diff --git a/src/runtime/export_test.go b/src/runtime/export_test.go
index 6a8d00c60d..67379796c7 100644
--- a/src/runtime/export_test.go
+++ b/src/runtime/export_test.go
@@ -975,3 +975,11 @@ func MapHashCheck(m interface{}, k interface{}) (uintptr, uintptr) {
 	y := typehash(kt, noescape(p), uintptr(mh.hash0))
 	return x, y
 }
+
+func MSpanCountAlloc(bits []byte) int {
+	s := mspan{
+		nelems:     uintptr(len(bits) * 8),
+		gcmarkBits: (*gcBits)(unsafe.Pointer(&bits[0])),
+	}
+	return s.countAlloc()
+}
diff --git a/src/runtime/gc_test.go b/src/runtime/gc_test.go
index 8ffb4f4a0f..4c281ce52b 100644
--- a/src/runtime/gc_test.go
+++ b/src/runtime/gc_test.go
@@ -6,6 +6,7 @@ package runtime_test
 
 import (
 	"fmt"
+	"math/rand"
 	"os"
 	"reflect"
 	"runtime"
@@ -751,6 +752,24 @@ func BenchmarkScanStackNoLocals(b *testing.B) {
 	close(teardown)
 }
 
+func BenchmarkMSpanCountAlloc(b *testing.B) {
+	// n is the number of bytes to benchmark against.
+	// n must always be a multiple of 8, since gcBits is
+	// always rounded up 8 bytes.
+	for _, n := range []int{8, 16, 32, 64, 128} {
+		b.Run(fmt.Sprintf("bits=%d", n*8), func(b *testing.B) {
+			// Initialize a new byte slice with pseduo-random data.
+			bits := make([]byte, n)
+			rand.Read(bits)
+
+			b.ResetTimer()
+			for i := 0; i < b.N; i++ {
+				runtime.MSpanCountAlloc(bits)
+			}
+		})
+	}
+}
+
 func countpwg(n *int, ready *sync.WaitGroup, teardown chan bool) {
 	if *n == 0 {
 		ready.Done()