Waketorture, not for inclusion

371 lines not on this page · raw message

index f433959e9322..88d3f9590322 100644
@@ -1306,7 +1306,8 @@ static void rcu_stall_kick_kthreads(struct rcu_state *rsp)
 	if (!rcu_kick_kthreads)
 		return;
 	j = READ_ONCE(rsp->jiffies_kick_kthreads);
-	if (time_after(jiffies, j) && rsp->gp_kthread) {
+	if (time_after(jiffies, j) && rsp->gp_kthread &&
+	    (rcu_gp_in_progress(rsp) || READ_ONCE(rsp->gp_flags))) {
 		WARN_ONCE(1, "Kicking %s grace-period kthread\n", rsp->name);
 		rcu_ftrace_dump(DUMP_ALL);
 		wake_up_process(rsp->gp_kthread);

index 75bc547206cc..3068a5c858c8 100644
@@ -81,7 +81,9 @@ static unsigned long *waiter_iter;	/* Number of wait iterations. */
 static bool *waiter_cts;		/* Waiter already checked. */
 static unsigned long *waiter_kicks;	/* Number of waiter starvations. */
 static unsigned long *waiter_ts;	/* Jiffies last run. */
-DEFINE_MUTEX(waiter_mutex);
+static DEFINE_MUTEX(waiter_mutex);
+static DEFINE_PER_CPU(u64, waiter_cputime); /* Nanoseconds. */
+static u64 starttime;
 
 static int torture_runnable = IS_ENABLED(MODULE);
 module_param(torture_runnable, int, 0444);

@@ -133,16 +135,23 @@ static int wake_torture_waiter(void *arg)
 {
 	int i;
 	long me = (long)arg;
+	u64 ts;
 
 	VERBOSE_TOROUT_STRING("wake_torture_waiter task started");
 	set_user_nice(current, MAX_NICE);
 
+	preempt_disable();
+	ts = trace_clock_local();
 	do {
 		waiter_ts[me] = jiffies;
 		smp_mb(); /* Ensure waiter_ts[] written before waiter_cts[]. */
 			  /* Pairs with [A]. */
 		waiter_cts[me] = false;
+		__this_cpu_add(waiter_cputime, trace_clock_local() - ts);
+		preempt_enable();
 		cur_ops->wait(wait_duration);
+		preempt_disable();
+		ts = trace_clock_local();
 		waiter_iter[me]++;
 		for (i = 0; i < nrealwaiters; i++) {
 			if (waiter_done[i] ||

@@ -164,8 +173,14 @@ static int wake_torture_waiter(void *arg)
 				mutex_unlock(&waiter_mutex);
 			}
 		}
+		__this_cpu_add(waiter_cputime, trace_clock_local() - ts);
+		preempt_enable();
 		torture_shutdown_absorb("wake_torture_waiter");
+		preempt_disable();
+		ts = trace_clock_local();
 	} while (!torture_must_stop());
+	__this_cpu_add(waiter_cputime, trace_clock_local() - ts);
+	preempt_enable();
 	mutex_lock(&waiter_mutex);
 	waiter_done[me] = true;
 	mutex_unlock(&waiter_mutex);

@@ -189,6 +204,8 @@ wake_torture_stats_print(void)
 {
 	int i;
 	bool tardy = false;
+	u64 timediff;
+	u64 timetot;
 
 	if (!waiter_done || !waiter_iter || !waiter_cts ||
 	    !waiter_kicks || !waiter_ts) {

@@ -212,6 +229,15 @@ wake_torture_stats_print(void)
 		pr_cont("\n");
 	else
 		TOROUT_STRING(" No tardy kthreads");
+	timediff = (trace_clock_global() - starttime) / 1000;
+	timetot = 0;
+	for_each_possible_cpu(i)
+		timetot += READ_ONCE(per_cpu(waiter_cputime, i));
+	timetot /= nr_cpu_ids;
+	timetot /= timediff;
+	pr_alert("%s" TORTURE_FLAG " timediff: %llu utilization: %llu.%llu nr_cpu_ids: %d\n",
+		 torture_type, timediff,
+		 timetot / 1000ULL, timetot % 1000ULL, nr_cpu_ids);
 }
 
 /*

@@ -290,6 +316,7 @@ wake_torture_init(void)
 
 	if (!torture_init_begin(torture_type, verbose, &torture_runnable))
 		return -EBUSY;
+	starttime = trace_clock_global();
 
 	/* Process args and tell the world that the torturer is on the job. */
 	for (i = 0; i < ARRAY_SIZE(torture_ops); i++) {

index 3068a5c858c8..bf04c65296bd 100644
@@ -62,12 +62,12 @@ torture_param(int, stat_interval, 60,
 	     "Number of seconds between stats printk()s");
 torture_param(bool, verbose, true,
 	     "Enable verbose debugging printk()s");
-torture_param(int, wait_duration, 3,
-	     "Number of jiffies to wait each iteration");
+torture_param(int, wait_duration, 127,
+	     "Number of microseconds to wait each iteration");
 torture_param(int, wait_grace, 20,
 	     "Number of jiffies before complaining about long wait");
 
-static char *torture_type = "sti";
+static char *torture_type = "sh";
 module_param(torture_type, charp, 0444);
 MODULE_PARM_DESC(torture_type, "Type of wait to torture (sti, stui, ...)");
 

@@ -94,18 +94,40 @@ MODULE_PARM_DESC(torture_runnable, "Start waketorture at boot");
  */
 
 struct wake_torture_ops {
-	signed long (*wait)(signed long timeout);
+	void (*wait)(void);
 	const char *name;
 };
 
 static struct wake_torture_ops *cur_ops;
 
 /*
+ * Definitions for schedule_hrtimeout() torture testing.
+ */
+
+static void wait_schedule_hrtimeout(void)
+{
+	ktime_t wait = ns_to_ktime(wait_duration * 1000);
+
+	set_current_state(TASK_UNINTERRUPTIBLE);
+	schedule_hrtimeout(&wait, HRTIMER_MODE_REL);
+}
+
+static struct wake_torture_ops sh_ops = {
+	.wait		= wait_schedule_hrtimeout,
+	.name		= "sh"
+};
+
+/*
  * Definitions for schedule_timeout_interruptible() torture testing.
  */
 
+static void wait_schedule_timeout_interruptible(void)
+{
+	schedule_timeout_interruptible((wait_duration + 999) / 1000);
+}
+
 static struct wake_torture_ops sti_ops = {
-	.wait		= schedule_timeout_interruptible,
+	.wait		= wait_schedule_timeout_interruptible,
 	.name		= "sti"
 };
 

@@ -113,8 +135,13 @@ static struct wake_torture_ops sti_ops = {
  * Definitions for schedule_timeout_uninterruptible() torture testing.
  */
 
+static void wait_schedule_timeout_uninterruptible(void)
+{
+	schedule_timeout_uninterruptible((wait_duration + 999) / 1000);
+}
+
 static struct wake_torture_ops stui_ops = {
-	.wait		= schedule_timeout_uninterruptible,
+	.wait		= wait_schedule_timeout_uninterruptible,
 	.name		= "stui"
 };
 

@@ -149,7 +176,7 @@ static int wake_torture_waiter(void *arg)
 		waiter_cts[me] = false;
 		__this_cpu_add(waiter_cputime, trace_clock_local() - ts);
 		preempt_enable();
-		cur_ops->wait(wait_duration);
+		cur_ops->wait();
 		preempt_disable();
 		ts = trace_clock_local();
 		waiter_iter[me]++;

@@ -167,7 +194,10 @@ static int wake_torture_waiter(void *arg)
 			} else {
 				waiter_cts[i] = true;
 				waiter_kicks[i]++;
-				pr_alert("%s%s wake_torture_waiter(): P%d failing to awaken!\n", torture_type, TORTURE_FLAG, waiter_tasks[i]->pid);
+				pr_alert("%s%s wake_torture_waiter(): P%d (%#lx) failing to awaken!\n",
+					 torture_type, TORTURE_FLAG,
+					 waiter_tasks[i]->pid,
+					 waiter_tasks[i]->state);
 				rcu_ftrace_dump(DUMP_ALL);
 				wake_up_process(waiter_tasks[i]);
 				mutex_unlock(&waiter_mutex);

@@ -312,7 +342,9 @@ wake_torture_init(void)
 {
 	int i;
 	int firsterr = 0;
-	static struct wake_torture_ops *torture_ops[] = { &sti_ops, &stui_ops };
+	static struct wake_torture_ops *torture_ops[] = {
+		&sh_ops, &sti_ops, &stui_ops
+	};
 
 	if (!torture_init_begin(torture_type, verbose, &torture_runnable))
 		return -EBUSY;

index bf04c65296bd..8568688607ac 100644
@@ -74,6 +74,7 @@ MODULE_PARM_DESC(torture_type, "Type of wait to torture (sti, stui, ...)");
 static int nrealwaiters;
 static struct task_struct **waiter_tasks;
 static struct task_struct *stats_task;
+static struct task_struct *onoff_task;
 
 /* Yes, these cache-thrash, and it is inherent to the concurrency design. */
 static bool *waiter_done;		/* Waiter is done, don't wake. */

@@ -85,6 +86,18 @@ static DEFINE_MUTEX(waiter_mutex);
 static DEFINE_PER_CPU(u64, waiter_cputime); /* Nanoseconds. */
 static u64 starttime;
 
+static int onoff_cpu = -1;
+static long n_offline_attempts;
+static long n_offline_successes;
+static unsigned long sum_offline;
+static int min_offline = -1;
+static int max_offline;
+static long n_online_attempts;
+static long n_online_successes;
+static unsigned long sum_online;
+static int min_online = -1;
+static int max_online;
+
 static int torture_runnable = IS_ENABLED(MODULE);
 module_param(torture_runnable, int, 0444);
 MODULE_PARM_DESC(torture_runnable, "Start waketorture at boot");

@@ -219,6 +232,67 @@ static int wake_torture_waiter(void *arg)
 }
 
 /*
+ * Find a hotpluggable CPU and repeatedly take it online and offline.
+ */
+static int wake_torture_onoff(void *args)
+{
+	int cpu;
+
+	VERBOSE_TOROUT_STRING("wake_torture_onoff task started");
+	if (onoff_holdoff > 0) {
+		VERBOSE_TOROUT_STRING("wake_torture_onoff begin holdoff");
+		schedule_timeout_interruptible(onoff_holdoff * HZ);
+		VERBOSE_TOROUT_STRING("wake_torture_onoff end holdoff");
+	}
+	for_each_online_cpu(cpu) {
+		if (cpu_is_hotpluggable(cpu))
+			onoff_cpu = cpu;
+	}
+	pr_alert("%s" TORTURE_FLAG " wake_torture_onoff: onoff_cpu: %d\n", torture_type, onoff_cpu);
+	if (onoff_cpu < 0)
+		VERBOSE_TOROUT_STRING("wake_torture_onoff: no hotpluggable CPUs!");
+	while (!torture_must_stop() && onoff_cpu >= 0) {
+		if (!torture_offline(onoff_cpu,
+				    &n_offline_attempts, &n_offline_successes,
+				    &sum_offline, &min_offline, &max_offline))
+			torture_online(onoff_cpu,
+				       &n_online_attempts, &n_online_successes,
+				       &sum_online, &min_online, &max_online);
+		schedule_timeout_interruptible(onoff_interval);
+	}
+	torture_kthread_stopping("wake_torture_onoff");
+	return 0;
+}
+
+/*
+ * Initiate waketorture-specific online-offline handling, which
+ * focuses on a single CPU.
+ */
+static int wake_torture_onoff_init(void)
+{
+	int ret = 0;
+
+	if (!IS_ENABLED(CONFIG_HOTPLUG_CPU))
+		return ret;
+	if (onoff_interval <= 0)
+		return 0;
+	ret = torture_create_kthread(wake_torture_onoff, NULL, onoff_task);
+	return ret;
+}
+
+/*
+ * Clean up after waketorture-specific online-offline handling.
+ */
+static void wake_torture_onoff_cleanup(void)
+{
+	if (!IS_ENABLED(CONFIG_HOTPLUG_CPU))
+		return;
+	VERBOSE_TOROUT_STRING("Stopping wake_torture_onoff task");
+	kthread_stop(onoff_task);
+	onoff_task = NULL;
+}
+
+/*
  * Print torture statistics.  Caller must ensure that there is only one
  * call to this function at a given time!!!  This is normally accomplished
  * by relying on the module system to only have one copy of the module

@@ -265,9 +339,14 @@ wake_torture_stats_print(void)
 		timetot += READ_ONCE(per_cpu(waiter_cputime, i));
 	timetot /= nr_cpu_ids;
 	timetot /= timediff;
-	pr_alert("%s" TORTURE_FLAG " timediff: %llu utilization: %llu.%llu nr_cpu_ids: %d\n",
+	pr_alert("%s" TORTURE_FLAG " timediff: %llu utilization: %llu.%llu nr_cpu_ids: %d onoff: %ld/%ld:%ld/%ld %d,%d:%d,%d %lu:%lu (HZ=%d)\n",
 		 torture_type, timediff,
-		 timetot / 1000ULL, timetot % 1000ULL, nr_cpu_ids);
+		 timetot / 1000ULL, timetot % 1000ULL, nr_cpu_ids,
+		 n_online_successes, n_online_attempts,
+		 n_offline_successes, n_offline_attempts,
+		 min_online, max_online,
+		 min_offline, max_offline,
+		 sum_online, sum_offline, HZ);
 }
 
 /*

@@ -307,6 +386,9 @@ wake_torture_cleanup(void)
 
 	(void)torture_cleanup_begin();
 
+	if (onoff_task)
+		wake_torture_onoff_cleanup();
+
 	if (waiter_tasks) {
 		for (i = 0; i < nrealwaiters; i++)
 			torture_stop_kthread(wake_torture_waiter,

@@ -414,7 +496,7 @@ wake_torture_init(void)
 	firsterr = torture_shutdown_init(shutdown_secs, wake_torture_cleanup);
 	if (firsterr)
 		goto unwind;
-	firsterr = torture_onoff_init(onoff_holdoff * HZ, onoff_interval);
+	firsterr = wake_torture_onoff_init();
 	if (firsterr)
 		goto unwind;
 	torture_init_end();

index 8568688607ac..f9843eb3f7b7 100644
@@ -248,9 +248,15 @@ static int wake_torture_onoff(void *args)
 		if (cpu_is_hotpluggable(cpu))
 			onoff_cpu = cpu;
 	}
-	pr_alert("%s" TORTURE_FLAG " wake_torture_onoff: onoff_cpu: %d\n", torture_type, onoff_cpu);
-	if (onoff_cpu < 0)
+	if (onoff_cpu < 0) {
 		VERBOSE_TOROUT_STRING("wake_torture_onoff: no hotpluggable CPUs!");
+		if (shutdown_secs > 0) {
+			VERBOSE_TOROUT_STRING("wake_torture_onoff: Shutting down");
+			kernel_power_off();
+			VERBOSE_TOROUT_STRING("wake_torture_onoff: Survived kernel_power_off()?");
+		}
+	}
+	pr_alert("%s" TORTURE_FLAG " wake_torture_onoff: onoff_cpu: %d\n", torture_type, onoff_cpu);
 	while (!torture_must_stop() && onoff_cpu >= 0) {
 		if (!torture_offline(onoff_cpu,
 				    &n_offline_attempts, &n_offline_successes,

index f9843eb3f7b7..68be40400f51 100644
@@ -236,7 +236,9 @@ static int wake_torture_waiter(void *arg)
  */
 static int wake_torture_onoff(void *args)
 {
+	cpumask_var_t cm;
 	int cpu;
+	int i;
 
 	VERBOSE_TOROUT_STRING("wake_torture_onoff task started");
 	if (onoff_holdoff > 0) {

@@ -244,6 +246,11 @@ static int wake_torture_onoff(void *args)
 		schedule_timeout_interruptible(onoff_holdoff * HZ);
 		VERBOSE_TOROUT_STRING("wake_torture_onoff end holdoff");
 	}
+
+	/*
+	 * Find the last hotpluggable CPU, and affinity the waiter
+	 * tasks elsewhere.
+	 */
 	for_each_online_cpu(cpu) {
 		if (cpu_is_hotpluggable(cpu))
 			onoff_cpu = cpu;

@@ -257,6 +264,18 @@ static int wake_torture_onoff(void *args)
 		}
 	}
 	pr_alert("%s" TORTURE_FLAG " wake_torture_onoff: onoff_cpu: %d\n", torture_type, onoff_cpu);
+	if (!zalloc_cpumask_var(&cm, GFP_KERNEL)) {
+		VERBOSE_TOROUT_STRING("wake_torture_onoff: Out of memory, no affinity");
+	} else {
+		cpumask_copy(cm, cpu_online_mask);
+		cpumask_clear_cpu(onoff_cpu, cm);
+		if (cpumask_weight(cm) == 0)
+			cpumask_setall(cm);
+		for (i = 0; i < nrealwaiters; i++)
+			set_cpus_allowed_ptr(waiter_tasks[i], cm);
+	}
+
+	/* Cycle the victim CPU online and offline! */
 	while (!torture_must_stop() && onoff_cpu >= 0) {
 		if (!torture_offline(onoff_cpu,
 				    &n_offline_attempts, &n_offline_successes,