x := s.base() + i*s.elemsize
				trace.HeapObjectExists(x, s.typePointersOfUnchecked(x).typ)
			}
			abits.advance()
		}
	}

	// Write out all the goroutine stacks.
	forEachGRace(func(gp *g) {
		trace.GoroutineStackExists(gp.stack.lo, gp.stack.hi-gp.stack.lo)
	})
	traceRelease(trace)
}

func traceSpanTypeAndClass(s *mspan) traceArg {
	if s.state.get() == mSpanInUse {

	//
	// Any goroutine that the scheduler tries to execute concurrently with this
	// call will start by adding itself to the profile (before the act of
	// executing can cause any changes in its stack).
	forEachGRace(func(gp1 *g) {
		tryRecordGoroutineProfile(gp1, pcbuf, Gosched)
	})

	stw = stopTheWorld(stwGoroutineProfileCleanup)
	endOffset := goroutineProfile.offset.Swap(0)
	goroutineProfile.active = false

	// throw/panic, where locking could be out-of-order or a direct
	// deadlock.
	//
	// Instead, use forEachGRace, which requires no locking. We don't lock
	// against concurrent creation of new Gs, but even with allglock we may
	// miss Gs created after this loop.
	forEachGRace(func(gp *g) {
		if gp == me || gp == curgp || readgstatus(gp) == _Gdead || isSystemGoroutine(gp, false) && level < 2 {
			return
		}

		gp           *g
		goid         uint64
		mid          int64
		stackID      uint64
		status       uint32
		waitreason   waitReason
		inMarkAssist bool
	}
	var untracedGs []untracedG
	forEachGRace(func(gp *g) {
		// Make absolutely sure all Gs are ready for the next
		// generation. We need to do this even for dead Gs because
		// they may come alive with a new identity, and its status

	//
	// We only check the first nStackRoots Gs that we should have scanned.
	// Since we don't care about newer Gs (see comment in
	// gcMarkRootPrepare), no locking is required.
	i := 0
	forEachGRace(func(gp *g) {
		if i >= work.nStackRoots {
			return
		}

		if !gp.gcscandone {
			println("gp", gp, "goid", gp.goid,
				"status", readgstatus(gp),
				"gcscandone", gp.gcscandone)

func forEachG(fn func(gp *g)) {
	lock(&allglock)
	for _, gp := range allgs {
		fn(gp)
	}
	unlock(&allglock)
}

// forEachGRace calls fn on every G from allgs.
//
// forEachGRace avoids locking, but does not exclude addition of new Gs during
// execution, which may be missed.
func forEachGRace(fn func(gp *g)) {
	ptr, length := atomicAllG()
	for i := uintptr(0); i < length; i++ {
		gp := atomicAllGIndex(ptr, i)
		fn(gp)