On Mon, 2019-09-23 at 12:58 -0700, John Hubbard wrote:
> 
> CPU 0                            CPU 1
> ------                         --------------
>                                READ(pte) (re-ordered at run time)
>                                atomic_inc(val) (no run-time memory barrier!)
>                            
> pmd_clear(pte)
> if (val)
>     run_on_all_cpus(): IPI
>                                local_irq_disable() (also not a mem barrier)
> 
>                                if(pte)
>                                   walk page tables

Let me see if I can understand,
On most patches, it would be:

CPU 0                            CPU 1
------				--------------
				ptep = __find_linux_pte  
				(re-ordered at run time)
				atomic_inc(val) 
pmd_clear(pte)
smp_mb()
if (val)
    run_on_all_cpus(): IPI
                               local_irq_disable() 

                               if(ptep)
                                  pte = *ptep;

Is that what you meant?