runtime: fix CPU underutilization

runtime.newproc/ready are deliberately sloppy about waking new M's,
they only ensure that there is at least 1 spinning M.
Currently to compensate for that, schedule() checks if the current P
has local work and there are no spinning M's, it wakes up another one.
It does not work if goroutines do not call schedule.
With this change a spinning M wakes up another M when it finds work to do.
It's also not ideal, but it fixes the underutilization.
A proper check would require to know the exact number of runnable G's,
but it's too expensive to maintain.
Fixes #5586.

R=rsc
CC=gobot, golang-dev
https://golang.org/cl/9776044

src/pkg/runtime/proc_test.go

@@ -93,6 +93,30 @@ func TestYieldLocked(t *testing.T) {
 	<-c
 }
 
+func TestGoroutineParallelism(t *testing.T) {
+	const P = 4
+	defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(P))
+	for try := 0; try < 10; try++ {
+		done := make(chan bool)
+		x := uint32(0)
+		for p := 0; p < P; p++ {
+			// Test that all P goroutines are scheduled at the same time
+			go func(p int) {
+				for i := 0; i < 3; i++ {
+					expected := uint32(P*i + p)
+					for atomic.LoadUint32(&x) != expected {
+					}
+					atomic.StoreUint32(&x, expected+1)
+				}
+				done <- true
+			}(p)
+		}
+		for p := 0; p < P; p++ {
+			<-done
+		}
+	}
+}
+
 func TestBlockLocked(t *testing.T) {
 	const N = 10
 	c := make(chan bool)