Skip to contentSkip to conversation

locking/osq: Use optimized spinning loop for arm64

PATCH Applied Waiman Long 3 participants 3 messages Jan 13, 2020 – Jan 17, 2020

Subsystems (2)
lore
Share & tools
lore

Waiman Long

author reviewer
#0 lore raw

PATCH v3 1/1 locking/osq: Use optimized spinning loop for arm64

Arm64 has a more optimized spinning loop (atomic_cond_read_acquire) using wfe for spinlock that can boost performance of sibling threads by putting the current cpu to a wait state that is broken only when the monitored variable changes or an external event happens.

OSQ has a more complicated spinning loop. Besides the lock value, it also checks for need_resched() and vcpu_is_preempted(). The check for need_resched() is not a problem as it is only set by the tick interrupt handler. That will be detected by the spinning cpu right after iret.

The vcpu_is_preempted() check, however, is a problem as changes to the preempt state of of previous node will not affect the wait state. For ARM64, vcpu_is_preempted is not currently defined and so is a no-op. Will has indicated that he is planning to para-virtualize wfe instead of defining vcpu_is_preempted for PV support. So just add a comment in arch/arm64/include/asm/spinlock.h to indicate that vcpu_is_preempted() should not be defined as suggested.

On a 2-socket 56-core 224-thread ARM64 system, a kernel mutex locking microbenchmark was run for 10s with and without the patch. The performance numbers before patch were:

Running locktest with mutex [runtime = 10s, load = 1] Threads = 224, Min/Mean/Max = 316/123,143/2,121,269 Threads = 224, Total Rate = 2,757 kop/s; Percpu Rate = 12 kop/s

After patch, the numbers were:

Running locktest with mutex [runtime = 10s, load = 1] Threads = 224, Min/Mean/Max = 334/147,836/1,304,787 Threads = 224, Total Rate = 3,311 kop/s; Percpu Rate = 15 kop/s

So there was about 20% performance improvement.

---
 arch/arm64/include/asm/spinlock.h |  9 +++++++++
 kernel/locking/osq_lock.c         | 17 ++++-------------
 2 files changed, 13 insertions(+), 13 deletions(-)
arch/arm64/include/asm/spinlock.h +9 -0
index b093b287babf..102404dc1e13 100644
@@ -11,4 +11,13 @@
 /* See include/linux/spinlock.h */
 #define smp_mb__after_spinlock()	smp_mb()
 
+/*
+ * Changing this will break osq_lock() thanks to the call inside
+ * smp_cond_load_relaxed().
+ *
+ * See:
+ * https://lore.kernel.org/lkml/[email protected]
+ */
+#define vcpu_is_preempted(cpu)	false
+
 #endif /* __ASM_SPINLOCK_H */
kernel/locking/osq_lock.c +4 -13
index 6ef600aa0f47..e42893f2fcbc 100644
@@ -134,20 +134,11 @@ bool osq_lock(struct optimistic_spin_queue *lock)
 	 * cmpxchg in an attempt to undo our queueing.
 	 */
 
-	while (!READ_ONCE(node->locked)) {
-		/*
-		 * If we need to reschedule bail... so we can block.
-		 * Use vcpu_is_preempted() to avoid waiting for a preempted
-		 * lock holder:
-		 */
-		if (need_resched() || vcpu_is_preempted(node_cpu(node->prev)))
-			goto unqueue;
-
-		cpu_relax();
-	}
-	return true;
+	if (smp_cond_load_relaxed(&node->locked, VAL || need_resched() ||
+				  vcpu_is_preempted(node_cpu(node->prev))))
+		return true;
 
-unqueue:
+	/* unqueue */
 	/*
 	 * Step - A  -- stabilize @prev
 	 *
Signature · 1 line
2.18.1
maintainer Acked-by on v3 1/1
#1 lore raw
Quoted 32 lines · Waiman Long

Arm64 has a more optimized spinning loop (atomic_cond_read_acquire) using wfe for spinlock that can boost performance of sibling threads by putting the current cpu to a wait state that is broken only when the monitored variable changes or an external event happens.

OSQ has a more complicated spinning loop. Besides the lock value, it also checks for need_resched() and vcpu_is_preempted(). The check for need_resched() is not a problem as it is only set by the tick interrupt handler. That will be detected by the spinning cpu right after iret.

The vcpu_is_preempted() check, however, is a problem as changes to the preempt state of of previous node will not affect the wait state. For ARM64, vcpu_is_preempted is not currently defined and so is a no-op. Will has indicated that he is planning to para-virtualize wfe instead of defining vcpu_is_preempted for PV support. So just add a comment in arch/arm64/include/asm/spinlock.h to indicate that vcpu_is_preempted() should not be defined as suggested.

On a 2-socket 56-core 224-thread ARM64 system, a kernel mutex locking microbenchmark was run for 10s with and without the patch. The performance numbers before patch were:

Running locktest with mutex [runtime = 10s, load = 1] Threads = 224, Min/Mean/Max = 316/123,143/2,121,269 Threads = 224, Total Rate = 2,757 kop/s; Percpu Rate = 12 kop/s

After patch, the numbers were:

Running locktest with mutex [runtime = 10s, load = 1] Threads = 224, Min/Mean/Max = 334/147,836/1,304,787 Threads = 224, Total Rate = 3,311 kop/s; Percpu Rate = 15 kop/s

Quoted 7 lines

So there was about 20% performance improvement.

---
 arch/arm64/include/asm/spinlock.h |  9 +++++++++
 kernel/locking/osq_lock.c         | 17 ++++-------------
 2 files changed, 13 insertions(+), 13 deletions(-)

Thanks,

Will

merge bot Applied
#2 lore raw
In reply to Waiman Long
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:     f5bfdc8e3947a7ae489cf8ae9cfd6b3fb357b952
Gitweb:        https://git.kernel.org/tip/f5bfdc8e3947a7ae489cf8ae9cfd6b3fb357b952
Author:        Waiman Long <[email protected]>
AuthorDate:    Mon, 13 Jan 2020 10:07:35 -05:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Fri, 17 Jan 2020 10:19:30 +01:00

locking/osq: Use optimized spinning loop for arm64

Arm64 has a more optimized spinning loop (atomic_cond_read_acquire) using wfe for spinlock that can boost performance of sibling threads by putting the current cpu to a wait state that is broken only when the monitored variable changes or an external event happens.

OSQ has a more complicated spinning loop. Besides the lock value, it also checks for need_resched() and vcpu_is_preempted(). The check for need_resched() is not a problem as it is only set by the tick interrupt handler. That will be detected by the spinning cpu right after iret.

The vcpu_is_preempted() check, however, is a problem as changes to the preempt state of of previous node will not affect the wait state. For ARM64, vcpu_is_preempted is not currently defined and so is a no-op. Will has indicated that he is planning to para-virtualize wfe instead of defining vcpu_is_preempted for PV support. So just add a comment in arch/arm64/include/asm/spinlock.h to indicate that vcpu_is_preempted() should not be defined as suggested.

On a 2-socket 56-core 224-thread ARM64 system, a kernel mutex locking microbenchmark was run for 10s with and without the patch. The performance numbers before patch were:

Running locktest with mutex [runtime = 10s, load = 1] Threads = 224, Min/Mean/Max = 316/123,143/2,121,269 Threads = 224, Total Rate = 2,757 kop/s; Percpu Rate = 12 kop/s

After patch, the numbers were:

Running locktest with mutex [runtime = 10s, load = 1] Threads = 224, Min/Mean/Max = 334/147,836/1,304,787 Threads = 224, Total Rate = 3,311 kop/s; Percpu Rate = 15 kop/s

So there was about 20% performance improvement.

---
 arch/arm64/include/asm/spinlock.h |  9 +++++++++
 kernel/locking/osq_lock.c         | 23 ++++++++++-------------
 2 files changed, 19 insertions(+), 13 deletions(-)
arch/arm64/include/asm/spinlock.h +9 -0
index b093b28..102404d 100644
@@ -11,4 +11,13 @@
 /* See include/linux/spinlock.h */
 #define smp_mb__after_spinlock()	smp_mb()
 
+/*
+ * Changing this will break osq_lock() thanks to the call inside
+ * smp_cond_load_relaxed().
+ *
+ * See:
+ * https://lore.kernel.org/lkml/[email protected]
+ */
+#define vcpu_is_preempted(cpu)	false
+
 #endif /* __ASM_SPINLOCK_H */
kernel/locking/osq_lock.c +10 -13
index 6ef600a..1f77349 100644
@@ -134,20 +134,17 @@ bool osq_lock(struct optimistic_spin_queue *lock)
 	 * cmpxchg in an attempt to undo our queueing.
 	 */
 
-	while (!READ_ONCE(node->locked)) {
-		/*
-		 * If we need to reschedule bail... so we can block.
-		 * Use vcpu_is_preempted() to avoid waiting for a preempted
-		 * lock holder:
-		 */
-		if (need_resched() || vcpu_is_preempted(node_cpu(node->prev)))
-			goto unqueue;
-
-		cpu_relax();
-	}
-	return true;
+	/*
+	 * Wait to acquire the lock or cancelation. Note that need_resched()
+	 * will come with an IPI, which will wake smp_cond_load_relaxed() if it
+	 * is implemented with a monitor-wait. vcpu_is_preempted() relies on
+	 * polling, be careful.
+	 */
+	if (smp_cond_load_relaxed(&node->locked, VAL || need_resched() ||
+				  vcpu_is_preempted(node_cpu(node->prev))))
+		return true;
 
-unqueue:
+	/* unqueue */
 	/*
 	 * Step - A  -- stabilize @prev
 	 *