[PATCH v2 1/2] hung_task: show the blocker task if the task is hung on rtmutex

[ruipengqi <ruipengqi3@gmail.com>]

From: Ruipeng Qi <ruipengqi3@gmail.com>

Currently, debug_show_blocker() only tracks mutex, semaphore and rwsem
lock types. Extend it to also cover rtmutex or rt_mutex-based lock
implementations on PREEMPT_RT, so that when a task is hung on an
rtmutex, the hung task detector can identify and report which task holds
the lock.

Add rt_mutex_task_owner() to get the task's lock owner when a task is
blocked by an rtmutex-based lock implementation.

Since the LSBs of task->blocker pointer are already fully utilized and
leave no room for new blocker types, this version avoids adding a new
blocker type by directly leveraging rt_mutex_task_owner() to retrieve
the lock owner information.

With this change, the hung task detector can now show blocker task's
info like below:

[ 3000.899985] INFO: task cat:195 blocked for more than 120 seconds.
[ 3000.900561]       Not tainted 7.2.0-rc4-00366-gf339a6eb59f3-dirty #22
[ 3000.900930] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[ 3000.901376] task:cat             state:D stack:13784 pid:195   tgid:195   ppid:1      task_flags:0x400000 flags:0x00080000
[ 3000.902081] Call Trace:
[ 3000.902233]  <TASK>
[ 3000.902383]  __schedule+0x514/0xf50
[ 3000.902677]  rt_mutex_schedule+0x1b/0x30
[ 3000.902916]  rt_mutex_slowlock_block.constprop.0+0x3b/0x1c0
[ 3000.903242]  __rt_mutex_slowlock_locked.constprop.0+0xa8/0x200
[ 3000.903716]  rt_mutex_slowlock.constprop.0+0x48/0xb0
[ 3000.904023]  rt_mutex_lock+0x32/0x40
[ 3000.904249]  read_dummy_rtmutex+0x2a/0x60 [hung_task_tests]
[ 3000.904647]  full_proxy_read+0x5b/0x90
[ 3000.904843]  vfs_read+0xb0/0x370
[ 3000.905051]  ? vm_mmap_pgoff+0xf1/0x1b0
[ 3000.905293]  ? vm_mmap_pgoff+0x122/0x1b0
[ 3000.905684]  ksys_read+0x68/0xe0
[ 3000.905980]  do_syscall_64+0xf9/0x540
[ 3000.906213]  entry_SYSCALL_64_after_hwframe+0x77/0x7f
[ 3000.906643] RIP: 0033:0x49a182
[ 3000.906844] RSP: 002b:00007ffc431f6528 EFLAGS: 00000246 ORIG_RAX: 0000000000000000
[ 3000.907251] RAX: ffffffffffffffda RBX: 0000000000000003 RCX: 000000000049a182
[ 3000.907673] RDX: 0000000000010000 RSI: 00007f69b2166000 RDI: 0000000000000003
[ 3000.908044] RBP: 00007f69b2166000 R08: 00000000ffffffff R09: 0000000000000000
[ 3000.908485] R10: 0000000000000022 R11: 0000000000000246 R12: 0000000000010000
[ 3000.908816] R13: 0000000000000003 R14: 0000000000000001 R15: 0000000001000000
[ 3000.909206]  </TASK>
[ 3000.909351] INFO: task cat:195 is blocked on a rtmutex likely owned by task cat:194.
[ 3000.909832] task:cat             state:S stack:13784 pid:194   tgid:194   ppid:1      task_flags:0x400000 flags:0x00080000
[ 3000.910426] Call Trace:
[ 3000.910618]  <TASK>
[ 3000.910727]  __schedule+0x514/0xf50
[ 3000.910912]  schedule+0x22/0xa0
[ 3000.911064]  schedule_timeout+0x81/0x100
[ 3000.911254]  ? __pfx_process_timeout+0x10/0x10
[ 3000.911522]  msleep_interruptible+0x28/0x50
[ 3000.911754]  read_dummy_rtmutex+0x34/0x60 [hung_task_tests]
[ 3000.912019]  full_proxy_read+0x5b/0x90
[ 3000.912212]  vfs_read+0xb0/0x370
[ 3000.912378]  ? vm_mmap_pgoff+0xf1/0x1b0
[ 3000.912630]  ? vm_mmap_pgoff+0x122/0x1b0
[ 3000.912839]  ksys_read+0x68/0xe0
[ 3000.913007]  do_syscall_64+0xf9/0x540
[ 3000.913187]  entry_SYSCALL_64_after_hwframe+0x77/0x7f
[ 3000.913461] RIP: 0033:0x49a182
[ 3000.913618] RSP: 002b:00007ffd0b4cf048 EFLAGS: 00000246 ORIG_RAX: 0000000000000000
[ 3000.913972] RAX: ffffffffffffffda RBX: 0000000000000003 RCX: 000000000049a182
[ 3000.914294] RDX: 0000000000010000 RSI: 00007fbf49a0d000 RDI: 0000000000000003
[ 3000.914648] RBP: 00007fbf49a0d000 R08: 00000000ffffffff R09: 0000000000000000
[ 3000.914955] R10: 0000000000000022 R11: 0000000000000246 R12: 0000000000010000
[ 3000.915208] R13: 0000000000000003 R14: 0000000000000001 R15: 0000000001000000
[ 3000.915439]  </TASK>

Suggested-by: Petr Mladek <pmladek@suse.com>
Suggested-by: Zhan Xusheng <zhanxusheng1024@gmail.com>
Signed-off-by: Ruipeng Qi <ruipengqi3@gmail.com>

---
v2 changes:
 - Reduce #ifdef-ery by using rt_mutex_task_owner() instead of adding a new blocker type.
 - Remove the unused function debug_trace_blocker().
 - Clean up and update relevant comments.
---
 include/linux/rtmutex.h      |  2 ++
 kernel/hung_task.c           | 26 +++++++++++++++++++++++---
 kernel/locking/rtmutex_api.c | 29 +++++++++++++++++++++++++++++
 lib/Kconfig.debug            |  1 -
 4 files changed, 54 insertions(+), 4 deletions(-)

diff --git a/include/linux/rtmutex.h b/include/linux/rtmutex.h
index 9e1f012f89db89..c179f946d2d040 100644
--- a/include/linux/rtmutex.h
+++ b/include/linux/rtmutex.h
@@ -53,6 +53,8 @@ static inline struct task_struct *rt_mutex_owner(struct rt_mutex_base *lock)
 
 	return (struct task_struct *) (owner & ~RT_MUTEX_HAS_WAITERS);
 }
+
+extern struct task_struct *rt_mutex_task_owner(struct task_struct *task);
 #endif
 extern void rt_mutex_base_init(struct rt_mutex_base *rtb);
 
diff --git a/kernel/hung_task.c b/kernel/hung_task.c
index 6fcc94ce4ca9d2..9467011f878c95 100644
--- a/kernel/hung_task.c
+++ b/kernel/hung_task.c
@@ -27,6 +27,7 @@
 #include <linux/sys_info.h>
 
 #include <trace/events/sched.h>
+#include <linux/rtmutex.h>
 
 /*
  * The number of tasks checked:
@@ -139,9 +140,19 @@ static void debug_show_blocker(struct task_struct *task, unsigned long timeout)
 	struct task_struct *g, *t;
 	unsigned long owner, blocker, blocker_type;
 	const char *rwsem_blocked_by, *rwsem_blocked_as;
+	bool is_rtmutex = false;
 
 	RCU_LOCKDEP_WARN(!rcu_read_lock_held(), "No rcu lock held");
 
+#if defined(CONFIG_RT_MUTEXES)
+	owner = (unsigned long)rt_mutex_task_owner(task);
+	if (owner) {
+		is_rtmutex = true;
+		blocker_type = -1;
+		goto found;
+	}
+#endif
+
 	blocker = READ_ONCE(task->blocker);
 	if (!blocker)
 		return;
@@ -149,12 +160,13 @@ static void debug_show_blocker(struct task_struct *task, unsigned long timeout)
 	blocker_type = hung_task_get_blocker_type(blocker);
 
 	switch (blocker_type) {
-	case BLOCKER_TYPE_MUTEX:
-		owner = mutex_get_owner(hung_task_blocker_to_lock(blocker));
-		break;
 	case BLOCKER_TYPE_SEM:
 		owner = sem_last_holder(hung_task_blocker_to_lock(blocker));
 		break;
+#ifndef CONFIG_PREEMPT_RT
+	case BLOCKER_TYPE_MUTEX:
+		owner = mutex_get_owner(hung_task_blocker_to_lock(blocker));
+		break;
 	case BLOCKER_TYPE_RWSEM_READER:
 	case BLOCKER_TYPE_RWSEM_WRITER:
 		owner = (unsigned long)rwsem_owner(
@@ -165,6 +177,7 @@ static void debug_show_blocker(struct task_struct *task, unsigned long timeout)
 					hung_task_blocker_to_lock(blocker)) ?
 					"reader" : "writer";
 		break;
+#endif
 	default:
 		WARN_ON_ONCE(1);
 		return;
@@ -190,6 +203,7 @@ static void debug_show_blocker(struct task_struct *task, unsigned long timeout)
 		return;
 	}
 
+found:
 	/* Ensure the owner information is correct. */
 	for_each_process_thread(g, t) {
 		if ((unsigned long)t != owner)
@@ -210,6 +224,12 @@ static void debug_show_blocker(struct task_struct *task, unsigned long timeout)
 			       task->comm, task->pid, rwsem_blocked_as, t->comm,
 			       t->pid, rwsem_blocked_by);
 			break;
+		default:
+			if (is_rtmutex) {
+				pr_err("INFO: task %s:%d is blocked on a rtmutex likely owned by task %s:%d.\n",
+						task->comm, task->pid, t->comm, t->pid);
+			}
+			break;
 		}
 		/* Avoid duplicated task dump, skip if the task is also hung. */
 		if (!task_is_hung(t, timeout))
diff --git a/kernel/locking/rtmutex_api.c b/kernel/locking/rtmutex_api.c
index 5d48d64725b125..d0826b31a148c0 100644
--- a/kernel/locking/rtmutex_api.c
+++ b/kernel/locking/rtmutex_api.c
@@ -530,6 +530,35 @@ void __sched rt_mutex_postunlock(struct rt_wake_q_head *wqh)
 	rt_mutex_wake_up_q(wqh);
 }
 
+/**
+ * rt_mutex_task_owner - Get the lock owner a task is blocked on
+ * @task: Task to inspect.
+ *
+ * Caller must hold an RCU read lock. The owner value is a snapshot read
+ * without exclusion, so it may become stale immediately. Use only for
+ * best-effort diagnostics (e.g., hung task detection).
+ *
+ * Return: The task_struct owning the lock, or NULL if not blocked.
+ */
+struct task_struct *rt_mutex_task_owner(struct task_struct *task)
+{
+        struct task_struct *owner = NULL;
+        struct rt_mutex_base *lock;
+        unsigned long flags;
+
+	lockdep_assert(rcu_read_lock_held());
+
+        raw_spin_lock_irqsave(&task->pi_lock, flags);
+        lock = task_blocked_on_lock(task);
+
+        if (lock)
+                owner = rt_mutex_owner(lock);
+        raw_spin_unlock_irqrestore(&task->pi_lock, flags);
+
+        return owner;
+}
+EXPORT_SYMBOL(rt_mutex_task_owner);
+
 #ifdef CONFIG_DEBUG_RT_MUTEXES
 void rt_mutex_debug_task_free(struct task_struct *task)
 {
diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug
index 1244dcac2294ad..92dff380027e78 100644
--- a/lib/Kconfig.debug
+++ b/lib/Kconfig.debug
@@ -1317,7 +1317,6 @@ config BOOTPARAM_HUNG_TASK_PANIC
 config DETECT_HUNG_TASK_BLOCKER
 	bool "Dump Hung Tasks Blocker"
 	depends on DETECT_HUNG_TASK
-	depends on !PREEMPT_RT
 	default y
 	help
 	  Say Y here to show the blocker task's stacktrace who acquires
-- 
2.25.1