Re: [PATCH 0/2] mm->def_flags cleanups (Was: Change khugepaged to respect MMF_THP_DISABLE flag)

[Oleg Nesterov <oleg@redhat.com>]

On 01/22, Alex Thorlton wrote:
>
> At a glance, without testing, it looks like a good idea to me.  By
> using def_flags, we leverage functionality that's already in place to
> achieve the same result.  We don't need to add any new checks into the
> fault path or into khugepaged, since we're just leveraging the
> VM_HUGEPAGE/NOHUGEPAGE flag, which we already check for.  We also get
> the behavior that you suggested (madvise is still respected, even with
> the new THP disable prctl set), for free with this method.

Yes, exactly.

> I like the idea, but I think that it should probably be a separate
> change from the other few cleanups that you proposed along with it,

Yes, sure, that is why I sent them separately,

> since
> they're somewhat unrelated to this particular issue.  Do you agree?

Not really. Note that without 1/2 VM_NOHUGEPAGE won't survive after
exec. And without 2/2 madvise(MADV_HUGEPAGE) won't work after
PR_SET_THP_DISABLE.

But again, I think that these 2 simple cleanups make sense even without
PR_SET_THP_DISABLE.

> > diff --git a/kernel/sys.c b/kernel/sys.c
> > index ac1842e..eb8b0fc 100644
> > --- a/kernel/sys.c
> > +++ b/kernel/sys.c
> > @@ -2029,6 +2029,19 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
> >  		if (arg2 || arg3 || arg4 || arg5)
> >  			return -EINVAL;
> >  		return current->no_new_privs ? 1 : 0;
> > +	case PR_SET_THP_DISABLE:
> > +	case PR_GET_THP_DISABLE:
> > +		down_write(&me->mm->mmap_sem);
> > +		if (option == PR_SET_THP_DISABLE) {
> > +			if (arg2)
> > +				me->mm->def_flags |= VM_NOHUGEPAGE;
> > +			else
> > +				me->mm->def_flags &= ~VM_NOHUGEPAGE;
> > +		} else {
> > +			error = !!(me->mm->flags && VM_NOHUGEPAGE);
>
> Should be:
>
> error = !!(me->mm->def_flags && VM_NOHUGEPAGE);

No, we need to return 1 if this bit is set ;)

Oleg.