Re: Regression :-) Re: [GIT PULL RESEND] x86/jumpmplabel changes for v3.12-rc1

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Wed, Sep 11, 2013 at 01:52:37PM -0400, Steven Rostedt wrote:
> On Wed, 11 Sep 2013 13:25:52 -0400
> Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> wrote:
> 
> 
> > commit 97ce2c88f9ad42e3c60a9beb9fca87abf3639faa
> > Author: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
> > Date:   Wed Oct 12 16:17:54 2011 -0700
> > 
> >     jump-label: initialize jump-label subsystem much earlier
> >     
> >     Initialize jump_labels much, much earlier, so they're available for use
> >     during system setup.
> >     
> >     Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
> >     Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
> > 
> > 
> > implies that yes.
> 
> Unfortunately it does not. All that patch did was move
> jump_label_init() up more. If anything, it implies "no".
> 
> The question is, can we enable jump_labels before jump_label_init()?
> 
> Note, we may still be able to (as it seems to work), the thing is, the
> only thing that static_key_slow_inc() does is to tell jump_label_init()
> to enable it. Before jump_label_init() is called, nothing has changed.
> No code modification, all users of paravirt_ticketlocks_enabled are
> still off.

<confused>

I am thins would still work:


47 static __always_inline void arch_spin_unlock(arch_spinlock_t *lock)             
148 {                                                                               
149         if (TICKET_SLOWPATH_FLAG &&                                             
150             static_key_false(&paravirt_ticketlocks_enabled)) {                  

(from arch/x86/include/asm/spinlock.h) as the static_key_false
would check the key->enabled. Which had been incremented?

Granted there are no patching done yet, but that should still allow
this code path to be taken?