sync: yield to the waiter when unlocking a starving mutex

When we have already assigned the semaphore ticket to a specific
waiter, we want to get the waiter running as fast as possible since
no other G waiting on the semaphore can acquire it optimistically.

The net effect is that, when a sync.Mutex is contented, the code in
the critical section guarded by the Mutex gets a priority boost.

Fixes #33747

Change-Id: I9967f0f763c25504010651bdd7f944ee0189cd45
Reviewed-on: https://go-review.googlesource.com/c/go/+/200577
Reviewed-by: Rhys Hiltner <rhys@justin.tv>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Emmanuel Odeke <emm.odeke@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

1 files changed, 80 insertions, 0 deletions

diff --git a/src/runtime/sema_test.go b/src/runtime/sema_test.go
new file mode 100644
index 0000000000..5cd2317269
--- /dev/null
+++ b/src/runtime/sema_test.go
@@ -0,0 +1,80 @@
+// Copyright 2019 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package runtime_test
+
+import (
+	. "runtime"
+	"sync/atomic"
+	"testing"
+)
+
+// TestSemaHandoff checks that when semrelease+handoff is
+// requested, the G that releases the semaphore yields its
+// P directly to the first waiter in line.
+// See issue 33747 for discussion.
+func TestSemaHandoff(t *testing.T) {
+	const iter = 10000
+	ok := 0
+	for i := 0; i < iter; i++ {
+		if testSemaHandoff() {
+			ok++
+		}
+	}
+	// As long as two thirds of handoffs are direct, we
+	// consider the test successful. The scheduler is
+	// nondeterministic, so this test checks that we get the
+	// desired outcome in a significant majority of cases.
+	// The actual ratio of direct handoffs is much higher
+	// (>90%) but we use a lower threshold to minimize the
+	// chances that unrelated changes in the runtime will
+	// cause the test to fail or become flaky.
+	if ok < iter*2/3 {
+		t.Fatal("direct handoff < 2/3:", ok, iter)
+	}
+}
+
+func TestSemaHandoff1(t *testing.T) {
+	if GOMAXPROCS(-1) <= 1 {
+		t.Skip("GOMAXPROCS <= 1")
+	}
+	defer GOMAXPROCS(GOMAXPROCS(-1))
+	GOMAXPROCS(1)
+	TestSemaHandoff(t)
+}
+
+func TestSemaHandoff2(t *testing.T) {
+	if GOMAXPROCS(-1) <= 2 {
+		t.Skip("GOMAXPROCS <= 2")
+	}
+	defer GOMAXPROCS(GOMAXPROCS(-1))
+	GOMAXPROCS(2)
+	TestSemaHandoff(t)
+}
+
+func testSemaHandoff() bool {
+	var sema, res uint32
+	done := make(chan struct{})
+
+	go func() {
+		Semacquire(&sema)
+		atomic.CompareAndSwapUint32(&res, 0, 1)
+
+		Semrelease1(&sema, true, 0)
+		close(done)
+	}()
+	for SemNwait(&sema) == 0 {
+		Gosched() // wait for goroutine to block in Semacquire
+	}
+
+	// The crux of the test: we release the semaphore with handoff
+	// and immediately perform a CAS both here and in the waiter; we
+	// want the CAS in the waiter to execute first.
+	Semrelease1(&sema, true, 0)
+	atomic.CompareAndSwapUint32(&res, 0, 2)
+
+	<-done // wait for goroutines to finish to avoid data races
+
+	return res == 1 // did the waiter run first?
+}