1 /// -*- mode: asm; asm-comment-char: ?/ -*-
3 /// ARM crypto-extension-based implementation of Rijndael
5 /// (c) 2016 Straylight/Edgeware
8 ///----- Licensing notice ---------------------------------------------------
10 /// This file is part of Catacomb.
12 /// Catacomb is free software; you can redistribute it and/or modify
13 /// it under the terms of the GNU Library General Public License as
14 /// published by the Free Software Foundation; either version 2 of the
15 /// License, or (at your option) any later version.
17 /// Catacomb is distributed in the hope that it will be useful,
18 /// but WITHOUT ANY WARRANTY; without even the implied warranty of
19 /// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
20 /// GNU Library General Public License for more details.
22 /// You should have received a copy of the GNU Library General Public
23 /// License along with Catacomb; if not, write to the Free
24 /// Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
25 /// MA 02111-1307, USA.
27 ///--------------------------------------------------------------------------
31 #include "asm-common.h"
34 .fpu crypto-neon-fp-armv8
37 .extern F(rijndael_rcon)
41 ///--------------------------------------------------------------------------
44 /// The ARM crypto extension implements a little-endian version of AES
45 /// (though the manual doesn't actually spell this out and you have to
46 /// experiment), but Catacomb's internal interface presents as big-endian so
47 /// as to work better with things like GCM. We therefore maintain the round
48 /// keys in little-endian form, and have to end-swap blocks in and out.
50 /// For added amusement, the crypto extension doesn't implement the larger-
51 /// block versions of Rijndael, so we have to end-swap the keys if we're
52 /// preparing for one of those.
55 .equ maxrounds, 16 // maximum number of rounds
56 .equ maxblksz, 32 // maximum block size, in bytes
57 .equ kbufsz, maxblksz*(maxrounds + 1) // size of key-sched buffer
60 .equ nr, 0 // number of rounds
61 .equ w, nr + 4 // encryption key words
62 .equ wi, w + kbufsz // decryption key words
64 ///--------------------------------------------------------------------------
67 FUNC(rijndael_setup_arm_crypto)
70 // r0 = pointer to context
71 // r1 = block size in words
72 // r2 = pointer to key material
73 // r3 = key size in words
77 // The initial round key material is taken directly from the input
78 // key, so copy it over. Unfortunately, the key material is not
79 // guaranteed to be aligned in any especially useful way, so we must
104 // Find out other useful things and prepare for the main loop.
105 9: ldr r7, [r0, #nr] // number of rounds
106 mla r2, r1, r7, r1 // total key size in words
107 leaextq r5, rijndael_rcon // round constants
108 sub r8, r2, r3 // minus what we've copied already
109 vmov.i32 q1, #0 // all-zero register for the key
110 add r8, r9, r8, lsl #2 // limit of the key buffer
111 mov r12, #0 // position in current cycle
113 // Main key expansion loop. Dispatch according to the position in
115 0: ldr r6, [r9, -r3, lsl #2] // word from previous cycle
116 cmp r12, #0 // first word of the cycle?
118 cmp r12, #4 // fourth word of the cycle?
120 cmp r3, #7 // seven or eight words of key?
123 // Fourth word of the cycle, seven or eight words of key. We must do
124 // the byte substitution.
126 aese.8 q0, q1 // effectively, just SubBytes
130 // First word of the cycle. Byte substitution, rotation, and round
132 1: ldrb r14, [r5], #1 // next round constant
134 aese.8 q0, q1 // effectively, just SubBytes
136 eor r4, r14, r4, ror #8
138 // Common ending: mix in the word from the previous cycle and store.
142 // Prepare for the next iteration. If we're done, then stop; if
143 // we've finished a cycle then reset the counter.
151 // Next job is to construct the decryption keys. The keys for the
152 // first and last rounds don't need to be mangled, but the remaining
153 // ones do -- and they all need to be reordered too.
155 // The plan of action, then, is to copy the final encryption round's
156 // keys into place first, then to do each of the intermediate rounds
157 // in reverse order, and finally do the first round.
159 // Do all the heavy lifting with NEON registers. The order we're
160 // doing this in means that it's OK if we read or write too much, and
161 // there's easily enough buffer space for the over-enthusiastic reads
162 // and writes because the context has space for 32-byte blocks, which
163 // is our maximum and an exact fit for two Q-class registers.
166 add r4, r4, r2, lsl #2
167 sub r4, r4, r1, lsl #2 // last round's keys
169 // Copy the last encryption round's keys.
171 vldmiaeq r4, {d0, d1}
173 vstmiaeq r5, {d0, d1}
176 // Update the loop variables and stop if we've finished.
177 0: sub r4, r4, r1, lsl #2
178 add r5, r5, r1, lsl #2
182 // Do another middle round's keys...
184 vldmiaeq r4, {d0, d1}
187 vstmiaeq r5, {d0, d1}
193 // Finally do the first encryption round.
195 vldmiaeq r4, {d0, d1}
197 vstmiaeq r5, {d0, d1}
200 // If the block size is not exactly four words then we must end-swap
201 // everything. We can use fancy NEON toys for this.
204 // End-swap the encryption keys.
208 // And the decryption keys
217 INTFUNC(endswap_block)
218 // End-swap R2 words starting at R1. R1 is clobbered; R2 is not.
219 // It's OK to work in 16-byte chunks.
222 0: vldmia r1, {d0, d1}
231 ///--------------------------------------------------------------------------
232 /// Encrypting and decrypting blocks.
234 .macro encdec op, aes, mc, koff
235 FUNC(rijndael_\op\()_arm_crypto)
238 // r0 = pointer to context
239 // r1 = pointer to input block
240 // r2 = pointer to output block
242 // Set things up ready.
248 // Check the number of rounds and dispatch.
251 addlo pc, pc, r3, lsl #2
261 11: vldmia r0!, {d16, d17}
267 12: vldmia r0!, {d16-d19}
275 13: vldmia r0!, {d16-d21}
284 // Fourteen rounds. (Drops through to the ten round case because
285 // this is the next most common.)
286 14: vldmia r0!, {d16-d23}
298 10: vldmia r0!, {d16-d25}
310 vldmia r0!, {d16-d27}
320 // Final round has no MixColumns, but is followed by final whitening.
332 encdec eblk, aese, aesmc, w
333 encdec dblk, aesd, aesimc, wi
335 ///----- That's all, folks --------------------------------------------------