With Unbreakable Enterprise Kernel Release 5 Update 1, Oracle continues to deliver kernel modifications to enable support for 64-bit Arm (aarch64) architecture. These changes are built and tested against existing Arm hardware and provide support for Oracle Linux for Arm. Features described in this document are available for Arm insofar as the hardware is capable of supporting the feature that is described. Limitations and items beyond the scope of current development work for Arm are described in more detail in Unusable or Unavailable Features for Arm.

Where specific changes have been made for Arm architecture, these are usually noted through this document. Particular effort has gone into updating and improving DTrace features to include support for the 64-bit Arm (aarch64) architecture, see DTrace for more information.

The following notable core kernel features are implemented in UEK R5U1:

• Cgroup v2 CPU controller backported to support kABI

  The cgroup v2 CPU controller has been backported in this release. This change was made to pre-populate structures that could potentially impact the Kernel Application Binary Interface (kABI). The update includes changes to the code that handles cgroup resource usage statistics and improves performance when handling frequent reads where there are many cgroups that are not active.

• Improved scheduler scalability for fast path

  This release includes scheduler scalability improvements for fast path. In addition, a new scheduler feature, SIS_CORE, is introduced to improve performance for certain workloads, such as Oracle database OLTP (online transaction processing). To provide additional performance improvement for OLTP, you can optionally disable this feature by setting NO_SIS_CORE in /sys/kernel/debug/sched_features. However, note that disabling the SIS_CORE feature is recommended for OLTP workloads only.

The following notable DTrace features and fixes are implemented in UEK R5U1:

• [aarch64] END and ENDPROC markers added

  DTrace has been enhanced for use on aarch64 platforms by adding or moving END[ENDPROC] markers in relevant assembly files. This change increases the likelihood that DTrace can map kernel instructions to a symbol.

• [aarch64] DTrace support for SDT probes added

  DTrace has been enhanced to include support for the use of SDT probes on aarch64 platforms.

• [aarch64] DTrace support for FBT entry probes added

  DTrace has been enhanced to include support for the use of FBT entry probes on aarch64 platforms.

• [aarch64] DTrace support for FBT return probes added

  DTrace has been enhanced to include support for the use of FBT return probes on aarch64 platforms.

• [aarch64] ustack() implementation

  DTrace has been enhanced to include the implementation of ustack() on aarch64 platforms.

• SDT macros updated to accurately support 18 arguments

  The DTrace implementation for defining SDT probes in the kernel source code has been improved to accurately accept up to, and including, 18 arguments. Although all 18 arguments were supported previously, any arguments beyond 8 were truncated. This enhancement applies to both the x86_64 and aarch64 platforms.

• New -x noresolve option added

  DTrace has been enhanced to include the -x noresolve runtime option, which is used to prevent automatic resolution of user-space symbols and to assist with symbol resolution for very large, statically linked binaries, where symbols take a long time to resolve. The option can also be used to reduce overhead when DTrace resolves addresses to symbol names for large applications that contain many symbols. The -x noresolve option was previously introduced in the Oracle Solaris 11.3 release.

• New -x ctfpath option added

  DTrace has been enhanced to include a new -x ctfpath option, which enables the use of user-provided Compressed Type Format (CTF) archive data. The ctfpath option allows you to explicitly specify which CTF archives to use for the running kernel, thus enabling the archive to be shared and shipped to non-standard locations.

The following sections detail the most notable features that have been implemented for file systems in UEK R5U1:

The following notable memory management features are implemented in UEK R5U1:

• Updates to hugetlbfs to resolve several race condition issues

  Several updates were applied to the hugetlbfs code to resolve race issues. These include a fix for an issue that resulted when a hugetlbfs file was truncated while another task was still performing a page fault process. This issue could cause a reserve map entry to be created before the cached page is deleted, resulting in the reserved page leaking indefinitely. The code was updated to prevent faults to a file while it is being truncated.

Remote Direct Memory Access (RDMA) is a feature that allows direct memory access between two systems that are connected by a network. RDMA facilitates high-throughput and low-latency networking in clusters.

Unbreakable Enterprise Kernel Release 5 Update 1 includes RDMA features that are provided in the upstream kernel, with the addition of Ksplice and DTrace functionality and Oracle's own RDMA features, which includes support for RDS and Shared-PD.

Oracle provides support for RDMA on InfiniBand on the following Oracle-branded HCAs as delivered inside Oracle Engineered Systems:

• Sun InfiniBand Dual Port 4x QDR Host Channel Adapters M2

• Oracle Dual Port QDR InfiniBand Adapter M3

Notable changes to RDMA implementation in UEK R5U1 include:

• resilient_rdmaip module fixes and improvements

  The Active-Active bonding code used by the resilient_rdmaip module, was updated to handle an issue that could cause a system panic if IPv6 addressing was disabled on the interface. Other improvements to logging and also to the module description information were also applied. A fix was also applied to include a check to prevent failover across ports within the failover group zero, which contains ports not intended for failover.

The following notable security features are implemented in Unbreakable Enterprise Kernel Release 5 Update 1:

• Secure boot update

  UEFI Secure Boot certificates used to sign the kernel and grub2 bootloader and which are verified using shim, have been updated to avoid expiration. This update can affect continued UEFI Secure boot functionality if the boot components involved are not all using the same valid Extended Validation (EV) certificate. For more information, see Oracle^® Linux: UEFI Secure Boot Update Notices.

The following notable storage features are implemented in Unbreakable Enterprise Kernel Release 5 Update 1:

• libnvdimm subsystem updated for PMEM and DAX

  The libnvdimm kernel subsystem, which is responsible for the detection, configuration, and management of Non-Volatile Dual Inline Memory Modules (NVDIMMs) is updated in UEK R5U1 to take advantage a large number of upstream patches, bug fixes and backports. Notably, these include fixes to /proc/smaps to reflect the actual PMEM page size and some work to improve Address Range Scrub (ARS). Also included as a notable improvement, is a fix that includes a new hand-shake mechanism that causes the fs thread to wait until DMA completion before waking up to update the fs metadata.

• NVMe updates

  General modifications to NVMe code are included in this update. Build configuration options were synchronized to ensure equivalence and compatibility in aarch64 environments. A patch was also applied to resolve an issue with NVMe hotplug/unplug functionality when used with a KVM guest.

The following notable virtualization features are implemented in Unbreakable Enterprise Kernel Release 5 Update 1:

• Backported bug fixes and features for KVM, Xen and Hyper-V

  Many bug fixes and virtualization features in the upstream 4.17, 4.18 and 4.19 kernels are backported into UEK R5U1. These patches offer better stability and resolve bugs and performance issues in KVM, Xen and Hyper-V. Fixes also include specific updates to better accommodate and support KVM on 64-bit Arm platforms.

• Added the net_failover driver module

  The addition of the net_failover driver module in this update release helps to enable virtio_net to act as a standby for a passthru device. This is important in cloud environments where implementing an accelerated data path to virtio-net enabled virtual machines should be possible with minimal guest user space modification. In particular, the net_failover module provides an automated failover mechanism for paravirtual drivers to manage network devices in a generic failover infrastructure.

• KVM Guest startup issue fixed for system where Transparent Huge Pages (THP) are enabled

  An issue that caused a large-memory KVM guest to hang on startup or restart when using Transparent Huge Pages (THP), has been resolved in this update. The issue resulted from a memory mapping heuristic that broke when a local node ran out of memory.

  To prevent this issue from occurring, where a kernel does not have a fix applied, run the following command on the host system before you attempt to start the guest:

  # echo defer > /sys/kernel/mm/transparent_hugepage/defrag

  A fix for this issue is included in this update release and systems using this kernel version, or later, do not require the workaround.