Red Hat Enterprise Linux 7.4 for IBM Power LE (POWER9) - Release Notes

Updated -

Red Hat Enterprise Linux 7.4 for IBM Power LE (POWER9) introduces Red Hat Enterprise Linux 7.4 user space with an updated kernel. The new kernel version 4.11 is provided by the kernel-alt packages. The offering is distributed with other updated packages but most of it is the standard Red Hat Enterprise Linux 7 Server RPMs.

Installation ISO images are available on the Customer Portal Downloads page.

For information about Red Hat Enterprise Linux 7.4 installation and user space, see the Release Notes, Installation Guide, and other Red Hat Enterprise Linux 7 documentation.

NOTE: Bare metal installations on IBM Power LE using an USB drive require you to specify the inst.stage2= boot option manually at the boot menu. See the Boot Options chapter in the Installation Guide for detailed information.

The following packages are being distributed with this release:

  • binutils
  • iproute
  • kernel-alt
  • libvirt
  • opal-prd
  • oprofile
  • papi and libpfm
  • qemu-kvm
  • rdma-core
  • redhat-release-server
  • systemtap

New Features and Updates

Platform Tools

  • OProfile now includes support for the IBM POWER9 processor. Note that the PM_RUN_INST_CMPL OProfile performance monitoring event cannot be setup and should not be used in this version of OProfile.
  • This update adds support for the IBM POWER9 performance monitoring hardware events to papi. It includes basic PAPI presets for events, such as instructions (PAPI_TOT_INS) or processor cycles (PAPI_TOT_CYC). Note that PAPI_L1_DCM preset does not currently work.
  • This version of libpfm includes support for the IBM POWER9 performance monitoring hardware events.
  • SystemTap includes backported compatibility fixes necessary for the kernel.

Kernel Configuration Changes

HW Enablement

  • DEVFREQ_GOV_SIMPLE_ONDEMAND (enabled)
  • GPIO IRQCHIP (enabled)
  • HID plantronic (disabled)
  • I2C sensors
    • JC42 (disabled)
      • NTC thermostat (enabled)
    • I2C MUX (enabled)
Networking Driver Support
  • Broadcom B44 driver (disabled)
  • Brocade BNA driver (disabled)
  • Calxeda driver(disabled)
  • IBM ethernet driver [ehea] (disabled)
  • Intel E1000 driver (disabled)
  • Mellanox driver [mlxsw] (disabled)
  • Netronoma driver [NFP] (disabled)
  • Qlogic [qla3xxx] driver (disabled)
  • SFC falcon driver (disabled)
  • Wireless (disabled)
    • WLAN (disabled)
    • Ath driver (disabled)
    • Ath10k driver (disabled)
    • Ath 9k driver (disabled)
    • Ath wil6210 (disabled)
    • Broadcom WLAN (disabled)
    • Broadcom brcm80211 (disabled)
    • Intel WLAN (disabled)
    • Intel iwlegacy (disabled)
    • Intel iwlwifi (disabled)
    • Marvell driver (disabled)
    • Marvell mwiflex (disabled)
    • Ralink WLAN driver (disabled)
    • Ralink rt2x00 driver (disabled)
    • Realtek driver (disabled)
    • Realtek rt1818x driver (disabled)
    • Realtek rtiwifi driver (disabled)
  • NVME driver + target driver (enabled)
  • ptp 1588 driver (disabled)
  • s390 HMC driver (disabled)
  • RTL8192e driver (disabled)
  • RTL8712u driver (disabled)
  • Serial UARTLITE driver (enabled)
  • USB LED trigger USBPORT (disabled)
  • USBIP driver (disabled)
  • Power Mgt Deubg + Adv Debug + Sleep Debug (enabled)

Core Kernel Support

  • Sched Imbalance patchset (enabled)
  • OPTprobes, kprobe on ftrace (enabled)
  • 64bit Aligned Access (disabled)
  • Arch Soft Dirty (enabled)
  • Arch MMAP Rnd Compat (disabled)
  • SWIOTLB (disabled)
  • Crypto: akcipher, rsa (enabled)
  • Compression:
    • Kernel gzip support (enabled)
    • Kernel XZ support (enabled)
  • Locking: Mutex spin on owner (enabled in debugging kernel)
  • Function Tracer (enabled)
  • Dynamic Ftrace (enabled)
  • Ftrace mcount record (enabled)
  • Common kernel Libaries
    • Rational (enabled)
    • Btree (enabled)
    • libfdt (enabled)
    • parman (disabled)
  • MM
    • NO_BOOTMEM (enabled)
    • MOVABLE NODE (enabled)
    • HMM (Hetrogenous Memory Management) (enabled)
      • HMM Mirrored (enabled)
      • Coherent Device Memory (CDM) (enabled)
      • Zone Device (enabled)
  • IMA (enabled)
  • YAMA (disabled)
Networking Stack Support
  • Compact Netlink Msg (disabled)
  • BPF_JIT (enabled)
  • DCCP (disabled)
  • CCIDS (disabled)
  • IPv6 NF target NPT (disabled)
  • Mac80211 (disabled)
Desktop/Graphic/GPU Support
  • DRM_DP_AUX_CHARDEV (enabled)
  • STK1160 video usb driver (disabled)
  • V412 BUF2_DMA_SG (enabled)
Storage Support
  • DAX (disabled)
  • NVDIMM + PFN + DAX (enabled)
  • SCSI
    • 3Ware 9xxx driver (disabled)
    • 3Ware sAS driver (disabled)
    • ARCMSR driver (disabled)
    • AIC79xx driver (disabled)
    • Broadcom Bnx2x driver (enabled)
    • Broadcom Bnx2 driver (disabled)
    • QED driver (disabled)
    • QEDI driver (disabled)
File Systems
  • BTRFS (disabled)
  • DLM (disabled)
  • GFS2 DLM locking support (disabled)
Virtualization and KVM Support
  • vhost [vsock] (disabled)
  • VMWare vsock (disabled)

Known Issues

Recovering from OOM situation fails due to incorrect function of OOM-killer

Recovering from an out-of-memory (OOM) situation does not work correctly on systems with large amounts of memory. Kernel's OOM-killer kills the process using the most memory and frees the memory to be used again. However, sometimes the OOM-killer does not wait long enough before killing a second process. Eventually, the OOM-killer kills all the processes on the system and logs this error:

Kernel panic - not syncing: Out of memory and no killable processes...

If this happens, the operating system must be rebooted. There is no available workaround. (BZ#1405748)

qemu-kvm does not support multiple threads per CPU core

The qemu-kvm emulator currently does not support more than one thread per CPU core on the POWER9 architecture with KVM. As a consequence, when the <topology threads='X'/> XML configuration is used with a value larger than 1, qemu-kvm fails to start the virtual machine with the following message:

qemu-kvm: Cannot support more than 1 threads on PPC with KVM

To work around this problem, do not specify more than one thread per core for the virtual machine. Note that when using one thread per CPU core, multiple CPU cores per virtual machine work as expected. (BZ#1450319)

Memory hot-unplug fails with PCI device passthrough

Removing memory from a running guest virtual machine, also known as memory hot-unplug, currently fails when a PCI device passthrough is configured on the guest, and the following error message is logged:

pseries-hotplug-mem: Memory indexed-count-remove failed, adding any removed LMBs

To work around this problem, edit the guest XML configuration file to add a memtune hard limit at least 1 gigabyte larger than the value of the <maxMemory> element. For example, if <maxMemory> is set to 20GB, set memtune hard limit to at least 21GB:

<maxMemory slots='16' unit='KiB'>20971520</maxMemory>
<memtune>
<hard_limit unit='KiB'>22020096</hard_limit>
</memtune>

(BZ#1490702)

Kdump saves the vmcore only if mpt3sas is blacklisted

When kdump kernel loads the mpt3sas driver, the kdump kernel crashes and fails to save the vmcore on certain POWER9 systems. To work around this problem, blacklist mpt3sas from the kdump kernel environment by appending the "module_blacklist=mpt3sas" string to the KDUMP_COMMANDLINE_APPEND variable in the /etc/sysconfig/kdump file:

KDUMP_COMMANDLINE_APPEND="irqpoll maxcpus=1 ... module_blacklist=mpt3sas"

Then restart the kdump service to pick up the changes to the configuration file by running the "systemctl restart" command as the root user:

 ~]# systemctl restart kdump.service

As a result, kdump is now able to save the vmcore on the POWER9 systems. (BZ#1496273)

Bug Fixes

Operating system no longer deadlocks if a CPU is hotplugged in parallel with nested blk-mq activity

When the online status of a CPU was changed in parallel with I/O activity involving a block device using the nested Multi-Queue Block IO Queueing Mechanism (blk-mq) infrastructure, Red Hat Enterprise Linux 7.4 deadlocked in the blk_mq_freeze_queue_wait() phase of the blk_mq_queue_reinit_work() function. Consequently, the installation failed on IBM Power VM configurations when all CPUs were not successfully enabled. This update fixes the deadlock, and the installation now proceeds as expected in the described scenario. (BZ#1466046)

kdump no longer hangs due to the attempts to read the memory from on-board devices

On the little-endian variants of IBM Power Systems hardware, the kdump mechanism became unresponsive because the kernel attempted to read the memory from on-board devices such as the GPU, and include it as a part of the vmcore. This update fixes kexec-tools to skip the on-board devices when attempting to read the memory during kdump. As a result, kdump now works correctly, the vmcore is saved to disk and the operating system reboots as expected. (BZ#1478049)

Installation of kernel updates is no longer slowed down due to depmod

On the IBM POWER architecture, the depmod utility did not correctly handle the alternate format of kernel modules symbols' versions used in Red Hat Enterprise Linux with the new kernel version provided by the kernel-alt package. The depmod utility reported problems of all symbols for all kernel modules and the weak-modules script parsed the depmod's output, which consequently slowed down the installation of kernel updates. The underlying source code in the kmod package has been fixed to make depmod correctly handle the alternate format of kernel modules symbols and to make the weak-modules script filter depmod's output for the modules only in the weak-modules working directories (weak-updates). As a result, the installation of kernel updates is no longer slowed down due to depmod. (BZ#1468305)

Comments