Getting RCU Further Out Of The Way

Although RCU has changed quite a bit to accommodate the needs of real-time workloads, it still has some bad habits from a real-time perspective. Ongoing work includes minimizing RCU's disturbance of other CPUs, including getting out of the way of changes that would help CPU hotplug avoid disturbing other CPUs.

All of this is goodness, but one major source of disturbance remains, namely invocation of the RCU callbacks. The need for this is real, in fact, it is pressing enough that some people actually maintain their own "JRCU" implementation. However, it would be better if this capability was available in mainline Linux and from the current RCU implementations in a normal distro kernel build. This talk will describe a way of doing that, hopefully with prototype available.

Topic Lead: Paul E. McKenney
Paul has been coding for almost four decades, more than half of that on parallel hardware, where his work has earned him a reputation among some as a flaming heretic. Over the past decade, Paul has been an IBM Distinguished Engineer at the IBM Linux Technology Center. Paul maintains the RCU implementation within the Linux kernel, where the variety of workloads present highly entertaining performance, scalability, real-time response, and energy-efficiency challenges. Prior to that, he worked on the DYNIX/ptx kernel at Sequent, and prior to that on packet-radio and Internet protocols (but long before it was polite to mention Internet at cocktail parties), system administration, business applications, and real-time systems. His hobbies include what passes for running at his age along with the usual house-wife-and-kids habit.