extend hung task blocker tracking to rwsems
Subsystems (2)
cover extend hung task blocker tracking to rwsems
Hi all,
Inspired by mutex blocker tracking[1], and having already extended it to semaphores, let's now add support for reader-writer semaphores (rwsems).
The approach is simple: when a task enters TASK_UNINTERRUPTIBLE while waiting for an rwsem, we just call hung_task_set_blocker(). The hung task detector can then query the rwsem's owner to identify the lock holder.
Tracking works reliably for writers, as there can only be a single writer holding the lock, and its task struct is stored in the owner field.
The main challenge lies with readers. The owner field points to only one of many concurrent readers, so we might lose track of the blocker if that specific reader unlocks, even while others remain. This is not a significant issue, however. In practice, long-lasting lock contention is almost always caused by a writer. Therefore, reliably tracking the writer is the primary goal of this patch series ;)
With this change, the hung task detector can now show blocker task's info like below:
[Thu Jun 12 11:01:33 2025] INFO: task rw_sem_thread2:36526 blocked for more than 122 seconds. [Thu Jun 12 11:01:33 2025] Tainted: G S O 6.16.0-rc1 #1 [Thu Jun 12 11:01:33 2025] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [Thu Jun 12 11:01:33 2025] task:rw_sem_thread2 state:D stack:0 pid:36526 tgid:36526 ppid:2 task_flags:0x208040 flags:0x00004000 [Thu Jun 12 11:01:33 2025] Call Trace: [Thu Jun 12 11:01:33 2025] <TASK> [Thu Jun 12 11:01:33 2025] __schedule+0x7c7/0x1930 [Thu Jun 12 11:01:33 2025] ? __pfx___schedule+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? _raw_spin_lock_irq+0x8a/0xe0 [Thu Jun 12 11:01:33 2025] ? __pfx__raw_spin_lock_irq+0x10/0x10 [Thu Jun 12 11:01:33 2025] schedule+0x6a/0x180 [Thu Jun 12 11:01:33 2025] schedule_preempt_disabled+0x15/0x30 [Thu Jun 12 11:01:33 2025] rwsem_down_write_slowpath+0x447/0x1090 [Thu Jun 12 11:01:33 2025] ? __pfx_rwsem_down_write_slowpath+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx___schedule+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx___might_resched+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_thread2_func+0x10/0x10 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] down_write+0x125/0x140 [Thu Jun 12 11:01:33 2025] ? __pfx_down_write+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? msleep+0x91/0xf0 [Thu Jun 12 11:01:33 2025] ? __raw_spin_lock_irqsave+0x8c/0xf0 [Thu Jun 12 11:01:33 2025] thread2_func+0x37/0x70 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] kthread+0x39f/0x750 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx__raw_spin_lock_irq+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork+0x25d/0x320 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork_asm+0x1a/0x30 [Thu Jun 12 11:01:33 2025] </TASK> [Thu Jun 12 11:01:33 2025] INFO: task rw_sem_thread2:36526 <writer> blocked on an rw-semaphore likely owned by task rw_sem_thread1:36525 <writer> [Thu Jun 12 11:01:33 2025] task:rw_sem_thread1 state:S stack:0 pid:36525 tgid:36525 ppid:2 task_flags:0x208040 flags:0x00004000 [Thu Jun 12 11:01:33 2025] Call Trace: [Thu Jun 12 11:01:33 2025] <TASK> [Thu Jun 12 11:01:33 2025] __schedule+0x7c7/0x1930 [Thu Jun 12 11:01:33 2025] ? __pfx___schedule+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __mod_timer+0x304/0xa80 [Thu Jun 12 11:01:33 2025] ? irq_work_queue+0x6a/0xa0 [Thu Jun 12 11:01:33 2025] ? __pfx_vprintk_emit+0x10/0x10 [Thu Jun 12 11:01:33 2025] schedule+0x6a/0x180 [Thu Jun 12 11:01:33 2025] schedule_timeout+0xfb/0x230 [Thu Jun 12 11:01:33 2025] ? __pfx_schedule_timeout+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_process_timeout+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? down_write+0xc4/0x140 [Thu Jun 12 11:01:33 2025] msleep_interruptible+0xbe/0x150 [Thu Jun 12 11:01:33 2025] ? __pfx_thread1_func+0x10/0x10 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] thread1_func+0x37/0x60 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] kthread+0x39f/0x750 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx__raw_spin_lock_irq+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork+0x25d/0x320 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork_asm+0x1a/0x30 [Thu Jun 12 11:01:33 2025] </TASK>
Thanks,
Lance
Lance Yang (3):
locking/rwsem: make owner helpers globally available locking/rwsem: clear reader-owner on unlock to reduce false positives
hung_task: extend hung task blocker tracking to rwsems
include/linux/hung_task.h | 18 +++++++++--------- include/linux/rwsem.h | 12 ++++++++++++ kernel/hung_task.c | 29 +++++++++++++++++++++++++---- kernel/locking/rwsem.c | 31 +++++++++++++++++++++++-------- 4 files changed, 69 insertions(+), 21 deletions(-)
Signature · 1 line
2.49.0
RFC PATCH 1/3 locking/rwsem: make owner helpers globally available
From: Lance Yang <[email protected]>
In preparation for extending blocker tracking to support rwsems, make the rwsem_owner() and is_rwsem_reader_owned() helpers globally available for determining if the blocker is a writer or one of the readers.
--- include/linux/rwsem.h | 12 ++++++++++++ kernel/locking/rwsem.c | 8 +++++--- 2 files changed, 17 insertions(+), 3 deletions(-)
include/linux/rwsem.h +12 -0
@@ -132,6 +132,18 @@ static inline int rwsem_is_contended(struct rw_semaphore *sem) return !list_empty(&sem->wait_list); } +#if defined(CONFIG_DEBUG_RWSEMS ) || defined(CONFIG_DETECT_HUNG_TASK_BLOCKER ) +/* +* Return just the real task structure pointer of the owner +*/ +extern struct task_struct *rwsem_owner (struct rw_semaphore *sem); + +/* +* Return true if the rwsem is owned by a reader. +*/ +extern bool is_rwsem_reader_owned (struct rw_semaphore *sem); +#endif +#else /* !CONFIG_PREEMPT_RT */ #include <linux/rwbase_rt.h>
kernel/locking/rwsem.c +5 -3
@@ -181,11 +181,11 @@ static inline void rwsem_set_reader_owned(struct rw_semaphore *sem)__rwsem_set_reader_owned (sem, current); } -#ifdef CONFIG_DEBUG_RWSEMS +#if defined(CONFIG_DEBUG_RWSEMS ) || defined(CONFIG_DETECT_HUNG_TASK_BLOCKER )/* * Return just the real task structure pointer of the owner */ -static inline struct task_struct *rwsem_owner (struct rw_semaphore *sem) +struct task_struct *rwsem_owner (struct rw_semaphore *sem) {return (struct task_struct *) (atomic_long_read (&sem->owner) & ~RWSEM_OWNER_FLAGS_MASK );
@@ -194,7 +194,7 @@ static inline struct task_struct *rwsem_owner(struct rw_semaphore *sem)/* * Return true if the rwsem is owned by a reader. */ -static inline bool is_rwsem_reader_owned (struct rw_semaphore *sem) +bool is_rwsem_reader_owned (struct rw_semaphore *sem) {/* * Check the count to see if it is write-locked.
@@ -205,7 +205,9 @@ static inline bool is_rwsem_reader_owned(struct rw_semaphore *sem)return false; return rwsem_test_oflags(sem, RWSEM_READER_OWNED); } +#endif +#ifdef CONFIG_DEBUG_RWSEMS /* * With CONFIG_DEBUG_RWSEMS configured, it will make sure that if there * is a task pointer in owner of a reader-owned rwsem, it will be the
Signature · 1 line
2.49.0
Re: [PATCH RFC 1/3] locking/rwsem: make owner helpers globally available
Quoted 6 lines · On Thu, 12 Jun 2025 12:19:24 +0800 Lance Yang <[email protected]> wrote:
From: Lance Yang <[email protected]>
In preparation for extending blocker tracking to support rwsems, make the
rwsem_owner()andis_rwsem_reader_owned()helpers globally available for determining if the blocker is a writer or one of the readers.
Looks good to me.
Thanks!
Quoted 69 lines
Signed-off-by Lance Yang <[email protected]>--- include/linux/rwsem.h | 12 ++++++++++++ kernel/locking/rwsem.c | 8 +++++--- 2 files changed, 17 insertions(+), 3 deletions(-)include/linux/rwsem.h +12
-0@@ -132,6 +132,18 @@ static inline int rwsem_is_contended(struct rw_semaphore *sem) return !list_empty(&sem->wait_list); } +#if defined(CONFIG_DEBUG_RWSEMS ) || defined(CONFIG_DETECT_HUNG_TASK_BLOCKER ) +/* +* Return just the real task structure pointer of the owner +*/ +extern struct task_struct *rwsem_owner (struct rw_semaphore *sem); + +/* +* Return true if the rwsem is owned by a reader. +*/ +extern bool is_rwsem_reader_owned (struct rw_semaphore *sem); +#endif +#else /* !CONFIG_PREEMPT_RT */ #include <linux/rwbase_rt.h> kernel/locking/rwsem.c +5
-3@@ -181,11 +181,11 @@ static inline void rwsem_set_reader_owned(struct rw_semaphore *sem)__rwsem_set_reader_owned (sem, current); } -#ifdef CONFIG_DEBUG_RWSEMS +#if defined(CONFIG_DEBUG_RWSEMS ) || defined(CONFIG_DETECT_HUNG_TASK_BLOCKER )/* * Return just the real task structure pointer of the owner */ -static inline struct task_struct *rwsem_owner (struct rw_semaphore *sem) +struct task_struct *rwsem_owner (struct rw_semaphore *sem) {return (struct task_struct *) (atomic_long_read (&sem->owner) & ~RWSEM_OWNER_FLAGS_MASK );@@ -194,7 +194,7 @@ static inline struct task_struct *rwsem_owner(struct rw_semaphore *sem)/* * Return true if the rwsem is owned by a reader. */ -static inline bool is_rwsem_reader_owned (struct rw_semaphore *sem) +bool is_rwsem_reader_owned (struct rw_semaphore *sem) {/* * Check the count to see if it is write-locked. @@ -205,7 +205,9 @@ static inline bool is_rwsem_reader_owned(struct rw_semaphore *sem)return false; return rwsem_test_oflags(sem, RWSEM_READER_OWNED); } +#endif +#ifdef CONFIG_DEBUG_RWSEMS /* * With CONFIG_DEBUG_RWSEMS configured, it will make sure that if there * is a task pointer in owner of a reader-owned rwsem, it will be the Signature · 1 line
2.49.0
Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
Lance Yang
maintainerRe: [PATCH RFC 1/3] locking/rwsem: make owner helpers globally available
Quoted 10 lines · Masami Hiramatsu (Google)
Quoted 6 lines · On Thu, 12 Jun 2025 12:19:24 +0800 Lance Yang <[email protected]> wrote:
From: Lance Yang <[email protected]>
In preparation for extending blocker tracking to support rwsems, make the
rwsem_owner()andis_rwsem_reader_owned()helpers globally available for determining if the blocker is a writer or one of the readers.
Quoted 5 lines
Looks good to me.
Reviewed-by Masami Hiramatsu (Google) <[email protected]>Thanks!
Thanks for taking the time to review!
Lance
Quoted 72 lines
Quoted 70 lines
Signed-off-by Lance Yang <[email protected]>--- include/linux/rwsem.h | 12 ++++++++++++ kernel/locking/rwsem.c | 8 +++++--- 2 files changed, 17 insertions(+), 3 deletions(-)include/linux/rwsem.h +12
-0@@ -132,6 +132,18 @@ static inline int rwsem_is_contended(struct rw_semaphore *sem) return !list_empty(&sem->wait_list); } +#if defined(CONFIG_DEBUG_RWSEMS ) || defined(CONFIG_DETECT_HUNG_TASK_BLOCKER ) +/* +* Return just the real task structure pointer of the owner +*/ +extern struct task_struct *rwsem_owner (struct rw_semaphore *sem); + +/* +* Return true if the rwsem is owned by a reader. +*/ +extern bool is_rwsem_reader_owned (struct rw_semaphore *sem); +#endif +#else /* !CONFIG_PREEMPT_RT */ #include <linux/rwbase_rt.h> kernel/locking/rwsem.c +5
-3@@ -181,11 +181,11 @@ static inline void rwsem_set_reader_owned(struct rw_semaphore *sem)__rwsem_set_reader_owned (sem, current); } -#ifdef CONFIG_DEBUG_RWSEMS +#if defined(CONFIG_DEBUG_RWSEMS ) || defined(CONFIG_DETECT_HUNG_TASK_BLOCKER )/* * Return just the real task structure pointer of the owner */ -static inline struct task_struct *rwsem_owner (struct rw_semaphore *sem) +struct task_struct *rwsem_owner (struct rw_semaphore *sem) {return (struct task_struct *) (atomic_long_read (&sem->owner) & ~RWSEM_OWNER_FLAGS_MASK );@@ -194,7 +194,7 @@ static inline struct task_struct *rwsem_owner(struct rw_semaphore *sem)/* * Return true if the rwsem is owned by a reader. */ -static inline bool is_rwsem_reader_owned (struct rw_semaphore *sem) +bool is_rwsem_reader_owned (struct rw_semaphore *sem) {/* * Check the count to see if it is write-locked. @@ -205,7 +205,9 @@ static inline bool is_rwsem_reader_owned(struct rw_semaphore *sem)return false; return rwsem_test_oflags(sem, RWSEM_READER_OWNED); } +#endif +#ifdef CONFIG_DEBUG_RWSEMS /* * With CONFIG_DEBUG_RWSEMS configured, it will make sure that if there * is a task pointer in owner of a reader-owned rwsem, it will be the Signature · 1 line
2.49.0
RFC PATCH 2/3 locking/rwsem: clear reader-owner on unlock to reduce false positives
In reply to Lance YangFrom: Lance Yang <[email protected]>
When CONFIG_DETECT_HUNG_TASK_BLOCKER is enabled, a stale owner pointer in a reader-owned rwsem can lead to false positives in blocker tracking.
To mitigate this, let’s try to clear the owner field on unlock, as a NULL owner is better than a stale one for diagnostics.
--- kernel/locking/rwsem.c | 10 ++++------ 1 file changed, 4 insertions(+), 6 deletions(-)
kernel/locking/rwsem.c +4 -6
@@ -205,14 +205,12 @@ bool is_rwsem_reader_owned(struct rw_semaphore *sem)return false;return rwsem_test_oflags (sem,RWSEM_READER_OWNED ); } -#endif -#ifdef CONFIG_DEBUG_RWSEMS /* -* With CONFIG_DEBUG_RWSEMS configured, it will make sure that if there -* is a task pointer in owner of a reader-owned rwsem, it will be the -* real owner or one of the real owners. The only exception is when the -* unlock is done by up_read_non_owner(). +* With CONFIG_DEBUG_RWSEMS or CONFIG_DETECT_HUNG_TASK_BLOCKER configured, +* it will make sure that the owner field of a reader-owned rwsem either +* points to a real reader-owner(s) or gets cleared. The only exception is +* when the unlock is done by up_read_non_owner(). */ static inline void rwsem_clear_reader_owned (struct rw_semaphore *sem) {
Signature · 1 line
2.49.0
Masami Hiramatsu (Google)
reviewerRe: [PATCH RFC 2/3] locking/rwsem: clear reader-owner on unlock to reduce false positives
Quoted 7 lines · On Thu, 12 Jun 2025 12:19:25 +0800 Lance Yang <[email protected]> wrote:
From: Lance Yang <[email protected]>
When
CONFIG_DETECT_HUNG_TASK_BLOCKERis enabled, a stale owner pointer in a reader-owned rwsem can lead to false positives in blocker tracking.To mitigate this, let’s try to clear the owner field on unlock, as a NULL owner is better than a stale one for diagnostics.
Can we merge this to [PATCH 1/3]? It seems that you removed #ifdef and remove it. This means in anyway we need the feature enabled by DEBUG_RWSEMS.
Thanks,
Quoted 32 lines
Signed-off-by Lance Yang <[email protected]>--- kernel/locking/rwsem.c | 10 ++++------ 1 file changed, 4 insertions(+), 6 deletions(-)kernel/locking/rwsem.c +4
-6@@ -205,14 +205,12 @@ bool is_rwsem_reader_owned(struct rw_semaphore *sem)return false;return rwsem_test_oflags (sem,RWSEM_READER_OWNED ); } -#endif -#ifdef CONFIG_DEBUG_RWSEMS /* -* With CONFIG_DEBUG_RWSEMS configured, it will make sure that if there -* is a task pointer in owner of a reader-owned rwsem, it will be the -* real owner or one of the real owners. The only exception is when the -* unlock is done by up_read_non_owner(). +* With CONFIG_DEBUG_RWSEMS or CONFIG_DETECT_HUNG_TASK_BLOCKER configured, +* it will make sure that the owner field of a reader-owned rwsem either +* points to a real reader-owner(s) or gets cleared. The only exception is +* when the unlock is done by up_read_non_owner(). */ static inline void rwsem_clear_reader_owned (struct rw_semaphore *sem) {Signature · 1 line
2.49.0
Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
Lance Yang
maintainerRe: [PATCH RFC 2/3] locking/rwsem: clear reader-owner on unlock to reduce false positives
Quoted 5 lines · Masami Hiramatsu (Google)
Quoted 2 lines · On Thu, 12 Jun 2025 12:19:25 +0800 Lance Yang <[email protected]> wrote:
From: Lance Yang <[email protected]>
Quoted 8 lines
Quoted 5 lines
When
CONFIG_DETECT_HUNG_TASK_BLOCKERis enabled, a stale owner pointer in a reader-owned rwsem can lead to false positives in blocker tracking.To mitigate this, let’s try to clear the owner field on unlock, as a NULL owner is better than a stale one for diagnostics.
Can we merge this to [PATCH 1/3]? It seems that you removed #ifdef and remove it. This means in anyway we need the feature enabled by
DEBUG_RWSEMS.
Thanks for the feedback! I see your point about the dependency ;)
Personlly, I'd perfer to keep them separate. The reasoning is that they addreess two distinct things, and I think splitting them makes this series clearer and easier to review ;)
Patch #1 focuses on "ownership tracking": Its only job is to make the existing owner-related helpers (rwsem_owner(), is_rwsem_reader_owned()) globally available when blocker tracking is enabled.
Patch #2, on the other hand, is about "reader-owner cleanup": It introduces a functional change to the unlock path, trying to clear the owner field for reader-owned rwsems.
Does this reasoning make sense to you?
Thanks,
Lance
Quoted 37 lines
Thanks,
Quoted 32 lines
Signed-off-by Lance Yang <[email protected]>--- kernel/locking/rwsem.c | 10 ++++------ 1 file changed, 4 insertions(+), 6 deletions(-)kernel/locking/rwsem.c +4
-6@@ -205,14 +205,12 @@ bool is_rwsem_reader_owned(struct rw_semaphore *sem)return false;return rwsem_test_oflags (sem,RWSEM_READER_OWNED ); } -#endif -#ifdef CONFIG_DEBUG_RWSEMS /* -* With CONFIG_DEBUG_RWSEMS configured, it will make sure that if there -* is a task pointer in owner of a reader-owned rwsem, it will be the -* real owner or one of the real owners. The only exception is when the -* unlock is done by up_read_non_owner(). +* With CONFIG_DEBUG_RWSEMS or CONFIG_DETECT_HUNG_TASK_BLOCKER configured, +* it will make sure that the owner field of a reader-owned rwsem either +* points to a real reader-owner(s) or gets cleared. The only exception is +* when the unlock is done by up_read_non_owner(). */ static inline void rwsem_clear_reader_owned (struct rw_semaphore *sem) {Signature · 1 line
2.49.0
Masami Hiramatsu (Google)
reviewerRe: [PATCH RFC 2/3] locking/rwsem: clear reader-owner on unlock to reduce false positives
Quoted 23 lines · On Tue, 24 Jun 2025 09:44:55 +0800 Lance Yang <[email protected]> wrote:
Quoted 5 lines · Masami Hiramatsu (Google)
Quoted 2 lines · On Thu, 12 Jun 2025 12:19:25 +0800 Lance Yang <[email protected]> wrote:
From: Lance Yang <[email protected]>
Quoted 8 lines
Quoted 5 lines
When
CONFIG_DETECT_HUNG_TASK_BLOCKERis enabled, a stale owner pointer in a reader-owned rwsem can lead to false positives in blocker tracking.To mitigate this, let’s try to clear the owner field on unlock, as a NULL owner is better than a stale one for diagnostics.
Can we merge this to [PATCH 1/3]? It seems that you removed #ifdef and remove it. This means in anyway we need the feature enabled by
DEBUG_RWSEMS.Thanks for the feedback! I see your point about the dependency ;)
Personlly, I'd perfer to keep them separate. The reasoning is that they addreess two distinct things, and I think splitting them makes this series clearer and easier to review ;)
Quoted 7 lines
Patch #1 focuses on "ownership tracking": Its only job is to make the existing owner-related helpers (
rwsem_owner(),is_rwsem_reader_owned()) globally available when blocker tracking is enabled.Patch #2, on the other hand, is about "reader-owner cleanup": It introduces a functional change to the unlock path, trying to clear the owner field for reader-owned rwsems.
But without clearing the owner, the owner information can be broken, right? Since CONFIG_DEBUG_RWSEMS is working as it is, I think those cannot be decoupled. For example, comparing the result of both DETECT_HUNG_TASK_BLOCKER and DEBUG_RWSEMS are enabled and only DETECT_HUNG_TASK_BLOCKER is enabled, the result is different.
Quoted 2 lines
Does this reasoning make sense to you?
Sorry, no. I think "reader-owner cleanup" is a part of "ownership tracking" as DEBUG_RWSEMS does (and that keeps consistency of the ownership tracking behavior same as DEBUG_RWSEM).
Thank you,
Quoted 42 lines
Thanks,
LanceQuoted 37 lines
Thanks,
Quoted 32 lines
Signed-off-by Lance Yang <[email protected]>--- kernel/locking/rwsem.c | 10 ++++------ 1 file changed, 4 insertions(+), 6 deletions(-)kernel/locking/rwsem.c +4
-6@@ -205,14 +205,12 @@ bool is_rwsem_reader_owned(struct rw_semaphore *sem)return false;return rwsem_test_oflags (sem,RWSEM_READER_OWNED ); } -#endif -#ifdef CONFIG_DEBUG_RWSEMS /* -* With CONFIG_DEBUG_RWSEMS configured, it will make sure that if there -* is a task pointer in owner of a reader-owned rwsem, it will be the -* real owner or one of the real owners. The only exception is when the -* unlock is done by up_read_non_owner(). +* With CONFIG_DEBUG_RWSEMS or CONFIG_DETECT_HUNG_TASK_BLOCKER configured, +* it will make sure that the owner field of a reader-owned rwsem either +* points to a real reader-owner(s) or gets cleared. The only exception is +* when the unlock is done by up_read_non_owner(). */ static inline void rwsem_clear_reader_owned (struct rw_semaphore *sem) {Signature · 1 line
2.49.0
Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
Lance Yang
maintainerRe: [PATCH RFC 2/3] locking/rwsem: clear reader-owner on unlock to reduce false positives
Quoted 30 lines · Masami Hiramatsu (Google)
Quoted 27 lines · On Tue, 24 Jun 2025 09:44:55 +0800 Lance Yang <[email protected]> wrote:
Quoted 5 lines · Masami Hiramatsu (Google)
Quoted 2 lines · On Thu, 12 Jun 2025 12:19:25 +0800 Lance Yang <[email protected]> wrote:
From: Lance Yang <[email protected]>
Quoted 8 lines
Quoted 5 lines
When
CONFIG_DETECT_HUNG_TASK_BLOCKERis enabled, a stale owner pointer in a reader-owned rwsem can lead to false positives in blocker tracking.To mitigate this, let’s try to clear the owner field on unlock, as a NULL owner is better than a stale one for diagnostics.
Can we merge this to [PATCH 1/3]? It seems that you removed #ifdef and remove it. This means in anyway we need the feature enabled by
DEBUG_RWSEMS.Thanks for the feedback! I see your point about the dependency ;)
Personlly, I'd perfer to keep them separate. The reasoning is that they addreess two distinct things, and I think splitting them makes this series clearer and easier to review ;)
Patch #1 focuses on "ownership tracking": Its only job is to make the existing owner-related helpers (
rwsem_owner(),is_rwsem_reader_owned()) globally available when blocker tracking is enabled.
Quoted 6 lines
Quoted 3 lines
Patch #2, on the other hand, is about "reader-owner cleanup": It introduces a functional change to the unlock path, trying to clear the owner field for reader-owned rwsems.
But without clearing the owner, the owner information can be broken, right? Since
CONFIG_DEBUG_RWSEMSis working as it is,
You're right, the owner info would be broken without the cleanup logic in patch #2. But ...
Quoted 4 lines
I think those cannot be decoupled. For example, comparing the result of both
DETECT_HUNG_TASK_BLOCKERandDEBUG_RWSEMSare enabled and onlyDETECT_HUNG_TASK_BLOCKERis enabled, the result is different.
The actual blocker tracking for rwsems is only turned on in patch #3. So, there's no case where the feature is active without the cleanup logic already being in place.
Quoted 7 lines
Quoted 3 lines
Does this reasoning make sense to you?
Sorry, no. I think "reader-owner cleanup" is a part of "ownership tracking" as
DEBUG_RWSEMSdoes (and that keeps consistency of the ownership tracking behavior same asDEBUG_RWSEM).
I thought this step-by-step approach was a bit cleaner, since there are currently only two users for these owner helpers (DEBUG_RWSEMS and DETECT_HUNG_TASK_BLOCKER).
Anyway, if you still feel strongly that they should be merged, I'm happy to rework the series as you suggested ;p
Thanks,
Lance
Quoted 47 lines
Thank you,
Quoted 42 lines
Thanks,
LanceQuoted 37 lines
Thanks,
Quoted 32 lines
Signed-off-by Lance Yang <[email protected]>--- kernel/locking/rwsem.c | 10 ++++------ 1 file changed, 4 insertions(+), 6 deletions(-)kernel/locking/rwsem.c +4
-6@@ -205,14 +205,12 @@ bool is_rwsem_reader_owned(struct rw_semaphore *sem)return false;return rwsem_test_oflags (sem,RWSEM_READER_OWNED ); } -#endif -#ifdef CONFIG_DEBUG_RWSEMS /* -* With CONFIG_DEBUG_RWSEMS configured, it will make sure that if there -* is a task pointer in owner of a reader-owned rwsem, it will be the -* real owner or one of the real owners. The only exception is when the -* unlock is done by up_read_non_owner(). +* With CONFIG_DEBUG_RWSEMS or CONFIG_DETECT_HUNG_TASK_BLOCKER configured, +* it will make sure that the owner field of a reader-owned rwsem either +* points to a real reader-owner(s) or gets cleared. The only exception is +* when the unlock is done by up_read_non_owner(). */ static inline void rwsem_clear_reader_owned (struct rw_semaphore *sem) {Signature · 1 line
2.49.0
Masami Hiramatsu (Google)
reviewerRe: [PATCH RFC 2/3] locking/rwsem: clear reader-owner on unlock to reduce false positives
Quoted 56 lines · On Tue, 24 Jun 2025 13:02:31 +0800 Lance Yang <[email protected]> wrote:
Quoted 30 lines · Masami Hiramatsu (Google)
Quoted 27 lines · On Tue, 24 Jun 2025 09:44:55 +0800 Lance Yang <[email protected]> wrote:
Quoted 5 lines · Masami Hiramatsu (Google)
Quoted 2 lines · On Thu, 12 Jun 2025 12:19:25 +0800 Lance Yang <[email protected]> wrote:
From: Lance Yang <[email protected]>
Quoted 8 lines
Quoted 5 lines
When
CONFIG_DETECT_HUNG_TASK_BLOCKERis enabled, a stale owner pointer in a reader-owned rwsem can lead to false positives in blocker tracking.To mitigate this, let’s try to clear the owner field on unlock, as a NULL owner is better than a stale one for diagnostics.
Can we merge this to [PATCH 1/3]? It seems that you removed #ifdef and remove it. This means in anyway we need the feature enabled by
DEBUG_RWSEMS.Thanks for the feedback! I see your point about the dependency ;)
Personlly, I'd perfer to keep them separate. The reasoning is that they addreess two distinct things, and I think splitting them makes this series clearer and easier to review ;)
Patch #1 focuses on "ownership tracking": Its only job is to make the existing owner-related helpers (
rwsem_owner(),is_rwsem_reader_owned()) globally available when blocker tracking is enabled.Quoted 6 lines
Quoted 3 lines
Patch #2, on the other hand, is about "reader-owner cleanup": It introduces a functional change to the unlock path, trying to clear the owner field for reader-owned rwsems.
But without clearing the owner, the owner information can be broken, right? Since
CONFIG_DEBUG_RWSEMSis working as it is,You're right, the owner info would be broken without the cleanup logic in patch #2. But ...
Quoted 4 lines
I think those cannot be decoupled. For example, comparing the result of both
DETECT_HUNG_TASK_BLOCKERandDEBUG_RWSEMSare enabled and onlyDETECT_HUNG_TASK_BLOCKERis enabled, the result is different.The actual blocker tracking for rwsems is only turned on in patch #3. So, there's no case where the feature is active without the cleanup logic already being in place.
Quoted 4 lines
Quoted 3 lines
Does this reasoning make sense to you?
Quoted 7 lines
Quoted 3 lines
Sorry, no. I think "reader-owner cleanup" is a part of "ownership tracking" as
DEBUG_RWSEMSdoes (and that keeps consistency of the ownership tracking behavior same asDEBUG_RWSEM).I thought this step-by-step approach was a bit cleaner, since there are currently only two users for these owner helpers (
DEBUG_RWSEMSandDETECT_HUNG_TASK_BLOCKER).
I think the step-by-step approach fits better if the feature is evolving (a working feature is already there.) I don't like the intermediate state which does not work correctly, because if we have a unit test( like kUnit) it should fail. If you can say "this finds the rwsem owner as same as what the CONFIG_DEBUG_RWSEM is doing", it is simpler to explain what you are doing, and easy to understand.
Quoted 3 lines
Anyway, if you still feel strongly that they should be merged, I'm happy to rework the series as you suggested ;p
Thanks,
Quoted 53 lines
Thanks,
LanceQuoted 47 lines
Thank you,
Quoted 42 lines
Thanks,
LanceQuoted 37 lines
Thanks,
Quoted 32 lines
Signed-off-by Lance Yang <[email protected]>--- kernel/locking/rwsem.c | 10 ++++------ 1 file changed, 4 insertions(+), 6 deletions(-)kernel/locking/rwsem.c +4
-6@@ -205,14 +205,12 @@ bool is_rwsem_reader_owned(struct rw_semaphore *sem)return false;return rwsem_test_oflags (sem,RWSEM_READER_OWNED ); } -#endif -#ifdef CONFIG_DEBUG_RWSEMS /* -* With CONFIG_DEBUG_RWSEMS configured, it will make sure that if there -* is a task pointer in owner of a reader-owned rwsem, it will be the -* real owner or one of the real owners. The only exception is when the -* unlock is done by up_read_non_owner(). +* With CONFIG_DEBUG_RWSEMS or CONFIG_DETECT_HUNG_TASK_BLOCKER configured, +* it will make sure that the owner field of a reader-owned rwsem either +* points to a real reader-owner(s) or gets cleared. The only exception is +* when the unlock is done by up_read_non_owner(). */ static inline void rwsem_clear_reader_owned (struct rw_semaphore *sem) {Signature · 1 line
2.49.0
Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
Lance Yang
maintainerRe: [PATCH RFC 2/3] locking/rwsem: clear reader-owner on unlock to reduce false positives
Quoted 63 lines · Masami Hiramatsu (Google)
Quoted 60 lines · On Tue, 24 Jun 2025 13:02:31 +0800 Lance Yang <[email protected]> wrote:
Quoted 30 lines · Masami Hiramatsu (Google)
Quoted 27 lines · On Tue, 24 Jun 2025 09:44:55 +0800 Lance Yang <[email protected]> wrote:
Quoted 5 lines · Masami Hiramatsu (Google)
Quoted 2 lines · On Thu, 12 Jun 2025 12:19:25 +0800 Lance Yang <[email protected]> wrote:
From: Lance Yang <[email protected]>
Quoted 8 lines
Quoted 5 lines
When
CONFIG_DETECT_HUNG_TASK_BLOCKERis enabled, a stale owner pointer in a reader-owned rwsem can lead to false positives in blocker tracking.To mitigate this, let’s try to clear the owner field on unlock, as a NULL owner is better than a stale one for diagnostics.
Can we merge this to [PATCH 1/3]? It seems that you removed #ifdef and remove it. This means in anyway we need the feature enabled by
DEBUG_RWSEMS.Thanks for the feedback! I see your point about the dependency ;)
Personlly, I'd perfer to keep them separate. The reasoning is that they addreess two distinct things, and I think splitting them makes this series clearer and easier to review ;)
Patch #1 focuses on "ownership tracking": Its only job is to make the existing owner-related helpers (
rwsem_owner(),is_rwsem_reader_owned()) globally available when blocker tracking is enabled.Quoted 6 lines
Quoted 3 lines
Patch #2, on the other hand, is about "reader-owner cleanup": It introduces a functional change to the unlock path, trying to clear the owner field for reader-owned rwsems.
But without clearing the owner, the owner information can be broken, right? Since
CONFIG_DEBUG_RWSEMSis working as it is,You're right, the owner info would be broken without the cleanup logic in patch #2. But ...
Quoted 4 lines
I think those cannot be decoupled. For example, comparing the result of both
DETECT_HUNG_TASK_BLOCKERandDEBUG_RWSEMSare enabled and onlyDETECT_HUNG_TASK_BLOCKERis enabled, the result is different.The actual blocker tracking for rwsems is only turned on in patch #3. So, there's no case where the feature is active without the cleanup logic already being in place.
Quoted 7 lines
Quoted 3 lines
Does this reasoning make sense to you?
Sorry, no. I think "reader-owner cleanup" is a part of "ownership tracking" as
DEBUG_RWSEMSdoes (and that keeps consistency of the ownership tracking behavior same asDEBUG_RWSEM).
Quoted 6 lines
Quoted 3 lines
I thought this step-by-step approach was a bit cleaner, since there are currently only two users for these owner helpers (
DEBUG_RWSEMSandDETECT_HUNG_TASK_BLOCKER).I think the step-by-step approach fits better if the feature is evolving (a working feature is already there.) I don't like the intermediate
Agreed.
Quoted 2 lines
state which does not work correctly, because if we have a unit test( like kUnit) it should fail. If you can say "this finds the rwsem
Ah, I missed that ...
Quoted 2 lines
owner as same as what the
CONFIG_DEBUG_RWSEMis doing", it is simpler to explain what you are doing, and easy to understand.
Thanks for the lesson! Will rework the series as you suggested ;) Lance
RFC PATCH 3/3 hung_task: extend hung task blocker tracking to rwsems
In reply to Lance YangFrom: Lance Yang <[email protected]>
Inspired by mutex blocker tracking[1], and having already extended it to semaphores, let's now add support for reader-writer semaphores (rwsems).
The approach is simple: when a task enters TASK_UNINTERRUPTIBLE while waiting for an rwsem, we just call hung_task_set_blocker(). The hung task detector can then query the rwsem's owner to identify the lock holder.
Tracking works reliably for writers, as there can only be a single writer holding the lock, and its task struct is stored in the owner field.
The main challenge lies with readers. The owner field points to only one of many concurrent readers, so we might lose track of the blocker if that specific reader unlocks, even while others remain. This is not a significant issue, however. In practice, long-lasting lock contention is almost always caused by a writer. Therefore, reliably tracking the writer is the primary goal of this patch series ;)
With this change, the hung task detector can now show blocker task's info like below:
[Thu Jun 12 11:01:33 2025] INFO: task rw_sem_thread2:36526 blocked for more than 122 seconds. [Thu Jun 12 11:01:33 2025] Tainted: G S O 6.16.0-rc1 #1 [Thu Jun 12 11:01:33 2025] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [Thu Jun 12 11:01:33 2025] task:rw_sem_thread2 state:D stack:0 pid:36526 tgid:36526 ppid:2 task_flags:0x208040 flags:0x00004000 [Thu Jun 12 11:01:33 2025] Call Trace: [Thu Jun 12 11:01:33 2025] <TASK> [Thu Jun 12 11:01:33 2025] __schedule+0x7c7/0x1930 [Thu Jun 12 11:01:33 2025] ? __pfx___schedule+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? _raw_spin_lock_irq+0x8a/0xe0 [Thu Jun 12 11:01:33 2025] ? __pfx__raw_spin_lock_irq+0x10/0x10 [Thu Jun 12 11:01:33 2025] schedule+0x6a/0x180 [Thu Jun 12 11:01:33 2025] schedule_preempt_disabled+0x15/0x30 [Thu Jun 12 11:01:33 2025] rwsem_down_write_slowpath+0x447/0x1090 [Thu Jun 12 11:01:33 2025] ? __pfx_rwsem_down_write_slowpath+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx___schedule+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx___might_resched+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_thread2_func+0x10/0x10 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] down_write+0x125/0x140 [Thu Jun 12 11:01:33 2025] ? __pfx_down_write+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? msleep+0x91/0xf0 [Thu Jun 12 11:01:33 2025] ? __raw_spin_lock_irqsave+0x8c/0xf0 [Thu Jun 12 11:01:33 2025] thread2_func+0x37/0x70 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] kthread+0x39f/0x750 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx__raw_spin_lock_irq+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork+0x25d/0x320 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork_asm+0x1a/0x30 [Thu Jun 12 11:01:33 2025] </TASK> [Thu Jun 12 11:01:33 2025] INFO: task rw_sem_thread2:36526 <writer> blocked on an rw-semaphore likely owned by task rw_sem_thread1:36525 <writer> [Thu Jun 12 11:01:33 2025] task:rw_sem_thread1 state:S stack:0 pid:36525 tgid:36525 ppid:2 task_flags:0x208040 flags:0x00004000 [Thu Jun 12 11:01:33 2025] Call Trace: [Thu Jun 12 11:01:33 2025] <TASK> [Thu Jun 12 11:01:33 2025] __schedule+0x7c7/0x1930 [Thu Jun 12 11:01:33 2025] ? __pfx___schedule+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __mod_timer+0x304/0xa80 [Thu Jun 12 11:01:33 2025] ? irq_work_queue+0x6a/0xa0 [Thu Jun 12 11:01:33 2025] ? __pfx_vprintk_emit+0x10/0x10 [Thu Jun 12 11:01:33 2025] schedule+0x6a/0x180 [Thu Jun 12 11:01:33 2025] schedule_timeout+0xfb/0x230 [Thu Jun 12 11:01:33 2025] ? __pfx_schedule_timeout+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_process_timeout+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? down_write+0xc4/0x140 [Thu Jun 12 11:01:33 2025] msleep_interruptible+0xbe/0x150 [Thu Jun 12 11:01:33 2025] ? __pfx_thread1_func+0x10/0x10 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] thread1_func+0x37/0x60 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] kthread+0x39f/0x750 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx__raw_spin_lock_irq+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork+0x25d/0x320 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork_asm+0x1a/0x30 [Thu Jun 12 11:01:33 2025] </TASK>
--- include/linux/hung_task.h | 18 +++++++++--------- kernel/hung_task.c | 29 +++++++++++++++++++++++++---- kernel/locking/rwsem.c | 17 ++++++++++++++++- 3 files changed, 50 insertions(+), 14 deletions(-)
include/linux/hung_task.h +9 -9
@@ -21,17 +21,17 @@* type. * * Type encoding: -* 00 - Blocked on mutex (BLOCKER_TYPE_MUTEX) -* 01 - Blocked on semaphore (BLOCKER_TYPE_SEM) -* 10 - Blocked on rt-mutex (BLOCKER_TYPE_RTMUTEX) -* 11 - Blocked on rw-semaphore (BLOCKER_TYPE_RWSEM) +* 00 - Blocked on mutex (BLOCKER_TYPE_MUTEX) +* 01 - Blocked on semaphore (BLOCKER_TYPE_SEM) +* 10 - Blocked on rw-semaphore as READER (BLOCKER_TYPE_RWSEM_READER) +* 11 - Blocked on rw-semaphore as WRITER (BLOCKER_TYPE_RWSEM_WRITER) */ -#define BLOCKER_TYPE_MUTEX 0x00UL -#define BLOCKER_TYPE_SEM 0x01UL -#define BLOCKER_TYPE_RTMUTEX 0x02UL -#define BLOCKER_TYPE_RWSEM 0x03UL +#define BLOCKER_TYPE_MUTEX 0x00UL +#define BLOCKER_TYPE_SEM 0x01UL +#define BLOCKER_TYPE_RWSEM_READER 0x02UL +#define BLOCKER_TYPE_RWSEM_WRITER 0x03UL -#define BLOCKER_TYPE_MASK 0x03UL +#define BLOCKER_TYPE_MASK 0x03UL#ifdef CONFIG_DETECT_HUNG_TASK_BLOCKER static inline void hung_task_set_blocker (void *lock,unsigned long type)
kernel/hung_task.c +25 -4
@@ -23,6 +23,7 @@#include <linux/sched/debug.h> #include <linux/sched/sysctl.h> #include <linux/hung_task.h> +#include <linux/rwsem.h> #include <trace/events/sched.h>
@@ -100,6 +101,7 @@ static void debug_show_blocker(struct task_struct *task) {struct task_struct *g, *t;unsigned long owner, blocker, blocker_type; +const char *rwsem_blocked_by, *rwsem_blocked_as;RCU_LOCKDEP_WARN (!rcu_read_lock_held (),"No rcu lock held" );
@@ -111,12 +113,20 @@ static void debug_show_blocker(struct task_struct *task)switch (blocker_type) {case BLOCKER_TYPE_MUTEX : - owner =mutex_get_owner ( - (struct mutex *)hung_task_blocker_to_lock (blocker)); + owner =mutex_get_owner (hung_task_blocker_to_lock (blocker));break ;case BLOCKER_TYPE_SEM : - owner =sem_last_holder ( - (struct semaphore *)hung_task_blocker_to_lock (blocker)); + owner =sem_last_holder (hung_task_blocker_to_lock (blocker)); +break ; +case BLOCKER_TYPE_RWSEM_READER : +case BLOCKER_TYPE_RWSEM_WRITER : + owner = (unsigned long )rwsem_owner ( +hung_task_blocker_to_lock (blocker)); + rwsem_blocked_as = (blocker_type ==BLOCKER_TYPE_RWSEM_READER ) ? +"reader" :"writer" ; + rwsem_blocked_by =is_rwsem_reader_owned ( +hung_task_blocker_to_lock (blocker)) ? +"reader" :"writer" ;break ;default :WARN_ON_ONCE (1 );
@@ -134,6 +144,11 @@ static void debug_show_blocker(struct task_struct *task)pr_err ("INFO: task %s:%d is blocked on a semaphore, but the last holder is not found.\n" , task->comm, task->pid);break ; +case BLOCKER_TYPE_RWSEM_READER : +case BLOCKER_TYPE_RWSEM_WRITER : +pr_err ("INFO: task %s:%d is blocked on an rw-semaphore, but the owner is not found.\n" , + task->comm, task->pid); +break ; }return ; }
@@ -152,6 +167,12 @@ static void debug_show_blocker(struct task_struct *task)pr_err ("INFO: task %s:%d blocked on a semaphore likely last held by task %s:%d\n" , task->comm, task->pid, t->comm, t->pid);break ; +case BLOCKER_TYPE_RWSEM_READER : +case BLOCKER_TYPE_RWSEM_WRITER : +pr_err ("INFO: task %s:%d <%s> blocked on an rw-semaphore likely owned by task %s:%d <%s>\n" , + task->comm, task->pid, rwsem_blocked_as, t->comm, + t->pid, rwsem_blocked_by); +break ; }sched_show_task (t);return ;
kernel/locking/rwsem.c +16 -1
@@ -27,6 +27,7 @@#include <linux/export.h> #include <linux/rwsem.h> #include <linux/atomic.h> +#include <linux/hung_task.h> #include <trace/events/lock.h> #ifndef CONFIG_PREEMPT_RT
@@ -1065,10 +1066,13 @@ rwsem_down_read_slowpath(struct rw_semaphore *sem, long count, unsigned int statwake_up_q (&wake_q);trace_contention_begin (sem,LCB_F_READ ); +set_current_state (state); + +if (state ==TASK_UNINTERRUPTIBLE ) +hung_task_set_blocker (sem,BLOCKER_TYPE_RWSEM_READER );/* wait to be given the lock */ for (;;) { -set_current_state (state);if (!smp_load_acquire (&waiter.task)) {/* Matches rwsem_mark_wake()'s smp_store_release(). */ break ;
@@ -1083,8 +1087,12 @@ rwsem_down_read_slowpath(struct rw_semaphore *sem, long count, unsigned int stat }schedule_preempt_disabled ();lockevent_inc (rwsem_sleep_reader); +set_current_state (state); } +if (state ==TASK_UNINTERRUPTIBLE ) +hung_task_clear_blocker (); +__set_current_state (TASK_RUNNING );lockevent_inc (rwsem_rlock);trace_contention_end (sem,0 );
@@ -1146,6 +1154,9 @@ rwsem_down_write_slowpath(struct rw_semaphore *sem, int state)set_current_state (state);trace_contention_begin (sem,LCB_F_WRITE ); +if (state ==TASK_UNINTERRUPTIBLE ) +hung_task_set_blocker (sem,BLOCKER_TYPE_RWSEM_WRITER ); +for (;;) {if (rwsem_try_write_lock (sem, &waiter)) {/* rwsem_try_write_lock() implies ACQUIRE on success */
@@ -1179,6 +1190,10 @@ rwsem_down_write_slowpath(struct rw_semaphore *sem, int state) trylock_again:raw_spin_lock_irq (&sem->wait_lock); } + +if (state ==TASK_UNINTERRUPTIBLE ) +hung_task_clear_blocker (); +__set_current_state (TASK_RUNNING );raw_spin_unlock_irq (&sem->wait_lock);lockevent_inc (rwsem_wlock);
Signature · 1 line
2.49.0
Re: [PATCH RFC 3/3] hung_task: extend hung task blocker tracking to rwsems
Quoted 12 lines · On Thu, 12 Jun 2025 12:19:26 +0800 Lance Yang <[email protected]> wrote:
From: Lance Yang <[email protected]>
Inspired by mutex blocker tracking[1], and having already extended it to semaphores, let's now add support for reader-writer semaphores (rwsems).
The approach is simple: when a task enters
TASK_UNINTERRUPTIBLEwhile waiting for an rwsem, we just callhung_task_set_blocker(). The hung task detector can then query the rwsem's owner to identify the lock holder.Tracking works reliably for writers, as there can only be a single writer holding the lock, and its task struct is stored in the owner field.
Quoted 6 lines
The main challenge lies with readers. The owner field points to only one of many concurrent readers, so we might lose track of the blocker if that specific reader unlocks, even while others remain. This is not a significant issue, however. In practice, long-lasting lock contention is almost always caused by a writer. Therefore, reliably tracking the writer is the primary goal of this patch series ;)
I think as far as it is reliable, it is better than nothing :) and that can help us to debug some part of kernel crashes.
Quoted 60 lines
With this change, the hung task detector can now show blocker task's info like below:
[Thu Jun 12 11:01:33 2025] INFO: task rw_sem_thread2:36526 blocked for more than 122 seconds. [Thu Jun 12 11:01:33 2025] Tainted: G S O 6.16.0-rc1 #1 [Thu Jun 12 11:01:33 2025] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [Thu Jun 12 11:01:33 2025] task:rw_sem_thread2 state:D stack:0 pid:36526 tgid:36526 ppid:2 task_flags:0x208040 flags:0x00004000 [Thu Jun 12 11:01:33 2025] Call Trace: [Thu Jun 12 11:01:33 2025] <TASK> [Thu Jun 12 11:01:33 2025] __schedule+0x7c7/0x1930 [Thu Jun 12 11:01:33 2025] ? __pfx___schedule+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? _raw_spin_lock_irq+0x8a/0xe0 [Thu Jun 12 11:01:33 2025] ? __pfx__raw_spin_lock_irq+0x10/0x10 [Thu Jun 12 11:01:33 2025] schedule+0x6a/0x180 [Thu Jun 12 11:01:33 2025] schedule_preempt_disabled+0x15/0x30 [Thu Jun 12 11:01:33 2025] rwsem_down_write_slowpath+0x447/0x1090 [Thu Jun 12 11:01:33 2025] ? __pfx_rwsem_down_write_slowpath+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx___schedule+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx___might_resched+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_thread2_func+0x10/0x10 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] down_write+0x125/0x140 [Thu Jun 12 11:01:33 2025] ? __pfx_down_write+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? msleep+0x91/0xf0 [Thu Jun 12 11:01:33 2025] ? __raw_spin_lock_irqsave+0x8c/0xf0 [Thu Jun 12 11:01:33 2025] thread2_func+0x37/0x70 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] kthread+0x39f/0x750 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx__raw_spin_lock_irq+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork+0x25d/0x320 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork_asm+0x1a/0x30 [Thu Jun 12 11:01:33 2025] </TASK> [Thu Jun 12 11:01:33 2025] INFO: task rw_sem_thread2:36526 <writer> blocked on an rw-semaphore likely owned by task rw_sem_thread1:36525 <writer> [Thu Jun 12 11:01:33 2025] task:rw_sem_thread1 state:S stack:0 pid:36525 tgid:36525 ppid:2 task_flags:0x208040 flags:0x00004000 [Thu Jun 12 11:01:33 2025] Call Trace: [Thu Jun 12 11:01:33 2025] <TASK> [Thu Jun 12 11:01:33 2025] __schedule+0x7c7/0x1930 [Thu Jun 12 11:01:33 2025] ? __pfx___schedule+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __mod_timer+0x304/0xa80 [Thu Jun 12 11:01:33 2025] ? irq_work_queue+0x6a/0xa0 [Thu Jun 12 11:01:33 2025] ? __pfx_vprintk_emit+0x10/0x10 [Thu Jun 12 11:01:33 2025] schedule+0x6a/0x180 [Thu Jun 12 11:01:33 2025] schedule_timeout+0xfb/0x230 [Thu Jun 12 11:01:33 2025] ? __pfx_schedule_timeout+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_process_timeout+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? down_write+0xc4/0x140 [Thu Jun 12 11:01:33 2025] msleep_interruptible+0xbe/0x150 [Thu Jun 12 11:01:33 2025] ? __pfx_thread1_func+0x10/0x10 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] thread1_func+0x37/0x60 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] kthread+0x39f/0x750 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx__raw_spin_lock_irq+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork+0x25d/0x320 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork_asm+0x1a/0x30 [Thu Jun 12 11:01:33 2025] </TASK>
Quoted 4 lines
Suggested-by Masami Hiramatsu (Google) <[email protected]>Signed-off-by Lance Yang <[email protected]>
Looks good to me.
BTW, can you also add a patch to extend the test module (samples/hung_task/hung_task_tests.c) to handle the rwsem blocking case ? (reader and writer side)
Thank you,
Quoted 171 lines
--- include/linux/hung_task.h | 18 +++++++++--------- kernel/hung_task.c | 29 +++++++++++++++++++++++++---- kernel/locking/rwsem.c | 17 ++++++++++++++++- 3 files changed, 50 insertions(+), 14 deletions(-)include/linux/hung_task.h +9
-9@@ -21,17 +21,17 @@* type. * * Type encoding: -* 00 - Blocked on mutex (BLOCKER_TYPE_MUTEX) -* 01 - Blocked on semaphore (BLOCKER_TYPE_SEM) -* 10 - Blocked on rt-mutex (BLOCKER_TYPE_RTMUTEX) -* 11 - Blocked on rw-semaphore (BLOCKER_TYPE_RWSEM) +* 00 - Blocked on mutex (BLOCKER_TYPE_MUTEX) +* 01 - Blocked on semaphore (BLOCKER_TYPE_SEM) +* 10 - Blocked on rw-semaphore as READER (BLOCKER_TYPE_RWSEM_READER) +* 11 - Blocked on rw-semaphore as WRITER (BLOCKER_TYPE_RWSEM_WRITER) */ -#define BLOCKER_TYPE_MUTEX 0x00UL -#define BLOCKER_TYPE_SEM 0x01UL -#define BLOCKER_TYPE_RTMUTEX 0x02UL -#define BLOCKER_TYPE_RWSEM 0x03UL +#define BLOCKER_TYPE_MUTEX 0x00UL +#define BLOCKER_TYPE_SEM 0x01UL +#define BLOCKER_TYPE_RWSEM_READER 0x02UL +#define BLOCKER_TYPE_RWSEM_WRITER 0x03UL -#define BLOCKER_TYPE_MASK 0x03UL +#define BLOCKER_TYPE_MASK 0x03UL#ifdef CONFIG_DETECT_HUNG_TASK_BLOCKER static inline void hung_task_set_blocker (void *lock,unsigned long type)kernel/hung_task.c +25
-4@@ -23,6 +23,7 @@#include <linux/sched/debug.h> #include <linux/sched/sysctl.h> #include <linux/hung_task.h> +#include <linux/rwsem.h> #include <trace/events/sched.h> @@ -100,6 +101,7 @@ static void debug_show_blocker(struct task_struct *task) {struct task_struct *g, *t;unsigned long owner, blocker, blocker_type; +const char *rwsem_blocked_by, *rwsem_blocked_as;RCU_LOCKDEP_WARN (!rcu_read_lock_held (),"No rcu lock held" );@@ -111,12 +113,20 @@ static void debug_show_blocker(struct task_struct *task)switch (blocker_type) {case BLOCKER_TYPE_MUTEX : - owner =mutex_get_owner ( - (struct mutex *)hung_task_blocker_to_lock (blocker)); + owner =mutex_get_owner (hung_task_blocker_to_lock (blocker));break ;case BLOCKER_TYPE_SEM : - owner =sem_last_holder ( - (struct semaphore *)hung_task_blocker_to_lock (blocker)); + owner =sem_last_holder (hung_task_blocker_to_lock (blocker)); +break ; +case BLOCKER_TYPE_RWSEM_READER : +case BLOCKER_TYPE_RWSEM_WRITER : + owner = (unsigned long )rwsem_owner ( +hung_task_blocker_to_lock (blocker)); + rwsem_blocked_as = (blocker_type ==BLOCKER_TYPE_RWSEM_READER ) ? +"reader" :"writer" ; + rwsem_blocked_by =is_rwsem_reader_owned ( +hung_task_blocker_to_lock (blocker)) ? +"reader" :"writer" ;break ;default :WARN_ON_ONCE (1 );@@ -134,6 +144,11 @@ static void debug_show_blocker(struct task_struct *task)pr_err ("INFO: task %s:%d is blocked on a semaphore, but the last holder is not found.\n" , task->comm, task->pid);break ; +case BLOCKER_TYPE_RWSEM_READER : +case BLOCKER_TYPE_RWSEM_WRITER : +pr_err ("INFO: task %s:%d is blocked on an rw-semaphore, but the owner is not found.\n" , + task->comm, task->pid); +break ; }return ; }@@ -152,6 +167,12 @@ static void debug_show_blocker(struct task_struct *task)pr_err ("INFO: task %s:%d blocked on a semaphore likely last held by task %s:%d\n" , task->comm, task->pid, t->comm, t->pid);break ; +case BLOCKER_TYPE_RWSEM_READER : +case BLOCKER_TYPE_RWSEM_WRITER : +pr_err ("INFO: task %s:%d <%s> blocked on an rw-semaphore likely owned by task %s:%d <%s>\n" , + task->comm, task->pid, rwsem_blocked_as, t->comm, + t->pid, rwsem_blocked_by); +break ; }sched_show_task (t);return ;kernel/locking/rwsem.c +16
-1@@ -27,6 +27,7 @@#include <linux/export.h> #include <linux/rwsem.h> #include <linux/atomic.h> +#include <linux/hung_task.h> #include <trace/events/lock.h> #ifndef CONFIG_PREEMPT_RT @@ -1065,10 +1066,13 @@ rwsem_down_read_slowpath(struct rw_semaphore *sem, long count, unsigned int statwake_up_q (&wake_q);trace_contention_begin (sem,LCB_F_READ ); +set_current_state (state); + +if (state ==TASK_UNINTERRUPTIBLE ) +hung_task_set_blocker (sem,BLOCKER_TYPE_RWSEM_READER );/* wait to be given the lock */ for (;;) { -set_current_state (state);if (!smp_load_acquire (&waiter.task)) {/* Matches rwsem_mark_wake()'s smp_store_release(). */ break ;@@ -1083,8 +1087,12 @@ rwsem_down_read_slowpath(struct rw_semaphore *sem, long count, unsigned int stat }schedule_preempt_disabled ();lockevent_inc (rwsem_sleep_reader); +set_current_state (state); } +if (state ==TASK_UNINTERRUPTIBLE ) +hung_task_clear_blocker (); +__set_current_state (TASK_RUNNING );lockevent_inc (rwsem_rlock);trace_contention_end (sem,0 );@@ -1146,6 +1154,9 @@ rwsem_down_write_slowpath(struct rw_semaphore *sem, int state)set_current_state (state);trace_contention_begin (sem,LCB_F_WRITE ); +if (state ==TASK_UNINTERRUPTIBLE ) +hung_task_set_blocker (sem,BLOCKER_TYPE_RWSEM_WRITER ); +for (;;) {if (rwsem_try_write_lock (sem, &waiter)) {/* rwsem_try_write_lock() implies ACQUIRE on success */ @@ -1179,6 +1190,10 @@ rwsem_down_write_slowpath(struct rw_semaphore *sem, int state) trylock_again:raw_spin_lock_irq (&sem->wait_lock); } + +if (state ==TASK_UNINTERRUPTIBLE ) +hung_task_clear_blocker (); +__set_current_state (TASK_RUNNING );raw_spin_unlock_irq (&sem->wait_lock);lockevent_inc (rwsem_wlock);Signature · 1 line
2.49.0
Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
Lance Yang
maintainerRe: [PATCH RFC 3/3] hung_task: extend hung task blocker tracking to rwsems
Quoted 22 lines · Masami Hiramatsu (Google)
Quoted 12 lines · On Thu, 12 Jun 2025 12:19:26 +0800 Lance Yang <[email protected]> wrote:
From: Lance Yang <[email protected]>
Inspired by mutex blocker tracking[1], and having already extended it to semaphores, let's now add support for reader-writer semaphores (rwsems).
The approach is simple: when a task enters
TASK_UNINTERRUPTIBLEwhile waiting for an rwsem, we just callhung_task_set_blocker(). The hung task detector can then query the rwsem's owner to identify the lock holder.Tracking works reliably for writers, as there can only be a single writer holding the lock, and its task struct is stored in the owner field.
Quoted 6 lines
The main challenge lies with readers. The owner field points to only one of many concurrent readers, so we might lose track of the blocker if that specific reader unlocks, even while others remain. This is not a significant issue, however. In practice, long-lasting lock contention is almost always caused by a writer. Therefore, reliably tracking the writer is the primary goal of this patch series ;)
Quoted 2 lines
I think as far as it is reliable, it is better than nothing :) and that can help us to debug some part of kernel crashes.
Yeah! That was the goal. Even if it's not a perfect solution for all reader scenarios, it provides much-needed visibility for writes and is a good improvement ;)
Quoted 66 lines
Quoted 61 lines
With this change, the hung task detector can now show blocker task's info like below:
[Thu Jun 12 11:01:33 2025] INFO: task rw_sem_thread2:36526 blocked for more than 122 seconds. [Thu Jun 12 11:01:33 2025] Tainted: G S O 6.16.0-rc1 #1 [Thu Jun 12 11:01:33 2025] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [Thu Jun 12 11:01:33 2025] task:rw_sem_thread2 state:D stack:0 pid:36526 tgid:36526 ppid:2 task_flags:0x208040 flags:0x00004000 [Thu Jun 12 11:01:33 2025] Call Trace: [Thu Jun 12 11:01:33 2025] <TASK> [Thu Jun 12 11:01:33 2025] __schedule+0x7c7/0x1930 [Thu Jun 12 11:01:33 2025] ? __pfx___schedule+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? _raw_spin_lock_irq+0x8a/0xe0 [Thu Jun 12 11:01:33 2025] ? __pfx__raw_spin_lock_irq+0x10/0x10 [Thu Jun 12 11:01:33 2025] schedule+0x6a/0x180 [Thu Jun 12 11:01:33 2025] schedule_preempt_disabled+0x15/0x30 [Thu Jun 12 11:01:33 2025] rwsem_down_write_slowpath+0x447/0x1090 [Thu Jun 12 11:01:33 2025] ? __pfx_rwsem_down_write_slowpath+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx___schedule+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx___might_resched+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_thread2_func+0x10/0x10 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] down_write+0x125/0x140 [Thu Jun 12 11:01:33 2025] ? __pfx_down_write+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? msleep+0x91/0xf0 [Thu Jun 12 11:01:33 2025] ? __raw_spin_lock_irqsave+0x8c/0xf0 [Thu Jun 12 11:01:33 2025] thread2_func+0x37/0x70 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] kthread+0x39f/0x750 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx__raw_spin_lock_irq+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork+0x25d/0x320 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork_asm+0x1a/0x30 [Thu Jun 12 11:01:33 2025] </TASK> [Thu Jun 12 11:01:33 2025] INFO: task rw_sem_thread2:36526 <writer> blocked on an rw-semaphore likely owned by task rw_sem_thread1:36525 <writer> [Thu Jun 12 11:01:33 2025] task:rw_sem_thread1 state:S stack:0 pid:36525 tgid:36525 ppid:2 task_flags:0x208040 flags:0x00004000 [Thu Jun 12 11:01:33 2025] Call Trace: [Thu Jun 12 11:01:33 2025] <TASK> [Thu Jun 12 11:01:33 2025] __schedule+0x7c7/0x1930 [Thu Jun 12 11:01:33 2025] ? __pfx___schedule+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __mod_timer+0x304/0xa80 [Thu Jun 12 11:01:33 2025] ? irq_work_queue+0x6a/0xa0 [Thu Jun 12 11:01:33 2025] ? __pfx_vprintk_emit+0x10/0x10 [Thu Jun 12 11:01:33 2025] schedule+0x6a/0x180 [Thu Jun 12 11:01:33 2025] schedule_timeout+0xfb/0x230 [Thu Jun 12 11:01:33 2025] ? __pfx_schedule_timeout+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_process_timeout+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? down_write+0xc4/0x140 [Thu Jun 12 11:01:33 2025] msleep_interruptible+0xbe/0x150 [Thu Jun 12 11:01:33 2025] ? __pfx_thread1_func+0x10/0x10 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] thread1_func+0x37/0x60 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] kthread+0x39f/0x750 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx__raw_spin_lock_irq+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork+0x25d/0x320 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork_asm+0x1a/0x30 [Thu Jun 12 11:01:33 2025] </TASK>Quoted 4 lines
Suggested-by Masami Hiramatsu (Google) <[email protected]>Signed-off-by Lance Yang <[email protected]>
Quoted 7 lines
Looks good to me.
Reviewed-by Masami Hiramatsu (Google) <[email protected]>BTW, can you also add a patch to extend the test module (
samples/hung_task/hung_task_tests.c) to handle the rwsem blocking case ? (reader and writer side)
Thanks! Will do.
Lance
Quoted 177 lines
Thank you,
Quoted 171 lines
--- include/linux/hung_task.h | 18 +++++++++--------- kernel/hung_task.c | 29 +++++++++++++++++++++++++---- kernel/locking/rwsem.c | 17 ++++++++++++++++- 3 files changed, 50 insertions(+), 14 deletions(-)include/linux/hung_task.h +9
-9@@ -21,17 +21,17 @@* type. * * Type encoding: -* 00 - Blocked on mutex (BLOCKER_TYPE_MUTEX) -* 01 - Blocked on semaphore (BLOCKER_TYPE_SEM) -* 10 - Blocked on rt-mutex (BLOCKER_TYPE_RTMUTEX) -* 11 - Blocked on rw-semaphore (BLOCKER_TYPE_RWSEM) +* 00 - Blocked on mutex (BLOCKER_TYPE_MUTEX) +* 01 - Blocked on semaphore (BLOCKER_TYPE_SEM) +* 10 - Blocked on rw-semaphore as READER (BLOCKER_TYPE_RWSEM_READER) +* 11 - Blocked on rw-semaphore as WRITER (BLOCKER_TYPE_RWSEM_WRITER) */ -#define BLOCKER_TYPE_MUTEX 0x00UL -#define BLOCKER_TYPE_SEM 0x01UL -#define BLOCKER_TYPE_RTMUTEX 0x02UL -#define BLOCKER_TYPE_RWSEM 0x03UL +#define BLOCKER_TYPE_MUTEX 0x00UL +#define BLOCKER_TYPE_SEM 0x01UL +#define BLOCKER_TYPE_RWSEM_READER 0x02UL +#define BLOCKER_TYPE_RWSEM_WRITER 0x03UL -#define BLOCKER_TYPE_MASK 0x03UL +#define BLOCKER_TYPE_MASK 0x03UL#ifdef CONFIG_DETECT_HUNG_TASK_BLOCKER static inline void hung_task_set_blocker (void *lock,unsigned long type)kernel/hung_task.c +25
-4@@ -23,6 +23,7 @@#include <linux/sched/debug.h> #include <linux/sched/sysctl.h> #include <linux/hung_task.h> +#include <linux/rwsem.h> #include <trace/events/sched.h> @@ -100,6 +101,7 @@ static void debug_show_blocker(struct task_struct *task) {struct task_struct *g, *t;unsigned long owner, blocker, blocker_type; +const char *rwsem_blocked_by, *rwsem_blocked_as;RCU_LOCKDEP_WARN (!rcu_read_lock_held (),"No rcu lock held" );@@ -111,12 +113,20 @@ static void debug_show_blocker(struct task_struct *task)switch (blocker_type) {case BLOCKER_TYPE_MUTEX : - owner =mutex_get_owner ( - (struct mutex *)hung_task_blocker_to_lock (blocker)); + owner =mutex_get_owner (hung_task_blocker_to_lock (blocker));break ;case BLOCKER_TYPE_SEM : - owner =sem_last_holder ( - (struct semaphore *)hung_task_blocker_to_lock (blocker)); + owner =sem_last_holder (hung_task_blocker_to_lock (blocker)); +break ; +case BLOCKER_TYPE_RWSEM_READER : +case BLOCKER_TYPE_RWSEM_WRITER : + owner = (unsigned long )rwsem_owner ( +hung_task_blocker_to_lock (blocker)); + rwsem_blocked_as = (blocker_type ==BLOCKER_TYPE_RWSEM_READER ) ? +"reader" :"writer" ; + rwsem_blocked_by =is_rwsem_reader_owned ( +hung_task_blocker_to_lock (blocker)) ? +"reader" :"writer" ;break ;default :WARN_ON_ONCE (1 );@@ -134,6 +144,11 @@ static void debug_show_blocker(struct task_struct *task)pr_err ("INFO: task %s:%d is blocked on a semaphore, but the last holder is not found.\n" , task->comm, task->pid);break ; +case BLOCKER_TYPE_RWSEM_READER : +case BLOCKER_TYPE_RWSEM_WRITER : +pr_err ("INFO: task %s:%d is blocked on an rw-semaphore, but the owner is not found.\n" , + task->comm, task->pid); +break ; }return ; }@@ -152,6 +167,12 @@ static void debug_show_blocker(struct task_struct *task)pr_err ("INFO: task %s:%d blocked on a semaphore likely last held by task %s:%d\n" , task->comm, task->pid, t->comm, t->pid);break ; +case BLOCKER_TYPE_RWSEM_READER : +case BLOCKER_TYPE_RWSEM_WRITER : +pr_err ("INFO: task %s:%d <%s> blocked on an rw-semaphore likely owned by task %s:%d <%s>\n" , + task->comm, task->pid, rwsem_blocked_as, t->comm, + t->pid, rwsem_blocked_by); +break ; }sched_show_task (t);return ;kernel/locking/rwsem.c +16
-1@@ -27,6 +27,7 @@#include <linux/export.h> #include <linux/rwsem.h> #include <linux/atomic.h> +#include <linux/hung_task.h> #include <trace/events/lock.h> #ifndef CONFIG_PREEMPT_RT @@ -1065,10 +1066,13 @@ rwsem_down_read_slowpath(struct rw_semaphore *sem, long count, unsigned int statwake_up_q (&wake_q);trace_contention_begin (sem,LCB_F_READ ); +set_current_state (state); + +if (state ==TASK_UNINTERRUPTIBLE ) +hung_task_set_blocker (sem,BLOCKER_TYPE_RWSEM_READER );/* wait to be given the lock */ for (;;) { -set_current_state (state);if (!smp_load_acquire (&waiter.task)) {/* Matches rwsem_mark_wake()'s smp_store_release(). */ break ;@@ -1083,8 +1087,12 @@ rwsem_down_read_slowpath(struct rw_semaphore *sem, long count, unsigned int stat }schedule_preempt_disabled ();lockevent_inc (rwsem_sleep_reader); +set_current_state (state); } +if (state ==TASK_UNINTERRUPTIBLE ) +hung_task_clear_blocker (); +__set_current_state (TASK_RUNNING );lockevent_inc (rwsem_rlock);trace_contention_end (sem,0 );@@ -1146,6 +1154,9 @@ rwsem_down_write_slowpath(struct rw_semaphore *sem, int state)set_current_state (state);trace_contention_begin (sem,LCB_F_WRITE ); +if (state ==TASK_UNINTERRUPTIBLE ) +hung_task_set_blocker (sem,BLOCKER_TYPE_RWSEM_WRITER ); +for (;;) {if (rwsem_try_write_lock (sem, &waiter)) {/* rwsem_try_write_lock() implies ACQUIRE on success */ @@ -1179,6 +1190,10 @@ rwsem_down_write_slowpath(struct rw_semaphore *sem, int state) trylock_again:raw_spin_lock_irq (&sem->wait_lock); } + +if (state ==TASK_UNINTERRUPTIBLE ) +hung_task_clear_blocker (); +__set_current_state (TASK_RUNNING );raw_spin_unlock_irq (&sem->wait_lock);lockevent_inc (rwsem_wlock);Signature · 1 line
2.49.0
Gently ping ;p
Quoted 98 lines · Lance Yang
Hi all,
Inspired by mutex blocker tracking[1], and having already extended it to semaphores, let's now add support for reader-writer semaphores (rwsems).
The approach is simple: when a task enters
TASK_UNINTERRUPTIBLEwhile waiting for an rwsem, we just callhung_task_set_blocker(). The hung task detector can then query the rwsem's owner to identify the lock holder.Tracking works reliably for writers, as there can only be a single writer holding the lock, and its task struct is stored in the owner field.
The main challenge lies with readers. The owner field points to only one of many concurrent readers, so we might lose track of the blocker if that specific reader unlocks, even while others remain. This is not a significant issue, however. In practice, long-lasting lock contention is almost always caused by a writer. Therefore, reliably tracking the writer is the primary goal of this patch series ;)
With this change, the hung task detector can now show blocker task's info like below:
[Thu Jun 12 11:01:33 2025] INFO: task rw_sem_thread2:36526 blocked for more than 122 seconds. [Thu Jun 12 11:01:33 2025] Tainted: G S O 6.16.0-rc1 #1 [Thu Jun 12 11:01:33 2025] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [Thu Jun 12 11:01:33 2025] task:rw_sem_thread2 state:D stack:0 pid:36526 tgid:36526 ppid:2 task_flags:0x208040 flags:0x00004000 [Thu Jun 12 11:01:33 2025] Call Trace: [Thu Jun 12 11:01:33 2025] <TASK> [Thu Jun 12 11:01:33 2025] __schedule+0x7c7/0x1930 [Thu Jun 12 11:01:33 2025] ? __pfx___schedule+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? _raw_spin_lock_irq+0x8a/0xe0 [Thu Jun 12 11:01:33 2025] ? __pfx__raw_spin_lock_irq+0x10/0x10 [Thu Jun 12 11:01:33 2025] schedule+0x6a/0x180 [Thu Jun 12 11:01:33 2025] schedule_preempt_disabled+0x15/0x30 [Thu Jun 12 11:01:33 2025] rwsem_down_write_slowpath+0x447/0x1090 [Thu Jun 12 11:01:33 2025] ? __pfx_rwsem_down_write_slowpath+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx___schedule+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx___might_resched+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_thread2_func+0x10/0x10 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] down_write+0x125/0x140 [Thu Jun 12 11:01:33 2025] ? __pfx_down_write+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? msleep+0x91/0xf0 [Thu Jun 12 11:01:33 2025] ? __raw_spin_lock_irqsave+0x8c/0xf0 [Thu Jun 12 11:01:33 2025] thread2_func+0x37/0x70 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] kthread+0x39f/0x750 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx__raw_spin_lock_irq+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork+0x25d/0x320 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork_asm+0x1a/0x30 [Thu Jun 12 11:01:33 2025] </TASK> [Thu Jun 12 11:01:33 2025] INFO: task rw_sem_thread2:36526 <writer> blocked on an rw-semaphore likely owned by task rw_sem_thread1:36525 <writer> [Thu Jun 12 11:01:33 2025] task:rw_sem_thread1 state:S stack:0 pid:36525 tgid:36525 ppid:2 task_flags:0x208040 flags:0x00004000 [Thu Jun 12 11:01:33 2025] Call Trace: [Thu Jun 12 11:01:33 2025] <TASK> [Thu Jun 12 11:01:33 2025] __schedule+0x7c7/0x1930 [Thu Jun 12 11:01:33 2025] ? __pfx___schedule+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __mod_timer+0x304/0xa80 [Thu Jun 12 11:01:33 2025] ? irq_work_queue+0x6a/0xa0 [Thu Jun 12 11:01:33 2025] ? __pfx_vprintk_emit+0x10/0x10 [Thu Jun 12 11:01:33 2025] schedule+0x6a/0x180 [Thu Jun 12 11:01:33 2025] schedule_timeout+0xfb/0x230 [Thu Jun 12 11:01:33 2025] ? __pfx_schedule_timeout+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_process_timeout+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? down_write+0xc4/0x140 [Thu Jun 12 11:01:33 2025] msleep_interruptible+0xbe/0x150 [Thu Jun 12 11:01:33 2025] ? __pfx_thread1_func+0x10/0x10 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] thread1_func+0x37/0x60 [rw_sem_test_2] [Thu Jun 12 11:01:33 2025] kthread+0x39f/0x750 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx__raw_spin_lock_irq+0x10/0x10 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork+0x25d/0x320 [Thu Jun 12 11:01:33 2025] ? __pfx_kthread+0x10/0x10 [Thu Jun 12 11:01:33 2025] ret_from_fork_asm+0x1a/0x30 [Thu Jun 12 11:01:33 2025] </TASK>Thanks,
LanceLance Yang (3):
locking/rwsem: make owner helpers globally available locking/rwsem: clear reader-owner on unlock to reduce false positives
hung_task: extend hung task blocker tracking to rwsemsinclude/linux/hung_task.h | 18 +++++++++--------- include/linux/rwsem.h | 12 ++++++++++++ kernel/hung_task.c | 29 +++++++++++++++++++++++++---- kernel/locking/rwsem.c | 31 +++++++++++++++++++++++-------- 4 files changed, 69 insertions(+), 21 deletions(-)--
2.49.0