Re: [PATCH 1/4] crypto: arm64/nhpoly1305 - add NEON-accelerated NHPoly1305

From: Ard Biesheuvel
Date: Fri Nov 30 2018 - 07:58:26 EST

On Thu, 29 Nov 2018 at 07:35, Eric Biggers <ebiggers@xxxxxxxxxx> wrote:
>
> From: Eric Biggers <ebiggers@xxxxxxxxxx>
>
> Add an ARM64 NEON implementation of NHPoly1305, an Î-almost-â-universal
> hash function used in the Adiantum encryption mode. For now, only the
> NH portion is actually NEON-accelerated; the Poly1305 part is less
> performance-critical so is just implemented in C.
>
> Signed-off-by: Eric Biggers <ebiggers@xxxxxxxxxx>
> ---
> arch/arm64/crypto/Kconfig | 5 ++
> arch/arm64/crypto/Makefile | 3 +
> arch/arm64/crypto/nh-neon-core.S | 103 +++++++++++++++++++++++
> arch/arm64/crypto/nhpoly1305-neon-glue.c | 77 +++++++++++++++++
> 4 files changed, 188 insertions(+)
> create mode 100644 arch/arm64/crypto/nh-neon-core.S
> create mode 100644 arch/arm64/crypto/nhpoly1305-neon-glue.c
>
> diff --git a/arch/arm64/crypto/Kconfig b/arch/arm64/crypto/Kconfig
> index a5606823ed4da..3f5aeb786192b 100644
> --- a/arch/arm64/crypto/Kconfig
> +++ b/arch/arm64/crypto/Kconfig
> @@ -106,6 +106,11 @@ config CRYPTO_CHACHA20_NEON
> select CRYPTO_BLKCIPHER
> select CRYPTO_CHACHA20
>
> +config CRYPTO_NHPOLY1305_NEON
> + tristate "NHPoly1305 hash function using NEON instructions (for Adiantum)"
> + depends on KERNEL_MODE_NEON
> + select CRYPTO_NHPOLY1305
> +
> config CRYPTO_AES_ARM64_BS
> tristate "AES in ECB/CBC/CTR/XTS modes using bit-sliced NEON algorithm"
> depends on KERNEL_MODE_NEON
> diff --git a/arch/arm64/crypto/Makefile b/arch/arm64/crypto/Makefile
> index f476fede09ba4..125dbb10a93ed 100644
> --- a/arch/arm64/crypto/Makefile
> +++ b/arch/arm64/crypto/Makefile
> @@ -53,6 +53,9 @@ sha512-arm64-y := sha512-glue.o sha512-core.o
> obj-$(CONFIG_CRYPTO_CHACHA20_NEON) += chacha20-neon.o
> chacha20-neon-y := chacha20-neon-core.o chacha20-neon-glue.o
>
> +obj-$(CONFIG_CRYPTO_NHPOLY1305_NEON) += nhpoly1305-neon.o
> +nhpoly1305-neon-y := nh-neon-core.o nhpoly1305-neon-glue.o
> +
> obj-$(CONFIG_CRYPTO_AES_ARM64) += aes-arm64.o
> aes-arm64-y := aes-cipher-core.o aes-cipher-glue.o
>
> diff --git a/arch/arm64/crypto/nh-neon-core.S b/arch/arm64/crypto/nh-neon-core.S
> new file mode 100644
> index 0000000000000..e08d57676127a
> --- /dev/null
> +++ b/arch/arm64/crypto/nh-neon-core.S
> @@ -0,0 +1,103 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +/*
> + * NH - Î-almost-universal hash function, ARM64 NEON accelerated version
> + *
> + * Copyright 2018 Google LLC
> + *
> + * Author: Eric Biggers <ebiggers@xxxxxxxxxx>
> + */
> +
> +#include <linux/linkage.h>
> +
> + KEY .req x0
> + MESSAGE .req x1
> + MESSAGE_LEN .req x2
> + HASH .req x3
> +
> + PASS0_SUMS .req v0
> + PASS1_SUMS .req v1
> + PASS2_SUMS .req v2
> + PASS3_SUMS .req v3
> + K0 .req v4
> + K1 .req v5
> + K2 .req v6
> + K3 .req v7
> + T0 .req v8
> + T1 .req v9
> + T2 .req v10
> + T3 .req v11
> + T4 .req v12
> + T5 .req v13
> + T6 .req v14
> + T7 .req v15
> +
> +.macro _nh_stride k0, k1, k2, k3
> +
> + // Load next message stride
> + ld1 {T3.16b}, [MESSAGE], #16
> +
> + // Load next key stride
> + ld1 {\k3\().4s}, [KEY], #16
> +
> + // Add message words to key words
> + add T0.4s, T3.4s, \k0\().4s
> + add T1.4s, T3.4s, \k1\().4s
> + add T2.4s, T3.4s, \k2\().4s
> + add T3.4s, T3.4s, \k3\().4s
> +
> + // Multiply 32x32 => 64 and accumulate
> + mov T4.d[0], T0.d[1]
> + mov T5.d[0], T1.d[1]
> + mov T6.d[0], T2.d[1]
> + mov T7.d[0], T3.d[1]

Nit: gmail mangles the whitespace anyway, but in the patch, there is
some whitespace damage here

With that fixed,

Reviewed-by: Ard Biesheuvel <ard.biesheuvel@xxxxxxxxxx>
Tested-by: Ard Biesheuvel <ard.biesheuvel@xxxxxxxxxx> # big-endian


> + umlal PASS0_SUMS.2d, T0.2s, T4.2s
> + umlal PASS1_SUMS.2d, T1.2s, T5.2s
> + umlal PASS2_SUMS.2d, T2.2s, T6.2s
> + umlal PASS3_SUMS.2d, T3.2s, T7.2s
> +.endm
> +
> +/*
> + * void nh_neon(const u32 *key, const u8 *message, size_t message_len,
> + * u8 hash[NH_HASH_BYTES])
> + *
> + * It's guaranteed that message_len % 16 == 0.
> + */
> +ENTRY(nh_neon)
> +
> + ld1 {K0.4s,K1.4s}, [KEY], #32
> + movi PASS0_SUMS.2d, #0
> + movi PASS1_SUMS.2d, #0
> + ld1 {K2.4s}, [KEY], #16
> + movi PASS2_SUMS.2d, #0
> + movi PASS3_SUMS.2d, #0
> +
> + subs MESSAGE_LEN, MESSAGE_LEN, #64
> + blt .Lloop4_done
> +.Lloop4:
> + _nh_stride K0, K1, K2, K3
> + _nh_stride K1, K2, K3, K0
> + _nh_stride K2, K3, K0, K1
> + _nh_stride K3, K0, K1, K2
> + subs MESSAGE_LEN, MESSAGE_LEN, #64
> + bge .Lloop4
> +
> +.Lloop4_done:
> + ands MESSAGE_LEN, MESSAGE_LEN, #63
> + beq .Ldone
> + _nh_stride K0, K1, K2, K3
> +
> + subs MESSAGE_LEN, MESSAGE_LEN, #16
> + beq .Ldone
> + _nh_stride K1, K2, K3, K0
> +
> + subs MESSAGE_LEN, MESSAGE_LEN, #16
> + beq .Ldone
> + _nh_stride K2, K3, K0, K1
> +
> +.Ldone:
> + // Sum the accumulators for each pass, then store the sums to 'hash'
> + addp T0.2d, PASS0_SUMS.2d, PASS1_SUMS.2d
> + addp T1.2d, PASS2_SUMS.2d, PASS3_SUMS.2d
> + st1 {T0.16b,T1.16b}, [HASH]
> + ret
> +ENDPROC(nh_neon)
> diff --git a/arch/arm64/crypto/nhpoly1305-neon-glue.c b/arch/arm64/crypto/nhpoly1305-neon-glue.c
> new file mode 100644
> index 0000000000000..22cc32ac9448d
> --- /dev/null
> +++ b/arch/arm64/crypto/nhpoly1305-neon-glue.c
> @@ -0,0 +1,77 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * NHPoly1305 - Î-almost-â-universal hash function for Adiantum
> + * (ARM64 NEON accelerated version)
> + *
> + * Copyright 2018 Google LLC
> + */
> +
> +#include <asm/neon.h>
> +#include <asm/simd.h>
> +#include <crypto/internal/hash.h>
> +#include <crypto/nhpoly1305.h>
> +#include <linux/module.h>
> +
> +asmlinkage void nh_neon(const u32 *key, const u8 *message, size_t message_len,
> + u8 hash[NH_HASH_BYTES]);
> +
> +/* wrapper to avoid indirect call to assembly, which doesn't work with CFI */
> +static void _nh_neon(const u32 *key, const u8 *message, size_t message_len,
> + __le64 hash[NH_NUM_PASSES])
> +{
> + nh_neon(key, message, message_len, (u8 *)hash);
> +}
> +
> +static int nhpoly1305_neon_update(struct shash_desc *desc,
> + const u8 *src, unsigned int srclen)
> +{
> + if (srclen < 64 || !may_use_simd())
> + return crypto_nhpoly1305_update(desc, src, srclen);
> +
> + do {
> + unsigned int n = min_t(unsigned int, srclen, PAGE_SIZE);
> +
> + kernel_neon_begin();
> + crypto_nhpoly1305_update_helper(desc, src, n, _nh_neon);
> + kernel_neon_end();
> + src += n;
> + srclen -= n;
> + } while (srclen);
> + return 0;
> +}
> +
> +static struct shash_alg nhpoly1305_alg = {
> + .base.cra_name = "nhpoly1305",
> + .base.cra_driver_name = "nhpoly1305-neon",
> + .base.cra_priority = 200,
> + .base.cra_ctxsize = sizeof(struct nhpoly1305_key),
> + .base.cra_module = THIS_MODULE,
> + .digestsize = POLY1305_DIGEST_SIZE,
> + .init = crypto_nhpoly1305_init,
> + .update = nhpoly1305_neon_update,
> + .final = crypto_nhpoly1305_final,
> + .setkey = crypto_nhpoly1305_setkey,
> + .descsize = sizeof(struct nhpoly1305_state),
> +};
> +
> +static int __init nhpoly1305_mod_init(void)
> +{
> + if (!(elf_hwcap & HWCAP_ASIMD))
> + return -ENODEV;
> +
> + return crypto_register_shash(&nhpoly1305_alg);
> +}
> +
> +static void __exit nhpoly1305_mod_exit(void)
> +{
> + crypto_unregister_shash(&nhpoly1305_alg);
> +}
> +
> +module_init(nhpoly1305_mod_init);
> +module_exit(nhpoly1305_mod_exit);
> +
> +MODULE_DESCRIPTION("NHPoly1305 Î-almost-â-universal hash function (NEON-accelerated)");
> +MODULE_LICENSE("GPL v2");
> +MODULE_AUTHOR("Eric Biggers <ebiggers@xxxxxxxxxx>");
> +MODULE_ALIAS_CRYPTO("nhpoly1305");
> +MODULE_ALIAS_CRYPTO("nhpoly1305-neon");
> --
> 2.19.2
>