[LU-8925] kernel update [RHEL7.3 3.10.0-514.2.2.el7]

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.10.0
• Severity: 3
• Assignee: Bob Glossman (Inactive)
• Reporter: Bob Glossman (Inactive)

Description

This update fixes the following bugs:

• Previously, using a floating IP address inside a network namespace in some
  cases failed with the following error message: "bind: Cannot assign requested
  address". With this update, the kernel respects setting of the
  net.ipv4.ip_nonlocal_bind parameter to 1 in namespaces, and the floating IP
  address is now assigned as expected. (BZ#1388591)

• In the OpenStack Platform (OSP) environment with multiple guests on the same
  network, "no route to host" errors occurred when attempting to ping all the
  guests from a single virtual machine. With this update, the deffering policy of
  the Open vSwitch data path has been fixed, and the described error no longer
  occurs. (BZ#1388592)

• Red Hat Enterprise Linux in some cases failed to boot on platforms due to High
  Performance Event Timer (HPET) clock access contention. On systems with a Time
  Stamp Counter (TSC) clock and an HPET, the HPET is used to calibrate the TSC. On
  some large CPU count systems, this in some cases led to access contention on
  HPET, which caused a system lockup followed by Nonmaskable Interrupt (NMI)
  softlockup watchdog. This update reduces HPET counter read contention, thus
  fixing this bug. As a result, Red Hat Enterprise Linux now boots as expected on
  affected platforms. (BZ#1388597)

• Previously, when two NFS shares with different security settings were mounted,
  the I/O operations to the kerberos-authenticated mount caused the
  RPC_CRED_KEY_EXPIRE_SOON parameter to be set, but the parameter was not unset
  when performing the I/O operations on the sec=sys mount. Consequently, writes to
  both NFS shares had the same parameters, regardless of their security settings.
  This update fixes this problem by moving the NO_CRKEY_TIMEOUT parameter to the
  auth->au_flags field. As a result, NFS shares with different security settings
  are now handled as expected. (BZ#1388604)

• Due to an incorrect value in the virtual time base (VTB) register, the kernel
  of an IBM POWER8 KVM guest in some cases incorrectly detected a soft lockup
  condition, logged a call trace, and reported the following error: "NMI watchdog:
  BUG: soft lockup - CPU#x stuck for XXs". This update applies a set of patches to
  fix these problems. As a result, the soft lockups and error messages no longer
  occur in the described situation. (BZ#1388997)

• Previously, a packet loss occurred on transmit when Red Hat Enterprise Linux
  had complex network configuration and was running as a guest on a Xen
  hypervisor. An upstream patch has been backported that fix the xen-netfront
  driver to avoid packet loss when ethernet header crosses page boundary. As a
  result, the packet loss no longer occurs in the described scenario. (BZ#1390257)

• After migration to Red Hat Enterprise Linux 7.3, the Unified Extensible
  Firmware Interface (UEFI) boot failed due to a bug that prevented the kernel
  from booting through the PE/COFF image entry, which is used by the Extensible
  Firmware Interface (EFI) ipxe boot loader. This update fixes the alignment
  parameter on the call to the efi_low_alloc() function, and the kernel now boots
  as expected. (BZ#1392044)