1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * x64 SIMD accelerated ChaCha and XChaCha stream ciphers, 4 * including ChaCha20 (RFC7539) 5 * 6 * Copyright (C) 2015 Martin Willi 7 */ 8 9 #include <crypto/algapi.h> 10 #include <crypto/internal/chacha.h> 11 #include <crypto/internal/simd.h> 12 #include <crypto/internal/skcipher.h> 13 #include <linux/kernel.h> 14 #include <linux/module.h> 15 #include <asm/simd.h> 16 17 #define CHACHA_STATE_ALIGN 16 18 19 asmlinkage void chacha_block_xor_ssse3(u32 *state, u8 *dst, const u8 *src, 20 unsigned int len, int nrounds); 21 asmlinkage void chacha_4block_xor_ssse3(u32 *state, u8 *dst, const u8 *src, 22 unsigned int len, int nrounds); 23 asmlinkage void hchacha_block_ssse3(const u32 *state, u32 *out, int nrounds); 24 25 asmlinkage void chacha_2block_xor_avx2(u32 *state, u8 *dst, const u8 *src, 26 unsigned int len, int nrounds); 27 asmlinkage void chacha_4block_xor_avx2(u32 *state, u8 *dst, const u8 *src, 28 unsigned int len, int nrounds); 29 asmlinkage void chacha_8block_xor_avx2(u32 *state, u8 *dst, const u8 *src, 30 unsigned int len, int nrounds); 31 32 asmlinkage void chacha_2block_xor_avx512vl(u32 *state, u8 *dst, const u8 *src, 33 unsigned int len, int nrounds); 34 asmlinkage void chacha_4block_xor_avx512vl(u32 *state, u8 *dst, const u8 *src, 35 unsigned int len, int nrounds); 36 asmlinkage void chacha_8block_xor_avx512vl(u32 *state, u8 *dst, const u8 *src, 37 unsigned int len, int nrounds); 38 39 static __ro_after_init DEFINE_STATIC_KEY_FALSE(chacha_use_simd); 40 static __ro_after_init DEFINE_STATIC_KEY_FALSE(chacha_use_avx2); 41 static __ro_after_init DEFINE_STATIC_KEY_FALSE(chacha_use_avx512vl); 42 43 static unsigned int chacha_advance(unsigned int len, unsigned int maxblocks) 44 { 45 len = min(len, maxblocks * CHACHA_BLOCK_SIZE); 46 return round_up(len, CHACHA_BLOCK_SIZE) / CHACHA_BLOCK_SIZE; 47 } 48 49 static void chacha_dosimd(u32 *state, u8 *dst, const u8 *src, 50 unsigned int bytes, int nrounds) 51 { 52 if (IS_ENABLED(CONFIG_AS_AVX512) && 53 static_branch_likely(&chacha_use_avx512vl)) { 54 while (bytes >= CHACHA_BLOCK_SIZE * 8) { 55 chacha_8block_xor_avx512vl(state, dst, src, bytes, 56 nrounds); 57 bytes -= CHACHA_BLOCK_SIZE * 8; 58 src += CHACHA_BLOCK_SIZE * 8; 59 dst += CHACHA_BLOCK_SIZE * 8; 60 state[12] += 8; 61 } 62 if (bytes > CHACHA_BLOCK_SIZE * 4) { 63 chacha_8block_xor_avx512vl(state, dst, src, bytes, 64 nrounds); 65 state[12] += chacha_advance(bytes, 8); 66 return; 67 } 68 if (bytes > CHACHA_BLOCK_SIZE * 2) { 69 chacha_4block_xor_avx512vl(state, dst, src, bytes, 70 nrounds); 71 state[12] += chacha_advance(bytes, 4); 72 return; 73 } 74 if (bytes) { 75 chacha_2block_xor_avx512vl(state, dst, src, bytes, 76 nrounds); 77 state[12] += chacha_advance(bytes, 2); 78 return; 79 } 80 } 81 82 if (static_branch_likely(&chacha_use_avx2)) { 83 while (bytes >= CHACHA_BLOCK_SIZE * 8) { 84 chacha_8block_xor_avx2(state, dst, src, bytes, nrounds); 85 bytes -= CHACHA_BLOCK_SIZE * 8; 86 src += CHACHA_BLOCK_SIZE * 8; 87 dst += CHACHA_BLOCK_SIZE * 8; 88 state[12] += 8; 89 } 90 if (bytes > CHACHA_BLOCK_SIZE * 4) { 91 chacha_8block_xor_avx2(state, dst, src, bytes, nrounds); 92 state[12] += chacha_advance(bytes, 8); 93 return; 94 } 95 if (bytes > CHACHA_BLOCK_SIZE * 2) { 96 chacha_4block_xor_avx2(state, dst, src, bytes, nrounds); 97 state[12] += chacha_advance(bytes, 4); 98 return; 99 } 100 if (bytes > CHACHA_BLOCK_SIZE) { 101 chacha_2block_xor_avx2(state, dst, src, bytes, nrounds); 102 state[12] += chacha_advance(bytes, 2); 103 return; 104 } 105 } 106 107 while (bytes >= CHACHA_BLOCK_SIZE * 4) { 108 chacha_4block_xor_ssse3(state, dst, src, bytes, nrounds); 109 bytes -= CHACHA_BLOCK_SIZE * 4; 110 src += CHACHA_BLOCK_SIZE * 4; 111 dst += CHACHA_BLOCK_SIZE * 4; 112 state[12] += 4; 113 } 114 if (bytes > CHACHA_BLOCK_SIZE) { 115 chacha_4block_xor_ssse3(state, dst, src, bytes, nrounds); 116 state[12] += chacha_advance(bytes, 4); 117 return; 118 } 119 if (bytes) { 120 chacha_block_xor_ssse3(state, dst, src, bytes, nrounds); 121 state[12]++; 122 } 123 } 124 125 void hchacha_block_arch(const u32 *state, u32 *stream, int nrounds) 126 { 127 state = PTR_ALIGN(state, CHACHA_STATE_ALIGN); 128 129 if (!static_branch_likely(&chacha_use_simd) || !crypto_simd_usable()) { 130 hchacha_block_generic(state, stream, nrounds); 131 } else { 132 kernel_fpu_begin(); 133 hchacha_block_ssse3(state, stream, nrounds); 134 kernel_fpu_end(); 135 } 136 } 137 EXPORT_SYMBOL(hchacha_block_arch); 138 139 void chacha_init_arch(u32 *state, const u32 *key, const u8 *iv) 140 { 141 state = PTR_ALIGN(state, CHACHA_STATE_ALIGN); 142 143 chacha_init_generic(state, key, iv); 144 } 145 EXPORT_SYMBOL(chacha_init_arch); 146 147 void chacha_crypt_arch(u32 *state, u8 *dst, const u8 *src, unsigned int bytes, 148 int nrounds) 149 { 150 state = PTR_ALIGN(state, CHACHA_STATE_ALIGN); 151 152 if (!static_branch_likely(&chacha_use_simd) || !crypto_simd_usable() || 153 bytes <= CHACHA_BLOCK_SIZE) 154 return chacha_crypt_generic(state, dst, src, bytes, nrounds); 155 156 do { 157 unsigned int todo = min_t(unsigned int, bytes, SZ_4K); 158 159 kernel_fpu_begin(); 160 chacha_dosimd(state, dst, src, todo, nrounds); 161 kernel_fpu_end(); 162 163 bytes -= todo; 164 src += todo; 165 dst += todo; 166 } while (bytes); 167 } 168 EXPORT_SYMBOL(chacha_crypt_arch); 169 170 static int chacha_simd_stream_xor(struct skcipher_request *req, 171 const struct chacha_ctx *ctx, const u8 *iv) 172 { 173 u32 *state, state_buf[16 + 2] __aligned(8); 174 struct skcipher_walk walk; 175 int err; 176 177 err = skcipher_walk_virt(&walk, req, false); 178 179 BUILD_BUG_ON(CHACHA_STATE_ALIGN != 16); 180 state = PTR_ALIGN(state_buf + 0, CHACHA_STATE_ALIGN); 181 182 chacha_init_generic(state, ctx->key, iv); 183 184 while (walk.nbytes > 0) { 185 unsigned int nbytes = walk.nbytes; 186 187 if (nbytes < walk.total) 188 nbytes = round_down(nbytes, walk.stride); 189 190 if (!static_branch_likely(&chacha_use_simd) || 191 !crypto_simd_usable()) { 192 chacha_crypt_generic(state, walk.dst.virt.addr, 193 walk.src.virt.addr, nbytes, 194 ctx->nrounds); 195 } else { 196 kernel_fpu_begin(); 197 chacha_dosimd(state, walk.dst.virt.addr, 198 walk.src.virt.addr, nbytes, 199 ctx->nrounds); 200 kernel_fpu_end(); 201 } 202 err = skcipher_walk_done(&walk, walk.nbytes - nbytes); 203 } 204 205 return err; 206 } 207 208 static int chacha_simd(struct skcipher_request *req) 209 { 210 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req); 211 struct chacha_ctx *ctx = crypto_skcipher_ctx(tfm); 212 213 return chacha_simd_stream_xor(req, ctx, req->iv); 214 } 215 216 static int xchacha_simd(struct skcipher_request *req) 217 { 218 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req); 219 struct chacha_ctx *ctx = crypto_skcipher_ctx(tfm); 220 u32 *state, state_buf[16 + 2] __aligned(8); 221 struct chacha_ctx subctx; 222 u8 real_iv[16]; 223 224 BUILD_BUG_ON(CHACHA_STATE_ALIGN != 16); 225 state = PTR_ALIGN(state_buf + 0, CHACHA_STATE_ALIGN); 226 chacha_init_generic(state, ctx->key, req->iv); 227 228 if (req->cryptlen > CHACHA_BLOCK_SIZE && crypto_simd_usable()) { 229 kernel_fpu_begin(); 230 hchacha_block_ssse3(state, subctx.key, ctx->nrounds); 231 kernel_fpu_end(); 232 } else { 233 hchacha_block_generic(state, subctx.key, ctx->nrounds); 234 } 235 subctx.nrounds = ctx->nrounds; 236 237 memcpy(&real_iv[0], req->iv + 24, 8); 238 memcpy(&real_iv[8], req->iv + 16, 8); 239 return chacha_simd_stream_xor(req, &subctx, real_iv); 240 } 241 242 static struct skcipher_alg algs[] = { 243 { 244 .base.cra_name = "chacha20", 245 .base.cra_driver_name = "chacha20-simd", 246 .base.cra_priority = 300, 247 .base.cra_blocksize = 1, 248 .base.cra_ctxsize = sizeof(struct chacha_ctx), 249 .base.cra_module = THIS_MODULE, 250 251 .min_keysize = CHACHA_KEY_SIZE, 252 .max_keysize = CHACHA_KEY_SIZE, 253 .ivsize = CHACHA_IV_SIZE, 254 .chunksize = CHACHA_BLOCK_SIZE, 255 .setkey = chacha20_setkey, 256 .encrypt = chacha_simd, 257 .decrypt = chacha_simd, 258 }, { 259 .base.cra_name = "xchacha20", 260 .base.cra_driver_name = "xchacha20-simd", 261 .base.cra_priority = 300, 262 .base.cra_blocksize = 1, 263 .base.cra_ctxsize = sizeof(struct chacha_ctx), 264 .base.cra_module = THIS_MODULE, 265 266 .min_keysize = CHACHA_KEY_SIZE, 267 .max_keysize = CHACHA_KEY_SIZE, 268 .ivsize = XCHACHA_IV_SIZE, 269 .chunksize = CHACHA_BLOCK_SIZE, 270 .setkey = chacha20_setkey, 271 .encrypt = xchacha_simd, 272 .decrypt = xchacha_simd, 273 }, { 274 .base.cra_name = "xchacha12", 275 .base.cra_driver_name = "xchacha12-simd", 276 .base.cra_priority = 300, 277 .base.cra_blocksize = 1, 278 .base.cra_ctxsize = sizeof(struct chacha_ctx), 279 .base.cra_module = THIS_MODULE, 280 281 .min_keysize = CHACHA_KEY_SIZE, 282 .max_keysize = CHACHA_KEY_SIZE, 283 .ivsize = XCHACHA_IV_SIZE, 284 .chunksize = CHACHA_BLOCK_SIZE, 285 .setkey = chacha12_setkey, 286 .encrypt = xchacha_simd, 287 .decrypt = xchacha_simd, 288 }, 289 }; 290 291 static int __init chacha_simd_mod_init(void) 292 { 293 if (!boot_cpu_has(X86_FEATURE_SSSE3)) 294 return 0; 295 296 static_branch_enable(&chacha_use_simd); 297 298 if (boot_cpu_has(X86_FEATURE_AVX) && 299 boot_cpu_has(X86_FEATURE_AVX2) && 300 cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM, NULL)) { 301 static_branch_enable(&chacha_use_avx2); 302 303 if (IS_ENABLED(CONFIG_AS_AVX512) && 304 boot_cpu_has(X86_FEATURE_AVX512VL) && 305 boot_cpu_has(X86_FEATURE_AVX512BW)) /* kmovq */ 306 static_branch_enable(&chacha_use_avx512vl); 307 } 308 return IS_REACHABLE(CONFIG_CRYPTO_SKCIPHER) ? 309 crypto_register_skciphers(algs, ARRAY_SIZE(algs)) : 0; 310 } 311 312 static void __exit chacha_simd_mod_fini(void) 313 { 314 if (IS_REACHABLE(CONFIG_CRYPTO_SKCIPHER) && boot_cpu_has(X86_FEATURE_SSSE3)) 315 crypto_unregister_skciphers(algs, ARRAY_SIZE(algs)); 316 } 317 318 module_init(chacha_simd_mod_init); 319 module_exit(chacha_simd_mod_fini); 320 321 MODULE_LICENSE("GPL"); 322 MODULE_AUTHOR("Martin Willi <martin@strongswan.org>"); 323 MODULE_DESCRIPTION("ChaCha and XChaCha stream ciphers (x64 SIMD accelerated)"); 324 MODULE_ALIAS_CRYPTO("chacha20"); 325 MODULE_ALIAS_CRYPTO("chacha20-simd"); 326 MODULE_ALIAS_CRYPTO("xchacha20"); 327 MODULE_ALIAS_CRYPTO("xchacha20-simd"); 328 MODULE_ALIAS_CRYPTO("xchacha12"); 329 MODULE_ALIAS_CRYPTO("xchacha12-simd"); 330