diff --git a/src/runtime/trace2.go b/src/runtime/trace2.go
index d40596f39be985..673205dda8a4a6 100644
--- a/src/runtime/trace2.go
+++ b/src/runtime/trace2.go
@@ -516,6 +516,9 @@ func traceAdvance(stopTrace bool) {
 	}
 	statusWriter.flush().end()
 
+	// Read everything out of the last gen's CPU profile buffer.
+	traceReadCPU(gen)
+
 	systemstack(func() {
 		// Flush CPU samples, stacks, and strings for the last generation. This is safe,
 		// because we're now certain no M is writing to the last generation.
diff --git a/src/runtime/trace2cpu.go b/src/runtime/trace2cpu.go
index 95c62c44b44114..4635662c08d56a 100644
--- a/src/runtime/trace2cpu.go
+++ b/src/runtime/trace2cpu.go
@@ -112,6 +112,9 @@ func traceStopReadCPU() {
 //
 // No more than one goroutine may be in traceReadCPU for the same
 // profBuf at a time.
+//
+// Must not run on the system stack because profBuf.read performs race
+// operations.
 func traceReadCPU(gen uintptr) bool {
 	var pcBuf [traceStackSize]uintptr
 
@@ -198,9 +201,6 @@ func traceReadCPU(gen uintptr) bool {
 //
 //go:systemstack
 func traceCPUFlush(gen uintptr) {
-	// Read everything out of the last gen's CPU profile buffer.
-	traceReadCPU(gen)
-
 	// Flush any remaining trace buffers containing CPU samples.
 	if buf := trace.cpuBuf[gen%2]; buf != nil {
 		lock(&trace.lock)