 {
845 	BUG_ON(percpu_init_rwsem(&pcpu_rwsem));
846 }
847 
848 static void torture_percpu_rwsem_exit(void)
849 {
850 	percpu_free_rwsem(&pcpu_rwsem);
851 }
852 
853 static int torture_percpu_rwsem_down_write(int tid __maybe_unused)
854 __acquires(pcpu_rwsem)
855 {
856 	percpu_down_write(&pcpu_rwsem);
857 	return 0;
858 }
859 
860 static void torture_percpu_rwsem_up_write(int tid __maybe_unused)
861 __releases(pcpu_rwsem)
862 {
863 	percpu_up_write(&pcpu_rwsem);
864 }
865 
866 static int torture_percpu_rwsem_down_read(int tid __maybe_unused)
867 __acquires(pcpu_rwsem)
868 {
869 	percpu_down_read(&pcpu_rwsem);
870 	return 0;
871 }
872 
873 static void torture_percpu_rwsem_up_read(int tid __maybe_unused)
874 __releases(pcpu_rwsem)
875 {
876 	percpu_up_read(&pcpu_rwsem);
877 }
878 
879 static struct lock_torture_ops percpu_rwsem_lock_ops = {
880 	.init		= torture_percpu_rwsem_init,
881 	.exit		= torture_percpu_rwsem_exit,
882 	.writelock	= torture_percpu_rwsem_down_write,
883 	.write_delay	= torture_rwsem_write_delay,
884 	.task_boost     = torture_rt_boost,
885 	.writeunlock	= torture_percpu_rwsem_up_write,
886 	.readlock       = torture_percpu_rwsem_down_read,
887 	.read_delay     = torture_rwsem_read_delay,
888 	.readunlock     = torture_percpu_rwsem_up_read,
889 	.name		= "percpu_rwsem_lock"
890 };
891 
892 /*
893  * Lock torture writer kthread.  Repeatedly acquires and releases
894  * the lock, checking for duplicate acquisitions.
895  */
896 static int lock_torture_writer(void *arg)
897 {
898 	unsigned long j;
899 	unsigned long j1;
900 	u32 lockset_mask;
901 	struct lock_stress_stats *lwsp = arg;
902 	DEFINE_TORTURE_RANDOM(rand);
903 	bool skip_main_lock;
904 	int tid = lwsp - cxt.lwsa;
905 
906 	VERBOSE_TOROUT_STRING("lock_torture_writer task started");
907 	if (!rt_task(current))
908 		set_user_nice(current, MAX_NICE);
909 
910 	do {
911 		if ((torture_random(&rand) & 0xfffff) == 0)
912 			schedule_timeout_uninterruptible(1);
913 
914 		lockset_mask = torture_random(&rand);
915 		/*
916 		 * When using nested_locks, we want to occasionally
917 		 * skip the main lock so we can avoid always serializing
918 		 * the lock chains on that central lock. By skipping the
919 		 * main lock occasionally, we can create different
920 		 * contention patterns (allowing for multiple disjoint
921 		 * blocked trees)
922 		 */
923 		skip_main_lock = (nested_locks &&
924 				 !(torture_random(&rand) % 100));
925 
926 		cxt.cur_ops->task_boost(&rand);
927 		if (cxt.cur_ops->nested_lock)
928 			cxt.cur_ops->nested_lock(tid, lockset_mask);
929 
930 		if (!skip_main_lock) {
931 			if (acq_writer_lim > 0)
932 				j = jiffies;
933 			cxt.cur_ops->writelock(tid);
934 			if (WARN_ON_ONCE(lock_is_write_held))
935 				lwsp->n_lock_fail++;
936 			lock_is_write_held = true;
937 			if (WARN_ON_ONCE(atomic_read(&lock_is_read_held)))
938 				lwsp->n_lock_fail++; /* rare, but... */
939 			if (acq_writer_lim > 0) {
940 				j1 = jiffies;
941 				WARN_ONCE(time_after(j1, j + acq_writer_lim),
942 					  "%s: Lock acquisition took %lu jiffies.\n",
943 					  __func__, j1 - j);
944 			}
945 			lwsp->n_lock_acquired++;
946 
947 			cxt.cur_ops->write_delay(&rand);
948 
949 			lock_is_write_held = false;
950 			WRITE_ONCE(last_lock_release, jiffies);
951 			cxt.cur_ops->writeunlock(tid);
952 		}
953 		if (cxt.cur_ops->nested_unlock)
954 			cxt.cur_ops->nested_unlock(tid, lockset_mask);
955 
956 		stutter_wait("lock_torture_writer");
957 	} while (!torture_must_stop());
958 
959 	cxt.cur_ops->task_boost(NULL); /* reset prio */
960 	torture_kthread_stopping("lock_torture_writer");
961 	return 0;
962 }
963 
964 /*
965  * Lock torture reader kthread.  Repeatedly acquires and releases
966  * the reader lock.
967  */
968 static int lock_torture_reader(void *arg)
969 {
970 	struct lock_stress_stats *lrsp = arg;
971 	int tid = lrsp - cxt.lrsa;
972 	DEFINE_TORTURE_RANDOM(rand);
973 
974 	VERBOSE_TOROUT_STRING("lock_torture_reader task started");
975 	set_user_nice(current, MAX_NICE);
976 
977 	do {
978 		if ((torture_random(&rand) & 0xfffff) == 0)
979 			schedule_timeout_uninterruptible(1);
980 
981 		cxt.cur_ops->readlock(tid);
982 		atomic_inc(&lock_is_read_held);
983 		if (WARN_ON_ONCE(lock_is_write_held))
984 			lrsp->n_lock_fail++; /* rare, but... */
985 
986 		lrsp->n_lock_acquired++;
987 		cxt.cur_ops->read_delay(&rand);
988 		atomic_dec(&lock_is_read_held);
989 		cxt.cur_ops->readunlock(tid);
990 
991 		stutter_wait("lock_torture_reader");
992 	} while (!torture_must_stop());
993 	torture_kthread_stopping("lock_torture_reader");
994 	return 0;
995 }
996 
997 /*
998  * Create an lock-torture-statistics message in the specified buffer.
999  */
1000 static void __torture_print_stats(char *page,
1001 				  struct lock_stress_stats *statp, bool write)
1002 {
1003 	long cur;
1004 	bool fail = false;
1005 	int i, n_stress;
1006 	long max = 0, min = statp ? data_race(statp[0].n_lock_acquired) : 0;
1007 	long long sum = 0;
1008 
1009 	n_stress = write ? cxt.nrealwriters_stress : cxt.nrealreaders_stress;
1010 	for (i = 0; i < n_stress; i++) {
1011 		if (data_race(statp[i].n_lock_fail))
1012 			fail = true;
1013 		cur = data_race(statp[i].n_lock_acquired);
1014 		sum += cur;
1015 		if (max < cur)
1016 			max = cur;
1017 		if (min > cur)
1018 			min = cur;
1019 	}
1020 	page += sprintf(page,
1021 			"%s:  Total: %lld  Max/Min: %ld/%ld %s  Fail: %d %s\n",
1022 			write ? "Writes" : "Reads ",
1023 			sum, max, min,
1024 			!onoff_interval && max / 2 > min ? "???" : "",
1025 			fail, fail ? "!!!" : "");
1026 	if (fail)
1027 		atomic_inc(&cxt.n_lock_torture_errors);
1028 }
1029 
1030 /*
1031  * Print torture statistics.  Caller must ensure that there is only one
1032  * call to this function at a given time!!!  This is normally accomplished
1033  * by relying on the module system to only have one copy of the module
1034  * loaded, and then by giving the lock_torture_stats kthread full control
1035  * (or the init/cleanup functions when lock_torture_stats thread is not
1036  * running).
1037  */
1038 static void lock_torture_stats_print(void)
1039 {
1040 	int size = cxt.nrealwriters_stress * 200 + 8192;
1041 	char *buf;
1042 
1043 	if (cxt.cur_ops->readlock)
1044 		size += cxt.nrealreaders_stress * 200 + 8192;
1045 
1046 	buf = kmalloc(size, GFP_KERNEL);
1047 	if (!buf) {
1048 		pr_err("lock_torture_stats_print: Out of memory, need: %d",
1049 		       size);
1050 		return;
1051 	}
1052 
1053 	__torture_print_stats(buf, cxt.lwsa, true);
1054 	pr_alert("%s", buf);
1055 	kfree(buf);
1056 
1057 	if (cxt.cur_ops->readlock) {
1058 		buf = kmalloc(size, GFP_KERNEL);
1059 		if (!buf) {
1060 			pr_err("lock_torture_stats_print: Out of memory, need: %d",
1061 			       size);
1062 			return;
1063 		}
1064 
1065 		__torture_print_stats(buf, cxt.lrsa, false);
1066 		pr_alert("%s", buf);
1067 		kfree(buf);
1068 	}
1069 }
1070 
1071 /*
1072  * Periodically prints torture statistics, if periodic statistics printing
1073  * was specified via the stat_interval module parameter.
1074  *
1075  * No need to worry about fullstop here, since this one doesn't reference
1076  * volatile state or register callbacks.
1077  */
1078 static int lock_torture_stats(void *arg)
1079 {
1080 	VERBOSE_TOROUT_STRING("lock_torture_stats task started");
1081 	do {
1082 		schedule_timeout_interruptible(stat_interval * HZ);
1083 		lock_torture_stats_print();
1084 		torture_shutdown_absorb("lock_torture_stats");
1085 	} while (!torture_must_stop());
1086 	torture_kthread_stopping("lock_torture_stats");
1087 	return 0;
1088 }
1089 
1090 
1091 static inline void
1092 lock_torture_print_module_parms(struct lock_torture_ops *cur_ops,
1093 				const char *tag)
1094 {
1095 	static cpumask_t cpumask_all;
1096 	cpumask_t *rcmp = cpumask_nonempty(bind_readers) ? bind_readers : &cpumask_all;
1097 	cpumask_t *wcmp = cpumask_nonempty(bind_writers) ? bind_writers : &cpumask_all;
1098 
1099 	cpumask_setall(&cpumask_all);
1100 	pr_alert("%s" TORTURE_FLAG
1101 		 "--- %s%s: acq_writer_lim=%d bind_readers=%*pbl bind_writers=%*pbl call_rcu_chains=%d long_hold=%d nested_locks=%d nreaders_stress=%d nwriters_stress=%d onoff_holdoff=%d onoff_interval=%d rt_boost=%d rt_boost_factor=%d shuffle_interval=%d shutdown_secs=%d stat_interval=%d stutter=%d verbose=%d writer_fifo=%d\n",
1102 		 torture_type, tag, cxt.debug_lock ? " [debug]": "",
1103 		 acq_writer_lim, cpumask_pr_args(rcmp), cpumask_pr_args(wcmp),
1104 		 call_rcu_chains, long_hold, nested_locks, cxt.nrealreaders_stress,
1105 		 cxt.nrealwriters_stress, onoff_holdoff, onoff_interval, rt_boost,
1106 		 rt_boost_factor, shuffle_interval, shutdown_secs, stat_interval, stutter,
1107 		 verbose, writer_fifo);
1108 }
1109 
1110 // If requested, maintain call_rcu() chains to keep a grace period always
1111 // in flight.  These increase the probability of getting an RCU CPU stall
1112 // warning and associated diagnostics when a locking primitive stalls.
1113 
1114 static void call_rcu_chain_cb(struct rcu_head *rhp)
1115 {
1116 	struct call_rcu_chain *crcp = container_of(rhp, struct call_rcu_chain, crc_rh);
1117 
1118 	if (!smp_load_acquire(&crcp->crc_stop)) {
1119 		(void)start_poll_synchronize_rcu(); // Start one grace period...
1120 		call_rcu(&crcp->crc_rh, call_rcu_chain_cb); // ... and later start another.
1121 	}
1122 }
1123 
1124 // Start the requested number of call_rcu() chains.
1125 static int call_rcu_chain_init(void)
1126 {
1127 	int i;
1128 
1129 	if (call_rcu_chains <= 0)
1130 		return 0;
1131 	call_rcu_chain_list = kzalloc_objs(*call_rcu_chain_list,
1132 					   call_rcu_chains, GFP_KERNEL);
1133 	if (!call_rcu_chain_list)
1134 		return -ENOMEM;
1135 	for (i = 0; i < call_rcu_chains; i++) {
1136 		call_rcu_chain_list[i].crc_stop = false;
1137 		call_rcu(&call_rcu_chain_list[i].crc_rh, call_rcu_chain_cb);
1138 	}
1139 	return 0;
1140 }
1141 
1142 // Stop all of the call_rcu() chains.
1143 static void call_rcu_chain_cleanup(void)
1144 {
1145 	int i;
1146 
1147 	if (!call_rcu_chain_list)
1148 		return;
1149 	for (i = 0; i < call_rcu_chains; i++)
1150 		smp_store_release(&call_rcu_chain_list[i].crc_stop, true);
1151 	rcu_barrier();
1152 	kfree(call_rcu_chain_list);
1153 	call_rcu_chain_list = NULL;
1154 }
1155 
1156 static void lock_torture_cleanup(void)
1157 {
1158 	int i;
1159 
1160 	if (torture_cleanup_begin())
1161 		return;
1162 
1163 	/*
1164 	 * Indicates early cleanup, meaning that the test has not run,
1165 	 * such as when passing bogus args when loading the module.
1166 	 * However cxt->cur_ops.init() may have been invoked, so beside
1167 	 * perform the underlying torture-specific cleanups, cur_ops.exit()
1168 	 * will be invoked if needed.
1169 	 */
1170 	if (!cxt.lwsa && !cxt.lrsa)
1171 		goto end;
1172 
1173 	if (writer_tasks) {
1174 		for (i = 0; i < cxt.nrealwriters_stress; i++)
1175 			torture_stop_kthread(lock_torture_writer, writer_tasks[i]);
1176 		kfree(writer_tasks);
1177 		writer_tasks = NULL;
1178 	}
1179 
1180 	if (reader_tasks) {
1181 		for (i = 0; i < cxt.nrealreaders_stress; i++)
1182 			torture_stop_kthread(lock_torture_reader,
1183 					     reader_tasks[i]);
1184 		kfree(reader_tasks);
1185 		reader_tasks = NULL;
1186 	}
1187 
1188 	torture_stop_kthread(lock_torture_stats, stats_task);
1189 	lock_torture_stats_print();  /* -After- the stats thread is stopped! */
1190 
1191 	if (atomic_read(&cxt.n_lock_torture_errors))
1192 		lock_torture_print_module_parms(cxt.cur_ops,
1193 						"End of test: FAILURE");
1194 	else if (torture_onoff_failures())
1195 		lock_torture_print_module_parms(cxt.cur_ops,
1196 						"End of test: LOCK_HOTPLUG");
1197 	else
1198 		lock_torture_print_module_parms(cxt.cur_ops,
1199 						"End of test: SUCCESS");
1200 
1201 	kfree(cxt.lwsa);
1202 	cxt.lwsa = NULL;
1203 	kfree(cxt.lrsa);
1204 	cxt.lrsa = NULL;
1205 
1206 	call_rcu_chain_cleanup();
1207 
1208 end:
1209 	if (cxt.init_called) {
1210 		if (cxt.cur_ops->exit)
1211 			cxt.cur_ops->exit();
1212 		cxt.init_called = false;
1213 	}
1214 
1215 	free_cpumask_var(bind_readers);
1216 	free_cpumask_var(bind_writers);
1217 
1218 	torture_cleanup_end();
1219 }
1220 
1221 static int __init lock_torture_init(void)
1222 {
1223 	int i, j;
1224 	int firsterr = 0;
1225 	static struct lock_torture_ops *torture_ops[] = {
1226 		&lock_busted_ops,
1227 		&spin_lock_ops, &spin_lock_irq_ops,
1228 		&raw_spin_lock_ops, &raw_spin_lock_irq_ops,
1229 #ifdef CONFIG_BPF_SYSCALL
1230 		&raw_res_spin_lock_ops, &raw_res_spin_lock_irq_ops,
1231 #endif
1232 		&rw_lock_ops, &rw_lock_irq_ops,
1233 		&mutex_lock_ops,
1234 		&ww_mutex_lock_ops,
1235 #ifdef CONFIG_RT_MUTEXES
1236 		&rtmutex_lock_ops,
1237 #endif
1238 		&rwsem_lock_ops,
1239 		&percpu_rwsem_lock_ops,
1240 	};
1241 
1242 	if (!torture_init_begin(torture_type, verbose))
1243 		return -EBUSY;
1244 
1245 	/* Process args and tell the world that the torturer is on the job. */
1246 	for (i = 0; i < ARRAY_SIZE(torture_ops); i++) {
1247 		cxt.cur_ops = torture_ops[i];
1248 		if (strcmp(torture_type, cxt.cur_ops->name) == 0)
1249 			break;
1250 	}
1251 	if (i == ARRAY_SIZE(torture_ops)) {
1252 		pr_alert("lock-torture: invalid torture type: \"%s\"\n",
1253 			 torture_type);
1254 		pr_alert("lock-torture types:");
1255 		for (i = 0; i < ARRAY_SIZE(torture_ops); i++)
1256 			pr_alert(" %s", torture_ops[i]->name);
1257 		pr_alert("\n");
1258 		firsterr = -EINVAL;
1259 		goto unwind;
1260 	}
1261 
1262 	if (nwriters_stress == 0 &&
1263 	    (!cxt.cur_ops->readlock || nreaders_stress == 0)) {
1264 		pr_alert("lock-torture: must run at least one locking thread\n");
1265 		firsterr = -EINVAL;
1266 		goto unwind;
1267 	}
1268 
1269 	if (nwriters_stress >= 0)
1270 		cxt.nrealwriters_stress = nwriters_stress;
1271 	else
1272 		cxt.nrealwriters_stress = 2 * num_online_cpus();
1273 
1274 	if (cxt.cur_ops->init) {
1275 		cxt.cur_ops->init();
1276 		cxt.init_called = true;
1277 	}
1278 
1279 #ifdef CONFIG_DEBUG_MUTEXES
1280 	if (str_has_prefix(torture_type, "mutex"))
1281 		cxt.debug_lock = true;
1282 #endif
1283 #ifdef CONFIG_DEBUG_RT_MUTEXES
1284 	if (str_has_prefix(torture_type, "rtmutex"))
1285 		cxt.debug_lock = true;
1286 #endif
1287 #ifdef CONFIG_DEBUG_SPINLOCK
1288 	if ((str_has_prefix(torture_type, "spin")) ||
1289 	    (str_has_prefix(torture_type, "rw_lock")))
1290 		cxt.debug_lock = true;
1291 #endif
1292 
1293 	/* Initialize the statistics so that each run gets its own numbers. */
1294 	if (nwriters_stress) {
1295 		lock_is_write_held = false;
1296 		cxt.lwsa = kmalloc_objs(*cxt.lwsa, cxt.nrealwriters_stress,
1297 					GFP_KERNEL);
1298 		if (cxt.lwsa == NULL) {
1299 			VERBOSE_TOROUT_STRING("cxt.lwsa: Out of memory");
1300 			firsterr = -ENOMEM;
1301 			goto unwind;
1302 		}
1303 
1304 		for (i = 0; i < cxt.nrealwriters_stress; i++) {
1305 			cxt.lwsa[i].n_lock_fail = 0;
1306 			cxt.lwsa[i].n_lock_acquired = 0;
1307 		}
1308 	}
1309 
1310 	if (cxt.cur_ops->readlock) {
1311 		if (nreaders_stress >= 0)
1312 			cxt.nrealreaders_stress = nreaders_stress;
1313 		else {
1314 			/*
1315 			 * By default distribute evenly the number of
1316 			 * readers and writers. We still run the same number
1317 			 * of threads as the writer-only locks default.
1318 			 */
1319 			if (nwriters_stress < 0) /* user doesn't care */
1320 				cxt.nrealwriters_stress = num_online_cpus();
1321 			cxt.nrealreaders_stress = cxt.nrealwriters_stress;
1322 		}
1323 
1324 		if (nreaders_stress) {
1325 			cxt.lrsa = kmalloc_objs(*cxt.lrsa,
1326 						cxt.nrealreaders_stress,
1327 						GFP_KERNEL);
1328 			if (cxt.lrsa == NULL) {
1329 				VERBOSE_TOROUT_STRING("cxt.lrsa: Out of memory");
1330 				firsterr = -ENOMEM;
1331 				kfree(cxt.lwsa);
1332 				cxt.lwsa = NULL;
1333 				goto unwind;
1334 			}
1335 
1336 			for (i = 0; i < cxt.nrealreaders_stress; i++) {
1337 				cxt.lrsa[i].n_lock_fail = 0;
1338 				cxt.lrsa[i].n_lock_acquired = 0;
1339 			}
1340 		}
1341 	}
1342 
1343 	firsterr = call_rcu_chain_init();
1344 	if (torture_init_error(firsterr))
1345 		goto unwind;
1346 
1347 	lock_torture_print_module_parms(cxt.cur_ops, "Start of test");
1348 
1349 	/* Prepare torture context. */
1350 	if (onoff_interval > 0) {
1351 		firsterr = torture_onoff_init(onoff_holdoff * HZ,
1352 					      onoff_interval * HZ, NULL);
1353 		if (torture_init_error(firsterr))
1354 			goto unwind;
1355 	}
1356 	if (shuffle_interval > 0) {
1357 		firsterr = torture_shuffle_init(shuffle_interval);
1358 		if (torture_init_error(firsterr))
1359 			goto unwind;
1360 	}
1361 	if (shutdown_secs > 0) {
1362 		firsterr = torture_shutdown_init(shutdown_secs,
1363 						 lock_torture_cleanup);
1364 		if (torture_init_error(firsterr))
1365 			goto unwind;
1366 	}
1367 	if (stutter > 0) {
1368 		firsterr = torture_stutter_init(stutter, stutter);
1369 		if (torture_init_error(firsterr))
1370 			goto unwind;
1371 	}
1372 
1373 	if (nwriters_stress) {
1374 		writer_tasks = kzalloc_objs(writer_tasks[0],
1375 					    cxt.nrealwriters_stress, GFP_KERNEL);
1376 		if (writer_tasks == NULL) {
1377 			TOROUT_ERRSTRING("writer_tasks: Out of memory");
1378 			firsterr = -ENOMEM;
1379 			goto unwind;
1380 		}
1381 	}
1382 
1383 	/* cap nested_locks to MAX_NESTED_LOCKS */
1384 	if (nested_locks > MAX_NESTED_LOCKS)
1385 		nested_locks = MAX_NESTED_LOCKS;
1386 
1387 	if (cxt.cur_ops->readlock) {
1388 		reader_tasks = kzalloc_objs(reader_tasks[0],
1389 					    cxt.nrealreaders_stress, GFP_KERNEL);
1390 		if (reader_tasks == NULL) {
1391 			TOROUT_ERRSTRING("reader_tasks: Out of memory");
1392 			kfree(writer_tasks);
1393 			writer_tasks = NULL;
1394 			firsterr = -ENOMEM;
1395 			goto unwind;
1396 		}
1397 	}
1398 
1399 	/*
1400 	 * Create the kthreads and start torturing (oh, those poor little locks).
1401 	 *
1402 	 * TODO: Note that we interleave writers with readers, giving writers a
1403 	 * slight advantage, by creating its kthread first. This can be modified
1404 	 * for very specific needs, or even let the user choose the policy, if
1405 	 * ever wanted.
1406 	 */
1407 	for (i = 0, j = 0; i < cxt.nrealwriters_stress ||
1408 		    j < cxt.nrealreaders_stress; i++, j++) {
1409 		if (i >= cxt.nrealwriters_stress)
1410 			goto create_reader;
1411 
1412 		/* Create writer. */
1413 		firsterr = torture_create_kthread_cb(lock_torture_writer, &cxt.lwsa[i],
1414 						     writer_tasks[i],
1415 						     writer_fifo ? sched_set_fifo : NULL);
1416 		if (torture_init_error(firsterr))
1417 			goto unwind;
1418 		if (cpumask_nonempty(bind_writers))
1419 			torture_sched_setaffinity(writer_tasks[i]->pid, bind_writers, true);
1420 
1421 	create_reader:
1422 		if (cxt.cur_ops->readlock == NULL || (j >= cxt.nrealreaders_stress))
1423 			continue;
1424 		/* Create reader. */
1425 		firsterr = torture_create_kthread(lock_torture_reader, &cxt.lrsa[j],
1426 						  reader_tasks[j]);
1427 		if (torture_init_error(firsterr))
1428 			goto unwind;
1429 		if (cpumask_nonempty(bind_readers))
1430 			torture_sched_setaffinity(reader_tasks[j]->pid, bind_readers, true);
1431 	}
1432 	if (stat_interval > 0) {
1433 		firsterr = torture_create_kthread(lock_torture_stats, NULL,
1434 						  stats_task);
1435 		if (torture_init_error(firsterr))
1436 			goto unwind;
1437 	}
1438 	torture_init_end();
1439 	return 0;
1440 
1441 unwind:
1442 	torture_init_end();
1443 	lock_torture_cleanup();
1444 	if (shutdown_secs) {
1445 		WARN_ON(!IS_MODULE(CONFIG_LOCK_TORTURE_TEST));
1446 		kernel_power_off();
1447 	}
1448 	return firsterr;
1449 }
1450 
1451 module_init(lock_torture_init);
1452 module_exit(lock_torture_cleanup);
1453