On Thu, May 15, 2025 at 11:52:54AM -0700, Eric Biggers wrote: > On Thu, May 15, 2025 at 04:27:03PM +0200, Ard Biesheuvel wrote: > > diff --git a/arch/arm64/crypto/sha512-ce-core.S b/arch/arm64/crypto/sha512-ce-core.S > > index 91ef68b15fcc..deb2469ab631 100644 > > --- a/arch/arm64/crypto/sha512-ce-core.S > > +++ b/arch/arm64/crypto/sha512-ce-core.S > > @@ -12,26 +12,7 @@ > > #include <linux/linkage.h> > > #include <asm/assembler.h> > > > > - .irp b,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19 > > - .set .Lq\b, \b > > - .set .Lv\b\().2d, \b > > - .endr > > - > > - .macro sha512h, rd, rn, rm > > - .inst 0xce608000 | .L\rd | (.L\rn << 5) | (.L\rm << 16) > > - .endm > > - > > - .macro sha512h2, rd, rn, rm > > - .inst 0xce608400 | .L\rd | (.L\rn << 5) | (.L\rm << 16) > > - .endm > > - > > - .macro sha512su0, rd, rn > > - .inst 0xcec08000 | .L\rd | (.L\rn << 5) > > - .endm > > - > > - .macro sha512su1, rd, rn, rm > > - .inst 0xce608800 | .L\rd | (.L\rn << 5) | (.L\rm << 16) > > - .endm > > + .arch armv8-a+sha3 > > This looked like a mistake: SHA-512 is part of SHA-2, not SHA-3. However, the > current versions of binutils and clang do indeed put it under sha3. There > should be a comment that mentions this unfortunate quirk. > > However, there's also the following commit which went into binutils 2.43: > > commit 0aac62aa3256719c37be9e0ce6af8b190f45c928 > Author: Andrew Carlotti <andrew.carlotti@xxxxxxx> > Date: Fri Jan 19 13:01:40 2024 +0000 > > aarch64: move SHA512 instructions to +sha3 > > SHA512 instructions were added to the architecture at the same time as SHA3 > instructions, but later than the SHA1 and SHA256 instructions. Furthermore, > implementations must support either both or neither of the SHA512 and SHA3 > instruction sets. However, SHA512 instructions were originally (and > incorrectly) added to Binutils under the +sha2 flag. > > This patch moves SHA512 instructions under the +sha3 flag, which matches the > architecture constraints and existing GCC and LLVM behaviour. > > So probably we need ".arch armv8-a+sha2+sha3" to support binutils 2.30 through > 2.42, as well as clang and the latest version of binutils? (I didn't test it > yet, but it seems likely...) Actually "sha2" isn't required here, since "sha3" implies "sha2". The kernel test robot did report a build error on this series. But it was with SHA-3, because in binutils 2.40 and earlier the SHA-3 instructions required both "sha3" and "armv8.2-a", not just "sha3" like they do in clang and in binutils 2.41 and later. For now, I split the SHA-512 part into a separate patch https://lore.kernel.org/r/20250718220706.475240-1-ebiggers@xxxxxxxxxx - Eric
