Skip to contentSkip to conversation

Fix boot regression for s390 and remove break_lock

PATCH In review Will Deacon 4 participants 8 messages Nov 28, 2017 – Dec 12, 2017 Locking primitives

lore
Share & tools
lore
#0 lore raw

cover Fix boot regression for s390 and remove break_lock

Hi all,

The following two patches do the following:

1. Fix boot breakage reported on s390 caused by a8a217c22116
2. Kill off the break_lock field entirely, since it's not actually that useful

I didn't go the whole hog and remove CONFIG_GENERIC_LOCKBREAK, since the "do something different with locks in preemptible kernels" isn't necessarily a bad idea, and PowerPC actually implements things like arch_spin_relax for that.

The first patch should go in to 4.15.

Thanks,

Will

--->8

Will Deacon (2):
locking/core: Fix deadlock during boot on systems with
GENERIC_LOCKBREAK
locking/core: Remove break_lock field when CONFIG_GENERIC_LOCKBREAK=y

 include/linux/rwlock_types.h   |  3 ---
 include/linux/spinlock.h       |  5 -----
 include/linux/spinlock_types.h |  3 ---
 kernel/locking/spinlock.c      | 13 +++----------
 4 files changed, 3 insertions(+), 21 deletions(-)
Signature · 1 line
2.1.4
#1 lore raw

PATCH 1/2 locking/core: Fix deadlock during boot on systems with GENERIC_LOCKBREAK

Commit a8a217c22116 ("locking/core: Remove {read,spin,write}_can_lock()") removed the definition of raw_spin_can_lock, causing the GENERIC_LOCKBREAK spin_lock routines to poll the break_lock field when waiting on a lock.

This has been reported to cause a deadlock during boot on s390, because the break_lock field is also set by the waiters, and can potentially remain set indefinitely if no other CPUs come in to take the lock after it has been released.

This patch removes the explicit spinning on break_lock from the waiters, instead relying on the outer trylock operation to determine when the lock is available.

Cc Martin Schwidefsky <[email protected]>
Cc Heiko Carstens <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Ingo Molnar <[email protected]>
Fixes a8a217c22116 ("locking/core: Remove {read,spin,write}_can_lock()")
---
 kernel/locking/spinlock.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
kernel/locking/spinlock.c +4 -4
index 1fd1a7543cdd..0ebb253e2199 100644
@@ -68,8 +68,8 @@ void __lockfunc __raw_##op##_lock(locktype##_t *lock)			\
 									\
 		if (!(lock)->break_lock)				\
 			(lock)->break_lock = 1;				\
-		while ((lock)->break_lock)				\
-			arch_##op##_relax(&lock->raw_lock);		\
+									\
+		arch_##op##_relax(&lock->raw_lock);			\
 	}								\
 	(lock)->break_lock = 0;						\
 }									\
@@ -88,8 +88,8 @@ unsigned long __lockfunc __raw_##op##_lock_irqsave(locktype##_t *lock)	\
 									\
 		if (!(lock)->break_lock)				\
 			(lock)->break_lock = 1;				\
-		while ((lock)->break_lock)				\
-			arch_##op##_relax(&lock->raw_lock);		\
+									\
+		arch_##op##_relax(&lock->raw_lock);			\
 	}								\
 	(lock)->break_lock = 0;						\
 	return flags;							\
Signature · 1 line
2.1.4
#2 lore raw

[tip:locking/urgent] locking/core: Fix deadlock during boot on systems with GENERIC_LOCKBREAK

locking/core: Fix deadlock during boot on systems with GENERIC_LOCKBREAK
Commit-ID:  f87f3a328dbbb3e79dd53e7e889ced9222512649
Gitweb:     https://git.kernel.org/tip/f87f3a328dbbb3e79dd53e7e889ced9222512649
Author:     Will Deacon <[email protected]>
AuthorDate: Tue, 28 Nov 2017 18:42:18 +0000
Committer:  Ingo Molnar <[email protected]>
CommitDate: Tue, 12 Dec 2017 11:24:01 +0100

locking/core: Fix deadlock during boot on systems with GENERIC_LOCKBREAK

Commit:

a8a217c22116 ("locking/core: Remove {read,spin,write}_can_lock()")

removed the definition of raw_spin_can_lock(), causing the GENERIC_LOCKBREAK spin_lock() routines to poll the ->break_lock field when waiting on a lock.

This has been reported to cause a deadlock during boot on s390, because the ->break_lock field is also set by the waiters, and can potentially remain set indefinitely if no other CPUs come in to take the lock after it has been released.

This patch removes the explicit spinning on ->break_lock from the waiters, instead relying on the outer trylock() operation to determine when the lock is available.

Cc Heiko Carstens <[email protected]>
Cc Linus Torvalds <[email protected]>
Cc Martin Schwidefsky <[email protected]>
Cc Thomas Gleixner <[email protected]>
Fixes a8a217c22116 ("locking/core: Remove {read,spin,write}_can_lock()")
---
 kernel/locking/spinlock.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
kernel/locking/spinlock.c +4 -4
index 1fd1a75..0ebb253 100644
@@ -68,8 +68,8 @@ void __lockfunc __raw_##op##_lock(locktype##_t *lock)			\
 									\
 		if (!(lock)->break_lock)				\
 			(lock)->break_lock = 1;				\
-		while ((lock)->break_lock)				\
-			arch_##op##_relax(&lock->raw_lock);		\
+									\
+		arch_##op##_relax(&lock->raw_lock);			\
 	}								\
 	(lock)->break_lock = 0;						\
 }									\
@@ -88,8 +88,8 @@ unsigned long __lockfunc __raw_##op##_lock_irqsave(locktype##_t *lock)	\
 									\
 		if (!(lock)->break_lock)				\
 			(lock)->break_lock = 1;				\
-		while ((lock)->break_lock)				\
-			arch_##op##_relax(&lock->raw_lock);		\
+									\
+		arch_##op##_relax(&lock->raw_lock);			\
 	}								\
 	(lock)->break_lock = 0;						\
 	return flags;							\
#3 lore raw

PATCH 2/2 locking/core: Remove break_lock field when CONFIG_GENERIC_LOCKBREAK=y

In reply to Will Deacon

When CONFIG_GENERIC_LOCKBEAK=y, locking structures grow an extra int break_lock field which is used to implement raw_spin_is_contended() by setting the field to 1 when waiting on a lock and clearing it to zero when holding a lock. However, there are a few problems with this approach:

- There is a write-write race between a CPU successfully taking the lock (and subsequently writing break_lock = 0) and a waiter waiting on the lock (and subsequently writing break_lock = 1). This could result in a contended lock being reported as uncontended and vice-versa.

- On machines with store buffers, nothing guarantees that the writes to break_lock are visible to other CPUs at any particular time.

- READ_ONCE/WRITE_ONCE are not used, so the field is potentially susceptible to harmful compiler optimisations,

Consequently, the usefulness of this field is unclear and we'd be better off removing it and allowing architectures to implement raw_spin_is_contended() by providing a definition of arch_spin_is_contended(), as they can when CONFIG_GENERIC_LOCKBREAK=n.

Cc Martin Schwidefsky <[email protected]>
Cc Heiko Carstens <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Sebastian Ott <[email protected]>
---
 include/linux/rwlock_types.h   | 3 ---
 include/linux/spinlock.h       | 5 -----
 include/linux/spinlock_types.h | 3 ---
 kernel/locking/spinlock.c      | 9 +--------
 4 files changed, 1 insertion(+), 19 deletions(-)
include/linux/rwlock_types.h +0 -3
index cc0072e93e36..857a72ceb794 100644
@@ -10,9 +10,6 @@
  */
 typedef struct {
 	arch_rwlock_t raw_lock;
-#ifdef CONFIG_GENERIC_LOCKBREAK
-	unsigned int break_lock;
-#endif
 #ifdef CONFIG_DEBUG_SPINLOCK
 	unsigned int magic, owner_cpu;
 	void *owner;
include/linux/spinlock.h +0 -5
index a39186194cd6..3bf273538840 100644
@@ -107,16 +107,11 @@ do {								\
 
 #define raw_spin_is_locked(lock)	arch_spin_is_locked(&(lock)->raw_lock)
 
-#ifdef CONFIG_GENERIC_LOCKBREAK
-#define raw_spin_is_contended(lock) ((lock)->break_lock)
-#else
-
 #ifdef arch_spin_is_contended
 #define raw_spin_is_contended(lock)	arch_spin_is_contended(&(lock)->raw_lock)
 #else
 #define raw_spin_is_contended(lock)	(((void)(lock), 0))
 #endif /*arch_spin_is_contended*/
-#endif
 
 /*
  * This barrier must provide two things:
include/linux/spinlock_types.h +0 -3
index 73548eb13a5d..24b4e6f2c1a2 100644
@@ -19,9 +19,6 @@
 
 typedef struct raw_spinlock {
 	arch_spinlock_t raw_lock;
-#ifdef CONFIG_GENERIC_LOCKBREAK
-	unsigned int break_lock;
-#endif
 #ifdef CONFIG_DEBUG_SPINLOCK
 	unsigned int magic, owner_cpu;
 	void *owner;
kernel/locking/spinlock.c +1 -8
index 0ebb253e2199..936f3d14dd6b 100644
@@ -66,12 +66,8 @@ void __lockfunc __raw_##op##_lock(locktype##_t *lock)			\
 			break;						\
 		preempt_enable();					\
 									\
-		if (!(lock)->break_lock)				\
-			(lock)->break_lock = 1;				\
-									\
 		arch_##op##_relax(&lock->raw_lock);			\
 	}								\
-	(lock)->break_lock = 0;						\
 }									\
 									\
 unsigned long __lockfunc __raw_##op##_lock_irqsave(locktype##_t *lock)	\
@@ -86,12 +82,9 @@ unsigned long __lockfunc __raw_##op##_lock_irqsave(locktype##_t *lock)	\
 		local_irq_restore(flags);				\
 		preempt_enable();					\
 									\
-		if (!(lock)->break_lock)				\
-			(lock)->break_lock = 1;				\
-									\
 		arch_##op##_relax(&lock->raw_lock);			\
 	}								\
-	(lock)->break_lock = 0;						\
+									\
 	return flags;							\
 }									\
 									\
Signature · 1 line
2.1.4
#4 lore raw

[tip:locking/urgent] locking/core: Remove break_lock field when CONFIG_GENERIC_LOCKBREAK=y

locking/core: Remove break_lock field when CONFIG_GENERIC_LOCKBREAK=y
Commit-ID:  d89c70356acf11b7cf47ca5cfcafae5062a85451
Gitweb:     https://git.kernel.org/tip/d89c70356acf11b7cf47ca5cfcafae5062a85451
Author:     Will Deacon <[email protected]>
AuthorDate: Tue, 28 Nov 2017 18:42:19 +0000
Committer:  Ingo Molnar <[email protected]>
CommitDate: Tue, 12 Dec 2017 11:24:01 +0100

locking/core: Remove break_lock field when CONFIG_GENERIC_LOCKBREAK=y

When CONFIG_GENERIC_LOCKBEAK=y, locking structures grow an extra int ->break_lock field which is used to implement raw_spin_is_contended() by setting the field to 1 when waiting on a lock and clearing it to zero when holding a lock. However, there are a few problems with this approach:

- There is a write-write race between a CPU successfully taking the lock (and subsequently writing break_lock = 0) and a waiter waiting on the lock (and subsequently writing break_lock = 1). This could result in a contended lock being reported as uncontended and vice-versa.

- On machines with store buffers, nothing guarantees that the writes to break_lock are visible to other CPUs at any particular time.

- READ_ONCE/WRITE_ONCE are not used, so the field is potentially susceptible to harmful compiler optimisations,

Consequently, the usefulness of this field is unclear and we'd be better off removing it and allowing architectures to implement raw_spin_is_contended() by providing a definition of arch_spin_is_contended(), as they can when CONFIG_GENERIC_LOCKBREAK=n.

---
 include/linux/rwlock_types.h   | 3 ---
 include/linux/spinlock.h       | 5 -----
 include/linux/spinlock_types.h | 3 ---
 kernel/locking/spinlock.c      | 9 +--------
 4 files changed, 1 insertion(+), 19 deletions(-)
include/linux/rwlock_types.h +0 -3
index cc0072e..857a72c 100644
@@ -10,9 +10,6 @@
  */
 typedef struct {
 	arch_rwlock_t raw_lock;
-#ifdef CONFIG_GENERIC_LOCKBREAK
-	unsigned int break_lock;
-#endif
 #ifdef CONFIG_DEBUG_SPINLOCK
 	unsigned int magic, owner_cpu;
 	void *owner;
include/linux/spinlock.h +0 -5
index a391861..3bf2735 100644
@@ -107,16 +107,11 @@ do {								\
 
 #define raw_spin_is_locked(lock)	arch_spin_is_locked(&(lock)->raw_lock)
 
-#ifdef CONFIG_GENERIC_LOCKBREAK
-#define raw_spin_is_contended(lock) ((lock)->break_lock)
-#else
-
 #ifdef arch_spin_is_contended
 #define raw_spin_is_contended(lock)	arch_spin_is_contended(&(lock)->raw_lock)
 #else
 #define raw_spin_is_contended(lock)	(((void)(lock), 0))
 #endif /*arch_spin_is_contended*/
-#endif
 
 /*
  * This barrier must provide two things:
include/linux/spinlock_types.h +0 -3
index 73548eb..24b4e6f 100644
@@ -19,9 +19,6 @@
 
 typedef struct raw_spinlock {
 	arch_spinlock_t raw_lock;
-#ifdef CONFIG_GENERIC_LOCKBREAK
-	unsigned int break_lock;
-#endif
 #ifdef CONFIG_DEBUG_SPINLOCK
 	unsigned int magic, owner_cpu;
 	void *owner;
kernel/locking/spinlock.c +1 -8
index 0ebb253..936f3d1 100644
@@ -66,12 +66,8 @@ void __lockfunc __raw_##op##_lock(locktype##_t *lock)			\
 			break;						\
 		preempt_enable();					\
 									\
-		if (!(lock)->break_lock)				\
-			(lock)->break_lock = 1;				\
-									\
 		arch_##op##_relax(&lock->raw_lock);			\
 	}								\
-	(lock)->break_lock = 0;						\
 }									\
 									\
 unsigned long __lockfunc __raw_##op##_lock_irqsave(locktype##_t *lock)	\
@@ -86,12 +82,9 @@ unsigned long __lockfunc __raw_##op##_lock_irqsave(locktype##_t *lock)	\
 		local_irq_restore(flags);				\
 		preempt_enable();					\
 									\
-		if (!(lock)->break_lock)				\
-			(lock)->break_lock = 1;				\
-									\
 		arch_##op##_relax(&lock->raw_lock);			\
 	}								\
-	(lock)->break_lock = 0;						\
+									\
 	return flags;							\
 }									\
 									\
#5 lore raw
In reply to Will Deacon
Quoted 26 lines · Will Deacon

Hi all,

The following two patches do the following:

1. Fix boot breakage reported on s390 caused by a8a217c22116
2. Kill off the break_lock field entirely, since it's not actually that useful

I didn't go the whole hog and remove CONFIG_GENERIC_LOCKBREAK, since the "do something different with locks in preemptible kernels" isn't necessarily a bad idea, and PowerPC actually implements things like arch_spin_relax for that.

The first patch should go in to 4.15.

Thanks,

Will

--->8

Will Deacon (2):
locking/core: Fix deadlock during boot on systems with
GENERIC_LOCKBREAK
locking/core: Remove break_lock field when CONFIG_GENERIC_LOCKBREAK=y

Quoted 5 lines
 include/linux/rwlock_types.h   |  3 ---
 include/linux/spinlock.h       |  5 -----
 include/linux/spinlock_types.h |  3 ---
 kernel/locking/spinlock.c      | 13 +++----------
 4 files changed, 3 insertions(+), 21 deletions(-)

*ping* ... these patches haven't been picked up yet as far as I can tell. Peter, Ingo?

#6 lore raw
Quoted 25 lines · Heiko Carstens
Quoted 24 lines · Will Deacon

The following two patches do the following:

1. Fix boot breakage reported on s390 caused by a8a217c22116
2. Kill off the break_lock field entirely, since it's not actually that useful

I didn't go the whole hog and remove CONFIG_GENERIC_LOCKBREAK, since the "do something different with locks in preemptible kernels" isn't necessarily a bad idea, and PowerPC actually implements things like arch_spin_relax for that.

The first patch should go in to 4.15.

Thanks,

Will

--->8

Will Deacon (2):
locking/core: Fix deadlock during boot on systems with
GENERIC_LOCKBREAK
locking/core: Remove break_lock field when CONFIG_GENERIC_LOCKBREAK=y

Quoted 8 lines
Quoted 5 lines
 include/linux/rwlock_types.h   |  3 ---
 include/linux/spinlock.h       |  5 -----
 include/linux/spinlock_types.h |  3 ---
 kernel/locking/spinlock.c      | 13 +++----------
 4 files changed, 3 insertions(+), 21 deletions(-)

*ping* ... these patches haven't been picked up yet as far as I can tell. Peter, Ingo?

Damn, this first patch didn't appear in -rc3.

Ingo -- please can you pick this up? Without it, s390 doesn't boot with mainline.

Thanks,

Will

#7 lore raw
Quoted 32 lines · Will Deacon
Quoted 31 lines · Heiko Carstens
Quoted 29 lines · Will Deacon

The following two patches do the following:

1. Fix boot breakage reported on s390 caused by a8a217c22116
2. Kill off the break_lock field entirely, since it's not actually that useful

I didn't go the whole hog and remove CONFIG_GENERIC_LOCKBREAK, since the "do something different with locks in preemptible kernels" isn't necessarily a bad idea, and PowerPC actually implements things like arch_spin_relax for that.

The first patch should go in to 4.15.

Thanks,

Will

--->8

Will Deacon (2):
locking/core: Fix deadlock during boot on systems with
GENERIC_LOCKBREAK
locking/core: Remove break_lock field when CONFIG_GENERIC_LOCKBREAK=y

 include/linux/rwlock_types.h   |  3 ---
 include/linux/spinlock.h       |  5 -----
 include/linux/spinlock_types.h |  3 ---
 kernel/locking/spinlock.c      | 13 +++----------
 4 files changed, 3 insertions(+), 21 deletions(-)
Quoted 7 lines
Quoted 2 lines

*ping* ... these patches haven't been picked up yet as far as I can tell. Peter, Ingo?

Damn, this first patch didn't appear in -rc3.

Ingo -- please can you pick this up? Without it, s390 doesn't boot with mainline.

Applied now, and will get it to Linus ASAP, sorry about that!

Thanks,

Ingo