1 // -------------------------------------------------------------------------
2 // Copyright (c) 2001, Dr Brian Gladman < >, Worcester, UK.
3 // All rights reserved.
7 // The free distribution and use of this software in both source and binary
8 // form is allowed (with or without changes) provided that:
10 // 1. distributions of this source code include the above copyright
11 // notice, this list of conditions and the following disclaimer//
13 // 2. distributions in binary form include the above copyright
14 // notice, this list of conditions and the following disclaimer
15 // in the documentation and/or other associated materials//
17 // 3. the copyright holder's name is not used to endorse products
18 // built using this software without specific written permission.
21 // ALTERNATIVELY, provided that this notice is retained in full, this product
22 // may be distributed under the terms of the GNU General Public License (GPL),
23 // in which case the provisions of the GPL apply INSTEAD OF those given above.
25 // Copyright (c) 2004 Linus Torvalds <torvalds@osdl.org>
26 // Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
30 // This software is provided 'as is' with no explicit or implied warranties
31 // in respect of its properties including, but not limited to, correctness
32 // and fitness for purpose.
33 // -------------------------------------------------------------------------
34 // Issue Date: 29/07/2002
36 .file "aes-i586-asm.S"
39 #include <asm/asm-offsets.h>
41 #define tlen 1024 // length of each of 4 'xor' arrays (256 32-bit words)
43 /* offsets to parameters with one register pushed onto stack */
48 /* offsets in crypto_tfm structure */
49 #define klen (crypto_tfm_ctx_offset + 0)
50 #define ekey (crypto_tfm_ctx_offset + 4)
51 #define dkey (crypto_tfm_ctx_offset + 244)
53 // register mapping for encrypt and decrypt subroutines
71 #define _h(reg) reg##h
72 #define h(reg) _h(reg)
74 #define _l(reg) reg##l
75 #define l(reg) _l(reg)
77 // This macro takes a 32-bit word representing a column and uses
78 // each of its four bytes to index into four tables of 256 32-bit
79 // words to obtain values that are then xored into the appropriate
80 // output registers r0, r1, r4 or r5.
83 // table table base address
88 // idx input register for the round (destroyed)
89 // tmp scratch register for the round
92 #define do_col(table, a1,a2,a3,a4, idx, tmp) \
94 xor table(,%tmp,4),%a1; \
97 xor table+tlen(,%tmp,4),%a2; \
100 xor table+2*tlen(,%tmp,4),%a3; \
101 xor table+3*tlen(,%idx,4),%a4;
103 // initialise output registers from the key schedule
104 // NB1: original value of a3 is in idx on exit
105 // NB2: original values of a1,a2,a4 aren't used
106 #define do_fcol(table, a1,a2,a3,a4, idx, tmp, sched) \
108 movzx %l(idx),%tmp; \
110 xor table(,%tmp,4),%a1; \
112 movzx %h(idx),%tmp; \
114 xor table+tlen(,%tmp,4),%a2; \
115 movzx %l(idx),%tmp; \
116 movzx %h(idx),%idx; \
117 xor table+3*tlen(,%idx,4),%a4; \
120 xor table+2*tlen(,%tmp,4),%a3;
122 // initialise output registers from the key schedule
123 // NB1: original value of a3 is in idx on exit
124 // NB2: original values of a1,a2,a4 aren't used
125 #define do_icol(table, a1,a2,a3,a4, idx, tmp, sched) \
127 movzx %l(idx),%tmp; \
129 xor table(,%tmp,4),%a1; \
131 movzx %h(idx),%tmp; \
133 xor table+tlen(,%tmp,4),%a2; \
134 movzx %l(idx),%tmp; \
135 movzx %h(idx),%idx; \
136 xor table+3*tlen(,%idx,4),%a4; \
139 xor table+2*tlen(,%tmp,4),%a3;
142 // original Gladman had conditional saves to MMX regs.
143 #define save(a1, a2) \
146 #define restore(a1, a2) \
149 // These macros perform a forward encryption cycle. They are entered with
150 // the first previous round column values in r0,r1,r4,r5 and
151 // exit with the final values in the same registers, using stack
152 // for temporary storage.
154 // round column values
155 // on entry: r0,r1,r4,r5
156 // on exit: r2,r1,r4,r5
157 #define fwd_rnd1(arg, table) \
161 /* compute new column values */ \
162 do_fcol(table, r2,r5,r4,r1, r0,r3, arg); /* idx=r0 */ \
163 do_col (table, r4,r1,r2,r5, r0,r3); /* idx=r4 */ \
165 do_col (table, r1,r2,r5,r4, r0,r3); /* idx=r1 */ \
167 do_col (table, r5,r4,r1,r2, r0,r3); /* idx=r5 */
169 // round column values
170 // on entry: r2,r1,r4,r5
171 // on exit: r0,r1,r4,r5
172 #define fwd_rnd2(arg, table) \
176 /* compute new column values */ \
177 do_fcol(table, r0,r5,r4,r1, r2,r3, arg); /* idx=r2 */ \
178 do_col (table, r4,r1,r0,r5, r2,r3); /* idx=r4 */ \
180 do_col (table, r1,r0,r5,r4, r2,r3); /* idx=r1 */ \
182 do_col (table, r5,r4,r1,r0, r2,r3); /* idx=r5 */
184 // These macros performs an inverse encryption cycle. They are entered with
185 // the first previous round column values in r0,r1,r4,r5 and
186 // exit with the final values in the same registers, using stack
187 // for temporary storage
189 // round column values
190 // on entry: r0,r1,r4,r5
191 // on exit: r2,r1,r4,r5
192 #define inv_rnd1(arg, table) \
196 /* compute new column values */ \
197 do_icol(table, r2,r1,r4,r5, r0,r3, arg); /* idx=r0 */ \
198 do_col (table, r4,r5,r2,r1, r0,r3); /* idx=r4 */ \
200 do_col (table, r1,r4,r5,r2, r0,r3); /* idx=r1 */ \
202 do_col (table, r5,r2,r1,r4, r0,r3); /* idx=r5 */
204 // round column values
205 // on entry: r2,r1,r4,r5
206 // on exit: r0,r1,r4,r5
207 #define inv_rnd2(arg, table) \
211 /* compute new column values */ \
212 do_icol(table, r0,r1,r4,r5, r2,r3, arg); /* idx=r2 */ \
213 do_col (table, r4,r5,r0,r1, r2,r3); /* idx=r4 */ \
215 do_col (table, r1,r4,r5,r0, r2,r3); /* idx=r1 */ \
217 do_col (table, r5,r0,r1,r4, r2,r3); /* idx=r5 */
219 // AES (Rijndael) Encryption Subroutine
220 /* void aes_enc_blk(struct crypto_tfm *tfm, u8 *out_blk, const u8 *in_blk) */
224 .extern crypto_ft_tab
225 .extern crypto_fl_tab
233 // CAUTION: the order and the values used in these assigns
234 // rely on the register mappings
237 mov in_blk+4(%esp),%r2
239 mov klen(%ebp),%r3 // key size
242 lea ekey(%ebp),%ebp // key pointer
245 // input four columns and xor in first round key
256 sub $8,%esp // space for register saves on stack
257 add $16,%ebp // increment to next round key
259 jb 4f // 10 rounds for 128-bit key
261 je 3f // 12 rounds for 192-bit key
264 2: fwd_rnd1( -64(%ebp), crypto_ft_tab) // 14 rounds for 256-bit key
265 fwd_rnd2( -48(%ebp), crypto_ft_tab)
266 3: fwd_rnd1( -32(%ebp), crypto_ft_tab) // 12 rounds for 192-bit key
267 fwd_rnd2( -16(%ebp), crypto_ft_tab)
268 4: fwd_rnd1( (%ebp), crypto_ft_tab) // 10 rounds for 128-bit key
269 fwd_rnd2( +16(%ebp), crypto_ft_tab)
270 fwd_rnd1( +32(%ebp), crypto_ft_tab)
271 fwd_rnd2( +48(%ebp), crypto_ft_tab)
272 fwd_rnd1( +64(%ebp), crypto_ft_tab)
273 fwd_rnd2( +80(%ebp), crypto_ft_tab)
274 fwd_rnd1( +96(%ebp), crypto_ft_tab)
275 fwd_rnd2(+112(%ebp), crypto_ft_tab)
276 fwd_rnd1(+128(%ebp), crypto_ft_tab)
277 fwd_rnd2(+144(%ebp), crypto_fl_tab) // last round uses a different table
279 // move final values to the output array. CAUTION: the
280 // order of these assigns rely on the register mappings
283 mov out_blk+12(%esp),%ebp
294 // AES (Rijndael) Decryption Subroutine
295 /* void aes_dec_blk(struct crypto_tfm *tfm, u8 *out_blk, const u8 *in_blk) */
299 .extern crypto_it_tab
300 .extern crypto_il_tab
308 // CAUTION: the order and the values used in these assigns
309 // rely on the register mappings
312 mov in_blk+4(%esp),%r2
314 mov klen(%ebp),%r3 // key size
317 lea dkey(%ebp),%ebp // key pointer
320 // input four columns and xor in first round key
331 sub $8,%esp // space for register saves on stack
332 add $16,%ebp // increment to next round key
334 jb 4f // 10 rounds for 128-bit key
336 je 3f // 12 rounds for 192-bit key
339 2: inv_rnd1( -64(%ebp), crypto_it_tab) // 14 rounds for 256-bit key
340 inv_rnd2( -48(%ebp), crypto_it_tab)
341 3: inv_rnd1( -32(%ebp), crypto_it_tab) // 12 rounds for 192-bit key
342 inv_rnd2( -16(%ebp), crypto_it_tab)
343 4: inv_rnd1( (%ebp), crypto_it_tab) // 10 rounds for 128-bit key
344 inv_rnd2( +16(%ebp), crypto_it_tab)
345 inv_rnd1( +32(%ebp), crypto_it_tab)
346 inv_rnd2( +48(%ebp), crypto_it_tab)
347 inv_rnd1( +64(%ebp), crypto_it_tab)
348 inv_rnd2( +80(%ebp), crypto_it_tab)
349 inv_rnd1( +96(%ebp), crypto_it_tab)
350 inv_rnd2(+112(%ebp), crypto_it_tab)
351 inv_rnd1(+128(%ebp), crypto_it_tab)
352 inv_rnd2(+144(%ebp), crypto_il_tab) // last round uses a different table
354 // move final values to the output array. CAUTION: the
355 // order of these assigns rely on the register mappings
358 mov out_blk+12(%esp),%ebp