Re: [PATCH v2 0/8] efi/libstub: simplify arm64 kernel image loading

[Ard Biesheuvel <ardb@kernel.org>]

On Mon, 13 Apr 2020 at 23:54, Atish Patra <atishp@atishpatra.org> wrote:
>
> On Mon, Apr 13, 2020 at 8:55 AM Ard Biesheuvel <ardb@kernel.org> wrote:
> >
> > On arm64, the kernel image used to be virtually mapped via the linear
> > region, making the two mappings correlated in a way that required the
> > kernel to be located at the start of the linear region, or the memory
> > below would not be accessible. For this reason, the EFI stub loader
> > code for arm64 has the notion of a 'preferred offset' for the physical
> > placement of the kernel image, and tries to put the kernel there, or
> > at least as low as possible in physical memory (unless KASLR is active,
> > in which case the placement is randomized)
> >
> > When KASLR was introduced, the virtual mapping of the kernel was moved
> > into the vmalloc region, and now, regardless of whether KASLR support
> > is built in or active, the kernel can be placed anywhere in physical
> > memory without any detrimental side effects on the linear region.
> >
> > This means that we can drop the notion of 'preferred offset' entirely,
> > and invoke the kernel in place if the PE/COFF loader loaded it at the
> > right offset. If not, we can invoke the ordinary UEFI top down page
> > allocator to reallocate it elsewhere in memory. By updating the PE/COFF
> > metadata, we can inform the PE/COFF loader about the desired alignment,
> > making it less likely that we need to move the kernel image in the first
> > place.
> >
> > Ard Biesheuvel (8):
> >   efi/libstub/random: align allocate size to EFI_ALLOC_ALIGN
> >   efi/libstub/random: increase random alloc granularity
> >   efi/libstub/arm64: replace 'preferred' offset with alignment check
> >   efi/libstub/arm64: simplify randomized loading of kernel image
> >   efi/libstub/arm64: align PE/COFF sections to segment alignment
> >   efi/libstub: add API function to allocate aligned memory
> >   efi/libstub/arm64: switch to ordinary page allocator for kernel image
> >   efi/libstub: move efi_relocate_kernel() into separate source file
> >
> >  arch/arm64/kernel/efi-header.S             |   2 +-
> >  arch/arm64/kernel/vmlinux.lds.S            |   3 +-
> >  drivers/firmware/efi/libstub/Makefile      |   3 +-
> >  drivers/firmware/efi/libstub/alignedmem.c  |  57 ++++++
> >  drivers/firmware/efi/libstub/arm64-stub.c  |  92 +++-------
> >  drivers/firmware/efi/libstub/efistub.h     |  18 +-
> >  drivers/firmware/efi/libstub/mem.c         | 191 +-------------------
> >  drivers/firmware/efi/libstub/randomalloc.c |   6 +-
> >  drivers/firmware/efi/libstub/relocate.c    | 174 ++++++++++++++++++
> >  9 files changed, 280 insertions(+), 266 deletions(-)
> >  create mode 100644 drivers/firmware/efi/libstub/alignedmem.c
> >  create mode 100644 drivers/firmware/efi/libstub/relocate.c
> >
> > --
> > 2.17.1
> >
>
> Oops. I just noticed this series after I sent out a v2.
> I see that efi_low_alloc is removed now and the handle_kernel_image is
> simplified for arm64.
> I will update the risc-v uefi series accordingly. Sorry for the noise.
>

No worries. And apologies for making this a moving target :-)

I realised you will probably need to bring back efi_low_alloc() as a
global symbol, as I don't think you will be able to switch to
efi_relocate_kernel().

In any case, this series is just set of patches on the list, so feel
free to propose changes if they are making your life too difficult.

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel