Re: [RFC PATCH v2] x86/sev: enforce RIP-relative accesses in early SEV/SME code

[Borislav Petkov]

On Sun, Jan 21, 2024 at 03:12:56PM +0100, Ard Biesheuvel wrote:
> My preliminary conclusion confirms that the SEV code is quite
> problematic in this regard (which is the reason for this patch, so we
> already knew that, of course).

So we can try to improve the situation gradually so that we don't
break current usages.

> TL;DR I think we will need a way to build certain objects with -fPIC
> (as we do in other places and on other architectures), but we should
> add instrumentation to ensure that these issues can be detected at
> build time. The SEV boot code is especially tricky here as very few
> people can even test it,

No worries about that - us, the Google cloud folks, AWS and a bunch of
others are people I could think of who could help out. :-)

> so applying this patch and hoping that the compiler will never
> generate reachable code paths that only work correctly when executed
> via the ordinary kernel virtual mapping is not sufficient.

..

> 1)
> WARNING: modpost: vmlinux: section mismatch in reference:
> startup_64_pi+0x33 (section: .pi.text) -> sme_enable (section:
> .init.text)

sme_enable() is in the 1:1 mapping TU
arch/x86/mm/mem_encrypt_identity.c, see

  1cd9c22fee3a ("x86/mm/encrypt: Move page table helpers into separate translation unit")

so might as well move it to .pi.text

The rest below look like they'd need more serious untangling.

Btw, I just had another idea: we could remove -mcmodel=kernel from the
build flags of the whole kernel once -fPIC is enabled so that gcc can be
forced to do rIP-relative addressing.

I'm being told the reason it doesn't allow mcmodel=kernel with -fPIC is
only a matter of removing that check and that it *should* otherwise work
but someone needs to try that. And then there are older gccs which we
cannot fix.

Thx.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette