Revert "crypto: aegis128 - add support for SIMD acceleration"
authorHerbert Xu <herbert@gondor.apana.org.au>
Fri, 2 Aug 2019 03:31:35 +0000 (13:31 +1000)
committerHerbert Xu <herbert@gondor.apana.org.au>
Fri, 2 Aug 2019 03:31:35 +0000 (13:31 +1000)
This reverts commit ecc8bc81f2fb3976737ef312f824ba6053aa3590
("crypto: aegis128 - provide a SIMD implementation based on NEON
intrinsics") and commit 7cdc0ddbf74a19cecb2f0e9efa2cae9d3c665189
("crypto: aegis128 - add support for SIMD acceleration").

They cause compile errors on platforms other than ARM because
the mechanism to selectively compile the SIMD code is broken.

Repoted-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
crypto/Kconfig
crypto/Makefile
crypto/aegis128-core.c [deleted file]
crypto/aegis128-neon-inner.c [deleted file]
crypto/aegis128-neon.c [deleted file]
crypto/aegis128.c [new file with mode: 0644]

index 455a3354e2914f1a369db27b87344327fe13af84..8880c1fc51d808d78c3e4ec6f201d4a52c4583cc 100644 (file)
@@ -306,11 +306,6 @@ 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
index b3e16b4fb414eaccc06c480378cc45bacf479e71..93375e124ff7f4374d87cc697571bd9b9867c9a8 100644 (file)
@@ -90,18 +90,6 @@ 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-core.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-$(CONFIG_CRYPTO_PCRYPT) += pcrypt.o
 obj-$(CONFIG_CRYPTO_CRYPTD) += cryptd.o
 obj-$(CONFIG_CRYPTO_DES) += des_generic.o
diff --git a/crypto/aegis128-core.c b/crypto/aegis128-core.c
deleted file mode 100644 (file)
index f815b46..0000000
+++ /dev/null
@@ -1,481 +0,0 @@
-// SPDX-License-Identifier: GPL-2.0-or-later
-/*
- * The AEGIS-128 Authenticated-Encryption Algorithm
- *
- * Copyright (c) 2017-2018 Ondrej Mosnacek <omosnacek@gmail.com>
- * Copyright (C) 2017-2018 Red Hat, Inc. All rights reserved.
- */
-
-#include <crypto/algapi.h>
-#include <crypto/internal/aead.h>
-#include <crypto/internal/simd.h>
-#include <crypto/internal/skcipher.h>
-#include <crypto/scatterwalk.h>
-#include <linux/err.h>
-#include <linux/init.h>
-#include <linux/kernel.h>
-#include <linux/module.h>
-#include <linux/scatterlist.h>
-#include <asm/simd.h>
-
-#include "aegis.h"
-
-#define AEGIS128_NONCE_SIZE 16
-#define AEGIS128_STATE_BLOCKS 5
-#define AEGIS128_KEY_SIZE 16
-#define AEGIS128_MIN_AUTH_SIZE 8
-#define AEGIS128_MAX_AUTH_SIZE 16
-
-struct aegis_state {
-       union aegis_block blocks[AEGIS128_STATE_BLOCKS];
-};
-
-struct aegis_ctx {
-       union aegis_block key;
-};
-
-struct aegis128_ops {
-       int (*skcipher_walk_init)(struct skcipher_walk *walk,
-                                 struct aead_request *req, bool atomic);
-
-       void (*crypt_chunk)(struct aegis_state *state, u8 *dst,
-                           const u8 *src, unsigned int size);
-};
-
-static bool have_simd;
-
-bool crypto_aegis128_have_simd(void);
-void crypto_aegis128_update_simd(struct aegis_state *state, const void *msg);
-void crypto_aegis128_encrypt_chunk_simd(struct aegis_state *state, u8 *dst,
-                                       const u8 *src, unsigned int size);
-void crypto_aegis128_decrypt_chunk_simd(struct aegis_state *state, u8 *dst,
-                                       const u8 *src, unsigned int size);
-
-static void crypto_aegis128_update(struct aegis_state *state)
-{
-       union aegis_block tmp;
-       unsigned int i;
-
-       tmp = state->blocks[AEGIS128_STATE_BLOCKS - 1];
-       for (i = AEGIS128_STATE_BLOCKS - 1; i > 0; i--)
-               crypto_aegis_aesenc(&state->blocks[i], &state->blocks[i - 1],
-                                   &state->blocks[i]);
-       crypto_aegis_aesenc(&state->blocks[0], &tmp, &state->blocks[0]);
-}
-
-static void crypto_aegis128_update_a(struct aegis_state *state,
-                                    const union aegis_block *msg)
-{
-       if (have_simd && crypto_simd_usable()) {
-               crypto_aegis128_update_simd(state, msg);
-               return;
-       }
-
-       crypto_aegis128_update(state);
-       crypto_aegis_block_xor(&state->blocks[0], msg);
-}
-
-static void crypto_aegis128_update_u(struct aegis_state *state, const void *msg)
-{
-       if (have_simd && crypto_simd_usable()) {
-               crypto_aegis128_update_simd(state, msg);
-               return;
-       }
-
-       crypto_aegis128_update(state);
-       crypto_xor(state->blocks[0].bytes, msg, AEGIS_BLOCK_SIZE);
-}
-
-static void crypto_aegis128_init(struct aegis_state *state,
-                                const union aegis_block *key,
-                                const u8 *iv)
-{
-       union aegis_block key_iv;
-       unsigned int i;
-
-       key_iv = *key;
-       crypto_xor(key_iv.bytes, iv, AEGIS_BLOCK_SIZE);
-
-       state->blocks[0] = key_iv;
-       state->blocks[1] = crypto_aegis_const[1];
-       state->blocks[2] = crypto_aegis_const[0];
-       state->blocks[3] = *key;
-       state->blocks[4] = *key;
-
-       crypto_aegis_block_xor(&state->blocks[3], &crypto_aegis_const[0]);
-       crypto_aegis_block_xor(&state->blocks[4], &crypto_aegis_const[1]);
-
-       for (i = 0; i < 5; i++) {
-               crypto_aegis128_update_a(state, key);
-               crypto_aegis128_update_a(state, &key_iv);
-       }
-}
-
-static void crypto_aegis128_ad(struct aegis_state *state,
-                              const u8 *src, unsigned int size)
-{
-       if (AEGIS_ALIGNED(src)) {
-               const union aegis_block *src_blk =
-                               (const union aegis_block *)src;
-
-               while (size >= AEGIS_BLOCK_SIZE) {
-                       crypto_aegis128_update_a(state, src_blk);
-
-                       size -= AEGIS_BLOCK_SIZE;
-                       src_blk++;
-               }
-       } else {
-               while (size >= AEGIS_BLOCK_SIZE) {
-                       crypto_aegis128_update_u(state, src);
-
-                       size -= AEGIS_BLOCK_SIZE;
-                       src += AEGIS_BLOCK_SIZE;
-               }
-       }
-}
-
-static void crypto_aegis128_encrypt_chunk(struct aegis_state *state, u8 *dst,
-                                         const u8 *src, unsigned int size)
-{
-       union aegis_block tmp;
-
-       if (AEGIS_ALIGNED(src) && AEGIS_ALIGNED(dst)) {
-               while (size >= AEGIS_BLOCK_SIZE) {
-                       union aegis_block *dst_blk =
-                                       (union aegis_block *)dst;
-                       const union aegis_block *src_blk =
-                                       (const union aegis_block *)src;
-
-                       tmp = state->blocks[2];
-                       crypto_aegis_block_and(&tmp, &state->blocks[3]);
-                       crypto_aegis_block_xor(&tmp, &state->blocks[4]);
-                       crypto_aegis_block_xor(&tmp, &state->blocks[1]);
-                       crypto_aegis_block_xor(&tmp, src_blk);
-
-                       crypto_aegis128_update_a(state, src_blk);
-
-                       *dst_blk = tmp;
-
-                       size -= AEGIS_BLOCK_SIZE;
-                       src += AEGIS_BLOCK_SIZE;
-                       dst += AEGIS_BLOCK_SIZE;
-               }
-       } else {
-               while (size >= AEGIS_BLOCK_SIZE) {
-                       tmp = state->blocks[2];
-                       crypto_aegis_block_and(&tmp, &state->blocks[3]);
-                       crypto_aegis_block_xor(&tmp, &state->blocks[4]);
-                       crypto_aegis_block_xor(&tmp, &state->blocks[1]);
-                       crypto_xor(tmp.bytes, src, AEGIS_BLOCK_SIZE);
-
-                       crypto_aegis128_update_u(state, src);
-
-                       memcpy(dst, tmp.bytes, AEGIS_BLOCK_SIZE);
-
-                       size -= AEGIS_BLOCK_SIZE;
-                       src += AEGIS_BLOCK_SIZE;
-                       dst += AEGIS_BLOCK_SIZE;
-               }
-       }
-
-       if (size > 0) {
-               union aegis_block msg = {};
-               memcpy(msg.bytes, src, size);
-
-               tmp = state->blocks[2];
-               crypto_aegis_block_and(&tmp, &state->blocks[3]);
-               crypto_aegis_block_xor(&tmp, &state->blocks[4]);
-               crypto_aegis_block_xor(&tmp, &state->blocks[1]);
-
-               crypto_aegis128_update_a(state, &msg);
-
-               crypto_aegis_block_xor(&msg, &tmp);
-
-               memcpy(dst, msg.bytes, size);
-       }
-}
-
-static void crypto_aegis128_decrypt_chunk(struct aegis_state *state, u8 *dst,
-                                         const u8 *src, unsigned int size)
-{
-       union aegis_block tmp;
-
-       if (AEGIS_ALIGNED(src) && AEGIS_ALIGNED(dst)) {
-               while (size >= AEGIS_BLOCK_SIZE) {
-                       union aegis_block *dst_blk =
-                                       (union aegis_block *)dst;
-                       const union aegis_block *src_blk =
-                                       (const union aegis_block *)src;
-
-                       tmp = state->blocks[2];
-                       crypto_aegis_block_and(&tmp, &state->blocks[3]);
-                       crypto_aegis_block_xor(&tmp, &state->blocks[4]);
-                       crypto_aegis_block_xor(&tmp, &state->blocks[1]);
-                       crypto_aegis_block_xor(&tmp, src_blk);
-
-                       crypto_aegis128_update_a(state, &tmp);
-
-                       *dst_blk = tmp;
-
-                       size -= AEGIS_BLOCK_SIZE;
-                       src += AEGIS_BLOCK_SIZE;
-                       dst += AEGIS_BLOCK_SIZE;
-               }
-       } else {
-               while (size >= AEGIS_BLOCK_SIZE) {
-                       tmp = state->blocks[2];
-                       crypto_aegis_block_and(&tmp, &state->blocks[3]);
-                       crypto_aegis_block_xor(&tmp, &state->blocks[4]);
-                       crypto_aegis_block_xor(&tmp, &state->blocks[1]);
-                       crypto_xor(tmp.bytes, src, AEGIS_BLOCK_SIZE);
-
-                       crypto_aegis128_update_a(state, &tmp);
-
-                       memcpy(dst, tmp.bytes, AEGIS_BLOCK_SIZE);
-
-                       size -= AEGIS_BLOCK_SIZE;
-                       src += AEGIS_BLOCK_SIZE;
-                       dst += AEGIS_BLOCK_SIZE;
-               }
-       }
-
-       if (size > 0) {
-               union aegis_block msg = {};
-               memcpy(msg.bytes, src, size);
-
-               tmp = state->blocks[2];
-               crypto_aegis_block_and(&tmp, &state->blocks[3]);
-               crypto_aegis_block_xor(&tmp, &state->blocks[4]);
-               crypto_aegis_block_xor(&tmp, &state->blocks[1]);
-               crypto_aegis_block_xor(&msg, &tmp);
-
-               memset(msg.bytes + size, 0, AEGIS_BLOCK_SIZE - size);
-
-               crypto_aegis128_update_a(state, &msg);
-
-               memcpy(dst, msg.bytes, size);
-       }
-}
-
-static void crypto_aegis128_process_ad(struct aegis_state *state,
-                                      struct scatterlist *sg_src,
-                                      unsigned int assoclen)
-{
-       struct scatter_walk walk;
-       union aegis_block buf;
-       unsigned int pos = 0;
-
-       scatterwalk_start(&walk, sg_src);
-       while (assoclen != 0) {
-               unsigned int size = scatterwalk_clamp(&walk, assoclen);
-               unsigned int left = size;
-               void *mapped = scatterwalk_map(&walk);
-               const u8 *src = (const u8 *)mapped;
-
-               if (pos + size >= AEGIS_BLOCK_SIZE) {
-                       if (pos > 0) {
-                               unsigned int fill = AEGIS_BLOCK_SIZE - pos;
-                               memcpy(buf.bytes + pos, src, fill);
-                               crypto_aegis128_update_a(state, &buf);
-                               pos = 0;
-                               left -= fill;
-                               src += fill;
-                       }
-
-                       crypto_aegis128_ad(state, src, left);
-                       src += left & ~(AEGIS_BLOCK_SIZE - 1);
-                       left &= AEGIS_BLOCK_SIZE - 1;
-               }
-
-               memcpy(buf.bytes + pos, src, left);
-
-               pos += left;
-               assoclen -= size;
-               scatterwalk_unmap(mapped);
-               scatterwalk_advance(&walk, size);
-               scatterwalk_done(&walk, 0, assoclen);
-       }
-
-       if (pos > 0) {
-               memset(buf.bytes + pos, 0, AEGIS_BLOCK_SIZE - pos);
-               crypto_aegis128_update_a(state, &buf);
-       }
-}
-
-static void crypto_aegis128_process_crypt(struct aegis_state *state,
-                                         struct aead_request *req,
-                                         const struct aegis128_ops *ops)
-{
-       struct skcipher_walk walk;
-
-       ops->skcipher_walk_init(&walk, req, false);
-
-       while (walk.nbytes) {
-               unsigned int nbytes = walk.nbytes;
-
-               if (nbytes < walk.total)
-                       nbytes = round_down(nbytes, walk.stride);
-
-               ops->crypt_chunk(state, walk.dst.virt.addr, walk.src.virt.addr,
-                                nbytes);
-
-               skcipher_walk_done(&walk, walk.nbytes - nbytes);
-       }
-}
-
-static void crypto_aegis128_final(struct aegis_state *state,
-                                 union aegis_block *tag_xor,
-                                 u64 assoclen, u64 cryptlen)
-{
-       u64 assocbits = assoclen * 8;
-       u64 cryptbits = cryptlen * 8;
-
-       union aegis_block tmp;
-       unsigned int i;
-
-       tmp.words64[0] = cpu_to_le64(assocbits);
-       tmp.words64[1] = cpu_to_le64(cryptbits);
-
-       crypto_aegis_block_xor(&tmp, &state->blocks[3]);
-
-       for (i = 0; i < 7; i++)
-               crypto_aegis128_update_a(state, &tmp);
-
-       for (i = 0; i < AEGIS128_STATE_BLOCKS; i++)
-               crypto_aegis_block_xor(tag_xor, &state->blocks[i]);
-}
-
-static int crypto_aegis128_setkey(struct crypto_aead *aead, const u8 *key,
-                                 unsigned int keylen)
-{
-       struct aegis_ctx *ctx = crypto_aead_ctx(aead);
-
-       if (keylen != AEGIS128_KEY_SIZE) {
-               crypto_aead_set_flags(aead, CRYPTO_TFM_RES_BAD_KEY_LEN);
-               return -EINVAL;
-       }
-
-       memcpy(ctx->key.bytes, key, AEGIS128_KEY_SIZE);
-       return 0;
-}
-
-static int crypto_aegis128_setauthsize(struct crypto_aead *tfm,
-                                      unsigned int authsize)
-{
-       if (authsize > AEGIS128_MAX_AUTH_SIZE)
-               return -EINVAL;
-       if (authsize < AEGIS128_MIN_AUTH_SIZE)
-               return -EINVAL;
-       return 0;
-}
-
-static void crypto_aegis128_crypt(struct aead_request *req,
-                                 union aegis_block *tag_xor,
-                                 unsigned int cryptlen,
-                                 const struct aegis128_ops *ops)
-{
-       struct crypto_aead *tfm = crypto_aead_reqtfm(req);
-       struct aegis_ctx *ctx = crypto_aead_ctx(tfm);
-       struct aegis_state state;
-
-       crypto_aegis128_init(&state, &ctx->key, req->iv);
-       crypto_aegis128_process_ad(&state, req->src, req->assoclen);
-       crypto_aegis128_process_crypt(&state, req, ops);
-       crypto_aegis128_final(&state, tag_xor, req->assoclen, cryptlen);
-}
-
-static int crypto_aegis128_encrypt(struct aead_request *req)
-{
-       const struct aegis128_ops *ops = &(struct aegis128_ops){
-               .skcipher_walk_init = skcipher_walk_aead_encrypt,
-               .crypt_chunk = crypto_aegis128_encrypt_chunk,
-       };
-
-       struct crypto_aead *tfm = crypto_aead_reqtfm(req);
-       union aegis_block tag = {};
-       unsigned int authsize = crypto_aead_authsize(tfm);
-       unsigned int cryptlen = req->cryptlen;
-
-       if (have_simd && crypto_simd_usable())
-               ops = &(struct aegis128_ops){
-                       .skcipher_walk_init = skcipher_walk_aead_encrypt,
-                       .crypt_chunk = crypto_aegis128_encrypt_chunk_simd };
-
-       crypto_aegis128_crypt(req, &tag, cryptlen, ops);
-
-       scatterwalk_map_and_copy(tag.bytes, req->dst, req->assoclen + cryptlen,
-                                authsize, 1);
-       return 0;
-}
-
-static int crypto_aegis128_decrypt(struct aead_request *req)
-{
-       const struct aegis128_ops *ops = &(struct aegis128_ops){
-               .skcipher_walk_init = skcipher_walk_aead_decrypt,
-               .crypt_chunk = crypto_aegis128_decrypt_chunk,
-       };
-       static const u8 zeros[AEGIS128_MAX_AUTH_SIZE] = {};
-
-       struct crypto_aead *tfm = crypto_aead_reqtfm(req);
-       union aegis_block tag;
-       unsigned int authsize = crypto_aead_authsize(tfm);
-       unsigned int cryptlen = req->cryptlen - authsize;
-
-       scatterwalk_map_and_copy(tag.bytes, req->src, req->assoclen + cryptlen,
-                                authsize, 0);
-
-       if (have_simd && crypto_simd_usable())
-               ops = &(struct aegis128_ops){
-                       .skcipher_walk_init = skcipher_walk_aead_decrypt,
-                       .crypt_chunk = crypto_aegis128_decrypt_chunk_simd };
-
-       crypto_aegis128_crypt(req, &tag, cryptlen, ops);
-
-       return crypto_memneq(tag.bytes, zeros, authsize) ? -EBADMSG : 0;
-}
-
-static struct aead_alg crypto_aegis128_alg = {
-       .setkey = crypto_aegis128_setkey,
-       .setauthsize = crypto_aegis128_setauthsize,
-       .encrypt = crypto_aegis128_encrypt,
-       .decrypt = crypto_aegis128_decrypt,
-
-       .ivsize = AEGIS128_NONCE_SIZE,
-       .maxauthsize = AEGIS128_MAX_AUTH_SIZE,
-       .chunksize = AEGIS_BLOCK_SIZE,
-
-       .base = {
-               .cra_blocksize = 1,
-               .cra_ctxsize = sizeof(struct aegis_ctx),
-               .cra_alignmask = 0,
-
-               .cra_priority = 100,
-
-               .cra_name = "aegis128",
-               .cra_driver_name = "aegis128-generic",
-
-               .cra_module = THIS_MODULE,
-       }
-};
-
-static int __init crypto_aegis128_module_init(void)
-{
-       if (IS_ENABLED(CONFIG_CRYPTO_AEGIS128_SIMD))
-               have_simd = crypto_aegis128_have_simd();
-
-       return crypto_register_aead(&crypto_aegis128_alg);
-}
-
-static void __exit crypto_aegis128_module_exit(void)
-{
-       crypto_unregister_aead(&crypto_aegis128_alg);
-}
-
-subsys_initcall(crypto_aegis128_module_init);
-module_exit(crypto_aegis128_module_exit);
-
-MODULE_LICENSE("GPL");
-MODULE_AUTHOR("Ondrej Mosnacek <omosnacek@gmail.com>");
-MODULE_DESCRIPTION("AEGIS-128 AEAD algorithm");
-MODULE_ALIAS_CRYPTO("aegis128");
-MODULE_ALIAS_CRYPTO("aegis128-generic");
diff --git a/crypto/aegis128-neon-inner.c b/crypto/aegis128-neon-inner.c
deleted file mode 100644 (file)
index 26e9450..0000000
+++ /dev/null
@@ -1,149 +0,0 @@
-// 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 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 struct aegis128_state aegis128_update_neon(struct aegis128_state st,
-                                                 uint8x16_t m)
-{
-       uint8x16_t t;
-
-       t        = 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] ^= aegis_aes_round(st.v[4]) ^ m;
-       st.v[4] ^= t;
-
-       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 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);
-       }
-
-       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 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);
-       }
-
-       aegis128_save_state_neon(st, state);
-}
diff --git a/crypto/aegis128-neon.c b/crypto/aegis128-neon.c
deleted file mode 100644 (file)
index c1c0a16..0000000
+++ /dev/null
@@ -1,43 +0,0 @@
-// 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();
-}
diff --git a/crypto/aegis128.c b/crypto/aegis128.c
new file mode 100644 (file)
index 0000000..32840d5
--- /dev/null
@@ -0,0 +1,447 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * The AEGIS-128 Authenticated-Encryption Algorithm
+ *
+ * Copyright (c) 2017-2018 Ondrej Mosnacek <omosnacek@gmail.com>
+ * Copyright (C) 2017-2018 Red Hat, Inc. All rights reserved.
+ */
+
+#include <crypto/algapi.h>
+#include <crypto/internal/aead.h>
+#include <crypto/internal/skcipher.h>
+#include <crypto/scatterwalk.h>
+#include <linux/err.h>
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/scatterlist.h>
+
+#include "aegis.h"
+
+#define AEGIS128_NONCE_SIZE 16
+#define AEGIS128_STATE_BLOCKS 5
+#define AEGIS128_KEY_SIZE 16
+#define AEGIS128_MIN_AUTH_SIZE 8
+#define AEGIS128_MAX_AUTH_SIZE 16
+
+struct aegis_state {
+       union aegis_block blocks[AEGIS128_STATE_BLOCKS];
+};
+
+struct aegis_ctx {
+       union aegis_block key;
+};
+
+struct aegis128_ops {
+       int (*skcipher_walk_init)(struct skcipher_walk *walk,
+                                 struct aead_request *req, bool atomic);
+
+       void (*crypt_chunk)(struct aegis_state *state, u8 *dst,
+                           const u8 *src, unsigned int size);
+};
+
+static void crypto_aegis128_update(struct aegis_state *state)
+{
+       union aegis_block tmp;
+       unsigned int i;
+
+       tmp = state->blocks[AEGIS128_STATE_BLOCKS - 1];
+       for (i = AEGIS128_STATE_BLOCKS - 1; i > 0; i--)
+               crypto_aegis_aesenc(&state->blocks[i], &state->blocks[i - 1],
+                                   &state->blocks[i]);
+       crypto_aegis_aesenc(&state->blocks[0], &tmp, &state->blocks[0]);
+}
+
+static void crypto_aegis128_update_a(struct aegis_state *state,
+                                    const union aegis_block *msg)
+{
+       crypto_aegis128_update(state);
+       crypto_aegis_block_xor(&state->blocks[0], msg);
+}
+
+static void crypto_aegis128_update_u(struct aegis_state *state, const void *msg)
+{
+       crypto_aegis128_update(state);
+       crypto_xor(state->blocks[0].bytes, msg, AEGIS_BLOCK_SIZE);
+}
+
+static void crypto_aegis128_init(struct aegis_state *state,
+                                const union aegis_block *key,
+                                const u8 *iv)
+{
+       union aegis_block key_iv;
+       unsigned int i;
+
+       key_iv = *key;
+       crypto_xor(key_iv.bytes, iv, AEGIS_BLOCK_SIZE);
+
+       state->blocks[0] = key_iv;
+       state->blocks[1] = crypto_aegis_const[1];
+       state->blocks[2] = crypto_aegis_const[0];
+       state->blocks[3] = *key;
+       state->blocks[4] = *key;
+
+       crypto_aegis_block_xor(&state->blocks[3], &crypto_aegis_const[0]);
+       crypto_aegis_block_xor(&state->blocks[4], &crypto_aegis_const[1]);
+
+       for (i = 0; i < 5; i++) {
+               crypto_aegis128_update_a(state, key);
+               crypto_aegis128_update_a(state, &key_iv);
+       }
+}
+
+static void crypto_aegis128_ad(struct aegis_state *state,
+                              const u8 *src, unsigned int size)
+{
+       if (AEGIS_ALIGNED(src)) {
+               const union aegis_block *src_blk =
+                               (const union aegis_block *)src;
+
+               while (size >= AEGIS_BLOCK_SIZE) {
+                       crypto_aegis128_update_a(state, src_blk);
+
+                       size -= AEGIS_BLOCK_SIZE;
+                       src_blk++;
+               }
+       } else {
+               while (size >= AEGIS_BLOCK_SIZE) {
+                       crypto_aegis128_update_u(state, src);
+
+                       size -= AEGIS_BLOCK_SIZE;
+                       src += AEGIS_BLOCK_SIZE;
+               }
+       }
+}
+
+static void crypto_aegis128_encrypt_chunk(struct aegis_state *state, u8 *dst,
+                                         const u8 *src, unsigned int size)
+{
+       union aegis_block tmp;
+
+       if (AEGIS_ALIGNED(src) && AEGIS_ALIGNED(dst)) {
+               while (size >= AEGIS_BLOCK_SIZE) {
+                       union aegis_block *dst_blk =
+                                       (union aegis_block *)dst;
+                       const union aegis_block *src_blk =
+                                       (const union aegis_block *)src;
+
+                       tmp = state->blocks[2];
+                       crypto_aegis_block_and(&tmp, &state->blocks[3]);
+                       crypto_aegis_block_xor(&tmp, &state->blocks[4]);
+                       crypto_aegis_block_xor(&tmp, &state->blocks[1]);
+                       crypto_aegis_block_xor(&tmp, src_blk);
+
+                       crypto_aegis128_update_a(state, src_blk);
+
+                       *dst_blk = tmp;
+
+                       size -= AEGIS_BLOCK_SIZE;
+                       src += AEGIS_BLOCK_SIZE;
+                       dst += AEGIS_BLOCK_SIZE;
+               }
+       } else {
+               while (size >= AEGIS_BLOCK_SIZE) {
+                       tmp = state->blocks[2];
+                       crypto_aegis_block_and(&tmp, &state->blocks[3]);
+                       crypto_aegis_block_xor(&tmp, &state->blocks[4]);
+                       crypto_aegis_block_xor(&tmp, &state->blocks[1]);
+                       crypto_xor(tmp.bytes, src, AEGIS_BLOCK_SIZE);
+
+                       crypto_aegis128_update_u(state, src);
+
+                       memcpy(dst, tmp.bytes, AEGIS_BLOCK_SIZE);
+
+                       size -= AEGIS_BLOCK_SIZE;
+                       src += AEGIS_BLOCK_SIZE;
+                       dst += AEGIS_BLOCK_SIZE;
+               }
+       }
+
+       if (size > 0) {
+               union aegis_block msg = {};
+               memcpy(msg.bytes, src, size);
+
+               tmp = state->blocks[2];
+               crypto_aegis_block_and(&tmp, &state->blocks[3]);
+               crypto_aegis_block_xor(&tmp, &state->blocks[4]);
+               crypto_aegis_block_xor(&tmp, &state->blocks[1]);
+
+               crypto_aegis128_update_a(state, &msg);
+
+               crypto_aegis_block_xor(&msg, &tmp);
+
+               memcpy(dst, msg.bytes, size);
+       }
+}
+
+static void crypto_aegis128_decrypt_chunk(struct aegis_state *state, u8 *dst,
+                                         const u8 *src, unsigned int size)
+{
+       union aegis_block tmp;
+
+       if (AEGIS_ALIGNED(src) && AEGIS_ALIGNED(dst)) {
+               while (size >= AEGIS_BLOCK_SIZE) {
+                       union aegis_block *dst_blk =
+                                       (union aegis_block *)dst;
+                       const union aegis_block *src_blk =
+                                       (const union aegis_block *)src;
+
+                       tmp = state->blocks[2];
+                       crypto_aegis_block_and(&tmp, &state->blocks[3]);
+                       crypto_aegis_block_xor(&tmp, &state->blocks[4]);
+                       crypto_aegis_block_xor(&tmp, &state->blocks[1]);
+                       crypto_aegis_block_xor(&tmp, src_blk);
+
+                       crypto_aegis128_update_a(state, &tmp);
+
+                       *dst_blk = tmp;
+
+                       size -= AEGIS_BLOCK_SIZE;
+                       src += AEGIS_BLOCK_SIZE;
+                       dst += AEGIS_BLOCK_SIZE;
+               }
+       } else {
+               while (size >= AEGIS_BLOCK_SIZE) {
+                       tmp = state->blocks[2];
+                       crypto_aegis_block_and(&tmp, &state->blocks[3]);
+                       crypto_aegis_block_xor(&tmp, &state->blocks[4]);
+                       crypto_aegis_block_xor(&tmp, &state->blocks[1]);
+                       crypto_xor(tmp.bytes, src, AEGIS_BLOCK_SIZE);
+
+                       crypto_aegis128_update_a(state, &tmp);
+
+                       memcpy(dst, tmp.bytes, AEGIS_BLOCK_SIZE);
+
+                       size -= AEGIS_BLOCK_SIZE;
+                       src += AEGIS_BLOCK_SIZE;
+                       dst += AEGIS_BLOCK_SIZE;
+               }
+       }
+
+       if (size > 0) {
+               union aegis_block msg = {};
+               memcpy(msg.bytes, src, size);
+
+               tmp = state->blocks[2];
+               crypto_aegis_block_and(&tmp, &state->blocks[3]);
+               crypto_aegis_block_xor(&tmp, &state->blocks[4]);
+               crypto_aegis_block_xor(&tmp, &state->blocks[1]);
+               crypto_aegis_block_xor(&msg, &tmp);
+
+               memset(msg.bytes + size, 0, AEGIS_BLOCK_SIZE - size);
+
+               crypto_aegis128_update_a(state, &msg);
+
+               memcpy(dst, msg.bytes, size);
+       }
+}
+
+static void crypto_aegis128_process_ad(struct aegis_state *state,
+                                      struct scatterlist *sg_src,
+                                      unsigned int assoclen)
+{
+       struct scatter_walk walk;
+       union aegis_block buf;
+       unsigned int pos = 0;
+
+       scatterwalk_start(&walk, sg_src);
+       while (assoclen != 0) {
+               unsigned int size = scatterwalk_clamp(&walk, assoclen);
+               unsigned int left = size;
+               void *mapped = scatterwalk_map(&walk);
+               const u8 *src = (const u8 *)mapped;
+
+               if (pos + size >= AEGIS_BLOCK_SIZE) {
+                       if (pos > 0) {
+                               unsigned int fill = AEGIS_BLOCK_SIZE - pos;
+                               memcpy(buf.bytes + pos, src, fill);
+                               crypto_aegis128_update_a(state, &buf);
+                               pos = 0;
+                               left -= fill;
+                               src += fill;
+                       }
+
+                       crypto_aegis128_ad(state, src, left);
+                       src += left & ~(AEGIS_BLOCK_SIZE - 1);
+                       left &= AEGIS_BLOCK_SIZE - 1;
+               }
+
+               memcpy(buf.bytes + pos, src, left);
+
+               pos += left;
+               assoclen -= size;
+               scatterwalk_unmap(mapped);
+               scatterwalk_advance(&walk, size);
+               scatterwalk_done(&walk, 0, assoclen);
+       }
+
+       if (pos > 0) {
+               memset(buf.bytes + pos, 0, AEGIS_BLOCK_SIZE - pos);
+               crypto_aegis128_update_a(state, &buf);
+       }
+}
+
+static void crypto_aegis128_process_crypt(struct aegis_state *state,
+                                         struct aead_request *req,
+                                         const struct aegis128_ops *ops)
+{
+       struct skcipher_walk walk;
+
+       ops->skcipher_walk_init(&walk, req, false);
+
+       while (walk.nbytes) {
+               unsigned int nbytes = walk.nbytes;
+
+               if (nbytes < walk.total)
+                       nbytes = round_down(nbytes, walk.stride);
+
+               ops->crypt_chunk(state, walk.dst.virt.addr, walk.src.virt.addr,
+                                nbytes);
+
+               skcipher_walk_done(&walk, walk.nbytes - nbytes);
+       }
+}
+
+static void crypto_aegis128_final(struct aegis_state *state,
+                                 union aegis_block *tag_xor,
+                                 u64 assoclen, u64 cryptlen)
+{
+       u64 assocbits = assoclen * 8;
+       u64 cryptbits = cryptlen * 8;
+
+       union aegis_block tmp;
+       unsigned int i;
+
+       tmp.words64[0] = cpu_to_le64(assocbits);
+       tmp.words64[1] = cpu_to_le64(cryptbits);
+
+       crypto_aegis_block_xor(&tmp, &state->blocks[3]);
+
+       for (i = 0; i < 7; i++)
+               crypto_aegis128_update_a(state, &tmp);
+
+       for (i = 0; i < AEGIS128_STATE_BLOCKS; i++)
+               crypto_aegis_block_xor(tag_xor, &state->blocks[i]);
+}
+
+static int crypto_aegis128_setkey(struct crypto_aead *aead, const u8 *key,
+                                 unsigned int keylen)
+{
+       struct aegis_ctx *ctx = crypto_aead_ctx(aead);
+
+       if (keylen != AEGIS128_KEY_SIZE) {
+               crypto_aead_set_flags(aead, CRYPTO_TFM_RES_BAD_KEY_LEN);
+               return -EINVAL;
+       }
+
+       memcpy(ctx->key.bytes, key, AEGIS128_KEY_SIZE);
+       return 0;
+}
+
+static int crypto_aegis128_setauthsize(struct crypto_aead *tfm,
+                                      unsigned int authsize)
+{
+       if (authsize > AEGIS128_MAX_AUTH_SIZE)
+               return -EINVAL;
+       if (authsize < AEGIS128_MIN_AUTH_SIZE)
+               return -EINVAL;
+       return 0;
+}
+
+static void crypto_aegis128_crypt(struct aead_request *req,
+                                 union aegis_block *tag_xor,
+                                 unsigned int cryptlen,
+                                 const struct aegis128_ops *ops)
+{
+       struct crypto_aead *tfm = crypto_aead_reqtfm(req);
+       struct aegis_ctx *ctx = crypto_aead_ctx(tfm);
+       struct aegis_state state;
+
+       crypto_aegis128_init(&state, &ctx->key, req->iv);
+       crypto_aegis128_process_ad(&state, req->src, req->assoclen);
+       crypto_aegis128_process_crypt(&state, req, ops);
+       crypto_aegis128_final(&state, tag_xor, req->assoclen, cryptlen);
+}
+
+static int crypto_aegis128_encrypt(struct aead_request *req)
+{
+       static const struct aegis128_ops ops = {
+               .skcipher_walk_init = skcipher_walk_aead_encrypt,
+               .crypt_chunk = crypto_aegis128_encrypt_chunk,
+       };
+
+       struct crypto_aead *tfm = crypto_aead_reqtfm(req);
+       union aegis_block tag = {};
+       unsigned int authsize = crypto_aead_authsize(tfm);
+       unsigned int cryptlen = req->cryptlen;
+
+       crypto_aegis128_crypt(req, &tag, cryptlen, &ops);
+
+       scatterwalk_map_and_copy(tag.bytes, req->dst, req->assoclen + cryptlen,
+                                authsize, 1);
+       return 0;
+}
+
+static int crypto_aegis128_decrypt(struct aead_request *req)
+{
+       static const struct aegis128_ops ops = {
+               .skcipher_walk_init = skcipher_walk_aead_decrypt,
+               .crypt_chunk = crypto_aegis128_decrypt_chunk,
+       };
+       static const u8 zeros[AEGIS128_MAX_AUTH_SIZE] = {};
+
+       struct crypto_aead *tfm = crypto_aead_reqtfm(req);
+       union aegis_block tag;
+       unsigned int authsize = crypto_aead_authsize(tfm);
+       unsigned int cryptlen = req->cryptlen - authsize;
+
+       scatterwalk_map_and_copy(tag.bytes, req->src, req->assoclen + cryptlen,
+                                authsize, 0);
+
+       crypto_aegis128_crypt(req, &tag, cryptlen, &ops);
+
+       return crypto_memneq(tag.bytes, zeros, authsize) ? -EBADMSG : 0;
+}
+
+static struct aead_alg crypto_aegis128_alg = {
+       .setkey = crypto_aegis128_setkey,
+       .setauthsize = crypto_aegis128_setauthsize,
+       .encrypt = crypto_aegis128_encrypt,
+       .decrypt = crypto_aegis128_decrypt,
+
+       .ivsize = AEGIS128_NONCE_SIZE,
+       .maxauthsize = AEGIS128_MAX_AUTH_SIZE,
+       .chunksize = AEGIS_BLOCK_SIZE,
+
+       .base = {
+               .cra_blocksize = 1,
+               .cra_ctxsize = sizeof(struct aegis_ctx),
+               .cra_alignmask = 0,
+
+               .cra_priority = 100,
+
+               .cra_name = "aegis128",
+               .cra_driver_name = "aegis128-generic",
+
+               .cra_module = THIS_MODULE,
+       }
+};
+
+static int __init crypto_aegis128_module_init(void)
+{
+       return crypto_register_aead(&crypto_aegis128_alg);
+}
+
+static void __exit crypto_aegis128_module_exit(void)
+{
+       crypto_unregister_aead(&crypto_aegis128_alg);
+}
+
+subsys_initcall(crypto_aegis128_module_init);
+module_exit(crypto_aegis128_module_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Ondrej Mosnacek <omosnacek@gmail.com>");
+MODULE_DESCRIPTION("AEGIS-128 AEAD algorithm");
+MODULE_ALIAS_CRYPTO("aegis128");
+MODULE_ALIAS_CRYPTO("aegis128-generic");