Commit | Line | Data |
---|---|---|
0e1227d3 HY |
1 | /* |
2 | * Accelerated GHASH implementation with Intel PCLMULQDQ-NI | |
3 | * instructions. This file contains glue code. | |
4 | * | |
5 | * Copyright (c) 2009 Intel Corp. | |
6 | * Author: Huang Ying <ying.huang@intel.com> | |
7 | * | |
8 | * This program is free software; you can redistribute it and/or modify it | |
9 | * under the terms of the GNU General Public License version 2 as published | |
10 | * by the Free Software Foundation. | |
11 | */ | |
12 | ||
52f6c5ad | 13 | #include <linux/err.h> |
0e1227d3 HY |
14 | #include <linux/module.h> |
15 | #include <linux/init.h> | |
16 | #include <linux/kernel.h> | |
17 | #include <linux/crypto.h> | |
18 | #include <crypto/algapi.h> | |
19 | #include <crypto/cryptd.h> | |
20 | #include <crypto/gf128mul.h> | |
21 | #include <crypto/internal/hash.h> | |
df6b35f4 | 22 | #include <asm/fpu/api.h> |
3bd391f0 | 23 | #include <asm/cpu_device_id.h> |
0e1227d3 HY |
24 | |
25 | #define GHASH_BLOCK_SIZE 16 | |
26 | #define GHASH_DIGEST_SIZE 16 | |
27 | ||
0ea48146 | 28 | void clmul_ghash_mul(char *dst, const u128 *shash); |
0e1227d3 HY |
29 | |
30 | void clmul_ghash_update(char *dst, const char *src, unsigned int srclen, | |
0ea48146 | 31 | const u128 *shash); |
0e1227d3 | 32 | |
0e1227d3 HY |
33 | struct ghash_async_ctx { |
34 | struct cryptd_ahash *cryptd_tfm; | |
35 | }; | |
36 | ||
37 | struct ghash_ctx { | |
0ea48146 | 38 | u128 shash; |
0e1227d3 HY |
39 | }; |
40 | ||
41 | struct ghash_desc_ctx { | |
42 | u8 buffer[GHASH_BLOCK_SIZE]; | |
43 | u32 bytes; | |
44 | }; | |
45 | ||
46 | static int ghash_init(struct shash_desc *desc) | |
47 | { | |
48 | struct ghash_desc_ctx *dctx = shash_desc_ctx(desc); | |
49 | ||
50 | memset(dctx, 0, sizeof(*dctx)); | |
51 | ||
52 | return 0; | |
53 | } | |
54 | ||
55 | static int ghash_setkey(struct crypto_shash *tfm, | |
56 | const u8 *key, unsigned int keylen) | |
57 | { | |
58 | struct ghash_ctx *ctx = crypto_shash_ctx(tfm); | |
8ceee728 AB |
59 | be128 *x = (be128 *)key; |
60 | u64 a, b; | |
0e1227d3 HY |
61 | |
62 | if (keylen != GHASH_BLOCK_SIZE) { | |
63 | crypto_shash_set_flags(tfm, CRYPTO_TFM_RES_BAD_KEY_LEN); | |
64 | return -EINVAL; | |
65 | } | |
66 | ||
8ceee728 AB |
67 | /* perform multiplication by 'x' in GF(2^128) */ |
68 | a = be64_to_cpu(x->a); | |
69 | b = be64_to_cpu(x->b); | |
70 | ||
0ea48146 HX |
71 | ctx->shash.a = (b << 1) | (a >> 63); |
72 | ctx->shash.b = (a << 1) | (b >> 63); | |
8ceee728 AB |
73 | |
74 | if (a >> 63) | |
0ea48146 | 75 | ctx->shash.b ^= ((u64)0xc2) << 56; |
0e1227d3 HY |
76 | |
77 | return 0; | |
78 | } | |
79 | ||
80 | static int ghash_update(struct shash_desc *desc, | |
81 | const u8 *src, unsigned int srclen) | |
82 | { | |
83 | struct ghash_desc_ctx *dctx = shash_desc_ctx(desc); | |
84 | struct ghash_ctx *ctx = crypto_shash_ctx(desc->tfm); | |
85 | u8 *dst = dctx->buffer; | |
86 | ||
87 | kernel_fpu_begin(); | |
88 | if (dctx->bytes) { | |
89 | int n = min(srclen, dctx->bytes); | |
90 | u8 *pos = dst + (GHASH_BLOCK_SIZE - dctx->bytes); | |
91 | ||
92 | dctx->bytes -= n; | |
93 | srclen -= n; | |
94 | ||
95 | while (n--) | |
96 | *pos++ ^= *src++; | |
97 | ||
98 | if (!dctx->bytes) | |
99 | clmul_ghash_mul(dst, &ctx->shash); | |
100 | } | |
101 | ||
102 | clmul_ghash_update(dst, src, srclen, &ctx->shash); | |
103 | kernel_fpu_end(); | |
104 | ||
105 | if (srclen & 0xf) { | |
106 | src += srclen - (srclen & 0xf); | |
107 | srclen &= 0xf; | |
108 | dctx->bytes = GHASH_BLOCK_SIZE - srclen; | |
109 | while (srclen--) | |
110 | *dst++ ^= *src++; | |
111 | } | |
112 | ||
113 | return 0; | |
114 | } | |
115 | ||
116 | static void ghash_flush(struct ghash_ctx *ctx, struct ghash_desc_ctx *dctx) | |
117 | { | |
118 | u8 *dst = dctx->buffer; | |
119 | ||
120 | if (dctx->bytes) { | |
121 | u8 *tmp = dst + (GHASH_BLOCK_SIZE - dctx->bytes); | |
122 | ||
123 | while (dctx->bytes--) | |
124 | *tmp++ ^= 0; | |
125 | ||
126 | kernel_fpu_begin(); | |
127 | clmul_ghash_mul(dst, &ctx->shash); | |
128 | kernel_fpu_end(); | |
129 | } | |
130 | ||
131 | dctx->bytes = 0; | |
132 | } | |
133 | ||
134 | static int ghash_final(struct shash_desc *desc, u8 *dst) | |
135 | { | |
136 | struct ghash_desc_ctx *dctx = shash_desc_ctx(desc); | |
137 | struct ghash_ctx *ctx = crypto_shash_ctx(desc->tfm); | |
138 | u8 *buf = dctx->buffer; | |
139 | ||
140 | ghash_flush(ctx, dctx); | |
141 | memcpy(dst, buf, GHASH_BLOCK_SIZE); | |
142 | ||
143 | return 0; | |
144 | } | |
145 | ||
146 | static struct shash_alg ghash_alg = { | |
147 | .digestsize = GHASH_DIGEST_SIZE, | |
148 | .init = ghash_init, | |
149 | .update = ghash_update, | |
150 | .final = ghash_final, | |
151 | .setkey = ghash_setkey, | |
152 | .descsize = sizeof(struct ghash_desc_ctx), | |
153 | .base = { | |
154 | .cra_name = "__ghash", | |
155 | .cra_driver_name = "__ghash-pclmulqdqni", | |
156 | .cra_priority = 0, | |
6a9b52b7 SM |
157 | .cra_flags = CRYPTO_ALG_TYPE_SHASH | |
158 | CRYPTO_ALG_INTERNAL, | |
0e1227d3 HY |
159 | .cra_blocksize = GHASH_BLOCK_SIZE, |
160 | .cra_ctxsize = sizeof(struct ghash_ctx), | |
161 | .cra_module = THIS_MODULE, | |
0e1227d3 HY |
162 | }, |
163 | }; | |
164 | ||
165 | static int ghash_async_init(struct ahash_request *req) | |
166 | { | |
167 | struct crypto_ahash *tfm = crypto_ahash_reqtfm(req); | |
168 | struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm); | |
169 | struct ahash_request *cryptd_req = ahash_request_ctx(req); | |
170 | struct cryptd_ahash *cryptd_tfm = ctx->cryptd_tfm; | |
171 | ||
01dd9582 | 172 | if (!irq_fpu_usable()) { |
0e1227d3 HY |
173 | memcpy(cryptd_req, req, sizeof(*req)); |
174 | ahash_request_set_tfm(cryptd_req, &cryptd_tfm->base); | |
175 | return crypto_ahash_init(cryptd_req); | |
176 | } else { | |
177 | struct shash_desc *desc = cryptd_shash_desc(cryptd_req); | |
178 | struct crypto_shash *child = cryptd_ahash_child(cryptd_tfm); | |
179 | ||
180 | desc->tfm = child; | |
181 | desc->flags = req->base.flags; | |
182 | return crypto_shash_init(desc); | |
183 | } | |
184 | } | |
185 | ||
186 | static int ghash_async_update(struct ahash_request *req) | |
187 | { | |
188 | struct ahash_request *cryptd_req = ahash_request_ctx(req); | |
189 | ||
01dd9582 | 190 | if (!irq_fpu_usable()) { |
0e1227d3 HY |
191 | struct crypto_ahash *tfm = crypto_ahash_reqtfm(req); |
192 | struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm); | |
193 | struct cryptd_ahash *cryptd_tfm = ctx->cryptd_tfm; | |
194 | ||
195 | memcpy(cryptd_req, req, sizeof(*req)); | |
196 | ahash_request_set_tfm(cryptd_req, &cryptd_tfm->base); | |
197 | return crypto_ahash_update(cryptd_req); | |
198 | } else { | |
199 | struct shash_desc *desc = cryptd_shash_desc(cryptd_req); | |
200 | return shash_ahash_update(req, desc); | |
201 | } | |
202 | } | |
203 | ||
204 | static int ghash_async_final(struct ahash_request *req) | |
205 | { | |
206 | struct ahash_request *cryptd_req = ahash_request_ctx(req); | |
207 | ||
01dd9582 | 208 | if (!irq_fpu_usable()) { |
0e1227d3 HY |
209 | struct crypto_ahash *tfm = crypto_ahash_reqtfm(req); |
210 | struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm); | |
211 | struct cryptd_ahash *cryptd_tfm = ctx->cryptd_tfm; | |
212 | ||
213 | memcpy(cryptd_req, req, sizeof(*req)); | |
214 | ahash_request_set_tfm(cryptd_req, &cryptd_tfm->base); | |
215 | return crypto_ahash_final(cryptd_req); | |
216 | } else { | |
217 | struct shash_desc *desc = cryptd_shash_desc(cryptd_req); | |
218 | return crypto_shash_final(desc, req->result); | |
219 | } | |
220 | } | |
221 | ||
222 | static int ghash_async_digest(struct ahash_request *req) | |
223 | { | |
224 | struct crypto_ahash *tfm = crypto_ahash_reqtfm(req); | |
225 | struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm); | |
226 | struct ahash_request *cryptd_req = ahash_request_ctx(req); | |
227 | struct cryptd_ahash *cryptd_tfm = ctx->cryptd_tfm; | |
228 | ||
01dd9582 | 229 | if (!irq_fpu_usable()) { |
0e1227d3 HY |
230 | memcpy(cryptd_req, req, sizeof(*req)); |
231 | ahash_request_set_tfm(cryptd_req, &cryptd_tfm->base); | |
232 | return crypto_ahash_digest(cryptd_req); | |
233 | } else { | |
234 | struct shash_desc *desc = cryptd_shash_desc(cryptd_req); | |
235 | struct crypto_shash *child = cryptd_ahash_child(cryptd_tfm); | |
236 | ||
237 | desc->tfm = child; | |
238 | desc->flags = req->base.flags; | |
239 | return shash_ahash_digest(req, desc); | |
240 | } | |
241 | } | |
242 | ||
243 | static int ghash_async_setkey(struct crypto_ahash *tfm, const u8 *key, | |
244 | unsigned int keylen) | |
245 | { | |
246 | struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm); | |
247 | struct crypto_ahash *child = &ctx->cryptd_tfm->base; | |
248 | int err; | |
249 | ||
250 | crypto_ahash_clear_flags(child, CRYPTO_TFM_REQ_MASK); | |
251 | crypto_ahash_set_flags(child, crypto_ahash_get_flags(tfm) | |
252 | & CRYPTO_TFM_REQ_MASK); | |
253 | err = crypto_ahash_setkey(child, key, keylen); | |
254 | crypto_ahash_set_flags(tfm, crypto_ahash_get_flags(child) | |
255 | & CRYPTO_TFM_RES_MASK); | |
256 | ||
c3e73e76 | 257 | return err; |
0e1227d3 HY |
258 | } |
259 | ||
260 | static int ghash_async_init_tfm(struct crypto_tfm *tfm) | |
261 | { | |
262 | struct cryptd_ahash *cryptd_tfm; | |
263 | struct ghash_async_ctx *ctx = crypto_tfm_ctx(tfm); | |
264 | ||
6a9b52b7 SM |
265 | cryptd_tfm = cryptd_alloc_ahash("__ghash-pclmulqdqni", |
266 | CRYPTO_ALG_INTERNAL, | |
267 | CRYPTO_ALG_INTERNAL); | |
0e1227d3 HY |
268 | if (IS_ERR(cryptd_tfm)) |
269 | return PTR_ERR(cryptd_tfm); | |
270 | ctx->cryptd_tfm = cryptd_tfm; | |
271 | crypto_ahash_set_reqsize(__crypto_ahash_cast(tfm), | |
272 | sizeof(struct ahash_request) + | |
273 | crypto_ahash_reqsize(&cryptd_tfm->base)); | |
274 | ||
275 | return 0; | |
276 | } | |
277 | ||
278 | static void ghash_async_exit_tfm(struct crypto_tfm *tfm) | |
279 | { | |
280 | struct ghash_async_ctx *ctx = crypto_tfm_ctx(tfm); | |
281 | ||
282 | cryptd_free_ahash(ctx->cryptd_tfm); | |
283 | } | |
284 | ||
285 | static struct ahash_alg ghash_async_alg = { | |
286 | .init = ghash_async_init, | |
287 | .update = ghash_async_update, | |
288 | .final = ghash_async_final, | |
289 | .setkey = ghash_async_setkey, | |
290 | .digest = ghash_async_digest, | |
291 | .halg = { | |
292 | .digestsize = GHASH_DIGEST_SIZE, | |
293 | .base = { | |
294 | .cra_name = "ghash", | |
295 | .cra_driver_name = "ghash-clmulni", | |
296 | .cra_priority = 400, | |
297 | .cra_flags = CRYPTO_ALG_TYPE_AHASH | CRYPTO_ALG_ASYNC, | |
298 | .cra_blocksize = GHASH_BLOCK_SIZE, | |
299 | .cra_type = &crypto_ahash_type, | |
300 | .cra_module = THIS_MODULE, | |
0e1227d3 HY |
301 | .cra_init = ghash_async_init_tfm, |
302 | .cra_exit = ghash_async_exit_tfm, | |
303 | }, | |
304 | }, | |
305 | }; | |
306 | ||
3bd391f0 AK |
307 | static const struct x86_cpu_id pcmul_cpu_id[] = { |
308 | X86_FEATURE_MATCH(X86_FEATURE_PCLMULQDQ), /* Pickle-Mickle-Duck */ | |
309 | {} | |
310 | }; | |
311 | MODULE_DEVICE_TABLE(x86cpu, pcmul_cpu_id); | |
312 | ||
0e1227d3 HY |
313 | static int __init ghash_pclmulqdqni_mod_init(void) |
314 | { | |
315 | int err; | |
316 | ||
3bd391f0 | 317 | if (!x86_match_cpu(pcmul_cpu_id)) |
0e1227d3 | 318 | return -ENODEV; |
0e1227d3 HY |
319 | |
320 | err = crypto_register_shash(&ghash_alg); | |
321 | if (err) | |
322 | goto err_out; | |
323 | err = crypto_register_ahash(&ghash_async_alg); | |
324 | if (err) | |
325 | goto err_shash; | |
326 | ||
327 | return 0; | |
328 | ||
329 | err_shash: | |
330 | crypto_unregister_shash(&ghash_alg); | |
331 | err_out: | |
332 | return err; | |
333 | } | |
334 | ||
335 | static void __exit ghash_pclmulqdqni_mod_exit(void) | |
336 | { | |
337 | crypto_unregister_ahash(&ghash_async_alg); | |
338 | crypto_unregister_shash(&ghash_alg); | |
339 | } | |
340 | ||
341 | module_init(ghash_pclmulqdqni_mod_init); | |
342 | module_exit(ghash_pclmulqdqni_mod_exit); | |
343 | ||
344 | MODULE_LICENSE("GPL"); | |
345 | MODULE_DESCRIPTION("GHASH Message Digest Algorithm, " | |
346 | "acclerated by PCLMULQDQ-NI"); | |
5d26a105 | 347 | MODULE_ALIAS_CRYPTO("ghash"); |