Re: [PATCH 06/12] x86/mm: Enable and use the arch_pgd_init_late() method

[Ingo Molnar <mingo@kernel.org>]

* Ingo Molnar <mingo@kernel.org> wrote:

> * Oleg Nesterov <oleg@redhat.com> wrote:
> 
> > On 06/11, Ingo Molnar wrote:
> > >
> > > +void arch_pgd_init_late(struct mm_struct *mm, pgd_t *pgd)
> > > +{
> > > +	/*
> > > +	 * This is called after a new MM has been made visible
> > > +	 * in fork() or exec().
> > > +	 *
> > > +	 * This barrier makes sure the MM is visible to new RCU
> > > +	 * walkers before we initialize it, so that we don't miss
> > > +	 * updates:
> > > +	 */
> > > +	smp_wmb();
> > 
> > I can't understand the comment and the barrier...
> > 
> > Afaics, we need to ensure that:
> > 
> > > +			if (pgd_val(*pgd_src))
> > > +				WRITE_ONCE(*pgd_dst, *pgd_src);
> > 
> > either we notice the recent update of this PGD, or (say) the subsequent
> > sync_global_pgds() can miss the child.
> > 
> > How the write barrier can help?
> 
> So the real thing this pairs with is the earlier:
> 
> 	tsk->mm = mm;
> 
> plus the linking of the new task in the task list.
> 
> _that_ write must become visible to others before we do the (conditional) copy 
> ourselves.
> 
> Granted, it happens quite a bit earlier, and the task linking's use of locking 
> is a natural barrier - but since this is lockless I didn't want to leave a 
> silent assumption in.
> 
> Perhaps remove the barrier and just leave a comment in that describes the 
> assumption on task-linking being a full barrier?

Ah, there's another detail I forgot. This might handle the fork case, but in 
exec() we have:

        tsk->mm = mm;
        arch_pgd_init_late(mm);

and since the task is already linked, here we need the barrier.

So how about I improve the comment to:

        /*
         * This function is called after a new MM has been made visible
         * in fork() or exec() via:
         *
         *   tsk->mm = mm;
         *
         * This barrier makes sure the MM is visible to new RCU
         * walkers before we initialize the pagetables below, so that
         * we don't miss updates:
         */
        smp_wmb();

and leave the barrier there?

Thanks,

	Ingo