1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Poly1305 authenticator algorithm, RFC7539, SIMD glue code 4 * 5 * Copyright (C) 2015 Martin Willi 6 */ 7 8 #include <crypto/algapi.h> 9 #include <crypto/internal/hash.h> 10 #include <crypto/internal/poly1305.h> 11 #include <crypto/internal/simd.h> 12 #include <linux/crypto.h> 13 #include <linux/jump_label.h> 14 #include <linux/kernel.h> 15 #include <linux/module.h> 16 #include <asm/simd.h> 17 18 asmlinkage void poly1305_block_sse2(u32 *h, const u8 *src, 19 const u32 *r, unsigned int blocks); 20 asmlinkage void poly1305_2block_sse2(u32 *h, const u8 *src, const u32 *r, 21 unsigned int blocks, const u32 *u); 22 asmlinkage void poly1305_4block_avx2(u32 *h, const u8 *src, const u32 *r, 23 unsigned int blocks, const u32 *u); 24 25 static __ro_after_init DEFINE_STATIC_KEY_FALSE(poly1305_use_simd); 26 static __ro_after_init DEFINE_STATIC_KEY_FALSE(poly1305_use_avx2); 27 28 static void poly1305_simd_mult(u32 *a, const u32 *b) 29 { 30 u8 m[POLY1305_BLOCK_SIZE]; 31 32 memset(m, 0, sizeof(m)); 33 /* The poly1305 block function adds a hi-bit to the accumulator which 34 * we don't need for key multiplication; compensate for it. */ 35 a[4] -= 1 << 24; 36 poly1305_block_sse2(a, m, b, 1); 37 } 38 39 static unsigned int poly1305_scalar_blocks(struct poly1305_desc_ctx *dctx, 40 const u8 *src, unsigned int srclen) 41 { 42 unsigned int datalen; 43 44 if (unlikely(!dctx->sset)) { 45 datalen = crypto_poly1305_setdesckey(dctx, src, srclen); 46 src += srclen - datalen; 47 srclen = datalen; 48 } 49 if (srclen >= POLY1305_BLOCK_SIZE) { 50 poly1305_core_blocks(&dctx->h, dctx->r, src, 51 srclen / POLY1305_BLOCK_SIZE, 1); 52 srclen %= POLY1305_BLOCK_SIZE; 53 } 54 return srclen; 55 } 56 57 static unsigned int poly1305_simd_blocks(struct poly1305_desc_ctx *dctx, 58 const u8 *src, unsigned int srclen) 59 { 60 unsigned int blocks, datalen; 61 62 if (unlikely(!dctx->sset)) { 63 datalen = crypto_poly1305_setdesckey(dctx, src, srclen); 64 src += srclen - datalen; 65 srclen = datalen; 66 } 67 68 if (IS_ENABLED(CONFIG_AS_AVX2) && 69 static_branch_likely(&poly1305_use_avx2) && 70 srclen >= POLY1305_BLOCK_SIZE * 4) { 71 if (unlikely(dctx->rset < 4)) { 72 if (dctx->rset < 2) { 73 dctx->r[1] = dctx->r[0]; 74 poly1305_simd_mult(dctx->r[1].r, dctx->r[0].r); 75 } 76 dctx->r[2] = dctx->r[1]; 77 poly1305_simd_mult(dctx->r[2].r, dctx->r[0].r); 78 dctx->r[3] = dctx->r[2]; 79 poly1305_simd_mult(dctx->r[3].r, dctx->r[0].r); 80 dctx->rset = 4; 81 } 82 blocks = srclen / (POLY1305_BLOCK_SIZE * 4); 83 poly1305_4block_avx2(dctx->h.h, src, dctx->r[0].r, blocks, 84 dctx->r[1].r); 85 src += POLY1305_BLOCK_SIZE * 4 * blocks; 86 srclen -= POLY1305_BLOCK_SIZE * 4 * blocks; 87 } 88 89 if (likely(srclen >= POLY1305_BLOCK_SIZE * 2)) { 90 if (unlikely(dctx->rset < 2)) { 91 dctx->r[1] = dctx->r[0]; 92 poly1305_simd_mult(dctx->r[1].r, dctx->r[0].r); 93 dctx->rset = 2; 94 } 95 blocks = srclen / (POLY1305_BLOCK_SIZE * 2); 96 poly1305_2block_sse2(dctx->h.h, src, dctx->r[0].r, 97 blocks, dctx->r[1].r); 98 src += POLY1305_BLOCK_SIZE * 2 * blocks; 99 srclen -= POLY1305_BLOCK_SIZE * 2 * blocks; 100 } 101 if (srclen >= POLY1305_BLOCK_SIZE) { 102 poly1305_block_sse2(dctx->h.h, src, dctx->r[0].r, 1); 103 srclen -= POLY1305_BLOCK_SIZE; 104 } 105 return srclen; 106 } 107 108 void poly1305_init_arch(struct poly1305_desc_ctx *desc, const u8 *key) 109 { 110 poly1305_init_generic(desc, key); 111 } 112 EXPORT_SYMBOL(poly1305_init_arch); 113 114 void poly1305_update_arch(struct poly1305_desc_ctx *dctx, const u8 *src, 115 unsigned int srclen) 116 { 117 unsigned int bytes; 118 119 if (unlikely(dctx->buflen)) { 120 bytes = min(srclen, POLY1305_BLOCK_SIZE - dctx->buflen); 121 memcpy(dctx->buf + dctx->buflen, src, bytes); 122 src += bytes; 123 srclen -= bytes; 124 dctx->buflen += bytes; 125 126 if (dctx->buflen == POLY1305_BLOCK_SIZE) { 127 if (static_branch_likely(&poly1305_use_simd) && 128 likely(crypto_simd_usable())) { 129 kernel_fpu_begin(); 130 poly1305_simd_blocks(dctx, dctx->buf, 131 POLY1305_BLOCK_SIZE); 132 kernel_fpu_end(); 133 } else { 134 poly1305_scalar_blocks(dctx, dctx->buf, 135 POLY1305_BLOCK_SIZE); 136 } 137 dctx->buflen = 0; 138 } 139 } 140 141 if (likely(srclen >= POLY1305_BLOCK_SIZE)) { 142 if (static_branch_likely(&poly1305_use_simd) && 143 likely(crypto_simd_usable())) { 144 kernel_fpu_begin(); 145 bytes = poly1305_simd_blocks(dctx, src, srclen); 146 kernel_fpu_end(); 147 } else { 148 bytes = poly1305_scalar_blocks(dctx, src, srclen); 149 } 150 src += srclen - bytes; 151 srclen = bytes; 152 } 153 154 if (unlikely(srclen)) { 155 dctx->buflen = srclen; 156 memcpy(dctx->buf, src, srclen); 157 } 158 } 159 EXPORT_SYMBOL(poly1305_update_arch); 160 161 void poly1305_final_arch(struct poly1305_desc_ctx *desc, u8 *digest) 162 { 163 poly1305_final_generic(desc, digest); 164 } 165 EXPORT_SYMBOL(poly1305_final_arch); 166 167 static int crypto_poly1305_init(struct shash_desc *desc) 168 { 169 struct poly1305_desc_ctx *dctx = shash_desc_ctx(desc); 170 171 poly1305_core_init(&dctx->h); 172 dctx->buflen = 0; 173 dctx->rset = 0; 174 dctx->sset = false; 175 176 return 0; 177 } 178 179 static int crypto_poly1305_final(struct shash_desc *desc, u8 *dst) 180 { 181 struct poly1305_desc_ctx *dctx = shash_desc_ctx(desc); 182 183 if (unlikely(!dctx->sset)) 184 return -ENOKEY; 185 186 poly1305_final_generic(dctx, dst); 187 return 0; 188 } 189 190 static int poly1305_simd_update(struct shash_desc *desc, 191 const u8 *src, unsigned int srclen) 192 { 193 struct poly1305_desc_ctx *dctx = shash_desc_ctx(desc); 194 195 poly1305_update_arch(dctx, src, srclen); 196 return 0; 197 } 198 199 static struct shash_alg alg = { 200 .digestsize = POLY1305_DIGEST_SIZE, 201 .init = crypto_poly1305_init, 202 .update = poly1305_simd_update, 203 .final = crypto_poly1305_final, 204 .descsize = sizeof(struct poly1305_desc_ctx), 205 .base = { 206 .cra_name = "poly1305", 207 .cra_driver_name = "poly1305-simd", 208 .cra_priority = 300, 209 .cra_blocksize = POLY1305_BLOCK_SIZE, 210 .cra_module = THIS_MODULE, 211 }, 212 }; 213 214 static int __init poly1305_simd_mod_init(void) 215 { 216 if (!boot_cpu_has(X86_FEATURE_XMM2)) 217 return 0; 218 219 static_branch_enable(&poly1305_use_simd); 220 221 if (IS_ENABLED(CONFIG_AS_AVX2) && 222 boot_cpu_has(X86_FEATURE_AVX) && 223 boot_cpu_has(X86_FEATURE_AVX2) && 224 cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM, NULL)) 225 static_branch_enable(&poly1305_use_avx2); 226 227 return IS_REACHABLE(CONFIG_CRYPTO_HASH) ? crypto_register_shash(&alg) : 0; 228 } 229 230 static void __exit poly1305_simd_mod_exit(void) 231 { 232 if (IS_REACHABLE(CONFIG_CRYPTO_HASH)) 233 crypto_unregister_shash(&alg); 234 } 235 236 module_init(poly1305_simd_mod_init); 237 module_exit(poly1305_simd_mod_exit); 238 239 MODULE_LICENSE("GPL"); 240 MODULE_AUTHOR("Martin Willi <martin@strongswan.org>"); 241 MODULE_DESCRIPTION("Poly1305 authenticator"); 242 MODULE_ALIAS_CRYPTO("poly1305"); 243 MODULE_ALIAS_CRYPTO("poly1305-simd"); 244