projects / openwrt / staging / pepe2k.git / blob
? search:


7f907f23641d6997c5cea47f1028eb1788146ebd
[openwrt/staging/pepe2k.git] /
1 From a92bd97c758d32511f0deeef84f25c3a1d5e7879 Mon Sep 17 00:00:00 2001
2 From: Ard Biesheuvel <ardb@kernel.org>
3 Date: Fri, 8 Nov 2019 13:22:14 +0100
4 Subject: [PATCH 008/124] crypto: arm/chacha - remove dependency on generic
5 ChaCha driver
6
7 commit b36d8c09e710c71f6a9690b6586fea2d1c9e1e27 upstream.
8
9 Instead of falling back to the generic ChaCha skcipher driver for
10 non-SIMD cases, use a fast scalar implementation for ARM authored
11 by Eric Biggers. This removes the module dependency on chacha-generic
12 altogether, which also simplifies things when we expose the ChaCha
13 library interface from this module.
14
15 Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
16 Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
18 ---
19 arch/arm/crypto/Kconfig | 4 +-
20 arch/arm/crypto/Makefile | 3 +-
21 arch/arm/crypto/chacha-glue.c | 304 +++++++++++++++++++++++++++
22 arch/arm/crypto/chacha-neon-glue.c | 202 ------------------
23 arch/arm/crypto/chacha-scalar-core.S | 65 +++---
24 arch/arm64/crypto/chacha-neon-glue.c | 2 +-
25 6 files changed, 340 insertions(+), 240 deletions(-)
26 create mode 100644 arch/arm/crypto/chacha-glue.c
27 delete mode 100644 arch/arm/crypto/chacha-neon-glue.c
28
29 --- a/arch/arm/crypto/Kconfig
30 +++ b/arch/arm/crypto/Kconfig
31 @@ -127,10 +127,8 @@ config CRYPTO_CRC32_ARM_CE
32 select CRYPTO_HASH
33
34 config CRYPTO_CHACHA20_NEON
35 - tristate "NEON accelerated ChaCha stream cipher algorithms"
36 - depends on KERNEL_MODE_NEON
37 + tristate "NEON and scalar accelerated ChaCha stream cipher algorithms"
38 select CRYPTO_BLKCIPHER
39 - select CRYPTO_CHACHA20
40
41 config CRYPTO_NHPOLY1305_NEON
42 tristate "NEON accelerated NHPoly1305 hash function (for Adiantum)"
43 --- a/arch/arm/crypto/Makefile
44 +++ b/arch/arm/crypto/Makefile
45 @@ -53,7 +53,8 @@ aes-arm-ce-y := aes-ce-core.o aes-ce-glu
46 ghash-arm-ce-y := ghash-ce-core.o ghash-ce-glue.o
47 crct10dif-arm-ce-y := crct10dif-ce-core.o crct10dif-ce-glue.o
48 crc32-arm-ce-y:= crc32-ce-core.o crc32-ce-glue.o
49 -chacha-neon-y := chacha-neon-core.o chacha-neon-glue.o
50 +chacha-neon-y := chacha-scalar-core.o chacha-glue.o
51 +chacha-neon-$(CONFIG_KERNEL_MODE_NEON) += chacha-neon-core.o
52 nhpoly1305-neon-y := nh-neon-core.o nhpoly1305-neon-glue.o
53
54 ifdef REGENERATE_ARM_CRYPTO
55 --- /dev/null
56 +++ b/arch/arm/crypto/chacha-glue.c
57 @@ -0,0 +1,304 @@
58 +// SPDX-License-Identifier: GPL-2.0
59 +/*
60 + * ARM NEON accelerated ChaCha and XChaCha stream ciphers,
61 + * including ChaCha20 (RFC7539)
62 + *
63 + * Copyright (C) 2016-2019 Linaro, Ltd. <ard.biesheuvel@linaro.org>
64 + * Copyright (C) 2015 Martin Willi
65 + */
66 +
67 +#include <crypto/algapi.h>
68 +#include <crypto/internal/chacha.h>
69 +#include <crypto/internal/simd.h>
70 +#include <crypto/internal/skcipher.h>
71 +#include <linux/kernel.h>
72 +#include <linux/module.h>
73 +
74 +#include <asm/cputype.h>
75 +#include <asm/hwcap.h>
76 +#include <asm/neon.h>
77 +#include <asm/simd.h>
78 +
79 +asmlinkage void chacha_block_xor_neon(const u32 *state, u8 *dst, const u8 *src,
80 + int nrounds);
81 +asmlinkage void chacha_4block_xor_neon(const u32 *state, u8 *dst, const u8 *src,
82 + int nrounds);
83 +asmlinkage void hchacha_block_arm(const u32 *state, u32 *out, int nrounds);
84 +asmlinkage void hchacha_block_neon(const u32 *state, u32 *out, int nrounds);
85 +
86 +asmlinkage void chacha_doarm(u8 *dst, const u8 *src, unsigned int bytes,
87 + const u32 *state, int nrounds);
88 +
89 +static inline bool neon_usable(void)
90 +{
91 + return crypto_simd_usable();
92 +}
93 +
94 +static void chacha_doneon(u32 *state, u8 *dst, const u8 *src,
95 + unsigned int bytes, int nrounds)
96 +{
97 + u8 buf[CHACHA_BLOCK_SIZE];
98 +
99 + while (bytes >= CHACHA_BLOCK_SIZE * 4) {
100 + chacha_4block_xor_neon(state, dst, src, nrounds);
101 + bytes -= CHACHA_BLOCK_SIZE * 4;
102 + src += CHACHA_BLOCK_SIZE * 4;
103 + dst += CHACHA_BLOCK_SIZE * 4;
104 + state[12] += 4;
105 + }
106 + while (bytes >= CHACHA_BLOCK_SIZE) {
107 + chacha_block_xor_neon(state, dst, src, nrounds);
108 + bytes -= CHACHA_BLOCK_SIZE;
109 + src += CHACHA_BLOCK_SIZE;
110 + dst += CHACHA_BLOCK_SIZE;
111 + state[12]++;
112 + }
113 + if (bytes) {
114 + memcpy(buf, src, bytes);
115 + chacha_block_xor_neon(state, buf, buf, nrounds);
116 + memcpy(dst, buf, bytes);
117 + }
118 +}
119 +
120 +static int chacha_stream_xor(struct skcipher_request *req,
121 + const struct chacha_ctx *ctx, const u8 *iv,
122 + bool neon)
123 +{
124 + struct skcipher_walk walk;
125 + u32 state[16];
126 + int err;
127 +
128 + err = skcipher_walk_virt(&walk, req, false);
129 +
130 + chacha_init_generic(state, ctx->key, iv);
131 +
132 + while (walk.nbytes > 0) {
133 + unsigned int nbytes = walk.nbytes;
134 +
135 + if (nbytes < walk.total)
136 + nbytes = round_down(nbytes, walk.stride);
137 +
138 + if (!neon) {
139 + chacha_doarm(walk.dst.virt.addr, walk.src.virt.addr,
140 + nbytes, state, ctx->nrounds);
141 + state[12] += DIV_ROUND_UP(nbytes, CHACHA_BLOCK_SIZE);
142 + } else {
143 + kernel_neon_begin();
144 + chacha_doneon(state, walk.dst.virt.addr,
145 + walk.src.virt.addr, nbytes, ctx->nrounds);
146 + kernel_neon_end();
147 + }
148 + err = skcipher_walk_done(&walk, walk.nbytes - nbytes);
149 + }
150 +
151 + return err;
152 +}
153 +
154 +static int do_chacha(struct skcipher_request *req, bool neon)
155 +{
156 + struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
157 + struct chacha_ctx *ctx = crypto_skcipher_ctx(tfm);
158 +
159 + return chacha_stream_xor(req, ctx, req->iv, neon);
160 +}
161 +
162 +static int chacha_arm(struct skcipher_request *req)
163 +{
164 + return do_chacha(req, false);
165 +}
166 +
167 +static int chacha_neon(struct skcipher_request *req)
168 +{
169 + return do_chacha(req, neon_usable());
170 +}
171 +
172 +static int do_xchacha(struct skcipher_request *req, bool neon)
173 +{
174 + struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
175 + struct chacha_ctx *ctx = crypto_skcipher_ctx(tfm);
176 + struct chacha_ctx subctx;
177 + u32 state[16];
178 + u8 real_iv[16];
179 +
180 + chacha_init_generic(state, ctx->key, req->iv);
181 +
182 + if (!neon) {
183 + hchacha_block_arm(state, subctx.key, ctx->nrounds);
184 + } else {
185 + kernel_neon_begin();
186 + hchacha_block_neon(state, subctx.key, ctx->nrounds);
187 + kernel_neon_end();
188 + }
189 + subctx.nrounds = ctx->nrounds;
190 +
191 + memcpy(&real_iv[0], req->iv + 24, 8);
192 + memcpy(&real_iv[8], req->iv + 16, 8);
193 + return chacha_stream_xor(req, &subctx, real_iv, neon);
194 +}
195 +
196 +static int xchacha_arm(struct skcipher_request *req)
197 +{
198 + return do_xchacha(req, false);
199 +}
200 +
201 +static int xchacha_neon(struct skcipher_request *req)
202 +{
203 + return do_xchacha(req, neon_usable());
204 +}
205 +
206 +static struct skcipher_alg arm_algs[] = {
207 + {
208 + .base.cra_name = "chacha20",
209 + .base.cra_driver_name = "chacha20-arm",
210 + .base.cra_priority = 200,
211 + .base.cra_blocksize = 1,
212 + .base.cra_ctxsize = sizeof(struct chacha_ctx),
213 + .base.cra_module = THIS_MODULE,
214 +
215 + .min_keysize = CHACHA_KEY_SIZE,
216 + .max_keysize = CHACHA_KEY_SIZE,
217 + .ivsize = CHACHA_IV_SIZE,
218 + .chunksize = CHACHA_BLOCK_SIZE,
219 + .setkey = chacha20_setkey,
220 + .encrypt = chacha_arm,
221 + .decrypt = chacha_arm,
222 + }, {
223 + .base.cra_name = "xchacha20",
224 + .base.cra_driver_name = "xchacha20-arm",
225 + .base.cra_priority = 200,
226 + .base.cra_blocksize = 1,
227 + .base.cra_ctxsize = sizeof(struct chacha_ctx),
228 + .base.cra_module = THIS_MODULE,
229 +
230 + .min_keysize = CHACHA_KEY_SIZE,
231 + .max_keysize = CHACHA_KEY_SIZE,
232 + .ivsize = XCHACHA_IV_SIZE,
233 + .chunksize = CHACHA_BLOCK_SIZE,
234 + .setkey = chacha20_setkey,
235 + .encrypt = xchacha_arm,
236 + .decrypt = xchacha_arm,
237 + }, {
238 + .base.cra_name = "xchacha12",
239 + .base.cra_driver_name = "xchacha12-arm",
240 + .base.cra_priority = 200,
241 + .base.cra_blocksize = 1,
242 + .base.cra_ctxsize = sizeof(struct chacha_ctx),
243 + .base.cra_module = THIS_MODULE,
244 +
245 + .min_keysize = CHACHA_KEY_SIZE,
246 + .max_keysize = CHACHA_KEY_SIZE,
247 + .ivsize = XCHACHA_IV_SIZE,
248 + .chunksize = CHACHA_BLOCK_SIZE,
249 + .setkey = chacha12_setkey,
250 + .encrypt = xchacha_arm,
251 + .decrypt = xchacha_arm,
252 + },
253 +};
254 +
255 +static struct skcipher_alg neon_algs[] = {
256 + {
257 + .base.cra_name = "chacha20",
258 + .base.cra_driver_name = "chacha20-neon",
259 + .base.cra_priority = 300,
260 + .base.cra_blocksize = 1,
261 + .base.cra_ctxsize = sizeof(struct chacha_ctx),
262 + .base.cra_module = THIS_MODULE,
263 +
264 + .min_keysize = CHACHA_KEY_SIZE,
265 + .max_keysize = CHACHA_KEY_SIZE,
266 + .ivsize = CHACHA_IV_SIZE,
267 + .chunksize = CHACHA_BLOCK_SIZE,
268 + .walksize = 4 * CHACHA_BLOCK_SIZE,
269 + .setkey = chacha20_setkey,
270 + .encrypt = chacha_neon,
271 + .decrypt = chacha_neon,
272 + }, {
273 + .base.cra_name = "xchacha20",
274 + .base.cra_driver_name = "xchacha20-neon",
275 + .base.cra_priority = 300,
276 + .base.cra_blocksize = 1,
277 + .base.cra_ctxsize = sizeof(struct chacha_ctx),
278 + .base.cra_module = THIS_MODULE,
279 +
280 + .min_keysize = CHACHA_KEY_SIZE,
281 + .max_keysize = CHACHA_KEY_SIZE,
282 + .ivsize = XCHACHA_IV_SIZE,
283 + .chunksize = CHACHA_BLOCK_SIZE,
284 + .walksize = 4 * CHACHA_BLOCK_SIZE,
285 + .setkey = chacha20_setkey,
286 + .encrypt = xchacha_neon,
287 + .decrypt = xchacha_neon,
288 + }, {
289 + .base.cra_name = "xchacha12",
290 + .base.cra_driver_name = "xchacha12-neon",
291 + .base.cra_priority = 300,
292 + .base.cra_blocksize = 1,
293 + .base.cra_ctxsize = sizeof(struct chacha_ctx),
294 + .base.cra_module = THIS_MODULE,
295 +
296 + .min_keysize = CHACHA_KEY_SIZE,
297 + .max_keysize = CHACHA_KEY_SIZE,
298 + .ivsize = XCHACHA_IV_SIZE,
299 + .chunksize = CHACHA_BLOCK_SIZE,
300 + .walksize = 4 * CHACHA_BLOCK_SIZE,
301 + .setkey = chacha12_setkey,
302 + .encrypt = xchacha_neon,
303 + .decrypt = xchacha_neon,
304 + }
305 +};
306 +
307 +static int __init chacha_simd_mod_init(void)
308 +{
309 + int err;
310 +
311 + err = crypto_register_skciphers(arm_algs, ARRAY_SIZE(arm_algs));
312 + if (err)
313 + return err;
314 +
315 + if (IS_ENABLED(CONFIG_KERNEL_MODE_NEON) && (elf_hwcap & HWCAP_NEON)) {
316 + int i;
317 +
318 + switch (read_cpuid_part()) {
319 + case ARM_CPU_PART_CORTEX_A7:
320 + case ARM_CPU_PART_CORTEX_A5:
321 + /*
322 + * The Cortex-A7 and Cortex-A5 do not perform well with
323 + * the NEON implementation but do incredibly with the
324 + * scalar one and use less power.
325 + */
326 + for (i = 0; i < ARRAY_SIZE(neon_algs); i++)
327 + neon_algs[i].base.cra_priority = 0;
328 + break;
329 + }
330 +
331 + err = crypto_register_skciphers(neon_algs, ARRAY_SIZE(neon_algs));
332 + if (err)
333 + crypto_unregister_skciphers(arm_algs, ARRAY_SIZE(arm_algs));
334 + }
335 + return err;
336 +}
337 +
338 +static void __exit chacha_simd_mod_fini(void)
339 +{
340 + crypto_unregister_skciphers(arm_algs, ARRAY_SIZE(arm_algs));
341 + if (IS_ENABLED(CONFIG_KERNEL_MODE_NEON) && (elf_hwcap & HWCAP_NEON))
342 + crypto_unregister_skciphers(neon_algs, ARRAY_SIZE(neon_algs));
343 +}
344 +
345 +module_init(chacha_simd_mod_init);
346 +module_exit(chacha_simd_mod_fini);
347 +
348 +MODULE_DESCRIPTION("ChaCha and XChaCha stream ciphers (scalar and NEON accelerated)");
349 +MODULE_AUTHOR("Ard Biesheuvel <ard.biesheuvel@linaro.org>");
350 +MODULE_LICENSE("GPL v2");
351 +MODULE_ALIAS_CRYPTO("chacha20");
352 +MODULE_ALIAS_CRYPTO("chacha20-arm");
353 +MODULE_ALIAS_CRYPTO("xchacha20");
354 +MODULE_ALIAS_CRYPTO("xchacha20-arm");
355 +MODULE_ALIAS_CRYPTO("xchacha12");
356 +MODULE_ALIAS_CRYPTO("xchacha12-arm");
357 +#ifdef CONFIG_KERNEL_MODE_NEON
358 +MODULE_ALIAS_CRYPTO("chacha20-neon");
359 +MODULE_ALIAS_CRYPTO("xchacha20-neon");
360 +MODULE_ALIAS_CRYPTO("xchacha12-neon");
361 +#endif
362 --- a/arch/arm/crypto/chacha-neon-glue.c
363 +++ /dev/null
364 @@ -1,202 +0,0 @@
365 -/*
366 - * ARM NEON accelerated ChaCha and XChaCha stream ciphers,
367 - * including ChaCha20 (RFC7539)
368 - *
369 - * Copyright (C) 2016 Linaro, Ltd. <ard.biesheuvel@linaro.org>
370 - *
371 - * This program is free software; you can redistribute it and/or modify
372 - * it under the terms of the GNU General Public License version 2 as
373 - * published by the Free Software Foundation.
374 - *
375 - * Based on:
376 - * ChaCha20 256-bit cipher algorithm, RFC7539, SIMD glue code
377 - *
378 - * Copyright (C) 2015 Martin Willi
379 - *
380 - * This program is free software; you can redistribute it and/or modify
381 - * it under the terms of the GNU General Public License as published by
382 - * the Free Software Foundation; either version 2 of the License, or
383 - * (at your option) any later version.
384 - */
385 -
386 -#include <crypto/algapi.h>
387 -#include <crypto/internal/chacha.h>
388 -#include <crypto/internal/simd.h>
389 -#include <crypto/internal/skcipher.h>
390 -#include <linux/kernel.h>
391 -#include <linux/module.h>
392 -
393 -#include <asm/hwcap.h>
394 -#include <asm/neon.h>
395 -#include <asm/simd.h>
396 -
397 -asmlinkage void chacha_block_xor_neon(const u32 *state, u8 *dst, const u8 *src,
398 - int nrounds);
399 -asmlinkage void chacha_4block_xor_neon(const u32 *state, u8 *dst, const u8 *src,
400 - int nrounds);
401 -asmlinkage void hchacha_block_neon(const u32 *state, u32 *out, int nrounds);
402 -
403 -static void chacha_doneon(u32 *state, u8 *dst, const u8 *src,
404 - unsigned int bytes, int nrounds)
405 -{
406 - u8 buf[CHACHA_BLOCK_SIZE];
407 -
408 - while (bytes >= CHACHA_BLOCK_SIZE * 4) {
409 - chacha_4block_xor_neon(state, dst, src, nrounds);
410 - bytes -= CHACHA_BLOCK_SIZE * 4;
411 - src += CHACHA_BLOCK_SIZE * 4;
412 - dst += CHACHA_BLOCK_SIZE * 4;
413 - state[12] += 4;
414 - }
415 - while (bytes >= CHACHA_BLOCK_SIZE) {
416 - chacha_block_xor_neon(state, dst, src, nrounds);
417 - bytes -= CHACHA_BLOCK_SIZE;
418 - src += CHACHA_BLOCK_SIZE;
419 - dst += CHACHA_BLOCK_SIZE;
420 - state[12]++;
421 - }
422 - if (bytes) {
423 - memcpy(buf, src, bytes);
424 - chacha_block_xor_neon(state, buf, buf, nrounds);
425 - memcpy(dst, buf, bytes);
426 - }
427 -}
428 -
429 -static int chacha_neon_stream_xor(struct skcipher_request *req,
430 - const struct chacha_ctx *ctx, const u8 *iv)
431 -{
432 - struct skcipher_walk walk;
433 - u32 state[16];
434 - int err;
435 -
436 - err = skcipher_walk_virt(&walk, req, false);
437 -
438 - crypto_chacha_init(state, ctx, iv);
439 -
440 - while (walk.nbytes > 0) {
441 - unsigned int nbytes = walk.nbytes;
442 -
443 - if (nbytes < walk.total)
444 - nbytes = round_down(nbytes, walk.stride);
445 -
446 - kernel_neon_begin();
447 - chacha_doneon(state, walk.dst.virt.addr, walk.src.virt.addr,
448 - nbytes, ctx->nrounds);
449 - kernel_neon_end();
450 - err = skcipher_walk_done(&walk, walk.nbytes - nbytes);
451 - }
452 -
453 - return err;
454 -}
455 -
456 -static int chacha_neon(struct skcipher_request *req)
457 -{
458 - struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
459 - struct chacha_ctx *ctx = crypto_skcipher_ctx(tfm);
460 -
461 - if (req->cryptlen <= CHACHA_BLOCK_SIZE || !crypto_simd_usable())
462 - return crypto_chacha_crypt(req);
463 -
464 - return chacha_neon_stream_xor(req, ctx, req->iv);
465 -}
466 -
467 -static int xchacha_neon(struct skcipher_request *req)
468 -{
469 - struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
470 - struct chacha_ctx *ctx = crypto_skcipher_ctx(tfm);
471 - struct chacha_ctx subctx;
472 - u32 state[16];
473 - u8 real_iv[16];
474 -
475 - if (req->cryptlen <= CHACHA_BLOCK_SIZE || !crypto_simd_usable())
476 - return crypto_xchacha_crypt(req);
477 -
478 - crypto_chacha_init(state, ctx, req->iv);
479 -
480 - kernel_neon_begin();
481 - hchacha_block_neon(state, subctx.key, ctx->nrounds);
482 - kernel_neon_end();
483 - subctx.nrounds = ctx->nrounds;
484 -
485 - memcpy(&real_iv[0], req->iv + 24, 8);
486 - memcpy(&real_iv[8], req->iv + 16, 8);
487 - return chacha_neon_stream_xor(req, &subctx, real_iv);
488 -}
489 -
490 -static struct skcipher_alg algs[] = {
491 - {
492 - .base.cra_name = "chacha20",
493 - .base.cra_driver_name = "chacha20-neon",
494 - .base.cra_priority = 300,
495 - .base.cra_blocksize = 1,
496 - .base.cra_ctxsize = sizeof(struct chacha_ctx),
497 - .base.cra_module = THIS_MODULE,
498 -
499 - .min_keysize = CHACHA_KEY_SIZE,
500 - .max_keysize = CHACHA_KEY_SIZE,
501 - .ivsize = CHACHA_IV_SIZE,
502 - .chunksize = CHACHA_BLOCK_SIZE,
503 - .walksize = 4 * CHACHA_BLOCK_SIZE,
504 - .setkey = crypto_chacha20_setkey,
505 - .encrypt = chacha_neon,
506 - .decrypt = chacha_neon,
507 - }, {
508 - .base.cra_name = "xchacha20",
509 - .base.cra_driver_name = "xchacha20-neon",
510 - .base.cra_priority = 300,
511 - .base.cra_blocksize = 1,
512 - .base.cra_ctxsize = sizeof(struct chacha_ctx),
513 - .base.cra_module = THIS_MODULE,
514 -
515 - .min_keysize = CHACHA_KEY_SIZE,
516 - .max_keysize = CHACHA_KEY_SIZE,
517 - .ivsize = XCHACHA_IV_SIZE,
518 - .chunksize = CHACHA_BLOCK_SIZE,
519 - .walksize = 4 * CHACHA_BLOCK_SIZE,
520 - .setkey = crypto_chacha20_setkey,
521 - .encrypt = xchacha_neon,
522 - .decrypt = xchacha_neon,
523 - }, {
524 - .base.cra_name = "xchacha12",
525 - .base.cra_driver_name = "xchacha12-neon",
526 - .base.cra_priority = 300,
527 - .base.cra_blocksize = 1,
528 - .base.cra_ctxsize = sizeof(struct chacha_ctx),
529 - .base.cra_module = THIS_MODULE,
530 -
531 - .min_keysize = CHACHA_KEY_SIZE,
532 - .max_keysize = CHACHA_KEY_SIZE,
533 - .ivsize = XCHACHA_IV_SIZE,
534 - .chunksize = CHACHA_BLOCK_SIZE,
535 - .walksize = 4 * CHACHA_BLOCK_SIZE,
536 - .setkey = crypto_chacha12_setkey,
537 - .encrypt = xchacha_neon,
538 - .decrypt = xchacha_neon,
539 - }
540 -};
541 -
542 -static int __init chacha_simd_mod_init(void)
543 -{
544 - if (!(elf_hwcap & HWCAP_NEON))
545 - return -ENODEV;
546 -
547 - return crypto_register_skciphers(algs, ARRAY_SIZE(algs));
548 -}
549 -
550 -static void __exit chacha_simd_mod_fini(void)
551 -{
552 - crypto_unregister_skciphers(algs, ARRAY_SIZE(algs));
553 -}
554 -
555 -module_init(chacha_simd_mod_init);
556 -module_exit(chacha_simd_mod_fini);
557 -
558 -MODULE_DESCRIPTION("ChaCha and XChaCha stream ciphers (NEON accelerated)");
559 -MODULE_AUTHOR("Ard Biesheuvel <ard.biesheuvel@linaro.org>");
560 -MODULE_LICENSE("GPL v2");
561 -MODULE_ALIAS_CRYPTO("chacha20");
562 -MODULE_ALIAS_CRYPTO("chacha20-neon");
563 -MODULE_ALIAS_CRYPTO("xchacha20");
564 -MODULE_ALIAS_CRYPTO("xchacha20-neon");
565 -MODULE_ALIAS_CRYPTO("xchacha12");
566 -MODULE_ALIAS_CRYPTO("xchacha12-neon");
567 --- a/arch/arm/crypto/chacha-scalar-core.S
568 +++ b/arch/arm/crypto/chacha-scalar-core.S
569 @@ -41,14 +41,6 @@
570 X14 .req r12
571 X15 .req r14
572
573 -.Lexpand_32byte_k:
574 - // "expand 32-byte k"
575 - .word 0x61707865, 0x3320646e, 0x79622d32, 0x6b206574
576 -
577 -#ifdef __thumb2__
578 -# define adrl adr
579 -#endif
580 -
581 .macro __rev out, in, t0, t1, t2
582 .if __LINUX_ARM_ARCH__ >= 6
583 rev \out, \in
584 @@ -391,61 +383,65 @@
585 .endm // _chacha
586
587 /*
588 - * void chacha20_arm(u8 *out, const u8 *in, size_t len, const u32 key[8],
589 - * const u32 iv[4]);
590 + * void chacha_doarm(u8 *dst, const u8 *src, unsigned int bytes,
591 + * const u32 *state, int nrounds);
592 */
593 -ENTRY(chacha20_arm)
594 +ENTRY(chacha_doarm)
595 cmp r2, #0 // len == 0?
596 reteq lr
597
598 + ldr ip, [sp]
599 + cmp ip, #12
600 +
601 push {r0-r2,r4-r11,lr}
602
603 // Push state x0-x15 onto stack.
604 // Also store an extra copy of x10-x11 just before the state.
605
606 - ldr r4, [sp, #48] // iv
607 - mov r0, sp
608 - sub sp, #80
609 -
610 - // iv: x12-x15
611 - ldm r4, {X12,X13,X14,X15}
612 - stmdb r0!, {X12,X13,X14,X15}
613 + add X12, r3, #48
614 + ldm X12, {X12,X13,X14,X15}
615 + push {X12,X13,X14,X15}
616 + sub sp, sp, #64
617
618 - // key: x4-x11
619 - __ldrd X8_X10, X9_X11, r3, 24
620 + __ldrd X8_X10, X9_X11, r3, 40
621 __strd X8_X10, X9_X11, sp, 8
622 - stmdb r0!, {X8_X10, X9_X11}
623 - ldm r3, {X4-X9_X11}
624 - stmdb r0!, {X4-X9_X11}
625 -
626 - // constants: x0-x3
627 - adrl X3, .Lexpand_32byte_k
628 - ldm X3, {X0-X3}
629 + __strd X8_X10, X9_X11, sp, 56
630 + ldm r3, {X0-X9_X11}
631 __strd X0, X1, sp, 16
632 __strd X2, X3, sp, 24
633 + __strd X4, X5, sp, 32
634 + __strd X6, X7, sp, 40
635 + __strd X8_X10, X9_X11, sp, 48
636
637 + beq 1f
638 _chacha 20
639
640 - add sp, #76
641 +0: add sp, #76
642 pop {r4-r11, pc}
643 -ENDPROC(chacha20_arm)
644 +
645 +1: _chacha 12
646 + b 0b
647 +ENDPROC(chacha_doarm)
648
649 /*
650 - * void hchacha20_arm(const u32 state[16], u32 out[8]);
651 + * void hchacha_block_arm(const u32 state[16], u32 out[8], int nrounds);
652 */
653 -ENTRY(hchacha20_arm)
654 +ENTRY(hchacha_block_arm)
655 push {r1,r4-r11,lr}
656
657 + cmp r2, #12 // ChaCha12 ?
658 +
659 mov r14, r0
660 ldmia r14!, {r0-r11} // load x0-x11
661 push {r10-r11} // store x10-x11 to stack
662 ldm r14, {r10-r12,r14} // load x12-x15
663 sub sp, #8
664
665 + beq 1f
666 _chacha_permute 20
667
668 // Skip over (unused0-unused1, x10-x11)
669 - add sp, #16
670 +0: add sp, #16
671
672 // Fix up rotations of x12-x15
673 ror X12, X12, #drot
674 @@ -458,4 +454,7 @@ ENTRY(hchacha20_arm)
675 stm r4, {X0,X1,X2,X3,X12,X13,X14,X15}
676
677 pop {r4-r11,pc}
678 -ENDPROC(hchacha20_arm)
679 +
680 +1: _chacha_permute 12
681 + b 0b
682 +ENDPROC(hchacha_block_arm)
683 --- a/arch/arm64/crypto/chacha-neon-glue.c
684 +++ b/arch/arm64/crypto/chacha-neon-glue.c
685 @@ -1,5 +1,5 @@
686 /*
687 - * ARM NEON accelerated ChaCha and XChaCha stream ciphers,
688 + * ARM NEON and scalar accelerated ChaCha and XChaCha stream ciphers,
689 * including ChaCha20 (RFC7539)
690 *
691 * Copyright (C) 2016 - 2017 Linaro, Ltd. <ard.biesheuvel@linaro.org>
Staging tree of Piotr Dymacz