[GIT PULL] [PATCH 0/9] Locking changes for v6.17

[GIT PULL] [PATCH 0/9] Locking changes for v6.17

[Boqun Feng]

Hi Ingo & Peter,

Please pull the lockdep changes for v6.17 into tip. One thing to notice
is that I would like to add Breno Leitao's workaround in
lockdep_unregister_key(), since the alternative requires hazptr and I'm
not near to get it done. I added comments explaining this and left a
TODO there, and I will replace it once the shazptr patch is in a good
shape. Peter, if you have any concern about it, either we can drop the
patch #9 or I can create another tag. Thanks!

The following changes since commit 7ff495e26a39f3e7a3d4058df59b5b6d6f943cab:

  local_lock: Move this_cpu_ptr() notation from internal to main header (2025-06-30 17:45:35 +0200)

are available in the Git repository at:

  git://git.kernel.org/pub/scm/linux/kernel/git/boqun/linux.git tags/lockdep-for-tip.2025.07.16

for you to fetch changes up to 7a3cedafccf8e7d038ad4cfec5b38052647ceac5:

  lockdep: Speed up lockdep_unregister_key() with expedited RCU synchronization (2025-07-14 21:57:29 -0700)

----------------------------------------------------------------
Locking changes for v6.17:

- General
  - Mark devm_mutex_init() as __must_check
  - Add #[must_use] to Lock::try_lock()
  - Remove OWNER_SPINNABLE in rwsem
  - Remove redundant #ifdefs in mutex
- Lockdep
  - Avoid returning struct in lock_stats()
  - Change `static const` into enum for LOCKF_*_IRQ_*
  - Temporarily use synchronize_rcu_expedited() in
    lockdep_unregister_key() to speed things up.

----------------------------------------------------------------

Regards,
Boqun

Arnd Bergmann (2):
      locking/lockdep: Avoid struct return in lock_stats()
      locking/lockdep: Change 'static const' variables to enum values

Breno Leitao (1):
      lockdep: Speed up lockdep_unregister_key() with expedited RCU synchronization

Jason Devers (1):
      rust: sync: Add #[must_use] to Lock::try_lock()

Jinliang Zheng (1):
      locking/rwsem: Use OWNER_NONSPINNABLE directly instead of OWNER_SPINNABLE

Ran Xiaokai (1):
      locking/mutex: Remove redundant #ifdefs

Thomas Weißschuh (3):
      spi: spi-nxp-fspi: Check return value of devm_mutex_init()
      leds: lp8860: Check return value of devm_mutex_init()
      locking/mutex: Mark devm_mutex_init() as __must_check

 drivers/leds/leds-lp8860.c         |  4 +++-
 drivers/spi/spi-nxp-fspi.c         |  4 +++-
 include/linux/lockdep_types.h      |  2 +-
 include/linux/mutex.h              | 11 +++++++----
 kernel/locking/lockdep.c           | 39 +++++++++++++++++++++-----------------
 kernel/locking/lockdep_internals.h | 18 ++++++++++--------
 kernel/locking/lockdep_proc.c      |  2 +-
 kernel/locking/mutex.c             |  4 ----
 kernel/locking/rwsem.c             |  4 +---
 rust/kernel/sync/lock.rs           |  2 ++
 10 files changed, 50 insertions(+), 40 deletions(-)
-- 
2.39.5 (Apple Git-154)

[PATCH 1/9] spi: spi-nxp-fspi: Check return value of devm_mutex_init()

[Boqun Feng]

From: Thomas Weißschuh <linux@weissschuh.net>

devm_mutex_init() can fail. With CONFIG_DEBUG_MUTEXES=y the mutex will
be marked as unusable and trigger errors on usage.

Add the missed check.

Fixes: 48900813abd2 ("spi: spi-nxp-fspi: remove the goto in probe")
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Reviewed-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Link: https://lore.kernel.org/r/20250617-must_check-devm_mutex_init-v7-1-d9e449f4d224@weissschuh.net
---
 drivers/spi/spi-nxp-fspi.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/spi/spi-nxp-fspi.c b/drivers/spi/spi-nxp-fspi.c
index e63c77e41823..f3d576505413 100644
--- a/drivers/spi/spi-nxp-fspi.c
+++ b/drivers/spi/spi-nxp-fspi.c
@@ -1273,7 +1273,9 @@ static int nxp_fspi_probe(struct platform_device *pdev)
 	if (ret)
 		return dev_err_probe(dev, ret, "Failed to request irq\n");
 
-	devm_mutex_init(dev, &f->lock);
+	ret = devm_mutex_init(dev, &f->lock);
+	if (ret)
+		return dev_err_probe(dev, ret, "Failed to initialize lock\n");
 
 	ctlr->bus_num = -1;
 	ctlr->num_chipselect = NXP_FSPI_MAX_CHIPSELECT;
-- 
2.39.5 (Apple Git-154)

[PATCH 2/9] leds: lp8860: Check return value of devm_mutex_init()

[Boqun Feng]

From: Thomas Weißschuh <linux@weissschuh.net>

devm_mutex_init() can fail. With CONFIG_DEBUG_MUTEXES=y the mutex will be
marked as unusable and trigger errors on usage.

Add the missed check.

Fixes: 87a59548af95 ("leds: lp8860: Use new mutex guards to cleanup function exits")
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Acked-by: Andrew Davis <afd@ti.com>
Acked-by: Lee Jones <lee@kernel.org>
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Link: https://lore.kernel.org/r/20250617-must_check-devm_mutex_init-v7-2-d9e449f4d224@weissschuh.net
---
 drivers/leds/leds-lp8860.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/leds/leds-lp8860.c b/drivers/leds/leds-lp8860.c
index 52b97c9f2a03..0962c00c215a 100644
--- a/drivers/leds/leds-lp8860.c
+++ b/drivers/leds/leds-lp8860.c
@@ -307,7 +307,9 @@ static int lp8860_probe(struct i2c_client *client)
 	led->client = client;
 	led->led_dev.brightness_set_blocking = lp8860_brightness_set;
 
-	devm_mutex_init(&client->dev, &led->lock);
+	ret = devm_mutex_init(&client->dev, &led->lock);
+	if (ret)
+		return dev_err_probe(&client->dev, ret, "Failed to initialize lock\n");
 
 	led->regmap = devm_regmap_init_i2c(client, &lp8860_regmap_config);
 	if (IS_ERR(led->regmap)) {
-- 
2.39.5 (Apple Git-154)

[PATCH 3/9] locking/mutex: Mark devm_mutex_init() as __must_check

[Boqun Feng]

From: Thomas Weißschuh <linux@weissschuh.net>

devm_mutex_init() can fail. With CONFIG_DEBUG_MUTEXES=y the mutex will be
marked as unusable and trigger errors on usage.
Enforce all callers check the return value through the compiler.

As devm_mutex_init() itself is a macro, it can not be annotated
directly. Annotate __devm_mutex_init() instead.
Unfortunately __must_check/warn_unused_result don't propagate through
statement expression. So move the statement expression into the argument
list of the call to __devm_mutex_init() through a helper macro.

Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Reviewed-by: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Link: https://lore.kernel.org/r/20250617-must_check-devm_mutex_init-v7-3-d9e449f4d224@weissschuh.net
---
 include/linux/mutex.h | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

diff --git a/include/linux/mutex.h b/include/linux/mutex.h
index a039fa8c1780..00afd341d293 100644
--- a/include/linux/mutex.h
+++ b/include/linux/mutex.h
@@ -126,11 +126,11 @@ do {							\
 
 #ifdef CONFIG_DEBUG_MUTEXES
 
-int __devm_mutex_init(struct device *dev, struct mutex *lock);
+int __must_check __devm_mutex_init(struct device *dev, struct mutex *lock);
 
 #else
 
-static inline int __devm_mutex_init(struct device *dev, struct mutex *lock)
+static inline int __must_check __devm_mutex_init(struct device *dev, struct mutex *lock)
 {
 	/*
 	 * When CONFIG_DEBUG_MUTEXES is off mutex_destroy() is just a nop so
@@ -141,14 +141,17 @@ static inline int __devm_mutex_init(struct device *dev, struct mutex *lock)
 
 #endif
 
-#define devm_mutex_init(dev, mutex)			\
+#define __mutex_init_ret(mutex)				\
 ({							\
 	typeof(mutex) mutex_ = (mutex);			\
 							\
 	mutex_init(mutex_);				\
-	__devm_mutex_init(dev, mutex_);			\
+	mutex_;						\
 })
 
+#define devm_mutex_init(dev, mutex) \
+	__devm_mutex_init(dev, __mutex_init_ret(mutex))
+
 /*
  * See kernel/locking/mutex.c for detailed documentation of these APIs.
  * Also see Documentation/locking/mutex-design.rst.
-- 
2.39.5 (Apple Git-154)

[PATCH 4/9] rust: sync: Add #[must_use] to Lock::try_lock()

[Boqun Feng]

From: Jason Devers <dev.json2@gmail.com>

The `Lock::try_lock()` function returns an `Option<Guard<...>>`, but it
currently does not issue a warning if the return value is unused.
To avoid potential bugs, the `#[must_use]` annotation is added to ensure
proper usage.

Note that `T` is `#[must_use]` but `Option<T>` is not.
For more context, see: https://github.com/rust-lang/rust/issues/71368.

Suggested-by: Alice Ryhl <aliceryhl@google.com>
Link: https://github.com/Rust-for-Linux/linux/issues/1133
Signed-off-by: Jason Devers <dev.json2@gmail.com>
Reviewed-by: Alice Ryhl <aliceryhl@google.com>
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Link: https://lore.kernel.org/r/20241212154753.139563-1-dev.json2@gmail.com
---
 rust/kernel/sync/lock.rs | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/rust/kernel/sync/lock.rs b/rust/kernel/sync/lock.rs
index e82fa5be289c..27202beef90c 100644
--- a/rust/kernel/sync/lock.rs
+++ b/rust/kernel/sync/lock.rs
@@ -175,6 +175,8 @@ pub fn lock(&self) -> Guard<'_, T, B> {
     /// Tries to acquire the lock.
     ///
     /// Returns a guard that can be used to access the data protected by the lock if successful.
+    // `Option<T>` is not `#[must_use]` even if `T` is, thus the attribute is needed here.
+    #[must_use = "if unused, the lock will be immediately unlocked"]
     pub fn try_lock(&self) -> Option<Guard<'_, T, B>> {
         // SAFETY: The constructor of the type calls `init`, so the existence of the object proves
         // that `init` was called.
-- 
2.39.5 (Apple Git-154)

[PATCH 5/9] locking/rwsem: Use OWNER_NONSPINNABLE directly instead of OWNER_SPINNABLE

[Boqun Feng]

From: Jinliang Zheng <alexjlzheng@tencent.com>

After commit 7d43f1ce9dd0 ("locking/rwsem: Enable time-based spinning on
reader-owned rwsem"), OWNER_SPINNABLE contains all possible values except
OWNER_NONSPINNABLE, namely OWNER_NULL | OWNER_WRITER | OWNER_READER.

Therefore, it is better to use OWNER_NONSPINNABLE directly to determine
whether to exit optimistic spin.

And, remove useless OWNER_SPINNABLE to simplify the code.

Signed-off-by: Jinliang Zheng <alexjlzheng@tencent.com>
Acked-by: Waiman Long <longman@redhat.com>
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Link: https://lore.kernel.org/r/20250610130158.4876-1-alexjlzheng@tencent.com
---
 kernel/locking/rwsem.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/kernel/locking/rwsem.c b/kernel/locking/rwsem.c
index 2ddb827e3bea..8572dba95af4 100644
--- a/kernel/locking/rwsem.c
+++ b/kernel/locking/rwsem.c
@@ -727,8 +727,6 @@ static inline bool rwsem_can_spin_on_owner(struct rw_semaphore *sem)
 	return ret;
 }
 
-#define OWNER_SPINNABLE		(OWNER_NULL | OWNER_WRITER | OWNER_READER)
-
 static inline enum owner_state
 rwsem_owner_state(struct task_struct *owner, unsigned long flags)
 {
@@ -835,7 +833,7 @@ static bool rwsem_optimistic_spin(struct rw_semaphore *sem)
 		enum owner_state owner_state;
 
 		owner_state = rwsem_spin_on_owner(sem);
-		if (!(owner_state & OWNER_SPINNABLE))
+		if (owner_state == OWNER_NONSPINNABLE)
 			break;
 
 		/*
-- 
2.39.5 (Apple Git-154)

[PATCH 6/9] locking/lockdep: Avoid struct return in lock_stats()

[Boqun Feng]

From: Arnd Bergmann <arnd@arndb.de>

Returning a large structure from the lock_stats() function causes clang
to have multiple copies of it on the stack and copy between them, which
can end up exceeding the frame size warning limit:

kernel/locking/lockdep.c:300:25: error: stack frame size (1464) exceeds limit (1280) in 'lock_stats' [-Werror,-Wframe-larger-than]
  300 | struct lock_class_stats lock_stats(struct lock_class *class)

Change the calling conventions to directly operate on the caller's copy,
which apparently is what gcc does already.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Link: https://lore.kernel.org/r/20250610092941.2642847-1-arnd@kernel.org
---
 include/linux/lockdep_types.h |  2 +-
 kernel/locking/lockdep.c      | 27 ++++++++++++---------------
 kernel/locking/lockdep_proc.c |  2 +-
 3 files changed, 14 insertions(+), 17 deletions(-)

diff --git a/include/linux/lockdep_types.h b/include/linux/lockdep_types.h
index 9f361d3ab9d9..eae115a26488 100644
--- a/include/linux/lockdep_types.h
+++ b/include/linux/lockdep_types.h
@@ -175,7 +175,7 @@ struct lock_class_stats {
 	unsigned long			bounces[nr_bounce_types];
 };
 
-struct lock_class_stats lock_stats(struct lock_class *class);
+void lock_stats(struct lock_class *class, struct lock_class_stats *stats);
 void clear_lock_stats(struct lock_class *class);
 #endif
 
diff --git a/kernel/locking/lockdep.c b/kernel/locking/lockdep.c
index dd2bbf73718b..0c941418a215 100644
--- a/kernel/locking/lockdep.c
+++ b/kernel/locking/lockdep.c
@@ -297,33 +297,30 @@ static inline void lock_time_add(struct lock_time *src, struct lock_time *dst)
 	dst->nr += src->nr;
 }
 
-struct lock_class_stats lock_stats(struct lock_class *class)
+void lock_stats(struct lock_class *class, struct lock_class_stats *stats)
 {
-	struct lock_class_stats stats;
 	int cpu, i;
 
-	memset(&stats, 0, sizeof(struct lock_class_stats));
+	memset(stats, 0, sizeof(struct lock_class_stats));
 	for_each_possible_cpu(cpu) {
 		struct lock_class_stats *pcs =
 			&per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
 
-		for (i = 0; i < ARRAY_SIZE(stats.contention_point); i++)
-			stats.contention_point[i] += pcs->contention_point[i];
+		for (i = 0; i < ARRAY_SIZE(stats->contention_point); i++)
+			stats->contention_point[i] += pcs->contention_point[i];
 
-		for (i = 0; i < ARRAY_SIZE(stats.contending_point); i++)
-			stats.contending_point[i] += pcs->contending_point[i];
+		for (i = 0; i < ARRAY_SIZE(stats->contending_point); i++)
+			stats->contending_point[i] += pcs->contending_point[i];
 
-		lock_time_add(&pcs->read_waittime, &stats.read_waittime);
-		lock_time_add(&pcs->write_waittime, &stats.write_waittime);
+		lock_time_add(&pcs->read_waittime, &stats->read_waittime);
+		lock_time_add(&pcs->write_waittime, &stats->write_waittime);
 
-		lock_time_add(&pcs->read_holdtime, &stats.read_holdtime);
-		lock_time_add(&pcs->write_holdtime, &stats.write_holdtime);
+		lock_time_add(&pcs->read_holdtime, &stats->read_holdtime);
+		lock_time_add(&pcs->write_holdtime, &stats->write_holdtime);
 
-		for (i = 0; i < ARRAY_SIZE(stats.bounces); i++)
-			stats.bounces[i] += pcs->bounces[i];
+		for (i = 0; i < ARRAY_SIZE(stats->bounces); i++)
+			stats->bounces[i] += pcs->bounces[i];
 	}
-
-	return stats;
 }
 
 void clear_lock_stats(struct lock_class *class)
diff --git a/kernel/locking/lockdep_proc.c b/kernel/locking/lockdep_proc.c
index b52c07c4707c..1916db9aa46b 100644
--- a/kernel/locking/lockdep_proc.c
+++ b/kernel/locking/lockdep_proc.c
@@ -657,7 +657,7 @@ static int lock_stat_open(struct inode *inode, struct file *file)
 			if (!test_bit(idx, lock_classes_in_use))
 				continue;
 			iter->class = class;
-			iter->stats = lock_stats(class);
+			lock_stats(class, &iter->stats);
 			iter++;
 		}
 
-- 
2.39.5 (Apple Git-154)

[PATCH 7/9] locking/lockdep: Change 'static const' variables to enum values

[Boqun Feng]

From: Arnd Bergmann <arnd@arndb.de>

gcc warns about 'static const' variables even in headers when building
with -Wunused-const-variables enabled:

In file included from kernel/locking/lockdep_proc.c:25:
kernel/locking/lockdep_internals.h:69:28: error: 'LOCKF_USED_IN_IRQ_READ' defined but not used [-Werror=unused-const-variable=]
   69 | static const unsigned long LOCKF_USED_IN_IRQ_READ =
      |                            ^~~~~~~~~~~~~~~~~~~~~~
kernel/locking/lockdep_internals.h:63:28: error: 'LOCKF_ENABLED_IRQ_READ' defined but not used [-Werror=unused-const-variable=]
   63 | static const unsigned long LOCKF_ENABLED_IRQ_READ =
      |                            ^~~~~~~~~~~~~~~~~~~~~~
kernel/locking/lockdep_internals.h:57:28: error: 'LOCKF_USED_IN_IRQ' defined but not used [-Werror=unused-const-variable=]
   57 | static const unsigned long LOCKF_USED_IN_IRQ =
      |                            ^~~~~~~~~~~~~~~~~
kernel/locking/lockdep_internals.h:51:28: error: 'LOCKF_ENABLED_IRQ' defined but not used [-Werror=unused-const-variable=]
   51 | static const unsigned long LOCKF_ENABLED_IRQ =
      |                            ^~~~~~~~~~~~~~~~~

This one is easy to avoid by changing the generated constant definition
into an equivalent enum.

Tested-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Link: https://lore.kernel.org/r/20250409122314.2848028-6-arnd@kernel.org
---
 kernel/locking/lockdep_internals.h | 18 ++++++++++--------
 1 file changed, 10 insertions(+), 8 deletions(-)

diff --git a/kernel/locking/lockdep_internals.h b/kernel/locking/lockdep_internals.h
index 82156caf77d1..0e5e6ffe91a3 100644
--- a/kernel/locking/lockdep_internals.h
+++ b/kernel/locking/lockdep_internals.h
@@ -47,29 +47,31 @@ enum {
 	__LOCKF(USED_READ)
 };
 
+enum {
 #define LOCKDEP_STATE(__STATE)	LOCKF_ENABLED_##__STATE |
-static const unsigned long LOCKF_ENABLED_IRQ =
+	LOCKF_ENABLED_IRQ =
 #include "lockdep_states.h"
-	0;
+	0,
 #undef LOCKDEP_STATE
 
 #define LOCKDEP_STATE(__STATE)	LOCKF_USED_IN_##__STATE |
-static const unsigned long LOCKF_USED_IN_IRQ =
+	LOCKF_USED_IN_IRQ =
 #include "lockdep_states.h"
-	0;
+	0,
 #undef LOCKDEP_STATE
 
 #define LOCKDEP_STATE(__STATE)	LOCKF_ENABLED_##__STATE##_READ |
-static const unsigned long LOCKF_ENABLED_IRQ_READ =
+	LOCKF_ENABLED_IRQ_READ =
 #include "lockdep_states.h"
-	0;
+	0,
 #undef LOCKDEP_STATE
 
 #define LOCKDEP_STATE(__STATE)	LOCKF_USED_IN_##__STATE##_READ |
-static const unsigned long LOCKF_USED_IN_IRQ_READ =
+	LOCKF_USED_IN_IRQ_READ =
 #include "lockdep_states.h"
-	0;
+	0,
 #undef LOCKDEP_STATE
+};
 
 #define LOCKF_ENABLED_IRQ_ALL (LOCKF_ENABLED_IRQ | LOCKF_ENABLED_IRQ_READ)
 #define LOCKF_USED_IN_IRQ_ALL (LOCKF_USED_IN_IRQ | LOCKF_USED_IN_IRQ_READ)
-- 
2.39.5 (Apple Git-154)

[PATCH 8/9] locking/mutex: Remove redundant #ifdefs

[Boqun Feng]

From: Ran Xiaokai <ran.xiaokai@zte.com.cn>

hung_task_{set,clear}_blocker() is already guarded by
CONFIG_DETECT_HUNG_TASK_BLOCKER in hung_task.h, So remove
the redudant check of #ifdef.

Signed-off-by: Ran Xiaokai <ran.xiaokai@zte.com.cn>
Acked-by: Waiman Long <longman@redhat.com>
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Link: https://lore.kernel.org/r/20250704015218.359754-1-ranxiaokai627@163.com
---
 kernel/locking/mutex.c | 4 ----
 1 file changed, 4 deletions(-)

diff --git a/kernel/locking/mutex.c b/kernel/locking/mutex.c
index a39ecccbd106..d4210dc97b6a 100644
--- a/kernel/locking/mutex.c
+++ b/kernel/locking/mutex.c
@@ -191,9 +191,7 @@ static void
 __mutex_add_waiter(struct mutex *lock, struct mutex_waiter *waiter,
 		   struct list_head *list)
 {
-#ifdef CONFIG_DETECT_HUNG_TASK_BLOCKER
 	hung_task_set_blocker(lock, BLOCKER_TYPE_MUTEX);
-#endif
 	debug_mutex_add_waiter(lock, waiter, current);
 
 	list_add_tail(&waiter->list, list);
@@ -209,9 +207,7 @@ __mutex_remove_waiter(struct mutex *lock, struct mutex_waiter *waiter)
 		__mutex_clear_flag(lock, MUTEX_FLAGS);
 
 	debug_mutex_remove_waiter(lock, waiter, current);
-#ifdef CONFIG_DETECT_HUNG_TASK_BLOCKER
 	hung_task_clear_blocker();
-#endif
 }
 
 /*
-- 
2.39.5 (Apple Git-154)

[PATCH 9/9] lockdep: Speed up lockdep_unregister_key() with expedited RCU synchronization

[Boqun Feng]

From: Breno Leitao <leitao@debian.org>

lockdep_unregister_key() is called from critical code paths, including
sections where rtnl_lock() is held. For example, when replacing a qdisc
in a network device, network egress traffic is disabled while
__qdisc_destroy() is called for every network queue.

If lockdep is enabled, __qdisc_destroy() calls lockdep_unregister_key(),
which gets blocked waiting for synchronize_rcu() to complete.

For example, a simple tc command to replace a qdisc could take 13
seconds:

  # time /usr/sbin/tc qdisc replace dev eth0 root handle 0x1: mq
    real    0m13.195s
    user    0m0.001s
    sys     0m2.746s

During this time, network egress is completely frozen while waiting for
RCU synchronization.

Use synchronize_rcu_expedited() instead to minimize the impact on
critical operations like network connectivity changes.

This improves 10x the function call to tc, when replacing the qdisc for
a network card.

   # time /usr/sbin/tc qdisc replace dev eth0 root handle 0x1: mq
     real     0m1.789s
     user     0m0.000s
     sys      0m1.613s

[boqun: Fixed the comment and add more information for the temporary
workaround, and add TODO information for hazptr]

Reported-by: Erik Lundgren <elundgren@meta.com>
Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Link: https://lore.kernel.org/r/20250321-lockdep-v1-1-78b732d195fb@debian.org
---
 kernel/locking/lockdep.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/kernel/locking/lockdep.c b/kernel/locking/lockdep.c
index 0c941418a215..2d4c5bab5af8 100644
--- a/kernel/locking/lockdep.c
+++ b/kernel/locking/lockdep.c
@@ -6616,8 +6616,16 @@ void lockdep_unregister_key(struct lock_class_key *key)
 	if (need_callback)
 		call_rcu(&delayed_free.rcu_head, free_zapped_rcu);
 
-	/* Wait until is_dynamic_key() has finished accessing k->hash_entry. */
-	synchronize_rcu();
+	/*
+	 * Wait until is_dynamic_key() has finished accessing k->hash_entry.
+	 *
+	 * Some operations like __qdisc_destroy() will call this in a debug
+	 * kernel, and the network traffic is disabled while waiting, hence
+	 * the delay of the wait matters in debugging cases. Currently use a
+	 * synchronize_rcu_expedited() to speed up the wait at the cost of
+	 * system IPIs. TODO: Replace RCU with hazptr for this.
+	 */
+	synchronize_rcu_expedited();
 }
 EXPORT_SYMBOL_GPL(lockdep_unregister_key);
 
-- 
2.39.5 (Apple Git-154)