Don't makr PPS thread activation depenent on the socket transport.

All regression tests pass.

1 files changed, 11 insertions, 3 deletions

diff --git a/gpsd.c b/gpsd.c
index 97f2ba59..9256fb81 100644
--- a/gpsd.c
+++ b/gpsd.c
@@ -908,9 +908,6 @@ static bool awaken(struct gps_device_t *device)
 			device->gpsdata.gps_fd);
 	    FD_SET(device->gpsdata.gps_fd, &all_fds);
 	    adjust_max_fd(device->gpsdata.gps_fd, true);
-#ifdef NTPSHM_ENABLE
-	    pps_thread_activate(device);
-#endif /* NTPSHM_ENABLE */
 	    return true;
 	}
     }
@@ -1487,6 +1484,17 @@ static void consume_packets(struct gps_device_t *device)
 			device->gpsdata.dev.path,
 			gps_maskdump(device->gpsdata.set));
 
+#ifdef NTPSHM_ENABLE
+	/*
+	 * This has to be called late becvause of a Linux kernel bug
+	 * in 2.6; the PPS thread will hang if a baud rate change
+	 * happens while it's waiting on the PPS strobe.
+	 */
+	if ((changed & DRIVER_IS) != 0)
+	    pps_thread_activate(device);
+#endif /* NTPSHM_ENABLE */
+
+
 #ifdef SOCKET_EXPORT_ENABLE
 	/* add any just-identified device to watcher lists */
 	if ((changed & DRIVER_IS) != 0) {