Re: [PATCH] lkdtm: Add more EFI runtime tests

[Ard Biesheuvel]

On Sun, 11 Oct 2026, at 16:48, MaeeFilho FL wrote:
> Hi Ard,
>
> The Code is excellent. A few points worth highlighting:
>
>> Two bugs were found in the arm64 EFI runtime support that weren't caught
>> by LKDTM. The existing EFI_RUNTIME_CRASH test instructs the firmware to
>> write to kernel memory that is mapped read-only, which usually fails as
>> expected. However, this does not exercise the runtime mapping logic in
>> the kernel, nor the fault recovery logic when the faulting address is
>> inside the EFI runtime memory regions itself.
>>
>> So add two more test cases:
>> - a write to R/O memory test that uses the firmware's own executable
>>   memory as an output buffer
>> - a branch to memory that is occupied by the firmware's runtime services
>>   function table, which is usually mapped non-execute.
>
> This is an excellent addition to the LKDTM suite. Forcing GetTime() to
> target efi.runtime->set_time is a very clever way to ensure that the
> firmware page tables are strictly enforcing permissions over its own
> executable boundaries, rather than just testing general kernel mappings.
>
> The execution tracking via the anonymous function pointer struct conversion
> to trap NX faults on efi.runtime data object is also rock solid and properly
> exercises the arm64 translation tables validation logic.
>
> Reviewed-by: João Guilherme da Silva Freitas Lima <joaoanandalima@xxxxxxxxx>

Thanks, but next time, please don't bother with the LLM generated reviews
unless there is actionable feedback.