x86, relocs: Remove an unused variable
[zen-stable.git] / lib / bch.c
blobbc89dfe4d1b38c1152f7f1940634a37ec988c8c3
1 /*
2 * Generic binary BCH encoding/decoding library
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of the GNU General Public License version 2 as published by
6 * the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
11 * more details.
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, write to the Free Software Foundation, Inc., 51
15 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
17 * Copyright © 2011 Parrot S.A.
19 * Author: Ivan Djelic <ivan.djelic@parrot.com>
21 * Description:
23 * This library provides runtime configurable encoding/decoding of binary
24 * Bose-Chaudhuri-Hocquenghem (BCH) codes.
26 * Call init_bch to get a pointer to a newly allocated bch_control structure for
27 * the given m (Galois field order), t (error correction capability) and
28 * (optional) primitive polynomial parameters.
30 * Call encode_bch to compute and store ecc parity bytes to a given buffer.
31 * Call decode_bch to detect and locate errors in received data.
33 * On systems supporting hw BCH features, intermediate results may be provided
34 * to decode_bch in order to skip certain steps. See decode_bch() documentation
35 * for details.
37 * Option CONFIG_BCH_CONST_PARAMS can be used to force fixed values of
38 * parameters m and t; thus allowing extra compiler optimizations and providing
39 * better (up to 2x) encoding performance. Using this option makes sense when
40 * (m,t) are fixed and known in advance, e.g. when using BCH error correction
41 * on a particular NAND flash device.
43 * Algorithmic details:
45 * Encoding is performed by processing 32 input bits in parallel, using 4
46 * remainder lookup tables.
48 * The final stage of decoding involves the following internal steps:
49 * a. Syndrome computation
50 * b. Error locator polynomial computation using Berlekamp-Massey algorithm
51 * c. Error locator root finding (by far the most expensive step)
53 * In this implementation, step c is not performed using the usual Chien search.
54 * Instead, an alternative approach described in [1] is used. It consists in
55 * factoring the error locator polynomial using the Berlekamp Trace algorithm
56 * (BTA) down to a certain degree (4), after which ad hoc low-degree polynomial
57 * solving techniques [2] are used. The resulting algorithm, called BTZ, yields
58 * much better performance than Chien search for usual (m,t) values (typically
59 * m >= 13, t < 32, see [1]).
61 * [1] B. Biswas, V. Herbert. Efficient root finding of polynomials over fields
62 * of characteristic 2, in: Western European Workshop on Research in Cryptology
63 * - WEWoRC 2009, Graz, Austria, LNCS, Springer, July 2009, to appear.
64 * [2] [Zin96] V.A. Zinoviev. On the solution of equations of degree 10 over
65 * finite fields GF(2^q). In Rapport de recherche INRIA no 2829, 1996.
68 #include <linux/kernel.h>
69 #include <linux/errno.h>
70 #include <linux/init.h>
71 #include <linux/module.h>
72 #include <linux/slab.h>
73 #include <linux/bitops.h>
74 #include <asm/byteorder.h>
75 #include <linux/bch.h>
77 #if defined(CONFIG_BCH_CONST_PARAMS)
78 #define GF_M(_p) (CONFIG_BCH_CONST_M)
79 #define GF_T(_p) (CONFIG_BCH_CONST_T)
80 #define GF_N(_p) ((1 << (CONFIG_BCH_CONST_M))-1)
81 #else
82 #define GF_M(_p) ((_p)->m)
83 #define GF_T(_p) ((_p)->t)
84 #define GF_N(_p) ((_p)->n)
85 #endif
87 #define BCH_ECC_WORDS(_p) DIV_ROUND_UP(GF_M(_p)*GF_T(_p), 32)
88 #define BCH_ECC_BYTES(_p) DIV_ROUND_UP(GF_M(_p)*GF_T(_p), 8)
90 #ifndef dbg
91 #define dbg(_fmt, args...) do {} while (0)
92 #endif
95 * represent a polynomial over GF(2^m)
97 struct gf_poly {
98 unsigned int deg; /* polynomial degree */
99 unsigned int c[0]; /* polynomial terms */
102 /* given its degree, compute a polynomial size in bytes */
103 #define GF_POLY_SZ(_d) (sizeof(struct gf_poly)+((_d)+1)*sizeof(unsigned int))
105 /* polynomial of degree 1 */
106 struct gf_poly_deg1 {
107 struct gf_poly poly;
108 unsigned int c[2];
112 * same as encode_bch(), but process input data one byte at a time
114 static void encode_bch_unaligned(struct bch_control *bch,
115 const unsigned char *data, unsigned int len,
116 uint32_t *ecc)
118 int i;
119 const uint32_t *p;
120 const int l = BCH_ECC_WORDS(bch)-1;
122 while (len--) {
123 p = bch->mod8_tab + (l+1)*(((ecc[0] >> 24)^(*data++)) & 0xff);
125 for (i = 0; i < l; i++)
126 ecc[i] = ((ecc[i] << 8)|(ecc[i+1] >> 24))^(*p++);
128 ecc[l] = (ecc[l] << 8)^(*p);
133 * convert ecc bytes to aligned, zero-padded 32-bit ecc words
135 static void load_ecc8(struct bch_control *bch, uint32_t *dst,
136 const uint8_t *src)
138 uint8_t pad[4] = {0, 0, 0, 0};
139 unsigned int i, nwords = BCH_ECC_WORDS(bch)-1;
141 for (i = 0; i < nwords; i++, src += 4)
142 dst[i] = (src[0] << 24)|(src[1] << 16)|(src[2] << 8)|src[3];
144 memcpy(pad, src, BCH_ECC_BYTES(bch)-4*nwords);
145 dst[nwords] = (pad[0] << 24)|(pad[1] << 16)|(pad[2] << 8)|pad[3];
149 * convert 32-bit ecc words to ecc bytes
151 static void store_ecc8(struct bch_control *bch, uint8_t *dst,
152 const uint32_t *src)
154 uint8_t pad[4];
155 unsigned int i, nwords = BCH_ECC_WORDS(bch)-1;
157 for (i = 0; i < nwords; i++) {
158 *dst++ = (src[i] >> 24);
159 *dst++ = (src[i] >> 16) & 0xff;
160 *dst++ = (src[i] >> 8) & 0xff;
161 *dst++ = (src[i] >> 0) & 0xff;
163 pad[0] = (src[nwords] >> 24);
164 pad[1] = (src[nwords] >> 16) & 0xff;
165 pad[2] = (src[nwords] >> 8) & 0xff;
166 pad[3] = (src[nwords] >> 0) & 0xff;
167 memcpy(dst, pad, BCH_ECC_BYTES(bch)-4*nwords);
171 * encode_bch - calculate BCH ecc parity of data
172 * @bch: BCH control structure
173 * @data: data to encode
174 * @len: data length in bytes
175 * @ecc: ecc parity data, must be initialized by caller
177 * The @ecc parity array is used both as input and output parameter, in order to
178 * allow incremental computations. It should be of the size indicated by member
179 * @ecc_bytes of @bch, and should be initialized to 0 before the first call.
181 * The exact number of computed ecc parity bits is given by member @ecc_bits of
182 * @bch; it may be less than m*t for large values of t.
184 void encode_bch(struct bch_control *bch, const uint8_t *data,
185 unsigned int len, uint8_t *ecc)
187 const unsigned int l = BCH_ECC_WORDS(bch)-1;
188 unsigned int i, mlen;
189 unsigned long m;
190 uint32_t w, r[l+1];
191 const uint32_t * const tab0 = bch->mod8_tab;
192 const uint32_t * const tab1 = tab0 + 256*(l+1);
193 const uint32_t * const tab2 = tab1 + 256*(l+1);
194 const uint32_t * const tab3 = tab2 + 256*(l+1);
195 const uint32_t *pdata, *p0, *p1, *p2, *p3;
197 if (ecc) {
198 /* load ecc parity bytes into internal 32-bit buffer */
199 load_ecc8(bch, bch->ecc_buf, ecc);
200 } else {
201 memset(bch->ecc_buf, 0, sizeof(r));
204 /* process first unaligned data bytes */
205 m = ((unsigned long)data) & 3;
206 if (m) {
207 mlen = (len < (4-m)) ? len : 4-m;
208 encode_bch_unaligned(bch, data, mlen, bch->ecc_buf);
209 data += mlen;
210 len -= mlen;
213 /* process 32-bit aligned data words */
214 pdata = (uint32_t *)data;
215 mlen = len/4;
216 data += 4*mlen;
217 len -= 4*mlen;
218 memcpy(r, bch->ecc_buf, sizeof(r));
221 * split each 32-bit word into 4 polynomials of weight 8 as follows:
223 * 31 ...24 23 ...16 15 ... 8 7 ... 0
224 * xxxxxxxx yyyyyyyy zzzzzzzz tttttttt
225 * tttttttt mod g = r0 (precomputed)
226 * zzzzzzzz 00000000 mod g = r1 (precomputed)
227 * yyyyyyyy 00000000 00000000 mod g = r2 (precomputed)
228 * xxxxxxxx 00000000 00000000 00000000 mod g = r3 (precomputed)
229 * xxxxxxxx yyyyyyyy zzzzzzzz tttttttt mod g = r0^r1^r2^r3
231 while (mlen--) {
232 /* input data is read in big-endian format */
233 w = r[0]^cpu_to_be32(*pdata++);
234 p0 = tab0 + (l+1)*((w >> 0) & 0xff);
235 p1 = tab1 + (l+1)*((w >> 8) & 0xff);
236 p2 = tab2 + (l+1)*((w >> 16) & 0xff);
237 p3 = tab3 + (l+1)*((w >> 24) & 0xff);
239 for (i = 0; i < l; i++)
240 r[i] = r[i+1]^p0[i]^p1[i]^p2[i]^p3[i];
242 r[l] = p0[l]^p1[l]^p2[l]^p3[l];
244 memcpy(bch->ecc_buf, r, sizeof(r));
246 /* process last unaligned bytes */
247 if (len)
248 encode_bch_unaligned(bch, data, len, bch->ecc_buf);
250 /* store ecc parity bytes into original parity buffer */
251 if (ecc)
252 store_ecc8(bch, ecc, bch->ecc_buf);
254 EXPORT_SYMBOL_GPL(encode_bch);
256 static inline int modulo(struct bch_control *bch, unsigned int v)
258 const unsigned int n = GF_N(bch);
259 while (v >= n) {
260 v -= n;
261 v = (v & n) + (v >> GF_M(bch));
263 return v;
267 * shorter and faster modulo function, only works when v < 2N.
269 static inline int mod_s(struct bch_control *bch, unsigned int v)
271 const unsigned int n = GF_N(bch);
272 return (v < n) ? v : v-n;
275 static inline int deg(unsigned int poly)
277 /* polynomial degree is the most-significant bit index */
278 return fls(poly)-1;
281 static inline int parity(unsigned int x)
284 * public domain code snippet, lifted from
285 * http://www-graphics.stanford.edu/~seander/bithacks.html
287 x ^= x >> 1;
288 x ^= x >> 2;
289 x = (x & 0x11111111U) * 0x11111111U;
290 return (x >> 28) & 1;
293 /* Galois field basic operations: multiply, divide, inverse, etc. */
295 static inline unsigned int gf_mul(struct bch_control *bch, unsigned int a,
296 unsigned int b)
298 return (a && b) ? bch->a_pow_tab[mod_s(bch, bch->a_log_tab[a]+
299 bch->a_log_tab[b])] : 0;
302 static inline unsigned int gf_sqr(struct bch_control *bch, unsigned int a)
304 return a ? bch->a_pow_tab[mod_s(bch, 2*bch->a_log_tab[a])] : 0;
307 static inline unsigned int gf_div(struct bch_control *bch, unsigned int a,
308 unsigned int b)
310 return a ? bch->a_pow_tab[mod_s(bch, bch->a_log_tab[a]+
311 GF_N(bch)-bch->a_log_tab[b])] : 0;
314 static inline unsigned int gf_inv(struct bch_control *bch, unsigned int a)
316 return bch->a_pow_tab[GF_N(bch)-bch->a_log_tab[a]];
319 static inline unsigned int a_pow(struct bch_control *bch, int i)
321 return bch->a_pow_tab[modulo(bch, i)];
324 static inline int a_log(struct bch_control *bch, unsigned int x)
326 return bch->a_log_tab[x];
329 static inline int a_ilog(struct bch_control *bch, unsigned int x)
331 return mod_s(bch, GF_N(bch)-bch->a_log_tab[x]);
335 * compute 2t syndromes of ecc polynomial, i.e. ecc(a^j) for j=1..2t
337 static void compute_syndromes(struct bch_control *bch, uint32_t *ecc,
338 unsigned int *syn)
340 int i, j, s;
341 unsigned int m;
342 uint32_t poly;
343 const int t = GF_T(bch);
345 s = bch->ecc_bits;
347 /* make sure extra bits in last ecc word are cleared */
348 m = ((unsigned int)s) & 31;
349 if (m)
350 ecc[s/32] &= ~((1u << (32-m))-1);
351 memset(syn, 0, 2*t*sizeof(*syn));
353 /* compute v(a^j) for j=1 .. 2t-1 */
354 do {
355 poly = *ecc++;
356 s -= 32;
357 while (poly) {
358 i = deg(poly);
359 for (j = 0; j < 2*t; j += 2)
360 syn[j] ^= a_pow(bch, (j+1)*(i+s));
362 poly ^= (1 << i);
364 } while (s > 0);
366 /* v(a^(2j)) = v(a^j)^2 */
367 for (j = 0; j < t; j++)
368 syn[2*j+1] = gf_sqr(bch, syn[j]);
371 static void gf_poly_copy(struct gf_poly *dst, struct gf_poly *src)
373 memcpy(dst, src, GF_POLY_SZ(src->deg));
376 static int compute_error_locator_polynomial(struct bch_control *bch,
377 const unsigned int *syn)
379 const unsigned int t = GF_T(bch);
380 const unsigned int n = GF_N(bch);
381 unsigned int i, j, tmp, l, pd = 1, d = syn[0];
382 struct gf_poly *elp = bch->elp;
383 struct gf_poly *pelp = bch->poly_2t[0];
384 struct gf_poly *elp_copy = bch->poly_2t[1];
385 int k, pp = -1;
387 memset(pelp, 0, GF_POLY_SZ(2*t));
388 memset(elp, 0, GF_POLY_SZ(2*t));
390 pelp->deg = 0;
391 pelp->c[0] = 1;
392 elp->deg = 0;
393 elp->c[0] = 1;
395 /* use simplified binary Berlekamp-Massey algorithm */
396 for (i = 0; (i < t) && (elp->deg <= t); i++) {
397 if (d) {
398 k = 2*i-pp;
399 gf_poly_copy(elp_copy, elp);
400 /* e[i+1](X) = e[i](X)+di*dp^-1*X^2(i-p)*e[p](X) */
401 tmp = a_log(bch, d)+n-a_log(bch, pd);
402 for (j = 0; j <= pelp->deg; j++) {
403 if (pelp->c[j]) {
404 l = a_log(bch, pelp->c[j]);
405 elp->c[j+k] ^= a_pow(bch, tmp+l);
408 /* compute l[i+1] = max(l[i]->c[l[p]+2*(i-p]) */
409 tmp = pelp->deg+k;
410 if (tmp > elp->deg) {
411 elp->deg = tmp;
412 gf_poly_copy(pelp, elp_copy);
413 pd = d;
414 pp = 2*i;
417 /* di+1 = S(2i+3)+elp[i+1].1*S(2i+2)+...+elp[i+1].lS(2i+3-l) */
418 if (i < t-1) {
419 d = syn[2*i+2];
420 for (j = 1; j <= elp->deg; j++)
421 d ^= gf_mul(bch, elp->c[j], syn[2*i+2-j]);
424 dbg("elp=%s\n", gf_poly_str(elp));
425 return (elp->deg > t) ? -1 : (int)elp->deg;
429 * solve a m x m linear system in GF(2) with an expected number of solutions,
430 * and return the number of found solutions
432 static int solve_linear_system(struct bch_control *bch, unsigned int *rows,
433 unsigned int *sol, int nsol)
435 const int m = GF_M(bch);
436 unsigned int tmp, mask;
437 int rem, c, r, p, k, param[m];
439 k = 0;
440 mask = 1 << m;
442 /* Gaussian elimination */
443 for (c = 0; c < m; c++) {
444 rem = 0;
445 p = c-k;
446 /* find suitable row for elimination */
447 for (r = p; r < m; r++) {
448 if (rows[r] & mask) {
449 if (r != p) {
450 tmp = rows[r];
451 rows[r] = rows[p];
452 rows[p] = tmp;
454 rem = r+1;
455 break;
458 if (rem) {
459 /* perform elimination on remaining rows */
460 tmp = rows[p];
461 for (r = rem; r < m; r++) {
462 if (rows[r] & mask)
463 rows[r] ^= tmp;
465 } else {
466 /* elimination not needed, store defective row index */
467 param[k++] = c;
469 mask >>= 1;
471 /* rewrite system, inserting fake parameter rows */
472 if (k > 0) {
473 p = k;
474 for (r = m-1; r >= 0; r--) {
475 if ((r > m-1-k) && rows[r])
476 /* system has no solution */
477 return 0;
479 rows[r] = (p && (r == param[p-1])) ?
480 p--, 1u << (m-r) : rows[r-p];
484 if (nsol != (1 << k))
485 /* unexpected number of solutions */
486 return 0;
488 for (p = 0; p < nsol; p++) {
489 /* set parameters for p-th solution */
490 for (c = 0; c < k; c++)
491 rows[param[c]] = (rows[param[c]] & ~1)|((p >> c) & 1);
493 /* compute unique solution */
494 tmp = 0;
495 for (r = m-1; r >= 0; r--) {
496 mask = rows[r] & (tmp|1);
497 tmp |= parity(mask) << (m-r);
499 sol[p] = tmp >> 1;
501 return nsol;
505 * this function builds and solves a linear system for finding roots of a degree
506 * 4 affine monic polynomial X^4+aX^2+bX+c over GF(2^m).
508 static int find_affine4_roots(struct bch_control *bch, unsigned int a,
509 unsigned int b, unsigned int c,
510 unsigned int *roots)
512 int i, j, k;
513 const int m = GF_M(bch);
514 unsigned int mask = 0xff, t, rows[16] = {0,};
516 j = a_log(bch, b);
517 k = a_log(bch, a);
518 rows[0] = c;
520 /* buid linear system to solve X^4+aX^2+bX+c = 0 */
521 for (i = 0; i < m; i++) {
522 rows[i+1] = bch->a_pow_tab[4*i]^
523 (a ? bch->a_pow_tab[mod_s(bch, k)] : 0)^
524 (b ? bch->a_pow_tab[mod_s(bch, j)] : 0);
525 j++;
526 k += 2;
529 * transpose 16x16 matrix before passing it to linear solver
530 * warning: this code assumes m < 16
532 for (j = 8; j != 0; j >>= 1, mask ^= (mask << j)) {
533 for (k = 0; k < 16; k = (k+j+1) & ~j) {
534 t = ((rows[k] >> j)^rows[k+j]) & mask;
535 rows[k] ^= (t << j);
536 rows[k+j] ^= t;
539 return solve_linear_system(bch, rows, roots, 4);
543 * compute root r of a degree 1 polynomial over GF(2^m) (returned as log(1/r))
545 static int find_poly_deg1_roots(struct bch_control *bch, struct gf_poly *poly,
546 unsigned int *roots)
548 int n = 0;
550 if (poly->c[0])
551 /* poly[X] = bX+c with c!=0, root=c/b */
552 roots[n++] = mod_s(bch, GF_N(bch)-bch->a_log_tab[poly->c[0]]+
553 bch->a_log_tab[poly->c[1]]);
554 return n;
558 * compute roots of a degree 2 polynomial over GF(2^m)
560 static int find_poly_deg2_roots(struct bch_control *bch, struct gf_poly *poly,
561 unsigned int *roots)
563 int n = 0, i, l0, l1, l2;
564 unsigned int u, v, r;
566 if (poly->c[0] && poly->c[1]) {
568 l0 = bch->a_log_tab[poly->c[0]];
569 l1 = bch->a_log_tab[poly->c[1]];
570 l2 = bch->a_log_tab[poly->c[2]];
572 /* using z=a/bX, transform aX^2+bX+c into z^2+z+u (u=ac/b^2) */
573 u = a_pow(bch, l0+l2+2*(GF_N(bch)-l1));
575 * let u = sum(li.a^i) i=0..m-1; then compute r = sum(li.xi):
576 * r^2+r = sum(li.(xi^2+xi)) = sum(li.(a^i+Tr(a^i).a^k)) =
577 * u + sum(li.Tr(a^i).a^k) = u+a^k.Tr(sum(li.a^i)) = u+a^k.Tr(u)
578 * i.e. r and r+1 are roots iff Tr(u)=0
580 r = 0;
581 v = u;
582 while (v) {
583 i = deg(v);
584 r ^= bch->xi_tab[i];
585 v ^= (1 << i);
587 /* verify root */
588 if ((gf_sqr(bch, r)^r) == u) {
589 /* reverse z=a/bX transformation and compute log(1/r) */
590 roots[n++] = modulo(bch, 2*GF_N(bch)-l1-
591 bch->a_log_tab[r]+l2);
592 roots[n++] = modulo(bch, 2*GF_N(bch)-l1-
593 bch->a_log_tab[r^1]+l2);
596 return n;
600 * compute roots of a degree 3 polynomial over GF(2^m)
602 static int find_poly_deg3_roots(struct bch_control *bch, struct gf_poly *poly,
603 unsigned int *roots)
605 int i, n = 0;
606 unsigned int a, b, c, a2, b2, c2, e3, tmp[4];
608 if (poly->c[0]) {
609 /* transform polynomial into monic X^3 + a2X^2 + b2X + c2 */
610 e3 = poly->c[3];
611 c2 = gf_div(bch, poly->c[0], e3);
612 b2 = gf_div(bch, poly->c[1], e3);
613 a2 = gf_div(bch, poly->c[2], e3);
615 /* (X+a2)(X^3+a2X^2+b2X+c2) = X^4+aX^2+bX+c (affine) */
616 c = gf_mul(bch, a2, c2); /* c = a2c2 */
617 b = gf_mul(bch, a2, b2)^c2; /* b = a2b2 + c2 */
618 a = gf_sqr(bch, a2)^b2; /* a = a2^2 + b2 */
620 /* find the 4 roots of this affine polynomial */
621 if (find_affine4_roots(bch, a, b, c, tmp) == 4) {
622 /* remove a2 from final list of roots */
623 for (i = 0; i < 4; i++) {
624 if (tmp[i] != a2)
625 roots[n++] = a_ilog(bch, tmp[i]);
629 return n;
633 * compute roots of a degree 4 polynomial over GF(2^m)
635 static int find_poly_deg4_roots(struct bch_control *bch, struct gf_poly *poly,
636 unsigned int *roots)
638 int i, l, n = 0;
639 unsigned int a, b, c, d, e = 0, f, a2, b2, c2, e4;
641 if (poly->c[0] == 0)
642 return 0;
644 /* transform polynomial into monic X^4 + aX^3 + bX^2 + cX + d */
645 e4 = poly->c[4];
646 d = gf_div(bch, poly->c[0], e4);
647 c = gf_div(bch, poly->c[1], e4);
648 b = gf_div(bch, poly->c[2], e4);
649 a = gf_div(bch, poly->c[3], e4);
651 /* use Y=1/X transformation to get an affine polynomial */
652 if (a) {
653 /* first, eliminate cX by using z=X+e with ae^2+c=0 */
654 if (c) {
655 /* compute e such that e^2 = c/a */
656 f = gf_div(bch, c, a);
657 l = a_log(bch, f);
658 l += (l & 1) ? GF_N(bch) : 0;
659 e = a_pow(bch, l/2);
661 * use transformation z=X+e:
662 * z^4+e^4 + a(z^3+ez^2+e^2z+e^3) + b(z^2+e^2) +cz+ce+d
663 * z^4 + az^3 + (ae+b)z^2 + (ae^2+c)z+e^4+be^2+ae^3+ce+d
664 * z^4 + az^3 + (ae+b)z^2 + e^4+be^2+d
665 * z^4 + az^3 + b'z^2 + d'
667 d = a_pow(bch, 2*l)^gf_mul(bch, b, f)^d;
668 b = gf_mul(bch, a, e)^b;
670 /* now, use Y=1/X to get Y^4 + b/dY^2 + a/dY + 1/d */
671 if (d == 0)
672 /* assume all roots have multiplicity 1 */
673 return 0;
675 c2 = gf_inv(bch, d);
676 b2 = gf_div(bch, a, d);
677 a2 = gf_div(bch, b, d);
678 } else {
679 /* polynomial is already affine */
680 c2 = d;
681 b2 = c;
682 a2 = b;
684 /* find the 4 roots of this affine polynomial */
685 if (find_affine4_roots(bch, a2, b2, c2, roots) == 4) {
686 for (i = 0; i < 4; i++) {
687 /* post-process roots (reverse transformations) */
688 f = a ? gf_inv(bch, roots[i]) : roots[i];
689 roots[i] = a_ilog(bch, f^e);
691 n = 4;
693 return n;
697 * build monic, log-based representation of a polynomial
699 static void gf_poly_logrep(struct bch_control *bch,
700 const struct gf_poly *a, int *rep)
702 int i, d = a->deg, l = GF_N(bch)-a_log(bch, a->c[a->deg]);
704 /* represent 0 values with -1; warning, rep[d] is not set to 1 */
705 for (i = 0; i < d; i++)
706 rep[i] = a->c[i] ? mod_s(bch, a_log(bch, a->c[i])+l) : -1;
710 * compute polynomial Euclidean division remainder in GF(2^m)[X]
712 static void gf_poly_mod(struct bch_control *bch, struct gf_poly *a,
713 const struct gf_poly *b, int *rep)
715 int la, p, m;
716 unsigned int i, j, *c = a->c;
717 const unsigned int d = b->deg;
719 if (a->deg < d)
720 return;
722 /* reuse or compute log representation of denominator */
723 if (!rep) {
724 rep = bch->cache;
725 gf_poly_logrep(bch, b, rep);
728 for (j = a->deg; j >= d; j--) {
729 if (c[j]) {
730 la = a_log(bch, c[j]);
731 p = j-d;
732 for (i = 0; i < d; i++, p++) {
733 m = rep[i];
734 if (m >= 0)
735 c[p] ^= bch->a_pow_tab[mod_s(bch,
736 m+la)];
740 a->deg = d-1;
741 while (!c[a->deg] && a->deg)
742 a->deg--;
746 * compute polynomial Euclidean division quotient in GF(2^m)[X]
748 static void gf_poly_div(struct bch_control *bch, struct gf_poly *a,
749 const struct gf_poly *b, struct gf_poly *q)
751 if (a->deg >= b->deg) {
752 q->deg = a->deg-b->deg;
753 /* compute a mod b (modifies a) */
754 gf_poly_mod(bch, a, b, NULL);
755 /* quotient is stored in upper part of polynomial a */
756 memcpy(q->c, &a->c[b->deg], (1+q->deg)*sizeof(unsigned int));
757 } else {
758 q->deg = 0;
759 q->c[0] = 0;
764 * compute polynomial GCD (Greatest Common Divisor) in GF(2^m)[X]
766 static struct gf_poly *gf_poly_gcd(struct bch_control *bch, struct gf_poly *a,
767 struct gf_poly *b)
769 struct gf_poly *tmp;
771 dbg("gcd(%s,%s)=", gf_poly_str(a), gf_poly_str(b));
773 if (a->deg < b->deg) {
774 tmp = b;
775 b = a;
776 a = tmp;
779 while (b->deg > 0) {
780 gf_poly_mod(bch, a, b, NULL);
781 tmp = b;
782 b = a;
783 a = tmp;
786 dbg("%s\n", gf_poly_str(a));
788 return a;
792 * Given a polynomial f and an integer k, compute Tr(a^kX) mod f
793 * This is used in Berlekamp Trace algorithm for splitting polynomials
795 static void compute_trace_bk_mod(struct bch_control *bch, int k,
796 const struct gf_poly *f, struct gf_poly *z,
797 struct gf_poly *out)
799 const int m = GF_M(bch);
800 int i, j;
802 /* z contains z^2j mod f */
803 z->deg = 1;
804 z->c[0] = 0;
805 z->c[1] = bch->a_pow_tab[k];
807 out->deg = 0;
808 memset(out, 0, GF_POLY_SZ(f->deg));
810 /* compute f log representation only once */
811 gf_poly_logrep(bch, f, bch->cache);
813 for (i = 0; i < m; i++) {
814 /* add a^(k*2^i)(z^(2^i) mod f) and compute (z^(2^i) mod f)^2 */
815 for (j = z->deg; j >= 0; j--) {
816 out->c[j] ^= z->c[j];
817 z->c[2*j] = gf_sqr(bch, z->c[j]);
818 z->c[2*j+1] = 0;
820 if (z->deg > out->deg)
821 out->deg = z->deg;
823 if (i < m-1) {
824 z->deg *= 2;
825 /* z^(2(i+1)) mod f = (z^(2^i) mod f)^2 mod f */
826 gf_poly_mod(bch, z, f, bch->cache);
829 while (!out->c[out->deg] && out->deg)
830 out->deg--;
832 dbg("Tr(a^%d.X) mod f = %s\n", k, gf_poly_str(out));
836 * factor a polynomial using Berlekamp Trace algorithm (BTA)
838 static void factor_polynomial(struct bch_control *bch, int k, struct gf_poly *f,
839 struct gf_poly **g, struct gf_poly **h)
841 struct gf_poly *f2 = bch->poly_2t[0];
842 struct gf_poly *q = bch->poly_2t[1];
843 struct gf_poly *tk = bch->poly_2t[2];
844 struct gf_poly *z = bch->poly_2t[3];
845 struct gf_poly *gcd;
847 dbg("factoring %s...\n", gf_poly_str(f));
849 *g = f;
850 *h = NULL;
852 /* tk = Tr(a^k.X) mod f */
853 compute_trace_bk_mod(bch, k, f, z, tk);
855 if (tk->deg > 0) {
856 /* compute g = gcd(f, tk) (destructive operation) */
857 gf_poly_copy(f2, f);
858 gcd = gf_poly_gcd(bch, f2, tk);
859 if (gcd->deg < f->deg) {
860 /* compute h=f/gcd(f,tk); this will modify f and q */
861 gf_poly_div(bch, f, gcd, q);
862 /* store g and h in-place (clobbering f) */
863 *h = &((struct gf_poly_deg1 *)f)[gcd->deg].poly;
864 gf_poly_copy(*g, gcd);
865 gf_poly_copy(*h, q);
871 * find roots of a polynomial, using BTZ algorithm; see the beginning of this
872 * file for details
874 static int find_poly_roots(struct bch_control *bch, unsigned int k,
875 struct gf_poly *poly, unsigned int *roots)
877 int cnt;
878 struct gf_poly *f1, *f2;
880 switch (poly->deg) {
881 /* handle low degree polynomials with ad hoc techniques */
882 case 1:
883 cnt = find_poly_deg1_roots(bch, poly, roots);
884 break;
885 case 2:
886 cnt = find_poly_deg2_roots(bch, poly, roots);
887 break;
888 case 3:
889 cnt = find_poly_deg3_roots(bch, poly, roots);
890 break;
891 case 4:
892 cnt = find_poly_deg4_roots(bch, poly, roots);
893 break;
894 default:
895 /* factor polynomial using Berlekamp Trace Algorithm (BTA) */
896 cnt = 0;
897 if (poly->deg && (k <= GF_M(bch))) {
898 factor_polynomial(bch, k, poly, &f1, &f2);
899 if (f1)
900 cnt += find_poly_roots(bch, k+1, f1, roots);
901 if (f2)
902 cnt += find_poly_roots(bch, k+1, f2, roots+cnt);
904 break;
906 return cnt;
909 #if defined(USE_CHIEN_SEARCH)
911 * exhaustive root search (Chien) implementation - not used, included only for
912 * reference/comparison tests
914 static int chien_search(struct bch_control *bch, unsigned int len,
915 struct gf_poly *p, unsigned int *roots)
917 int m;
918 unsigned int i, j, syn, syn0, count = 0;
919 const unsigned int k = 8*len+bch->ecc_bits;
921 /* use a log-based representation of polynomial */
922 gf_poly_logrep(bch, p, bch->cache);
923 bch->cache[p->deg] = 0;
924 syn0 = gf_div(bch, p->c[0], p->c[p->deg]);
926 for (i = GF_N(bch)-k+1; i <= GF_N(bch); i++) {
927 /* compute elp(a^i) */
928 for (j = 1, syn = syn0; j <= p->deg; j++) {
929 m = bch->cache[j];
930 if (m >= 0)
931 syn ^= a_pow(bch, m+j*i);
933 if (syn == 0) {
934 roots[count++] = GF_N(bch)-i;
935 if (count == p->deg)
936 break;
939 return (count == p->deg) ? count : 0;
941 #define find_poly_roots(_p, _k, _elp, _loc) chien_search(_p, len, _elp, _loc)
942 #endif /* USE_CHIEN_SEARCH */
945 * decode_bch - decode received codeword and find bit error locations
946 * @bch: BCH control structure
947 * @data: received data, ignored if @calc_ecc is provided
948 * @len: data length in bytes, must always be provided
949 * @recv_ecc: received ecc, if NULL then assume it was XORed in @calc_ecc
950 * @calc_ecc: calculated ecc, if NULL then calc_ecc is computed from @data
951 * @syn: hw computed syndrome data (if NULL, syndrome is calculated)
952 * @errloc: output array of error locations
954 * Returns:
955 * The number of errors found, or -EBADMSG if decoding failed, or -EINVAL if
956 * invalid parameters were provided
958 * Depending on the available hw BCH support and the need to compute @calc_ecc
959 * separately (using encode_bch()), this function should be called with one of
960 * the following parameter configurations -
962 * by providing @data and @recv_ecc only:
963 * decode_bch(@bch, @data, @len, @recv_ecc, NULL, NULL, @errloc)
965 * by providing @recv_ecc and @calc_ecc:
966 * decode_bch(@bch, NULL, @len, @recv_ecc, @calc_ecc, NULL, @errloc)
968 * by providing ecc = recv_ecc XOR calc_ecc:
969 * decode_bch(@bch, NULL, @len, NULL, ecc, NULL, @errloc)
971 * by providing syndrome results @syn:
972 * decode_bch(@bch, NULL, @len, NULL, NULL, @syn, @errloc)
974 * Once decode_bch() has successfully returned with a positive value, error
975 * locations returned in array @errloc should be interpreted as follows -
977 * if (errloc[n] >= 8*len), then n-th error is located in ecc (no need for
978 * data correction)
980 * if (errloc[n] < 8*len), then n-th error is located in data and can be
981 * corrected with statement data[errloc[n]/8] ^= 1 << (errloc[n] % 8);
983 * Note that this function does not perform any data correction by itself, it
984 * merely indicates error locations.
986 int decode_bch(struct bch_control *bch, const uint8_t *data, unsigned int len,
987 const uint8_t *recv_ecc, const uint8_t *calc_ecc,
988 const unsigned int *syn, unsigned int *errloc)
990 const unsigned int ecc_words = BCH_ECC_WORDS(bch);
991 unsigned int nbits;
992 int i, err, nroots;
993 uint32_t sum;
995 /* sanity check: make sure data length can be handled */
996 if (8*len > (bch->n-bch->ecc_bits))
997 return -EINVAL;
999 /* if caller does not provide syndromes, compute them */
1000 if (!syn) {
1001 if (!calc_ecc) {
1002 /* compute received data ecc into an internal buffer */
1003 if (!data || !recv_ecc)
1004 return -EINVAL;
1005 encode_bch(bch, data, len, NULL);
1006 } else {
1007 /* load provided calculated ecc */
1008 load_ecc8(bch, bch->ecc_buf, calc_ecc);
1010 /* load received ecc or assume it was XORed in calc_ecc */
1011 if (recv_ecc) {
1012 load_ecc8(bch, bch->ecc_buf2, recv_ecc);
1013 /* XOR received and calculated ecc */
1014 for (i = 0, sum = 0; i < (int)ecc_words; i++) {
1015 bch->ecc_buf[i] ^= bch->ecc_buf2[i];
1016 sum |= bch->ecc_buf[i];
1018 if (!sum)
1019 /* no error found */
1020 return 0;
1022 compute_syndromes(bch, bch->ecc_buf, bch->syn);
1023 syn = bch->syn;
1026 err = compute_error_locator_polynomial(bch, syn);
1027 if (err > 0) {
1028 nroots = find_poly_roots(bch, 1, bch->elp, errloc);
1029 if (err != nroots)
1030 err = -1;
1032 if (err > 0) {
1033 /* post-process raw error locations for easier correction */
1034 nbits = (len*8)+bch->ecc_bits;
1035 for (i = 0; i < err; i++) {
1036 if (errloc[i] >= nbits) {
1037 err = -1;
1038 break;
1040 errloc[i] = nbits-1-errloc[i];
1041 errloc[i] = (errloc[i] & ~7)|(7-(errloc[i] & 7));
1044 return (err >= 0) ? err : -EBADMSG;
1046 EXPORT_SYMBOL_GPL(decode_bch);
1049 * generate Galois field lookup tables
1051 static int build_gf_tables(struct bch_control *bch, unsigned int poly)
1053 unsigned int i, x = 1;
1054 const unsigned int k = 1 << deg(poly);
1056 /* primitive polynomial must be of degree m */
1057 if (k != (1u << GF_M(bch)))
1058 return -1;
1060 for (i = 0; i < GF_N(bch); i++) {
1061 bch->a_pow_tab[i] = x;
1062 bch->a_log_tab[x] = i;
1063 if (i && (x == 1))
1064 /* polynomial is not primitive (a^i=1 with 0<i<2^m-1) */
1065 return -1;
1066 x <<= 1;
1067 if (x & k)
1068 x ^= poly;
1070 bch->a_pow_tab[GF_N(bch)] = 1;
1071 bch->a_log_tab[0] = 0;
1073 return 0;
1077 * compute generator polynomial remainder tables for fast encoding
1079 static void build_mod8_tables(struct bch_control *bch, const uint32_t *g)
1081 int i, j, b, d;
1082 uint32_t data, hi, lo, *tab;
1083 const int l = BCH_ECC_WORDS(bch);
1084 const int plen = DIV_ROUND_UP(bch->ecc_bits+1, 32);
1085 const int ecclen = DIV_ROUND_UP(bch->ecc_bits, 32);
1087 memset(bch->mod8_tab, 0, 4*256*l*sizeof(*bch->mod8_tab));
1089 for (i = 0; i < 256; i++) {
1090 /* p(X)=i is a small polynomial of weight <= 8 */
1091 for (b = 0; b < 4; b++) {
1092 /* we want to compute (p(X).X^(8*b+deg(g))) mod g(X) */
1093 tab = bch->mod8_tab + (b*256+i)*l;
1094 data = i << (8*b);
1095 while (data) {
1096 d = deg(data);
1097 /* subtract X^d.g(X) from p(X).X^(8*b+deg(g)) */
1098 data ^= g[0] >> (31-d);
1099 for (j = 0; j < ecclen; j++) {
1100 hi = (d < 31) ? g[j] << (d+1) : 0;
1101 lo = (j+1 < plen) ?
1102 g[j+1] >> (31-d) : 0;
1103 tab[j] ^= hi|lo;
1111 * build a base for factoring degree 2 polynomials
1113 static int build_deg2_base(struct bch_control *bch)
1115 const int m = GF_M(bch);
1116 int i, j, r;
1117 unsigned int sum, x, y, remaining, ak = 0, xi[m];
1119 /* find k s.t. Tr(a^k) = 1 and 0 <= k < m */
1120 for (i = 0; i < m; i++) {
1121 for (j = 0, sum = 0; j < m; j++)
1122 sum ^= a_pow(bch, i*(1 << j));
1124 if (sum) {
1125 ak = bch->a_pow_tab[i];
1126 break;
1129 /* find xi, i=0..m-1 such that xi^2+xi = a^i+Tr(a^i).a^k */
1130 remaining = m;
1131 memset(xi, 0, sizeof(xi));
1133 for (x = 0; (x <= GF_N(bch)) && remaining; x++) {
1134 y = gf_sqr(bch, x)^x;
1135 for (i = 0; i < 2; i++) {
1136 r = a_log(bch, y);
1137 if (y && (r < m) && !xi[r]) {
1138 bch->xi_tab[r] = x;
1139 xi[r] = 1;
1140 remaining--;
1141 dbg("x%d = %x\n", r, x);
1142 break;
1144 y ^= ak;
1147 /* should not happen but check anyway */
1148 return remaining ? -1 : 0;
1151 static void *bch_alloc(size_t size, int *err)
1153 void *ptr;
1155 ptr = kmalloc(size, GFP_KERNEL);
1156 if (ptr == NULL)
1157 *err = 1;
1158 return ptr;
1162 * compute generator polynomial for given (m,t) parameters.
1164 static uint32_t *compute_generator_polynomial(struct bch_control *bch)
1166 const unsigned int m = GF_M(bch);
1167 const unsigned int t = GF_T(bch);
1168 int n, err = 0;
1169 unsigned int i, j, nbits, r, word, *roots;
1170 struct gf_poly *g;
1171 uint32_t *genpoly;
1173 g = bch_alloc(GF_POLY_SZ(m*t), &err);
1174 roots = bch_alloc((bch->n+1)*sizeof(*roots), &err);
1175 genpoly = bch_alloc(DIV_ROUND_UP(m*t+1, 32)*sizeof(*genpoly), &err);
1177 if (err) {
1178 kfree(genpoly);
1179 genpoly = NULL;
1180 goto finish;
1183 /* enumerate all roots of g(X) */
1184 memset(roots , 0, (bch->n+1)*sizeof(*roots));
1185 for (i = 0; i < t; i++) {
1186 for (j = 0, r = 2*i+1; j < m; j++) {
1187 roots[r] = 1;
1188 r = mod_s(bch, 2*r);
1191 /* build generator polynomial g(X) */
1192 g->deg = 0;
1193 g->c[0] = 1;
1194 for (i = 0; i < GF_N(bch); i++) {
1195 if (roots[i]) {
1196 /* multiply g(X) by (X+root) */
1197 r = bch->a_pow_tab[i];
1198 g->c[g->deg+1] = 1;
1199 for (j = g->deg; j > 0; j--)
1200 g->c[j] = gf_mul(bch, g->c[j], r)^g->c[j-1];
1202 g->c[0] = gf_mul(bch, g->c[0], r);
1203 g->deg++;
1206 /* store left-justified binary representation of g(X) */
1207 n = g->deg+1;
1208 i = 0;
1210 while (n > 0) {
1211 nbits = (n > 32) ? 32 : n;
1212 for (j = 0, word = 0; j < nbits; j++) {
1213 if (g->c[n-1-j])
1214 word |= 1u << (31-j);
1216 genpoly[i++] = word;
1217 n -= nbits;
1219 bch->ecc_bits = g->deg;
1221 finish:
1222 kfree(g);
1223 kfree(roots);
1225 return genpoly;
1229 * init_bch - initialize a BCH encoder/decoder
1230 * @m: Galois field order, should be in the range 5-15
1231 * @t: maximum error correction capability, in bits
1232 * @prim_poly: user-provided primitive polynomial (or 0 to use default)
1234 * Returns:
1235 * a newly allocated BCH control structure if successful, NULL otherwise
1237 * This initialization can take some time, as lookup tables are built for fast
1238 * encoding/decoding; make sure not to call this function from a time critical
1239 * path. Usually, init_bch() should be called on module/driver init and
1240 * free_bch() should be called to release memory on exit.
1242 * You may provide your own primitive polynomial of degree @m in argument
1243 * @prim_poly, or let init_bch() use its default polynomial.
1245 * Once init_bch() has successfully returned a pointer to a newly allocated
1246 * BCH control structure, ecc length in bytes is given by member @ecc_bytes of
1247 * the structure.
1249 struct bch_control *init_bch(int m, int t, unsigned int prim_poly)
1251 int err = 0;
1252 unsigned int i, words;
1253 uint32_t *genpoly;
1254 struct bch_control *bch = NULL;
1256 const int min_m = 5;
1257 const int max_m = 15;
1259 /* default primitive polynomials */
1260 static const unsigned int prim_poly_tab[] = {
1261 0x25, 0x43, 0x83, 0x11d, 0x211, 0x409, 0x805, 0x1053, 0x201b,
1262 0x402b, 0x8003,
1265 #if defined(CONFIG_BCH_CONST_PARAMS)
1266 if ((m != (CONFIG_BCH_CONST_M)) || (t != (CONFIG_BCH_CONST_T))) {
1267 printk(KERN_ERR "bch encoder/decoder was configured to support "
1268 "parameters m=%d, t=%d only!\n",
1269 CONFIG_BCH_CONST_M, CONFIG_BCH_CONST_T);
1270 goto fail;
1272 #endif
1273 if ((m < min_m) || (m > max_m))
1275 * values of m greater than 15 are not currently supported;
1276 * supporting m > 15 would require changing table base type
1277 * (uint16_t) and a small patch in matrix transposition
1279 goto fail;
1281 /* sanity checks */
1282 if ((t < 1) || (m*t >= ((1 << m)-1)))
1283 /* invalid t value */
1284 goto fail;
1286 /* select a primitive polynomial for generating GF(2^m) */
1287 if (prim_poly == 0)
1288 prim_poly = prim_poly_tab[m-min_m];
1290 bch = kzalloc(sizeof(*bch), GFP_KERNEL);
1291 if (bch == NULL)
1292 goto fail;
1294 bch->m = m;
1295 bch->t = t;
1296 bch->n = (1 << m)-1;
1297 words = DIV_ROUND_UP(m*t, 32);
1298 bch->ecc_bytes = DIV_ROUND_UP(m*t, 8);
1299 bch->a_pow_tab = bch_alloc((1+bch->n)*sizeof(*bch->a_pow_tab), &err);
1300 bch->a_log_tab = bch_alloc((1+bch->n)*sizeof(*bch->a_log_tab), &err);
1301 bch->mod8_tab = bch_alloc(words*1024*sizeof(*bch->mod8_tab), &err);
1302 bch->ecc_buf = bch_alloc(words*sizeof(*bch->ecc_buf), &err);
1303 bch->ecc_buf2 = bch_alloc(words*sizeof(*bch->ecc_buf2), &err);
1304 bch->xi_tab = bch_alloc(m*sizeof(*bch->xi_tab), &err);
1305 bch->syn = bch_alloc(2*t*sizeof(*bch->syn), &err);
1306 bch->cache = bch_alloc(2*t*sizeof(*bch->cache), &err);
1307 bch->elp = bch_alloc((t+1)*sizeof(struct gf_poly_deg1), &err);
1309 for (i = 0; i < ARRAY_SIZE(bch->poly_2t); i++)
1310 bch->poly_2t[i] = bch_alloc(GF_POLY_SZ(2*t), &err);
1312 if (err)
1313 goto fail;
1315 err = build_gf_tables(bch, prim_poly);
1316 if (err)
1317 goto fail;
1319 /* use generator polynomial for computing encoding tables */
1320 genpoly = compute_generator_polynomial(bch);
1321 if (genpoly == NULL)
1322 goto fail;
1324 build_mod8_tables(bch, genpoly);
1325 kfree(genpoly);
1327 err = build_deg2_base(bch);
1328 if (err)
1329 goto fail;
1331 return bch;
1333 fail:
1334 free_bch(bch);
1335 return NULL;
1337 EXPORT_SYMBOL_GPL(init_bch);
1340 * free_bch - free the BCH control structure
1341 * @bch: BCH control structure to release
1343 void free_bch(struct bch_control *bch)
1345 unsigned int i;
1347 if (bch) {
1348 kfree(bch->a_pow_tab);
1349 kfree(bch->a_log_tab);
1350 kfree(bch->mod8_tab);
1351 kfree(bch->ecc_buf);
1352 kfree(bch->ecc_buf2);
1353 kfree(bch->xi_tab);
1354 kfree(bch->syn);
1355 kfree(bch->cache);
1356 kfree(bch->elp);
1358 for (i = 0; i < ARRAY_SIZE(bch->poly_2t); i++)
1359 kfree(bch->poly_2t[i]);
1361 kfree(bch);
1364 EXPORT_SYMBOL_GPL(free_bch);
1366 MODULE_LICENSE("GPL");
1367 MODULE_AUTHOR("Ivan Djelic <ivan.djelic@parrot.com>");
1368 MODULE_DESCRIPTION("Binary BCH encoder/decoder");