| Description: | The kernel-rt packages contain the Linux kernel, the core of any Linux operating system.
A buffer overflow flaw was found in the way the Linux kernel's Intel AES-NI instructions optimized version of the RFC4106 GCM mode decryption functionality handled fragmented packets. A remote attacker could use this flaw to crash, or potentially escalate their privileges on, a system over a connection with an active AEC-GCM mode IPSec security association. (CVE-2015-3331, Important)
The kernel-rt packages have been upgraded to version 3.10.0-229.4.1, which provides a number of bug fixes and enhancements over the previous version, including:
Audit subsystem not resolving path name on directory watches audit watches do not track correctly after a rename auditctl output is changed in RHEL 7 megaraid_sas: non-booting system with intel_iommu=on kernel parameter GFS2: kernel NULL pointer dereference in gfs2_inplace_reserve Crypto adapter cannot be brought online - affect all HW crypto/seqiv.c: wrong check of return code from crypto_rng_get_bytes Backport crypto: sha256_ssse3 - also test for BMI2 Null pointer at team_handle_frame+0x62/0x100 [team] AES CTR x86_64 "by8" AVX optimization Intel RDSEED - Fix for entropy counting Intel SHA1 multi-buffer crypto implementation Intel SHA1 AVX2 optimization support mlx4_en: HW timestamp ends up in error queue of socket which does not have SO_TIMESTAMPING enabled
(BZ#1209963)
This update also fixes the following bugs:
Prior to this update, heavy lock contention occurred on systems with greater than 32 cores when large numbers of tasks went idle simultaneously. Consequently, all the idle CPUs attempted to acquire the run-queue (rq) lock of a CPU with extra tasks in order to pull those run-able tasks. This increased scheduler latency due to the lock contention. Instead of each idle CPU attempting to acquire the run-queue lock, now each idle CPU will send an IPI to let the overloaded CPU select one core to pull tasks from it. The result is less spin-lock contention on the rq lock and produces improved scheduler response time. (BZ#1210924)
The CONFIG_NO_HZ logic enabled/disabled the timer tick every time a CPU went into an idle state. This timer tick manipulation caused the system performance (throughput) to suffer. The CONFIG_NO_HZ configuration setting is now turned off by default, which increases the throughput due to the lower idle overhead while allowing system administrators to enable it selectively in their environment. (BZ#1210597)
All kernel-rt users are advised to upgrade to these updated packages, which correct these issues and add these enhancements. The system must be rebooted for this update to take effect.
|