1 Reply Latest reply on Dec 28, 2016 4:28 AM by Intel Corporation

    Are HW transmission timestamps for sequential packets on XL710 strictly increasing


      Upon reading your documentation for the XL710 and the driver and the Linux Kernel ( ) , it appears that HW Timestamping exists for the XL710 on transmit.  Specifically , Intel's driver uses it to implement PTP protocol in the driver file i40e_ptp.c for the Linux i40e driver.  My question is a follow up to anther poster's question ( Suresh ) in conjunction with explaining some of the uncertainties about the driver spec and the Linux kernel spec:  if one specifies the socket option SOF_TIMESTAMPING_TX_HARDWARE  ( with SO_TIMESTAMP also specified ) how is timestamp  included when "the packet is provided to the Network adapter" as defined in https://www.kernel.org/doc/Documentation/networking/timestamping.txt  ?


      My question pertains to this timestamping functionality, which according to linuxptp.sourceforge.net,  is considered to be applied at the MAC level and not the PHY. Can you let me know what that means ?  Are these timestamps always incrementing for packets as they are sent out over the wire or is there the potential that these timestamps will not be "strictly increasing" for consecutive packets sent out by the XL710 NIC ? 


      I am working on a project that doesn't care about the packets received, only that we know the ordering of the packets sent out.  If we know that two consecutive packets arriving at MAC layer will always have the first packet leaving and timestamped with a lower timestamp than it's successor, then we are happy.  Is this the case ? I wasn't sure i there is some reordering that my occur in the NIC between the MAC layer and Phy submission.


      Thank you for your feedback,