[PATCH v2 08/12] crypto: x86/sm4 - fix crash with CFI enabled

[Sami Tolvanen]

On Fri, Nov 18, 2022 at 12:10 PM Eric Biggers <ebiggers at kernel.org> wrote:
>
> On Fri, Nov 18, 2022 at 11:44:17AM -0800, Eric Biggers wrote:
> > From: Eric Biggers <ebiggers at google.com>
> >
> > sm4_aesni_avx_ctr_enc_blk8(), sm4_aesni_avx_cbc_dec_blk8(),
> > sm4_aesni_avx_cfb_dec_blk8(), sm4_aesni_avx2_ctr_enc_blk16(),
> > sm4_aesni_avx2_cbc_dec_blk16(), and sm4_aesni_avx2_cfb_dec_blk16() are
> > called via indirect function calls.  Therefore they need to use
> > SYM_TYPED_FUNC_START instead of SYM_FUNC_START to cause their type
> > hashes to be emitted when the kernel is built with CONFIG_CFI_CLANG=y.
> > Otherwise, the code crashes with a CFI failure.
> >
> > (Or at least that should be the case.  For some reason the CFI checks in
> > sm4_avx_cbc_decrypt(), sm4_avx_cfb_decrypt(), and sm4_avx_ctr_crypt()
> > are not always being generated, using current tip-of-tree clang.
> > Anyway, this patch is a good idea anyway.)
>
> Sami, is it expected that a CFI check isn't being generated for the indirect
> call to 'func' in sm4_avx_cbc_decrypt()?  I'm using LLVM commit 4a7be42d922af0.

If the compiler emits an indirect call, it should also emit a CFI
check. What's the assembly code it generates here?

Sami