[linux.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>
#include <linux/file.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_lock: 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");

struct file *test_file;
static struct inode *test_inode;
static char test_file_path[256];
module_param_string(file_path, test_file_path, sizeof(test_file_path), 0400);
MODULE_PARM_DESC(file_path, "file path to test");

static bool test_lock_inode;
module_param_named(lock_inode, test_lock_inode, bool, 0400);
MODULE_PARM_DESC(lock_inode, "lock file -> inode -> i_rwsem");

static bool test_lock_mapping;
module_param_named(lock_mapping, test_lock_mapping, bool, 0400);
MODULE_PARM_DESC(lock_mapping, "lock file -> mapping -> i_mmap_rwsem");

static bool test_lock_sb_umount;
module_param_named(lock_sb_umount, test_lock_sb_umount, bool, 0400);
MODULE_PARM_DESC(lock_sb_umount, "lock file -> sb -> s_umount");

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 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_lock pid=%d\n", main_task->pid);
		if (lock_read)
			mmap_read_lock(main_task->mm);
		else
			mmap_write_lock(main_task->mm);
	}

	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)
			mmap_read_unlock(main_task->mm);
		else
			mmap_write_unlock(main_task->mm);
		if (verbose)
			pr_notice("unlock mmap_lock 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) ||
	    get_kernel_nofault(buf, ptr) ||
	    get_kernel_nofault(buf, ptr + size - 1)) {
		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 (get_kernel_nofault(magic, ptr) || 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_lock\n");
		return -EINVAL;
	}

	if (test_file_path[0]) {
		test_file = filp_open(test_file_path, O_RDONLY, 0);
		if (IS_ERR(test_file)) {
			pr_err("cannot find file_path\n");
			return -EINVAL;
		}
		test_inode = file_inode(test_file);
	} else if (test_lock_inode ||
		   test_lock_mapping ||
		   test_lock_sb_umount) {
		pr_err("no file to lock\n");
		return -EINVAL;
	}

	if (test_lock_inode && test_inode)
		lock_rwsem_ptr = (unsigned long)&test_inode->i_rwsem;

	if (test_lock_mapping && test_file && test_file->f_mapping)
		lock_rwsem_ptr = (unsigned long)&test_file->f_mapping->i_mmap_rwsem;

	if (test_lock_sb_umount && test_inode)
		lock_rwsem_ptr = (unsigned long)&test_inode->i_sb->s_umount;

	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",
		  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 (test_file)
		fput(test_file);

	if (signal_pending(main_task))
		return -EINTR;

	return -EAGAIN;
}
module_init(test_lockup_init);

MODULE_LICENSE("GPL");
MODULE_AUTHOR("Konstantin Khlebnikov <[email protected]>");
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>
aecd42df	17	#include <linux/file.h>
30428ef5 KK	18
	19	static unsigned int time_secs;
	20	module_param(time_secs, uint, 0600);
	21	MODULE_PARM_DESC(time_secs, "lockup time in seconds, default 0");
	22
	23	static unsigned int time_nsecs;
	24	module_param(time_nsecs, uint, 0600);
	25	MODULE_PARM_DESC(time_nsecs, "nanoseconds part of lockup time, default 0");
	26
	27	static unsigned int cooldown_secs;
	28	module_param(cooldown_secs, uint, 0600);
	29	MODULE_PARM_DESC(cooldown_secs, "cooldown time between iterations in seconds, default 0");
	30
	31	static unsigned int cooldown_nsecs;
	32	module_param(cooldown_nsecs, uint, 0600);
	33	MODULE_PARM_DESC(cooldown_nsecs, "nanoseconds part of cooldown, default 0");
	34
	35	static unsigned int iterations = 1;
	36	module_param(iterations, uint, 0600);
	37	MODULE_PARM_DESC(iterations, "lockup iterations, default 1");
	38
	39	static bool all_cpus;
	40	module_param(all_cpus, bool, 0400);
	41	MODULE_PARM_DESC(all_cpus, "trigger lockup at all cpus at once");
	42
	43	static int wait_state;
	44	static char *state = "R";
	45	module_param(state, charp, 0400);
	46	MODULE_PARM_DESC(state, "wait in 'R' running (default), 'D' uninterruptible, 'K' killable, 'S' interruptible state");
	47
	48	static bool use_hrtimer;
	49	module_param(use_hrtimer, bool, 0400);
	50	MODULE_PARM_DESC(use_hrtimer, "use high-resolution timer for sleeping");
	51
	52	static bool iowait;
	53	module_param(iowait, bool, 0400);
	54	MODULE_PARM_DESC(iowait, "account sleep time as iowait");
	55
	56	static bool lock_read;
	57	module_param(lock_read, bool, 0400);
	58	MODULE_PARM_DESC(lock_read, "lock read-write locks for read");
	59
	60	static bool lock_single;
	61	module_param(lock_single, bool, 0400);
	62	MODULE_PARM_DESC(lock_single, "acquire locks only at one cpu");
	63
	64	static bool reacquire_locks;
	65	module_param(reacquire_locks, bool, 0400);
	66	MODULE_PARM_DESC(reacquire_locks, "release and reacquire locks/irq/preempt between iterations");
	67
	68	static bool touch_softlockup;
	69	module_param(touch_softlockup, bool, 0600);
	70	MODULE_PARM_DESC(touch_softlockup, "touch soft-lockup watchdog between iterations");
	71
	72	static bool touch_hardlockup;
	73	module_param(touch_hardlockup, bool, 0600);
	74	MODULE_PARM_DESC(touch_hardlockup, "touch hard-lockup watchdog between iterations");
	75
	76	static bool call_cond_resched;
	77	module_param(call_cond_resched, bool, 0600);
	78	MODULE_PARM_DESC(call_cond_resched, "call cond_resched() between iterations");
	79
	80	static bool measure_lock_wait;
	81	module_param(measure_lock_wait, bool, 0400);
82	MODULE_PARM_DESC(measure_lock_wait, "measure lock wait time");
83
84	static unsigned long lock_wait_threshold = ULONG_MAX;
85	module_param(lock_wait_threshold, ulong, 0400);
86	MODULE_PARM_DESC(lock_wait_threshold, "print lock wait time longer than this in nanoseconds, default off");
87
88	static bool test_disable_irq;
89	module_param_named(disable_irq, test_disable_irq, bool, 0400);
90	MODULE_PARM_DESC(disable_irq, "disable interrupts: generate hard-lockups");
91
92	static bool disable_softirq;
93	module_param(disable_softirq, bool, 0400);
94	MODULE_PARM_DESC(disable_softirq, "disable bottom-half irq handlers");
95
96	static bool disable_preempt;
97	module_param(disable_preempt, bool, 0400);
98	MODULE_PARM_DESC(disable_preempt, "disable preemption: generate soft-lockups");
99
100	static bool lock_rcu;
101	module_param(lock_rcu, bool, 0400);
102	MODULE_PARM_DESC(lock_rcu, "grab rcu_read_lock: generate rcu stalls");
103
104	static bool lock_mmap_sem;
105	module_param(lock_mmap_sem, bool, 0400);
c1e8d7c6	106	MODULE_PARM_DESC(lock_mmap_sem, "lock mm->mmap_lock: block procfs interfaces");
30428ef5 KK	107
	108	static unsigned long lock_rwsem_ptr;
	109	module_param_unsafe(lock_rwsem_ptr, ulong, 0400);
	110	MODULE_PARM_DESC(lock_rwsem_ptr, "lock rw_semaphore at address");
	111
	112	static unsigned long lock_mutex_ptr;
	113	module_param_unsafe(lock_mutex_ptr, ulong, 0400);
	114	MODULE_PARM_DESC(lock_mutex_ptr, "lock mutex at address");
	115
	116	static unsigned long lock_spinlock_ptr;
	117	module_param_unsafe(lock_spinlock_ptr, ulong, 0400);
	118	MODULE_PARM_DESC(lock_spinlock_ptr, "lock spinlock at address");
	119
	120	static unsigned long lock_rwlock_ptr;
	121	module_param_unsafe(lock_rwlock_ptr, ulong, 0400);
	122	MODULE_PARM_DESC(lock_rwlock_ptr, "lock rwlock at address");
	123
	124	static unsigned int alloc_pages_nr;
	125	module_param_unsafe(alloc_pages_nr, uint, 0600);
	126	MODULE_PARM_DESC(alloc_pages_nr, "allocate and free pages under locks");
	127
	128	static unsigned int alloc_pages_order;
	129	module_param(alloc_pages_order, uint, 0400);
	130	MODULE_PARM_DESC(alloc_pages_order, "page order to allocate");
	131
	132	static gfp_t alloc_pages_gfp = GFP_KERNEL;
	133	module_param_unsafe(alloc_pages_gfp, uint, 0400);
	134	MODULE_PARM_DESC(alloc_pages_gfp, "allocate pages with this gfp_mask, default GFP_KERNEL");
	135
	136	static bool alloc_pages_atomic;
	137	module_param(alloc_pages_atomic, bool, 0400);
	138	MODULE_PARM_DESC(alloc_pages_atomic, "allocate pages with GFP_ATOMIC");
	139
	140	static bool reallocate_pages;
	141	module_param(reallocate_pages, bool, 0400);
	142	MODULE_PARM_DESC(reallocate_pages, "free and allocate pages between iterations");
	143
aecd42df	144	struct file *test_file;
02223e36	145	static struct inode *test_inode;
aecd42df KK	146	static char test_file_path[256];
	147	module_param_string(file_path, test_file_path, sizeof(test_file_path), 0400);
	148	MODULE_PARM_DESC(file_path, "file path to test");
	149
	150	static bool test_lock_inode;
	151	module_param_named(lock_inode, test_lock_inode, bool, 0400);
	152	MODULE_PARM_DESC(lock_inode, "lock file -> inode -> i_rwsem");
	153
	154	static bool test_lock_mapping;
	155	module_param_named(lock_mapping, test_lock_mapping, bool, 0400);
	156	MODULE_PARM_DESC(lock_mapping, "lock file -> mapping -> i_mmap_rwsem");
	157
	158	static bool test_lock_sb_umount;
	159	module_param_named(lock_sb_umount, test_lock_sb_umount, bool, 0400);
	160	MODULE_PARM_DESC(lock_sb_umount, "lock file -> sb -> s_umount");
	161
30428ef5 KK	162	static atomic_t alloc_pages_failed = ATOMIC_INIT(0);
	163
	164	static atomic64_t max_lock_wait = ATOMIC64_INIT(0);
	165
	166	static struct task_struct *main_task;
	167	static int master_cpu;
	168
	169	static void test_lock(bool master, bool verbose)
	170	{
3f649ab7	171	u64 wait_start;
30428ef5 KK	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)
c1e8d7c6	194	pr_notice("lock mmap_lock pid=%d\n", main_task->pid);
30428ef5	195	if (lock_read)
d8ed45c5	196	mmap_read_lock(main_task->mm);
30428ef5	197	else
d8ed45c5	198	mmap_write_lock(main_task->mm);
30428ef5 KK	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
	246	static 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)
d8ed45c5	279	mmap_read_unlock(main_task->mm);
30428ef5	280	else
d8ed45c5	281	mmap_write_unlock(main_task->mm);
30428ef5	282	if (verbose)
c1e8d7c6	283	pr_notice("unlock mmap_lock pid=%d\n", main_task->pid);
30428ef5 KK	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
	304	static 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
	319	static 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
	328	static 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
349	static 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
403	DEFINE_PER_CPU(struct work_struct, test_works);
404
405	static 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
411	static 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 (access_ok(ptr, 1) \|\|
421	access_ok(ptr + size - 1, 1) \|\|
25f12ae4 CH	422	get_kernel_nofault(buf, ptr) \|\|
25f12ae4 CH	423	get_kernel_nofault(buf, ptr + size - 1)) {
30428ef5 KK	424	pr_err("invalid kernel ptr: %#lx\n", addr);
	425	return true;
	426	}
	427
	428	return false;
	429	}
	430
	431	static bool __maybe_unused test_magic(unsigned long addr, int offset,
	432	unsigned int expected)
	433	{
	434	void ptr = (void )addr + offset;
	435	unsigned int magic = 0;
	436
	437	if (!addr)
	438	return false;
	439
25f12ae4	440	if (get_kernel_nofault(magic, ptr) \|\| magic != expected) {
30428ef5 KK	441	pr_err("invalid magic at %#lx + %#x = %#x, expected %#x\n",
	442	addr, offset, magic, expected);
	443	return true;
	444	}
	445
	446	return false;
	447	}
	448
	449	static int __init test_lockup_init(void)
	450	{
	451	u64 test_start = local_clock();
	452
	453	main_task = current;
	454
	455	switch (state[0]) {
	456	case 'S':
	457	wait_state = TASK_INTERRUPTIBLE;
	458	break;
	459	case 'D':
	460	wait_state = TASK_UNINTERRUPTIBLE;
	461	break;
	462	case 'K':
	463	wait_state = TASK_KILLABLE;
	464	break;
	465	case 'R':
	466	wait_state = TASK_RUNNING;
	467	break;
	468	default:
	469	pr_err("unknown state=%s\n", state);
	470	return -EINVAL;
	471	}
	472
	473	if (alloc_pages_atomic)
	474	alloc_pages_gfp = GFP_ATOMIC;
	475
	476	if (test_kernel_ptr(lock_spinlock_ptr, sizeof(spinlock_t)) \|\|
	477	test_kernel_ptr(lock_rwlock_ptr, sizeof(rwlock_t)) \|\|
	478	test_kernel_ptr(lock_mutex_ptr, sizeof(struct mutex)) \|\|
	479	test_kernel_ptr(lock_rwsem_ptr, sizeof(struct rw_semaphore)))
	480	return -EINVAL;
	481
	482	#ifdef CONFIG_DEBUG_SPINLOCK
	483	if (test_magic(lock_spinlock_ptr,
	484	offsetof(spinlock_t, rlock.magic),
	485	SPINLOCK_MAGIC) \|\|
	486	test_magic(lock_rwlock_ptr,
	487	offsetof(rwlock_t, magic),
	488	RWLOCK_MAGIC) \|\|
	489	test_magic(lock_mutex_ptr,
	490	offsetof(struct mutex, wait_lock.rlock.magic),
	491	SPINLOCK_MAGIC) \|\|
	492	test_magic(lock_rwsem_ptr,
	493	offsetof(struct rw_semaphore, wait_lock.magic),
	494	SPINLOCK_MAGIC))
	495	return -EINVAL;
	496	#endif
	497
	498	if ((wait_state != TASK_RUNNING \|\|
	499	(call_cond_resched && !reacquire_locks) \|\|
	500	(alloc_pages_nr && gfpflags_allow_blocking(alloc_pages_gfp))) &&
	501	(test_disable_irq \|\| disable_softirq \|\| disable_preempt \|\|
	502	lock_rcu \|\| lock_spinlock_ptr \|\| lock_rwlock_ptr)) {
	503	pr_err("refuse to sleep in atomic context\n");
	504	return -EINVAL;
505	}
506
507	if (lock_mmap_sem && !main_task->mm) {
c1e8d7c6	508	pr_err("no mm to lock mmap_lock\n");
30428ef5 KK	509	return -EINVAL;
	510	}
	511
aecd42df KK	512	if (test_file_path[0]) {
	513	test_file = filp_open(test_file_path, O_RDONLY, 0);
	514	if (IS_ERR(test_file)) {
	515	pr_err("cannot find file_path\n");
	516	return -EINVAL;
	517	}
	518	test_inode = file_inode(test_file);
	519	} else if (test_lock_inode \|\|
	520	test_lock_mapping \|\|
	521	test_lock_sb_umount) {
	522	pr_err("no file to lock\n");
	523	return -EINVAL;
	524	}
	525
	526	if (test_lock_inode && test_inode)
	527	lock_rwsem_ptr = (unsigned long)&test_inode->i_rwsem;
	528
	529	if (test_lock_mapping && test_file && test_file->f_mapping)
	530	lock_rwsem_ptr = (unsigned long)&test_file->f_mapping->i_mmap_rwsem;
	531
	532	if (test_lock_sb_umount && test_inode)
	533	lock_rwsem_ptr = (unsigned long)&test_inode->i_sb->s_umount;
	534
ad3f434b	535	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",
30428ef5 KK	536	main_task->pid, time_secs, time_nsecs,
	537	cooldown_secs, cooldown_nsecs, iterations, state,
	538	all_cpus ? "all_cpus " : "",
	539	iowait ? "iowait " : "",
	540	test_disable_irq ? "disable_irq " : "",
	541	disable_softirq ? "disable_softirq " : "",
	542	disable_preempt ? "disable_preempt " : "",
	543	lock_rcu ? "lock_rcu " : "",
	544	lock_read ? "lock_read " : "",
	545	touch_softlockup ? "touch_softlockup " : "",
	546	touch_hardlockup ? "touch_hardlockup " : "",
	547	call_cond_resched ? "call_cond_resched " : "",
	548	reacquire_locks ? "reacquire_locks " : "");
	549
	550	if (alloc_pages_nr)
	551	pr_notice("ALLOCATE PAGES nr=%u order=%u gfp=%pGg %s\n",
	552	alloc_pages_nr, alloc_pages_order, &alloc_pages_gfp,
	553	reallocate_pages ? "reallocate_pages " : "");
	554
	555	if (all_cpus) {
	556	unsigned int cpu;
	557
	558	cpus_read_lock();
	559
	560	preempt_disable();
	561	master_cpu = smp_processor_id();
	562	for_each_online_cpu(cpu) {
	563	INIT_WORK(per_cpu_ptr(&test_works, cpu), test_work_fn);
	564	queue_work_on(cpu, system_highpri_wq,
	565	per_cpu_ptr(&test_works, cpu));
	566	}
	567	preempt_enable();
	568
	569	for_each_online_cpu(cpu)
	570	flush_work(per_cpu_ptr(&test_works, cpu));
	571
	572	cpus_read_unlock();
	573	} else {
	574	test_lockup(true);
	575	}
	576
	577	if (measure_lock_wait)
	578	pr_notice("Maximum lock wait: %lld ns\n",
	579	atomic64_read(&max_lock_wait));
	580
	581	if (alloc_pages_nr)
	582	pr_notice("Page allocation failed %u times\n",
	583	atomic_read(&alloc_pages_failed));
	584
	585	pr_notice("FINISH in %llu ns\n", local_clock() - test_start);
	586
aecd42df KK	587	if (test_file)
	588	fput(test_file);
	589
30428ef5 KK	590	if (signal_pending(main_task))
	591	return -EINTR;
	592
	593	return -EAGAIN;
	594	}
	595	module_init(test_lockup_init);
	596
	597	MODULE_LICENSE("GPL");
	598	MODULE_AUTHOR("Konstantin Khlebnikov <[email protected]>");
	599	MODULE_DESCRIPTION("Test module to generate lockups");