Failure in WDF_SIMPLE_IO (IcmpSendEcho) on BCM5716C - adapter just does not reliably come back online RRS feed

  • Question

  • I am trying to get through HCK testing for a NDIS6 LWF.

    At present I am seeing failures in the WDF_SIMPLE_IO test.   After resume from sleep or standby, pings to the default gateway fail for too long a period of time.

    I have 4 identical systems (Dell PE R210 II) with WIn8x86, Win8x64, Win81x86 and Win8x64.

    Three of them have passed this test.  The outlier is the Win8x64 system.

    These are dual-nic systems.  Both nics have IPv6 addresses and default gateways.  Prior to starting the test I can ping both default gateways without issue.    After the tests the same is true.

    What I have found is that after resume from sleep/standby, no traffic passes through NIC for a long time.  Sometimes a very long time (as in the failure cases).   Packet counts on the NICs are zero.  I am looking at a case now where it has been zero for 30 minutes.   At the same time, the Media State shows enabled.   802.1x is disabled.

    In this particular case I have simulated the behavior manually by starting two continuous ping -t to the default gateways while everything was working.   My LWF is unbound from these NICs to eliminate it.   So this is base system behavior.   If I disable / re-enable the NICs, the pings both (eventually) succeed.   I let the system run in this state for a couple of minutes then put it to sleep.   I wait a minute, then wake it up.

    After waking I open the ncpa and bring up status on the two NICs.   The pings are running but failing alternately showing timed out or host unreachable.

    IPCONFIG shows both NICs active and with their assigned IPv6 addresses and gateways.

    The status panes, however, show zero packets in & out.

    The BCM drivers are latest (provided by Windows Update I presume, I did not do anything to get them).

    Curiously, from this state if I enabled (bind) my LWF or unbind Pacer, it 'fixes' the stack.  In other words, running the stack through Pause/Resume seems to make it all nice.

    If I run this test with both Pacer and my LWF unbound, it still occurs.

    Anyone else seen such an effect?   If so, how did you get around it?

    Dave Cattley

    Monday, January 6, 2014 8:56 PM

All replies

  • I found the issue.  It turned out to be the NIC driver on that one system (Win8 x64).  Windows Update had deployed updated drivers for the BCM5716C on all of the other platforms but not this one.   After getting a driver update package directly from BCM and installing it, the tests pass.

    Dave Cattley

    Tuesday, January 7, 2014 1:28 AM