Re: [PATCH 00/33] Compile-time stack metadata validation

[Josh Poimboeuf]

On Fri, Feb 12, 2016 at 06:10:37PM +0100, Peter Zijlstra wrote:
> On Fri, Feb 12, 2016 at 08:45:43AM -0600, Josh Poimboeuf wrote:
> > On Fri, Feb 12, 2016 at 11:36:24AM +0100, Jiri Slaby wrote:
> > 
> > This seems like a real frame pointer bug caused by the following line in
> > arch/x86/include/asm/preempt.h:
> > 
> >   # define __preempt_schedule() asm ("call ___preempt_schedule")
> 
> The purpose there is that:
> 
> 	preempt_enable();
> 
> turns into:
> 
> 	decl	__percpu_prefix:__preempt_count
> 	jnz	1f:
> 	call	___preempt_schedule
> 1:
> 
> See arch/x86/include/asm/preempt.h:__preempt_count_dec_and_test()

Sorry, I'm kind of confused.  Do you mean that's what preempt_enable()
would turn into *without* the above define?

What I actually see in the listing is:

 	decl	__percpu_prefix:__preempt_count
 	je	1f:
	....
 1:
 	call	___preempt_schedule

So it puts the "call ___preempt_schedule" in the slow path.

I also don't see how that would be related to the use of the asm
statement in the __preempt_schedule() macro.  Doesn't the use of
unlikely() in preempt_enable() put the call in the slow path?

  #define preempt_enable() \
  do { \
	  barrier(); \
	  if (unlikely(preempt_count_dec_and_test())) \
		  preempt_schedule(); \
  } while (0)

Also, why is the thunk needed?  Any reason why preempt_enable() can't be
called directly from C?

-- 
Josh