+++ This bug was initially created as a clone of Bug #1344077 +++ Description of problem: Although rcu no callbacks is useful for the majority of -rt workloads. There are cases where this feature can cause bad impact in the -rt performance. For example, if the rcuc/ thread needs to wake up many offload threads, for instance in a large box with dozens CPUs, this can cause an interface that can be higher than the 150 microseconds we guarantee in the -rt products. As we can enable the rcu no callbacks via kernel command line, we still can have the same behavior of CONFIG_RCU_NOCB_CPU_ALL=y by using the rcu_nocbs=<allcpus>. So it is safe to turn it off. In fact, this gives us more flexibility than have all CPUs rcu_nocbs by default
This change in kernel-rt-3.10.0-439.rt56.327.el7
RHEL-RT 7.3 disables CONFIG_RCU_NOCB_CPU_ALL=y. The MRG RT kernel inherited this config in the MRG-RT kernel based on RHEL-RT 7.3-zstream #2. The current MRG-RT kernel config is: CONFIG_RCU_NOCB_CPU=y CONFIG_RCU_NOCB_CPU_NONE=y # CONFIG_RCU_NOCB_CPU_ZERO is not set # CONFIG_RCU_NOCB_CPU_ALL is not set The fix was released with the following ERRATA: https://errata.devel.redhat.com/advisory/25966
A release has been made for this BZ, so I am closing it currentrelease.