On Thu, Feb 25, 2016 at 4:02 PM, Marty McFadden <mcfadd...@llnl.gov> wrote: > Provides a new ioctl interface through /dev/cpu/msr_batch. > > This implementation will cause an Inter Processor Interrupt to be sent > to each destination processor and will wait until all processors have > finished processing their respective batch of MSR operations before > returning.
This comes up periodicially, and it needs to be benchmarked against the sane way to do it in userspace: For each CPU: sched_setaffinity to bind to that CPU, and then a bunch of conventional MSR accesses. Optionally do all the CPUs at once using lots of threads. Please do this benchmark on a kernel that is *not* using nohz_full. The nohz_full implementaion in current kernels has abysmal syscall performance, but that's improving rapidly. --Andy
