diff mbox series

[5/6] crypto: aegis128 - provide a SIMD implementation based on NEON intrinsics

Message ID 20190624073818.29296-6-ard.biesheuvel@linaro.org (mailing list archive)
State Changes Requested
Delegated to: Herbert Xu
Headers show
Series crypto: aegis128 - add NEON intrinsics version for ARM/arm64 | expand

Commit Message

Ard Biesheuvel June 24, 2019, 7:38 a.m. UTC 
Provide an accelerated implementation of aegis128 by wiring up the
SIMD hooks in the generic driver to an implementation based on NEON
intrinsics, which can be compiled to both ARM and arm64 code.

This results in a performance of 2.2 cycles per byte on Cortex-A53,
which is a performance increase of ~11x compared to the generic
code.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
---
 crypto/Kconfig               |   5 +
 crypto/Makefile              |  12 ++
 crypto/aegis128-neon-inner.c | 142 ++++++++++++++++++++
 crypto/aegis128-neon.c       |  43 ++++++
 4 files changed, 202 insertions(+)

Comments

Ard Biesheuvel June 24, 2019, 2:37 p.m. UTC | #1 
On Mon, 24 Jun 2019 at 09:38, Ard Biesheuvel <ard.biesheuvel@linaro.org> wrote:
>
> Provide an accelerated implementation of aegis128 by wiring up the
> SIMD hooks in the generic driver to an implementation based on NEON
> intrinsics, which can be compiled to both ARM and arm64 code.
>
> This results in a performance of 2.2 cycles per byte on Cortex-A53,
> which is a performance increase of ~11x compared to the generic
> code.
>
> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
> ---
>  crypto/Kconfig               |   5 +
>  crypto/Makefile              |  12 ++
>  crypto/aegis128-neon-inner.c | 142 ++++++++++++++++++++
>  crypto/aegis128-neon.c       |  43 ++++++
>  4 files changed, 202 insertions(+)
>
...
> diff --git a/crypto/Makefile b/crypto/Makefile
> index 266a4cdbb9e2..f4a55cfb7f17 100644
> --- a/crypto/Makefile
> +++ b/crypto/Makefile
> @@ -92,6 +92,18 @@ obj-$(CONFIG_CRYPTO_GCM) += gcm.o
>  obj-$(CONFIG_CRYPTO_CCM) += ccm.o
>  obj-$(CONFIG_CRYPTO_CHACHA20POLY1305) += chacha20poly1305.o
>  obj-$(CONFIG_CRYPTO_AEGIS128) += aegis128.o
> +aegis128-y := aegis128.o
> +

This doesn't actually work when building a module. I'll have to rename
the .c file so that the module that combines the objects can retain
its name
diff mbox series

Patch

diff --git a/crypto/Kconfig b/crypto/Kconfig
index 3d056e7da65f..c4b96f2e1344 100644
--- a/crypto/Kconfig
+++ b/crypto/Kconfig
@@ -311,6 +311,11 @@  config CRYPTO_AEGIS128
 	help
 	 Support for the AEGIS-128 dedicated AEAD algorithm.
 
+config CRYPTO_AEGIS128_SIMD
+	bool "Support SIMD acceleration for AEGIS-128"
+	depends on CRYPTO_AEGIS128 && ((ARM || ARM64) && KERNEL_MODE_NEON)
+	default y
+
 config CRYPTO_AEGIS128L
 	tristate "AEGIS-128L AEAD algorithm"
 	select CRYPTO_AEAD
diff --git a/crypto/Makefile b/crypto/Makefile
index 266a4cdbb9e2..f4a55cfb7f17 100644
--- a/crypto/Makefile
+++ b/crypto/Makefile
@@ -92,6 +92,18 @@  obj-$(CONFIG_CRYPTO_GCM) += gcm.o
 obj-$(CONFIG_CRYPTO_CCM) += ccm.o
 obj-$(CONFIG_CRYPTO_CHACHA20POLY1305) += chacha20poly1305.o
 obj-$(CONFIG_CRYPTO_AEGIS128) += aegis128.o
+aegis128-y := aegis128.o
+
+ifeq ($(ARCH),arm)
+CFLAGS_aegis128-neon-inner.o += -ffreestanding -march=armv7-a -mfloat-abi=softfp -mfpu=crypto-neon-fp-armv8
+aegis128-$(CONFIG_CRYPTO_AEGIS128_SIMD) += aegis128-neon.o aegis128-neon-inner.o
+endif
+ifeq ($(ARCH),arm64)
+CFLAGS_aegis128-neon-inner.o += -ffreestanding -mcpu=generic+crypto
+CFLAGS_REMOVE_aegis128-neon-inner.o += -mgeneral-regs-only
+aegis128-$(CONFIG_CRYPTO_AEGIS128_SIMD) += aegis128-neon.o aegis128-neon-inner.o
+endif
+
 obj-$(CONFIG_CRYPTO_AEGIS128L) += aegis128l.o
 obj-$(CONFIG_CRYPTO_AEGIS256) += aegis256.o
 obj-$(CONFIG_CRYPTO_MORUS640) += morus640.o
diff --git a/crypto/aegis128-neon-inner.c b/crypto/aegis128-neon-inner.c
new file mode 100644
index 000000000000..c6d90390ac38
--- /dev/null
+++ b/crypto/aegis128-neon-inner.c
@@ -0,0 +1,142 @@ 
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright (C) 2019 Linaro, Ltd. <ard.biesheuvel@linaro.org>
+ */
+
+#ifdef CONFIG_ARM64
+#include <asm/neon-intrinsics.h>
+#else
+#include <arm_neon.h>
+#endif
+
+#define AEGIS_BLOCK_SIZE	16
+
+#include <stddef.h>
+
+void *memcpy(void *dest, const void *src, size_t n);
+void *memset(void *s, int c, size_t n);
+
+struct aegis128_state {
+	uint8x16_t v[5];
+};
+
+static struct aegis128_state aegis128_update_neon(struct aegis128_state st,
+						  uint8x16_t m)
+{
+	uint8x16_t z = {};
+	uint8x16_t t;
+
+	t        = vaesmcq_u8(vaeseq_u8(st.v[3], z));
+	st.v[3] ^= vaesmcq_u8(vaeseq_u8(st.v[2], z));
+	st.v[2] ^= vaesmcq_u8(vaeseq_u8(st.v[1], z));
+	st.v[1] ^= vaesmcq_u8(vaeseq_u8(st.v[0], z));
+	st.v[0] ^= vaesmcq_u8(vaeseq_u8(st.v[4], z)) ^ m;
+	st.v[4] ^= t;
+
+	return st;
+}
+
+void crypto_aegis128_update_neon(void *state, const void *msg)
+{
+	struct aegis128_state st = { {
+		vld1q_u8(state),
+		vld1q_u8(state + 16),
+		vld1q_u8(state + 32),
+		vld1q_u8(state + 48),
+		vld1q_u8(state + 64)
+	} };
+
+	st = aegis128_update_neon(st, vld1q_u8(msg));
+
+	vst1q_u8(state, st.v[0]);
+	vst1q_u8(state + 16, st.v[1]);
+	vst1q_u8(state + 32, st.v[2]);
+	vst1q_u8(state + 48, st.v[3]);
+	vst1q_u8(state + 64, st.v[4]);
+}
+
+void crypto_aegis128_encrypt_chunk_neon(void *state, void *dst, const void *src,
+					unsigned int size)
+{
+	struct aegis128_state st = { {
+		vld1q_u8(state),
+		vld1q_u8(state + 16),
+		vld1q_u8(state + 32),
+		vld1q_u8(state + 48),
+		vld1q_u8(state + 64)
+	} };
+	uint8x16_t tmp;
+
+	while (size >= AEGIS_BLOCK_SIZE) {
+		uint8x16_t s = vld1q_u8(src);
+
+		tmp = s ^ st.v[1] ^ (st.v[2] & st.v[3]) ^ st.v[4];
+		st = aegis128_update_neon(st, s);
+		vst1q_u8(dst, tmp);
+
+		size -= AEGIS_BLOCK_SIZE;
+		src += AEGIS_BLOCK_SIZE;
+		dst += AEGIS_BLOCK_SIZE;
+	}
+
+	if (size > 0) {
+		uint8_t buf[AEGIS_BLOCK_SIZE] = {};
+		uint8x16_t msg;
+
+		memcpy(buf, src, size);
+		msg = vld1q_u8(buf);
+		tmp = msg ^ st.v[1] ^ (st.v[2] & st.v[3]) ^ st.v[4];
+		st = aegis128_update_neon(st, msg);
+		vst1q_u8(buf, tmp);
+		memcpy(dst, buf, size);
+	}
+
+	vst1q_u8(state, st.v[0]);
+	vst1q_u8(state + 16, st.v[1]);
+	vst1q_u8(state + 32, st.v[2]);
+	vst1q_u8(state + 48, st.v[3]);
+	vst1q_u8(state + 64, st.v[4]);
+}
+
+void crypto_aegis128_decrypt_chunk_neon(void *state, void *dst, const void *src,
+					unsigned int size)
+{
+	struct aegis128_state st = { {
+		vld1q_u8(state),
+		vld1q_u8(state + 16),
+		vld1q_u8(state + 32),
+		vld1q_u8(state + 48),
+		vld1q_u8(state + 64)
+	} };
+	uint8x16_t tmp;
+
+	while (size >= AEGIS_BLOCK_SIZE) {
+		tmp = vld1q_u8(src) ^ st.v[1] ^ (st.v[2] & st.v[3]) ^ st.v[4];
+		st = aegis128_update_neon(st, tmp);
+		vst1q_u8(dst, tmp);
+
+		size -= AEGIS_BLOCK_SIZE;
+		src += AEGIS_BLOCK_SIZE;
+		dst += AEGIS_BLOCK_SIZE;
+	}
+
+	if (size > 0) {
+		uint8_t buf[AEGIS_BLOCK_SIZE] = {};
+		uint8x16_t msg;
+
+		memcpy(buf, src, size);
+		msg = vld1q_u8(buf) ^ st.v[1] ^ (st.v[2] & st.v[3]) ^ st.v[4];
+		vst1q_u8(buf, msg);
+		memcpy(dst, buf, size);
+
+		memset(buf + size, 0, AEGIS_BLOCK_SIZE - size);
+		msg = vld1q_u8(buf);
+		st = aegis128_update_neon(st, msg);
+	}
+
+	vst1q_u8(state, st.v[0]);
+	vst1q_u8(state + 16, st.v[1]);
+	vst1q_u8(state + 32, st.v[2]);
+	vst1q_u8(state + 48, st.v[3]);
+	vst1q_u8(state + 64, st.v[4]);
+}
diff --git a/crypto/aegis128-neon.c b/crypto/aegis128-neon.c
new file mode 100644
index 000000000000..c1c0a1686f67
--- /dev/null
+++ b/crypto/aegis128-neon.c
@@ -0,0 +1,43 @@ 
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright (C) 2019 Linaro Ltd <ard.biesheuvel@linaro.org>
+ */
+
+#include <asm/cpufeature.h>
+#include <asm/neon.h>
+
+#include "aegis.h"
+
+void crypto_aegis128_update_neon(void *state, const void *msg);
+void crypto_aegis128_encrypt_chunk_neon(void *state, void *dst, const void *src,
+					unsigned int size);
+void crypto_aegis128_decrypt_chunk_neon(void *state, void *dst, const void *src,
+					unsigned int size);
+
+bool crypto_aegis128_have_simd(void)
+{
+	return cpu_have_feature(cpu_feature(AES));
+}
+
+void crypto_aegis128_update_simd(union aegis_block *state, const void *msg)
+{
+	kernel_neon_begin();
+	crypto_aegis128_update_neon(state, msg);
+	kernel_neon_end();
+}
+
+void crypto_aegis128_encrypt_chunk_simd(union aegis_block *state, u8 *dst,
+					const u8 *src, unsigned int size)
+{
+	kernel_neon_begin();
+	crypto_aegis128_encrypt_chunk_neon(state, dst, src, size);
+	kernel_neon_end();
+}
+
+void crypto_aegis128_decrypt_chunk_simd(union aegis_block *state, u8 *dst,
+					const u8 *src, unsigned int size)
+{
+	kernel_neon_begin();
+	crypto_aegis128_decrypt_chunk_neon(state, dst, src, size);
+	kernel_neon_end();
+}