arm/crypto/blake2s-core.S

1 /* SPDX-License-Identifier: GPL-2.0-or-later */
15 	// load the words on-demand.
21 	.word	0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A
22 	.word	0x510E527F, 0x9B05688C, 0x1F83D9AB, 0x5BE0CD19
24 .macro __ldrd		a, b, src, offset
26 	ldrd		\a, \b, [\src, #\offset]
28 	ldr		\a, [\src, #\offset]
33 .macro __strd		a, b, dst, offset
35 	strd		\a, \b, [\dst, #\offset]
37 	str		\a, [\dst, #\offset]
42 .macro _le32_bswap	a, tmp
44 	rev_l		\a, \tmp
48 .macro _le32_bswap_8x	a, b, c, d, e, f, g, h,  tmp
49 	_le32_bswap	\a, \tmp
52 	_le32_bswap	\d, \tmp
59 // Execute a quarter-round of BLAKE2s by mixing two columns or two diagonals.
61 // columns/diagonals.  s0-s1 are the word offsets to the message words the first
62 // column/diagonal needs, and likewise s2-s3 for the second column/diagonal.
73 	// a += b + m[blake2s_sigma[r][2*i + 0]];
79 	// d = ror32(d ^ a, 16);
83 	// c += d;
94 	// a += b + m[blake2s_sigma[r][2*i + 1]];
100 	// d = ror32(d ^ a, 8);
104 	// c += d;
113 // Execute one round of BLAKE2s by updating the state matrix v[0..15].  v[0..9]
115 // spilling v[8..9], then to v[9..15], then to the message block.  r10-r12 and
116 // r14 are free to use.  The macro arguments s0-s15 give the order in which the
122 // rows of the BLAKE2s state matrix (rows 'b' and 'd') to temporarily have the
126 // similarly for row 'd'.  (brot, drot) start out as (0, 0) but we make it such
132 	// (v[0], v[4], v[8], v[12]) and (v[1], v[5], v[9], v[13]).
136 	__strd		r8, r9, sp, 0
152 	// (v[0], v[5], v[10], v[15]) and (v[1], v[6], v[11], v[12]).
161 	// (v[2], v[7], v[8], v[13]) and (v[3], v[4], v[9], v[14]).
162 	__ldrd		r8, r9, sp, 0		// load v[8] and v[9]
180 	push		{r0-r2,r4-r11,lr}	// keep this an even number
187 	// Load and increment the counter t[0..1].
190 	adc		r11, r11, #0
194 	// to the stack to save a register during the rounds.  This also has the
200 	ldmia		r1!, {r2-r9}
202 	stmia		r12!, {r2-r9}
203 	ldmia		r1!, {r2-r9}
205 	stmia		r12, {r2-r9}
209 	// Calculate v[8..15].  Push v[9..15] onto the stack, and leave space
210 	// for spilling v[8..9].  Leave v[8..9] in r8-r9.
213 	ldmia		r12!, {r8-r9}		// load IV[0..1]
214 	__ldrd		r0, r1, r14, 40		// load f[0..1]
215 	ldm		r12, {r2-r7}		// load IV[3..7]
216 	eor		r4, r4, r10		// v[12] = IV[4] ^ t[0]
218 	eor		r6, r6, r0		// v[14] = IV[6] ^ f[0]
220 	push		{r2-r7}			// push v[9..15]
221 	sub		sp, sp, #8		// leave space for v[8..9]
223 	// Load h[0..7] == v[0..7].
224 	ldm		r14, {r0-r7}
228 	.set brot, 0
229 	.set drot, 0
230 	_blake2s_round	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
231 	_blake2s_round	14, 10, 4, 8, 9, 15, 13, 6, 1, 12, 0, 2, 11, 7, 5, 3
232 	_blake2s_round	11, 8, 12, 0, 5, 2, 15, 13, 10, 14, 3, 6, 7, 1, 9, 4
233 	_blake2s_round	7, 9, 3, 1, 13, 12, 11, 14, 2, 6, 5, 10, 4, 0, 15, 8
234 	_blake2s_round	9, 0, 5, 7, 2, 4, 10, 15, 14, 1, 11, 12, 6, 8, 3, 13
235 	_blake2s_round	2, 12, 6, 10, 0, 11, 8, 3, 4, 13, 7, 5, 15, 14, 1, 9
236 	_blake2s_round	12, 5, 1, 15, 14, 13, 4, 10, 0, 7, 6, 3, 9, 2, 8, 11
237 	_blake2s_round	13, 11, 7, 14, 12, 1, 3, 9, 5, 0, 15, 4, 8, 6, 2, 10
238 	_blake2s_round	6, 15, 14, 9, 11, 3, 0, 8, 12, 2, 13, 7, 1, 4, 10, 5
239 	_blake2s_round	10, 2, 8, 4, 7, 6, 1, 5, 15, 11, 9, 14, 3, 12, 13, 0
243 	//	for (i = 0; i < 8; i++)
246 	ldr		r14, [sp, #96]		// r14 = &h[0]
247 	add		sp, sp, #8		// v[8..9] are already loaded.
248 	pop		{r10-r11}		// load v[10..11]
253 	ldm		r14, {r8-r11}		// load h[0..3]
258 	stmia		r14!, {r0-r3}		// store new h[0..3]
259 	ldm		r14, {r0-r3}		// load old h[4..7]
260 	pop		{r8-r11}		// load v[12..15]
270 	stm		r14, {r0-r3}		// store new h[4..7]
274 	// 64.  So we can simply set it to 64 without re-loading it.
277 	subs		r2, r2, #1		// nblocks--
279 	bne		.Lnext_block		// nblocks != 0?
281 	pop		{r0-r2,r4-r11,pc}
283 	// The next message block (pointed to by r1) isn't 4-byte aligned, so it
285 	// by r12) using an alternative method.  r2-r9 are free to use.
293 	ldrb		r3, [r1, #0]