Re: [PATCH 2/3] hung_task: Add hung_task_sys_info sysctl to dump sys info on task-hung

[Feng Tang <feng.tang@linux.alibaba.com>]

On Wed, Nov 12, 2025 at 03:44:09PM +0100, Petr Mladek wrote:
> On Wed 2025-11-12 19:25:27, Feng Tang wrote:
> > On Mon, Nov 10, 2025 at 06:55:57PM +0100, Petr Mladek wrote:
> > > On Thu 2025-11-06 10:30:31, Feng Tang wrote:
> > [...]
> > > @@ -315,8 +295,10 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout)
> > >  {
> > >  	int max_count = sysctl_hung_task_check_count;
> > >  	unsigned long last_break = jiffies;
> > > +	unsigned long total_hung_task;
> > >  	struct task_struct *g, *t;
> > >  	unsigned long prev_detect_count = sysctl_hung_task_detect_count;
> > > +	unsigned long si_mask;
> > >  
> > >  	/*
> > >  	 * If the system crashed already then all bets are off,
> > > @@ -325,6 +307,14 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout)
> > >  	if (test_taint(TAINT_DIE) || did_panic)
> > >  		return;
> > >  
> > > +	si_mask = hung_task_si_mask;
> > > +	if (sysctl_hung_task_warnings || hung_task_call_panic) {
> > > +		si_mask |= SYS_INFO_LOCKS;
> > > +
> > > +		if (sysctl_hung_task_all_cpu_backtrace)
> > > +			si_mask |= SYS_INFO_ALL_BT;
> > > +	}
> > 
> > This probably needs to be moved to after the loop check of
> > check_hung_task(). 
> 
> I did it on purpose because "sysctl_hung_task_warnings" might get
> decremented down to "0" in the loop below. But IMHO, we need to print
> the information if it was non-zero at the beginning.
> 
> It might be worth to add a comment why it has to be done
> before the cycle.
 
I see your point. Yes, that could happen and should be handled.

My concern was:
1. 'hung_task_call_panic' is actually set in the following loop of 
   checking, and should be checked after the loop
2. when 'sysctl_hung_task_warnings' is not 0 (likely), the
	si_mask |= SYS_INFO_LOCKS
  will make it always call sys_info() will non-zero value, while the
  'hung_task_si_mask' could be pre-set. I just run a simple hung task
  test and can confirm this.

So I made some small changes based on your patches, please help to
review.

diff --git a/kernel/hung_task.c b/kernel/hung_task.c
index 5f0275b2c742..5b3a7785d3a2 100644
--- a/kernel/hung_task.c
+++ b/kernel/hung_task.c
@@ -71,12 +71,6 @@ static struct task_struct *watchdog_task;
  */
 static unsigned long hung_task_si_mask;
 
-/*
- * There are several sysctl knobs, and this serves as the runtime
- * effective sys_info knob
- */
-static unsigned long cur_si_mask;
-
 #ifdef CONFIG_SMP
 /*
  * Should we dump all CPUs backtraces in a hung task event?
@@ -229,11 +223,8 @@ static inline void debug_show_blocker(struct task_struct *task, unsigned long ti
 }
 #endif
 
-static void check_hung_task(struct task_struct *t, unsigned long timeout,
-		unsigned long prev_detect_count)
+static void check_hung_task(struct task_struct *t, unsigned long timeout)
 {
-	unsigned long total_hung_task;
-
 	if (!task_is_hung(t, timeout))
 		return;
 
@@ -243,21 +234,13 @@ static void check_hung_task(struct task_struct *t, unsigned long timeout,
 	 */
 	sysctl_hung_task_detect_count++;
 
-	total_hung_task = sysctl_hung_task_detect_count - prev_detect_count;
 	trace_sched_process_hang(t);
 
-	cur_si_mask = hung_task_si_mask;
-	if (sysctl_hung_task_panic && total_hung_task >= sysctl_hung_task_panic) {
-		console_verbose();
-		cur_si_mask |= SYS_INFO_LOCKS;
-		hung_task_call_panic = true;
-	}
-
 	/*
 	 * Ok, the task did not get scheduled for more than 2 minutes,
 	 * complain:
 	 */
-	if (sysctl_hung_task_warnings || hung_task_call_panic) {
+	if (sysctl_hung_task_warnings) {
 		if (sysctl_hung_task_warnings > 0)
 			sysctl_hung_task_warnings--;
 		pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
@@ -272,10 +255,7 @@ static void check_hung_task(struct task_struct *t, unsigned long timeout,
 			" disables this message.\n");
 		sched_show_task(t);
 		debug_show_blocker(t, timeout);
-		cur_si_mask |= SYS_INFO_LOCKS;
 
-		if (sysctl_hung_task_all_cpu_backtrace)
-			cur_si_mask |= SYS_INFO_ALL_BT;
 		if (!sysctl_hung_task_warnings)
 			pr_info("Future hung task reports are suppressed, see sysctl kernel.hung_task_warnings\n");
 	}
@@ -315,8 +295,11 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout)
 {
 	int max_count = sysctl_hung_task_check_count;
 	unsigned long last_break = jiffies;
+	unsigned long total_hung_task;
 	struct task_struct *g, *t;
 	unsigned long prev_detect_count = sysctl_hung_task_detect_count;
+	int need_warning = sysctl_hung_task_warnings;
+	unsigned long si_mask = hung_task_si_mask;
 
 	/*
 	 * If the system crashed already then all bets are off,
@@ -325,6 +308,7 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout)
 	if (test_taint(TAINT_DIE) || did_panic)
 		return;
 
+
 	rcu_read_lock();
 	for_each_process_thread(g, t) {
 
@@ -336,16 +320,29 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout)
 			last_break = jiffies;
 		}
 
-		check_hung_task(t, timeout, prev_detect_count);
+		check_hung_task(t, timeout);
 	}
  unlock:
 	rcu_read_unlock();
 
-	if (unlikely(cur_si_mask)) {
-		sys_info(cur_si_mask);
-		cur_si_mask = 0;
+	total_hung_task = sysctl_hung_task_detect_count - prev_detect_count;
+	if (!total_hung_task)
+		return;
+
+	if (sysctl_hung_task_panic && total_hung_task >= sysctl_hung_task_panic) {
+		console_verbose();
+		hung_task_call_panic = true;
 	}
 
+	if (need_warning || hung_task_call_panic) {
+		si_mask |= SYS_INFO_LOCKS;
+
+		if (sysctl_hung_task_all_cpu_backtrace)
+			si_mask |= SYS_INFO_ALL_BT;
+	}
+
+	sys_info(si_mask);
+
 	if (hung_task_call_panic)
 		panic("hung_task: blocked tasks");
 }

Thanks,
Feng