Re: [PATCH] efi/libstub: Indicate clang the relocation mode for arm64

[Ard Biesheuvel]

On 18 May 2017 at 00:24, Greg Hackmann <ghackmann@xxxxxxxxxx> wrote:
> On 05/11/2017 06:51 AM, Ard Biesheuvel wrote:
> [snip]
>>>>>>
>>>>>> In my opinion, the correct fix would be to make -fpie (as opposed to
>>>>>> -fpic) imply hidden visibility, given that PIE executables don't
>>>>>> export symbols in the first place, and so the preemption rules do not
>>>>>> apply. It is worth a try whether -fpie works as expected in this case
>>>>>> on Clang, but the last time I tried it on GCC, it behaved exactly like
>>>>>> -fpic.
>>>>>
>>>>>
>>>>> Thanks a lot for the detailed description and your suggestions!
>>>>>
>>>>> A clang build with -fpie for the EFI stub succeeds without complaints
>>>>> about GOT entries. I will send out an updated patch (with -fpie only
>>>>> for clang) later.
>>>>>
>>>>
>>>> Good! I never liked the visibility hack, which is why I never upstreamed
>>>> it.
>>>>
>>>> Could you please check how recent GCC behaves?
>>>
>>>
>>> I tried GCC v4.9.4 and v6.3.1, both build the EFI stub with -fpie
>>> without errors.
>>>
>>> Are you suggesting to use -fpie for both clang and GCC? Do you know
>>> what the minimum required GCC version is for building an arm64 kernel?
>>
>>
>> Yes. Up until now, we have been relying on the position independent
>> nature of small model code, but it would be better to specify it
>> explicitly, so if -fpie gives us mostly identical code and does not
>> need visibility hacks, I would prefer to add it for all compilers and
>> not have an exception only for Clang. Note that the same applies to
>> the entire kernel when built in KASLR mode, so it would also be good
>> to know our options here.
>>
>> Arnd, Will, what is the oldest GCC version we claim to support for arm64?
>>
>
> Unfortunately, after looking into this a bit more, -fpie by itself doesn't
> force clang to disable symbol preeemption.  For example when building the
> EFI stub from 4.9 with clang, -fpie gives me a stub that crashes with a
> synchronous exception inside handle_kernel_image().  The faulting
> instruction is a read from __nokaslr that still goes through the GOT.
>
> Right now you'll get a usable EFI stub with -fpie anyway, since 60f38de7a8d4
> ("efi/libstub: Unify command line param parsing") masked the problem when it
> moved __nokaslr behind a helper function.  But AIUI there's nothing really
> preventing a similar problem in the future.
>
> You *can* force clang to disable symbol preemption using "-fpie
> -mpie-copy-relocations".  That said, I don't know enough about EFI to say
> whether this is actually appropriate for building the EFI stub.

Thanks for digging into this. It is really quite unfortunate that it
is so difficult to force Clang (or GCC for that matter) to generate
relative references without the compiler assuming that you are
building a shared library. Perhaps we need a stronger version of
-fvisibility=hidden, i.e., one that applies to extern declarations as
well.

For the stub, we could simply replace all remaining extern symbol
references (if any) with accessor functions, such as the one I added
for __nokaslr (which is actually needed for x86 as well, for different
reasons). Let me look into this.