Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1716282 > unrolled thread

[PATCH] RFC: hung task: Check specific tasks for long uninterruptible sleep state

Back to article view | Back to linux.kernel

Contents

  [PATCH] RFC: hung task: Check specific tasks for long uninterruptible sleep state Imran Khan <kimran@codeaurora.org> - 2017-08-21 12:30 +0200
    RE: [PATCH] RFC: hung task: Check specific tasks for long  uninterruptible sleep state "Zhang, Shile (NSB - CN/Hangzhou)" <shile.zhang@nokia-sbell.com> - 2017-08-22 04:00 +0200
      Re: [PATCH] RFC: hung task: Check specific tasks for long  uninterruptible sleep state Imran Khan <kimran@codeaurora.org> - 2017-08-22 08:10 +0200
        RE: [PATCH] RFC: hung task: Check specific tasks for long  uninterruptible sleep state "Zhang, Shile (NSB - CN/Hangzhou)" <shile.zhang@nokia-sbell.com> - 2017-08-22 08:50 +0200
    Re: [PATCH] RFC: hung task: Check specific tasks for long  uninterruptible sleep state Peter Zijlstra <peterz@infradead.org> - 2017-08-22 22:40 +0200
      Re: [PATCH] RFC: hung task: Check specific tasks for long  uninterruptible sleep state Imran Khan <kimran@codeaurora.org> - 2017-08-23 14:30 +0200

#1716282 — [PATCH] RFC: hung task: Check specific tasks for long uninterruptible sleep state

khungtask by default monitors either all tasks or no tasks at all
for long unterruptible sleeps.
For Android like environments this arrangement is not optimal because
on one hand it may be permissible to have some background(bg) task
in uninterruptible sleep state for long duration while on the other
hand it may not be permissible to have some foreground(fg) task like
surfaceflinger in uninterruptible sleep state for long duration.
So it would be good to have some arrangement so that few specified tasks
can be monitored by khungtaskd, on a need basis.
This change introduces a sysctl option, /proc/sys/kernel/
hung_task_check_selected, to enable monitoring of selected tasks using
khungtask daemon. If this sysctl option is enabled then only the tasks
specified in /proc/hung_task_monitor_list are monitored otherwise all
tasks are monitored, just like the default case.

Signed-off-by: Imran Khan <kimran@codeaurora.org>
---
 include/linux/sched/sysctl.h |   1 +
 kernel/hung_task.c           | 121 ++++++++++++++++++++++++++++++++++++++++++-
 kernel/sysctl.c              |   8 +++
 3 files changed, 128 insertions(+), 2 deletions(-)

diff --git a/include/linux/sched/sysctl.h b/include/linux/sched/sysctl.h
index 0f5ecd4..05892f1 100644
--- a/include/linux/sched/sysctl.h
+++ b/include/linux/sched/sysctl.h
@@ -10,6 +10,7 @@
 extern unsigned int  sysctl_hung_task_panic;
 extern unsigned long sysctl_hung_task_timeout_secs;
 extern int sysctl_hung_task_warnings;
+extern int sysctl_hung_task_check_selected;
 extern int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
 					 void __user *buffer,
 					 size_t *lenp, loff_t *ppos);
diff --git a/kernel/hung_task.c b/kernel/hung_task.c
index 751593e..49f13fb 100644
--- a/kernel/hung_task.c
+++ b/kernel/hung_task.c
@@ -16,12 +16,28 @@
 #include <linux/export.h>
 #include <linux/sysctl.h>
 #include <linux/utsname.h>
+#include <linux/uaccess.h>
+#include <linux/slab.h>
+#include <linux/proc_fs.h>
 #include <linux/sched/signal.h>
 #include <linux/sched/debug.h>
 
 #include <trace/events/sched.h>
 
 /*
+ * Hung task that needs monitoring
+ */
+struct hung_task {
+	struct list_head list;
+	char comm[TASK_COMM_LEN];
+};
+
+static struct hung_task	*monitor_list;
+int sysctl_hung_task_check_selected;
+
+
+
+/*
  * The number of tasks checked:
  */
 int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
@@ -76,6 +92,92 @@ static int __init hung_task_panic_setup(char *str)
 	.notifier_call = hung_task_panic,
 };
 
+static void hung_task_monitor_setup(void)
+{
+	monitor_list = kmalloc(sizeof(*monitor_list), GFP_KERNEL);
+	if (monitor_list) {
+		INIT_LIST_HEAD(&monitor_list->list);
+		memset(monitor_list->comm, 0, TASK_COMM_LEN);
+	}
+}
+
+
+static  int hung_task_info_show(struct seq_file *m, void *v)
+{
+	struct hung_task *ht;
+
+	ht = list_entry(v, struct hung_task, list);
+	seq_puts(m, ht->comm);
+
+	return 0;
+}
+
+static void *hung_task_info_start(struct seq_file *m, loff_t *pos)
+{
+	return seq_list_start_head(&monitor_list->list, *pos);
+}
+
+static void *hung_task_info_next(struct seq_file *m, void *v, loff_t *pos)
+{
+	return seq_list_next(v, &monitor_list->list, pos);
+}
+
+static void hung_task_info_stop(struct seq_file *m, void *v)
+{
+}
+
+const struct seq_operations hung_task_info_op = {
+	.start	= hung_task_info_start,
+	.next	= hung_task_info_next,
+	.stop	= hung_task_info_stop,
+	.show	= hung_task_info_show
+};
+
+static int hung_task_info_open(struct inode *inode, struct file *file)
+{
+	return seq_open(file, &hung_task_info_op);
+}
+
+static ssize_t
+hung_task_info_write(struct file *file, const char __user *buf, size_t count,
+	     loff_t *offs)
+{
+	struct task_struct *g, *t;
+	struct hung_task *ht = kmalloc(sizeof(*ht), GFP_KERNEL);
+
+	if (!ht)
+		return -ENOMEM;
+
+	if (copy_from_user(ht->comm, buf, count))
+		return -EFAULT;
+	ht->comm[count] = '\0';
+
+	for_each_process_thread(g, t) {
+		if (!strncmp(t->comm, ht->comm, strlen(t->comm))) {
+			list_add_tail(&ht->list, &monitor_list->list);
+			return count;
+		}
+	}
+
+	pr_err("Non-existing task: %s can't be monitored\n", ht->comm);
+	return count;
+}
+
+static const struct file_operations hung_task_info_operations = {
+	.open		= hung_task_info_open,
+	.read		= seq_read,
+	.write		= hung_task_info_write,
+	.llseek		= seq_lseek,
+	.release	= seq_release,
+};
+
+static int __init proc_hung_task_info_init(void)
+{
+	proc_create("hung_task_monitor_list", 0644, NULL,
+		    &hung_task_info_operations);
+	return 0;
+}
+
 static void check_hung_task(struct task_struct *t, unsigned long timeout)
 {
 	unsigned long switch_count = t->nvcsw + t->nivcsw;
@@ -167,6 +269,7 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout)
 	int max_count = sysctl_hung_task_check_count;
 	int batch_count = HUNG_TASK_BATCHING;
 	struct task_struct *g, *t;
+	struct hung_task *ht, *tmp;
 
 	/*
 	 * If the system crashed already then all bets are off,
@@ -186,9 +289,21 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout)
 				goto unlock;
 		}
 		/* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
-		if (t->state == TASK_UNINTERRUPTIBLE)
-			check_hung_task(t, timeout);
+		if (t->state == TASK_UNINTERRUPTIBLE) {
+			if (sysctl_hung_task_check_selected) {
+				list_for_each_entry_safe(ht, tmp,
+							 &monitor_list->list,
+							 list)
+					if (!strncmp(ht->comm, t->comm,
+					    strlen(t->comm)))
+					/* Task belongs to the selected group */
+						check_hung_task(t, timeout);
+			} else {
+				check_hung_task(t, timeout);
+			}
+		}
 	}
+
  unlock:
 	rcu_read_unlock();
 	if (hung_task_show_lock)
@@ -259,6 +374,8 @@ static int watchdog(void *dummy)
 static int __init hung_task_init(void)
 {
 	atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
+	hung_task_monitor_setup();
+	proc_hung_task_info_init();
 	watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
 
 	return 0;
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index 6648fbb..ab45774 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -1096,6 +1096,14 @@ static int sysrq_sysctl_handler(struct ctl_table *table, int write,
 		.proc_handler	= proc_dointvec_minmax,
 		.extra1		= &neg_one,
 	},
+	{
+		.procname	= "hung_task_check_selected",
+		.data		= &sysctl_hung_task_check_selected,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &neg_one,
+	},
 #endif
 #ifdef CONFIG_RT_MUTEXES
 	{
-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

[toc] | [next] | [standalone]

#1716969 — RE: [PATCH] RFC: hung task: Check specific tasks for long uninterruptible sleep state

Hi, Imran,

I think a "unmonitored list" is better than "monitor list", because we want khungtaskd can find out the "unexpected" hung task, but not few in a list.
Then, for the fg tasks, which can put it in the "unmonitored list", for the bg tasks, I think we can tweak the timeout to control the duration.

Thanks!

BR, Shile

-----Original Message-----
From: Imran Khan [mailto:kimran@codeaurora.org] 
Sent: Monday, August 21, 2017 6:26 PM
To: mingo@kernel.org
Cc: imrank140517@gmail.com; Imran Khan <kimran@codeaurora.org>; Luis R. Rodriguez <mcgrof@kernel.org>; Kees Cook <keescook@chromium.org>; Peter Zijlstra (Intel) <peterz@infradead.org>; Zhang, Shile (NSB - CN/Hangzhou) <shile.zhang@nokia-sbell.com>; Matt Fleming <matt@codeblueprint.co.uk>; Andrew Morton <akpm@linux-foundation.org>; Vegard Nossum <vegard.nossum@oracle.com>; Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>; John Siddle <jsiddle@redhat.com>; open list <linux-kernel@vger.kernel.org>; open list:PROC SYSCTL <linux-fsdevel@vger.kernel.org>
Subject: [PATCH] RFC: hung task: Check specific tasks for long uninterruptible sleep state

khungtask by default monitors either all tasks or no tasks at all
for long unterruptible sleeps.
For Android like environments this arrangement is not optimal because
on one hand it may be permissible to have some background(bg) task
in uninterruptible sleep state for long duration while on the other
hand it may not be permissible to have some foreground(fg) task like
surfaceflinger in uninterruptible sleep state for long duration.
So it would be good to have some arrangement so that few specified tasks
can be monitored by khungtaskd, on a need basis.
This change introduces a sysctl option, /proc/sys/kernel/
hung_task_check_selected, to enable monitoring of selected tasks using
khungtask daemon. If this sysctl option is enabled then only the tasks
specified in /proc/hung_task_monitor_list are monitored otherwise all
tasks are monitored, just like the default case.

Signed-off-by: Imran Khan <kimran@codeaurora.org>
---
 include/linux/sched/sysctl.h |   1 +
 kernel/hung_task.c           | 121 ++++++++++++++++++++++++++++++++++++++++++-
 kernel/sysctl.c              |   8 +++
 3 files changed, 128 insertions(+), 2 deletions(-)

diff --git a/include/linux/sched/sysctl.h b/include/linux/sched/sysctl.h
index 0f5ecd4..05892f1 100644
--- a/include/linux/sched/sysctl.h
+++ b/include/linux/sched/sysctl.h
@@ -10,6 +10,7 @@
 extern unsigned int  sysctl_hung_task_panic;
 extern unsigned long sysctl_hung_task_timeout_secs;
 extern int sysctl_hung_task_warnings;
+extern int sysctl_hung_task_check_selected;
 extern int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
 					 void __user *buffer,
 					 size_t *lenp, loff_t *ppos);
diff --git a/kernel/hung_task.c b/kernel/hung_task.c
index 751593e..49f13fb 100644
--- a/kernel/hung_task.c
+++ b/kernel/hung_task.c
@@ -16,12 +16,28 @@
 #include <linux/export.h>
 #include <linux/sysctl.h>
 #include <linux/utsname.h>
+#include <linux/uaccess.h>
+#include <linux/slab.h>
+#include <linux/proc_fs.h>
 #include <linux/sched/signal.h>
 #include <linux/sched/debug.h>
 
 #include <trace/events/sched.h>
 
 /*
+ * Hung task that needs monitoring
+ */
+struct hung_task {
+	struct list_head list;
+	char comm[TASK_COMM_LEN];
+};
+
+static struct hung_task	*monitor_list;
+int sysctl_hung_task_check_selected;
+
+
+
+/*
  * The number of tasks checked:
  */
 int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
@@ -76,6 +92,92 @@ static int __init hung_task_panic_setup(char *str)
 	.notifier_call = hung_task_panic,
 };
 
+static void hung_task_monitor_setup(void)
+{
+	monitor_list = kmalloc(sizeof(*monitor_list), GFP_KERNEL);
+	if (monitor_list) {
+		INIT_LIST_HEAD(&monitor_list->list);
+		memset(monitor_list->comm, 0, TASK_COMM_LEN);
+	}
+}
+
+
+static  int hung_task_info_show(struct seq_file *m, void *v)
+{
+	struct hung_task *ht;
+
+	ht = list_entry(v, struct hung_task, list);
+	seq_puts(m, ht->comm);
+
+	return 0;
+}
+
+static void *hung_task_info_start(struct seq_file *m, loff_t *pos)
+{
+	return seq_list_start_head(&monitor_list->list, *pos);
+}
+
+static void *hung_task_info_next(struct seq_file *m, void *v, loff_t *pos)
+{
+	return seq_list_next(v, &monitor_list->list, pos);
+}
+
+static void hung_task_info_stop(struct seq_file *m, void *v)
+{
+}
+
+const struct seq_operations hung_task_info_op = {
+	.start	= hung_task_info_start,
+	.next	= hung_task_info_next,
+	.stop	= hung_task_info_stop,
+	.show	= hung_task_info_show
+};
+
+static int hung_task_info_open(struct inode *inode, struct file *file)
+{
+	return seq_open(file, &hung_task_info_op);
+}
+
+static ssize_t
+hung_task_info_write(struct file *file, const char __user *buf, size_t count,
+	     loff_t *offs)
+{
+	struct task_struct *g, *t;
+	struct hung_task *ht = kmalloc(sizeof(*ht), GFP_KERNEL);
+
+	if (!ht)
+		return -ENOMEM;
+
+	if (copy_from_user(ht->comm, buf, count))
+		return -EFAULT;
+	ht->comm[count] = '\0';
+
+	for_each_process_thread(g, t) {
+		if (!strncmp(t->comm, ht->comm, strlen(t->comm))) {
+			list_add_tail(&ht->list, &monitor_list->list);
+			return count;
+		}
+	}
+
+	pr_err("Non-existing task: %s can't be monitored\n", ht->comm);
+	return count;
+}
+
+static const struct file_operations hung_task_info_operations = {
+	.open		= hung_task_info_open,
+	.read		= seq_read,
+	.write		= hung_task_info_write,
+	.llseek		= seq_lseek,
+	.release	= seq_release,
+};
+
+static int __init proc_hung_task_info_init(void)
+{
+	proc_create("hung_task_monitor_list", 0644, NULL,
+		    &hung_task_info_operations);
+	return 0;
+}
+
 static void check_hung_task(struct task_struct *t, unsigned long timeout)
 {
 	unsigned long switch_count = t->nvcsw + t->nivcsw;
@@ -167,6 +269,7 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout)
 	int max_count = sysctl_hung_task_check_count;
 	int batch_count = HUNG_TASK_BATCHING;
 	struct task_struct *g, *t;
+	struct hung_task *ht, *tmp;
 
 	/*
 	 * If the system crashed already then all bets are off,
@@ -186,9 +289,21 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout)
 				goto unlock;
 		}
 		/* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
-		if (t->state == TASK_UNINTERRUPTIBLE)
-			check_hung_task(t, timeout);
+		if (t->state == TASK_UNINTERRUPTIBLE) {
+			if (sysctl_hung_task_check_selected) {
+				list_for_each_entry_safe(ht, tmp,
+							 &monitor_list->list,
+							 list)
+					if (!strncmp(ht->comm, t->comm,
+					    strlen(t->comm)))
+					/* Task belongs to the selected group */
+						check_hung_task(t, timeout);
+			} else {
+				check_hung_task(t, timeout);
+			}
+		}
 	}
+
  unlock:
 	rcu_read_unlock();
 	if (hung_task_show_lock)
@@ -259,6 +374,8 @@ static int watchdog(void *dummy)
 static int __init hung_task_init(void)
 {
 	atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
+	hung_task_monitor_setup();
+	proc_hung_task_info_init();
 	watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
 
 	return 0;
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index 6648fbb..ab45774 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -1096,6 +1096,14 @@ static int sysrq_sysctl_handler(struct ctl_table *table, int write,
 		.proc_handler	= proc_dointvec_minmax,
 		.extra1		= &neg_one,
 	},
+	{
+		.procname	= "hung_task_check_selected",
+		.data		= &sysctl_hung_task_check_selected,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &neg_one,
+	},
 #endif
 #ifdef CONFIG_RT_MUTEXES
 	{
-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

[toc] | [prev] | [next] | [standalone]

#1717052 — Re: [PATCH] RFC: hung task: Check specific tasks for long uninterruptible sleep state

On 8/22/2017 7:24 AM, Zhang, Shile (NSB - CN/Hangzhou) wrote:

> Hi, Imran,
> 
Hi Shile,
Thanks for getting back on this.
> I think a "unmonitored list" is better than "monitor list", because we want khungtaskd can find out the "unexpected" hung task, but not few in a list.
> Then, for the fg tasks, which can put it in the "unmonitored list", for the bg tasks, I think we can tweak the timeout to control the duration.
> 
Here fg and bg tasks were just an example and may not be applicable for all environments (non-android). But main intention here is to have
some mechanism that would allow to monitor only specific tasks. As we have a sysctl option to enable/disable this mechanism we can at any time
fall back to default mechanism of monitoring all tasks.

Alternatively this can be done on task priority basis also, where we have a minimum priority and only tasks with priorities higher than
this min priority would be monitored and if we keep this min priority to the least priority , all tasks would be monitored like the 
default case. Please let me know your feedback regarding this approach.

Thanks and Regards,
Imran 
> Thanks!
> 
> BR, Shile
> 
> -----Original Message-----
> From: Imran Khan [mailto:kimran@codeaurora.org] 
> Sent: Monday, August 21, 2017 6:26 PM
> To: mingo@kernel.org
> Cc: imrank140517@gmail.com; Imran Khan <kimran@codeaurora.org>; Luis R. Rodriguez <mcgrof@kernel.org>; Kees Cook <keescook@chromium.org>; Peter Zijlstra (Intel) <peterz@infradead.org>; Zhang, Shile (NSB - CN/Hangzhou) <shile.zhang@nokia-sbell.com>; Matt Fleming <matt@codeblueprint.co.uk>; Andrew Morton <akpm@linux-foundation.org>; Vegard Nossum <vegard.nossum@oracle.com>; Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>; John Siddle <jsiddle@redhat.com>; open list <linux-kernel@vger.kernel.org>; open list:PROC SYSCTL <linux-fsdevel@vger.kernel.org>
> Subject: [PATCH] RFC: hung task: Check specific tasks for long uninterruptible sleep state
> 
> khungtask by default monitors either all tasks or no tasks at all
> for long unterruptible sleeps.
> For Android like environments this arrangement is not optimal because
> on one hand it may be permissible to have some background(bg) task
> in uninterruptible sleep state for long duration while on the other
> hand it may not be permissible to have some foreground(fg) task like
> surfaceflinger in uninterruptible sleep state for long duration.
> So it would be good to have some arrangement so that few specified tasks
> can be monitored by khungtaskd, on a need basis.
> This change introduces a sysctl option, /proc/sys/kernel/
> hung_task_check_selected, to enable monitoring of selected tasks using
> khungtask daemon. If this sysctl option is enabled then only the tasks
> specified in /proc/hung_task_monitor_list are monitored otherwise all
> tasks are monitored, just like the default case.
> 
> Signed-off-by: Imran Khan <kimran@codeaurora.org>
> ---
>  include/linux/sched/sysctl.h |   1 +
>  kernel/hung_task.c           | 121 ++++++++++++++++++++++++++++++++++++++++++-
>  kernel/sysctl.c              |   8 +++
>  3 files changed, 128 insertions(+), 2 deletions(-)
> 
> diff --git a/include/linux/sched/sysctl.h b/include/linux/sched/sysctl.h
> index 0f5ecd4..05892f1 100644
> --- a/include/linux/sched/sysctl.h
> +++ b/include/linux/sched/sysctl.h
> @@ -10,6 +10,7 @@
>  extern unsigned int  sysctl_hung_task_panic;
>  extern unsigned long sysctl_hung_task_timeout_secs;
>  extern int sysctl_hung_task_warnings;
> +extern int sysctl_hung_task_check_selected;
>  extern int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
>  					 void __user *buffer,
>  					 size_t *lenp, loff_t *ppos);
> diff --git a/kernel/hung_task.c b/kernel/hung_task.c
> index 751593e..49f13fb 100644
> --- a/kernel/hung_task.c
> +++ b/kernel/hung_task.c
> @@ -16,12 +16,28 @@
>  #include <linux/export.h>
>  #include <linux/sysctl.h>
>  #include <linux/utsname.h>
> +#include <linux/uaccess.h>
> +#include <linux/slab.h>
> +#include <linux/proc_fs.h>
>  #include <linux/sched/signal.h>
>  #include <linux/sched/debug.h>
>  
>  #include <trace/events/sched.h>
>  
>  /*
> + * Hung task that needs monitoring
> + */
> +struct hung_task {
> +	struct list_head list;
> +	char comm[TASK_COMM_LEN];
> +};
> +
> +static struct hung_task	*monitor_list;
> +int sysctl_hung_task_check_selected;
> +
> +
> +
> +/*
>   * The number of tasks checked:
>   */
>  int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
> @@ -76,6 +92,92 @@ static int __init hung_task_panic_setup(char *str)
>  	.notifier_call = hung_task_panic,
>  };
>  
> +static void hung_task_monitor_setup(void)
> +{
> +	monitor_list = kmalloc(sizeof(*monitor_list), GFP_KERNEL);
> +	if (monitor_list) {
> +		INIT_LIST_HEAD(&monitor_list->list);
> +		memset(monitor_list->comm, 0, TASK_COMM_LEN);
> +	}
> +}
> +
> +
> +static  int hung_task_info_show(struct seq_file *m, void *v)
> +{
> +	struct hung_task *ht;
> +
> +	ht = list_entry(v, struct hung_task, list);
> +	seq_puts(m, ht->comm);
> +
> +	return 0;
> +}
> +
> +static void *hung_task_info_start(struct seq_file *m, loff_t *pos)
> +{
> +	return seq_list_start_head(&monitor_list->list, *pos);
> +}
> +
> +static void *hung_task_info_next(struct seq_file *m, void *v, loff_t *pos)
> +{
> +	return seq_list_next(v, &monitor_list->list, pos);
> +}
> +
> +static void hung_task_info_stop(struct seq_file *m, void *v)
> +{
> +}
> +
> +const struct seq_operations hung_task_info_op = {
> +	.start	= hung_task_info_start,
> +	.next	= hung_task_info_next,
> +	.stop	= hung_task_info_stop,
> +	.show	= hung_task_info_show
> +};
> +
> +static int hung_task_info_open(struct inode *inode, struct file *file)
> +{
> +	return seq_open(file, &hung_task_info_op);
> +}
> +
> +static ssize_t
> +hung_task_info_write(struct file *file, const char __user *buf, size_t count,
> +	     loff_t *offs)
> +{
> +	struct task_struct *g, *t;
> +	struct hung_task *ht = kmalloc(sizeof(*ht), GFP_KERNEL);
> +
> +	if (!ht)
> +		return -ENOMEM;
> +
> +	if (copy_from_user(ht->comm, buf, count))
> +		return -EFAULT;
> +	ht->comm[count] = '\0';
> +
> +	for_each_process_thread(g, t) {
> +		if (!strncmp(t->comm, ht->comm, strlen(t->comm))) {
> +			list_add_tail(&ht->list, &monitor_list->list);
> +			return count;
> +		}
> +	}
> +
> +	pr_err("Non-existing task: %s can't be monitored\n", ht->comm);
> +	return count;
> +}
> +
> +static const struct file_operations hung_task_info_operations = {
> +	.open		= hung_task_info_open,
> +	.read		= seq_read,
> +	.write		= hung_task_info_write,
> +	.llseek		= seq_lseek,
> +	.release	= seq_release,
> +};
> +
> +static int __init proc_hung_task_info_init(void)
> +{
> +	proc_create("hung_task_monitor_list", 0644, NULL,
> +		    &hung_task_info_operations);
> +	return 0;
> +}
> +
>  static void check_hung_task(struct task_struct *t, unsigned long timeout)
>  {
>  	unsigned long switch_count = t->nvcsw + t->nivcsw;
> @@ -167,6 +269,7 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout)
>  	int max_count = sysctl_hung_task_check_count;
>  	int batch_count = HUNG_TASK_BATCHING;
>  	struct task_struct *g, *t;
> +	struct hung_task *ht, *tmp;
>  
>  	/*
>  	 * If the system crashed already then all bets are off,
> @@ -186,9 +289,21 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout)
>  				goto unlock;
>  		}
>  		/* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
> -		if (t->state == TASK_UNINTERRUPTIBLE)
> -			check_hung_task(t, timeout);
> +		if (t->state == TASK_UNINTERRUPTIBLE) {
> +			if (sysctl_hung_task_check_selected) {
> +				list_for_each_entry_safe(ht, tmp,
> +							 &monitor_list->list,
> +							 list)
> +					if (!strncmp(ht->comm, t->comm,
> +					    strlen(t->comm)))
> +					/* Task belongs to the selected group */
> +						check_hung_task(t, timeout);
> +			} else {
> +				check_hung_task(t, timeout);
> +			}
> +		}
>  	}
> +
>   unlock:
>  	rcu_read_unlock();
>  	if (hung_task_show_lock)
> @@ -259,6 +374,8 @@ static int watchdog(void *dummy)
>  static int __init hung_task_init(void)
>  {
>  	atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
> +	hung_task_monitor_setup();
> +	proc_hung_task_info_init();
>  	watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
>  
>  	return 0;
> diff --git a/kernel/sysctl.c b/kernel/sysctl.c
> index 6648fbb..ab45774 100644
> --- a/kernel/sysctl.c
> +++ b/kernel/sysctl.c
> @@ -1096,6 +1096,14 @@ static int sysrq_sysctl_handler(struct ctl_table *table, int write,
>  		.proc_handler	= proc_dointvec_minmax,
>  		.extra1		= &neg_one,
>  	},
> +	{
> +		.procname	= "hung_task_check_selected",
> +		.data		= &sysctl_hung_task_check_selected,
> +		.maxlen		= sizeof(int),
> +		.mode		= 0644,
> +		.proc_handler	= proc_dointvec_minmax,
> +		.extra1		= &neg_one,
> +	},
>  #endif
>  #ifdef CONFIG_RT_MUTEXES
>  	{
> 


-- 
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a\nmember of the Code Aurora Forum, hosted by The Linux Foundation

[toc] | [prev] | [next] | [standalone]

#1717072 — RE: [PATCH] RFC: hung task: Check specific tasks for long uninterruptible sleep state

Hi, Imran,

Seems the alternatively approach is better than former.

But sorry, I still think, like this kind of "monitoring", its better/easy to mark some "privileged task" not be monitored, but not list the one should be monitored.
Because It's easy to know which task should NOT be monitored (for some special purpose), but it's hard to which one should be monitored, right? Likes some task created runtime.

Just for your info.

BR, Shile

-----Original Message-----
From: Imran Khan [mailto:kimran@codeaurora.org] 
Sent: Tuesday, August 22, 2017 2:04 PM
To: Zhang, Shile (NSB - CN/Hangzhou) <shile.zhang@nokia-sbell.com>; mingo@kernel.org
Cc: imrank140517@gmail.com; Luis R. Rodriguez <mcgrof@kernel.org>; Kees Cook <keescook@chromium.org>; Peter Zijlstra (Intel) <peterz@infradead.org>; Matt Fleming <matt@codeblueprint.co.uk>; Andrew Morton <akpm@linux-foundation.org>; Vegard Nossum <vegard.nossum@oracle.com>; Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>; John Siddle <jsiddle@redhat.com>; open list <linux-kernel@vger.kernel.org>; open list:PROC SYSCTL <linux-fsdevel@vger.kernel.org>
Subject: Re: [PATCH] RFC: hung task: Check specific tasks for long uninterruptible sleep state

On 8/22/2017 7:24 AM, Zhang, Shile (NSB - CN/Hangzhou) wrote:

> Hi, Imran,
> 
Hi Shile,
Thanks for getting back on this.
> I think a "unmonitored list" is better than "monitor list", because we want khungtaskd can find out the "unexpected" hung task, but not few in a list.
> Then, for the fg tasks, which can put it in the "unmonitored list", for the bg tasks, I think we can tweak the timeout to control the duration.
> 
Here fg and bg tasks were just an example and may not be applicable for all environments (non-android). But main intention here is to have
some mechanism that would allow to monitor only specific tasks. As we have a sysctl option to enable/disable this mechanism we can at any time
fall back to default mechanism of monitoring all tasks.

Alternatively this can be done on task priority basis also, where we have a minimum priority and only tasks with priorities higher than
this min priority would be monitored and if we keep this min priority to the least priority , all tasks would be monitored like the 
default case. Please let me know your feedback regarding this approach.

Thanks and Regards,
Imran 
> Thanks!
> 
> BR, Shile
> 
> -----Original Message-----
> From: Imran Khan [mailto:kimran@codeaurora.org] 
> Sent: Monday, August 21, 2017 6:26 PM
> To: mingo@kernel.org
> Cc: imrank140517@gmail.com; Imran Khan <kimran@codeaurora.org>; Luis R. Rodriguez <mcgrof@kernel.org>; Kees Cook <keescook@chromium.org>; Peter Zijlstra (Intel) <peterz@infradead.org>; Zhang, Shile (NSB - CN/Hangzhou) <shile.zhang@nokia-sbell.com>; Matt Fleming <matt@codeblueprint.co.uk>; Andrew Morton <akpm@linux-foundation.org>; Vegard Nossum <vegard.nossum@oracle.com>; Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>; John Siddle <jsiddle@redhat.com>; open list <linux-kernel@vger.kernel.org>; open list:PROC SYSCTL <linux-fsdevel@vger.kernel.org>
> Subject: [PATCH] RFC: hung task: Check specific tasks for long uninterruptible sleep state
> 
> khungtask by default monitors either all tasks or no tasks at all
> for long unterruptible sleeps.
> For Android like environments this arrangement is not optimal because
> on one hand it may be permissible to have some background(bg) task
> in uninterruptible sleep state for long duration while on the other
> hand it may not be permissible to have some foreground(fg) task like
> surfaceflinger in uninterruptible sleep state for long duration.
> So it would be good to have some arrangement so that few specified tasks
> can be monitored by khungtaskd, on a need basis.
> This change introduces a sysctl option, /proc/sys/kernel/
> hung_task_check_selected, to enable monitoring of selected tasks using
> khungtask daemon. If this sysctl option is enabled then only the tasks
> specified in /proc/hung_task_monitor_list are monitored otherwise all
> tasks are monitored, just like the default case.
> 
> Signed-off-by: Imran Khan <kimran@codeaurora.org>
> ---
>  include/linux/sched/sysctl.h |   1 +
>  kernel/hung_task.c           | 121 ++++++++++++++++++++++++++++++++++++++++++-
>  kernel/sysctl.c              |   8 +++
>  3 files changed, 128 insertions(+), 2 deletions(-)
> 
> diff --git a/include/linux/sched/sysctl.h b/include/linux/sched/sysctl.h
> index 0f5ecd4..05892f1 100644
> --- a/include/linux/sched/sysctl.h
> +++ b/include/linux/sched/sysctl.h
> @@ -10,6 +10,7 @@
>  extern unsigned int  sysctl_hung_task_panic;
>  extern unsigned long sysctl_hung_task_timeout_secs;
>  extern int sysctl_hung_task_warnings;
> +extern int sysctl_hung_task_check_selected;
>  extern int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
>  					 void __user *buffer,
>  					 size_t *lenp, loff_t *ppos);
> diff --git a/kernel/hung_task.c b/kernel/hung_task.c
> index 751593e..49f13fb 100644
> --- a/kernel/hung_task.c
> +++ b/kernel/hung_task.c
> @@ -16,12 +16,28 @@
>  #include <linux/export.h>
>  #include <linux/sysctl.h>
>  #include <linux/utsname.h>
> +#include <linux/uaccess.h>
> +#include <linux/slab.h>
> +#include <linux/proc_fs.h>
>  #include <linux/sched/signal.h>
>  #include <linux/sched/debug.h>
>  
>  #include <trace/events/sched.h>
>  
>  /*
> + * Hung task that needs monitoring
> + */
> +struct hung_task {
> +	struct list_head list;
> +	char comm[TASK_COMM_LEN];
> +};
> +
> +static struct hung_task	*monitor_list;
> +int sysctl_hung_task_check_selected;
> +
> +
> +
> +/*
>   * The number of tasks checked:
>   */
>  int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
> @@ -76,6 +92,92 @@ static int __init hung_task_panic_setup(char *str)
>  	.notifier_call = hung_task_panic,
>  };
>  
> +static void hung_task_monitor_setup(void)
> +{
> +	monitor_list = kmalloc(sizeof(*monitor_list), GFP_KERNEL);
> +	if (monitor_list) {
> +		INIT_LIST_HEAD(&monitor_list->list);
> +		memset(monitor_list->comm, 0, TASK_COMM_LEN);
> +	}
> +}
> +
> +
> +static  int hung_task_info_show(struct seq_file *m, void *v)
> +{
> +	struct hung_task *ht;
> +
> +	ht = list_entry(v, struct hung_task, list);
> +	seq_puts(m, ht->comm);
> +
> +	return 0;
> +}
> +
> +static void *hung_task_info_start(struct seq_file *m, loff_t *pos)
> +{
> +	return seq_list_start_head(&monitor_list->list, *pos);
> +}
> +
> +static void *hung_task_info_next(struct seq_file *m, void *v, loff_t *pos)
> +{
> +	return seq_list_next(v, &monitor_list->list, pos);
> +}
> +
> +static void hung_task_info_stop(struct seq_file *m, void *v)
> +{
> +}
> +
> +const struct seq_operations hung_task_info_op = {
> +	.start	= hung_task_info_start,
> +	.next	= hung_task_info_next,
> +	.stop	= hung_task_info_stop,
> +	.show	= hung_task_info_show
> +};
> +
> +static int hung_task_info_open(struct inode *inode, struct file *file)
> +{
> +	return seq_open(file, &hung_task_info_op);
> +}
> +
> +static ssize_t
> +hung_task_info_write(struct file *file, const char __user *buf, size_t count,
> +	     loff_t *offs)
> +{
> +	struct task_struct *g, *t;
> +	struct hung_task *ht = kmalloc(sizeof(*ht), GFP_KERNEL);
> +
> +	if (!ht)
> +		return -ENOMEM;
> +
> +	if (copy_from_user(ht->comm, buf, count))
> +		return -EFAULT;
> +	ht->comm[count] = '\0';
> +
> +	for_each_process_thread(g, t) {
> +		if (!strncmp(t->comm, ht->comm, strlen(t->comm))) {
> +			list_add_tail(&ht->list, &monitor_list->list);
> +			return count;
> +		}
> +	}
> +
> +	pr_err("Non-existing task: %s can't be monitored\n", ht->comm);
> +	return count;
> +}
> +
> +static const struct file_operations hung_task_info_operations = {
> +	.open		= hung_task_info_open,
> +	.read		= seq_read,
> +	.write		= hung_task_info_write,
> +	.llseek		= seq_lseek,
> +	.release	= seq_release,
> +};
> +
> +static int __init proc_hung_task_info_init(void)
> +{
> +	proc_create("hung_task_monitor_list", 0644, NULL,
> +		    &hung_task_info_operations);
> +	return 0;
> +}
> +
>  static void check_hung_task(struct task_struct *t, unsigned long timeout)
>  {
>  	unsigned long switch_count = t->nvcsw + t->nivcsw;
> @@ -167,6 +269,7 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout)
>  	int max_count = sysctl_hung_task_check_count;
>  	int batch_count = HUNG_TASK_BATCHING;
>  	struct task_struct *g, *t;
> +	struct hung_task *ht, *tmp;
>  
>  	/*
>  	 * If the system crashed already then all bets are off,
> @@ -186,9 +289,21 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout)
>  				goto unlock;
>  		}
>  		/* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
> -		if (t->state == TASK_UNINTERRUPTIBLE)
> -			check_hung_task(t, timeout);
> +		if (t->state == TASK_UNINTERRUPTIBLE) {
> +			if (sysctl_hung_task_check_selected) {
> +				list_for_each_entry_safe(ht, tmp,
> +							 &monitor_list->list,
> +							 list)
> +					if (!strncmp(ht->comm, t->comm,
> +					    strlen(t->comm)))
> +					/* Task belongs to the selected group */
> +						check_hung_task(t, timeout);
> +			} else {
> +				check_hung_task(t, timeout);
> +			}
> +		}
>  	}
> +
>   unlock:
>  	rcu_read_unlock();
>  	if (hung_task_show_lock)
> @@ -259,6 +374,8 @@ static int watchdog(void *dummy)
>  static int __init hung_task_init(void)
>  {
>  	atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
> +	hung_task_monitor_setup();
> +	proc_hung_task_info_init();
>  	watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
>  
>  	return 0;
> diff --git a/kernel/sysctl.c b/kernel/sysctl.c
> index 6648fbb..ab45774 100644
> --- a/kernel/sysctl.c
> +++ b/kernel/sysctl.c
> @@ -1096,6 +1096,14 @@ static int sysrq_sysctl_handler(struct ctl_table *table, int write,
>  		.proc_handler	= proc_dointvec_minmax,
>  		.extra1		= &neg_one,
>  	},
> +	{
> +		.procname	= "hung_task_check_selected",
> +		.data		= &sysctl_hung_task_check_selected,
> +		.maxlen		= sizeof(int),
> +		.mode		= 0644,
> +		.proc_handler	= proc_dointvec_minmax,
> +		.extra1		= &neg_one,
> +	},
>  #endif
>  #ifdef CONFIG_RT_MUTEXES
>  	{
> 


-- 
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a\nmember of the Code Aurora Forum, hosted by The Linux Foundation

[toc] | [prev] | [next] | [standalone]

#1717821 — Re: [PATCH] RFC: hung task: Check specific tasks for long uninterruptible sleep state

On Mon, Aug 21, 2017 at 03:55:53PM +0530, Imran Khan wrote:
> khungtask by default monitors either all tasks or no tasks at all
> for long unterruptible sleeps.
> For Android like environments this arrangement is not optimal because
> on one hand it may be permissible to have some background(bg) task
> in uninterruptible sleep state for long duration while on the other
> hand it may not be permissible to have some foreground(fg) task like
> surfaceflinger in uninterruptible sleep state for long duration.

How are you getting tasks in UNINTERRUPTIBLE state for a long time to
begin with? That's not a good thing, background or not.

> So it would be good to have some arrangement so that few specified tasks
> can be monitored by khungtaskd, on a need basis.
> This change introduces a sysctl option, /proc/sys/kernel/
> hung_task_check_selected, to enable monitoring of selected tasks using
> khungtask daemon. If this sysctl option is enabled then only the tasks
> specified in /proc/hung_task_monitor_list are monitored otherwise all
> tasks are monitored, just like the default case.

That's horrible. Why not a file in /proc/$PID/ that excludes it from
things?

[toc] | [prev] | [next] | [standalone]

#1718314 — Re: [PATCH] RFC: hung task: Check specific tasks for long uninterruptible sleep state

On 8/23/2017 2:06 AM, Peter Zijlstra wrote:
> On Mon, Aug 21, 2017 at 03:55:53PM +0530, Imran Khan wrote:
>> khungtask by default monitors either all tasks or no tasks at all
>> for long unterruptible sleeps.
>> For Android like environments this arrangement is not optimal because
>> on one hand it may be permissible to have some background(bg) task
>> in uninterruptible sleep state for long duration while on the other
>> hand it may not be permissible to have some foreground(fg) task like
>> surfaceflinger in uninterruptible sleep state for long duration.
> 
> How are you getting tasks in UNINTERRUPTIBLE state for a long time to
> begin with? That's not a good thing, background or not.
> 
As of now, I don't have the details of exact test case as most of these issues are
coming in android aging and stress tests but I see even in a non-android
setup I get few tasks in uninterruptible sleep state for long duration (60 secs).
For example, I see there is a thread waiting for mmc requests to appear from block
layer. Of course this can be a faulty design on the driver side too but my main
intention here is to know if we can enable/disable this monitoring for 
specific tasks and whether such an option would be useful and agreeable to
community or not.

Also in current scheme of things we use same timeout for all tasks which may
not be optimal. For example if some task is in disk sleep state it may be okay
for it to wait for 2 minutes but if surfaceflinger (android) is stuck even for
1 minute, it results in a very poor UI experience for user of the phone.
>> So it would be good to have some arrangement so that few specified tasks
>> can be monitored by khungtaskd, on a need basis.
>> This change introduces a sysctl option, /proc/sys/kernel/
>> hung_task_check_selected, to enable monitoring of selected tasks using
>> khungtask daemon. If this sysctl option is enabled then only the tasks
>> specified in /proc/hung_task_monitor_list are monitored otherwise all
>> tasks are monitored, just like the default case.
> 
> That's horrible. Why not a file in /proc/$PID/ that excludes it from
> things?
> 
To be honest, I also had apprehension about my approach and did not think
it through. I agree that /proc/$PID/ based approach is much better. If it is
agreed that selective monitoring of hung tasks has valid use case(s), I can
send the /proc/$PID based patch for review.
Thanks for your time and feedback.

Regards,
Imran


-- 
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a\nmember of the Code Aurora Forum, hosted by The Linux Foundation

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web