Skip to contentSkip to conversation

crypto: curve25519 - Work around link failure

PATCH In review Arnd Bergmann 2 participants 3 messages Jan 7, 2020

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 crypto: curve25519 - Work around link failure

The curve25519 selftest causes a link failure when one of the two implementations is built-in and the other one is a loadable module, as then the library gets built in as well but cannot call into the module:

lib/crypto/curve25519-selftest.o: In function `curve25519_selftest':
curve25519-selftest.c:(.init.text+0x5c): undefined reference to curve25519_arch' curve25519-selftest.c:(.init.text+0xfd): undefined reference to curve25519_base_arch' curve25519-selftest.c:(.init.text+0x15a): undefined reference to `curve25519_arch'

There is probably a better fix, but this is the local workaround that I used to get a clean randconfig build again, using Makefile tricks to make all the curve25519 code built-in if any of the implementations are.

Fixes aa127963f1ca ("crypto: lib/curve25519 - re-add selftests")
---
 arch/arm/crypto/Makefile | 4 +++-
 arch/x86/crypto/Makefile | 4 +++-
 crypto/Makefile          | 5 ++++-
 3 files changed, 10 insertions(+), 3 deletions(-)
arch/arm/crypto/Makefile +3 -1
index b745c17d356f..a7b3957aca58 100644
@@ -12,7 +12,9 @@ obj-$(CONFIG_CRYPTO_SHA512_ARM) += sha512-arm.o
 obj-$(CONFIG_CRYPTO_CHACHA20_NEON) += chacha-neon.o
 obj-$(CONFIG_CRYPTO_POLY1305_ARM) += poly1305-arm.o
 obj-$(CONFIG_CRYPTO_NHPOLY1305_NEON) += nhpoly1305-neon.o
-obj-$(CONFIG_CRYPTO_CURVE25519_NEON) += curve25519-neon.o
+ifdef CONFIG_CRYPTO_CURVE25519_NEON
+obj-$(CONFIG_CRYPTO_LIB_CURVE25519_GENERIC) += curve25519-neon.o
+endif
 
 obj-$(CONFIG_CRYPTO_AES_ARM_CE) += aes-arm-ce.o
 obj-$(CONFIG_CRYPTO_SHA1_ARM_CE) += sha1-arm-ce.o
arch/x86/crypto/Makefile +3 -1
index 958440eae27e..7546c276e2f0 100644
@@ -39,7 +39,9 @@ obj-$(CONFIG_CRYPTO_AEGIS128_AESNI_SSE2) += aegis128-aesni.o
 
 obj-$(CONFIG_CRYPTO_NHPOLY1305_SSE2) += nhpoly1305-sse2.o
 obj-$(CONFIG_CRYPTO_NHPOLY1305_AVX2) += nhpoly1305-avx2.o
-obj-$(CONFIG_CRYPTO_CURVE25519_X86) += curve25519-x86_64.o
+ifdef CONFIG_CRYPTO_CURVE25519_X86
+obj-$(CONFIG_CRYPTO_LIB_CURVE25519_GENERIC) += curve25519-x86_64.o
+endif
 
 # These modules require assembler to support AVX.
 ifeq ($(avx_supported),yes)
crypto/Makefile +4 -1
index 4ca12b6044f7..93ecbfe50285 100644
@@ -166,7 +166,10 @@ obj-$(CONFIG_CRYPTO_ZSTD) += zstd.o
 obj-$(CONFIG_CRYPTO_OFB) += ofb.o
 obj-$(CONFIG_CRYPTO_ECC) += ecc.o
 obj-$(CONFIG_CRYPTO_ESSIV) += essiv.o
-obj-$(CONFIG_CRYPTO_CURVE25519) += curve25519-generic.o
+
+ifdef CONFIG_CRYPTO_CURVE25519
+obj-$(CONFIG_CRYPTO_LIB_CURVE25519_GENERIC) += curve25519-generic.o
+endif
 
 ecdh_generic-y += ecdh.o
 ecdh_generic-y += ecdh_helper.o
Signature · 1 line
2.20.0
#2 lore raw
Quoted 9 lines · Jason A. Donenfeld

Hey Arnd,

Another solution to this was already posted:

https://lore.kernel.org/linux-crypto/CAHmME9pg4KWw1zNVybxn1WLGusyGCjqeAHLQXY=Dr4zznUM82g@mail.gmail.com/T/#t

That might be slightly cleaner, though yours is shorter. I'm alright with either one.

Yes, I agree the other approach looks nicer. I've added that to my randconfig builder in place of my patch to see if it's sufficient. If not, I'll let you know.

Arnd