Re: [PATCH] x86/kaslr: try process e820 entries if can not get suitable regions from efi

[Arvind Sankar <nivedita@alum.mit.edu>]

On Tue, Jan 05, 2021 at 09:54:52AM +0100, Ard Biesheuvel wrote:
> (cc Arvind)
> 
> On Tue, 5 Jan 2021 at 09:54, Lin Feng <linfeng23@huawei.com> wrote:
> >
> > On efi64 x86_64 system, the EFI_CONVENTIONAL_MEMORY regions will not
> > be mapped when making EFI runtime calls. So kexec-tools can not get
> > these from /sys/firmware/efi/runtime-map. Then compressed boot os
> > can not get suitable regions in process_efi_entries and print debug
> > message as follow:
> >         Physical KASLR disabled: no suitable memory region!
> > To enable physical kaslr with kexec, call process_e820_entries when
> > no suitable regions in efi memmaps.
> >
> > Signed-off-by: Lin Feng <linfeng23@huawei.com>
> >
> > ---
> >
> > I find a regular of Kernel code and data placement with kexec. It
> > seems unsafe. The reason is showed above.
> >
> > I'm not familiar with efi firmware. I wonder if there are some risks
> > to get regions according to e820 when there is no suitable region
> > in efi memmaps.
> > ---
> >  arch/x86/boot/compressed/kaslr.c | 4 +++-
> >  1 file changed, 3 insertions(+), 1 deletion(-)
> >
> > diff --git a/arch/x86/boot/compressed/kaslr.c b/arch/x86/boot/compressed/kaslr.c
> > index b92fffbe761f..dbd7244b71aa 100644
> > --- a/arch/x86/boot/compressed/kaslr.c
> > +++ b/arch/x86/boot/compressed/kaslr.c
> > @@ -685,6 +685,7 @@ process_efi_entries(unsigned long minimum, unsigned long image_size)
> >  {
> >         struct efi_info *e = &boot_params->efi_info;
> >         bool efi_mirror_found = false;
> > +       bool efi_mem_region_found = false;
> >         struct mem_vector region;
> >         efi_memory_desc_t *md;
> >         unsigned long pmap;
> > @@ -742,12 +743,13 @@ process_efi_entries(unsigned long minimum, unsigned long image_size)
> >                     !(md->attribute & EFI_MEMORY_MORE_RELIABLE))
> >                         continue;
> >
> > +               efi_mem_region_found = false;
					   ^^ this should be true, not false.

Other than that, I think this should be okay. The reason EFI memmap is
preferred over E820, according to commit

  0982adc74673 ("x86/boot/KASLR: Work around firmware bugs by excluding EFI_BOOT_SERVICES_* and EFI_LOADER_* from KASLR's choice")

was to avoid allocating inside EFI_BOOT_SERVICES/EFI_LOADER_DATA etc.
That's not a danger during kexec, and I believe runtime services regions
should be marked as reserved in the E820 map, right?

Also, something a little fishy-looking here is that the first loop to
see if there is any EFI_MEMORY_MORE_RELIABLE region does not apply any
of the checks on the memory region type/attributes. If there is a mirror
region but it isn't conventional memory, or if it was soft-reserved, we
shouldn't be setting efi_mirror_found.


> >                 region.start = md->phys_addr;
> >                 region.size = md->num_pages << EFI_PAGE_SHIFT;
> >                 if (process_mem_region(&region, minimum, image_size))
> >                         break;
> >         }
> > -       return true;
> > +       return efi_mem_region_found;
> >  }
> >  #else
> >  static inline bool
> > --
> > 2.23.0
> >