runtime: give proper messages when user defined DYLD_INSERT_LIBRARIES on Darwin

       also gives clear error when bsdthread_register fails on Darwin
       Fixes #2992.

R=rsc, krautz
CC=golang-dev
https://golang.org/cl/5966067

src/pkg/runtime/os_darwin.h

@@ -6,7 +6,7 @@
 #define SIG_IGN ((void*)1)
 
 int32	runtime·bsdthread_create(void*, M*, G*, void(*)(void));
-void	runtime·bsdthread_register(void);
+int32	runtime·bsdthread_register(void);
 int32	runtime·mach_msg_trap(MachHeader*, int32, uint32, uint32, uint32, uint32, uint32);
 uint32	runtime·mach_reply_port(void);
 int32	runtime·mach_semacquire(uint32, int64);

src/pkg/runtime/sys_darwin_386.s

@@ -268,8 +268,10 @@ TEXT runtime·bsdthread_register(SB),7,$40
 	MOVL	$0, 20(SP)	// targetconc_ptr
 	MOVL	$0, 24(SP)	// dispatchqueue_offset
 	INT	$0x80
-	JAE	2(PC)
-	MOVL	$0xf1, 0xf1  // crash
+	JAE	3(PC)
+	NEGL	AX
+	RET
+	MOVL	$0, AX
 	RET
 
 // Invoke Mach system call.

src/pkg/runtime/sys_darwin_amd64.s

@@ -265,8 +265,10 @@ TEXT runtime·bsdthread_register(SB),7,$0
 	MOVQ	$0, R9	// dispatchqueue_offset
 	MOVQ	$(0x2000000+366), AX	// bsdthread_register
 	SYSCALL
-	JCC 2(PC)
-	MOVL	$0xf1, 0xf1  // crash
+	JCC 3(PC)
+	NEGL	AX
+	RET
+	MOVL	$0, AX
 	RET
 
 // Mach system calls use 0x1000000 instead of the BSD's 0x2000000.

src/pkg/runtime/thread_darwin.c

@@ -50,11 +50,8 @@ runtime·semacreate(void)
 void
 runtime·osinit(void)
 {
-	// Register our thread-creation callback (see sys_darwin_{amd64,386}.s)
-	// but only if we're not using cgo.  If we are using cgo we need
-	// to let the C pthread libary install its own thread-creation callback.
-	if(!runtime·iscgo)
-		runtime·bsdthread_register();
+	// bsdthread_register delayed until end of goenvs so that we
+	// can look at the environment first.
 
 	// Use sysctl to fetch hw.ncpu.
 	uint32 mib[2];
@@ -75,6 +72,18 @@ void
 runtime·goenvs(void)
 {
 	runtime·goenvs_unix();
+
+	// Register our thread-creation callback (see sys_darwin_{amd64,386}.s)
+	// but only if we're not using cgo.  If we are using cgo we need
+	// to let the C pthread libary install its own thread-creation callback.
+	if(!runtime·iscgo) {
+		if(runtime·bsdthread_register() != 0) {
+			if(runtime·getenv("DYLD_INSERT_LIBRARIES"))
+				runtime·throw("runtime: bsdthread_register error (unset DYLD_INSERT_LIBRARIES)");
+			runtime·throw("runtime: bsdthread_register error");
+		}
+	}
+
 }
 
 void