arch/x86/crypto/chacha20_glue.c

   1 /*
   2  * ChaCha20 256-bit cipher algorithm, RFC7539, SIMD glue code
   3  *
   4  * Copyright (C) 2015 Martin Willi
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  */
  11
  12 #include <crypto/algapi.h>
  13 #include <crypto/chacha20.h>
  14 #include <crypto/internal/skcipher.h>
  15 #include <linux/kernel.h>
  16 #include <linux/module.h>
  17 #include <asm/fpu/api.h>
  18 #include <asm/simd.h>
  19
  20 #define CHACHA20_STATE_ALIGN 16
  21
  22 asmlinkage void chacha20_block_xor_ssse3(u32 *state, u8 *dst, const u8 *src,
  23                                          unsigned int len);
  24 asmlinkage void chacha20_4block_xor_ssse3(u32 *state, u8 *dst, const u8 *src,
  25                                           unsigned int len);
  26 #ifdef CONFIG_AS_AVX2
  27 asmlinkage void chacha20_2block_xor_avx2(u32 *state, u8 *dst, const u8 *src,
  28                                          unsigned int len);
  29 asmlinkage void chacha20_8block_xor_avx2(u32 *state, u8 *dst, const u8 *src,
  30                                          unsigned int len);
  31 static bool chacha20_use_avx2;
  32 #endif
  33
  34 static unsigned int chacha20_advance(unsigned int len, unsigned int maxblocks)
  35 {
  36         len = min(len, maxblocks * CHACHA20_BLOCK_SIZE);
  37         return round_up(len, CHACHA20_BLOCK_SIZE) / CHACHA20_BLOCK_SIZE;
  38 }
  39
  40 static void chacha20_dosimd(u32 *state, u8 *dst, const u8 *src,
  41                             unsigned int bytes)
  42 {
  43 #ifdef CONFIG_AS_AVX2
  44         if (chacha20_use_avx2) {
  45                 while (bytes >= CHACHA20_BLOCK_SIZE * 8) {
  46                         chacha20_8block_xor_avx2(state, dst, src, bytes);
  47                         bytes -= CHACHA20_BLOCK_SIZE * 8;
  48                         src += CHACHA20_BLOCK_SIZE * 8;
  49                         dst += CHACHA20_BLOCK_SIZE * 8;
  50                         state[12] += 8;
  51                 }
  52                 if (bytes > CHACHA20_BLOCK_SIZE * 4) {
  53                         chacha20_8block_xor_avx2(state, dst, src, bytes);
  54                         state[12] += chacha20_advance(bytes, 8);
  55                         return;
  56                 }
  57                 if (bytes > CHACHA20_BLOCK_SIZE) {
  58                         chacha20_2block_xor_avx2(state, dst, src, bytes);
  59                         state[12] += chacha20_advance(bytes, 2);
  60                         return;
  61                 }
  62         }
  63 #endif
  64         while (bytes >= CHACHA20_BLOCK_SIZE * 4) {
  65                 chacha20_4block_xor_ssse3(state, dst, src, bytes);
  66                 bytes -= CHACHA20_BLOCK_SIZE * 4;
  67                 src += CHACHA20_BLOCK_SIZE * 4;
  68                 dst += CHACHA20_BLOCK_SIZE * 4;
  69                 state[12] += 4;
  70         }
  71         if (bytes > CHACHA20_BLOCK_SIZE) {
  72                 chacha20_4block_xor_ssse3(state, dst, src, bytes);
  73                 state[12] += chacha20_advance(bytes, 4);
  74                 return;
  75         }
  76         if (bytes) {
  77                 chacha20_block_xor_ssse3(state, dst, src, bytes);
  78                 state[12]++;
  79         }
  80 }
  81
  82 static int chacha20_simd(struct skcipher_request *req)
  83 {
  84         struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
  85         struct chacha20_ctx *ctx = crypto_skcipher_ctx(tfm);
  86         u32 *state, state_buf[16 + 2] __aligned(8);
  87         struct skcipher_walk walk;
  88         int err;
  89
  90         BUILD_BUG_ON(CHACHA20_STATE_ALIGN != 16);
  91         state = PTR_ALIGN(state_buf + 0, CHACHA20_STATE_ALIGN);
  92
  93         if (req->cryptlen <= CHACHA20_BLOCK_SIZE || !may_use_simd())
  94                 return crypto_chacha20_crypt(req);
  95
  96         err = skcipher_walk_virt(&walk, req, true);
  97
  98         crypto_chacha20_init(state, ctx, walk.iv);
  99
 100         kernel_fpu_begin();
 101
 102         while (walk.nbytes > 0) {
 103                 unsigned int nbytes = walk.nbytes;
 104
 105                 if (nbytes < walk.total)
 106                         nbytes = round_down(nbytes, walk.stride);
 107
 108                 chacha20_dosimd(state, walk.dst.virt.addr, walk.src.virt.addr,
 109                                 nbytes);
 110
 111                 err = skcipher_walk_done(&walk, walk.nbytes - nbytes);
 112         }
 113
 114         kernel_fpu_end();
 115
 116         return err;
 117 }
 118
 119 static struct skcipher_alg alg = {
 120         .base.cra_name          = "chacha20",
 121         .base.cra_driver_name   = "chacha20-simd",
 122         .base.cra_priority      = 300,
 123         .base.cra_blocksize     = 1,
 124         .base.cra_ctxsize       = sizeof(struct chacha20_ctx),
 125         .base.cra_module        = THIS_MODULE,
 126
 127         .min_keysize            = CHACHA20_KEY_SIZE,
 128         .max_keysize            = CHACHA20_KEY_SIZE,
 129         .ivsize                 = CHACHA20_IV_SIZE,
 130         .chunksize              = CHACHA20_BLOCK_SIZE,
 131         .setkey                 = crypto_chacha20_setkey,
 132         .encrypt                = chacha20_simd,
 133         .decrypt                = chacha20_simd,
 134 };
 135
 136 static int __init chacha20_simd_mod_init(void)
 137 {
 138         if (!boot_cpu_has(X86_FEATURE_SSSE3))
 139                 return -ENODEV;
 140
 141 #ifdef CONFIG_AS_AVX2
 142         chacha20_use_avx2 = boot_cpu_has(X86_FEATURE_AVX) &&
 143                             boot_cpu_has(X86_FEATURE_AVX2) &&
 144                             cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM, NULL);
 145 #endif
 146         return crypto_register_skcipher(&alg);
 147 }
 148
 149 static void __exit chacha20_simd_mod_fini(void)
 150 {
 151         crypto_unregister_skcipher(&alg);
 152 }
 153
 154 module_init(chacha20_simd_mod_init);
 155 module_exit(chacha20_simd_mod_fini);
 156
 157 MODULE_LICENSE("GPL");
 158 MODULE_AUTHOR("Martin Willi <martin@strongswan.org>");
 159 MODULE_DESCRIPTION("chacha20 cipher algorithm, SIMD accelerated");
 160 MODULE_ALIAS_CRYPTO("chacha20");
 161 MODULE_ALIAS_CRYPTO("chacha20-simd");