From: Johannes Berg <johannes.berg@xxxxxxxxx>
There's a window of time, that probably never really happens
in real life, where we can TX before the kernel has run the
linkwatch_event(), which sets up the qdisc in dev_activate().
If we happen to attempt to do connectivity checks before the
event has been processed, all frames are dropped in the noop
qdisc, causing this to fail.
Signed-off-by: Johannes Berg <johannes.berg@xxxxxxxxx>
---
tests/hwsim/hwsim_utils.py | 12 ++++++++++++
1 file changed, 12 insertions(+)
diff --git a/tests/hwsim/hwsim_utils.py b/tests/hwsim/hwsim_utils.py
index dccdefc71c56..5070b3a6ff8e 100644
--- a/tests/hwsim/hwsim_utils.py
+++ b/tests/hwsim/hwsim_utils.py
@@ -15,6 +15,18 @@ def run_connectivity_test(dev1, dev2, tos, dev1group=False, dev2group=False,
ifname1=None, ifname2=None, config=True, timeout=5,
multicast_to_unicast=False, broadcast=True,
send_len=None):
+ # If running on a single-CPU system, and in particular on UML with
+ # purely virtual time (where there's no real preemption), we might
+ # be *really* fast to get to this point after starting things up,
+ # and that means that after the call to the linkwatch_fire_event()
+ # function in the kernel we still continue processing immediately
+ # in userspace (i.e. here) and manage to try to test the transmit
+ # before we have a real qdisc assigned to the netdev queues...
+ # Sleep a tiny little bit here to let the kernel actually run the
+ # scheduled work function (linkwatch_event()) after having queued
+ # it from the carrier on event.
+ time.sleep(0.01)
+
addr1 = dev1.own_addr()
if not dev1group and isinstance(dev1, WpaSupplicant):
addr = dev1.get_driver_status_field('addr')
--
2.17.2
_______________________________________________
Hostap mailing list
Hostap@xxxxxxxxxxxxxxxxxxx
http://lists.infradead.org/mailman/listinfo/hostap
