|
This update fixes the following bugs:
In an environment with an IPsec tunnel between two Red Hat Enterprise Linux 7.4 nodes and a ccm module, the system previously panicked after disabling ipsec.service. This update moves the algorithm reference counting to the crypto_check_alg() function. As a result, removing ccm no longer causes a system panic. (BZ#1475668)
Due to a changed deallocating path in the cxgb4 driver, Dynamic Logical Partitioning (DLPAR) remove operation previously freed the interrupts of Upper-Layer Drivers (ULD) when unregistering cxgb4 or when shutting down the PCI handler. As a consequence, a kernel panic occurred and the LPAR was rebooted. The provided update refactors the shutdown path of ULD on cxgb4 by properly freeing and disabling interrupts on PCI remove handler. (BZ#1475671)
After injecting an Enhanced Error Handling (EEH) error, IBM Power RAID (IPR) adapter previously became offline or disappeared from the lspci command output and could not be seen or accessed with the iprconfig tool. The provided set of patches fixes the partial hotplug criterion and the incorrect argument passed to the eeh_rmv_device() function, so that the information about IPR is displayed correctly. (BZ#1475672)
Previously, the code in dst_ifdown placed a hold on the loopback device. Consequently, when removing network namespaces (netns), Daylight Saving Time (DST) garbage collection failed to clean DST on time, which held the loopback device (attempting to be unregistered) for a long time. In addition, netdev_wait_allrefs() emmited a pr_emerg warning message. This update resets the time interval so that DST cleanups occur in a more timely fashion, thus preventing the pr_emerg warning messages from appearing. (BZ#1475675)
Kernel version 3.10.0-498.el7 separated CPU and TSC frequency and introduced the x86_platform.calibrate_cpu function pointer which points by default to the native_calibrate_cpu() function. As a consequence, time synchronization bugs appeared on Red Hat Enterprise Linux 7.3 virtual machine running on ESXi causing a time offset shortly after boot. An upstream patch has been applied, which sets x86_platform.calibrate_cpu pointer on ESXi guests to the proper function, thus fixing this bug. (BZ#1479246)
A system having more than 128 CPUs could previously experience a crash during shutdown after the Intelligent Platform Management Interface (IPMI) service was stopped. The provided update fixes a race condition in the IPMI smi_timeout() function, allowing the system to shut down as expected. (BZ#1479761)
Allocating memory on node1 with the MPOL_PREFERRED policy did not work with Transparent Huge Pages (THP) enabled, but always fell back to the node0 local node. Consequently, workload performance for multinode systems was significantly impacted. The backported patch ensures MPOL_PREFERRED policy with non-local node is respected, and system performance is back to optimal. (BZ#1479763)
|