diff --git a/src/testing/benchmark.go b/src/testing/benchmark.go
index c033ce5fec..bcebb418c4 100644
--- a/src/testing/benchmark.go
+++ b/src/testing/benchmark.go
@@ -219,7 +219,7 @@ func (b *B) run1() bool {
 	}
 	// Only print the output if we know we are not going to proceed.
 	// Otherwise it is printed in processBench.
-	if b.hasSub || b.finished {
+	if atomic.LoadInt32(&b.hasSub) != 0 || b.finished {
 		tag := "BENCH"
 		if b.skipped {
 			tag = "SKIP"
@@ -460,10 +460,13 @@ func (ctx *benchContext) processBench(b *B) {
 //
 // A subbenchmark is like any other benchmark. A benchmark that calls Run at
 // least once will not be measured itself and will be called once with N=1.
+//
+// Run may be called simultaneously from multiple goroutines, but all such
+// calls must happen before the outer benchmark function for b returns.
 func (b *B) Run(name string, f func(b *B)) bool {
 	// Since b has subbenchmarks, we will no longer run it as a benchmark itself.
 	// Release the lock and acquire it on exit to ensure locks stay paired.
-	b.hasSub = true
+	atomic.StoreInt32(&b.hasSub, 1)
 	benchmarkLock.Unlock()
 	defer benchmarkLock.Lock()
 
diff --git a/src/testing/sub_test.go b/src/testing/sub_test.go
index 8d5d9206f0..bb7b3e0925 100644
--- a/src/testing/sub_test.go
+++ b/src/testing/sub_test.go
@@ -6,6 +6,7 @@ package testing
 
 import (
 	"bytes"
+	"fmt"
 	"regexp"
 	"strings"
 	"sync/atomic"
@@ -515,3 +516,19 @@ func TestBenchmarkOutput(t *T) {
 	Benchmark(func(b *B) { b.Error("do not print this output") })
 	Benchmark(func(b *B) {})
 }
+
+func TestParallelSub(t *T) {
+	c := make(chan int)
+	block := make(chan int)
+	for i := 0; i < 10; i++ {
+		go func(i int) {
+			<-block
+			t.Run(fmt.Sprint(i), func(t *T) {})
+			c <- 1
+		}(i)
+	}
+	close(block)
+	for i := 0; i < 10; i++ {
+		<-c
+	}
+}
diff --git a/src/testing/testing.go b/src/testing/testing.go
index c972b2737f..ddbdc25bf1 100644
--- a/src/testing/testing.go
+++ b/src/testing/testing.go
@@ -216,6 +216,7 @@ import (
 	"strconv"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 )
 
@@ -267,8 +268,8 @@ type common struct {
 	skipped    bool         // Test of benchmark has been skipped.
 	finished   bool         // Test function has completed.
 	done       bool         // Test is finished and all subtests have completed.
-	hasSub     bool
-	raceErrors int // number of races detected during test
+	hasSub     int32        // written atomically
+	raceErrors int          // number of races detected during test
 
 	parent   *common
 	level    int       // Nesting depth of test or benchmark.
@@ -645,7 +646,7 @@ func tRunner(t *T, fn func(t *T)) {
 		// Do not lock t.done to allow race detector to detect race in case
 		// the user does not appropriately synchronizes a goroutine.
 		t.done = true
-		if t.parent != nil && !t.hasSub {
+		if t.parent != nil && atomic.LoadInt32(&t.hasSub) == 0 {
 			t.setRan()
 		}
 		t.signal <- true
@@ -659,8 +660,11 @@ func tRunner(t *T, fn func(t *T)) {
 
 // Run runs f as a subtest of t called name. It reports whether f succeeded.
 // Run will block until all its parallel subtests have completed.
+//
+// Run may be called simultaneously from multiple goroutines, but all such
+// calls must happen before the outer test function for t returns.
 func (t *T) Run(name string, f func(t *T)) bool {
-	t.hasSub = true
+	atomic.StoreInt32(&t.hasSub, 1)
 	testName, ok := t.context.match.fullName(&t.common, name)
 	if !ok {
 		return true