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 kernel_fpu_begin(); 157 chacha_dosimd(state, dst, src, bytes, nrounds); 158 kernel_fpu_end(); 159 } 160 EXPORT_SYMBOL(chacha_crypt_arch); 161 162 static int chacha_simd_stream_xor(struct skcipher_request *req, 163 const struct chacha_ctx *ctx, const u8 *iv) 164 { 165 u32 *state, state_buf[16 + 2] __aligned(8); 166 struct skcipher_walk walk; 167 int err; 168 169 err = skcipher_walk_virt(&walk, req, false); 170 171 BUILD_BUG_ON(CHACHA_STATE_ALIGN != 16); 172 state = PTR_ALIGN(state_buf + 0, CHACHA_STATE_ALIGN); 173 174 chacha_init_generic(state, ctx->key, iv); 175 176 while (walk.nbytes > 0) { 177 unsigned int nbytes = walk.nbytes; 178 179 if (nbytes < walk.total) 180 nbytes = round_down(nbytes, walk.stride); 181 182 if (!static_branch_likely(&chacha_use_simd) || 183 !crypto_simd_usable()) { 184 chacha_crypt_generic(state, walk.dst.virt.addr, 185 walk.src.virt.addr, nbytes, 186 ctx->nrounds); 187 } else { 188 kernel_fpu_begin(); 189 chacha_dosimd(state, walk.dst.virt.addr, 190 walk.src.virt.addr, nbytes, 191 ctx->nrounds); 192 kernel_fpu_end(); 193 } 194 err = skcipher_walk_done(&walk, walk.nbytes - nbytes); 195 } 196 197 return err; 198 } 199 200 static int chacha_simd(struct skcipher_request *req) 201 { 202 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req); 203 struct chacha_ctx *ctx = crypto_skcipher_ctx(tfm); 204 205 return chacha_simd_stream_xor(req, ctx, req->iv); 206 } 207 208 static int xchacha_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 u32 *state, state_buf[16 + 2] __aligned(8); 213 struct chacha_ctx subctx; 214 u8 real_iv[16]; 215 216 BUILD_BUG_ON(CHACHA_STATE_ALIGN != 16); 217 state = PTR_ALIGN(state_buf + 0, CHACHA_STATE_ALIGN); 218 chacha_init_generic(state, ctx->key, req->iv); 219 220 if (req->cryptlen > CHACHA_BLOCK_SIZE && crypto_simd_usable()) { 221 kernel_fpu_begin(); 222 hchacha_block_ssse3(state, subctx.key, ctx->nrounds); 223 kernel_fpu_end(); 224 } else { 225 hchacha_block_generic(state, subctx.key, ctx->nrounds); 226 } 227 subctx.nrounds = ctx->nrounds; 228 229 memcpy(&real_iv[0], req->iv + 24, 8); 230 memcpy(&real_iv[8], req->iv + 16, 8); 231 return chacha_simd_stream_xor(req, &subctx, real_iv); 232 } 233 234 static struct skcipher_alg algs[] = { 235 { 236 .base.cra_name = "chacha20", 237 .base.cra_driver_name = "chacha20-simd", 238 .base.cra_priority = 300, 239 .base.cra_blocksize = 1, 240 .base.cra_ctxsize = sizeof(struct chacha_ctx), 241 .base.cra_module = THIS_MODULE, 242 243 .min_keysize = CHACHA_KEY_SIZE, 244 .max_keysize = CHACHA_KEY_SIZE, 245 .ivsize = CHACHA_IV_SIZE, 246 .chunksize = CHACHA_BLOCK_SIZE, 247 .setkey = chacha20_setkey, 248 .encrypt = chacha_simd, 249 .decrypt = chacha_simd, 250 }, { 251 .base.cra_name = "xchacha20", 252 .base.cra_driver_name = "xchacha20-simd", 253 .base.cra_priority = 300, 254 .base.cra_blocksize = 1, 255 .base.cra_ctxsize = sizeof(struct chacha_ctx), 256 .base.cra_module = THIS_MODULE, 257 258 .min_keysize = CHACHA_KEY_SIZE, 259 .max_keysize = CHACHA_KEY_SIZE, 260 .ivsize = XCHACHA_IV_SIZE, 261 .chunksize = CHACHA_BLOCK_SIZE, 262 .setkey = chacha20_setkey, 263 .encrypt = xchacha_simd, 264 .decrypt = xchacha_simd, 265 }, { 266 .base.cra_name = "xchacha12", 267 .base.cra_driver_name = "xchacha12-simd", 268 .base.cra_priority = 300, 269 .base.cra_blocksize = 1, 270 .base.cra_ctxsize = sizeof(struct chacha_ctx), 271 .base.cra_module = THIS_MODULE, 272 273 .min_keysize = CHACHA_KEY_SIZE, 274 .max_keysize = CHACHA_KEY_SIZE, 275 .ivsize = XCHACHA_IV_SIZE, 276 .chunksize = CHACHA_BLOCK_SIZE, 277 .setkey = chacha12_setkey, 278 .encrypt = xchacha_simd, 279 .decrypt = xchacha_simd, 280 }, 281 }; 282 283 static int __init chacha_simd_mod_init(void) 284 { 285 if (!boot_cpu_has(X86_FEATURE_SSSE3)) 286 return 0; 287 288 static_branch_enable(&chacha_use_simd); 289 290 if (boot_cpu_has(X86_FEATURE_AVX) && 291 boot_cpu_has(X86_FEATURE_AVX2) && 292 cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM, NULL)) { 293 static_branch_enable(&chacha_use_avx2); 294 295 if (IS_ENABLED(CONFIG_AS_AVX512) && 296 boot_cpu_has(X86_FEATURE_AVX512VL) && 297 boot_cpu_has(X86_FEATURE_AVX512BW)) /* kmovq */ 298 static_branch_enable(&chacha_use_avx512vl); 299 } 300 return IS_REACHABLE(CONFIG_CRYPTO_SKCIPHER) ? 301 crypto_register_skciphers(algs, ARRAY_SIZE(algs)) : 0; 302 } 303 304 static void __exit chacha_simd_mod_fini(void) 305 { 306 if (IS_REACHABLE(CONFIG_CRYPTO_SKCIPHER) && boot_cpu_has(X86_FEATURE_SSSE3)) 307 crypto_unregister_skciphers(algs, ARRAY_SIZE(algs)); 308 } 309 310 module_init(chacha_simd_mod_init); 311 module_exit(chacha_simd_mod_fini); 312 313 MODULE_LICENSE("GPL"); 314 MODULE_AUTHOR("Martin Willi <martin@strongswan.org>"); 315 MODULE_DESCRIPTION("ChaCha and XChaCha stream ciphers (x64 SIMD accelerated)"); 316 MODULE_ALIAS_CRYPTO("chacha20"); 317 MODULE_ALIAS_CRYPTO("chacha20-simd"); 318 MODULE_ALIAS_CRYPTO("xchacha20"); 319 MODULE_ALIAS_CRYPTO("xchacha20-simd"); 320 MODULE_ALIAS_CRYPTO("xchacha12"); 321 MODULE_ALIAS_CRYPTO("xchacha12-simd"); 322