Skip to contentSkip to conversation

asm-generic: ticket-lock: Optimize arch_spin_value_unlocked

PATCH Applied [email protected] 2 participants 2 messages Sep 8, 2023 – Sep 21, 2023

Subsystems (2)
RISC-V Generic include/asm header files
lore
Share & tools
lore

[email protected]

author
#0 lore raw

PATCH v3 1/1 asm-generic: ticket-lock: Optimize arch_spin_value_unlocked

From: Guo Ren <[email protected]>

The arch_spin_value_unlocked of ticket-lock would cause the compiler to generate inefficient asm code in riscv architecture because of unnecessary memory access to the contended value.

Before the patch:
------
void lockref_get(struct lockref *lockref)
{

  78:   fd010113                add     sp,sp,-48
  7c:   02813023                sd      s0,32(sp)
  80:   02113423                sd      ra,40(sp)
  84:   03010413                add     s0,sp,48

0000000000000088 <.LBB296>:

        CMPXCHG_LOOP(

88: 00053783 ld a5,0(a0)
------

After the patch:
------
void lockref_get(struct lockref *lockref)

{
        CMPXCHG_LOOP(

78: 00053783 ld a5,0(a0)
------

After the patch, the lockref_get could get in a fast path instead of the function's prologue. This is because ticket lock complex logic would limit compiler optimization for the spinlock fast path, and qspinlock won't.

The caller of arch_spin_value_unlocked() could benefit from this change. Currently, the only caller is lockref.

Acked-by Waiman Long <[email protected]>
Acked-by Will Deacon <[email protected]>
Signed-off-by Guo Ren <[email protected]>
Signed-off-by Guo Ren <[email protected]>

---
Changelog
V3:
- Add Acked-by tags
- Optimize commit log

V2:
- Fixup commit log with Waiman advice.
- Add Waiman comment in the commit msg.

---
 include/asm-generic/spinlock.h | 16 +++++++++-------
 1 file changed, 9 insertions(+), 7 deletions(-)
include/asm-generic/spinlock.h +9 -7
index fdfebcb050f4..90803a826ba0 100644
@@ -68,11 +68,18 @@ static __always_inline void arch_spin_unlock(arch_spinlock_t *lock)
 	smp_store_release(ptr, (u16)val + 1);
 }
 
+static __always_inline int arch_spin_value_unlocked(arch_spinlock_t lock)
+{
+	u32 val = lock.counter;
+
+	return ((val >> 16) == (val & 0xffff));
+}
+
 static __always_inline int arch_spin_is_locked(arch_spinlock_t *lock)
 {
-	u32 val = atomic_read(lock);
+	arch_spinlock_t val = READ_ONCE(*lock);
 
-	return ((val >> 16) != (val & 0xffff));
+	return !arch_spin_value_unlocked(val);
 }
 
 static __always_inline int arch_spin_is_contended(arch_spinlock_t *lock)

@@ -82,11 +89,6 @@ static __always_inline int arch_spin_is_contended(arch_spinlock_t *lock)
 	return (s16)((val >> 16) - (val & 0xffff)) > 1;
 }
 
-static __always_inline int arch_spin_value_unlocked(arch_spinlock_t lock)
-{
-	return !arch_spin_is_locked(&lock);
-}
-
 #include <asm/qrwlock.h>
 
 #endif /* __ASM_GENERIC_SPINLOCK_H */
Signature · 1 line
2.36.1
merge bot Applied
#1 lore raw

[tip: locking/core] asm-generic: ticket-lock: Optimize arch_spin_value_unlocked()

The following commit has been merged into the locking/core branch of tip:

The following commit has been merged into the locking/core branch of tip:

Commit-ID:     c6f4a90022524d06f6d9de323b1757031dcf0c26
Gitweb:        https://git.kernel.org/tip/c6f4a90022524d06f6d9de323b1757031dcf0c26
Author:        Guo Ren <[email protected]>
AuthorDate:    Fri, 08 Sep 2023 11:43:39 -04:00
Committer:     Ingo Molnar <[email protected]>
CommitterDate: Thu, 21 Sep 2023 10:17:00 +02:00

asm-generic: ticket-lock: Optimize arch_spin_value_unlocked()

The arch_spin_value_unlocked() of ticket-lock would cause the compiler to generate inefficient asm code in riscv architecture because of unnecessary memory access to the contended value.

Before the patch:

	void lockref_get(struct lockref *lockref)
	{
	  78:   fd010113                add     sp,sp,-48
	  7c:   02813023                sd      s0,32(sp)
	  80:   02113423                sd      ra,40(sp)
	  84:   03010413                add     s0,sp,48

	0000000000000088 <.LBB296>:
		CMPXCHG_LOOP(
	  88:   00053783                ld      a5,0(a0)

After the patch:

	void lockref_get(struct lockref *lockref)
	{
		CMPXCHG_LOOP(
	  78:   00053783                ld      a5,0(a0)

After the patch, the lockref_get() could get in a fast path instead of the function's prologue. This is because ticket lock complex logic would limit compiler optimization for the spinlock fast path, and qspinlock won't.

The caller of arch_spin_value_unlocked() could benefit from this change. Currently, the only caller is lockref.

Signed-off-by Guo Ren <[email protected]>
Signed-off-by Guo Ren <[email protected]>
Signed-off-by Ingo Molnar <[email protected]>
Acked-by Waiman Long <[email protected]>
Acked-by Will Deacon <[email protected]>
Link https://lore.kernel.org/r/[email protected] open here
---
 include/asm-generic/spinlock.h | 16 +++++++++-------
 1 file changed, 9 insertions(+), 7 deletions(-)
include/asm-generic/spinlock.h +9 -7
index fdfebcb..90803a8 100644
@@ -68,11 +68,18 @@ static __always_inline void arch_spin_unlock(arch_spinlock_t *lock)
 	smp_store_release(ptr, (u16)val + 1);
 }
 
+static __always_inline int arch_spin_value_unlocked(arch_spinlock_t lock)
+{
+	u32 val = lock.counter;
+
+	return ((val >> 16) == (val & 0xffff));
+}
+
 static __always_inline int arch_spin_is_locked(arch_spinlock_t *lock)
 {
-	u32 val = atomic_read(lock);
+	arch_spinlock_t val = READ_ONCE(*lock);
 
-	return ((val >> 16) != (val & 0xffff));
+	return !arch_spin_value_unlocked(val);
 }
 
 static __always_inline int arch_spin_is_contended(arch_spinlock_t *lock)

@@ -82,11 +89,6 @@ static __always_inline int arch_spin_is_contended(arch_spinlock_t *lock)
 	return (s16)((val >> 16) - (val & 0xffff)) > 1;
 }
 
-static __always_inline int arch_spin_value_unlocked(arch_spinlock_t lock)
-{
-	return !arch_spin_is_locked(&lock);
-}
-
 #include <asm/qrwlock.h>
 
 #endif /* __ASM_GENERIC_SPINLOCK_H */