runtime: make NumGoroutine and Stack agree not to include system goroutines

Before, NumGoroutine counted system goroutines and Stack (usually) didn't show them,
which was inconsistent and confusing.

To resolve which way they should be consistent, it seems like

	package main
	import "runtime"
	func main() { println(runtime.NumGoroutine()) }

should print 1 regardless of internal runtime details. Make it so.

Fixes #11706.

Change-Id: I6bfe26a901de517728192cfb26a5568c4ef4fe47
Reviewed-on: https://go-review.googlesource.com/18343
Reviewed-by: Austin Clements <austin@google.com>

src/runtime/mprof.go

@@ -576,12 +576,17 @@ func Stack(buf []byte, all bool) int {
 		pc := getcallerpc(unsafe.Pointer(&buf))
 		systemstack(func() {
 			g0 := getg()
+			// Force traceback=1 to override GOTRACEBACK setting,
+			// so that Stack's results are consistent.
+			// GOTRACEBACK is only about crash dumps.
+			g0.m.traceback = 1
 			g0.writebuf = buf[0:0:len(buf)]
 			goroutineheader(gp)
 			traceback(pc, sp, 0, gp)
 			if all {
 				tracebackothers(gp)
 			}
+			g0.m.traceback = 0
 			n = len(g0.writebuf)
 			g0.writebuf = nil
 		})

src/runtime/proc.go

@@ -2162,6 +2162,9 @@ func goexit0(gp *g) {
 	_g_ := getg()
 
 	casgstatus(gp, _Grunning, _Gdead)
+	if isSystemGoroutine(gp) {
+		atomic.Xadd(&sched.ngsys, -1)
+	}
 	gp.m = nil
 	gp.lockedm = nil
 	_g_.m.lockedg = nil
@@ -2693,6 +2696,9 @@ func newproc1(fn *funcval, argp *uint8, narg int32, nret int32, callerpc uintptr
 	gostartcallfn(&newg.sched, fn)
 	newg.gopc = callerpc
 	newg.startpc = fn.fn
+	if isSystemGoroutine(newg) {
+		atomic.Xadd(&sched.ngsys, +1)
+	}
 	casgstatus(newg, _Gdead, _Grunnable)
 
 	if _p_.goidcache == _p_.goidcacheend {
@@ -2885,7 +2891,7 @@ func badunlockosthread() {
 }
 
 func gcount() int32 {
-	n := int32(allglen) - sched.ngfree
+	n := int32(allglen) - sched.ngfree - int32(atomic.Load(&sched.ngsys))
 	for i := 0; ; i++ {
 		_p_ := allp[i]
 		if _p_ == nil {

src/runtime/proc_test.go

@@ -9,6 +9,7 @@ import (
 	"net"
 	"runtime"
 	"runtime/debug"
+	"strings"
 	"sync"
 	"sync/atomic"
 	"syscall"
@@ -336,6 +337,23 @@ func TestGCFairness(t *testing.T) {
 	}
 }
 
+func TestNumGoroutine(t *testing.T) {
+	output := runTestProg(t, "testprog", "NumGoroutine")
+	want := "1\n"
+	if output != want {
+		t.Fatalf("want %q, got %q", want, output)
+	}
+
+	buf := make([]byte, 1<<20)
+	buf = buf[:runtime.Stack(buf, true)]
+
+	n := runtime.NumGoroutine()
+
+	if nstk := strings.Count(string(buf), "goroutine "); n != nstk {
+		t.Fatalf("NumGoroutine=%d, but found %d goroutines in stack dump", n, nstk)
+	}
+}
+
 func TestPingPongHog(t *testing.T) {
 	if testing.Short() {
 		t.Skip("skipping in -short mode")

src/runtime/runtime2.go

@@ -418,6 +418,8 @@ type schedt struct {
 	mcount       int32    // number of m's that have been created
 	maxmcount    int32    // maximum number of m's allowed (or die)
 
+	ngsys uint32 // number of system goroutines; updated atomically
+
 	pidle      puintptr // idle p's
 	npidle     uint32
 	nmspinning uint32 // See "Worker thread parking/unparking" comment in proc.go.

src/runtime/testdata/testprog/misc.go

@@ -0,0 +1,15 @@
+// Copyright 2016 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 main
+
+import "runtime"
+
+func init() {
+	register("NumGoroutine", NumGoroutine)
+}
+
+func NumGoroutine() {
+	println(runtime.NumGoroutine())
+}