I was wondering why the Linux kernel was spending 21ms starting up each secondary CPU when we boot it virtualized with qemu -smp >= 2.
It turns out the time was lost in check_tsc_sync_target(). This is easily avoided by using `tsc=reliable' on the command line, saving 63ms when starting a guest with -smp 4 for example. Is this option safe? If so, why don't we enable it automatically in the kernel when we detect that we are a KVM guest? Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com virt-builder quickly builds VMs from scratch http://libguestfs.org/virt-builder.1.html
