1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Glue code for AES implementation for SPE instructions (PPC)
4 *
5 * Based on generic implementation. The assembler module takes care
6 * about the SPE registers so it can run from interrupt context.
7 *
8 * Copyright (c) 2015 Markus Stockhausen <stockhausen@collogia.de>
9 */
10
11 #include <crypto/aes.h>
12 #include <linux/module.h>
13 #include <linux/init.h>
14 #include <linux/types.h>
15 #include <linux/errno.h>
16 #include <linux/crypto.h>
17 #include <asm/byteorder.h>
18 #include <asm/switch_to.h>
19 #include <crypto/algapi.h>
20 #include <crypto/internal/skcipher.h>
21 #include <crypto/xts.h>
22 #include <crypto/gf128mul.h>
23 #include <crypto/scatterwalk.h>
24
25 /*
26 * MAX_BYTES defines the number of bytes that are allowed to be processed
27 * between preempt_disable() and preempt_enable(). e500 cores can issue two
28 * instructions per clock cycle using one 32/64 bit unit (SU1) and one 32
29 * bit unit (SU2). One of these can be a memory access that is executed via
30 * a single load and store unit (LSU). XTS-AES-256 takes ~780 operations per
31 * 16 byte block or 25 cycles per byte. Thus 768 bytes of input data
32 * will need an estimated maximum of 20,000 cycles. Headroom for cache misses
33 * included. Even with the low end model clocked at 667 MHz this equals to a
34 * critical time window of less than 30us. The value has been chosen to
35 * process a 512 byte disk block in one or a large 1400 bytes IPsec network
36 * packet in two runs.
37 *
38 */
39 #define MAX_BYTES 768
40
41 struct ppc_aes_ctx {
42 u32 key_enc[AES_MAX_KEYLENGTH_U32];
43 u32 key_dec[AES_MAX_KEYLENGTH_U32];
44 u32 rounds;
45 };
46
47 struct ppc_xts_ctx {
48 u32 key_enc[AES_MAX_KEYLENGTH_U32];
49 u32 key_dec[AES_MAX_KEYLENGTH_U32];
50 u32 key_twk[AES_MAX_KEYLENGTH_U32];
51 u32 rounds;
52 };
53
54 extern void ppc_encrypt_aes(u8 *out, const u8 *in, u32 *key_enc, u32 rounds);
55 extern void ppc_decrypt_aes(u8 *out, const u8 *in, u32 *key_dec, u32 rounds);
56 extern void ppc_encrypt_ecb(u8 *out, const u8 *in, u32 *key_enc, u32 rounds,
57 u32 bytes);
58 extern void ppc_decrypt_ecb(u8 *out, const u8 *in, u32 *key_dec, u32 rounds,
59 u32 bytes);
60 extern void ppc_encrypt_cbc(u8 *out, const u8 *in, u32 *key_enc, u32 rounds,
61 u32 bytes, u8 *iv);
62 extern void ppc_decrypt_cbc(u8 *out, const u8 *in, u32 *key_dec, u32 rounds,
63 u32 bytes, u8 *iv);
64 extern void ppc_crypt_ctr (u8 *out, const u8 *in, u32 *key_enc, u32 rounds,
65 u32 bytes, u8 *iv);
66 extern void ppc_encrypt_xts(u8 *out, const u8 *in, u32 *key_enc, u32 rounds,
67 u32 bytes, u8 *iv, u32 *key_twk);
68 extern void ppc_decrypt_xts(u8 *out, const u8 *in, u32 *key_dec, u32 rounds,
69 u32 bytes, u8 *iv, u32 *key_twk);
70
71 extern void ppc_expand_key_128(u32 *key_enc, const u8 *key);
72 extern void ppc_expand_key_192(u32 *key_enc, const u8 *key);
73 extern void ppc_expand_key_256(u32 *key_enc, const u8 *key);
74
75 extern void ppc_generate_decrypt_key(u32 *key_dec,u32 *key_enc,
76 unsigned int key_len);
77
78 static void spe_begin(void)
79 {
80 /* disable preemption and save users SPE registers if required */
81 preempt_disable();
82 enable_kernel_spe();
83 }
84
85 static void spe_end(void)
86 {
87 disable_kernel_spe();
88 /* reenable preemption */
89 preempt_enable();
90 }
91
92 static int ppc_aes_setkey(struct crypto_tfm *tfm, const u8 *in_key,
93 unsigned int key_len)
94 {
95 struct ppc_aes_ctx *ctx = crypto_tfm_ctx(tfm);
96
97 switch (key_len) {
98 case AES_KEYSIZE_128:
99 ctx->rounds = 4;
100 ppc_expand_key_128(ctx->key_enc, in_key);
101 break;
102 case AES_KEYSIZE_192:
103 ctx->rounds = 5;
104 ppc_expand_key_192(ctx->key_enc, in_key);
105 break;
106 case AES_KEYSIZE_256:
107 ctx->rounds = 6;
108 ppc_expand_key_256(ctx->key_enc, in_key);
109 break;
110 default:
111 return -EINVAL;
112 }
113
114 ppc_generate_decrypt_key(ctx->key_dec, ctx->key_enc, key_len);
115
116 return 0;
117 }
118
119 static int ppc_aes_setkey_skcipher(struct crypto_skcipher *tfm,
120 const u8 *in_key, unsigned int key_len)
121 {
122 return ppc_aes_setkey(crypto_skcipher_tfm(tfm), in_key, key_len);
123 }
124
125 static int ppc_xts_setkey(struct crypto_skcipher *tfm, const u8 *in_key,
126 unsigned int key_len)
127 {
128 struct ppc_xts_ctx *ctx = crypto_skcipher_ctx(tfm);
129 int err;
130
131 err = xts_verify_key(tfm, in_key, key_len);
132 if (err)
133 return err;
134
135 key_len >>= 1;
136
137 switch (key_len) {
138 case AES_KEYSIZE_128:
139 ctx->rounds = 4;
140 ppc_expand_key_128(ctx->key_enc, in_key);
141 ppc_expand_key_128(ctx->key_twk, in_key + AES_KEYSIZE_128);
142 break;
143 case AES_KEYSIZE_192:
144 ctx->rounds = 5;
145 ppc_expand_key_192(ctx->key_enc, in_key);
146 ppc_expand_key_192(ctx->key_twk, in_key + AES_KEYSIZE_192);
147 break;
148 case AES_KEYSIZE_256:
149 ctx->rounds = 6;
150 ppc_expand_key_256(ctx->key_enc, in_key);
151 ppc_expand_key_256(ctx->key_twk, in_key + AES_KEYSIZE_256);
152 break;
153 default:
154 return -EINVAL;
155 }
156
157 ppc_generate_decrypt_key(ctx->key_dec, ctx->key_enc, key_len);
158
159 return 0;
160 }
161
162 static void ppc_aes_encrypt(struct crypto_tfm *tfm, u8 *out, const u8 *in)
163 {
164 struct ppc_aes_ctx *ctx = crypto_tfm_ctx(tfm);
165
166 spe_begin();
167 ppc_encrypt_aes(out, in, ctx->key_enc, ctx->rounds);
168 spe_end();
169 }
170
171 static void ppc_aes_decrypt(struct crypto_tfm *tfm, u8 *out, const u8 *in)
172 {
173 struct ppc_aes_ctx *ctx = crypto_tfm_ctx(tfm);
174
175 spe_begin();
176 ppc_decrypt_aes(out, in, ctx->key_dec, ctx->rounds);
177 spe_end();
178 }
179
180 static int ppc_ecb_crypt(struct skcipher_request *req, bool enc)
181 {
182 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
183 struct ppc_aes_ctx *ctx = crypto_skcipher_ctx(tfm);
184 struct skcipher_walk walk;
185 unsigned int nbytes;
186 int err;
187
188 err = skcipher_walk_virt(&walk, req, false);
189
190 while ((nbytes = walk.nbytes) != 0) {
191 nbytes = min_t(unsigned int, nbytes, MAX_BYTES);
192 nbytes = round_down(nbytes, AES_BLOCK_SIZE);
193
194 spe_begin();
195 if (enc)
196 ppc_encrypt_ecb(walk.dst.virt.addr, walk.src.virt.addr,
197 ctx->key_enc, ctx->rounds, nbytes);
198 else
199 ppc_decrypt_ecb(walk.dst.virt.addr, walk.src.virt.addr,
200 ctx->key_dec, ctx->rounds, nbytes);
201 spe_end();
202
203 err = skcipher_walk_done(&walk, walk.nbytes - nbytes);
204 }
205
206 return err;
207 }
208
209 static int ppc_ecb_encrypt(struct skcipher_request *req)
210 {
211 return ppc_ecb_crypt(req, true);
212 }
213
214 static int ppc_ecb_decrypt(struct skcipher_request *req)
215 {
216 return ppc_ecb_crypt(req, false);
217 }
218
219 static int ppc_cbc_crypt(struct skcipher_request *req, bool enc)
220 {
221 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
222 struct ppc_aes_ctx *ctx = crypto_skcipher_ctx(tfm);
223 struct skcipher_walk walk;
224 unsigned int nbytes;
225 int err;
226
227 err = skcipher_walk_virt(&walk, req, false);
228
229 while ((nbytes = walk.nbytes) != 0) {
230 nbytes = min_t(unsigned int, nbytes, MAX_BYTES);
231 nbytes = round_down(nbytes, AES_BLOCK_SIZE);
232
233 spe_begin();
234 if (enc)
235 ppc_encrypt_cbc(walk.dst.virt.addr, walk.src.virt.addr,
236 ctx->key_enc, ctx->rounds, nbytes,
237 walk.iv);
238 else
239 ppc_decrypt_cbc(walk.dst.virt.addr, walk.src.virt.addr,
240 ctx->key_dec, ctx->rounds, nbytes,
241 walk.iv);
242 spe_end();
243
244 err = skcipher_walk_done(&walk, walk.nbytes - nbytes);
245 }
246
247 return err;
248 }
249
250 static int ppc_cbc_encrypt(struct skcipher_request *req)
251 {
252 return ppc_cbc_crypt(req, true);
253 }
254
255 static int ppc_cbc_decrypt(struct skcipher_request *req)
256 {
257 return ppc_cbc_crypt(req, false);
258 }
259
260 static int ppc_ctr_crypt(struct skcipher_request *req)
261 {
262 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
263 struct ppc_aes_ctx *ctx = crypto_skcipher_ctx(tfm);
264 struct skcipher_walk walk;
265 unsigned int nbytes;
266 int err;
267
268 err = skcipher_walk_virt(&walk, req, false);
269
270 while ((nbytes = walk.nbytes) != 0) {
271 nbytes = min_t(unsigned int, nbytes, MAX_BYTES);
272 if (nbytes < walk.total)
273 nbytes = round_down(nbytes, AES_BLOCK_SIZE);
274
275 spe_begin();
276 ppc_crypt_ctr(walk.dst.virt.addr, walk.src.virt.addr,
277 ctx->key_enc, ctx->rounds, nbytes, walk.iv);
278 spe_end();
279
280 err = skcipher_walk_done(&walk, walk.nbytes - nbytes);
281 }
282
283 return err;
284 }
285
286 static int ppc_xts_crypt(struct skcipher_request *req, bool enc)
287 {
288 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
289 struct ppc_xts_ctx *ctx = crypto_skcipher_ctx(tfm);
290 struct skcipher_walk walk;
291 unsigned int nbytes;
292 int err;
293 u32 *twk;
294
295 err = skcipher_walk_virt(&walk, req, false);
296 twk = ctx->key_twk;
297
298 while ((nbytes = walk.nbytes) != 0) {
299 nbytes = min_t(unsigned int, nbytes, MAX_BYTES);
300 nbytes = round_down(nbytes, AES_BLOCK_SIZE);
301
302 spe_begin();
303 if (enc)
304 ppc_encrypt_xts(walk.dst.virt.addr, walk.src.virt.addr,
305 ctx->key_enc, ctx->rounds, nbytes,
306 walk.iv, twk);
307 else
308 ppc_decrypt_xts(walk.dst.virt.addr, walk.src.virt.addr,
309 ctx->key_dec, ctx->rounds, nbytes,
310 walk.iv, twk);
311 spe_end();
312
313 twk = NULL;
314 err = skcipher_walk_done(&walk, walk.nbytes - nbytes);
315 }
316
317 return err;
318 }
319
320 static int ppc_xts_encrypt(struct skcipher_request *req)
321 {
322 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
323 struct ppc_xts_ctx *ctx = crypto_skcipher_ctx(tfm);
324 int tail = req->cryptlen % AES_BLOCK_SIZE;
325 int offset = req->cryptlen - tail - AES_BLOCK_SIZE;
326 struct skcipher_request subreq;
327 u8 b[2][AES_BLOCK_SIZE];
328 int err;
329
330 if (req->cryptlen < AES_BLOCK_SIZE)
331 return -EINVAL;
332
333 if (tail) {
334 subreq = *req;
335 skcipher_request_set_crypt(&subreq, req->src, req->dst,
336 req->cryptlen - tail, req->iv);
337 req = &subreq;
338 }
339
340 err = ppc_xts_crypt(req, true);
341 if (err || !tail)
342 return err;
343
344 scatterwalk_map_and_copy(b[0], req->dst, offset, AES_BLOCK_SIZE, 0);
345 memcpy(b[1], b[0], tail);
346 scatterwalk_map_and_copy(b[0], req->src, offset + AES_BLOCK_SIZE, tail, 0);
347
348 spe_begin();
349 ppc_encrypt_xts(b[0], b[0], ctx->key_enc, ctx->rounds, AES_BLOCK_SIZE,
350 req->iv, NULL);
351 spe_end();
352
353 scatterwalk_map_and_copy(b[0], req->dst, offset, AES_BLOCK_SIZE + tail, 1);
354
355 return 0;
356 }
357
358 static int ppc_xts_decrypt(struct skcipher_request *req)
359 {
360 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
361 struct ppc_xts_ctx *ctx = crypto_skcipher_ctx(tfm);
362 int tail = req->cryptlen % AES_BLOCK_SIZE;
363 int offset = req->cryptlen - tail - AES_BLOCK_SIZE;
364 struct skcipher_request subreq;
365 u8 b[3][AES_BLOCK_SIZE];
366 le128 twk;
367 int err;
368
369 if (req->cryptlen < AES_BLOCK_SIZE)
370 return -EINVAL;
371
372 if (tail) {
373 subreq = *req;
374 skcipher_request_set_crypt(&subreq, req->src, req->dst,
375 offset, req->iv);
376 req = &subreq;
377 }
378
379 err = ppc_xts_crypt(req, false);
380 if (err || !tail)
381 return err;
382
383 scatterwalk_map_and_copy(b[1], req->src, offset, AES_BLOCK_SIZE + tail, 0);
384
385 spe_begin();
386 if (!offset)
387 ppc_encrypt_ecb(req->iv, req->iv, ctx->key_twk, ctx->rounds,
388 AES_BLOCK_SIZE);
389
390 gf128mul_x_ble(&twk, (le128 *)req->iv);
391
392 ppc_decrypt_xts(b[1], b[1], ctx->key_dec, ctx->rounds, AES_BLOCK_SIZE,
393 (u8 *)&twk, NULL);
394 memcpy(b[0], b[2], tail);
395 memcpy(b[0] + tail, b[1] + tail, AES_BLOCK_SIZE - tail);
396 ppc_decrypt_xts(b[0], b[0], ctx->key_dec, ctx->rounds, AES_BLOCK_SIZE,
397 req->iv, NULL);
398 spe_end();
399
400 scatterwalk_map_and_copy(b[0], req->dst, offset, AES_BLOCK_SIZE + tail, 1);
401
402 return 0;
403 }
404
405 /*
406 * Algorithm definitions. Disabling alignment (cra_alignmask=0) was chosen
407 * because the e500 platform can handle unaligned reads/writes very efficiently.
408 * This improves IPsec thoughput by another few percent. Additionally we assume
409 * that AES context is always aligned to at least 8 bytes because it is created
410 * with kmalloc() in the crypto infrastructure
411 */
412
413 static struct crypto_alg aes_cipher_alg = {
414 .cra_name = "aes",
415 .cra_driver_name = "aes-ppc-spe",
416 .cra_priority = 300,
417 .cra_flags = CRYPTO_ALG_TYPE_CIPHER,
418 .cra_blocksize = AES_BLOCK_SIZE,
419 .cra_ctxsize = sizeof(struct ppc_aes_ctx),
420 .cra_alignmask = 0,
421 .cra_module = THIS_MODULE,
422 .cra_u = {
423 .cipher = {
424 .cia_min_keysize = AES_MIN_KEY_SIZE,
425 .cia_max_keysize = AES_MAX_KEY_SIZE,
426 .cia_setkey = ppc_aes_setkey,
427 .cia_encrypt = ppc_aes_encrypt,
428 .cia_decrypt = ppc_aes_decrypt
429 }
430 }
431 };
432
433 static struct skcipher_alg aes_skcipher_algs[] = {
434 {
435 .base.cra_name = "ecb(aes)",
436 .base.cra_driver_name = "ecb-ppc-spe",
437 .base.cra_priority = 300,
438 .base.cra_blocksize = AES_BLOCK_SIZE,
439 .base.cra_ctxsize = sizeof(struct ppc_aes_ctx),
440 .base.cra_module = THIS_MODULE,
441 .min_keysize = AES_MIN_KEY_SIZE,
442 .max_keysize = AES_MAX_KEY_SIZE,
443 .setkey = ppc_aes_setkey_skcipher,
444 .encrypt = ppc_ecb_encrypt,
445 .decrypt = ppc_ecb_decrypt,
446 }, {
447 .base.cra_name = "cbc(aes)",
448 .base.cra_driver_name = "cbc-ppc-spe",
449 .base.cra_priority = 300,
450 .base.cra_blocksize = AES_BLOCK_SIZE,
451 .base.cra_ctxsize = sizeof(struct ppc_aes_ctx),
452 .base.cra_module = THIS_MODULE,
453 .min_keysize = AES_MIN_KEY_SIZE,
454 .max_keysize = AES_MAX_KEY_SIZE,
455 .ivsize = AES_BLOCK_SIZE,
456 .setkey = ppc_aes_setkey_skcipher,
457 .encrypt = ppc_cbc_encrypt,
458 .decrypt = ppc_cbc_decrypt,
459 }, {
460 .base.cra_name = "ctr(aes)",
461 .base.cra_driver_name = "ctr-ppc-spe",
462 .base.cra_priority = 300,
463 .base.cra_blocksize = 1,
464 .base.cra_ctxsize = sizeof(struct ppc_aes_ctx),
465 .base.cra_module = THIS_MODULE,
466 .min_keysize = AES_MIN_KEY_SIZE,
467 .max_keysize = AES_MAX_KEY_SIZE,
468 .ivsize = AES_BLOCK_SIZE,
469 .setkey = ppc_aes_setkey_skcipher,
470 .encrypt = ppc_ctr_crypt,
471 .decrypt = ppc_ctr_crypt,
472 .chunksize = AES_BLOCK_SIZE,
473 }, {
474 .base.cra_name = "xts(aes)",
475 .base.cra_driver_name = "xts-ppc-spe",
476 .base.cra_priority = 300,
477 .base.cra_blocksize = AES_BLOCK_SIZE,
478 .base.cra_ctxsize = sizeof(struct ppc_xts_ctx),
479 .base.cra_module = THIS_MODULE,
480 .min_keysize = AES_MIN_KEY_SIZE * 2,
481 .max_keysize = AES_MAX_KEY_SIZE * 2,
482 .ivsize = AES_BLOCK_SIZE,
483 .setkey = ppc_xts_setkey,
484 .encrypt = ppc_xts_encrypt,
485 .decrypt = ppc_xts_decrypt,
486 }
487 };
488
489 static int __init ppc_aes_mod_init(void)
490 {
491 int err;
492
493 err = crypto_register_alg(&aes_cipher_alg);
494 if (err)
495 return err;
496
497 err = crypto_register_skciphers(aes_skcipher_algs,
498 ARRAY_SIZE(aes_skcipher_algs));
499 if (err)
500 crypto_unregister_alg(&aes_cipher_alg);
501 return err;
502 }
503
504 static void __exit ppc_aes_mod_fini(void)
505 {
506 crypto_unregister_alg(&aes_cipher_alg);
507 crypto_unregister_skciphers(aes_skcipher_algs,
508 ARRAY_SIZE(aes_skcipher_algs));
509 }
510
511 module_init(ppc_aes_mod_init);
512 module_exit(ppc_aes_mod_fini);
513
514 MODULE_LICENSE("GPL");
515 MODULE_DESCRIPTION("AES-ECB/CBC/CTR/XTS, SPE optimized");
516
517 MODULE_ALIAS_CRYPTO("aes");
518 MODULE_ALIAS_CRYPTO("ecb(aes)");
519 MODULE_ALIAS_CRYPTO("cbc(aes)");
520 MODULE_ALIAS_CRYPTO("ctr(aes)");
521 MODULE_ALIAS_CRYPTO("xts(aes)");
522 MODULE_ALIAS_CRYPTO("aes-ppc-spe");
523
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.