crypto: aegis128 - provide a SIMD implementation based on NEON intrinsics

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.

Reviewed-by: Ondrej Mosnacek <omosnace@redhat.com>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

crypto/Kconfig

@@ -306,6 +306,11 @@ 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_AEGIS128_AESNI_SSE2
 	tristate "AEGIS-128 AEAD algorithm (x86_64 AESNI+SSE2 implementation)"
 	depends on X86 && 64BIT

crypto/Makefile

@@ -91,6 +91,18 @@ obj-$(CONFIG_CRYPTO_CCM) += ccm.o
 obj-$(CONFIG_CRYPTO_CHACHA20POLY1305) += chacha20poly1305.o
 obj-$(CONFIG_CRYPTO_AEGIS128) += aegis128.o
 aegis128-y := aegis128-core.o
+
+ifeq ($(ARCH),arm)
+CFLAGS_aegis128-neon-inner.o += -ffreestanding -march=armv7-a -mfloat-abi=softfp
+CFLAGS_aegis128-neon-inner.o += -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_PCRYPT) += pcrypt.o
 obj-$(CONFIG_CRYPTO_CRYPTD) += cryptd.o
 obj-$(CONFIG_CRYPTO_DES) += des_generic.o

crypto/aegis128-neon-inner.c

@@ -0,0 +1,147 @@
+// 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>
+
+#define AES_ROUND	"aese %0.16b, %1.16b \n\t aesmc %0.16b, %0.16b"
+#else
+#include <arm_neon.h>
+
+#define AES_ROUND	"aese.8 %q0, %q1 \n\t aesmc.8 %q0, %q0"
+#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_load_state_neon(const void *state)
+{
+	return (struct aegis128_state){ {
+		vld1q_u8(state),
+		vld1q_u8(state + 16),
+		vld1q_u8(state + 32),
+		vld1q_u8(state + 48),
+		vld1q_u8(state + 64)
+	} };
+}
+
+static void aegis128_save_state_neon(struct aegis128_state st, void *state)
+{
+	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]);
+}
+
+static inline __attribute__((always_inline))
+uint8x16_t aegis_aes_round(uint8x16_t w)
+{
+	uint8x16_t z = {};
+
+	/*
+	 * We use inline asm here instead of the vaeseq_u8/vaesmcq_u8 intrinsics
+	 * to force the compiler to issue the aese/aesmc instructions in pairs.
+	 * This is much faster on many cores, where the instruction pair can
+	 * execute in a single cycle.
+	 */
+	asm(AES_ROUND : "+w"(w) : "w"(z));
+	return w;
+}
+
+static inline __attribute__((always_inline))
+struct aegis128_state aegis128_update_neon(struct aegis128_state st,
+					   uint8x16_t m)
+{
+	m       ^= aegis_aes_round(st.v[4]);
+	st.v[4] ^= aegis_aes_round(st.v[3]);
+	st.v[3] ^= aegis_aes_round(st.v[2]);
+	st.v[2] ^= aegis_aes_round(st.v[1]);
+	st.v[1] ^= aegis_aes_round(st.v[0]);
+	st.v[0] ^= m;
+
+	return st;
+}
+
+void crypto_aegis128_update_neon(void *state, const void *msg)
+{
+	struct aegis128_state st = aegis128_load_state_neon(state);
+
+	st = aegis128_update_neon(st, vld1q_u8(msg));
+
+	aegis128_save_state_neon(st, state);
+}
+
+void crypto_aegis128_encrypt_chunk_neon(void *state, void *dst, const void *src,
+					unsigned int size)
+{
+	struct aegis128_state st = aegis128_load_state_neon(state);
+	uint8x16_t msg;
+
+	while (size >= AEGIS_BLOCK_SIZE) {
+		uint8x16_t s = st.v[1] ^ (st.v[2] & st.v[3]) ^ st.v[4];
+
+		msg = vld1q_u8(src);
+		st = aegis128_update_neon(st, msg);
+		vst1q_u8(dst, msg ^ s);
+
+		size -= AEGIS_BLOCK_SIZE;
+		src += AEGIS_BLOCK_SIZE;
+		dst += AEGIS_BLOCK_SIZE;
+	}
+
+	if (size > 0) {
+		uint8x16_t s = st.v[1] ^ (st.v[2] & st.v[3]) ^ st.v[4];
+		uint8_t buf[AEGIS_BLOCK_SIZE] = {};
+
+		memcpy(buf, src, size);
+		msg = vld1q_u8(buf);
+		st = aegis128_update_neon(st, msg);
+		vst1q_u8(buf, msg ^ s);
+		memcpy(dst, buf, size);
+	}
+
+	aegis128_save_state_neon(st, state);
+}
+
+void crypto_aegis128_decrypt_chunk_neon(void *state, void *dst, const void *src,
+					unsigned int size)
+{
+	struct aegis128_state st = aegis128_load_state_neon(state);
+	uint8x16_t msg;
+
+	while (size >= AEGIS_BLOCK_SIZE) {
+		msg = vld1q_u8(src) ^ st.v[1] ^ (st.v[2] & st.v[3]) ^ st.v[4];
+		st = aegis128_update_neon(st, msg);
+		vst1q_u8(dst, msg);
+
+		size -= AEGIS_BLOCK_SIZE;
+		src += AEGIS_BLOCK_SIZE;
+		dst += AEGIS_BLOCK_SIZE;
+	}
+
+	if (size > 0) {
+		uint8x16_t s = st.v[1] ^ (st.v[2] & st.v[3]) ^ st.v[4];
+		uint8_t buf[AEGIS_BLOCK_SIZE];
+
+		vst1q_u8(buf, s);
+		memcpy(buf, src, size);
+		msg = vld1q_u8(buf) ^ s;
+		vst1q_u8(buf, msg);
+		memcpy(dst, buf, size);
+
+		st = aegis128_update_neon(st, msg);
+	}
+
+	aegis128_save_state_neon(st, state);
+}

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();
+}