Re: + mm-revert-thp-make-madv_hugepage-check-for-mm-def_flags.patch added to -mm tree

[Gerald Schaefer <gerald.schaefer@de.ibm.com>]

On Wed, 26 Feb 2014 19:06:03 +0100
Oleg Nesterov <oleg@redhat.com> wrote:

> On 02/26, Alex Thorlton wrote:
> >
> > + * MADV_HUGEPAGE after PRCTL_THP_DISABLE is broken on s390 because
> > + * qemu blindly does madvise(MADV_HUGEPAGE) after s390_enable_sie().
> > + */
> > +#ifdef CONFIG_S390
> > +               if (mm_has_pgste(vma->vm_mm))
> >                         return -EINVAL;
> > +#endif
> 
> The comment is not really right...
> 
> And personally I think that
> 
> 	@@ -504,6 +504,9 @@ static int gmap_connect_pgtable(unsigned long address, unsigned long segment,
> 		if (!pmd_present(*pmd) &&
> 		    __pte_alloc(mm, vma, pmd, vmaddr))
> 			return -ENOMEM;
> 	+       /* large pmds cannot yet be handled */
> 	+       if (pmd_large(*pmd))
> 	+               return -EFAULT;
> 
> change still makes sense, so that we can simply revert this s390-
> specific hack in hugepage_madvise().

Yes, agreed.

> 
> I'd suggest the patch below on top of your changes, but I won't argue.
> 
> It would be nice to also change thp_split_mm() to not not play with
> mm->def_flags, but I am not sure if we can do this.

Hmm, I'm also wondering about this. Basically, we only need VM_NOHUGEPAGE
in vma->vm_flags, which is done for all existing vmas in thp_split_mm().
But if there should be new vmas created afterwards, it would still be
necessary to also have VM_NOHUGEPAGE in mm->def_flags, because the
vm_flags for new vmas will be set via OR of mm->def_flags, e.g. in
do_brk() and do_mmap_pgoff().

I guess the question is if new vmas can be created for the qemu/kvm host
process?

Anyway, this would then have to be a separate patch, to keep the
"revertability" of this hack.

> 
> Oleg.
> ---
> 
> Subject: [PATCH] s390: make sure MADV_HUGEPAGE fails after s390_enable_sie()
> 
> As Christian pointed out, the recent 'Revert "thp: make MADV_HUGEPAGE
> check for mm->def_flags"' breaks qemu, it does QEMU_MADV_HUGEPAGE for
> all kvm pages but this doesn't work after s390_enable_sie/thp_split_mm.
> 
> Reported-by: Christian Borntraeger <borntraeger@de.ibm.com>
> Suggested-by: Christian Borntraeger <borntraeger@de.ibm.com>
> Signed-off-by: Oleg Nesterov <oleg@redhat.com>
> 
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index a4310a5..0e08d92 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -1970,11 +1970,22 @@ int hugepage_madvise(struct vm_area_struct *vma,
>  {
>  	switch (advice) {
>  	case MADV_HUGEPAGE:
> +#ifdef CONFIG_S390
> +		/*
> +		 * MADV_HUGEPAGE is broken after s390_enable_sie(), qemu
> +		 * blindly does madvise(MADV_HUGEPAGE) for for all kvm pages
> +		 * and expects it must fail on s390. Avoid a possible SIGSEGV
> +		 * until qemu is changed.
> +		 */
> +		if (mm_has_pgste(vma->vm_mm))
> +			return -EINVAL;
> +#endif
>  		/*
>  		 * Be somewhat over-protective like KSM for now!
>  		 */
>  		if (*vm_flags & (VM_HUGEPAGE | VM_NO_THP))
>  			return -EINVAL;
> +
>  		*vm_flags &= ~VM_NOHUGEPAGE;
>  		*vm_flags |= VM_HUGEPAGE;
>  		/*
>