4 * Glue code for the SHA1 Secure Hash Algorithm assembler implementation using
5 * Supplemental SSE3 instructions.
7 * This file is based on sha1_generic.c
9 * Copyright (c) Alan Smithee.
10 * Copyright (c) Andrew McDonald <andrew@mcdonald.org.uk>
11 * Copyright (c) Jean-Francois Dive <jef@linuxbe.org>
12 * Copyright (c) Mathias Krause <minipli@googlemail.com>
14 * This program is free software; you can redistribute it and/or modify it
15 * under the terms of the GNU General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option)
21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
23 #include <crypto/internal/hash.h>
24 #include <linux/init.h>
25 #include <linux/module.h>
27 #include <linux/cryptohash.h>
28 #include <linux/types.h>
29 #include <crypto/sha.h>
30 #include <asm/byteorder.h>
33 #include <asm/xsave.h>
36 asmlinkage void sha1_transform_ssse3(u32 *digest, const char *data,
39 asmlinkage void sha1_transform_avx(u32 *digest, const char *data,
43 static asmlinkage void (*sha1_transform_asm)(u32 *, const char *, unsigned int);
46 static int sha1_ssse3_init(struct shash_desc *desc)
48 struct sha1_state *sctx = shash_desc_ctx(desc);
50 *sctx = (struct sha1_state){
51 .state = { SHA1_H0, SHA1_H1, SHA1_H2, SHA1_H3, SHA1_H4 },
57 static int __sha1_ssse3_update(struct shash_desc *desc, const u8 *data,
58 unsigned int len, unsigned int partial)
60 struct sha1_state *sctx = shash_desc_ctx(desc);
61 unsigned int done = 0;
66 done = SHA1_BLOCK_SIZE - partial;
67 memcpy(sctx->buffer + partial, data, done);
68 sha1_transform_asm(sctx->state, sctx->buffer, 1);
71 if (len - done >= SHA1_BLOCK_SIZE) {
72 const unsigned int rounds = (len - done) / SHA1_BLOCK_SIZE;
74 sha1_transform_asm(sctx->state, data + done, rounds);
75 done += rounds * SHA1_BLOCK_SIZE;
78 memcpy(sctx->buffer, data + done, len - done);
83 static int sha1_ssse3_update(struct shash_desc *desc, const u8 *data,
86 struct sha1_state *sctx = shash_desc_ctx(desc);
87 unsigned int partial = sctx->count % SHA1_BLOCK_SIZE;
90 /* Handle the fast case right here */
91 if (partial + len < SHA1_BLOCK_SIZE) {
93 memcpy(sctx->buffer + partial, data, len);
98 if (!irq_fpu_usable()) {
99 res = crypto_sha1_update(desc, data, len);
102 res = __sha1_ssse3_update(desc, data, len, partial);
110 /* Add padding and return the message digest. */
111 static int sha1_ssse3_final(struct shash_desc *desc, u8 *out)
113 struct sha1_state *sctx = shash_desc_ctx(desc);
114 unsigned int i, index, padlen;
115 __be32 *dst = (__be32 *)out;
117 static const u8 padding[SHA1_BLOCK_SIZE] = { 0x80, };
119 bits = cpu_to_be64(sctx->count << 3);
121 /* Pad out to 56 mod 64 and append length */
122 index = sctx->count % SHA1_BLOCK_SIZE;
123 padlen = (index < 56) ? (56 - index) : ((SHA1_BLOCK_SIZE+56) - index);
124 if (!irq_fpu_usable()) {
125 crypto_sha1_update(desc, padding, padlen);
126 crypto_sha1_update(desc, (const u8 *)&bits, sizeof(bits));
129 /* We need to fill a whole block for __sha1_ssse3_update() */
131 sctx->count += padlen;
132 memcpy(sctx->buffer + index, padding, padlen);
134 __sha1_ssse3_update(desc, padding, padlen, index);
136 __sha1_ssse3_update(desc, (const u8 *)&bits, sizeof(bits), 56);
140 /* Store state in digest */
141 for (i = 0; i < 5; i++)
142 dst[i] = cpu_to_be32(sctx->state[i]);
145 memset(sctx, 0, sizeof(*sctx));
150 static int sha1_ssse3_export(struct shash_desc *desc, void *out)
152 struct sha1_state *sctx = shash_desc_ctx(desc);
154 memcpy(out, sctx, sizeof(*sctx));
159 static int sha1_ssse3_import(struct shash_desc *desc, const void *in)
161 struct sha1_state *sctx = shash_desc_ctx(desc);
163 memcpy(sctx, in, sizeof(*sctx));
168 static struct shash_alg alg = {
169 .digestsize = SHA1_DIGEST_SIZE,
170 .init = sha1_ssse3_init,
171 .update = sha1_ssse3_update,
172 .final = sha1_ssse3_final,
173 .export = sha1_ssse3_export,
174 .import = sha1_ssse3_import,
175 .descsize = sizeof(struct sha1_state),
176 .statesize = sizeof(struct sha1_state),
179 .cra_driver_name= "sha1-ssse3",
181 .cra_flags = CRYPTO_ALG_TYPE_SHASH,
182 .cra_blocksize = SHA1_BLOCK_SIZE,
183 .cra_module = THIS_MODULE,
188 static bool __init avx_usable(void)
192 if (!cpu_has_avx || !cpu_has_osxsave)
195 xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
196 if ((xcr0 & (XSTATE_SSE | XSTATE_YMM)) != (XSTATE_SSE | XSTATE_YMM)) {
197 pr_info("AVX detected but unusable.\n");
206 static int __init sha1_ssse3_mod_init(void)
208 /* test for SSSE3 first */
210 sha1_transform_asm = sha1_transform_ssse3;
213 /* allow AVX to override SSSE3, it's a little faster */
215 sha1_transform_asm = sha1_transform_avx;
218 if (sha1_transform_asm) {
219 pr_info("Using %s optimized SHA-1 implementation\n",
220 sha1_transform_asm == sha1_transform_ssse3 ? "SSSE3"
222 return crypto_register_shash(&alg);
224 pr_info("Neither AVX nor SSSE3 is available/usable.\n");
229 static void __exit sha1_ssse3_mod_fini(void)
231 crypto_unregister_shash(&alg);
234 module_init(sha1_ssse3_mod_init);
235 module_exit(sha1_ssse3_mod_fini);
237 MODULE_LICENSE("GPL");
238 MODULE_DESCRIPTION("SHA1 Secure Hash Algorithm, Supplemental SSE3 accelerated");
240 MODULE_ALIAS("sha1");