xref: /openbmc/linux/include/crypto/gf128mul.h (revision 27ab1c1c)
1 /* gf128mul.h - GF(2^128) multiplication functions
2  *
3  * Copyright (c) 2003, Dr Brian Gladman, Worcester, UK.
4  * Copyright (c) 2006 Rik Snel <rsnel@cube.dyndns.org>
5  *
6  * Based on Dr Brian Gladman's (GPL'd) work published at
7  * http://fp.gladman.plus.com/cryptography_technology/index.htm
8  * See the original copyright notice below.
9  *
10  * This program is free software; you can redistribute it and/or modify it
11  * under the terms of the GNU General Public License as published by the Free
12  * Software Foundation; either version 2 of the License, or (at your option)
13  * any later version.
14  */
15 /*
16  ---------------------------------------------------------------------------
17  Copyright (c) 2003, Dr Brian Gladman, Worcester, UK.   All rights reserved.
18 
19  LICENSE TERMS
20 
21  The free distribution and use of this software in both source and binary
22  form is allowed (with or without changes) provided that:
23 
24    1. distributions of this source code include the above copyright
25       notice, this list of conditions and the following disclaimer;
26 
27    2. distributions in binary form include the above copyright
28       notice, this list of conditions and the following disclaimer
29       in the documentation and/or other associated materials;
30 
31    3. the copyright holder's name is not used to endorse products
32       built using this software without specific written permission.
33 
34  ALTERNATIVELY, provided that this notice is retained in full, this product
35  may be distributed under the terms of the GNU General Public License (GPL),
36  in which case the provisions of the GPL apply INSTEAD OF those given above.
37 
38  DISCLAIMER
39 
40  This software is provided 'as is' with no explicit or implied warranties
41  in respect of its properties, including, but not limited to, correctness
42  and/or fitness for purpose.
43  ---------------------------------------------------------------------------
44  Issue Date: 31/01/2006
45 
46  An implementation of field multiplication in Galois Field GF(2^128)
47 */
48 
49 #ifndef _CRYPTO_GF128MUL_H
50 #define _CRYPTO_GF128MUL_H
51 
52 #include <asm/byteorder.h>
53 #include <crypto/b128ops.h>
54 #include <linux/slab.h>
55 
56 /* Comment by Rik:
57  *
58  * For some background on GF(2^128) see for example:
59  * http://csrc.nist.gov/groups/ST/toolkit/BCM/documents/proposedmodes/gcm/gcm-revised-spec.pdf
60  *
61  * The elements of GF(2^128) := GF(2)[X]/(X^128-X^7-X^2-X^1-1) can
62  * be mapped to computer memory in a variety of ways. Let's examine
63  * three common cases.
64  *
65  * Take a look at the 16 binary octets below in memory order. The msb's
66  * are left and the lsb's are right. char b[16] is an array and b[0] is
67  * the first octet.
68  *
69  * 10000000 00000000 00000000 00000000 .... 00000000 00000000 00000000
70  *   b[0]     b[1]     b[2]     b[3]          b[13]    b[14]    b[15]
71  *
72  * Every bit is a coefficient of some power of X. We can store the bits
73  * in every byte in little-endian order and the bytes themselves also in
74  * little endian order. I will call this lle (little-little-endian).
75  * The above buffer represents the polynomial 1, and X^7+X^2+X^1+1 looks
76  * like 11100001 00000000 .... 00000000 = { 0xE1, 0x00, }.
77  * This format was originally implemented in gf128mul and is used
78  * in GCM (Galois/Counter mode) and in ABL (Arbitrary Block Length).
79  *
80  * Another convention says: store the bits in bigendian order and the
81  * bytes also. This is bbe (big-big-endian). Now the buffer above
82  * represents X^127. X^7+X^2+X^1+1 looks like 00000000 .... 10000111,
83  * b[15] = 0x87 and the rest is 0. LRW uses this convention and bbe
84  * is partly implemented.
85  *
86  * Both of the above formats are easy to implement on big-endian
87  * machines.
88  *
89  * XTS and EME (the latter of which is patent encumbered) use the ble
90  * format (bits are stored in big endian order and the bytes in little
91  * endian). The above buffer represents X^7 in this case and the
92  * primitive polynomial is b[0] = 0x87.
93  *
94  * The common machine word-size is smaller than 128 bits, so to make
95  * an efficient implementation we must split into machine word sizes.
96  * This implementation uses 64-bit words for the moment. Machine
97  * endianness comes into play. The lle format in relation to machine
98  * endianness is discussed below by the original author of gf128mul Dr
99  * Brian Gladman.
100  *
101  * Let's look at the bbe and ble format on a little endian machine.
102  *
103  * bbe on a little endian machine u32 x[4]:
104  *
105  *  MS            x[0]           LS  MS            x[1]		  LS
106  *  ms   ls ms   ls ms   ls ms   ls  ms   ls ms   ls ms   ls ms   ls
107  *  103..96 111.104 119.112 127.120  71...64 79...72 87...80 95...88
108  *
109  *  MS            x[2]           LS  MS            x[3]		  LS
110  *  ms   ls ms   ls ms   ls ms   ls  ms   ls ms   ls ms   ls ms   ls
111  *  39...32 47...40 55...48 63...56  07...00 15...08 23...16 31...24
112  *
113  * ble on a little endian machine
114  *
115  *  MS            x[0]           LS  MS            x[1]		  LS
116  *  ms   ls ms   ls ms   ls ms   ls  ms   ls ms   ls ms   ls ms   ls
117  *  31...24 23...16 15...08 07...00  63...56 55...48 47...40 39...32
118  *
119  *  MS            x[2]           LS  MS            x[3]		  LS
120  *  ms   ls ms   ls ms   ls ms   ls  ms   ls ms   ls ms   ls ms   ls
121  *  95...88 87...80 79...72 71...64  127.120 199.112 111.104 103..96
122  *
123  * Multiplications in GF(2^128) are mostly bit-shifts, so you see why
124  * ble (and lbe also) are easier to implement on a little-endian
125  * machine than on a big-endian machine. The converse holds for bbe
126  * and lle.
127  *
128  * Note: to have good alignment, it seems to me that it is sufficient
129  * to keep elements of GF(2^128) in type u64[2]. On 32-bit wordsize
130  * machines this will automatically aligned to wordsize and on a 64-bit
131  * machine also.
132  */
133 /*	Multiply a GF(2^128) field element by x. Field elements are
134     held in arrays of bytes in which field bits 8n..8n + 7 are held in
135     byte[n], with lower indexed bits placed in the more numerically
136     significant bit positions within bytes.
137 
138     On little endian machines the bit indexes translate into the bit
139     positions within four 32-bit words in the following way
140 
141     MS            x[0]           LS  MS            x[1]		  LS
142     ms   ls ms   ls ms   ls ms   ls  ms   ls ms   ls ms   ls ms   ls
143     24...31 16...23 08...15 00...07  56...63 48...55 40...47 32...39
144 
145     MS            x[2]           LS  MS            x[3]		  LS
146     ms   ls ms   ls ms   ls ms   ls  ms   ls ms   ls ms   ls ms   ls
147     88...95 80...87 72...79 64...71  120.127 112.119 104.111 96..103
148 
149     On big endian machines the bit indexes translate into the bit
150     positions within four 32-bit words in the following way
151 
152     MS            x[0]           LS  MS            x[1]		  LS
153     ms   ls ms   ls ms   ls ms   ls  ms   ls ms   ls ms   ls ms   ls
154     00...07 08...15 16...23 24...31  32...39 40...47 48...55 56...63
155 
156     MS            x[2]           LS  MS            x[3]		  LS
157     ms   ls ms   ls ms   ls ms   ls  ms   ls ms   ls ms   ls ms   ls
158     64...71 72...79 80...87 88...95  96..103 104.111 112.119 120.127
159 */
160 
161 /*	A slow generic version of gf_mul, implemented for lle and bbe
162  * 	It multiplies a and b and puts the result in a */
163 void gf128mul_lle(be128 *a, const be128 *b);
164 
165 void gf128mul_bbe(be128 *a, const be128 *b);
166 
167 /*
168  * The following functions multiply a field element by x in
169  * the polynomial field representation.  They use 64-bit word operations
170  * to gain speed but compensate for machine endianness and hence work
171  * correctly on both styles of machine.
172  *
173  * They are defined here for performance.
174  */
175 
176 static inline u64 gf128mul_mask_from_bit(u64 x, int which)
177 {
178 	/* a constant-time version of 'x & ((u64)1 << which) ? (u64)-1 : 0' */
179 	return ((s64)(x << (63 - which)) >> 63);
180 }
181 
182 static inline void gf128mul_x_lle(be128 *r, const be128 *x)
183 {
184 	u64 a = be64_to_cpu(x->a);
185 	u64 b = be64_to_cpu(x->b);
186 
187 	/* equivalent to gf128mul_table_le[(b << 7) & 0xff] << 48
188 	 * (see crypto/gf128mul.c): */
189 	u64 _tt = gf128mul_mask_from_bit(b, 0) & ((u64)0xe1 << 56);
190 
191 	r->b = cpu_to_be64((b >> 1) | (a << 63));
192 	r->a = cpu_to_be64((a >> 1) ^ _tt);
193 }
194 
195 static inline void gf128mul_x_bbe(be128 *r, const be128 *x)
196 {
197 	u64 a = be64_to_cpu(x->a);
198 	u64 b = be64_to_cpu(x->b);
199 
200 	/* equivalent to gf128mul_table_be[a >> 63] (see crypto/gf128mul.c): */
201 	u64 _tt = gf128mul_mask_from_bit(a, 63) & 0x87;
202 
203 	r->a = cpu_to_be64((a << 1) | (b >> 63));
204 	r->b = cpu_to_be64((b << 1) ^ _tt);
205 }
206 
207 /* needed by XTS */
208 static inline void gf128mul_x_ble(le128 *r, const le128 *x)
209 {
210 	u64 a = le64_to_cpu(x->a);
211 	u64 b = le64_to_cpu(x->b);
212 
213 	/* equivalent to gf128mul_table_be[b >> 63] (see crypto/gf128mul.c): */
214 	u64 _tt = gf128mul_mask_from_bit(a, 63) & 0x87;
215 
216 	r->a = cpu_to_le64((a << 1) | (b >> 63));
217 	r->b = cpu_to_le64((b << 1) ^ _tt);
218 }
219 
220 /* 4k table optimization */
221 
222 struct gf128mul_4k {
223 	be128 t[256];
224 };
225 
226 struct gf128mul_4k *gf128mul_init_4k_lle(const be128 *g);
227 struct gf128mul_4k *gf128mul_init_4k_bbe(const be128 *g);
228 void gf128mul_4k_lle(be128 *a, const struct gf128mul_4k *t);
229 void gf128mul_4k_bbe(be128 *a, const struct gf128mul_4k *t);
230 void gf128mul_x8_ble(le128 *r, const le128 *x);
231 static inline void gf128mul_free_4k(struct gf128mul_4k *t)
232 {
233 	kfree_sensitive(t);
234 }
235 
236 
237 /* 64k table optimization, implemented for bbe */
238 
239 struct gf128mul_64k {
240 	struct gf128mul_4k *t[16];
241 };
242 
243 /* First initialize with the constant factor with which you
244  * want to multiply and then call gf128mul_64k_bbe with the other
245  * factor in the first argument, and the table in the second.
246  * Afterwards, the result is stored in *a.
247  */
248 struct gf128mul_64k *gf128mul_init_64k_bbe(const be128 *g);
249 void gf128mul_free_64k(struct gf128mul_64k *t);
250 void gf128mul_64k_bbe(be128 *a, const struct gf128mul_64k *t);
251 
252 #endif /* _CRYPTO_GF128MUL_H */
253