1 /***********************************************************************
3 ** Implementation of the Skein block functions.
5 ** Source code author: Doug Whiting, 2008.
7 ** This algorithm and source code is released to the public domain.
9 ** Compile-time switches:
11 ** SKEIN_USE_ASM -- set bits (256/512/1024) to select which
12 ** versions use ASM code for block processing
13 ** [default: use C for all block sizes]
15 ************************************************************************/
17 #include <linux/string.h>
18 #include <linux/bitops.h>
19 #include "skein_base.h"
20 #include "skein_block.h"
23 #define SKEIN_USE_ASM (0) /* default is all C code (no ASM) */
27 #define SKEIN_LOOP 001 /* default: unroll 256 and 512, but not 1024 */
30 #define BLK_BITS (WCNT * 64) /* some useful definitions for code here */
31 #define KW_TWK_BASE (0)
32 #define KW_KEY_BASE (3)
33 #define ks (kw + KW_KEY_BASE)
34 #define ts (kw + KW_TWK_BASE)
37 #define debug_save_tweak(ctx) \
39 ctx->h.tweak[0] = ts[0]; \
40 ctx->h.tweak[1] = ts[1]; \
43 #define debug_save_tweak(ctx)
46 #if !(SKEIN_USE_ASM & 256)
48 #define RCNT (SKEIN_256_ROUNDS_TOTAL / 8)
49 #ifdef SKEIN_LOOP /* configure how much to unroll the loop */
50 #define SKEIN_UNROLL_256 (((SKEIN_LOOP) / 100) % 10)
52 #define SKEIN_UNROLL_256 (0)
56 #if (RCNT % SKEIN_UNROLL_256)
57 #error "Invalid SKEIN_UNROLL_256" /* sanity check on unroll count */
60 #define ROUND256(p0, p1, p2, p3, ROT, r_num) \
63 X##p1 = rol64(X##p1, ROT##_0); \
66 X##p3 = rol64(X##p3, ROT##_1); \
70 #if SKEIN_UNROLL_256 == 0
71 #define R256(p0, p1, p2, p3, ROT, r_num) /* fully unrolled */ \
72 ROUND256(p0, p1, p2, p3, ROT, r_num)
76 /* inject the key schedule value */ \
77 X0 += ks[((R) + 1) % 5]; \
78 X1 += ks[((R) + 2) % 5] + ts[((R) + 1) % 3]; \
79 X2 += ks[((R) + 3) % 5] + ts[((R) + 2) % 3]; \
80 X3 += ks[((R) + 4) % 5] + (R) + 1; \
84 #define R256(p0, p1, p2, p3, ROT, r_num) ROUND256(p0, p1, p2, p3, ROT, r_num)
88 /* inject the key schedule value */ \
89 X0 += ks[r + (R) + 0]; \
90 X1 += ks[r + (R) + 1] + ts[r + (R) + 0];\
91 X2 += ks[r + (R) + 2] + ts[r + (R) + 1];\
92 X3 += ks[r + (R) + 3] + r + (R); \
93 /* rotate key schedule */ \
94 ks[r + (R) + 4] = ks[r + (R) - 1]; \
95 ts[r + (R) + 2] = ts[r + (R) - 1]; \
98 #define R256_8_ROUNDS(R) \
100 R256(0, 1, 2, 3, R_256_0, 8 * (R) + 1); \
101 R256(0, 3, 2, 1, R_256_1, 8 * (R) + 2); \
102 R256(0, 1, 2, 3, R_256_2, 8 * (R) + 3); \
103 R256(0, 3, 2, 1, R_256_3, 8 * (R) + 4); \
105 R256(0, 1, 2, 3, R_256_4, 8 * (R) + 5); \
106 R256(0, 3, 2, 1, R_256_5, 8 * (R) + 6); \
107 R256(0, 1, 2, 3, R_256_6, 8 * (R) + 7); \
108 R256(0, 3, 2, 1, R_256_7, 8 * (R) + 8); \
112 #define R256_UNROLL_R(NN) \
113 ((SKEIN_UNROLL_256 == 0 && \
114 SKEIN_256_ROUNDS_TOTAL / 8 > (NN)) || \
115 (SKEIN_UNROLL_256 > (NN)))
117 #if (SKEIN_UNROLL_256 > 14)
118 #error "need more unrolling in skein_256_process_block"
122 #if !(SKEIN_USE_ASM & 512)
124 #define RCNT (SKEIN_512_ROUNDS_TOTAL / 8)
126 #ifdef SKEIN_LOOP /* configure how much to unroll the loop */
127 #define SKEIN_UNROLL_512 (((SKEIN_LOOP) / 10) % 10)
129 #define SKEIN_UNROLL_512 (0)
133 #if (RCNT % SKEIN_UNROLL_512)
134 #error "Invalid SKEIN_UNROLL_512" /* sanity check on unroll count */
137 #define ROUND512(p0, p1, p2, p3, p4, p5, p6, p7, ROT, r_num) \
140 X##p1 = rol64(X##p1, ROT##_0); \
143 X##p3 = rol64(X##p3, ROT##_1); \
146 X##p5 = rol64(X##p5, ROT##_2); \
149 X##p7 = rol64(X##p7, ROT##_3); \
153 #if SKEIN_UNROLL_512 == 0
154 #define R512(p0, p1, p2, p3, p4, p5, p6, p7, ROT, r_num) /* unrolled */ \
155 ROUND512(p0, p1, p2, p3, p4, p5, p6, p7, ROT, r_num)
159 /* inject the key schedule value */ \
160 X0 += ks[((R) + 1) % 9]; \
161 X1 += ks[((R) + 2) % 9]; \
162 X2 += ks[((R) + 3) % 9]; \
163 X3 += ks[((R) + 4) % 9]; \
164 X4 += ks[((R) + 5) % 9]; \
165 X5 += ks[((R) + 6) % 9] + ts[((R) + 1) % 3]; \
166 X6 += ks[((R) + 7) % 9] + ts[((R) + 2) % 3]; \
167 X7 += ks[((R) + 8) % 9] + (R) + 1; \
170 #else /* looping version */
171 #define R512(p0, p1, p2, p3, p4, p5, p6, p7, ROT, r_num) \
172 ROUND512(p0, p1, p2, p3, p4, p5, p6, p7, ROT, r_num) \
176 /* inject the key schedule value */ \
177 X0 += ks[r + (R) + 0]; \
178 X1 += ks[r + (R) + 1]; \
179 X2 += ks[r + (R) + 2]; \
180 X3 += ks[r + (R) + 3]; \
181 X4 += ks[r + (R) + 4]; \
182 X5 += ks[r + (R) + 5] + ts[r + (R) + 0]; \
183 X6 += ks[r + (R) + 6] + ts[r + (R) + 1]; \
184 X7 += ks[r + (R) + 7] + r + (R); \
185 /* rotate key schedule */ \
186 ks[r + (R) + 8] = ks[r + (R) - 1]; \
187 ts[r + (R) + 2] = ts[r + (R) - 1]; \
189 #endif /* end of looped code definitions */
190 #define R512_8_ROUNDS(R) /* do 8 full rounds */ \
192 R512(0, 1, 2, 3, 4, 5, 6, 7, R_512_0, 8 * (R) + 1); \
193 R512(2, 1, 4, 7, 6, 5, 0, 3, R_512_1, 8 * (R) + 2); \
194 R512(4, 1, 6, 3, 0, 5, 2, 7, R_512_2, 8 * (R) + 3); \
195 R512(6, 1, 0, 7, 2, 5, 4, 3, R_512_3, 8 * (R) + 4); \
197 R512(0, 1, 2, 3, 4, 5, 6, 7, R_512_4, 8 * (R) + 5); \
198 R512(2, 1, 4, 7, 6, 5, 0, 3, R_512_5, 8 * (R) + 6); \
199 R512(4, 1, 6, 3, 0, 5, 2, 7, R_512_6, 8 * (R) + 7); \
200 R512(6, 1, 0, 7, 2, 5, 4, 3, R_512_7, 8 * (R) + 8); \
201 I512(2 * (R) + 1); /* and key injection */ \
203 #define R512_UNROLL_R(NN) \
204 ((SKEIN_UNROLL_512 == 0 && \
205 SKEIN_512_ROUNDS_TOTAL / 8 > (NN)) || \
206 (SKEIN_UNROLL_512 > (NN)))
208 #if (SKEIN_UNROLL_512 > 14)
209 #error "need more unrolling in skein_512_process_block"
213 #if !(SKEIN_USE_ASM & 1024)
215 #define RCNT (SKEIN_1024_ROUNDS_TOTAL / 8)
216 #ifdef SKEIN_LOOP /* configure how much to unroll the loop */
217 #define SKEIN_UNROLL_1024 ((SKEIN_LOOP) % 10)
219 #define SKEIN_UNROLL_1024 (0)
222 #if (SKEIN_UNROLL_1024 != 0)
223 #if (RCNT % SKEIN_UNROLL_1024)
224 #error "Invalid SKEIN_UNROLL_1024" /* sanity check on unroll count */
227 #define ROUND1024(p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pA, pB, pC, pD, pE, \
231 X##p1 = rol64(X##p1, ROT##_0); \
234 X##p3 = rol64(X##p3, ROT##_1); \
237 X##p5 = rol64(X##p5, ROT##_2); \
240 X##p7 = rol64(X##p7, ROT##_3); \
243 X##p9 = rol64(X##p9, ROT##_4); \
246 X##pB = rol64(X##pB, ROT##_5); \
249 X##pD = rol64(X##pD, ROT##_6); \
252 X##pF = rol64(X##pF, ROT##_7); \
256 #if SKEIN_UNROLL_1024 == 0
257 #define R1024(p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pA, pB, pC, pD, pE, pF, \
259 ROUND1024(p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pA, pB, pC, pD, pE, \
264 /* inject the key schedule value */ \
265 X00 += ks[((R) + 1) % 17]; \
266 X01 += ks[((R) + 2) % 17]; \
267 X02 += ks[((R) + 3) % 17]; \
268 X03 += ks[((R) + 4) % 17]; \
269 X04 += ks[((R) + 5) % 17]; \
270 X05 += ks[((R) + 6) % 17]; \
271 X06 += ks[((R) + 7) % 17]; \
272 X07 += ks[((R) + 8) % 17]; \
273 X08 += ks[((R) + 9) % 17]; \
274 X09 += ks[((R) + 10) % 17]; \
275 X10 += ks[((R) + 11) % 17]; \
276 X11 += ks[((R) + 12) % 17]; \
277 X12 += ks[((R) + 13) % 17]; \
278 X13 += ks[((R) + 14) % 17] + ts[((R) + 1) % 3]; \
279 X14 += ks[((R) + 15) % 17] + ts[((R) + 2) % 3]; \
280 X15 += ks[((R) + 16) % 17] + (R) + 1; \
282 #else /* looping version */
283 #define R1024(p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pA, pB, pC, pD, pE, pF, \
285 ROUND1024(p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pA, pB, pC, pD, pE, \
290 /* inject the key schedule value */ \
291 X00 += ks[r + (R) + 0]; \
292 X01 += ks[r + (R) + 1]; \
293 X02 += ks[r + (R) + 2]; \
294 X03 += ks[r + (R) + 3]; \
295 X04 += ks[r + (R) + 4]; \
296 X05 += ks[r + (R) + 5]; \
297 X06 += ks[r + (R) + 6]; \
298 X07 += ks[r + (R) + 7]; \
299 X08 += ks[r + (R) + 8]; \
300 X09 += ks[r + (R) + 9]; \
301 X10 += ks[r + (R) + 10]; \
302 X11 += ks[r + (R) + 11]; \
303 X12 += ks[r + (R) + 12]; \
304 X13 += ks[r + (R) + 13] + ts[r + (R) + 0]; \
305 X14 += ks[r + (R) + 14] + ts[r + (R) + 1]; \
306 X15 += ks[r + (R) + 15] + r + (R); \
307 /* rotate key schedule */ \
308 ks[r + (R) + 16] = ks[r + (R) - 1]; \
309 ts[r + (R) + 2] = ts[r + (R) - 1]; \
313 #define R1024_8_ROUNDS(R) \
315 R1024(00, 01, 02, 03, 04, 05, 06, 07, 08, 09, 10, 11, 12, \
316 13, 14, 15, R1024_0, 8 * (R) + 1); \
317 R1024(00, 09, 02, 13, 06, 11, 04, 15, 10, 07, 12, 03, 14, \
318 05, 08, 01, R1024_1, 8 * (R) + 2); \
319 R1024(00, 07, 02, 05, 04, 03, 06, 01, 12, 15, 14, 13, 08, \
320 11, 10, 09, R1024_2, 8 * (R) + 3); \
321 R1024(00, 15, 02, 11, 06, 13, 04, 09, 14, 01, 08, 05, 10, \
322 03, 12, 07, R1024_3, 8 * (R) + 4); \
324 R1024(00, 01, 02, 03, 04, 05, 06, 07, 08, 09, 10, 11, 12, \
325 13, 14, 15, R1024_4, 8 * (R) + 5); \
326 R1024(00, 09, 02, 13, 06, 11, 04, 15, 10, 07, 12, 03, 14, \
327 05, 08, 01, R1024_5, 8 * (R) + 6); \
328 R1024(00, 07, 02, 05, 04, 03, 06, 01, 12, 15, 14, 13, 08, \
329 11, 10, 09, R1024_6, 8 * (R) + 7); \
330 R1024(00, 15, 02, 11, 06, 13, 04, 09, 14, 01, 08, 05, 10, \
331 03, 12, 07, R1024_7, 8 * (R) + 8); \
332 I1024(2 * (R) + 1); \
335 #define R1024_UNROLL_R(NN) \
336 ((SKEIN_UNROLL_1024 == 0 && \
337 SKEIN_1024_ROUNDS_TOTAL / 8 > (NN)) || \
338 (SKEIN_UNROLL_1024 > (NN)))
340 #if (SKEIN_UNROLL_1024 > 14)
341 #error "need more unrolling in Skein_1024_Process_Block"
345 /***************************** SKEIN_256 ******************************/
346 #if !(SKEIN_USE_ASM & 256)
347 void skein_256_process_block(struct skein_256_ctx *ctx, const u8 *blk_ptr,
348 size_t blk_cnt, size_t byte_cnt_add)
351 WCNT = SKEIN_256_STATE_WORDS
355 /* key schedule: chaining vars + tweak + "rot"*/
356 u64 kw[WCNT + 4 + (RCNT * 2)];
358 /* key schedule words : chaining vars + tweak */
361 u64 X0, X1, X2, X3; /* local copy of context vars, for speed */
362 u64 w[WCNT]; /* local copy of input block */
364 const u64 *X_ptr[4]; /* use for debugging (help cc put Xn in regs) */
371 skein_assert(blk_cnt != 0); /* never call with blk_cnt == 0! */
372 ts[0] = ctx->h.tweak[0];
373 ts[1] = ctx->h.tweak[1];
376 * this implementation only supports 2**64 input bytes
377 * (no carry out here)
379 ts[0] += byte_cnt_add; /* update processed length */
381 /* precompute the key schedule for this block */
386 ks[4] = ks[0] ^ ks[1] ^ ks[2] ^ ks[3] ^ SKEIN_KS_PARITY;
388 ts[2] = ts[0] ^ ts[1];
390 /* get input block in little-endian format */
391 skein_get64_lsb_first(w, blk_ptr, WCNT);
392 debug_save_tweak(ctx);
394 /* do the first full key injection */
396 X1 = w[1] + ks[1] + ts[0];
397 X2 = w[2] + ks[2] + ts[1];
400 blk_ptr += SKEIN_256_BLOCK_BYTES;
404 r < (SKEIN_UNROLL_256 ? 2 * RCNT : 2);
405 r += (SKEIN_UNROLL_256 ? 2 * SKEIN_UNROLL_256 : 1)) {
434 #if R256_UNROLL_R(10)
437 #if R256_UNROLL_R(11)
440 #if R256_UNROLL_R(12)
443 #if R256_UNROLL_R(13)
446 #if R256_UNROLL_R(14)
450 /* do the final "feedforward" xor, update context chaining */
451 ctx->x[0] = X0 ^ w[0];
452 ctx->x[1] = X1 ^ w[1];
453 ctx->x[2] = X2 ^ w[2];
454 ctx->x[3] = X3 ^ w[3];
456 ts[1] &= ~SKEIN_T1_FLAG_FIRST;
458 ctx->h.tweak[0] = ts[0];
459 ctx->h.tweak[1] = ts[1];
462 #if defined(SKEIN_CODE_SIZE) || defined(SKEIN_PERF)
463 size_t skein_256_process_block_code_size(void)
465 return ((u8 *)skein_256_process_block_code_size) -
466 ((u8 *)skein_256_process_block);
469 unsigned int skein_256_unroll_cnt(void)
471 return SKEIN_UNROLL_256;
476 /***************************** SKEIN_512 ******************************/
477 #if !(SKEIN_USE_ASM & 512)
478 void skein_512_process_block(struct skein_512_ctx *ctx, const u8 *blk_ptr,
479 size_t blk_cnt, size_t byte_cnt_add)
482 WCNT = SKEIN_512_STATE_WORDS
486 /* key sched: chaining vars + tweak + "rot"*/
487 u64 kw[WCNT + 4 + RCNT * 2];
489 /* key schedule words : chaining vars + tweak */
492 u64 X0, X1, X2, X3, X4, X5, X6, X7; /* local copies, for speed */
493 u64 w[WCNT]; /* local copy of input block */
495 const u64 *X_ptr[8]; /* use for debugging (help cc put Xn in regs) */
507 skein_assert(blk_cnt != 0); /* never call with blk_cnt == 0! */
508 ts[0] = ctx->h.tweak[0];
509 ts[1] = ctx->h.tweak[1];
512 * this implementation only supports 2**64 input bytes
513 * (no carry out here)
515 ts[0] += byte_cnt_add; /* update processed length */
517 /* precompute the key schedule for this block */
526 ks[8] = ks[0] ^ ks[1] ^ ks[2] ^ ks[3] ^
527 ks[4] ^ ks[5] ^ ks[6] ^ ks[7] ^ SKEIN_KS_PARITY;
529 ts[2] = ts[0] ^ ts[1];
531 /* get input block in little-endian format */
532 skein_get64_lsb_first(w, blk_ptr, WCNT);
533 debug_save_tweak(ctx);
535 /* do the first full key injection */
541 X5 = w[5] + ks[5] + ts[0];
542 X6 = w[6] + ks[6] + ts[1];
545 blk_ptr += SKEIN_512_BLOCK_BYTES;
549 r < (SKEIN_UNROLL_512 ? 2 * RCNT : 2);
550 r += (SKEIN_UNROLL_512 ? 2 * SKEIN_UNROLL_512 : 1)) {
580 #if R512_UNROLL_R(10)
583 #if R512_UNROLL_R(11)
586 #if R512_UNROLL_R(12)
589 #if R512_UNROLL_R(13)
592 #if R512_UNROLL_R(14)
597 /* do the final "feedforward" xor, update context chaining */
598 ctx->x[0] = X0 ^ w[0];
599 ctx->x[1] = X1 ^ w[1];
600 ctx->x[2] = X2 ^ w[2];
601 ctx->x[3] = X3 ^ w[3];
602 ctx->x[4] = X4 ^ w[4];
603 ctx->x[5] = X5 ^ w[5];
604 ctx->x[6] = X6 ^ w[6];
605 ctx->x[7] = X7 ^ w[7];
607 ts[1] &= ~SKEIN_T1_FLAG_FIRST;
609 ctx->h.tweak[0] = ts[0];
610 ctx->h.tweak[1] = ts[1];
613 #if defined(SKEIN_CODE_SIZE) || defined(SKEIN_PERF)
614 size_t skein_512_process_block_code_size(void)
616 return ((u8 *)skein_512_process_block_code_size) -
617 ((u8 *)skein_512_process_block);
620 unsigned int skein_512_unroll_cnt(void)
622 return SKEIN_UNROLL_512;
627 /***************************** SKEIN_1024 ******************************/
628 #if !(SKEIN_USE_ASM & 1024)
629 void skein_1024_process_block(struct skein_1024_ctx *ctx, const u8 *blk_ptr,
630 size_t blk_cnt, size_t byte_cnt_add)
631 { /* do it in C, always looping (unrolled is bigger AND slower!) */
633 WCNT = SKEIN_1024_STATE_WORDS
636 #if (SKEIN_UNROLL_1024 != 0)
637 /* key sched: chaining vars + tweak + "rot" */
638 u64 kw[WCNT + 4 + (RCNT * 2)];
640 /* key schedule words : chaining vars + tweak */
644 /* local copy of vars, for speed */
645 u64 X00, X01, X02, X03, X04, X05, X06, X07,
646 X08, X09, X10, X11, X12, X13, X14, X15;
647 u64 w[WCNT]; /* local copy of input block */
649 skein_assert(blk_cnt != 0); /* never call with blk_cnt == 0! */
650 ts[0] = ctx->h.tweak[0];
651 ts[1] = ctx->h.tweak[1];
654 * this implementation only supports 2**64 input bytes
655 * (no carry out here)
657 ts[0] += byte_cnt_add; /* update processed length */
659 /* precompute the key schedule for this block */
676 ks[16] = ks[0] ^ ks[1] ^ ks[2] ^ ks[3] ^
677 ks[4] ^ ks[5] ^ ks[6] ^ ks[7] ^
678 ks[8] ^ ks[9] ^ ks[10] ^ ks[11] ^
679 ks[12] ^ ks[13] ^ ks[14] ^ ks[15] ^ SKEIN_KS_PARITY;
681 ts[2] = ts[0] ^ ts[1];
683 /* get input block in little-endian format */
684 skein_get64_lsb_first(w, blk_ptr, WCNT);
685 debug_save_tweak(ctx);
687 /* do the first full key injection */
698 X10 = w[10] + ks[10];
699 X11 = w[11] + ks[11];
700 X12 = w[12] + ks[12];
701 X13 = w[13] + ks[13] + ts[0];
702 X14 = w[14] + ks[14] + ts[1];
703 X15 = w[15] + ks[15];
706 r < (SKEIN_UNROLL_1024 ? 2 * RCNT : 2);
707 r += (SKEIN_UNROLL_1024 ? 2 * SKEIN_UNROLL_1024 : 1)) {
709 #if R1024_UNROLL_R(1)
712 #if R1024_UNROLL_R(2)
715 #if R1024_UNROLL_R(3)
718 #if R1024_UNROLL_R(4)
721 #if R1024_UNROLL_R(5)
724 #if R1024_UNROLL_R(6)
727 #if R1024_UNROLL_R(7)
730 #if R1024_UNROLL_R(8)
733 #if R1024_UNROLL_R(9)
736 #if R1024_UNROLL_R(10)
739 #if R1024_UNROLL_R(11)
742 #if R1024_UNROLL_R(12)
745 #if R1024_UNROLL_R(13)
748 #if R1024_UNROLL_R(14)
752 /* do the final "feedforward" xor, update context chaining */
754 ctx->x[0] = X00 ^ w[0];
755 ctx->x[1] = X01 ^ w[1];
756 ctx->x[2] = X02 ^ w[2];
757 ctx->x[3] = X03 ^ w[3];
758 ctx->x[4] = X04 ^ w[4];
759 ctx->x[5] = X05 ^ w[5];
760 ctx->x[6] = X06 ^ w[6];
761 ctx->x[7] = X07 ^ w[7];
762 ctx->x[8] = X08 ^ w[8];
763 ctx->x[9] = X09 ^ w[9];
764 ctx->x[10] = X10 ^ w[10];
765 ctx->x[11] = X11 ^ w[11];
766 ctx->x[12] = X12 ^ w[12];
767 ctx->x[13] = X13 ^ w[13];
768 ctx->x[14] = X14 ^ w[14];
769 ctx->x[15] = X15 ^ w[15];
771 ts[1] &= ~SKEIN_T1_FLAG_FIRST;
772 blk_ptr += SKEIN_1024_BLOCK_BYTES;
774 ctx->h.tweak[0] = ts[0];
775 ctx->h.tweak[1] = ts[1];
778 #if defined(SKEIN_CODE_SIZE) || defined(SKEIN_PERF)
779 size_t skein_1024_process_block_code_size(void)
781 return ((u8 *)skein_1024_process_block_code_size) -
782 ((u8 *)skein_1024_process_block);
785 unsigned int skein_1024_unroll_cnt(void)
787 return SKEIN_UNROLL_1024;