xref: /openbmc/u-boot/lib/bch.c (revision ed09a554)
1 /*
2  * Generic binary BCH encoding/decoding library
3  *
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.
7  *
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.
12  *
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.
16  *
17  * Copyright © 2011 Parrot S.A.
18  *
19  * Author: Ivan Djelic <ivan.djelic@parrot.com>
20  *
21  * Description:
22  *
23  * This library provides runtime configurable encoding/decoding of binary
24  * Bose-Chaudhuri-Hocquenghem (BCH) codes.
25  *
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.
29  *
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.
32  *
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.
36  *
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.
42  *
43  * Algorithmic details:
44  *
45  * Encoding is performed by processing 32 input bits in parallel, using 4
46  * remainder lookup tables.
47  *
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)
52  *
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]).
60  *
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.
66  */
67 
68 #include <common.h>
69 #include <ubi_uboot.h>
70 
71 #include <linux/bitops.h>
72 #include <asm/byteorder.h>
73 #include <linux/bch.h>
74 
75 #if defined(CONFIG_BCH_CONST_PARAMS)
76 #define GF_M(_p)               (CONFIG_BCH_CONST_M)
77 #define GF_T(_p)               (CONFIG_BCH_CONST_T)
78 #define GF_N(_p)               ((1 << (CONFIG_BCH_CONST_M))-1)
79 #else
80 #define GF_M(_p)               ((_p)->m)
81 #define GF_T(_p)               ((_p)->t)
82 #define GF_N(_p)               ((_p)->n)
83 #endif
84 
85 #define BCH_ECC_WORDS(_p)      DIV_ROUND_UP(GF_M(_p)*GF_T(_p), 32)
86 #define BCH_ECC_BYTES(_p)      DIV_ROUND_UP(GF_M(_p)*GF_T(_p), 8)
87 
88 #ifndef dbg
89 #define dbg(_fmt, args...)     do {} while (0)
90 #endif
91 
92 /*
93  * represent a polynomial over GF(2^m)
94  */
95 struct gf_poly {
96 	unsigned int deg;    /* polynomial degree */
97 	unsigned int c[0];   /* polynomial terms */
98 };
99 
100 /* given its degree, compute a polynomial size in bytes */
101 #define GF_POLY_SZ(_d) (sizeof(struct gf_poly)+((_d)+1)*sizeof(unsigned int))
102 
103 /* polynomial of degree 1 */
104 struct gf_poly_deg1 {
105 	struct gf_poly poly;
106 	unsigned int   c[2];
107 };
108 
109 /*
110  * same as encode_bch(), but process input data one byte at a time
111  */
112 static void encode_bch_unaligned(struct bch_control *bch,
113 				 const unsigned char *data, unsigned int len,
114 				 uint32_t *ecc)
115 {
116 	int i;
117 	const uint32_t *p;
118 	const int l = BCH_ECC_WORDS(bch)-1;
119 
120 	while (len--) {
121 		p = bch->mod8_tab + (l+1)*(((ecc[0] >> 24)^(*data++)) & 0xff);
122 
123 		for (i = 0; i < l; i++)
124 			ecc[i] = ((ecc[i] << 8)|(ecc[i+1] >> 24))^(*p++);
125 
126 		ecc[l] = (ecc[l] << 8)^(*p);
127 	}
128 }
129 
130 /*
131  * convert ecc bytes to aligned, zero-padded 32-bit ecc words
132  */
133 static void load_ecc8(struct bch_control *bch, uint32_t *dst,
134 		      const uint8_t *src)
135 {
136 	uint8_t pad[4] = {0, 0, 0, 0};
137 	unsigned int i, nwords = BCH_ECC_WORDS(bch)-1;
138 
139 	for (i = 0; i < nwords; i++, src += 4)
140 		dst[i] = (src[0] << 24)|(src[1] << 16)|(src[2] << 8)|src[3];
141 
142 	memcpy(pad, src, BCH_ECC_BYTES(bch)-4*nwords);
143 	dst[nwords] = (pad[0] << 24)|(pad[1] << 16)|(pad[2] << 8)|pad[3];
144 }
145 
146 /*
147  * convert 32-bit ecc words to ecc bytes
148  */
149 static void store_ecc8(struct bch_control *bch, uint8_t *dst,
150 		       const uint32_t *src)
151 {
152 	uint8_t pad[4];
153 	unsigned int i, nwords = BCH_ECC_WORDS(bch)-1;
154 
155 	for (i = 0; i < nwords; i++) {
156 		*dst++ = (src[i] >> 24);
157 		*dst++ = (src[i] >> 16) & 0xff;
158 		*dst++ = (src[i] >>  8) & 0xff;
159 		*dst++ = (src[i] >>  0) & 0xff;
160 	}
161 	pad[0] = (src[nwords] >> 24);
162 	pad[1] = (src[nwords] >> 16) & 0xff;
163 	pad[2] = (src[nwords] >>  8) & 0xff;
164 	pad[3] = (src[nwords] >>  0) & 0xff;
165 	memcpy(dst, pad, BCH_ECC_BYTES(bch)-4*nwords);
166 }
167 
168 /**
169  * encode_bch - calculate BCH ecc parity of data
170  * @bch:   BCH control structure
171  * @data:  data to encode
172  * @len:   data length in bytes
173  * @ecc:   ecc parity data, must be initialized by caller
174  *
175  * The @ecc parity array is used both as input and output parameter, in order to
176  * allow incremental computations. It should be of the size indicated by member
177  * @ecc_bytes of @bch, and should be initialized to 0 before the first call.
178  *
179  * The exact number of computed ecc parity bits is given by member @ecc_bits of
180  * @bch; it may be less than m*t for large values of t.
181  */
182 void encode_bch(struct bch_control *bch, const uint8_t *data,
183 		unsigned int len, uint8_t *ecc)
184 {
185 	const unsigned int l = BCH_ECC_WORDS(bch)-1;
186 	unsigned int i, mlen;
187 	unsigned long m;
188 	uint32_t w, r[l+1];
189 	const uint32_t * const tab0 = bch->mod8_tab;
190 	const uint32_t * const tab1 = tab0 + 256*(l+1);
191 	const uint32_t * const tab2 = tab1 + 256*(l+1);
192 	const uint32_t * const tab3 = tab2 + 256*(l+1);
193 	const uint32_t *pdata, *p0, *p1, *p2, *p3;
194 
195 	if (ecc) {
196 		/* load ecc parity bytes into internal 32-bit buffer */
197 		load_ecc8(bch, bch->ecc_buf, ecc);
198 	} else {
199 		memset(bch->ecc_buf, 0, sizeof(r));
200 	}
201 
202 	/* process first unaligned data bytes */
203 	m = ((unsigned long)data) & 3;
204 	if (m) {
205 		mlen = (len < (4-m)) ? len : 4-m;
206 		encode_bch_unaligned(bch, data, mlen, bch->ecc_buf);
207 		data += mlen;
208 		len  -= mlen;
209 	}
210 
211 	/* process 32-bit aligned data words */
212 	pdata = (uint32_t *)data;
213 	mlen  = len/4;
214 	data += 4*mlen;
215 	len  -= 4*mlen;
216 	memcpy(r, bch->ecc_buf, sizeof(r));
217 
218 	/*
219 	 * split each 32-bit word into 4 polynomials of weight 8 as follows:
220 	 *
221 	 * 31 ...24  23 ...16  15 ... 8  7 ... 0
222 	 * xxxxxxxx  yyyyyyyy  zzzzzzzz  tttttttt
223 	 *                               tttttttt  mod g = r0 (precomputed)
224 	 *                     zzzzzzzz  00000000  mod g = r1 (precomputed)
225 	 *           yyyyyyyy  00000000  00000000  mod g = r2 (precomputed)
226 	 * xxxxxxxx  00000000  00000000  00000000  mod g = r3 (precomputed)
227 	 * xxxxxxxx  yyyyyyyy  zzzzzzzz  tttttttt  mod g = r0^r1^r2^r3
228 	 */
229 	while (mlen--) {
230 		/* input data is read in big-endian format */
231 		w = r[0]^cpu_to_be32(*pdata++);
232 		p0 = tab0 + (l+1)*((w >>  0) & 0xff);
233 		p1 = tab1 + (l+1)*((w >>  8) & 0xff);
234 		p2 = tab2 + (l+1)*((w >> 16) & 0xff);
235 		p3 = tab3 + (l+1)*((w >> 24) & 0xff);
236 
237 		for (i = 0; i < l; i++)
238 			r[i] = r[i+1]^p0[i]^p1[i]^p2[i]^p3[i];
239 
240 		r[l] = p0[l]^p1[l]^p2[l]^p3[l];
241 	}
242 	memcpy(bch->ecc_buf, r, sizeof(r));
243 
244 	/* process last unaligned bytes */
245 	if (len)
246 		encode_bch_unaligned(bch, data, len, bch->ecc_buf);
247 
248 	/* store ecc parity bytes into original parity buffer */
249 	if (ecc)
250 		store_ecc8(bch, ecc, bch->ecc_buf);
251 }
252 
253 static inline int modulo(struct bch_control *bch, unsigned int v)
254 {
255 	const unsigned int n = GF_N(bch);
256 	while (v >= n) {
257 		v -= n;
258 		v = (v & n) + (v >> GF_M(bch));
259 	}
260 	return v;
261 }
262 
263 /*
264  * shorter and faster modulo function, only works when v < 2N.
265  */
266 static inline int mod_s(struct bch_control *bch, unsigned int v)
267 {
268 	const unsigned int n = GF_N(bch);
269 	return (v < n) ? v : v-n;
270 }
271 
272 static inline int deg(unsigned int poly)
273 {
274 	/* polynomial degree is the most-significant bit index */
275 	return fls(poly)-1;
276 }
277 
278 static inline int parity(unsigned int x)
279 {
280 	/*
281 	 * public domain code snippet, lifted from
282 	 * http://www-graphics.stanford.edu/~seander/bithacks.html
283 	 */
284 	x ^= x >> 1;
285 	x ^= x >> 2;
286 	x = (x & 0x11111111U) * 0x11111111U;
287 	return (x >> 28) & 1;
288 }
289 
290 /* Galois field basic operations: multiply, divide, inverse, etc. */
291 
292 static inline unsigned int gf_mul(struct bch_control *bch, unsigned int a,
293 				  unsigned int b)
294 {
295 	return (a && b) ? bch->a_pow_tab[mod_s(bch, bch->a_log_tab[a]+
296 					       bch->a_log_tab[b])] : 0;
297 }
298 
299 static inline unsigned int gf_sqr(struct bch_control *bch, unsigned int a)
300 {
301 	return a ? bch->a_pow_tab[mod_s(bch, 2*bch->a_log_tab[a])] : 0;
302 }
303 
304 static inline unsigned int gf_div(struct bch_control *bch, unsigned int a,
305 				  unsigned int b)
306 {
307 	return a ? bch->a_pow_tab[mod_s(bch, bch->a_log_tab[a]+
308 					GF_N(bch)-bch->a_log_tab[b])] : 0;
309 }
310 
311 static inline unsigned int gf_inv(struct bch_control *bch, unsigned int a)
312 {
313 	return bch->a_pow_tab[GF_N(bch)-bch->a_log_tab[a]];
314 }
315 
316 static inline unsigned int a_pow(struct bch_control *bch, int i)
317 {
318 	return bch->a_pow_tab[modulo(bch, i)];
319 }
320 
321 static inline int a_log(struct bch_control *bch, unsigned int x)
322 {
323 	return bch->a_log_tab[x];
324 }
325 
326 static inline int a_ilog(struct bch_control *bch, unsigned int x)
327 {
328 	return mod_s(bch, GF_N(bch)-bch->a_log_tab[x]);
329 }
330 
331 /*
332  * compute 2t syndromes of ecc polynomial, i.e. ecc(a^j) for j=1..2t
333  */
334 static void compute_syndromes(struct bch_control *bch, uint32_t *ecc,
335 			      unsigned int *syn)
336 {
337 	int i, j, s;
338 	unsigned int m;
339 	uint32_t poly;
340 	const int t = GF_T(bch);
341 
342 	s = bch->ecc_bits;
343 
344 	/* make sure extra bits in last ecc word are cleared */
345 	m = ((unsigned int)s) & 31;
346 	if (m)
347 		ecc[s/32] &= ~((1u << (32-m))-1);
348 	memset(syn, 0, 2*t*sizeof(*syn));
349 
350 	/* compute v(a^j) for j=1 .. 2t-1 */
351 	do {
352 		poly = *ecc++;
353 		s -= 32;
354 		while (poly) {
355 			i = deg(poly);
356 			for (j = 0; j < 2*t; j += 2)
357 				syn[j] ^= a_pow(bch, (j+1)*(i+s));
358 
359 			poly ^= (1 << i);
360 		}
361 	} while (s > 0);
362 
363 	/* v(a^(2j)) = v(a^j)^2 */
364 	for (j = 0; j < t; j++)
365 		syn[2*j+1] = gf_sqr(bch, syn[j]);
366 }
367 
368 static void gf_poly_copy(struct gf_poly *dst, struct gf_poly *src)
369 {
370 	memcpy(dst, src, GF_POLY_SZ(src->deg));
371 }
372 
373 static int compute_error_locator_polynomial(struct bch_control *bch,
374 					    const unsigned int *syn)
375 {
376 	const unsigned int t = GF_T(bch);
377 	const unsigned int n = GF_N(bch);
378 	unsigned int i, j, tmp, l, pd = 1, d = syn[0];
379 	struct gf_poly *elp = bch->elp;
380 	struct gf_poly *pelp = bch->poly_2t[0];
381 	struct gf_poly *elp_copy = bch->poly_2t[1];
382 	int k, pp = -1;
383 
384 	memset(pelp, 0, GF_POLY_SZ(2*t));
385 	memset(elp, 0, GF_POLY_SZ(2*t));
386 
387 	pelp->deg = 0;
388 	pelp->c[0] = 1;
389 	elp->deg = 0;
390 	elp->c[0] = 1;
391 
392 	/* use simplified binary Berlekamp-Massey algorithm */
393 	for (i = 0; (i < t) && (elp->deg <= t); i++) {
394 		if (d) {
395 			k = 2*i-pp;
396 			gf_poly_copy(elp_copy, elp);
397 			/* e[i+1](X) = e[i](X)+di*dp^-1*X^2(i-p)*e[p](X) */
398 			tmp = a_log(bch, d)+n-a_log(bch, pd);
399 			for (j = 0; j <= pelp->deg; j++) {
400 				if (pelp->c[j]) {
401 					l = a_log(bch, pelp->c[j]);
402 					elp->c[j+k] ^= a_pow(bch, tmp+l);
403 				}
404 			}
405 			/* compute l[i+1] = max(l[i]->c[l[p]+2*(i-p]) */
406 			tmp = pelp->deg+k;
407 			if (tmp > elp->deg) {
408 				elp->deg = tmp;
409 				gf_poly_copy(pelp, elp_copy);
410 				pd = d;
411 				pp = 2*i;
412 			}
413 		}
414 		/* di+1 = S(2i+3)+elp[i+1].1*S(2i+2)+...+elp[i+1].lS(2i+3-l) */
415 		if (i < t-1) {
416 			d = syn[2*i+2];
417 			for (j = 1; j <= elp->deg; j++)
418 				d ^= gf_mul(bch, elp->c[j], syn[2*i+2-j]);
419 		}
420 	}
421 	dbg("elp=%s\n", gf_poly_str(elp));
422 	return (elp->deg > t) ? -1 : (int)elp->deg;
423 }
424 
425 /*
426  * solve a m x m linear system in GF(2) with an expected number of solutions,
427  * and return the number of found solutions
428  */
429 static int solve_linear_system(struct bch_control *bch, unsigned int *rows,
430 			       unsigned int *sol, int nsol)
431 {
432 	const int m = GF_M(bch);
433 	unsigned int tmp, mask;
434 	int rem, c, r, p, k, param[m];
435 
436 	k = 0;
437 	mask = 1 << m;
438 
439 	/* Gaussian elimination */
440 	for (c = 0; c < m; c++) {
441 		rem = 0;
442 		p = c-k;
443 		/* find suitable row for elimination */
444 		for (r = p; r < m; r++) {
445 			if (rows[r] & mask) {
446 				if (r != p) {
447 					tmp = rows[r];
448 					rows[r] = rows[p];
449 					rows[p] = tmp;
450 				}
451 				rem = r+1;
452 				break;
453 			}
454 		}
455 		if (rem) {
456 			/* perform elimination on remaining rows */
457 			tmp = rows[p];
458 			for (r = rem; r < m; r++) {
459 				if (rows[r] & mask)
460 					rows[r] ^= tmp;
461 			}
462 		} else {
463 			/* elimination not needed, store defective row index */
464 			param[k++] = c;
465 		}
466 		mask >>= 1;
467 	}
468 	/* rewrite system, inserting fake parameter rows */
469 	if (k > 0) {
470 		p = k;
471 		for (r = m-1; r >= 0; r--) {
472 			if ((r > m-1-k) && rows[r])
473 				/* system has no solution */
474 				return 0;
475 
476 			rows[r] = (p && (r == param[p-1])) ?
477 				p--, 1u << (m-r) : rows[r-p];
478 		}
479 	}
480 
481 	if (nsol != (1 << k))
482 		/* unexpected number of solutions */
483 		return 0;
484 
485 	for (p = 0; p < nsol; p++) {
486 		/* set parameters for p-th solution */
487 		for (c = 0; c < k; c++)
488 			rows[param[c]] = (rows[param[c]] & ~1)|((p >> c) & 1);
489 
490 		/* compute unique solution */
491 		tmp = 0;
492 		for (r = m-1; r >= 0; r--) {
493 			mask = rows[r] & (tmp|1);
494 			tmp |= parity(mask) << (m-r);
495 		}
496 		sol[p] = tmp >> 1;
497 	}
498 	return nsol;
499 }
500 
501 /*
502  * this function builds and solves a linear system for finding roots of a degree
503  * 4 affine monic polynomial X^4+aX^2+bX+c over GF(2^m).
504  */
505 static int find_affine4_roots(struct bch_control *bch, unsigned int a,
506 			      unsigned int b, unsigned int c,
507 			      unsigned int *roots)
508 {
509 	int i, j, k;
510 	const int m = GF_M(bch);
511 	unsigned int mask = 0xff, t, rows[16] = {0,};
512 
513 	j = a_log(bch, b);
514 	k = a_log(bch, a);
515 	rows[0] = c;
516 
517 	/* buid linear system to solve X^4+aX^2+bX+c = 0 */
518 	for (i = 0; i < m; i++) {
519 		rows[i+1] = bch->a_pow_tab[4*i]^
520 			(a ? bch->a_pow_tab[mod_s(bch, k)] : 0)^
521 			(b ? bch->a_pow_tab[mod_s(bch, j)] : 0);
522 		j++;
523 		k += 2;
524 	}
525 	/*
526 	 * transpose 16x16 matrix before passing it to linear solver
527 	 * warning: this code assumes m < 16
528 	 */
529 	for (j = 8; j != 0; j >>= 1, mask ^= (mask << j)) {
530 		for (k = 0; k < 16; k = (k+j+1) & ~j) {
531 			t = ((rows[k] >> j)^rows[k+j]) & mask;
532 			rows[k] ^= (t << j);
533 			rows[k+j] ^= t;
534 		}
535 	}
536 	return solve_linear_system(bch, rows, roots, 4);
537 }
538 
539 /*
540  * compute root r of a degree 1 polynomial over GF(2^m) (returned as log(1/r))
541  */
542 static int find_poly_deg1_roots(struct bch_control *bch, struct gf_poly *poly,
543 				unsigned int *roots)
544 {
545 	int n = 0;
546 
547 	if (poly->c[0])
548 		/* poly[X] = bX+c with c!=0, root=c/b */
549 		roots[n++] = mod_s(bch, GF_N(bch)-bch->a_log_tab[poly->c[0]]+
550 				   bch->a_log_tab[poly->c[1]]);
551 	return n;
552 }
553 
554 /*
555  * compute roots of a degree 2 polynomial over GF(2^m)
556  */
557 static int find_poly_deg2_roots(struct bch_control *bch, struct gf_poly *poly,
558 				unsigned int *roots)
559 {
560 	int n = 0, i, l0, l1, l2;
561 	unsigned int u, v, r;
562 
563 	if (poly->c[0] && poly->c[1]) {
564 
565 		l0 = bch->a_log_tab[poly->c[0]];
566 		l1 = bch->a_log_tab[poly->c[1]];
567 		l2 = bch->a_log_tab[poly->c[2]];
568 
569 		/* using z=a/bX, transform aX^2+bX+c into z^2+z+u (u=ac/b^2) */
570 		u = a_pow(bch, l0+l2+2*(GF_N(bch)-l1));
571 		/*
572 		 * let u = sum(li.a^i) i=0..m-1; then compute r = sum(li.xi):
573 		 * r^2+r = sum(li.(xi^2+xi)) = sum(li.(a^i+Tr(a^i).a^k)) =
574 		 * u + sum(li.Tr(a^i).a^k) = u+a^k.Tr(sum(li.a^i)) = u+a^k.Tr(u)
575 		 * i.e. r and r+1 are roots iff Tr(u)=0
576 		 */
577 		r = 0;
578 		v = u;
579 		while (v) {
580 			i = deg(v);
581 			r ^= bch->xi_tab[i];
582 			v ^= (1 << i);
583 		}
584 		/* verify root */
585 		if ((gf_sqr(bch, r)^r) == u) {
586 			/* reverse z=a/bX transformation and compute log(1/r) */
587 			roots[n++] = modulo(bch, 2*GF_N(bch)-l1-
588 					    bch->a_log_tab[r]+l2);
589 			roots[n++] = modulo(bch, 2*GF_N(bch)-l1-
590 					    bch->a_log_tab[r^1]+l2);
591 		}
592 	}
593 	return n;
594 }
595 
596 /*
597  * compute roots of a degree 3 polynomial over GF(2^m)
598  */
599 static int find_poly_deg3_roots(struct bch_control *bch, struct gf_poly *poly,
600 				unsigned int *roots)
601 {
602 	int i, n = 0;
603 	unsigned int a, b, c, a2, b2, c2, e3, tmp[4];
604 
605 	if (poly->c[0]) {
606 		/* transform polynomial into monic X^3 + a2X^2 + b2X + c2 */
607 		e3 = poly->c[3];
608 		c2 = gf_div(bch, poly->c[0], e3);
609 		b2 = gf_div(bch, poly->c[1], e3);
610 		a2 = gf_div(bch, poly->c[2], e3);
611 
612 		/* (X+a2)(X^3+a2X^2+b2X+c2) = X^4+aX^2+bX+c (affine) */
613 		c = gf_mul(bch, a2, c2);           /* c = a2c2      */
614 		b = gf_mul(bch, a2, b2)^c2;        /* b = a2b2 + c2 */
615 		a = gf_sqr(bch, a2)^b2;            /* a = a2^2 + b2 */
616 
617 		/* find the 4 roots of this affine polynomial */
618 		if (find_affine4_roots(bch, a, b, c, tmp) == 4) {
619 			/* remove a2 from final list of roots */
620 			for (i = 0; i < 4; i++) {
621 				if (tmp[i] != a2)
622 					roots[n++] = a_ilog(bch, tmp[i]);
623 			}
624 		}
625 	}
626 	return n;
627 }
628 
629 /*
630  * compute roots of a degree 4 polynomial over GF(2^m)
631  */
632 static int find_poly_deg4_roots(struct bch_control *bch, struct gf_poly *poly,
633 				unsigned int *roots)
634 {
635 	int i, l, n = 0;
636 	unsigned int a, b, c, d, e = 0, f, a2, b2, c2, e4;
637 
638 	if (poly->c[0] == 0)
639 		return 0;
640 
641 	/* transform polynomial into monic X^4 + aX^3 + bX^2 + cX + d */
642 	e4 = poly->c[4];
643 	d = gf_div(bch, poly->c[0], e4);
644 	c = gf_div(bch, poly->c[1], e4);
645 	b = gf_div(bch, poly->c[2], e4);
646 	a = gf_div(bch, poly->c[3], e4);
647 
648 	/* use Y=1/X transformation to get an affine polynomial */
649 	if (a) {
650 		/* first, eliminate cX by using z=X+e with ae^2+c=0 */
651 		if (c) {
652 			/* compute e such that e^2 = c/a */
653 			f = gf_div(bch, c, a);
654 			l = a_log(bch, f);
655 			l += (l & 1) ? GF_N(bch) : 0;
656 			e = a_pow(bch, l/2);
657 			/*
658 			 * use transformation z=X+e:
659 			 * z^4+e^4 + a(z^3+ez^2+e^2z+e^3) + b(z^2+e^2) +cz+ce+d
660 			 * z^4 + az^3 + (ae+b)z^2 + (ae^2+c)z+e^4+be^2+ae^3+ce+d
661 			 * z^4 + az^3 + (ae+b)z^2 + e^4+be^2+d
662 			 * z^4 + az^3 +     b'z^2 + d'
663 			 */
664 			d = a_pow(bch, 2*l)^gf_mul(bch, b, f)^d;
665 			b = gf_mul(bch, a, e)^b;
666 		}
667 		/* now, use Y=1/X to get Y^4 + b/dY^2 + a/dY + 1/d */
668 		if (d == 0)
669 			/* assume all roots have multiplicity 1 */
670 			return 0;
671 
672 		c2 = gf_inv(bch, d);
673 		b2 = gf_div(bch, a, d);
674 		a2 = gf_div(bch, b, d);
675 	} else {
676 		/* polynomial is already affine */
677 		c2 = d;
678 		b2 = c;
679 		a2 = b;
680 	}
681 	/* find the 4 roots of this affine polynomial */
682 	if (find_affine4_roots(bch, a2, b2, c2, roots) == 4) {
683 		for (i = 0; i < 4; i++) {
684 			/* post-process roots (reverse transformations) */
685 			f = a ? gf_inv(bch, roots[i]) : roots[i];
686 			roots[i] = a_ilog(bch, f^e);
687 		}
688 		n = 4;
689 	}
690 	return n;
691 }
692 
693 /*
694  * build monic, log-based representation of a polynomial
695  */
696 static void gf_poly_logrep(struct bch_control *bch,
697 			   const struct gf_poly *a, int *rep)
698 {
699 	int i, d = a->deg, l = GF_N(bch)-a_log(bch, a->c[a->deg]);
700 
701 	/* represent 0 values with -1; warning, rep[d] is not set to 1 */
702 	for (i = 0; i < d; i++)
703 		rep[i] = a->c[i] ? mod_s(bch, a_log(bch, a->c[i])+l) : -1;
704 }
705 
706 /*
707  * compute polynomial Euclidean division remainder in GF(2^m)[X]
708  */
709 static void gf_poly_mod(struct bch_control *bch, struct gf_poly *a,
710 			const struct gf_poly *b, int *rep)
711 {
712 	int la, p, m;
713 	unsigned int i, j, *c = a->c;
714 	const unsigned int d = b->deg;
715 
716 	if (a->deg < d)
717 		return;
718 
719 	/* reuse or compute log representation of denominator */
720 	if (!rep) {
721 		rep = bch->cache;
722 		gf_poly_logrep(bch, b, rep);
723 	}
724 
725 	for (j = a->deg; j >= d; j--) {
726 		if (c[j]) {
727 			la = a_log(bch, c[j]);
728 			p = j-d;
729 			for (i = 0; i < d; i++, p++) {
730 				m = rep[i];
731 				if (m >= 0)
732 					c[p] ^= bch->a_pow_tab[mod_s(bch,
733 								     m+la)];
734 			}
735 		}
736 	}
737 	a->deg = d-1;
738 	while (!c[a->deg] && a->deg)
739 		a->deg--;
740 }
741 
742 /*
743  * compute polynomial Euclidean division quotient in GF(2^m)[X]
744  */
745 static void gf_poly_div(struct bch_control *bch, struct gf_poly *a,
746 			const struct gf_poly *b, struct gf_poly *q)
747 {
748 	if (a->deg >= b->deg) {
749 		q->deg = a->deg-b->deg;
750 		/* compute a mod b (modifies a) */
751 		gf_poly_mod(bch, a, b, NULL);
752 		/* quotient is stored in upper part of polynomial a */
753 		memcpy(q->c, &a->c[b->deg], (1+q->deg)*sizeof(unsigned int));
754 	} else {
755 		q->deg = 0;
756 		q->c[0] = 0;
757 	}
758 }
759 
760 /*
761  * compute polynomial GCD (Greatest Common Divisor) in GF(2^m)[X]
762  */
763 static struct gf_poly *gf_poly_gcd(struct bch_control *bch, struct gf_poly *a,
764 				   struct gf_poly *b)
765 {
766 	struct gf_poly *tmp;
767 
768 	dbg("gcd(%s,%s)=", gf_poly_str(a), gf_poly_str(b));
769 
770 	if (a->deg < b->deg) {
771 		tmp = b;
772 		b = a;
773 		a = tmp;
774 	}
775 
776 	while (b->deg > 0) {
777 		gf_poly_mod(bch, a, b, NULL);
778 		tmp = b;
779 		b = a;
780 		a = tmp;
781 	}
782 
783 	dbg("%s\n", gf_poly_str(a));
784 
785 	return a;
786 }
787 
788 /*
789  * Given a polynomial f and an integer k, compute Tr(a^kX) mod f
790  * This is used in Berlekamp Trace algorithm for splitting polynomials
791  */
792 static void compute_trace_bk_mod(struct bch_control *bch, int k,
793 				 const struct gf_poly *f, struct gf_poly *z,
794 				 struct gf_poly *out)
795 {
796 	const int m = GF_M(bch);
797 	int i, j;
798 
799 	/* z contains z^2j mod f */
800 	z->deg = 1;
801 	z->c[0] = 0;
802 	z->c[1] = bch->a_pow_tab[k];
803 
804 	out->deg = 0;
805 	memset(out, 0, GF_POLY_SZ(f->deg));
806 
807 	/* compute f log representation only once */
808 	gf_poly_logrep(bch, f, bch->cache);
809 
810 	for (i = 0; i < m; i++) {
811 		/* add a^(k*2^i)(z^(2^i) mod f) and compute (z^(2^i) mod f)^2 */
812 		for (j = z->deg; j >= 0; j--) {
813 			out->c[j] ^= z->c[j];
814 			z->c[2*j] = gf_sqr(bch, z->c[j]);
815 			z->c[2*j+1] = 0;
816 		}
817 		if (z->deg > out->deg)
818 			out->deg = z->deg;
819 
820 		if (i < m-1) {
821 			z->deg *= 2;
822 			/* z^(2(i+1)) mod f = (z^(2^i) mod f)^2 mod f */
823 			gf_poly_mod(bch, z, f, bch->cache);
824 		}
825 	}
826 	while (!out->c[out->deg] && out->deg)
827 		out->deg--;
828 
829 	dbg("Tr(a^%d.X) mod f = %s\n", k, gf_poly_str(out));
830 }
831 
832 /*
833  * factor a polynomial using Berlekamp Trace algorithm (BTA)
834  */
835 static void factor_polynomial(struct bch_control *bch, int k, struct gf_poly *f,
836 			      struct gf_poly **g, struct gf_poly **h)
837 {
838 	struct gf_poly *f2 = bch->poly_2t[0];
839 	struct gf_poly *q  = bch->poly_2t[1];
840 	struct gf_poly *tk = bch->poly_2t[2];
841 	struct gf_poly *z  = bch->poly_2t[3];
842 	struct gf_poly *gcd;
843 
844 	dbg("factoring %s...\n", gf_poly_str(f));
845 
846 	*g = f;
847 	*h = NULL;
848 
849 	/* tk = Tr(a^k.X) mod f */
850 	compute_trace_bk_mod(bch, k, f, z, tk);
851 
852 	if (tk->deg > 0) {
853 		/* compute g = gcd(f, tk) (destructive operation) */
854 		gf_poly_copy(f2, f);
855 		gcd = gf_poly_gcd(bch, f2, tk);
856 		if (gcd->deg < f->deg) {
857 			/* compute h=f/gcd(f,tk); this will modify f and q */
858 			gf_poly_div(bch, f, gcd, q);
859 			/* store g and h in-place (clobbering f) */
860 			*h = &((struct gf_poly_deg1 *)f)[gcd->deg].poly;
861 			gf_poly_copy(*g, gcd);
862 			gf_poly_copy(*h, q);
863 		}
864 	}
865 }
866 
867 /*
868  * find roots of a polynomial, using BTZ algorithm; see the beginning of this
869  * file for details
870  */
871 static int find_poly_roots(struct bch_control *bch, unsigned int k,
872 			   struct gf_poly *poly, unsigned int *roots)
873 {
874 	int cnt;
875 	struct gf_poly *f1, *f2;
876 
877 	switch (poly->deg) {
878 		/* handle low degree polynomials with ad hoc techniques */
879 	case 1:
880 		cnt = find_poly_deg1_roots(bch, poly, roots);
881 		break;
882 	case 2:
883 		cnt = find_poly_deg2_roots(bch, poly, roots);
884 		break;
885 	case 3:
886 		cnt = find_poly_deg3_roots(bch, poly, roots);
887 		break;
888 	case 4:
889 		cnt = find_poly_deg4_roots(bch, poly, roots);
890 		break;
891 	default:
892 		/* factor polynomial using Berlekamp Trace Algorithm (BTA) */
893 		cnt = 0;
894 		if (poly->deg && (k <= GF_M(bch))) {
895 			factor_polynomial(bch, k, poly, &f1, &f2);
896 			if (f1)
897 				cnt += find_poly_roots(bch, k+1, f1, roots);
898 			if (f2)
899 				cnt += find_poly_roots(bch, k+1, f2, roots+cnt);
900 		}
901 		break;
902 	}
903 	return cnt;
904 }
905 
906 #if defined(USE_CHIEN_SEARCH)
907 /*
908  * exhaustive root search (Chien) implementation - not used, included only for
909  * reference/comparison tests
910  */
911 static int chien_search(struct bch_control *bch, unsigned int len,
912 			struct gf_poly *p, unsigned int *roots)
913 {
914 	int m;
915 	unsigned int i, j, syn, syn0, count = 0;
916 	const unsigned int k = 8*len+bch->ecc_bits;
917 
918 	/* use a log-based representation of polynomial */
919 	gf_poly_logrep(bch, p, bch->cache);
920 	bch->cache[p->deg] = 0;
921 	syn0 = gf_div(bch, p->c[0], p->c[p->deg]);
922 
923 	for (i = GF_N(bch)-k+1; i <= GF_N(bch); i++) {
924 		/* compute elp(a^i) */
925 		for (j = 1, syn = syn0; j <= p->deg; j++) {
926 			m = bch->cache[j];
927 			if (m >= 0)
928 				syn ^= a_pow(bch, m+j*i);
929 		}
930 		if (syn == 0) {
931 			roots[count++] = GF_N(bch)-i;
932 			if (count == p->deg)
933 				break;
934 		}
935 	}
936 	return (count == p->deg) ? count : 0;
937 }
938 #define find_poly_roots(_p, _k, _elp, _loc) chien_search(_p, len, _elp, _loc)
939 #endif /* USE_CHIEN_SEARCH */
940 
941 /**
942  * decode_bch - decode received codeword and find bit error locations
943  * @bch:      BCH control structure
944  * @data:     received data, ignored if @calc_ecc is provided
945  * @len:      data length in bytes, must always be provided
946  * @recv_ecc: received ecc, if NULL then assume it was XORed in @calc_ecc
947  * @calc_ecc: calculated ecc, if NULL then calc_ecc is computed from @data
948  * @syn:      hw computed syndrome data (if NULL, syndrome is calculated)
949  * @errloc:   output array of error locations
950  *
951  * Returns:
952  *  The number of errors found, or -EBADMSG if decoding failed, or -EINVAL if
953  *  invalid parameters were provided
954  *
955  * Depending on the available hw BCH support and the need to compute @calc_ecc
956  * separately (using encode_bch()), this function should be called with one of
957  * the following parameter configurations -
958  *
959  * by providing @data and @recv_ecc only:
960  *   decode_bch(@bch, @data, @len, @recv_ecc, NULL, NULL, @errloc)
961  *
962  * by providing @recv_ecc and @calc_ecc:
963  *   decode_bch(@bch, NULL, @len, @recv_ecc, @calc_ecc, NULL, @errloc)
964  *
965  * by providing ecc = recv_ecc XOR calc_ecc:
966  *   decode_bch(@bch, NULL, @len, NULL, ecc, NULL, @errloc)
967  *
968  * by providing syndrome results @syn:
969  *   decode_bch(@bch, NULL, @len, NULL, NULL, @syn, @errloc)
970  *
971  * Once decode_bch() has successfully returned with a positive value, error
972  * locations returned in array @errloc should be interpreted as follows -
973  *
974  * if (errloc[n] >= 8*len), then n-th error is located in ecc (no need for
975  * data correction)
976  *
977  * if (errloc[n] < 8*len), then n-th error is located in data and can be
978  * corrected with statement data[errloc[n]/8] ^= 1 << (errloc[n] % 8);
979  *
980  * Note that this function does not perform any data correction by itself, it
981  * merely indicates error locations.
982  */
983 int decode_bch(struct bch_control *bch, const uint8_t *data, unsigned int len,
984 	       const uint8_t *recv_ecc, const uint8_t *calc_ecc,
985 	       const unsigned int *syn, unsigned int *errloc)
986 {
987 	const unsigned int ecc_words = BCH_ECC_WORDS(bch);
988 	unsigned int nbits;
989 	int i, err, nroots;
990 	uint32_t sum;
991 
992 	/* sanity check: make sure data length can be handled */
993 	if (8*len > (bch->n-bch->ecc_bits))
994 		return -EINVAL;
995 
996 	/* if caller does not provide syndromes, compute them */
997 	if (!syn) {
998 		if (!calc_ecc) {
999 			/* compute received data ecc into an internal buffer */
1000 			if (!data || !recv_ecc)
1001 				return -EINVAL;
1002 			encode_bch(bch, data, len, NULL);
1003 		} else {
1004 			/* load provided calculated ecc */
1005 			load_ecc8(bch, bch->ecc_buf, calc_ecc);
1006 		}
1007 		/* load received ecc or assume it was XORed in calc_ecc */
1008 		if (recv_ecc) {
1009 			load_ecc8(bch, bch->ecc_buf2, recv_ecc);
1010 			/* XOR received and calculated ecc */
1011 			for (i = 0, sum = 0; i < (int)ecc_words; i++) {
1012 				bch->ecc_buf[i] ^= bch->ecc_buf2[i];
1013 				sum |= bch->ecc_buf[i];
1014 			}
1015 			if (!sum)
1016 				/* no error found */
1017 				return 0;
1018 		}
1019 		compute_syndromes(bch, bch->ecc_buf, bch->syn);
1020 		syn = bch->syn;
1021 	}
1022 
1023 	err = compute_error_locator_polynomial(bch, syn);
1024 	if (err > 0) {
1025 		nroots = find_poly_roots(bch, 1, bch->elp, errloc);
1026 		if (err != nroots)
1027 			err = -1;
1028 	}
1029 	if (err > 0) {
1030 		/* post-process raw error locations for easier correction */
1031 		nbits = (len*8)+bch->ecc_bits;
1032 		for (i = 0; i < err; i++) {
1033 			if (errloc[i] >= nbits) {
1034 				err = -1;
1035 				break;
1036 			}
1037 			errloc[i] = nbits-1-errloc[i];
1038 			errloc[i] = (errloc[i] & ~7)|(7-(errloc[i] & 7));
1039 		}
1040 	}
1041 	return (err >= 0) ? err : -EBADMSG;
1042 }
1043 
1044 /*
1045  * generate Galois field lookup tables
1046  */
1047 static int build_gf_tables(struct bch_control *bch, unsigned int poly)
1048 {
1049 	unsigned int i, x = 1;
1050 	const unsigned int k = 1 << deg(poly);
1051 
1052 	/* primitive polynomial must be of degree m */
1053 	if (k != (1u << GF_M(bch)))
1054 		return -1;
1055 
1056 	for (i = 0; i < GF_N(bch); i++) {
1057 		bch->a_pow_tab[i] = x;
1058 		bch->a_log_tab[x] = i;
1059 		if (i && (x == 1))
1060 			/* polynomial is not primitive (a^i=1 with 0<i<2^m-1) */
1061 			return -1;
1062 		x <<= 1;
1063 		if (x & k)
1064 			x ^= poly;
1065 	}
1066 	bch->a_pow_tab[GF_N(bch)] = 1;
1067 	bch->a_log_tab[0] = 0;
1068 
1069 	return 0;
1070 }
1071 
1072 /*
1073  * compute generator polynomial remainder tables for fast encoding
1074  */
1075 static void build_mod8_tables(struct bch_control *bch, const uint32_t *g)
1076 {
1077 	int i, j, b, d;
1078 	uint32_t data, hi, lo, *tab;
1079 	const int l = BCH_ECC_WORDS(bch);
1080 	const int plen = DIV_ROUND_UP(bch->ecc_bits+1, 32);
1081 	const int ecclen = DIV_ROUND_UP(bch->ecc_bits, 32);
1082 
1083 	memset(bch->mod8_tab, 0, 4*256*l*sizeof(*bch->mod8_tab));
1084 
1085 	for (i = 0; i < 256; i++) {
1086 		/* p(X)=i is a small polynomial of weight <= 8 */
1087 		for (b = 0; b < 4; b++) {
1088 			/* we want to compute (p(X).X^(8*b+deg(g))) mod g(X) */
1089 			tab = bch->mod8_tab + (b*256+i)*l;
1090 			data = i << (8*b);
1091 			while (data) {
1092 				d = deg(data);
1093 				/* subtract X^d.g(X) from p(X).X^(8*b+deg(g)) */
1094 				data ^= g[0] >> (31-d);
1095 				for (j = 0; j < ecclen; j++) {
1096 					hi = (d < 31) ? g[j] << (d+1) : 0;
1097 					lo = (j+1 < plen) ?
1098 						g[j+1] >> (31-d) : 0;
1099 					tab[j] ^= hi|lo;
1100 				}
1101 			}
1102 		}
1103 	}
1104 }
1105 
1106 /*
1107  * build a base for factoring degree 2 polynomials
1108  */
1109 static int build_deg2_base(struct bch_control *bch)
1110 {
1111 	const int m = GF_M(bch);
1112 	int i, j, r;
1113 	unsigned int sum, x, y, remaining, ak = 0, xi[m];
1114 
1115 	/* find k s.t. Tr(a^k) = 1 and 0 <= k < m */
1116 	for (i = 0; i < m; i++) {
1117 		for (j = 0, sum = 0; j < m; j++)
1118 			sum ^= a_pow(bch, i*(1 << j));
1119 
1120 		if (sum) {
1121 			ak = bch->a_pow_tab[i];
1122 			break;
1123 		}
1124 	}
1125 	/* find xi, i=0..m-1 such that xi^2+xi = a^i+Tr(a^i).a^k */
1126 	remaining = m;
1127 	memset(xi, 0, sizeof(xi));
1128 
1129 	for (x = 0; (x <= GF_N(bch)) && remaining; x++) {
1130 		y = gf_sqr(bch, x)^x;
1131 		for (i = 0; i < 2; i++) {
1132 			r = a_log(bch, y);
1133 			if (y && (r < m) && !xi[r]) {
1134 				bch->xi_tab[r] = x;
1135 				xi[r] = 1;
1136 				remaining--;
1137 				dbg("x%d = %x\n", r, x);
1138 				break;
1139 			}
1140 			y ^= ak;
1141 		}
1142 	}
1143 	/* should not happen but check anyway */
1144 	return remaining ? -1 : 0;
1145 }
1146 
1147 static void *bch_alloc(size_t size, int *err)
1148 {
1149 	void *ptr;
1150 
1151 	ptr = kmalloc(size, GFP_KERNEL);
1152 	if (ptr == NULL)
1153 		*err = 1;
1154 	return ptr;
1155 }
1156 
1157 /*
1158  * compute generator polynomial for given (m,t) parameters.
1159  */
1160 static uint32_t *compute_generator_polynomial(struct bch_control *bch)
1161 {
1162 	const unsigned int m = GF_M(bch);
1163 	const unsigned int t = GF_T(bch);
1164 	int n, err = 0;
1165 	unsigned int i, j, nbits, r, word, *roots;
1166 	struct gf_poly *g;
1167 	uint32_t *genpoly;
1168 
1169 	g = bch_alloc(GF_POLY_SZ(m*t), &err);
1170 	roots = bch_alloc((bch->n+1)*sizeof(*roots), &err);
1171 	genpoly = bch_alloc(DIV_ROUND_UP(m*t+1, 32)*sizeof(*genpoly), &err);
1172 
1173 	if (err) {
1174 		kfree(genpoly);
1175 		genpoly = NULL;
1176 		goto finish;
1177 	}
1178 
1179 	/* enumerate all roots of g(X) */
1180 	memset(roots , 0, (bch->n+1)*sizeof(*roots));
1181 	for (i = 0; i < t; i++) {
1182 		for (j = 0, r = 2*i+1; j < m; j++) {
1183 			roots[r] = 1;
1184 			r = mod_s(bch, 2*r);
1185 		}
1186 	}
1187 	/* build generator polynomial g(X) */
1188 	g->deg = 0;
1189 	g->c[0] = 1;
1190 	for (i = 0; i < GF_N(bch); i++) {
1191 		if (roots[i]) {
1192 			/* multiply g(X) by (X+root) */
1193 			r = bch->a_pow_tab[i];
1194 			g->c[g->deg+1] = 1;
1195 			for (j = g->deg; j > 0; j--)
1196 				g->c[j] = gf_mul(bch, g->c[j], r)^g->c[j-1];
1197 
1198 			g->c[0] = gf_mul(bch, g->c[0], r);
1199 			g->deg++;
1200 		}
1201 	}
1202 	/* store left-justified binary representation of g(X) */
1203 	n = g->deg+1;
1204 	i = 0;
1205 
1206 	while (n > 0) {
1207 		nbits = (n > 32) ? 32 : n;
1208 		for (j = 0, word = 0; j < nbits; j++) {
1209 			if (g->c[n-1-j])
1210 				word |= 1u << (31-j);
1211 		}
1212 		genpoly[i++] = word;
1213 		n -= nbits;
1214 	}
1215 	bch->ecc_bits = g->deg;
1216 
1217 finish:
1218 	kfree(g);
1219 	kfree(roots);
1220 
1221 	return genpoly;
1222 }
1223 
1224 /**
1225  * init_bch - initialize a BCH encoder/decoder
1226  * @m:          Galois field order, should be in the range 5-15
1227  * @t:          maximum error correction capability, in bits
1228  * @prim_poly:  user-provided primitive polynomial (or 0 to use default)
1229  *
1230  * Returns:
1231  *  a newly allocated BCH control structure if successful, NULL otherwise
1232  *
1233  * This initialization can take some time, as lookup tables are built for fast
1234  * encoding/decoding; make sure not to call this function from a time critical
1235  * path. Usually, init_bch() should be called on module/driver init and
1236  * free_bch() should be called to release memory on exit.
1237  *
1238  * You may provide your own primitive polynomial of degree @m in argument
1239  * @prim_poly, or let init_bch() use its default polynomial.
1240  *
1241  * Once init_bch() has successfully returned a pointer to a newly allocated
1242  * BCH control structure, ecc length in bytes is given by member @ecc_bytes of
1243  * the structure.
1244  */
1245 struct bch_control *init_bch(int m, int t, unsigned int prim_poly)
1246 {
1247 	int err = 0;
1248 	unsigned int i, words;
1249 	uint32_t *genpoly;
1250 	struct bch_control *bch = NULL;
1251 
1252 	const int min_m = 5;
1253 	const int max_m = 15;
1254 
1255 	/* default primitive polynomials */
1256 	static const unsigned int prim_poly_tab[] = {
1257 		0x25, 0x43, 0x83, 0x11d, 0x211, 0x409, 0x805, 0x1053, 0x201b,
1258 		0x402b, 0x8003,
1259 	};
1260 
1261 #if defined(CONFIG_BCH_CONST_PARAMS)
1262 	if ((m != (CONFIG_BCH_CONST_M)) || (t != (CONFIG_BCH_CONST_T))) {
1263 		printk(KERN_ERR "bch encoder/decoder was configured to support "
1264 		       "parameters m=%d, t=%d only!\n",
1265 		       CONFIG_BCH_CONST_M, CONFIG_BCH_CONST_T);
1266 		goto fail;
1267 	}
1268 #endif
1269 	if ((m < min_m) || (m > max_m))
1270 		/*
1271 		 * values of m greater than 15 are not currently supported;
1272 		 * supporting m > 15 would require changing table base type
1273 		 * (uint16_t) and a small patch in matrix transposition
1274 		 */
1275 		goto fail;
1276 
1277 	/* sanity checks */
1278 	if ((t < 1) || (m*t >= ((1 << m)-1)))
1279 		/* invalid t value */
1280 		goto fail;
1281 
1282 	/* select a primitive polynomial for generating GF(2^m) */
1283 	if (prim_poly == 0)
1284 		prim_poly = prim_poly_tab[m-min_m];
1285 
1286 	bch = kzalloc(sizeof(*bch), GFP_KERNEL);
1287 	if (bch == NULL)
1288 		goto fail;
1289 
1290 	bch->m = m;
1291 	bch->t = t;
1292 	bch->n = (1 << m)-1;
1293 	words  = DIV_ROUND_UP(m*t, 32);
1294 	bch->ecc_bytes = DIV_ROUND_UP(m*t, 8);
1295 	bch->a_pow_tab = bch_alloc((1+bch->n)*sizeof(*bch->a_pow_tab), &err);
1296 	bch->a_log_tab = bch_alloc((1+bch->n)*sizeof(*bch->a_log_tab), &err);
1297 	bch->mod8_tab  = bch_alloc(words*1024*sizeof(*bch->mod8_tab), &err);
1298 	bch->ecc_buf   = bch_alloc(words*sizeof(*bch->ecc_buf), &err);
1299 	bch->ecc_buf2  = bch_alloc(words*sizeof(*bch->ecc_buf2), &err);
1300 	bch->xi_tab    = bch_alloc(m*sizeof(*bch->xi_tab), &err);
1301 	bch->syn       = bch_alloc(2*t*sizeof(*bch->syn), &err);
1302 	bch->cache     = bch_alloc(2*t*sizeof(*bch->cache), &err);
1303 	bch->elp       = bch_alloc((t+1)*sizeof(struct gf_poly_deg1), &err);
1304 
1305 	for (i = 0; i < ARRAY_SIZE(bch->poly_2t); i++)
1306 		bch->poly_2t[i] = bch_alloc(GF_POLY_SZ(2*t), &err);
1307 
1308 	if (err)
1309 		goto fail;
1310 
1311 	err = build_gf_tables(bch, prim_poly);
1312 	if (err)
1313 		goto fail;
1314 
1315 	/* use generator polynomial for computing encoding tables */
1316 	genpoly = compute_generator_polynomial(bch);
1317 	if (genpoly == NULL)
1318 		goto fail;
1319 
1320 	build_mod8_tables(bch, genpoly);
1321 	kfree(genpoly);
1322 
1323 	err = build_deg2_base(bch);
1324 	if (err)
1325 		goto fail;
1326 
1327 	return bch;
1328 
1329 fail:
1330 	free_bch(bch);
1331 	return NULL;
1332 }
1333 
1334 /**
1335  *  free_bch - free the BCH control structure
1336  *  @bch:    BCH control structure to release
1337  */
1338 void free_bch(struct bch_control *bch)
1339 {
1340 	unsigned int i;
1341 
1342 	if (bch) {
1343 		kfree(bch->a_pow_tab);
1344 		kfree(bch->a_log_tab);
1345 		kfree(bch->mod8_tab);
1346 		kfree(bch->ecc_buf);
1347 		kfree(bch->ecc_buf2);
1348 		kfree(bch->xi_tab);
1349 		kfree(bch->syn);
1350 		kfree(bch->cache);
1351 		kfree(bch->elp);
1352 
1353 		for (i = 0; i < ARRAY_SIZE(bch->poly_2t); i++)
1354 			kfree(bch->poly_2t[i]);
1355 
1356 		kfree(bch);
1357 	}
1358 }
1359