[mirror_ubuntu-artful-kernel.git] / kernel / rcu / rcuperf.c

/*
 * Read-Copy Update module-based performance-test facility
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, you can access it online at
 * http://www.gnu.org/licenses/gpl-2.0.html.
 *
 * Copyright (C) IBM Corporation, 2015
 *
 * Authors: Paul E. McKenney <paulmck@us.ibm.com>
 */
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/module.h>
#include <linux/kthread.h>
#include <linux/err.h>
#include <linux/spinlock.h>
#include <linux/smp.h>
#include <linux/rcupdate.h>
#include <linux/interrupt.h>
#include <linux/sched.h>
#include <uapi/linux/sched/types.h>
#include <linux/atomic.h>
#include <linux/bitops.h>
#include <linux/completion.h>
#include <linux/moduleparam.h>
#include <linux/percpu.h>
#include <linux/notifier.h>
#include <linux/reboot.h>
#include <linux/freezer.h>
#include <linux/cpu.h>
#include <linux/delay.h>
#include <linux/stat.h>
#include <linux/srcu.h>
#include <linux/slab.h>
#include <asm/byteorder.h>
#include <linux/torture.h>
#include <linux/vmalloc.h>

MODULE_LICENSE("GPL");
MODULE_AUTHOR("Paul E. McKenney <paulmck@linux.vnet.ibm.com>");

#define PERF_FLAG "-perf:"
#define PERFOUT_STRING(s) \
	pr_alert("%s" PERF_FLAG " %s\n", perf_type, s)
#define VERBOSE_PERFOUT_STRING(s) \
	do { if (verbose) pr_alert("%s" PERF_FLAG " %s\n", perf_type, s); } while (0)
#define VERBOSE_PERFOUT_ERRSTRING(s) \
	do { if (verbose) pr_alert("%s" PERF_FLAG "!!! %s\n", perf_type, s); } while (0)

torture_param(bool, gp_exp, false, "Use expedited GP wait primitives");
torture_param(int, holdoff, 10, "Holdoff time before test start (s)");
torture_param(int, nreaders, -1, "Number of RCU reader threads");
torture_param(int, nwriters, -1, "Number of RCU updater threads");
torture_param(bool, shutdown, false, "Shutdown at end of performance tests.");
torture_param(bool, verbose, true, "Enable verbose debugging printk()s");

static char *perf_type = "rcu";
module_param(perf_type, charp, 0444);
MODULE_PARM_DESC(perf_type, "Type of RCU to performance-test (rcu, rcu_bh, ...)");

static int nrealreaders;
static int nrealwriters;
static struct task_struct **writer_tasks;
static struct task_struct **reader_tasks;
static struct task_struct *shutdown_task;

static u64 **writer_durations;
static int *writer_n_durations;
static atomic_t n_rcu_perf_reader_started;
static atomic_t n_rcu_perf_writer_started;
static atomic_t n_rcu_perf_writer_finished;
static wait_queue_head_t shutdown_wq;
static u64 t_rcu_perf_writer_started;
static u64 t_rcu_perf_writer_finished;
static unsigned long b_rcu_perf_writer_started;
static unsigned long b_rcu_perf_writer_finished;

static int rcu_perf_writer_state;
#define RTWS_INIT		0
#define RTWS_EXP_SYNC		1
#define RTWS_SYNC		2
#define RTWS_IDLE		2
#define RTWS_STOPPING		3

#define MAX_MEAS 10000
#define MIN_MEAS 100

static int perf_runnable = IS_ENABLED(MODULE);
module_param(perf_runnable, int, 0444);
MODULE_PARM_DESC(perf_runnable, "Start rcuperf at boot");

/*
 * Operations vector for selecting different types of tests.
 */

struct rcu_perf_ops {
	int ptype;
	void (*init)(void);
	void (*cleanup)(void);
	int (*readlock)(void);
	void (*readunlock)(int idx);
	unsigned long (*started)(void);
	unsigned long (*completed)(void);
	unsigned long (*exp_completed)(void);
	void (*sync)(void);
	void (*exp_sync)(void);
	const char *name;
};

static struct rcu_perf_ops *cur_ops;

/*
 * Definitions for rcu perf testing.
 */

static int rcu_perf_read_lock(void) __acquires(RCU)
{
	rcu_read_lock();
	return 0;
}

static void rcu_perf_read_unlock(int idx) __releases(RCU)
{
	rcu_read_unlock();
}

static unsigned long __maybe_unused rcu_no_completed(void)
{
	return 0;
}

static void rcu_sync_perf_init(void)
{
}

static struct rcu_perf_ops rcu_ops = {
	.ptype		= RCU_FLAVOR,
	.init		= rcu_sync_perf_init,
	.readlock	= rcu_perf_read_lock,
	.readunlock	= rcu_perf_read_unlock,
	.started	= rcu_batches_started,
	.completed	= rcu_batches_completed,
	.exp_completed	= rcu_exp_batches_completed,
	.sync		= synchronize_rcu,
	.exp_sync	= synchronize_rcu_expedited,
	.name		= "rcu"
};

/*
 * Definitions for rcu_bh perf testing.
 */

static int rcu_bh_perf_read_lock(void) __acquires(RCU_BH)
{
	rcu_read_lock_bh();
	return 0;
}

static void rcu_bh_perf_read_unlock(int idx) __releases(RCU_BH)
{
	rcu_read_unlock_bh();
}

static struct rcu_perf_ops rcu_bh_ops = {
	.ptype		= RCU_BH_FLAVOR,
	.init		= rcu_sync_perf_init,
	.readlock	= rcu_bh_perf_read_lock,
	.readunlock	= rcu_bh_perf_read_unlock,
	.started	= rcu_batches_started_bh,
	.completed	= rcu_batches_completed_bh,
	.exp_completed	= rcu_exp_batches_completed_sched,
	.sync		= synchronize_rcu_bh,
	.exp_sync	= synchronize_rcu_bh_expedited,
	.name		= "rcu_bh"
};

/*
 * Definitions for srcu perf testing.
 */

DEFINE_STATIC_SRCU(srcu_ctl_perf);
static struct srcu_struct *srcu_ctlp = &srcu_ctl_perf;

static int srcu_perf_read_lock(void) __acquires(srcu_ctlp)
{
	return srcu_read_lock(srcu_ctlp);
}

static void srcu_perf_read_unlock(int idx) __releases(srcu_ctlp)
{
	srcu_read_unlock(srcu_ctlp, idx);
}

static unsigned long srcu_perf_completed(void)
{
	return srcu_batches_completed(srcu_ctlp);
}

static void srcu_perf_synchronize(void)
{
	synchronize_srcu(srcu_ctlp);
}

static void srcu_perf_synchronize_expedited(void)
{
	synchronize_srcu_expedited(srcu_ctlp);
}

static struct rcu_perf_ops srcu_ops = {
	.ptype		= SRCU_FLAVOR,
	.init		= rcu_sync_perf_init,
	.readlock	= srcu_perf_read_lock,
	.readunlock	= srcu_perf_read_unlock,
	.started	= NULL,
	.completed	= srcu_perf_completed,
	.exp_completed	= srcu_perf_completed,
	.sync		= srcu_perf_synchronize,
	.exp_sync	= srcu_perf_synchronize_expedited,
	.name		= "srcu"
};

/*
 * Definitions for sched perf testing.
 */

static int sched_perf_read_lock(void)
{
	preempt_disable();
	return 0;
}

static void sched_perf_read_unlock(int idx)
{
	preempt_enable();
}

static struct rcu_perf_ops sched_ops = {
	.ptype		= RCU_SCHED_FLAVOR,
	.init		= rcu_sync_perf_init,
	.readlock	= sched_perf_read_lock,
	.readunlock	= sched_perf_read_unlock,
	.started	= rcu_batches_started_sched,
	.completed	= rcu_batches_completed_sched,
	.exp_completed	= rcu_exp_batches_completed_sched,
	.sync		= synchronize_sched,
	.exp_sync	= synchronize_sched_expedited,
	.name		= "sched"
};

#ifdef CONFIG_TASKS_RCU

/*
 * Definitions for RCU-tasks perf testing.
 */

static int tasks_perf_read_lock(void)
{
	return 0;
}

static void tasks_perf_read_unlock(int idx)
{
}

static struct rcu_perf_ops tasks_ops = {
	.ptype		= RCU_TASKS_FLAVOR,
	.init		= rcu_sync_perf_init,
	.readlock	= tasks_perf_read_lock,
	.readunlock	= tasks_perf_read_unlock,
	.started	= rcu_no_completed,
	.completed	= rcu_no_completed,
	.sync		= synchronize_rcu_tasks,
	.exp_sync	= synchronize_rcu_tasks,
	.name		= "tasks"
};

#define RCUPERF_TASKS_OPS &tasks_ops,

static bool __maybe_unused torturing_tasks(void)
{
	return cur_ops == &tasks_ops;
}

#else /* #ifdef CONFIG_TASKS_RCU */

#define RCUPERF_TASKS_OPS

static bool __maybe_unused torturing_tasks(void)
{
	return false;
}

#endif /* #else #ifdef CONFIG_TASKS_RCU */

/*
 * If performance tests complete, wait for shutdown to commence.
 */
static void rcu_perf_wait_shutdown(void)
{
	cond_resched_rcu_qs();
	if (atomic_read(&n_rcu_perf_writer_finished) < nrealwriters)
		return;
	while (!torture_must_stop())
		schedule_timeout_uninterruptible(1);
}

/*
 * RCU perf reader kthread.  Repeatedly does empty RCU read-side
 * critical section, minimizing update-side interference.
 */
static int
rcu_perf_reader(void *arg)
{
	unsigned long flags;
	int idx;
	long me = (long)arg;

	VERBOSE_PERFOUT_STRING("rcu_perf_reader task started");
	set_cpus_allowed_ptr(current, cpumask_of(me % nr_cpu_ids));
	set_user_nice(current, MAX_NICE);
	atomic_inc(&n_rcu_perf_reader_started);

	do {
		local_irq_save(flags);
		idx = cur_ops->readlock();
		cur_ops->readunlock(idx);
		local_irq_restore(flags);
		rcu_perf_wait_shutdown();
	} while (!torture_must_stop());
	torture_kthread_stopping("rcu_perf_reader");
	return 0;
}

/*
 * RCU perf writer kthread.  Repeatedly does a grace period.
 */
static int
rcu_perf_writer(void *arg)
{
	int i = 0;
	int i_max;
	long me = (long)arg;
	struct sched_param sp;
	bool started = false, done = false, alldone = false;
	u64 t;
	u64 *wdp;
	u64 *wdpp = writer_durations[me];

	VERBOSE_PERFOUT_STRING("rcu_perf_writer task started");
	WARN_ON(!wdpp);
	set_cpus_allowed_ptr(current, cpumask_of(me % nr_cpu_ids));
	sp.sched_priority = 1;
	sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);

	if (holdoff)
		schedule_timeout_uninterruptible(holdoff * HZ);

	t = ktime_get_mono_fast_ns();
	if (atomic_inc_return(&n_rcu_perf_writer_started) >= nrealwriters) {
		t_rcu_perf_writer_started = t;
		if (gp_exp) {
			b_rcu_perf_writer_started =
				cur_ops->exp_completed() / 2;
		} else {
			b_rcu_perf_writer_started =
				cur_ops->completed();
		}
	}

	do {
		wdp = &wdpp[i];
		*wdp = ktime_get_mono_fast_ns();
		if (gp_exp) {
			rcu_perf_writer_state = RTWS_EXP_SYNC;
			cur_ops->exp_sync();
		} else {
			rcu_perf_writer_state = RTWS_SYNC;
			cur_ops->sync();
		}
		rcu_perf_writer_state = RTWS_IDLE;
		t = ktime_get_mono_fast_ns();
		*wdp = t - *wdp;
		i_max = i;
		if (!started &&
		    atomic_read(&n_rcu_perf_writer_started) >= nrealwriters)
			started = true;
		if (!done && i >= MIN_MEAS) {
			done = true;
			sp.sched_priority = 0;
			sched_setscheduler_nocheck(current,
						   SCHED_NORMAL, &sp);
			pr_alert("%s%s rcu_perf_writer %ld has %d measurements\n",
				 perf_type, PERF_FLAG, me, MIN_MEAS);
			if (atomic_inc_return(&n_rcu_perf_writer_finished) >=
			    nrealwriters) {
				schedule_timeout_interruptible(10);
				rcu_ftrace_dump(DUMP_ALL);
				PERFOUT_STRING("Test complete");
				t_rcu_perf_writer_finished = t;
				if (gp_exp) {
					b_rcu_perf_writer_finished =
						cur_ops->exp_completed() / 2;
				} else {
					b_rcu_perf_writer_finished =
						cur_ops->completed();
				}
				if (shutdown) {
					smp_mb(); /* Assign before wake. */
					wake_up(&shutdown_wq);
				}
			}
		}
		if (done && !alldone &&
		    atomic_read(&n_rcu_perf_writer_finished) >= nrealwriters)
			alldone = true;
		if (started && !alldone && i < MAX_MEAS - 1)
			i++;
		rcu_perf_wait_shutdown();
	} while (!torture_must_stop());
	rcu_perf_writer_state = RTWS_STOPPING;
	writer_n_durations[me] = i_max;
	torture_kthread_stopping("rcu_perf_writer");
	return 0;
}

static inline void
rcu_perf_print_module_parms(struct rcu_perf_ops *cur_ops, const char *tag)
{
	pr_alert("%s" PERF_FLAG
		 "--- %s: nreaders=%d nwriters=%d verbose=%d shutdown=%d\n",
		 perf_type, tag, nrealreaders, nrealwriters, verbose, shutdown);
}

static void
rcu_perf_cleanup(void)
{
	int i;
	int j;
	int ngps = 0;
	u64 *wdp;
	u64 *wdpp;

	if (torture_cleanup_begin())
		return;

	if (reader_tasks) {
		for (i = 0; i < nrealreaders; i++)
			torture_stop_kthread(rcu_perf_reader,
					     reader_tasks[i]);
		kfree(reader_tasks);
	}

	if (writer_tasks) {
		for (i = 0; i < nrealwriters; i++) {
			torture_stop_kthread(rcu_perf_writer,
					     writer_tasks[i]);
			if (!writer_n_durations)
				continue;
			j = writer_n_durations[i];
			pr_alert("%s%s writer %d gps: %d\n",
				 perf_type, PERF_FLAG, i, j);
			ngps += j;
		}
		pr_alert("%s%s start: %llu end: %llu duration: %llu gps: %d batches: %ld\n",
			 perf_type, PERF_FLAG,
			 t_rcu_perf_writer_started, t_rcu_perf_writer_finished,
			 t_rcu_perf_writer_finished -
			 t_rcu_perf_writer_started,
			 ngps,
			 b_rcu_perf_writer_finished -
			 b_rcu_perf_writer_started);
		for (i = 0; i < nrealwriters; i++) {
			if (!writer_durations)
				break;
			if (!writer_n_durations)
				continue;
			wdpp = writer_durations[i];
			if (!wdpp)
				continue;
			for (j = 0; j <= writer_n_durations[i]; j++) {
				wdp = &wdpp[j];
				pr_alert("%s%s %4d writer-duration: %5d %llu\n",
					perf_type, PERF_FLAG,
					i, j, *wdp);
				if (j % 100 == 0)
					schedule_timeout_uninterruptible(1);
			}
			kfree(writer_durations[i]);
		}
		kfree(writer_tasks);
		kfree(writer_durations);
		kfree(writer_n_durations);
	}

	/* Do flavor-specific cleanup operations.  */
	if (cur_ops->cleanup != NULL)
		cur_ops->cleanup();

	torture_cleanup_end();
}

/*
 * Return the number if non-negative.  If -1, the number of CPUs.
 * If less than -1, that much less than the number of CPUs, but
 * at least one.
 */
static int compute_real(int n)
{
	int nr;

	if (n >= 0) {
		nr = n;
	} else {
		nr = num_online_cpus() + 1 + n;
		if (nr <= 0)
			nr = 1;
	}
	return nr;
}

/*
 * RCU perf shutdown kthread.  Just waits to be awakened, then shuts
 * down system.
 */
static int
rcu_perf_shutdown(void *arg)
{
	do {
		wait_event(shutdown_wq,
			   atomic_read(&n_rcu_perf_writer_finished) >=
			   nrealwriters);
	} while (atomic_read(&n_rcu_perf_writer_finished) < nrealwriters);
	smp_mb(); /* Wake before output. */
	rcu_perf_cleanup();
	kernel_power_off();
	return -EINVAL;
}

static int __init
rcu_perf_init(void)
{
	long i;
	int firsterr = 0;
	static struct rcu_perf_ops *perf_ops[] = {
		&rcu_ops, &rcu_bh_ops, &srcu_ops, &sched_ops,
		RCUPERF_TASKS_OPS
	};

	if (!torture_init_begin(perf_type, verbose, &perf_runnable))
		return -EBUSY;

	/* Process args and tell the world that the perf'er is on the job. */
	for (i = 0; i < ARRAY_SIZE(perf_ops); i++) {
		cur_ops = perf_ops[i];
		if (strcmp(perf_type, cur_ops->name) == 0)
			break;
	}
	if (i == ARRAY_SIZE(perf_ops)) {
		pr_alert("rcu-perf: invalid perf type: \"%s\"\n",
			 perf_type);
		pr_alert("rcu-perf types:");
		for (i = 0; i < ARRAY_SIZE(perf_ops); i++)
			pr_alert(" %s", perf_ops[i]->name);
		pr_alert("\n");
		firsterr = -EINVAL;
		goto unwind;
	}
	if (cur_ops->init)
		cur_ops->init();

	nrealwriters = compute_real(nwriters);
	nrealreaders = compute_real(nreaders);
	atomic_set(&n_rcu_perf_reader_started, 0);
	atomic_set(&n_rcu_perf_writer_started, 0);
	atomic_set(&n_rcu_perf_writer_finished, 0);
	rcu_perf_print_module_parms(cur_ops, "Start of test");

	/* Start up the kthreads. */

	if (shutdown) {
		init_waitqueue_head(&shutdown_wq);
		firsterr = torture_create_kthread(rcu_perf_shutdown, NULL,
						  shutdown_task);
		if (firsterr)
			goto unwind;
		schedule_timeout_uninterruptible(1);
	}
	reader_tasks = kcalloc(nrealreaders, sizeof(reader_tasks[0]),
			       GFP_KERNEL);
	if (reader_tasks == NULL) {
		VERBOSE_PERFOUT_ERRSTRING("out of memory");
		firsterr = -ENOMEM;
		goto unwind;
	}
	for (i = 0; i < nrealreaders; i++) {
		firsterr = torture_create_kthread(rcu_perf_reader, (void *)i,
						  reader_tasks[i]);
		if (firsterr)
			goto unwind;
	}
	while (atomic_read(&n_rcu_perf_reader_started) < nrealreaders)
		schedule_timeout_uninterruptible(1);
	writer_tasks = kcalloc(nrealwriters, sizeof(reader_tasks[0]),
			       GFP_KERNEL);
	writer_durations = kcalloc(nrealwriters, sizeof(*writer_durations),
				   GFP_KERNEL);
	writer_n_durations =
		kcalloc(nrealwriters, sizeof(*writer_n_durations),
			GFP_KERNEL);
	if (!writer_tasks || !writer_durations || !writer_n_durations) {
		VERBOSE_PERFOUT_ERRSTRING("out of memory");
		firsterr = -ENOMEM;
		goto unwind;
	}
	if (rcu_gp_is_expedited() && !rcu_gp_is_normal() && !gp_exp) {
		VERBOSE_PERFOUT_ERRSTRING("All grace periods expedited, no normal ones to measure!");
		firsterr = -EINVAL;
		goto unwind;
	}
	if (rcu_gp_is_normal() && gp_exp) {
		VERBOSE_PERFOUT_ERRSTRING("All grace periods normal, no expedited ones to measure!");
		firsterr = -EINVAL;
		goto unwind;
	}
	for (i = 0; i < nrealwriters; i++) {
		writer_durations[i] =
			kcalloc(MAX_MEAS, sizeof(*writer_durations[i]),
				GFP_KERNEL);
		if (!writer_durations[i]) {
			firsterr = -ENOMEM;
			goto unwind;
		}
		firsterr = torture_create_kthread(rcu_perf_writer, (void *)i,
						  writer_tasks[i]);
		if (firsterr)
			goto unwind;
	}
	torture_init_end();
	return 0;

unwind:
	torture_init_end();
	rcu_perf_cleanup();
	return firsterr;
}

module_init(rcu_perf_init);
module_exit(rcu_perf_cleanup);
Commit	Line	Data
8704baab PM	1	/*
	2	* Read-Copy Update module-based performance-test facility
	3	*
	4	* This program is free software; you can redistribute it and/or modify
	5	* it under the terms of the GNU General Public License as published by
	6	* the Free Software Foundation; either version 2 of the License, or
	7	* (at your option) any later version.
	8	*
	9	* This program is distributed in the hope that it will be useful,
	10	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	11	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
	12	* GNU General Public License for more details.
	13	*
	14	* You should have received a copy of the GNU General Public License
	15	* along with this program; if not, you can access it online at
	16	* http://www.gnu.org/licenses/gpl-2.0.html.
	17	*
	18	* Copyright (C) IBM Corporation, 2015
	19	*
	20	* Authors: Paul E. McKenney <paulmck@us.ibm.com>
	21	*/
	22	#include <linux/types.h>
	23	#include <linux/kernel.h>
	24	#include <linux/init.h>
	25	#include <linux/module.h>
	26	#include <linux/kthread.h>
	27	#include <linux/err.h>
	28	#include <linux/spinlock.h>
	29	#include <linux/smp.h>
	30	#include <linux/rcupdate.h>
	31	#include <linux/interrupt.h>
	32	#include <linux/sched.h>
ae7e81c0	33	#include <uapi/linux/sched/types.h>
8704baab PM	34	#include <linux/atomic.h>
	35	#include <linux/bitops.h>
	36	#include <linux/completion.h>
	37	#include <linux/moduleparam.h>
	38	#include <linux/percpu.h>
	39	#include <linux/notifier.h>
	40	#include <linux/reboot.h>
	41	#include <linux/freezer.h>
	42	#include <linux/cpu.h>
	43	#include <linux/delay.h>
	44	#include <linux/stat.h>
	45	#include <linux/srcu.h>
	46	#include <linux/slab.h>
	47	#include <asm/byteorder.h>
	48	#include <linux/torture.h>
	49	#include <linux/vmalloc.h>
	50
	51	MODULE_LICENSE("GPL");
	52	MODULE_AUTHOR("Paul E. McKenney <paulmck@linux.vnet.ibm.com>");
	53
	54	#define PERF_FLAG "-perf:"
	55	#define PERFOUT_STRING(s) \
a56fefa2	56	pr_alert("%s" PERF_FLAG " %s\n", perf_type, s)
8704baab PM	57	#define VERBOSE_PERFOUT_STRING(s) \
	58	do { if (verbose) pr_alert("%s" PERF_FLAG " %s\n", perf_type, s); } while (0)
	59	#define VERBOSE_PERFOUT_ERRSTRING(s) \
	60	do { if (verbose) pr_alert("%s" PERF_FLAG "!!! %s\n", perf_type, s); } while (0)
	61
af06d4f7	62	torture_param(bool, gp_exp, false, "Use expedited GP wait primitives");
df37e66b	63	torture_param(int, holdoff, 10, "Holdoff time before test start (s)");
8704baab PM	64	torture_param(int, nreaders, -1, "Number of RCU reader threads");
	65	torture_param(int, nwriters, -1, "Number of RCU updater threads");
	66	torture_param(bool, shutdown, false, "Shutdown at end of performance tests.");
	67	torture_param(bool, verbose, true, "Enable verbose debugging printk()s");
	68
	69	static char *perf_type = "rcu";
	70	module_param(perf_type, charp, 0444);
	71	MODULE_PARM_DESC(perf_type, "Type of RCU to performance-test (rcu, rcu_bh, ...)");
	72
	73	static int nrealreaders;
	74	static int nrealwriters;
	75	static struct task_struct **writer_tasks;
	76	static struct task_struct **reader_tasks;
	77	static struct task_struct *shutdown_task;
	78
	79	static u64 **writer_durations;
	80	static int *writer_n_durations;
	81	static atomic_t n_rcu_perf_reader_started;
	82	static atomic_t n_rcu_perf_writer_started;
	83	static atomic_t n_rcu_perf_writer_finished;
	84	static wait_queue_head_t shutdown_wq;
	85	static u64 t_rcu_perf_writer_started;
	86	static u64 t_rcu_perf_writer_finished;
	87	static unsigned long b_rcu_perf_writer_started;
	88	static unsigned long b_rcu_perf_writer_finished;
	89
	90	static int rcu_perf_writer_state;
	91	#define RTWS_INIT 0
	92	#define RTWS_EXP_SYNC 1
	93	#define RTWS_SYNC 2
	94	#define RTWS_IDLE 2
	95	#define RTWS_STOPPING 3
	96
	97	#define MAX_MEAS 10000
	98	#define MIN_MEAS 100
	99
f8cbdee9	100	static int perf_runnable = IS_ENABLED(MODULE);
8704baab PM	101	module_param(perf_runnable, int, 0444);
	102	MODULE_PARM_DESC(perf_runnable, "Start rcuperf at boot");
	103
	104	/*
	105	* Operations vector for selecting different types of tests.
	106	*/
	107
	108	struct rcu_perf_ops {
	109	int ptype;
	110	void (*init)(void);
	111	void (*cleanup)(void);
	112	int (*readlock)(void);
	113	void (*readunlock)(int idx);
	114	unsigned long (*started)(void);
	115	unsigned long (*completed)(void);
	116	unsigned long (*exp_completed)(void);
	117	void (*sync)(void);
	118	void (*exp_sync)(void);
	119	const char *name;
	120	};
	121
	122	static struct rcu_perf_ops *cur_ops;
	123
	124	/*
	125	* Definitions for rcu perf testing.
	126	*/
	127
	128	static int rcu_perf_read_lock(void) __acquires(RCU)
	129	{
	130	rcu_read_lock();
	131	return 0;
	132	}
	133
	134	static void rcu_perf_read_unlock(int idx) __releases(RCU)
	135	{
	136	rcu_read_unlock();
	137	}
	138
	139	static unsigned long __maybe_unused rcu_no_completed(void)
	140	{
	141	return 0;
	142	}
	143
	144	static void rcu_sync_perf_init(void)
	145	{
	146	}
	147
	148	static struct rcu_perf_ops rcu_ops = {
	149	.ptype = RCU_FLAVOR,
	150	.init = rcu_sync_perf_init,
	151	.readlock = rcu_perf_read_lock,
	152	.readunlock = rcu_perf_read_unlock,
	153	.started = rcu_batches_started,
	154	.completed = rcu_batches_completed,
	155	.exp_completed = rcu_exp_batches_completed,
	156	.sync = synchronize_rcu,
	157	.exp_sync = synchronize_rcu_expedited,
	158	.name = "rcu"
	159	};
	160
	161	/*
	162	* Definitions for rcu_bh perf testing.
	163	*/
	164
165	static int rcu_bh_perf_read_lock(void) __acquires(RCU_BH)
166	{
167	rcu_read_lock_bh();
168	return 0;
169	}
170
171	static void rcu_bh_perf_read_unlock(int idx) __releases(RCU_BH)
172	{
173	rcu_read_unlock_bh();
174	}
175
176	static struct rcu_perf_ops rcu_bh_ops = {
177	.ptype = RCU_BH_FLAVOR,
178	.init = rcu_sync_perf_init,
179	.readlock = rcu_bh_perf_read_lock,
180	.readunlock = rcu_bh_perf_read_unlock,
181	.started = rcu_batches_started_bh,
182	.completed = rcu_batches_completed_bh,
183	.exp_completed = rcu_exp_batches_completed_sched,
184	.sync = synchronize_rcu_bh,
185	.exp_sync = synchronize_rcu_bh_expedited,
186	.name = "rcu_bh"
187	};
188
189	/*
190	* Definitions for srcu perf testing.
191	*/
192
193	DEFINE_STATIC_SRCU(srcu_ctl_perf);
194	static struct srcu_struct *srcu_ctlp = &srcu_ctl_perf;
195
196	static int srcu_perf_read_lock(void) __acquires(srcu_ctlp)
197	{
198	return srcu_read_lock(srcu_ctlp);
199	}
200
201	static void srcu_perf_read_unlock(int idx) __releases(srcu_ctlp)
202	{
203	srcu_read_unlock(srcu_ctlp, idx);
204	}
205
206	static unsigned long srcu_perf_completed(void)
207	{
208	return srcu_batches_completed(srcu_ctlp);
209	}
210
211	static void srcu_perf_synchronize(void)
212	{
213	synchronize_srcu(srcu_ctlp);
214	}
215
216	static void srcu_perf_synchronize_expedited(void)
217	{
218	synchronize_srcu_expedited(srcu_ctlp);
219	}
220
221	static struct rcu_perf_ops srcu_ops = {
222	.ptype = SRCU_FLAVOR,
223	.init = rcu_sync_perf_init,
224	.readlock = srcu_perf_read_lock,
225	.readunlock = srcu_perf_read_unlock,
226	.started = NULL,
227	.completed = srcu_perf_completed,
228	.exp_completed = srcu_perf_completed,
229	.sync = srcu_perf_synchronize,
230	.exp_sync = srcu_perf_synchronize_expedited,
231	.name = "srcu"
232	};
233
234	/*
235	* Definitions for sched perf testing.
236	*/
237
238	static int sched_perf_read_lock(void)
239	{
240	preempt_disable();
241	return 0;
242	}
243
244	static void sched_perf_read_unlock(int idx)
245	{
246	preempt_enable();
247	}
248
249	static struct rcu_perf_ops sched_ops = {
250	.ptype = RCU_SCHED_FLAVOR,
251	.init = rcu_sync_perf_init,
252	.readlock = sched_perf_read_lock,
253	.readunlock = sched_perf_read_unlock,
254	.started = rcu_batches_started_sched,
255	.completed = rcu_batches_completed_sched,
256	.exp_completed = rcu_exp_batches_completed_sched,
257	.sync = synchronize_sched,
258	.exp_sync = synchronize_sched_expedited,
259	.name = "sched"
260	};
261
262	#ifdef CONFIG_TASKS_RCU
263
264	/*
265	* Definitions for RCU-tasks perf testing.
266	*/
267
268	static int tasks_perf_read_lock(void)
269	{
270	return 0;
271	}
272
273	static void tasks_perf_read_unlock(int idx)
274	{
275	}
276
277	static struct rcu_perf_ops tasks_ops = {
278	.ptype = RCU_TASKS_FLAVOR,
279	.init = rcu_sync_perf_init,
280	.readlock = tasks_perf_read_lock,
281	.readunlock = tasks_perf_read_unlock,
282	.started = rcu_no_completed,
283	.completed = rcu_no_completed,
284	.sync = synchronize_rcu_tasks,
285	.exp_sync = synchronize_rcu_tasks,
286	.name = "tasks"
287	};
288
289	#define RCUPERF_TASKS_OPS &tasks_ops,
290
291	static bool __maybe_unused torturing_tasks(void)
292	{
293	return cur_ops == &tasks_ops;
294	}
295
296	#else /* #ifdef CONFIG_TASKS_RCU */
297
298	#define RCUPERF_TASKS_OPS
299
300	static bool __maybe_unused torturing_tasks(void)
301	{
302	return false;
303	}
304
305	#endif /* #else #ifdef CONFIG_TASKS_RCU */
306
307	/*
308	* If performance tests complete, wait for shutdown to commence.
309	*/
310	static void rcu_perf_wait_shutdown(void)
311	{
312	cond_resched_rcu_qs();
313	if (atomic_read(&n_rcu_perf_writer_finished) < nrealwriters)
314	return;
315	while (!torture_must_stop())
316	schedule_timeout_uninterruptible(1);
317	}
318
319	/*
320	* RCU perf reader kthread. Repeatedly does empty RCU read-side
321	* critical section, minimizing update-side interference.
322	*/
323	static int
324	rcu_perf_reader(void *arg)
325	{
326	unsigned long flags;
327	int idx;
6b558c4c	328	long me = (long)arg;
8704baab PM	329
8704baab PM	330	VERBOSE_PERFOUT_STRING("rcu_perf_reader task started");
6b558c4c	331	set_cpus_allowed_ptr(current, cpumask_of(me % nr_cpu_ids));
8704baab PM	332	set_user_nice(current, MAX_NICE);
	333	atomic_inc(&n_rcu_perf_reader_started);
	334
	335	do {
	336	local_irq_save(flags);
	337	idx = cur_ops->readlock();
	338	cur_ops->readunlock(idx);
	339	local_irq_restore(flags);
	340	rcu_perf_wait_shutdown();
	341	} while (!torture_must_stop());
	342	torture_kthread_stopping("rcu_perf_reader");
	343	return 0;
	344	}
	345
	346	/*
	347	* RCU perf writer kthread. Repeatedly does a grace period.
	348	*/
	349	static int
	350	rcu_perf_writer(void *arg)
	351	{
	352	int i = 0;
	353	int i_max;
	354	long me = (long)arg;
2094c995	355	struct sched_param sp;
8704baab PM	356	bool started = false, done = false, alldone = false;
	357	u64 t;
	358	u64 *wdp;
	359	u64 *wdpp = writer_durations[me];
	360
	361	VERBOSE_PERFOUT_STRING("rcu_perf_writer task started");
8704baab	362	WARN_ON(!wdpp);
6b558c4c	363	set_cpus_allowed_ptr(current, cpumask_of(me % nr_cpu_ids));
2094c995 PM	364	sp.sched_priority = 1;
2094c995 PM	365	sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);
df37e66b PM	366
	367	if (holdoff)
	368	schedule_timeout_uninterruptible(holdoff * HZ);
	369
8704baab PM	370	t = ktime_get_mono_fast_ns();
	371	if (atomic_inc_return(&n_rcu_perf_writer_started) >= nrealwriters) {
	372	t_rcu_perf_writer_started = t;
	373	if (gp_exp) {
	374	b_rcu_perf_writer_started =
	375	cur_ops->exp_completed() / 2;
	376	} else {
	377	b_rcu_perf_writer_started =
	378	cur_ops->completed();
	379	}
	380	}
	381
	382	do {
	383	wdp = &wdpp[i];
	384	*wdp = ktime_get_mono_fast_ns();
	385	if (gp_exp) {
	386	rcu_perf_writer_state = RTWS_EXP_SYNC;
	387	cur_ops->exp_sync();
	388	} else {
	389	rcu_perf_writer_state = RTWS_SYNC;
	390	cur_ops->sync();
	391	}
	392	rcu_perf_writer_state = RTWS_IDLE;
	393	t = ktime_get_mono_fast_ns();
	394	wdp = t - wdp;
	395	i_max = i;
	396	if (!started &&
	397	atomic_read(&n_rcu_perf_writer_started) >= nrealwriters)
	398	started = true;
	399	if (!done && i >= MIN_MEAS) {
	400	done = true;
620316e5 PM	401	sp.sched_priority = 0;
	402	sched_setscheduler_nocheck(current,
	403	SCHED_NORMAL, &sp);
a56fefa2 SP	404	pr_alert("%s%s rcu_perf_writer %ld has %d measurements\n",
a56fefa2 SP	405	perf_type, PERF_FLAG, me, MIN_MEAS);
8704baab PM	406	if (atomic_inc_return(&n_rcu_perf_writer_finished) >=
8704baab PM	407	nrealwriters) {
620316e5	408	schedule_timeout_interruptible(10);
ac2bb275	409	rcu_ftrace_dump(DUMP_ALL);
8704baab PM	410	PERFOUT_STRING("Test complete");
	411	t_rcu_perf_writer_finished = t;
	412	if (gp_exp) {
	413	b_rcu_perf_writer_finished =
	414	cur_ops->exp_completed() / 2;
	415	} else {
	416	b_rcu_perf_writer_finished =
	417	cur_ops->completed();
	418	}
e6fb1fc1 AS	419	if (shutdown) {
	420	smp_mb(); /* Assign before wake. */
	421	wake_up(&shutdown_wq);
	422	}
8704baab PM	423	}
	424	}
	425	if (done && !alldone &&
	426	atomic_read(&n_rcu_perf_writer_finished) >= nrealwriters)
	427	alldone = true;
	428	if (started && !alldone && i < MAX_MEAS - 1)
	429	i++;
	430	rcu_perf_wait_shutdown();
	431	} while (!torture_must_stop());
	432	rcu_perf_writer_state = RTWS_STOPPING;
	433	writer_n_durations[me] = i_max;
	434	torture_kthread_stopping("rcu_perf_writer");
	435	return 0;
	436	}
	437
	438	static inline void
	439	rcu_perf_print_module_parms(struct rcu_perf_ops cur_ops, const char tag)
	440	{
	441	pr_alert("%s" PERF_FLAG
	442	"--- %s: nreaders=%d nwriters=%d verbose=%d shutdown=%d\n",
	443	perf_type, tag, nrealreaders, nrealwriters, verbose, shutdown);
	444	}
	445
	446	static void
	447	rcu_perf_cleanup(void)
	448	{
	449	int i;
	450	int j;
	451	int ngps = 0;
	452	u64 *wdp;
	453	u64 *wdpp;
	454
	455	if (torture_cleanup_begin())
	456	return;
	457
	458	if (reader_tasks) {
	459	for (i = 0; i < nrealreaders; i++)
	460	torture_stop_kthread(rcu_perf_reader,
	461	reader_tasks[i]);
	462	kfree(reader_tasks);
	463	}
	464
	465	if (writer_tasks) {
	466	for (i = 0; i < nrealwriters; i++) {
	467	torture_stop_kthread(rcu_perf_writer,
	468	writer_tasks[i]);
	469	if (!writer_n_durations)
	470	continue;
	471	j = writer_n_durations[i];
	472	pr_alert("%s%s writer %d gps: %d\n",
	473	perf_type, PERF_FLAG, i, j);
	474	ngps += j;
	475	}
	476	pr_alert("%s%s start: %llu end: %llu duration: %llu gps: %d batches: %ld\n",
	477	perf_type, PERF_FLAG,
	478	t_rcu_perf_writer_started, t_rcu_perf_writer_finished,
	479	t_rcu_perf_writer_finished -
	480	t_rcu_perf_writer_started,
	481	ngps,
	482	b_rcu_perf_writer_finished -
	483	b_rcu_perf_writer_started);
	484	for (i = 0; i < nrealwriters; i++) {
	485	if (!writer_durations)
	486	break;
487	if (!writer_n_durations)
488	continue;
489	wdpp = writer_durations[i];
490	if (!wdpp)
491	continue;
492	for (j = 0; j <= writer_n_durations[i]; j++) {
493	wdp = &wdpp[j];
494	pr_alert("%s%s %4d writer-duration: %5d %llu\n",
495	perf_type, PERF_FLAG,
496	i, j, *wdp);
497	if (j % 100 == 0)
498	schedule_timeout_uninterruptible(1);
499	}
500	kfree(writer_durations[i]);
501	}
502	kfree(writer_tasks);
503	kfree(writer_durations);
504	kfree(writer_n_durations);
505	}
506
507	/* Do flavor-specific cleanup operations. */
508	if (cur_ops->cleanup != NULL)
509	cur_ops->cleanup();
510
511	torture_cleanup_end();
512	}
513
514	/*
515	* Return the number if non-negative. If -1, the number of CPUs.
516	* If less than -1, that much less than the number of CPUs, but
517	* at least one.
518	*/
519	static int compute_real(int n)
520	{
521	int nr;
522
523	if (n >= 0) {
524	nr = n;
525	} else {
526	nr = num_online_cpus() + 1 + n;
527	if (nr <= 0)
528	nr = 1;
529	}
530	return nr;
531	}
532
533	/*
534	* RCU perf shutdown kthread. Just waits to be awakened, then shuts
535	* down system.
536	*/
537	static int
538	rcu_perf_shutdown(void *arg)
539	{
540	do {
541	wait_event(shutdown_wq,
542	atomic_read(&n_rcu_perf_writer_finished) >=
543	nrealwriters);
544	} while (atomic_read(&n_rcu_perf_writer_finished) < nrealwriters);
545	smp_mb(); /* Wake before output. */
546	rcu_perf_cleanup();
547	kernel_power_off();
548	return -EINVAL;
549	}
550
551	static int __init
552	rcu_perf_init(void)
553	{
554	long i;
555	int firsterr = 0;
556	static struct rcu_perf_ops *perf_ops[] = {
557	&rcu_ops, &rcu_bh_ops, &srcu_ops, &sched_ops,
558	RCUPERF_TASKS_OPS
559	};
560
561	if (!torture_init_begin(perf_type, verbose, &perf_runnable))
562	return -EBUSY;
563
564	/* Process args and tell the world that the perf'er is on the job. */
565	for (i = 0; i < ARRAY_SIZE(perf_ops); i++) {
566	cur_ops = perf_ops[i];
567	if (strcmp(perf_type, cur_ops->name) == 0)
568	break;
569	}
570	if (i == ARRAY_SIZE(perf_ops)) {
571	pr_alert("rcu-perf: invalid perf type: \"%s\"\n",
572	perf_type);
573	pr_alert("rcu-perf types:");
574	for (i = 0; i < ARRAY_SIZE(perf_ops); i++)
575	pr_alert(" %s", perf_ops[i]->name);
576	pr_alert("\n");
577	firsterr = -EINVAL;
578	goto unwind;
579	}
580	if (cur_ops->init)
581	cur_ops->init();
582
583	nrealwriters = compute_real(nwriters);
584	nrealreaders = compute_real(nreaders);
585	atomic_set(&n_rcu_perf_reader_started, 0);
586	atomic_set(&n_rcu_perf_writer_started, 0);
587	atomic_set(&n_rcu_perf_writer_finished, 0);
588	rcu_perf_print_module_parms(cur_ops, "Start of test");
589
590	/* Start up the kthreads. */
591
592	if (shutdown) {
593	init_waitqueue_head(&shutdown_wq);
594	firsterr = torture_create_kthread(rcu_perf_shutdown, NULL,
595	shutdown_task);
596	if (firsterr)
597	goto unwind;
598	schedule_timeout_uninterruptible(1);
599	}
600	reader_tasks = kcalloc(nrealreaders, sizeof(reader_tasks[0]),
601	GFP_KERNEL);
602	if (reader_tasks == NULL) {
603	VERBOSE_PERFOUT_ERRSTRING("out of memory");
604	firsterr = -ENOMEM;
605	goto unwind;
606	}
607	for (i = 0; i < nrealreaders; i++) {
6b558c4c	608	firsterr = torture_create_kthread(rcu_perf_reader, (void *)i,
8704baab PM	609	reader_tasks[i]);
	610	if (firsterr)
	611	goto unwind;
	612	}
	613	while (atomic_read(&n_rcu_perf_reader_started) < nrealreaders)
	614	schedule_timeout_uninterruptible(1);
	615	writer_tasks = kcalloc(nrealwriters, sizeof(reader_tasks[0]),
	616	GFP_KERNEL);
	617	writer_durations = kcalloc(nrealwriters, sizeof(*writer_durations),
	618	GFP_KERNEL);
	619	writer_n_durations =
	620	kcalloc(nrealwriters, sizeof(*writer_n_durations),
	621	GFP_KERNEL);
	622	if (!writer_tasks \|\| !writer_durations \|\| !writer_n_durations) {
	623	VERBOSE_PERFOUT_ERRSTRING("out of memory");
	624	firsterr = -ENOMEM;
	625	goto unwind;
	626	}
af06d4f7 BF	627	if (rcu_gp_is_expedited() && !rcu_gp_is_normal() && !gp_exp) {
	628	VERBOSE_PERFOUT_ERRSTRING("All grace periods expedited, no normal ones to measure!");
	629	firsterr = -EINVAL;
	630	goto unwind;
	631	}
	632	if (rcu_gp_is_normal() && gp_exp) {
	633	VERBOSE_PERFOUT_ERRSTRING("All grace periods normal, no expedited ones to measure!");
	634	firsterr = -EINVAL;
	635	goto unwind;
	636	}
8704baab PM	637	for (i = 0; i < nrealwriters; i++) {
	638	writer_durations[i] =
	639	kcalloc(MAX_MEAS, sizeof(*writer_durations[i]),
	640	GFP_KERNEL);
05dbbfe7 WY	641	if (!writer_durations[i]) {
05dbbfe7 WY	642	firsterr = -ENOMEM;
8704baab	643	goto unwind;
05dbbfe7	644	}
8704baab PM	645	firsterr = torture_create_kthread(rcu_perf_writer, (void *)i,
	646	writer_tasks[i]);
	647	if (firsterr)
	648	goto unwind;
	649	}
	650	torture_init_end();
	651	return 0;
	652
	653	unwind:
	654	torture_init_end();
	655	rcu_perf_cleanup();
	656	return firsterr;
	657	}
	658
	659	module_init(rcu_perf_init);
	660	module_exit(rcu_perf_cleanup);