On Sun, 11 Mar 2012 18:24:03 -0400, Christoffer Dall <c.dall at virtualopensystems.com> wrote: > On Thu, Feb 23, 2012 at 11:00 PM, Rusty Russell <rusty at rustcorp.com.au> wrote: > > On Thu, 23 Feb 2012 02:32:39 -0500, Christoffer Dall <c.dall at virtualopensystems.com> wrote: > >> + ? ? /* > >> + ? ? ?* Allocate stack pages for Hypervisor-mode > >> + ? ? ?*/ > >> + ? ? for_each_possible_cpu(cpu) > >> + ? ? ? ? ? ? per_cpu(kvm_arm_hyp_stack_page, cpu) = NULL; > > > > This is weird; we can't call this init function multiple times without > > reloading the module. > > > > what you're saying is that static DEFINE_PER_CPU will still be > initialized to zero? I don't actually have experience with this, but I > assumed that the data..percpu section would not necessarily be zeroed > out when loading a module... Now I tried to look, but I couldn't > convince myself. Can you enlighten me? Absolutely and definitely. It's placed in a special section, and from there copied into the per-cpu sections by the module loader but still follows the C semantics of being zero-initialized. Cheers, Rusty. -- How could I marry someone with more hair than me? http://baldalex.org
