[PATCH] arm64: mm: assign PXN permission in trans_pgd

[Catalin Marinas]

On Thu, Nov 11, 2021 at 05:13:02PM +0100, Ard Biesheuvel wrote:
> On Thu, 11 Nov 2021 at 17:09, Catalin Marinas <catalin.marinas at arm.com> wrote:
> > On Thu, Nov 11, 2021 at 02:03:16PM +0800, Pingfan Liu wrote:
> > > diff --git a/arch/arm64/mm/trans_pgd.c b/arch/arm64/mm/trans_pgd.c
> > > index d7da8ca40d2e..7b43d179e94d 100644
> > > --- a/arch/arm64/mm/trans_pgd.c
> > > +++ b/arch/arm64/mm/trans_pgd.c
> > > @@ -67,7 +67,7 @@ static int copy_pte(struct trans_pgd_info *info, pmd_t *dst_pmdp,
> > >       dst_ptep = trans_alloc(info);
> > >       if (!dst_ptep)
> > >               return -ENOMEM;
> > > -     pmd_populate_kernel(NULL, dst_pmdp, dst_ptep);
> > > +     pmd_populate_kernel(&init_mm, dst_pmdp, dst_ptep);
> > >       dst_ptep = pte_offset_kernel(dst_pmdp, start);
> >
> > Commit 5de59884ac0e ("arm64: trans_pgd: pass NULL instead of init_mm to
> > *_populate functions") actually replaced &init_mm with NULL. Commit
> > 59511cfd08f3 ("arm64: mm: use XN table mapping attributes for
> > user/kernel mappings") introduced the VM_BUG_ON.
> >
> > So we either revert Pasha's trans_pgd commit (as per your patch) or we
> > fix Ard's commit to make it aware of a NULL mm and either skip the table
> > UXN altogether or treat it as init_mm.
> >
> 
> That code in pmd_populate_kernel() does not actually *use* the mm
> argument at all, it just BUGs if its value != init_mm. So we can just
> relax the BUG() to disregard mm ==  NULL.

We also have pud_populate() that ends up setting PUD_TABLE_PXN when mm
is NULL. Fortunately we don't execute anything from those mappings, they
are just a copy of the linear map.

BTW, in line with PAGE_KERNEL, pud_populate() could set both UXN and PXN
for init_mm mappings.

-- 
Catalin