[PATCH 0/2] efi: preserve NEON registers on UEFI services calls

[Ard Biesheuvel]

On 23 June 2014 16:18, Ard Biesheuvel <ard.biesheuvel at linaro.org> wrote:
> The current UEFI implementation for arm64 fails to preserve/restore the contents
> of the NEON register file, which may result in data corruption, especially now
> that those contents are lazily restored for user processes.
>
> This series proposes to fix this by wrapping all runtime services calls, and
> adding kernel_neon_begin()/kernel_neon_end() pairs to the wrappers.
>
> The first patch moves the existing x86 versions of those wrappers to generic
> code, so that the second patch can easily enable them by supplying a definition
> for  efi_call_virt and adding a call to efi_native_runtime_setup().
>

CC'ing Olivier and Mark (with correct email address this time).

Also, as an additional note, the UEFI spec section 2.3.6.4 mandates
that 'any additional execution state context' should be saved and
restored by the callee, which would imply that doing it in the kernel
is redundant. But current implementations of Tianocore/EDK2 don't seem
to honor that requirement, and considering GCC's tendency to spill
state to FPSIMD registers, we may choose to do so anyway to be on the
safe side.

-- 
Ard.

> Ard Biesheuvel (2):
>   efi/x86: move UEFI Runtime Services wrappers to generic code
>   efi/arm64: preserve NEON registers on UEFI runtime services calls
>
>  arch/arm64/include/asm/efi.h   |  21 ++++++
>  arch/arm64/kernel/efi.c        |  14 +---
>  arch/x86/platform/efi/efi.c    | 144 +----------------------------------
>  drivers/firmware/efi/Makefile  |   2 +-
>  drivers/firmware/efi/runtime.c | 167 +++++++++++++++++++++++++++++++++++++++++
>  include/linux/efi.h            |   2 +
>  6 files changed, 194 insertions(+), 156 deletions(-)
>  create mode 100644 drivers/firmware/efi/runtime.c
>
> --
> 1.8.3.2
>