src/runtime/proc.c    | 32 ++++++++++++++++++++++++++++++++
 src/runtime/runtime.h |  2 ++
 src/runtime/stack.c   |  7 +------

diff --git a/src/runtime/proc.c b/src/runtime/proc.c
index 91e3fe16d69ce1fb1b7c2efb7d8c70a4b20be8da..8462c4b1d61b7913415d3056c5be5c2740137e85 100644
--- a/src/runtime/proc.c
+++ b/src/runtime/proc.c
@@ -402,6 +402,7 @@ }
 
 static void badcasgstatus(void);
 static void helpcasgstatus(void);
+static void badgstatusrunnable(void);
 
 // If asked to move to or from a Gscanstatus this will throw. Use the castogscanstatus
 // and casfromgscanstatus instead.
@@ -423,6 +424,10 @@
 	// loop if gp->atomicstatus is in a scan state giving
 	// GC time to finish and change the state to oldval.
 	while(!runtime·cas(&gp->atomicstatus, oldval, newval)) {
+		if(oldval == Gwaiting && gp->atomicstatus == Grunnable) {
+			fn = badgstatusrunnable;
+			runtime·onM(&fn);
+		}
 		// Help GC if needed. 
 		if(gp->preemptscan && !gp->gcworkdone && (oldval == Grunning || oldval == Gsyscall)) {
 			gp->preemptscan = false;
@@ -431,6 +436,33 @@ 			fn = helpcasgstatus;
 			runtime·onM(&fn);
 		}
 	}	
+}
+
+static void
+badgstatusrunnable(void)
+{
+	runtime·throw("casgstatus: waiting for Gwaiting but is Grunnable");
+}
+
+// casgstatus(gp, oldstatus, Gcopystack), assuming oldstatus is Gwaiting or Grunnable.
+// Returns old status. Cannot call casgstatus directly, because we are racing with an
+// async wakeup that might come in from netpoll. If we see Gwaiting from the readgstatus,
+// it might have become Grunnable by the time we get to the cas. If we called casgstatus,
+// it would loop waiting for the status to go back to Gwaiting, which it never will.
+#pragma textflag NOSPLIT
+uint32
+runtime·casgcopystack(G *gp)
+{
+	uint32 oldstatus;
+
+	for(;;) {
+		oldstatus = runtime·readgstatus(gp) & ~Gscan;
+		if(oldstatus != Gwaiting && oldstatus != Grunnable)
+			runtime·throw("copystack: bad status, not Gwaiting or Grunnable");
+		if(runtime·cas(&gp->atomicstatus, oldstatus, Gcopystack))
+			break;
+	}
+	return oldstatus;
 }
 
 static void
diff --git a/src/runtime/runtime.h b/src/runtime/runtime.h
index 977c4547dfbacf55f2627968e9c4d3fd39c72016..177a1287eca0c41c9c1821d40980bcee1818d4e3 100644
--- a/src/runtime/runtime.h
+++ b/src/runtime/runtime.h
@@ -666,6 +666,8 @@ };
 
 uint32  runtime·readgstatus(G*);
 void    runtime·casgstatus(G*, uint32, uint32);
+void    runtime·casgstatus(G*, uint32, uint32);
+uint32	runtime·casgcopystack(G*);
 void    runtime·quiesce(G*);
 bool    runtime·stopg(G*);
 void    runtime·restartg(G*);
diff --git a/src/runtime/stack.c b/src/runtime/stack.c
index 072bc242bc2a5d3cad6bd004a2d2011b17fd5f92..cb9557243b5e921a6348134c9e547c72beaf29d9 100644
--- a/src/runtime/stack.c
+++ b/src/runtime/stack.c
@@ -637,12 +637,7 @@ 			*p++ = 0xfb;
 	}
 	runtime·memmove((byte*)new.hi - used, (byte*)old.hi - used, used);
 
-	oldstatus = runtime·readgstatus(gp);
-	oldstatus &= ~Gscan;
-	if(oldstatus == Gwaiting || oldstatus == Grunnable)
-		runtime·casgstatus(gp, oldstatus, Gcopystack); // oldstatus is Gwaiting or Grunnable
-	else
-		runtime·throw("copystack: bad status, not Gwaiting or Grunnable");
+	oldstatus = runtime·casgcopystack(gp); // cas from Gwaiting or Grunnable to Gcopystack, return old status
 
 	// Swap out old stack for new one
 	gp->stack = new;