Protect calls to fork with a gc_lock on MacOS

On MacOS running under Rosetta there is a hang if you are calling fork
from one thread and thread_suspend from another. (I believe the suspend
call is attempting to suspend the forking thread)

We call thread_suspend on all managed threads when doing a GC. We can
avoid the collision of calls if we use mono_gc_invoke_with_gc_lock
to protect the fork call.

I do not believe this is needed on other platforms or chips. I suspect
this is a Rosetta problem that we need to work around.

Author: bholmes

mono/metadata/w32process-unix.c

@@ -1488,6 +1488,15 @@ close_my_fds (void)
 }
 #endif
 
+#if defined (HAVE_FORK) 
+static void* fork_helper(void* data)
+{
+	pid_t* pid = (pid_t*)data;
+	*pid = fork();
+	return NULL;
+}
+#endif
+
 static gboolean
 process_create (const gunichar2 *appname, const gunichar2 *cmdline,
 	const gunichar2 *cwd, StartupHandles *startup_handles, MonoW32ProcessInfo *process_info)
@@ -1849,7 +1858,12 @@ process_create (const gunichar2 *appname, const gunichar2 *cmdline,
 		mono_trace (G_LOG_LEVEL_DEBUG, MONO_TRACE_IO_LAYER_PROCESS, "%s: new process startup not synchronized. We may not notice if the newly created process exits immediately.", __func__);
 	}
 
-	switch (pid = fork ()) {
+#if defined(HOST_DARWIN) && defined(TARGET_AMD64)
+	mono_gc_invoke_with_gc_lock(fork_helper, &pid);
+#else
+	pid = fork();
+#endif
+	switch (pid) {
 	case -1: /* Error */ {
 		mono_w32error_set_last (ERROR_OUTOFMEMORY);
 		ret = FALSE;