[PATCH] arm64: correct modules range of kernel virtual memory layout

[Will Deacon]

On Mon, Aug 07, 2017 at 02:18:00PM +0100, Ard Biesheuvel wrote:
> On 7 August 2017 at 14:16, Will Deacon <will.deacon at arm.com> wrote:
> > On Mon, Aug 07, 2017 at 07:04:46PM +0800, Miles Chen wrote:
> >> The commit f80fb3a3d508 ("arm64: add support for kernel ASLR")
> >> moved module virtual address to
> >> [module_alloc_base, module_alloc_base + MODULES_VSIZE).
> >>
> >> Display module information of the virtual kernel
> >> memory layout by using module_alloc_base.
> >>
> >> testing output:
> >> 1) Current implementation:
> >> Virtual kernel memory layout:
> >>       modules : 0xffffff8000000000 - 0xffffff8008000000   (   128 MB)
> >> 2) this patch + KASLR:
> >> Virtual kernel memory layout:
> >>       modules : 0xffffff8000560000 - 0xffffff8008560000   (   128 MB)
> >> 3) this patch + KASLR and a dummy seed:
> >> Virtual kernel memory layout:
> >>       modules : 0xffffffa7df637000 - 0xffffffa7e7637000   (   128 MB)
> >>
> >> Signed-off-by: Miles Chen <miles.chen at mediatek.com>
> >> ---
> >>  arch/arm64/mm/init.c | 5 +++--
> >>  1 file changed, 3 insertions(+), 2 deletions(-)
> >
> > Does this mean the modules code in our pt dumper is busted
> > (arch/arm64/mm/dump.c)? Also, what about KASAN, which uses these addresses
> > too (in kasan_init)? Should we just remove MODULES_VADDR and MODULES_END
> > altogether?
> >
> 
> I don't think we need this patch. The 'module' line simply prints the
> VA region that is reserved for modules. The fact that we end up
> putting them elsewhere when running randomized does not necessarily
> mean this line should reflect that.

I was more concerned by other users of MODULES_VADDR tbh, although I see
now that we don't randomize the module region if kasan is enabled. Still,
the kcore code adds the modules region as a separate area (distinct from
vmalloc) if MODULES_VADDR is defined, the page table dumping code uses
MODULES_VADDR to identify the module region and I think we'll get false
positives from is_vmalloc_or_module_addr, which again uses the static
region.

So, given that MODULES_VADDR never points at the module area, can't we get
rid of it?

Will