x86/apic related update:

  - A single commit which reduces cacheline misses in
    __x2apic_send_IPI_mask() significantly by converting
    x86_cpu_to_logical_apicid() to an array instead of using per CPU
    storage. This reduces the cost for a full broadcast on a dual socket
    system with 256 CPUs from 33 down to 11 microseconds.