Red Hat Training

A Red Hat training course is available for Red Hat Enterprise Linux

Chapter 56. Kernel

Cache information is missing in sysfs if firmware does not support ACPI PPTT

The kernel-alt package has been updated to use the Advanced Configuration and Power Interface Processor Properties Topology Table (ACPI PPTT) to populate CPU topology including the CPU's cache information. Consequently, on systems whose firmware does not support ACPI PPTT, the /sys/devices/system/cpu/cpu0/cache file does not contain the cache information. To work around this problem, check for updated firmware that includes ACPI PPTT support with your hardware vendor. (BZ#1615370)

PCI-passthrough of devices connected to PCIe slots is not possible with default settings of HPE ProLiant Gen8 and Gen9

Default settings of HPE ProLiant Gen8 and Gen9 systems disallow use of PCI-passthrough for devices connected to PCIe slots. Consequently, any attempt to pass through such devices fails with the following message in the kernel log: 
Device is ineligible for IOMMU domain attach due to platform RMRR requirement. Contact your platform vendor.
To work around this problem:
  • In case of HPE ProLiant Gen8, reconfigure mentioned system settings with the conrep tool provided by HPE.
  • In case of HPE ProLiant Gen9, update system firmware or NICs firmware depending on type of used NICs.
For more details about the workaround, see https://support.hpe.com/hpsc/doc/public/display?docId=emr_na-c04781229. (BZ#1615210)

Attaching a non-RoCE device to RXE driver no longer causes a kernel to panic

When a user created a Soft RDMA Over Converged Ethernet (Soft RoCE) interface and attached a non-RoCE device, certain issues were observed in the RXE driver. As a consequence, a kernel panicked when rebooting or shutting down a host. With this update, disabling the Soft RoCE interface before rebooting or shutting down a host fixes the issue. As a result, the host no longer panics in the described scenario. (BZ#1520302)

Enabling the BCC packages for the 64-bit AMD and Intel architectures only

The BPF Compiler Collection (BCC) library and the pcp-pmda-bcc plugins use the bpf() system call, which is enabled only on the 64-bit AMD and Intel CPU architectures. As a result, Red Hat Enterprise Linux 7 only supports BCC and pcp-pmda-bcc for the 64-bit AMD and Intel CPU architectures. (BZ#1633185)

Branch prediction of ternary operators no longer causes a system panic

Previously, the branch prediction of ternary operators caused that the compiler incorrectly called the blk_queue_nonrot() function before checking the mddev->queue structure. As a consequence, the system panicked. With this update, checking mddev->queue and then calling blk_queue_nonrot() prevents the bug from appearing. As a result, the system no longer panics in the described scenario. (BZ#1627563)

RAID1 write-behind causes a kernel panic

Write-behind mode in the Redundant Array of Independent Disks Mode 1 (RAID1) virtualization technology uses the upper layer bio structures, which are freed immediately after the bio structures written to bottom layer disks come back. As a consequence, a kernel panic is triggered and the write-behind function cannot be used. (BZ#1632575)

The i40iw module does not load automatically on boot

Some i40e NICs do not support iWarp and the i40iw module does not fully support suspend and resume operations. Consequently, the i40iw module is not automatically loaded by default to ensure suspend and resume operations work properly. To work around this problem, edit the /lib/udev/rules.d/90-rdma-hw-modules.rules file to enable automated load of i40iw.
Also note that if there is another RDMA device installed with an i40e device on the same machine, the non-i40e RDMA device triggers the rdma service, which loads all enabled RDMA stack modules, including the i40iw module. (BZ#1622413)