1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Test module to generate lockups
4 */
5#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
6
7#include <linux/kernel.h>
8#include <linux/module.h>
9#include <linux/delay.h>
10#include <linux/sched.h>
11#include <linux/sched/signal.h>
12#include <linux/sched/clock.h>
13#include <linux/cpu.h>
14#include <linux/nmi.h>
15#include <linux/mm.h>
16#include <linux/uaccess.h>
17#include <linux/file.h>
18
19static unsigned int time_secs;
20module_param(time_secs, uint, 0600);
21MODULE_PARM_DESC(time_secs, "lockup time in seconds, default 0");
22
23static unsigned int time_nsecs;
24module_param(time_nsecs, uint, 0600);
25MODULE_PARM_DESC(time_nsecs, "nanoseconds part of lockup time, default 0");
26
27static unsigned int cooldown_secs;
28module_param(cooldown_secs, uint, 0600);
29MODULE_PARM_DESC(cooldown_secs, "cooldown time between iterations in seconds, default 0");
30
31static unsigned int cooldown_nsecs;
32module_param(cooldown_nsecs, uint, 0600);
33MODULE_PARM_DESC(cooldown_nsecs, "nanoseconds part of cooldown, default 0");
34
35static unsigned int iterations = 1;
36module_param(iterations, uint, 0600);
37MODULE_PARM_DESC(iterations, "lockup iterations, default 1");
38
39static bool all_cpus;
40module_param(all_cpus, bool, 0400);
41MODULE_PARM_DESC(all_cpus, "trigger lockup at all cpus at once");
42
43static int wait_state;
44static char *state = "R";
45module_param(state, charp, 0400);
46MODULE_PARM_DESC(state, "wait in 'R' running (default), 'D' uninterruptible, 'K' killable, 'S' interruptible state");
47
48static bool use_hrtimer;
49module_param(use_hrtimer, bool, 0400);
50MODULE_PARM_DESC(use_hrtimer, "use high-resolution timer for sleeping");
51
52static bool iowait;
53module_param(iowait, bool, 0400);
54MODULE_PARM_DESC(iowait, "account sleep time as iowait");
55
56static bool lock_read;
57module_param(lock_read, bool, 0400);
58MODULE_PARM_DESC(lock_read, "lock read-write locks for read");
59
60static bool lock_single;
61module_param(lock_single, bool, 0400);
62MODULE_PARM_DESC(lock_single, "acquire locks only at one cpu");
63
64static bool reacquire_locks;
65module_param(reacquire_locks, bool, 0400);
66MODULE_PARM_DESC(reacquire_locks, "release and reacquire locks/irq/preempt between iterations");
67
68static bool touch_softlockup;
69module_param(touch_softlockup, bool, 0600);
70MODULE_PARM_DESC(touch_softlockup, "touch soft-lockup watchdog between iterations");
71
72static bool touch_hardlockup;
73module_param(touch_hardlockup, bool, 0600);
74MODULE_PARM_DESC(touch_hardlockup, "touch hard-lockup watchdog between iterations");
75
76static bool call_cond_resched;
77module_param(call_cond_resched, bool, 0600);
78MODULE_PARM_DESC(call_cond_resched, "call cond_resched() between iterations");
79
80static bool measure_lock_wait;
81module_param(measure_lock_wait, bool, 0400);
82MODULE_PARM_DESC(measure_lock_wait, "measure lock wait time");
83
84static unsigned long lock_wait_threshold = ULONG_MAX;
85module_param(lock_wait_threshold, ulong, 0400);
86MODULE_PARM_DESC(lock_wait_threshold, "print lock wait time longer than this in nanoseconds, default off");
87
88static bool test_disable_irq;
89module_param_named(disable_irq, test_disable_irq, bool, 0400);
90MODULE_PARM_DESC(disable_irq, "disable interrupts: generate hard-lockups");
91
92static bool disable_softirq;
93module_param(disable_softirq, bool, 0400);
94MODULE_PARM_DESC(disable_softirq, "disable bottom-half irq handlers");
95
96static bool disable_preempt;
97module_param(disable_preempt, bool, 0400);
98MODULE_PARM_DESC(disable_preempt, "disable preemption: generate soft-lockups");
99
100static bool lock_rcu;
101module_param(lock_rcu, bool, 0400);
102MODULE_PARM_DESC(lock_rcu, "grab rcu_read_lock: generate rcu stalls");
103
104static bool lock_mmap_sem;
105module_param(lock_mmap_sem, bool, 0400);
106MODULE_PARM_DESC(lock_mmap_sem, "lock mm->mmap_lock: block procfs interfaces");
107
108static unsigned long lock_rwsem_ptr;
109module_param_unsafe(lock_rwsem_ptr, ulong, 0400);
110MODULE_PARM_DESC(lock_rwsem_ptr, "lock rw_semaphore at address");
111
112static unsigned long lock_mutex_ptr;
113module_param_unsafe(lock_mutex_ptr, ulong, 0400);
114MODULE_PARM_DESC(lock_mutex_ptr, "lock mutex at address");
115
116static unsigned long lock_spinlock_ptr;
117module_param_unsafe(lock_spinlock_ptr, ulong, 0400);
118MODULE_PARM_DESC(lock_spinlock_ptr, "lock spinlock at address");
119
120static unsigned long lock_rwlock_ptr;
121module_param_unsafe(lock_rwlock_ptr, ulong, 0400);
122MODULE_PARM_DESC(lock_rwlock_ptr, "lock rwlock at address");
123
124static unsigned int alloc_pages_nr;
125module_param_unsafe(alloc_pages_nr, uint, 0600);
126MODULE_PARM_DESC(alloc_pages_nr, "allocate and free pages under locks");
127
128static unsigned int alloc_pages_order;
129module_param(alloc_pages_order, uint, 0400);
130MODULE_PARM_DESC(alloc_pages_order, "page order to allocate");
131
132static gfp_t alloc_pages_gfp = GFP_KERNEL;
133module_param_unsafe(alloc_pages_gfp, uint, 0400);
134MODULE_PARM_DESC(alloc_pages_gfp, "allocate pages with this gfp_mask, default GFP_KERNEL");
135
136static bool alloc_pages_atomic;
137module_param(alloc_pages_atomic, bool, 0400);
138MODULE_PARM_DESC(alloc_pages_atomic, "allocate pages with GFP_ATOMIC");
139
140static bool reallocate_pages;
141module_param(reallocate_pages, bool, 0400);
142MODULE_PARM_DESC(reallocate_pages, "free and allocate pages between iterations");
143
144struct file *test_file;
145static struct inode *test_inode;
146static char test_file_path[256];
147module_param_string(file_path, test_file_path, sizeof(test_file_path), 0400);
148MODULE_PARM_DESC(file_path, "file path to test");
149
150static bool test_lock_inode;
151module_param_named(lock_inode, test_lock_inode, bool, 0400);
152MODULE_PARM_DESC(lock_inode, "lock file -> inode -> i_rwsem");
153
154static bool test_lock_mapping;
155module_param_named(lock_mapping, test_lock_mapping, bool, 0400);
156MODULE_PARM_DESC(lock_mapping, "lock file -> mapping -> i_mmap_rwsem");
157
158static bool test_lock_sb_umount;
159module_param_named(lock_sb_umount, test_lock_sb_umount, bool, 0400);
160MODULE_PARM_DESC(lock_sb_umount, "lock file -> sb -> s_umount");
161
162static atomic_t alloc_pages_failed = ATOMIC_INIT(0);
163
164static atomic64_t max_lock_wait = ATOMIC64_INIT(0);
165
166static struct task_struct *main_task;
167static int master_cpu;
168
169static void test_lock(bool master, bool verbose)
170{
171	u64 wait_start;
172
173	if (measure_lock_wait)
174		wait_start = local_clock();
175
176	if (lock_mutex_ptr && master) {
177		if (verbose)
178			pr_notice("lock mutex %ps\n", (void *)lock_mutex_ptr);
179		mutex_lock((struct mutex *)lock_mutex_ptr);
180	}
181
182	if (lock_rwsem_ptr && master) {
183		if (verbose)
184			pr_notice("lock rw_semaphore %ps\n",
185				  (void *)lock_rwsem_ptr);
186		if (lock_read)
187			down_read((struct rw_semaphore *)lock_rwsem_ptr);
188		else
189			down_write((struct rw_semaphore *)lock_rwsem_ptr);
190	}
191
192	if (lock_mmap_sem && master) {
193		if (verbose)
194			pr_notice("lock mmap_lock pid=%d\n", main_task->pid);
195		if (lock_read)
196			mmap_read_lock(main_task->mm);
197		else
198			mmap_write_lock(main_task->mm);
199	}
200
201	if (test_disable_irq)
202		local_irq_disable();
203
204	if (disable_softirq)
205		local_bh_disable();
206
207	if (disable_preempt)
208		preempt_disable();
209
210	if (lock_rcu)
211		rcu_read_lock();
212
213	if (lock_spinlock_ptr && master) {
214		if (verbose)
215			pr_notice("lock spinlock %ps\n",
216				  (void *)lock_spinlock_ptr);
217		spin_lock((spinlock_t *)lock_spinlock_ptr);
218	}
219
220	if (lock_rwlock_ptr && master) {
221		if (verbose)
222			pr_notice("lock rwlock %ps\n",
223				  (void *)lock_rwlock_ptr);
224		if (lock_read)
225			read_lock((rwlock_t *)lock_rwlock_ptr);
226		else
227			write_lock((rwlock_t *)lock_rwlock_ptr);
228	}
229
230	if (measure_lock_wait) {
231		s64 cur_wait = local_clock() - wait_start;
232		s64 max_wait = atomic64_read(&max_lock_wait);
233
234		do {
235			if (cur_wait < max_wait)
236				break;
237			max_wait = atomic64_cmpxchg(&max_lock_wait,
238						    max_wait, cur_wait);
239		} while (max_wait != cur_wait);
240
241		if (cur_wait > lock_wait_threshold)
242			pr_notice_ratelimited("lock wait %lld ns\n", cur_wait);
243	}
244}
245
246static void test_unlock(bool master, bool verbose)
247{
248	if (lock_rwlock_ptr && master) {
249		if (lock_read)
250			read_unlock((rwlock_t *)lock_rwlock_ptr);
251		else
252			write_unlock((rwlock_t *)lock_rwlock_ptr);
253		if (verbose)
254			pr_notice("unlock rwlock %ps\n",
255				  (void *)lock_rwlock_ptr);
256	}
257
258	if (lock_spinlock_ptr && master) {
259		spin_unlock((spinlock_t *)lock_spinlock_ptr);
260		if (verbose)
261			pr_notice("unlock spinlock %ps\n",
262				  (void *)lock_spinlock_ptr);
263	}
264
265	if (lock_rcu)
266		rcu_read_unlock();
267
268	if (disable_preempt)
269		preempt_enable();
270
271	if (disable_softirq)
272		local_bh_enable();
273
274	if (test_disable_irq)
275		local_irq_enable();
276
277	if (lock_mmap_sem && master) {
278		if (lock_read)
279			mmap_read_unlock(main_task->mm);
280		else
281			mmap_write_unlock(main_task->mm);
282		if (verbose)
283			pr_notice("unlock mmap_lock pid=%d\n", main_task->pid);
284	}
285
286	if (lock_rwsem_ptr && master) {
287		if (lock_read)
288			up_read((struct rw_semaphore *)lock_rwsem_ptr);
289		else
290			up_write((struct rw_semaphore *)lock_rwsem_ptr);
291		if (verbose)
292			pr_notice("unlock rw_semaphore %ps\n",
293				  (void *)lock_rwsem_ptr);
294	}
295
296	if (lock_mutex_ptr && master) {
297		mutex_unlock((struct mutex *)lock_mutex_ptr);
298		if (verbose)
299			pr_notice("unlock mutex %ps\n",
300				  (void *)lock_mutex_ptr);
301	}
302}
303
304static void test_alloc_pages(struct list_head *pages)
305{
306	struct page *page;
307	unsigned int i;
308
309	for (i = 0; i < alloc_pages_nr; i++) {
310		page = alloc_pages(alloc_pages_gfp, alloc_pages_order);
311		if (!page) {
312			atomic_inc(&alloc_pages_failed);
313			break;
314		}
315		list_add(&page->lru, pages);
316	}
317}
318
319static void test_free_pages(struct list_head *pages)
320{
321	struct page *page, *next;
322
323	list_for_each_entry_safe(page, next, pages, lru)
324		__free_pages(page, alloc_pages_order);
325	INIT_LIST_HEAD(pages);
326}
327
328static void test_wait(unsigned int secs, unsigned int nsecs)
329{
330	if (wait_state == TASK_RUNNING) {
331		if (secs)
332			mdelay(secs * MSEC_PER_SEC);
333		if (nsecs)
334			ndelay(nsecs);
335		return;
336	}
337
338	__set_current_state(wait_state);
339	if (use_hrtimer) {
340		ktime_t time;
341
342		time = ns_to_ktime((u64)secs * NSEC_PER_SEC + nsecs);
343		schedule_hrtimeout(&time, HRTIMER_MODE_REL);
344	} else {
345		schedule_timeout(secs * HZ + nsecs_to_jiffies(nsecs));
346	}
347}
348
349static void test_lockup(bool master)
350{
351	u64 lockup_start = local_clock();
352	unsigned int iter = 0;
353	LIST_HEAD(pages);
354
355	pr_notice("Start on CPU%d\n", raw_smp_processor_id());
356
357	test_lock(master, true);
358
359	test_alloc_pages(&pages);
360
361	while (iter++ < iterations && !signal_pending(main_task)) {
362
363		if (iowait)
364			current->in_iowait = 1;
365
366		test_wait(time_secs, time_nsecs);
367
368		if (iowait)
369			current->in_iowait = 0;
370
371		if (reallocate_pages)
372			test_free_pages(&pages);
373
374		if (reacquire_locks)
375			test_unlock(master, false);
376
377		if (touch_softlockup)
378			touch_softlockup_watchdog();
379
380		if (touch_hardlockup)
381			touch_nmi_watchdog();
382
383		if (call_cond_resched)
384			cond_resched();
385
386		test_wait(cooldown_secs, cooldown_nsecs);
387
388		if (reacquire_locks)
389			test_lock(master, false);
390
391		if (reallocate_pages)
392			test_alloc_pages(&pages);
393	}
394
395	pr_notice("Finish on CPU%d in %lld ns\n", raw_smp_processor_id(),
396		  local_clock() - lockup_start);
397
398	test_free_pages(&pages);
399
400	test_unlock(master, true);
401}
402
403static DEFINE_PER_CPU(struct work_struct, test_works);
404
405static void test_work_fn(struct work_struct *work)
406{
407	test_lockup(!lock_single ||
408		    work == per_cpu_ptr(&test_works, master_cpu));
409}
410
411static bool test_kernel_ptr(unsigned long addr, int size)
412{
413	void *ptr = (void *)addr;
414	char buf;
415
416	if (!addr)
417		return false;
418
419	/* should be at least readable kernel address */
420	if (!IS_ENABLED(CONFIG_ALTERNATE_USER_ADDRESS_SPACE) &&
421	    (access_ok((void __user *)ptr, 1) ||
422	     access_ok((void __user *)ptr + size - 1, 1))) {
423		pr_err("user space ptr invalid in kernel: %#lx\n", addr);
424		return true;
425	}
426
427	if (get_kernel_nofault(buf, ptr) ||
428	    get_kernel_nofault(buf, ptr + size - 1)) {
429		pr_err("invalid kernel ptr: %#lx\n", addr);
430		return true;
431	}
432
433	return false;
434}
435
436static bool __maybe_unused test_magic(unsigned long addr, int offset,
437				      unsigned int expected)
438{
439	void *ptr = (void *)addr + offset;
440	unsigned int magic = 0;
441
442	if (!addr)
443		return false;
444
445	if (get_kernel_nofault(magic, ptr) || magic != expected) {
446		pr_err("invalid magic at %#lx + %#x = %#x, expected %#x\n",
447		       addr, offset, magic, expected);
448		return true;
449	}
450
451	return false;
452}
453
454static int __init test_lockup_init(void)
455{
456	u64 test_start = local_clock();
457
458	main_task = current;
459
460	switch (state[0]) {
461	case 'S':
462		wait_state = TASK_INTERRUPTIBLE;
463		break;
464	case 'D':
465		wait_state = TASK_UNINTERRUPTIBLE;
466		break;
467	case 'K':
468		wait_state = TASK_KILLABLE;
469		break;
470	case 'R':
471		wait_state = TASK_RUNNING;
472		break;
473	default:
474		pr_err("unknown state=%s\n", state);
475		return -EINVAL;
476	}
477
478	if (alloc_pages_atomic)
479		alloc_pages_gfp = GFP_ATOMIC;
480
481	if (test_kernel_ptr(lock_spinlock_ptr, sizeof(spinlock_t)) ||
482	    test_kernel_ptr(lock_rwlock_ptr, sizeof(rwlock_t)) ||
483	    test_kernel_ptr(lock_mutex_ptr, sizeof(struct mutex)) ||
484	    test_kernel_ptr(lock_rwsem_ptr, sizeof(struct rw_semaphore)))
485		return -EINVAL;
486
487#ifdef CONFIG_DEBUG_SPINLOCK
488#ifdef CONFIG_PREEMPT_RT
489	if (test_magic(lock_spinlock_ptr,
490		       offsetof(spinlock_t, lock.wait_lock.magic),
491		       SPINLOCK_MAGIC) ||
492	    test_magic(lock_rwlock_ptr,
493		       offsetof(rwlock_t, rwbase.rtmutex.wait_lock.magic),
494		       SPINLOCK_MAGIC) ||
495	    test_magic(lock_mutex_ptr,
496		       offsetof(struct mutex, rtmutex.wait_lock.magic),
497		       SPINLOCK_MAGIC) ||
498	    test_magic(lock_rwsem_ptr,
499		       offsetof(struct rw_semaphore, rwbase.rtmutex.wait_lock.magic),
500		       SPINLOCK_MAGIC))
501		return -EINVAL;
502#else
503	if (test_magic(lock_spinlock_ptr,
504		       offsetof(spinlock_t, rlock.magic),
505		       SPINLOCK_MAGIC) ||
506	    test_magic(lock_rwlock_ptr,
507		       offsetof(rwlock_t, magic),
508		       RWLOCK_MAGIC) ||
509	    test_magic(lock_mutex_ptr,
510		       offsetof(struct mutex, wait_lock.magic),
511		       SPINLOCK_MAGIC) ||
512	    test_magic(lock_rwsem_ptr,
513		       offsetof(struct rw_semaphore, wait_lock.magic),
514		       SPINLOCK_MAGIC))
515		return -EINVAL;
516#endif
517#endif
518
519	if ((wait_state != TASK_RUNNING ||
520	     (call_cond_resched && !reacquire_locks) ||
521	     (alloc_pages_nr && gfpflags_allow_blocking(alloc_pages_gfp))) &&
522	    (test_disable_irq || disable_softirq || disable_preempt ||
523	     lock_rcu || lock_spinlock_ptr || lock_rwlock_ptr)) {
524		pr_err("refuse to sleep in atomic context\n");
525		return -EINVAL;
526	}
527
528	if (lock_mmap_sem && !main_task->mm) {
529		pr_err("no mm to lock mmap_lock\n");
530		return -EINVAL;
531	}
532
533	if (test_file_path[0]) {
534		test_file = filp_open(test_file_path, O_RDONLY, 0);
535		if (IS_ERR(test_file)) {
536			pr_err("failed to open %s: %ld\n", test_file_path, PTR_ERR(test_file));
537			return PTR_ERR(test_file);
538		}
539		test_inode = file_inode(test_file);
540	} else if (test_lock_inode ||
541		   test_lock_mapping ||
542		   test_lock_sb_umount) {
543		pr_err("no file to lock\n");
544		return -EINVAL;
545	}
546
547	if (test_lock_inode && test_inode)
548		lock_rwsem_ptr = (unsigned long)&test_inode->i_rwsem;
549
550	if (test_lock_mapping && test_file && test_file->f_mapping)
551		lock_rwsem_ptr = (unsigned long)&test_file->f_mapping->i_mmap_rwsem;
552
553	if (test_lock_sb_umount && test_inode)
554		lock_rwsem_ptr = (unsigned long)&test_inode->i_sb->s_umount;
555
556	pr_notice("START pid=%d time=%u +%u ns cooldown=%u +%u ns iterations=%u state=%s %s%s%s%s%s%s%s%s%s%s%s\n",
557		  main_task->pid, time_secs, time_nsecs,
558		  cooldown_secs, cooldown_nsecs, iterations, state,
559		  all_cpus ? "all_cpus " : "",
560		  iowait ? "iowait " : "",
561		  test_disable_irq ? "disable_irq " : "",
562		  disable_softirq ? "disable_softirq " : "",
563		  disable_preempt ? "disable_preempt " : "",
564		  lock_rcu ? "lock_rcu " : "",
565		  lock_read ? "lock_read " : "",
566		  touch_softlockup ? "touch_softlockup " : "",
567		  touch_hardlockup ? "touch_hardlockup " : "",
568		  call_cond_resched ? "call_cond_resched " : "",
569		  reacquire_locks ? "reacquire_locks " : "");
570
571	if (alloc_pages_nr)
572		pr_notice("ALLOCATE PAGES nr=%u order=%u gfp=%pGg %s\n",
573			  alloc_pages_nr, alloc_pages_order, &alloc_pages_gfp,
574			  reallocate_pages ? "reallocate_pages " : "");
575
576	if (all_cpus) {
577		unsigned int cpu;
578
579		cpus_read_lock();
580
581		preempt_disable();
582		master_cpu = smp_processor_id();
583		for_each_online_cpu(cpu) {
584			INIT_WORK(per_cpu_ptr(&test_works, cpu), test_work_fn);
585			queue_work_on(cpu, system_highpri_wq,
586				      per_cpu_ptr(&test_works, cpu));
587		}
588		preempt_enable();
589
590		for_each_online_cpu(cpu)
591			flush_work(per_cpu_ptr(&test_works, cpu));
592
593		cpus_read_unlock();
594	} else {
595		test_lockup(true);
596	}
597
598	if (measure_lock_wait)
599		pr_notice("Maximum lock wait: %lld ns\n",
600			  atomic64_read(&max_lock_wait));
601
602	if (alloc_pages_nr)
603		pr_notice("Page allocation failed %u times\n",
604			  atomic_read(&alloc_pages_failed));
605
606	pr_notice("FINISH in %llu ns\n", local_clock() - test_start);
607
608	if (test_file)
609		fput(test_file);
610
611	if (signal_pending(main_task))
612		return -EINTR;
613
614	return -EAGAIN;
615}
616module_init(test_lockup_init);
617
618MODULE_LICENSE("GPL");
619MODULE_AUTHOR("Konstantin Khlebnikov <khlebnikov@yandex-team.ru>");
620MODULE_DESCRIPTION("Test module to generate lockups");
621