[linux-block.git] / lib / test_lockup.c

// SPDX-License-Identifier: GPL-2.0
/*
 * Test module to generate lockups
 */
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt

#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/delay.h>
#include <linux/sched.h>
#include <linux/sched/signal.h>
#include <linux/sched/clock.h>
#include <linux/cpu.h>
#include <linux/nmi.h>
#include <linux/mm.h>
#include <linux/uaccess.h>

static unsigned int time_secs;
module_param(time_secs, uint, 0600);
MODULE_PARM_DESC(time_secs, "lockup time in seconds, default 0");

static unsigned int time_nsecs;
module_param(time_nsecs, uint, 0600);
MODULE_PARM_DESC(time_nsecs, "nanoseconds part of lockup time, default 0");

static unsigned int cooldown_secs;
module_param(cooldown_secs, uint, 0600);
MODULE_PARM_DESC(cooldown_secs, "cooldown time between iterations in seconds, default 0");

static unsigned int cooldown_nsecs;
module_param(cooldown_nsecs, uint, 0600);
MODULE_PARM_DESC(cooldown_nsecs, "nanoseconds part of cooldown, default 0");

static unsigned int iterations = 1;
module_param(iterations, uint, 0600);
MODULE_PARM_DESC(iterations, "lockup iterations, default 1");

static bool all_cpus;
module_param(all_cpus, bool, 0400);
MODULE_PARM_DESC(all_cpus, "trigger lockup at all cpus at once");

static int wait_state;
static char *state = "R";
module_param(state, charp, 0400);
MODULE_PARM_DESC(state, "wait in 'R' running (default), 'D' uninterruptible, 'K' killable, 'S' interruptible state");

static bool use_hrtimer;
module_param(use_hrtimer, bool, 0400);
MODULE_PARM_DESC(use_hrtimer, "use high-resolution timer for sleeping");

static bool iowait;
module_param(iowait, bool, 0400);
MODULE_PARM_DESC(iowait, "account sleep time as iowait");

static bool lock_read;
module_param(lock_read, bool, 0400);
MODULE_PARM_DESC(lock_read, "lock read-write locks for read");

static bool lock_single;
module_param(lock_single, bool, 0400);
MODULE_PARM_DESC(lock_single, "acquire locks only at one cpu");

static bool reacquire_locks;
module_param(reacquire_locks, bool, 0400);
MODULE_PARM_DESC(reacquire_locks, "release and reacquire locks/irq/preempt between iterations");

static bool touch_softlockup;
module_param(touch_softlockup, bool, 0600);
MODULE_PARM_DESC(touch_softlockup, "touch soft-lockup watchdog between iterations");

static bool touch_hardlockup;
module_param(touch_hardlockup, bool, 0600);
MODULE_PARM_DESC(touch_hardlockup, "touch hard-lockup watchdog between iterations");

static bool call_cond_resched;
module_param(call_cond_resched, bool, 0600);
MODULE_PARM_DESC(call_cond_resched, "call cond_resched() between iterations");

static bool measure_lock_wait;
module_param(measure_lock_wait, bool, 0400);
MODULE_PARM_DESC(measure_lock_wait, "measure lock wait time");

static unsigned long lock_wait_threshold = ULONG_MAX;
module_param(lock_wait_threshold, ulong, 0400);
MODULE_PARM_DESC(lock_wait_threshold, "print lock wait time longer than this in nanoseconds, default off");

static bool test_disable_irq;
module_param_named(disable_irq, test_disable_irq, bool, 0400);
MODULE_PARM_DESC(disable_irq, "disable interrupts: generate hard-lockups");

static bool disable_softirq;
module_param(disable_softirq, bool, 0400);
MODULE_PARM_DESC(disable_softirq, "disable bottom-half irq handlers");

static bool disable_preempt;
module_param(disable_preempt, bool, 0400);
MODULE_PARM_DESC(disable_preempt, "disable preemption: generate soft-lockups");

static bool lock_rcu;
module_param(lock_rcu, bool, 0400);
MODULE_PARM_DESC(lock_rcu, "grab rcu_read_lock: generate rcu stalls");

static bool lock_mmap_sem;
module_param(lock_mmap_sem, bool, 0400);
MODULE_PARM_DESC(lock_mmap_sem, "lock mm->mmap_sem: block procfs interfaces");

static unsigned long lock_rwsem_ptr;
module_param_unsafe(lock_rwsem_ptr, ulong, 0400);
MODULE_PARM_DESC(lock_rwsem_ptr, "lock rw_semaphore at address");

static unsigned long lock_mutex_ptr;
module_param_unsafe(lock_mutex_ptr, ulong, 0400);
MODULE_PARM_DESC(lock_mutex_ptr, "lock mutex at address");

static unsigned long lock_spinlock_ptr;
module_param_unsafe(lock_spinlock_ptr, ulong, 0400);
MODULE_PARM_DESC(lock_spinlock_ptr, "lock spinlock at address");

static unsigned long lock_rwlock_ptr;
module_param_unsafe(lock_rwlock_ptr, ulong, 0400);
MODULE_PARM_DESC(lock_rwlock_ptr, "lock rwlock at address");

static unsigned int alloc_pages_nr;
module_param_unsafe(alloc_pages_nr, uint, 0600);
MODULE_PARM_DESC(alloc_pages_nr, "allocate and free pages under locks");

static unsigned int alloc_pages_order;
module_param(alloc_pages_order, uint, 0400);
MODULE_PARM_DESC(alloc_pages_order, "page order to allocate");

static gfp_t alloc_pages_gfp = GFP_KERNEL;
module_param_unsafe(alloc_pages_gfp, uint, 0400);
MODULE_PARM_DESC(alloc_pages_gfp, "allocate pages with this gfp_mask, default GFP_KERNEL");

static bool alloc_pages_atomic;
module_param(alloc_pages_atomic, bool, 0400);
MODULE_PARM_DESC(alloc_pages_atomic, "allocate pages with GFP_ATOMIC");

static bool reallocate_pages;
module_param(reallocate_pages, bool, 0400);
MODULE_PARM_DESC(reallocate_pages, "free and allocate pages between iterations");

static atomic_t alloc_pages_failed = ATOMIC_INIT(0);

static atomic64_t max_lock_wait = ATOMIC64_INIT(0);

static struct task_struct *main_task;
static int master_cpu;

static void test_lock(bool master, bool verbose)
{
	u64 uninitialized_var(wait_start);

	if (measure_lock_wait)
		wait_start = local_clock();

	if (lock_mutex_ptr && master) {
		if (verbose)
			pr_notice("lock mutex %ps\n", (void *)lock_mutex_ptr);
		mutex_lock((struct mutex *)lock_mutex_ptr);
	}

	if (lock_rwsem_ptr && master) {
		if (verbose)
			pr_notice("lock rw_semaphore %ps\n",
				  (void *)lock_rwsem_ptr);
		if (lock_read)
			down_read((struct rw_semaphore *)lock_rwsem_ptr);
		else
			down_write((struct rw_semaphore *)lock_rwsem_ptr);
	}

	if (lock_mmap_sem && master) {
		if (verbose)
			pr_notice("lock mmap_sem pid=%d\n", main_task->pid);
		if (lock_read)
			down_read(&main_task->mm->mmap_sem);
		else
			down_write(&main_task->mm->mmap_sem);
	}

	if (test_disable_irq)
		local_irq_disable();

	if (disable_softirq)
		local_bh_disable();

	if (disable_preempt)
		preempt_disable();

	if (lock_rcu)
		rcu_read_lock();

	if (lock_spinlock_ptr && master) {
		if (verbose)
			pr_notice("lock spinlock %ps\n",
				  (void *)lock_spinlock_ptr);
		spin_lock((spinlock_t *)lock_spinlock_ptr);
	}

	if (lock_rwlock_ptr && master) {
		if (verbose)
			pr_notice("lock rwlock %ps\n",
				  (void *)lock_rwlock_ptr);
		if (lock_read)
			read_lock((rwlock_t *)lock_rwlock_ptr);
		else
			write_lock((rwlock_t *)lock_rwlock_ptr);
	}

	if (measure_lock_wait) {
		s64 cur_wait = local_clock() - wait_start;
		s64 max_wait = atomic64_read(&max_lock_wait);

		do {
			if (cur_wait < max_wait)
				break;
			max_wait = atomic64_cmpxchg(&max_lock_wait,
						    max_wait, cur_wait);
		} while (max_wait != cur_wait);

		if (cur_wait > lock_wait_threshold)
			pr_notice_ratelimited("lock wait %lld ns\n", cur_wait);
	}
}

static void test_unlock(bool master, bool verbose)
{
	if (lock_rwlock_ptr && master) {
		if (lock_read)
			read_unlock((rwlock_t *)lock_rwlock_ptr);
		else
			write_unlock((rwlock_t *)lock_rwlock_ptr);
		if (verbose)
			pr_notice("unlock rwlock %ps\n",
				  (void *)lock_rwlock_ptr);
	}

	if (lock_spinlock_ptr && master) {
		spin_unlock((spinlock_t *)lock_spinlock_ptr);
		if (verbose)
			pr_notice("unlock spinlock %ps\n",
				  (void *)lock_spinlock_ptr);
	}

	if (lock_rcu)
		rcu_read_unlock();

	if (disable_preempt)
		preempt_enable();

	if (disable_softirq)
		local_bh_enable();

	if (test_disable_irq)
		local_irq_enable();

	if (lock_mmap_sem && master) {
		if (lock_read)
			up_read(&main_task->mm->mmap_sem);
		else
			up_write(&main_task->mm->mmap_sem);
		if (verbose)
			pr_notice("unlock mmap_sem pid=%d\n", main_task->pid);
	}

	if (lock_rwsem_ptr && master) {
		if (lock_read)
			up_read((struct rw_semaphore *)lock_rwsem_ptr);
		else
			up_write((struct rw_semaphore *)lock_rwsem_ptr);
		if (verbose)
			pr_notice("unlock rw_semaphore %ps\n",
				  (void *)lock_rwsem_ptr);
	}

	if (lock_mutex_ptr && master) {
		mutex_unlock((struct mutex *)lock_mutex_ptr);
		if (verbose)
			pr_notice("unlock mutex %ps\n",
				  (void *)lock_mutex_ptr);
	}
}

static void test_alloc_pages(struct list_head *pages)
{
	struct page *page;
	unsigned int i;

	for (i = 0; i < alloc_pages_nr; i++) {
		page = alloc_pages(alloc_pages_gfp, alloc_pages_order);
		if (!page) {
			atomic_inc(&alloc_pages_failed);
			break;
		}
		list_add(&page->lru, pages);
	}
}

static void test_free_pages(struct list_head *pages)
{
	struct page *page, *next;

	list_for_each_entry_safe(page, next, pages, lru)
		__free_pages(page, alloc_pages_order);
	INIT_LIST_HEAD(pages);
}

static void test_wait(unsigned int secs, unsigned int nsecs)
{
	if (wait_state == TASK_RUNNING) {
		if (secs)
			mdelay(secs * MSEC_PER_SEC);
		if (nsecs)
			ndelay(nsecs);
		return;
	}

	__set_current_state(wait_state);
	if (use_hrtimer) {
		ktime_t time;

		time = ns_to_ktime((u64)secs * NSEC_PER_SEC + nsecs);
		schedule_hrtimeout(&time, HRTIMER_MODE_REL);
	} else {
		schedule_timeout(secs * HZ + nsecs_to_jiffies(nsecs));
	}
}

static void test_lockup(bool master)
{
	u64 lockup_start = local_clock();
	unsigned int iter = 0;
	LIST_HEAD(pages);

	pr_notice("Start on CPU%d\n", raw_smp_processor_id());

	test_lock(master, true);

	test_alloc_pages(&pages);

	while (iter++ < iterations && !signal_pending(main_task)) {

		if (iowait)
			current->in_iowait = 1;

		test_wait(time_secs, time_nsecs);

		if (iowait)
			current->in_iowait = 0;

		if (reallocate_pages)
			test_free_pages(&pages);

		if (reacquire_locks)
			test_unlock(master, false);

		if (touch_softlockup)
			touch_softlockup_watchdog();

		if (touch_hardlockup)
			touch_nmi_watchdog();

		if (call_cond_resched)
			cond_resched();

		test_wait(cooldown_secs, cooldown_nsecs);

		if (reacquire_locks)
			test_lock(master, false);

		if (reallocate_pages)
			test_alloc_pages(&pages);
	}

	pr_notice("Finish on CPU%d in %lld ns\n", raw_smp_processor_id(),
		  local_clock() - lockup_start);

	test_free_pages(&pages);

	test_unlock(master, true);
}

DEFINE_PER_CPU(struct work_struct, test_works);

static void test_work_fn(struct work_struct *work)
{
	test_lockup(!lock_single ||
		    work == per_cpu_ptr(&test_works, master_cpu));
}

static bool test_kernel_ptr(unsigned long addr, int size)
{
	void *ptr = (void *)addr;
	char buf;

	if (!addr)
		return false;

	/* should be at least readable kernel address */
	if (access_ok(ptr, 1) ||
	    access_ok(ptr + size - 1, 1) ||
	    probe_kernel_address(ptr, buf) ||
	    probe_kernel_address(ptr + size - 1, buf)) {
		pr_err("invalid kernel ptr: %#lx\n", addr);
		return true;
	}

	return false;
}

static bool __maybe_unused test_magic(unsigned long addr, int offset,
				      unsigned int expected)
{
	void *ptr = (void *)addr + offset;
	unsigned int magic = 0;

	if (!addr)
		return false;

	if (probe_kernel_address(ptr, magic) || magic != expected) {
		pr_err("invalid magic at %#lx + %#x = %#x, expected %#x\n",
		       addr, offset, magic, expected);
		return true;
	}

	return false;
}

static int __init test_lockup_init(void)
{
	u64 test_start = local_clock();

	main_task = current;

	switch (state[0]) {
	case 'S':
		wait_state = TASK_INTERRUPTIBLE;
		break;
	case 'D':
		wait_state = TASK_UNINTERRUPTIBLE;
		break;
	case 'K':
		wait_state = TASK_KILLABLE;
		break;
	case 'R':
		wait_state = TASK_RUNNING;
		break;
	default:
		pr_err("unknown state=%s\n", state);
		return -EINVAL;
	}

	if (alloc_pages_atomic)
		alloc_pages_gfp = GFP_ATOMIC;

	if (test_kernel_ptr(lock_spinlock_ptr, sizeof(spinlock_t)) ||
	    test_kernel_ptr(lock_rwlock_ptr, sizeof(rwlock_t)) ||
	    test_kernel_ptr(lock_mutex_ptr, sizeof(struct mutex)) ||
	    test_kernel_ptr(lock_rwsem_ptr, sizeof(struct rw_semaphore)))
		return -EINVAL;

#ifdef CONFIG_DEBUG_SPINLOCK
	if (test_magic(lock_spinlock_ptr,
		       offsetof(spinlock_t, rlock.magic),
		       SPINLOCK_MAGIC) ||
	    test_magic(lock_rwlock_ptr,
		       offsetof(rwlock_t, magic),
		       RWLOCK_MAGIC) ||
	    test_magic(lock_mutex_ptr,
		       offsetof(struct mutex, wait_lock.rlock.magic),
		       SPINLOCK_MAGIC) ||
	    test_magic(lock_rwsem_ptr,
		       offsetof(struct rw_semaphore, wait_lock.magic),
		       SPINLOCK_MAGIC))
		return -EINVAL;
#endif

	if ((wait_state != TASK_RUNNING ||
	     (call_cond_resched && !reacquire_locks) ||
	     (alloc_pages_nr && gfpflags_allow_blocking(alloc_pages_gfp))) &&
	    (test_disable_irq || disable_softirq || disable_preempt ||
	     lock_rcu || lock_spinlock_ptr || lock_rwlock_ptr)) {
		pr_err("refuse to sleep in atomic context\n");
		return -EINVAL;
	}

	if (lock_mmap_sem && !main_task->mm) {
		pr_err("no mm to lock mmap_sem\n");
		return -EINVAL;
	}

	pr_notice("START pid=%d time=%u +%u ns cooldown=%u +%u ns iteraions=%u state=%s %s%s%s%s%s%s%s%s%s%s%s\n",
		  main_task->pid, time_secs, time_nsecs,
		  cooldown_secs, cooldown_nsecs, iterations, state,
		  all_cpus ? "all_cpus " : "",
		  iowait ? "iowait " : "",
		  test_disable_irq ? "disable_irq " : "",
		  disable_softirq ? "disable_softirq " : "",
		  disable_preempt ? "disable_preempt " : "",
		  lock_rcu ? "lock_rcu " : "",
		  lock_read ? "lock_read " : "",
		  touch_softlockup ? "touch_softlockup " : "",
		  touch_hardlockup ? "touch_hardlockup " : "",
		  call_cond_resched ? "call_cond_resched " : "",
		  reacquire_locks ? "reacquire_locks " : "");

	if (alloc_pages_nr)
		pr_notice("ALLOCATE PAGES nr=%u order=%u gfp=%pGg %s\n",
			  alloc_pages_nr, alloc_pages_order, &alloc_pages_gfp,
			  reallocate_pages ? "reallocate_pages " : "");

	if (all_cpus) {
		unsigned int cpu;

		cpus_read_lock();

		preempt_disable();
		master_cpu = smp_processor_id();
		for_each_online_cpu(cpu) {
			INIT_WORK(per_cpu_ptr(&test_works, cpu), test_work_fn);
			queue_work_on(cpu, system_highpri_wq,
				      per_cpu_ptr(&test_works, cpu));
		}
		preempt_enable();

		for_each_online_cpu(cpu)
			flush_work(per_cpu_ptr(&test_works, cpu));

		cpus_read_unlock();
	} else {
		test_lockup(true);
	}

	if (measure_lock_wait)
		pr_notice("Maximum lock wait: %lld ns\n",
			  atomic64_read(&max_lock_wait));

	if (alloc_pages_nr)
		pr_notice("Page allocation failed %u times\n",
			  atomic_read(&alloc_pages_failed));

	pr_notice("FINISH in %llu ns\n", local_clock() - test_start);

	if (signal_pending(main_task))
		return -EINTR;

	return -EAGAIN;
}
module_init(test_lockup_init);

MODULE_LICENSE("GPL");
MODULE_AUTHOR("Konstantin Khlebnikov <khlebnikov@yandex-team.ru>");
MODULE_DESCRIPTION("Test module to generate lockups");
Commit	Line	Data
30428ef5 KK	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
	18	static unsigned int time_secs;
	19	module_param(time_secs, uint, 0600);
	20	MODULE_PARM_DESC(time_secs, "lockup time in seconds, default 0");
	21
	22	static unsigned int time_nsecs;
	23	module_param(time_nsecs, uint, 0600);
	24	MODULE_PARM_DESC(time_nsecs, "nanoseconds part of lockup time, default 0");
	25
	26	static unsigned int cooldown_secs;
	27	module_param(cooldown_secs, uint, 0600);
	28	MODULE_PARM_DESC(cooldown_secs, "cooldown time between iterations in seconds, default 0");
	29
	30	static unsigned int cooldown_nsecs;
	31	module_param(cooldown_nsecs, uint, 0600);
	32	MODULE_PARM_DESC(cooldown_nsecs, "nanoseconds part of cooldown, default 0");
	33
	34	static unsigned int iterations = 1;
	35	module_param(iterations, uint, 0600);
	36	MODULE_PARM_DESC(iterations, "lockup iterations, default 1");
	37
	38	static bool all_cpus;
	39	module_param(all_cpus, bool, 0400);
	40	MODULE_PARM_DESC(all_cpus, "trigger lockup at all cpus at once");
	41
	42	static int wait_state;
	43	static char *state = "R";
	44	module_param(state, charp, 0400);
	45	MODULE_PARM_DESC(state, "wait in 'R' running (default), 'D' uninterruptible, 'K' killable, 'S' interruptible state");
	46
	47	static bool use_hrtimer;
	48	module_param(use_hrtimer, bool, 0400);
	49	MODULE_PARM_DESC(use_hrtimer, "use high-resolution timer for sleeping");
	50
	51	static bool iowait;
	52	module_param(iowait, bool, 0400);
	53	MODULE_PARM_DESC(iowait, "account sleep time as iowait");
	54
	55	static bool lock_read;
	56	module_param(lock_read, bool, 0400);
	57	MODULE_PARM_DESC(lock_read, "lock read-write locks for read");
	58
	59	static bool lock_single;
	60	module_param(lock_single, bool, 0400);
	61	MODULE_PARM_DESC(lock_single, "acquire locks only at one cpu");
	62
	63	static bool reacquire_locks;
	64	module_param(reacquire_locks, bool, 0400);
65	MODULE_PARM_DESC(reacquire_locks, "release and reacquire locks/irq/preempt between iterations");
66
67	static bool touch_softlockup;
68	module_param(touch_softlockup, bool, 0600);
69	MODULE_PARM_DESC(touch_softlockup, "touch soft-lockup watchdog between iterations");
70
71	static bool touch_hardlockup;
72	module_param(touch_hardlockup, bool, 0600);
73	MODULE_PARM_DESC(touch_hardlockup, "touch hard-lockup watchdog between iterations");
74
75	static bool call_cond_resched;
76	module_param(call_cond_resched, bool, 0600);
77	MODULE_PARM_DESC(call_cond_resched, "call cond_resched() between iterations");
78
79	static bool measure_lock_wait;
80	module_param(measure_lock_wait, bool, 0400);
81	MODULE_PARM_DESC(measure_lock_wait, "measure lock wait time");
82
83	static unsigned long lock_wait_threshold = ULONG_MAX;
84	module_param(lock_wait_threshold, ulong, 0400);
85	MODULE_PARM_DESC(lock_wait_threshold, "print lock wait time longer than this in nanoseconds, default off");
86
87	static bool test_disable_irq;
88	module_param_named(disable_irq, test_disable_irq, bool, 0400);
89	MODULE_PARM_DESC(disable_irq, "disable interrupts: generate hard-lockups");
90
91	static bool disable_softirq;
92	module_param(disable_softirq, bool, 0400);
93	MODULE_PARM_DESC(disable_softirq, "disable bottom-half irq handlers");
94
95	static bool disable_preempt;
96	module_param(disable_preempt, bool, 0400);
97	MODULE_PARM_DESC(disable_preempt, "disable preemption: generate soft-lockups");
98
99	static bool lock_rcu;
100	module_param(lock_rcu, bool, 0400);
101	MODULE_PARM_DESC(lock_rcu, "grab rcu_read_lock: generate rcu stalls");
102
103	static bool lock_mmap_sem;
104	module_param(lock_mmap_sem, bool, 0400);
105	MODULE_PARM_DESC(lock_mmap_sem, "lock mm->mmap_sem: block procfs interfaces");
106
107	static unsigned long lock_rwsem_ptr;
108	module_param_unsafe(lock_rwsem_ptr, ulong, 0400);
109	MODULE_PARM_DESC(lock_rwsem_ptr, "lock rw_semaphore at address");
110
111	static unsigned long lock_mutex_ptr;
112	module_param_unsafe(lock_mutex_ptr, ulong, 0400);
113	MODULE_PARM_DESC(lock_mutex_ptr, "lock mutex at address");
114
115	static unsigned long lock_spinlock_ptr;
116	module_param_unsafe(lock_spinlock_ptr, ulong, 0400);
117	MODULE_PARM_DESC(lock_spinlock_ptr, "lock spinlock at address");
118
119	static unsigned long lock_rwlock_ptr;
120	module_param_unsafe(lock_rwlock_ptr, ulong, 0400);
121	MODULE_PARM_DESC(lock_rwlock_ptr, "lock rwlock at address");
122
123	static unsigned int alloc_pages_nr;
124	module_param_unsafe(alloc_pages_nr, uint, 0600);
125	MODULE_PARM_DESC(alloc_pages_nr, "allocate and free pages under locks");
126
127	static unsigned int alloc_pages_order;
128	module_param(alloc_pages_order, uint, 0400);
129	MODULE_PARM_DESC(alloc_pages_order, "page order to allocate");
130
131	static gfp_t alloc_pages_gfp = GFP_KERNEL;
132	module_param_unsafe(alloc_pages_gfp, uint, 0400);
133	MODULE_PARM_DESC(alloc_pages_gfp, "allocate pages with this gfp_mask, default GFP_KERNEL");
134
135	static bool alloc_pages_atomic;
136	module_param(alloc_pages_atomic, bool, 0400);
137	MODULE_PARM_DESC(alloc_pages_atomic, "allocate pages with GFP_ATOMIC");
138
139	static bool reallocate_pages;
140	module_param(reallocate_pages, bool, 0400);
141	MODULE_PARM_DESC(reallocate_pages, "free and allocate pages between iterations");
142
143	static atomic_t alloc_pages_failed = ATOMIC_INIT(0);
144
145	static atomic64_t max_lock_wait = ATOMIC64_INIT(0);
146
147	static struct task_struct *main_task;
148	static int master_cpu;
149
150	static void test_lock(bool master, bool verbose)
151	{
152	u64 uninitialized_var(wait_start);
153
154	if (measure_lock_wait)
155	wait_start = local_clock();
156
157	if (lock_mutex_ptr && master) {
158	if (verbose)
159	pr_notice("lock mutex %ps\n", (void *)lock_mutex_ptr);
160	mutex_lock((struct mutex *)lock_mutex_ptr);
161	}
162
163	if (lock_rwsem_ptr && master) {
164	if (verbose)
165	pr_notice("lock rw_semaphore %ps\n",
166	(void *)lock_rwsem_ptr);
167	if (lock_read)
168	down_read((struct rw_semaphore *)lock_rwsem_ptr);
169	else
170	down_write((struct rw_semaphore *)lock_rwsem_ptr);
171	}
172
173	if (lock_mmap_sem && master) {
174	if (verbose)
175	pr_notice("lock mmap_sem pid=%d\n", main_task->pid);
176	if (lock_read)
177	down_read(&main_task->mm->mmap_sem);
178	else
179	down_write(&main_task->mm->mmap_sem);
180	}
181
182	if (test_disable_irq)
183	local_irq_disable();
184
185	if (disable_softirq)
186	local_bh_disable();
187
188	if (disable_preempt)
189	preempt_disable();
190
191	if (lock_rcu)
192	rcu_read_lock();
193
194	if (lock_spinlock_ptr && master) {
195	if (verbose)
196	pr_notice("lock spinlock %ps\n",
197	(void *)lock_spinlock_ptr);
198	spin_lock((spinlock_t *)lock_spinlock_ptr);
199	}
200
201	if (lock_rwlock_ptr && master) {
202	if (verbose)
203	pr_notice("lock rwlock %ps\n",
204	(void *)lock_rwlock_ptr);
205	if (lock_read)
206	read_lock((rwlock_t *)lock_rwlock_ptr);
207	else
208	write_lock((rwlock_t *)lock_rwlock_ptr);
209	}
210
211	if (measure_lock_wait) {
212	s64 cur_wait = local_clock() - wait_start;
213	s64 max_wait = atomic64_read(&max_lock_wait);
214
215	do {
216	if (cur_wait < max_wait)
217	break;
218	max_wait = atomic64_cmpxchg(&max_lock_wait,
219	max_wait, cur_wait);
220	} while (max_wait != cur_wait);
221
222	if (cur_wait > lock_wait_threshold)
223	pr_notice_ratelimited("lock wait %lld ns\n", cur_wait);
224	}
225	}
226
227	static void test_unlock(bool master, bool verbose)
228	{
229	if (lock_rwlock_ptr && master) {
230	if (lock_read)
231	read_unlock((rwlock_t *)lock_rwlock_ptr);
232	else
233	write_unlock((rwlock_t *)lock_rwlock_ptr);
234	if (verbose)
235	pr_notice("unlock rwlock %ps\n",
236	(void *)lock_rwlock_ptr);
237	}
238
239	if (lock_spinlock_ptr && master) {
240	spin_unlock((spinlock_t *)lock_spinlock_ptr);
241	if (verbose)
242	pr_notice("unlock spinlock %ps\n",
243	(void *)lock_spinlock_ptr);
244	}
245
246	if (lock_rcu)
247	rcu_read_unlock();
248
249	if (disable_preempt)
250	preempt_enable();
251
252	if (disable_softirq)
253	local_bh_enable();
254
255	if (test_disable_irq)
256	local_irq_enable();
257
258	if (lock_mmap_sem && master) {
259	if (lock_read)
260	up_read(&main_task->mm->mmap_sem);
261	else
262	up_write(&main_task->mm->mmap_sem);
263	if (verbose)
264	pr_notice("unlock mmap_sem pid=%d\n", main_task->pid);
265	}
266
267	if (lock_rwsem_ptr && master) {
268	if (lock_read)
269	up_read((struct rw_semaphore *)lock_rwsem_ptr);
270	else
271	up_write((struct rw_semaphore *)lock_rwsem_ptr);
272	if (verbose)
273	pr_notice("unlock rw_semaphore %ps\n",
274	(void *)lock_rwsem_ptr);
275	}
276
277	if (lock_mutex_ptr && master) {
278	mutex_unlock((struct mutex *)lock_mutex_ptr);
279	if (verbose)
280	pr_notice("unlock mutex %ps\n",
281	(void *)lock_mutex_ptr);
282	}
283	}
284
285	static void test_alloc_pages(struct list_head *pages)
286	{
287	struct page *page;
288	unsigned int i;
289
290	for (i = 0; i < alloc_pages_nr; i++) {
291	page = alloc_pages(alloc_pages_gfp, alloc_pages_order);
292	if (!page) {
293	atomic_inc(&alloc_pages_failed);
294	break;
295	}
296	list_add(&page->lru, pages);
297	}
298	}
299
300	static void test_free_pages(struct list_head *pages)
301	{
302	struct page page, next;
303
304	list_for_each_entry_safe(page, next, pages, lru)
305	__free_pages(page, alloc_pages_order);
306	INIT_LIST_HEAD(pages);
307	}
308
309	static void test_wait(unsigned int secs, unsigned int nsecs)
310	{
311	if (wait_state == TASK_RUNNING) {
312	if (secs)
313	mdelay(secs * MSEC_PER_SEC);
314	if (nsecs)
315	ndelay(nsecs);
316	return;
317	}
318
319	__set_current_state(wait_state);
320	if (use_hrtimer) {
321	ktime_t time;
322
323	time = ns_to_ktime((u64)secs * NSEC_PER_SEC + nsecs);
324	schedule_hrtimeout(&time, HRTIMER_MODE_REL);
325	} else {
326	schedule_timeout(secs * HZ + nsecs_to_jiffies(nsecs));
327	}
328	}
329
330	static void test_lockup(bool master)
331	{
332	u64 lockup_start = local_clock();
333	unsigned int iter = 0;
334	LIST_HEAD(pages);
335
336	pr_notice("Start on CPU%d\n", raw_smp_processor_id());
337
338	test_lock(master, true);
339
340	test_alloc_pages(&pages);
341
342	while (iter++ < iterations && !signal_pending(main_task)) {
343
344	if (iowait)
345	current->in_iowait = 1;
346
347	test_wait(time_secs, time_nsecs);
348
349	if (iowait)
350	current->in_iowait = 0;
351
352	if (reallocate_pages)
353	test_free_pages(&pages);
354
355	if (reacquire_locks)
356	test_unlock(master, false);
357
358	if (touch_softlockup)
359	touch_softlockup_watchdog();
360
361	if (touch_hardlockup)
362	touch_nmi_watchdog();
363
364	if (call_cond_resched)
365	cond_resched();
366
367	test_wait(cooldown_secs, cooldown_nsecs);
368
369	if (reacquire_locks)
370	test_lock(master, false);
371
372	if (reallocate_pages)
373	test_alloc_pages(&pages);
374	}
375
376	pr_notice("Finish on CPU%d in %lld ns\n", raw_smp_processor_id(),
377	local_clock() - lockup_start);
378
379	test_free_pages(&pages);
380
381	test_unlock(master, true);
382	}
383
384	DEFINE_PER_CPU(struct work_struct, test_works);
385
386	static void test_work_fn(struct work_struct *work)
387	{
388	test_lockup(!lock_single \|\|
389	work == per_cpu_ptr(&test_works, master_cpu));
390	}
391
392	static bool test_kernel_ptr(unsigned long addr, int size)
393	{
394	void ptr = (void )addr;
395	char buf;
396
397	if (!addr)
398	return false;
399
400	/* should be at least readable kernel address */
401	if (access_ok(ptr, 1) \|\|
402	access_ok(ptr + size - 1, 1) \|\|
403	probe_kernel_address(ptr, buf) \|\|
404	probe_kernel_address(ptr + size - 1, buf)) {
405	pr_err("invalid kernel ptr: %#lx\n", addr);
406	return true;
407	}
408
409	return false;
410	}
411
412	static bool __maybe_unused test_magic(unsigned long addr, int offset,
413	unsigned int expected)
414	{
415	void ptr = (void )addr + offset;
416	unsigned int magic = 0;
417
418	if (!addr)
419	return false;
420
421	if (probe_kernel_address(ptr, magic) \|\| magic != expected) {
422	pr_err("invalid magic at %#lx + %#x = %#x, expected %#x\n",
423	addr, offset, magic, expected);
424	return true;
425	}
426
427	return false;
428	}
429
430	static int __init test_lockup_init(void)
431	{
432	u64 test_start = local_clock();
433
434	main_task = current;
435
436	switch (state[0]) {
437	case 'S':
438	wait_state = TASK_INTERRUPTIBLE;
439	break;
440	case 'D':
441	wait_state = TASK_UNINTERRUPTIBLE;
442	break;
443	case 'K':
444	wait_state = TASK_KILLABLE;
445	break;
446	case 'R':
447	wait_state = TASK_RUNNING;
448	break;
449	default:
450	pr_err("unknown state=%s\n", state);
451	return -EINVAL;
452	}
453
454	if (alloc_pages_atomic)
455	alloc_pages_gfp = GFP_ATOMIC;
456
457	if (test_kernel_ptr(lock_spinlock_ptr, sizeof(spinlock_t)) \|\|
458	test_kernel_ptr(lock_rwlock_ptr, sizeof(rwlock_t)) \|\|
459	test_kernel_ptr(lock_mutex_ptr, sizeof(struct mutex)) \|\|
460	test_kernel_ptr(lock_rwsem_ptr, sizeof(struct rw_semaphore)))
461	return -EINVAL;
462
463	#ifdef CONFIG_DEBUG_SPINLOCK
464	if (test_magic(lock_spinlock_ptr,
465	offsetof(spinlock_t, rlock.magic),
466	SPINLOCK_MAGIC) \|\|
467	test_magic(lock_rwlock_ptr,
468	offsetof(rwlock_t, magic),
469	RWLOCK_MAGIC) \|\|
470	test_magic(lock_mutex_ptr,
471	offsetof(struct mutex, wait_lock.rlock.magic),
472	SPINLOCK_MAGIC) \|\|
473	test_magic(lock_rwsem_ptr,
474	offsetof(struct rw_semaphore, wait_lock.magic),
475	SPINLOCK_MAGIC))
476	return -EINVAL;
477	#endif
478
479	if ((wait_state != TASK_RUNNING \|\|
480	(call_cond_resched && !reacquire_locks) \|\|
481	(alloc_pages_nr && gfpflags_allow_blocking(alloc_pages_gfp))) &&
482	(test_disable_irq \|\| disable_softirq \|\| disable_preempt \|\|
483	lock_rcu \|\| lock_spinlock_ptr \|\| lock_rwlock_ptr)) {
484	pr_err("refuse to sleep in atomic context\n");
485	return -EINVAL;
486	}
487
488	if (lock_mmap_sem && !main_task->mm) {
489	pr_err("no mm to lock mmap_sem\n");
490	return -EINVAL;
491	}
492
493	pr_notice("START pid=%d time=%u +%u ns cooldown=%u +%u ns iteraions=%u state=%s %s%s%s%s%s%s%s%s%s%s%s\n",
494	main_task->pid, time_secs, time_nsecs,
495	cooldown_secs, cooldown_nsecs, iterations, state,
496	all_cpus ? "all_cpus " : "",
497	iowait ? "iowait " : "",
498	test_disable_irq ? "disable_irq " : "",
499	disable_softirq ? "disable_softirq " : "",
500	disable_preempt ? "disable_preempt " : "",
501	lock_rcu ? "lock_rcu " : "",
502	lock_read ? "lock_read " : "",
503	touch_softlockup ? "touch_softlockup " : "",
504	touch_hardlockup ? "touch_hardlockup " : "",
505	call_cond_resched ? "call_cond_resched " : "",
506	reacquire_locks ? "reacquire_locks " : "");
507
508	if (alloc_pages_nr)
509	pr_notice("ALLOCATE PAGES nr=%u order=%u gfp=%pGg %s\n",
510	alloc_pages_nr, alloc_pages_order, &alloc_pages_gfp,
511	reallocate_pages ? "reallocate_pages " : "");
512
513	if (all_cpus) {
514	unsigned int cpu;
515
516	cpus_read_lock();
517
518	preempt_disable();
519	master_cpu = smp_processor_id();
520	for_each_online_cpu(cpu) {
521	INIT_WORK(per_cpu_ptr(&test_works, cpu), test_work_fn);
522	queue_work_on(cpu, system_highpri_wq,
523	per_cpu_ptr(&test_works, cpu));
524	}
525	preempt_enable();
526
527	for_each_online_cpu(cpu)
528	flush_work(per_cpu_ptr(&test_works, cpu));
529
530	cpus_read_unlock();
531	} else {
532	test_lockup(true);
533	}
534
535	if (measure_lock_wait)
536	pr_notice("Maximum lock wait: %lld ns\n",
537	atomic64_read(&max_lock_wait));
538
539	if (alloc_pages_nr)
540	pr_notice("Page allocation failed %u times\n",
541	atomic_read(&alloc_pages_failed));
542
543	pr_notice("FINISH in %llu ns\n", local_clock() - test_start);
544
545	if (signal_pending(main_task))
546	return -EINTR;
547
548	return -EAGAIN;
549	}
550	module_init(test_lockup_init);
551
552	MODULE_LICENSE("GPL");
553	MODULE_AUTHOR("Konstantin Khlebnikov <khlebnikov@yandex-team.ru>");
554	MODULE_DESCRIPTION("Test module to generate lockups");