Debian's clang carries a patch that makes the default FPU mode
'vfp3-d16' instead of 'neon' for 'armv7-a' to avoid generating NEON
instructions on hardware that does not support them:
5a61ca6f21/debian/patches/clang-arm-default-vfp3-on-armv7a.patchhttps://bugs.debian.org/841474https://bugs.debian.org/842142https://bugs.debian.org/914268
This results in the following build error when clang's integrated
assembler is used because the '.arch' directive overrides the '.fpu'
directive:
arch/arm/crypto/curve25519-core.S:25:2: error: instruction requires: NEON
vmov.i32 q0, #1
^
arch/arm/crypto/curve25519-core.S:26:2: error: instruction requires: NEON
vshr.u64 q1, q0, #7
^
arch/arm/crypto/curve25519-core.S:27:2: error: instruction requires: NEON
vshr.u64 q0, q0, #8
^
arch/arm/crypto/curve25519-core.S:28:2: error: instruction requires: NEON
vmov.i32 d4, #19
^
Shuffle the order of the '.arch' and '.fpu' directives so that the code
builds regardless of the default FPU mode. This has been tested against
both clang with and without Debian's patch and GCC.
Cc: stable@vger.kernel.org
Fixes: d8f1308a02 ("crypto: arm/curve25519 - wire up NEON implementation")
Link: https://github.com/ClangBuiltLinux/continuous-integration2/issues/118
Reported-by: Arnd Bergmann <arnd@arndb.de>
Suggested-by: Arnd Bergmann <arnd@arndb.de>
Suggested-by: Jessica Clarke <jrtc27@jrtc27.com>
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Acked-by: Jason A. Donenfeld <Jason@zx2c4.com>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Tested-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This ports the SUPERCOP implementation for usage in kernel space. In
addition to the usual header, macro, and style changes required for
kernel space, it makes a few small changes to the code:
- The stack alignment is relaxed to 16 bytes.
- Superfluous mov statements have been removed.
- ldr for constants has been replaced with movw.
- ldreq has been replaced with moveq.
- The str epilogue has been made more idiomatic.
- SIMD registers are not pushed and popped at the beginning and end.
- The prologue and epilogue have been made idiomatic.
- A hole has been removed from the stack, saving 32 bytes.
- We write-back the base register whenever possible for vld1.8.
- Some multiplications have been reordered for better A7 performance.
There are more opportunities for cleanup, since this code is from qhasm,
which doesn't always do the most opportune thing. But even prior to
extensive hand optimizations, this code delivers significant performance
improvements (given in get_cycles() per call):
----------- -------------
| generic C | this commit |
------------ ----------- -------------
| Cortex-A7 | 49136 | 22395 |
------------ ----------- -------------
| Cortex-A17 | 17326 | 4983 |
------------ ----------- -------------
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
[ardb: - move to arch/arm/crypto
- wire into lib/crypto framework
- implement crypto API KPP hooks ]
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This comes from Dan Bernstein and Peter Schwabe's public domain NEON
code, and is included here in raw form so that subsequent commits that
fix these up for the kernel can see how it has changed. This code does
have some entirely cosmetic formatting differences, adding indentation
and so forth, so that when we actually port it for use in the kernel in
the subsequent commit, it's obvious what's changed in the process.
This code originates from SUPERCOP 20180818, available at
<https://bench.cr.yp.to/supercop.html>.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Nathan Chancellor
Jason A. Donenfeld