IV82694: VIO SERVER USING 10 GB PCIE ADAPTERS AND LARGE_SEND MAY CRASH APPLIES TO AIX 6100-09

APAR status

• Closed as program error.

Error description

• **************************************************************
  * USERS AFFECTED:
    * Systems running the AIX 6100-09 Technology Level
    * or VIOS 2.2.4.x with bos.net.tcp.client between
    * and including the 6.1.9.100 and 6.1.9.102 levels.
    **************************************************************
    * PROBLEM DESCRIPTION:
    * A VIO Server running 2.2.4 and using a 10Gb PCIe adapter
    * may crash or have other unexpected issues during normal
    * network communications.
    *
    * This happens when chksum_offload is enabled on the
    * VIO server and client LPARs, which is typically done to
    * enable large_send.
    *
    * It is recommended that every system running VIOS 2.2.4
    * with a 10GB PCIe adapter install this fix.
    *
    * The specific adapters that we have seen this problem with
    * are:
    * EC27 PCIe2 LP 2-Port 10GbE RoCE SFP+ Adapter
    * EC28 PCIe2 2-Port 10 GbE RoCE SFP+ adapter
    * EC37 (Copper) PCIe3 2-port 10 GbE NIC and RoCE SFP+ Copper
    * EC38 (Copper) PCIe3 LP 2-port 10 GbE NIC and RoCE SFP+
    Copper
    * EL3X (Copper) PCIe3 LP 2-Port 10 GbE NIC and RoCE SFP+
    Copper
    * EC2M (Fiber) PCIe3 2-port 10 GbE NIC and RoCE SR
    * EC2N (Fiber) PCIe3 2-port 10 GbE NIC and RoCE SR
    * EL40 (Fiber) PCIe3 LP 2-port 10 GbE NIC and RoCE SR
    *
    * The fix needs to be applied to both the VIO Server with the
    * above adapter as well as all VIO Client LPARs sharing
    * this ethernet adapter using SEA.
    **************************************************************
    * RECOMMENDATION:
    * Install APAR IV82694.
    * Prior to fix availability, an interim fix is available from
    * either
    * ftp://aix.software.ibm.com/aix/ifixes/iv82694/
    * https://aix.software.ibm.com/aix/ifixes/iv82694/
    * Installation of the ifix requires a reboot.
    **************************************************************
  

Local fix

Problem summary

•   **************************************************************
    * USERS AFFECTED:
    * Systems running the AIX 6100-09 Technology Level
    * or VIOS 2.2.4.x with bos.net.tcp.client between
    * and including the 6.1.9.100 and 6.1.9.102 levels.
    **************************************************************
    * PROBLEM DESCRIPTION:
    * A VIO Server running 2.2.4 and using a 10Gb PCIe adapter
    * may crash or have other unexpected issues during normal
    * network communications.
    *
    * This happens when chksum_offload is enabled on the
    * VIO server and client LPARs, which is typically done to
    * enable large_send.
    *
    * It is recommended that every system running VIOS 2.2.4
    * with a 10GB PCIe adapter install this fix.
    *
    * The specific adapters that we have seen this problem with
    * are:
    * EC27 PCIe2 LP 2-Port 10GbE RoCE SFP+ Adapter
    * EC28 PCIe2 2-Port 10 GbE RoCE SFP+ adapter
    * EC37 (Copper) PCIe3 2-port 10 GbE NIC and RoCE SFP+ Copper
    * EC38 (Copper) PCIe3 LP 2-port 10 GbE NIC and RoCE SFP+
    Copper
    * EL3X (Copper) PCIe3 LP 2-Port 10 GbE NIC and RoCE SFP+
    Copper
    * EC2M (Fiber) PCIe3 2-port 10 GbE NIC and RoCE SR
    * EC2N (Fiber) PCIe3 2-port 10 GbE NIC and RoCE SR
    * EL40 (Fiber) PCIe3 LP 2-port 10 GbE NIC and RoCE SR
    *
    * The fix needs to be applied to both the VIO Server with the
    * above adapter as well as all VIO Client LPARs sharing
    * this ethernet adapter using SEA.
    **************************************************************
    * RECOMMENDATION:
    * Install APAR IV82694.
    * Prior to fix availability, an interim fix is available from
    * either
    * ftp://aix.software.ibm.com/aix/ifixes/iv82694/
    * https://aix.software.ibm.com/aix/ifixes/iv82694/
    * Installation of the ifix requires a reboot.
    **************************************************************
  

Problem conclusion

• Code modified to not set M_LARGESEND flag on the
  receive path.  So even if the same memory is used to
  send data back to the sender, it will not lead to a crash.
  

Temporary fix

•   *********
    * HIPER *
    *********
  

Comments

• 6100-09 - use AIX APAR IV82694
  6100-09 - use AIX APAR IV82694
  7100-04 - use AIX APAR IV86856
  7100-04 - use AIX APAR IV86856
  7200-01 - use AIX APAR IV86941
  

APAR Information

• APAR is sysrouted FROM one or more of the following:

• APAR is sysrouted TO one or more of the following:

  IV86856 IV86941 IV86960 IV87017 IV87162 IV87165 U875079

Fix information

• Fixed component name

  VIRTUAL I/O SER

• Fixed component ID

  5765G3400

Applicable component levels