1########################################################################
2# Implement fast SHA-512 with SSSE3 instructions. (x86_64)
3#
4# Copyright (C) 2013 Intel Corporation.
5#
6# Authors:
7#     James Guilford <james.guilford@intel.com>
8#     Kirk Yap <kirk.s.yap@intel.com>
9#     David Cote <david.m.cote@intel.com>
10#     Tim Chen <tim.c.chen@linux.intel.com>
11#
12# This software is available to you under a choice of one of two
13# licenses.  You may choose to be licensed under the terms of the GNU
14# General Public License (GPL) Version 2, available from the file
15# COPYING in the main directory of this source tree, or the
16# OpenIB.org BSD license below:
17#
18#     Redistribution and use in source and binary forms, with or
19#     without modification, are permitted provided that the following
20#     conditions are met:
21#
22#      - Redistributions of source code must retain the above
23#        copyright notice, this list of conditions and the following
24#        disclaimer.
25#
26#      - Redistributions in binary form must reproduce the above
27#        copyright notice, this list of conditions and the following
28#        disclaimer in the documentation and/or other materials
29#        provided with the distribution.
30#
31# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
32# EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
33# MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
34# NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
35# BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
36# ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
37# CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
38# SOFTWARE.
39#
40########################################################################
41#
42# This code is described in an Intel White-Paper:
43# "Fast SHA-512 Implementations on Intel Architecture Processors"
44#
45# To find it, surf to http://www.intel.com/p/en_US/embedded
46# and search for that title.
47#
48########################################################################
49
50#include <linux/linkage.h>
51
52.text
53
54# Virtual Registers
55# ARG1
56msg =		%rdi
57# ARG2
58digest =	%rsi
59# ARG3
60msglen =	%rdx
61T1 =		%rcx
62T2 =		%r8
63a_64 =		%r9
64b_64 =		%r10
65c_64 =		%r11
66d_64 =		%r12
67e_64 =		%r13
68f_64 =		%r14
69g_64 =		%r15
70h_64 =		%rbx
71tmp0 =		%rax
72
73# Local variables (stack frame)
74
75W_SIZE = 80*8
76WK_SIZE = 2*8
77RSPSAVE_SIZE = 1*8
78GPRSAVE_SIZE = 5*8
79
80frame_W = 0
81frame_WK = frame_W + W_SIZE
82frame_RSPSAVE = frame_WK + WK_SIZE
83frame_GPRSAVE = frame_RSPSAVE + RSPSAVE_SIZE
84frame_size = frame_GPRSAVE + GPRSAVE_SIZE
85
86# Useful QWORD "arrays" for simpler memory references
87# MSG, DIGEST, K_t, W_t are arrays
88# WK_2(t) points to 1 of 2 qwords at frame.WK depdending on t being odd/even
89
90# Input message (arg1)
91#define MSG(i)    8*i(msg)
92
93# Output Digest (arg2)
94#define DIGEST(i) 8*i(digest)
95
96# SHA Constants (static mem)
97#define K_t(i)    8*i+K512(%rip)
98
99# Message Schedule (stack frame)
100#define W_t(i)    8*i+frame_W(%rsp)
101
102# W[t]+K[t] (stack frame)
103#define WK_2(i)   8*((i%2))+frame_WK(%rsp)
104
105.macro RotateState
106	# Rotate symbols a..h right
107	TMP   = h_64
108	h_64  = g_64
109	g_64  = f_64
110	f_64  = e_64
111	e_64  = d_64
112	d_64  = c_64
113	c_64  = b_64
114	b_64  = a_64
115	a_64  = TMP
116.endm
117
118.macro SHA512_Round rnd
119
120	# Compute Round %%t
121	mov	f_64, T1          # T1 = f
122	mov	e_64, tmp0        # tmp = e
123	xor	g_64, T1          # T1 = f ^ g
124	ror	$23, tmp0 # 41    # tmp = e ror 23
125	and	e_64, T1          # T1 = (f ^ g) & e
126	xor	e_64, tmp0        # tmp = (e ror 23) ^ e
127	xor	g_64, T1          # T1 = ((f ^ g) & e) ^ g = CH(e,f,g)
128	idx = \rnd
129	add	WK_2(idx), T1     # W[t] + K[t] from message scheduler
130	ror	$4, tmp0  # 18    # tmp = ((e ror 23) ^ e) ror 4
131	xor	e_64, tmp0        # tmp = (((e ror 23) ^ e) ror 4) ^ e
132	mov	a_64, T2          # T2 = a
133	add	h_64, T1          # T1 = CH(e,f,g) + W[t] + K[t] + h
134	ror	$14, tmp0 # 14    # tmp = ((((e ror23)^e)ror4)^e)ror14 = S1(e)
135	add	tmp0, T1          # T1 = CH(e,f,g) + W[t] + K[t] + S1(e)
136	mov	a_64, tmp0        # tmp = a
137	xor	c_64, T2          # T2 = a ^ c
138	and	c_64, tmp0        # tmp = a & c
139	and	b_64, T2          # T2 = (a ^ c) & b
140	xor	tmp0, T2          # T2 = ((a ^ c) & b) ^ (a & c) = Maj(a,b,c)
141	mov	a_64, tmp0        # tmp = a
142	ror	$5, tmp0 # 39     # tmp = a ror 5
143	xor	a_64, tmp0        # tmp = (a ror 5) ^ a
144	add	T1, d_64          # e(next_state) = d + T1
145	ror	$6, tmp0 # 34     # tmp = ((a ror 5) ^ a) ror 6
146	xor	a_64, tmp0        # tmp = (((a ror 5) ^ a) ror 6) ^ a
147	lea	(T1, T2), h_64    # a(next_state) = T1 + Maj(a,b,c)
148	ror	$28, tmp0 # 28    # tmp = ((((a ror5)^a)ror6)^a)ror28 = S0(a)
149	add	tmp0, h_64        # a(next_state) = T1 + Maj(a,b,c) S0(a)
150	RotateState
151.endm
152
153.macro SHA512_2Sched_2Round_sse rnd
154
155	# Compute rounds t-2 and t-1
156	# Compute message schedule QWORDS t and t+1
157
158	#   Two rounds are computed based on the values for K[t-2]+W[t-2] and
159	# K[t-1]+W[t-1] which were previously stored at WK_2 by the message
160	# scheduler.
161	#   The two new schedule QWORDS are stored at [W_t(%%t)] and [W_t(%%t+1)].
162	# They are then added to their respective SHA512 constants at
163	# [K_t(%%t)] and [K_t(%%t+1)] and stored at dqword [WK_2(%%t)]
164	#   For brievity, the comments following vectored instructions only refer to
165	# the first of a pair of QWORDS.
166	# Eg. XMM2=W[t-2] really means XMM2={W[t-2]|W[t-1]}
167	#   The computation of the message schedule and the rounds are tightly
168	# stitched to take advantage of instruction-level parallelism.
169	# For clarity, integer instructions (for the rounds calculation) are indented
170	# by one tab. Vectored instructions (for the message scheduler) are indented
171	# by two tabs.
172
173	mov	f_64, T1
174	idx = \rnd -2
175	movdqa	W_t(idx), %xmm2		    # XMM2 = W[t-2]
176	xor	g_64, T1
177	and	e_64, T1
178	movdqa	%xmm2, %xmm0	            # XMM0 = W[t-2]
179	xor	g_64, T1
180	idx = \rnd
181	add	WK_2(idx), T1
182	idx = \rnd - 15
183	movdqu	W_t(idx), %xmm5		    # XMM5 = W[t-15]
184	mov	e_64, tmp0
185	ror	$23, tmp0 # 41
186	movdqa	%xmm5, %xmm3	            # XMM3 = W[t-15]
187	xor	e_64, tmp0
188	ror	$4, tmp0 # 18
189	psrlq	$61-19, %xmm0		    # XMM0 = W[t-2] >> 42
190	xor	e_64, tmp0
191	ror	$14, tmp0 # 14
192	psrlq	$(8-7), %xmm3		    # XMM3 = W[t-15] >> 1
193	add	tmp0, T1
194	add	h_64, T1
195	pxor	%xmm2, %xmm0                # XMM0 = (W[t-2] >> 42) ^ W[t-2]
196	mov	a_64, T2
197	xor	c_64, T2
198	pxor	%xmm5, %xmm3                # XMM3 = (W[t-15] >> 1) ^ W[t-15]
199	and	b_64, T2
200	mov	a_64, tmp0
201	psrlq	$(19-6), %xmm0		    # XMM0 = ((W[t-2]>>42)^W[t-2])>>13
202	and	c_64, tmp0
203	xor	tmp0, T2
204	psrlq	$(7-1), %xmm3		    # XMM3 = ((W[t-15]>>1)^W[t-15])>>6
205	mov	a_64, tmp0
206	ror	$5, tmp0 # 39
207	pxor	%xmm2, %xmm0	            # XMM0 = (((W[t-2]>>42)^W[t-2])>>13)^W[t-2]
208	xor	a_64, tmp0
209	ror	$6, tmp0 # 34
210	pxor	%xmm5, %xmm3                # XMM3 = (((W[t-15]>>1)^W[t-15])>>6)^W[t-15]
211	xor	a_64, tmp0
212	ror	$28, tmp0 # 28
213	psrlq	$6, %xmm0                   # XMM0 = ((((W[t-2]>>42)^W[t-2])>>13)^W[t-2])>>6
214	add	tmp0, T2
215	add	T1, d_64
216	psrlq	$1, %xmm3                   # XMM3 = (((W[t-15]>>1)^W[t-15])>>6)^W[t-15]>>1
217	lea	(T1, T2), h_64
218	RotateState
219	movdqa	%xmm2, %xmm1	            # XMM1 = W[t-2]
220	mov	f_64, T1
221	xor	g_64, T1
222	movdqa	%xmm5, %xmm4		    # XMM4 = W[t-15]
223	and	e_64, T1
224	xor	g_64, T1
225	psllq	$(64-19)-(64-61) , %xmm1    # XMM1 = W[t-2] << 42
226	idx = \rnd + 1
227	add	WK_2(idx), T1
228	mov	e_64, tmp0
229	psllq	$(64-1)-(64-8), %xmm4	    # XMM4 = W[t-15] << 7
230	ror	$23, tmp0 # 41
231	xor	e_64, tmp0
232	pxor	%xmm2, %xmm1		    # XMM1 = (W[t-2] << 42)^W[t-2]
233	ror	$4, tmp0 # 18
234	xor	e_64, tmp0
235	pxor	%xmm5, %xmm4		    # XMM4 = (W[t-15]<<7)^W[t-15]
236	ror	$14, tmp0 # 14
237	add	tmp0, T1
238	psllq	$(64-61), %xmm1		    # XMM1 = ((W[t-2] << 42)^W[t-2])<<3
239	add	h_64, T1
240	mov	a_64, T2
241	psllq	$(64-8), %xmm4		    # XMM4 = ((W[t-15]<<7)^W[t-15])<<56
242	xor	c_64, T2
243	and	b_64, T2
244	pxor	%xmm1, %xmm0		    # XMM0 = s1(W[t-2])
245	mov	a_64, tmp0
246	and	c_64, tmp0
247	idx = \rnd - 7
248	movdqu	W_t(idx), %xmm1		    # XMM1 = W[t-7]
249	xor	tmp0, T2
250	pxor	%xmm4, %xmm3                # XMM3 = s0(W[t-15])
251	mov	a_64, tmp0
252	paddq	%xmm3, %xmm0		    # XMM0 = s1(W[t-2]) + s0(W[t-15])
253	ror	$5, tmp0 # 39
254	idx =\rnd-16
255	paddq	W_t(idx), %xmm0		    # XMM0 = s1(W[t-2]) + s0(W[t-15]) + W[t-16]
256	xor	a_64, tmp0
257	paddq	%xmm1, %xmm0	            # XMM0 = s1(W[t-2]) + W[t-7] + s0(W[t-15]) + W[t-16]
258	ror	$6, tmp0 # 34
259	movdqa	%xmm0, W_t(\rnd)	    # Store scheduled qwords
260	xor	a_64, tmp0
261	paddq	K_t(\rnd), %xmm0	    # Compute W[t]+K[t]
262	ror	$28, tmp0 # 28
263	idx = \rnd
264	movdqa	%xmm0, WK_2(idx)	    # Store W[t]+K[t] for next rounds
265	add	tmp0, T2
266	add	T1, d_64
267	lea	(T1, T2), h_64
268	RotateState
269.endm
270
271########################################################################
272# void sha512_transform_ssse3(const void* M, void* D, u64 L)#
273# Purpose: Updates the SHA512 digest stored at D with the message stored in M.
274# The size of the message pointed to by M must be an integer multiple of SHA512
275#   message blocks.
276# L is the message length in SHA512 blocks.
277########################################################################
278ENTRY(sha512_transform_ssse3)
279
280	cmp $0, msglen
281	je nowork
282
283	# Allocate Stack Space
284	mov	%rsp, %rax
285	sub	$frame_size, %rsp
286	and	$~(0x20 - 1), %rsp
287	mov	%rax, frame_RSPSAVE(%rsp)
288
289	# Save GPRs
290	mov	%rbx, frame_GPRSAVE(%rsp)
291	mov	%r12, frame_GPRSAVE +8*1(%rsp)
292	mov	%r13, frame_GPRSAVE +8*2(%rsp)
293	mov	%r14, frame_GPRSAVE +8*3(%rsp)
294	mov	%r15, frame_GPRSAVE +8*4(%rsp)
295
296updateblock:
297
298# Load state variables
299	mov	DIGEST(0), a_64
300	mov	DIGEST(1), b_64
301	mov	DIGEST(2), c_64
302	mov	DIGEST(3), d_64
303	mov	DIGEST(4), e_64
304	mov	DIGEST(5), f_64
305	mov	DIGEST(6), g_64
306	mov	DIGEST(7), h_64
307
308	t = 0
309	.rept 80/2 + 1
310	# (80 rounds) / (2 rounds/iteration) + (1 iteration)
311	# +1 iteration because the scheduler leads hashing by 1 iteration
312		.if t < 2
313			# BSWAP 2 QWORDS
314			movdqa	XMM_QWORD_BSWAP(%rip), %xmm1
315			movdqu	MSG(t), %xmm0
316			pshufb	%xmm1, %xmm0	# BSWAP
317			movdqa	%xmm0, W_t(t)	# Store Scheduled Pair
318			paddq	K_t(t), %xmm0	# Compute W[t]+K[t]
319			movdqa	%xmm0, WK_2(t)	# Store into WK for rounds
320		.elseif t < 16
321			# BSWAP 2 QWORDS# Compute 2 Rounds
322			movdqu	MSG(t), %xmm0
323			pshufb	%xmm1, %xmm0	# BSWAP
324			SHA512_Round t-2	# Round t-2
325			movdqa	%xmm0, W_t(t)	# Store Scheduled Pair
326			paddq	K_t(t), %xmm0	# Compute W[t]+K[t]
327			SHA512_Round t-1	# Round t-1
328			movdqa	%xmm0, WK_2(t)	# Store W[t]+K[t] into WK
329		.elseif t < 79
330			# Schedule 2 QWORDS# Compute 2 Rounds
331			SHA512_2Sched_2Round_sse t
332		.else
333			# Compute 2 Rounds
334			SHA512_Round t-2
335			SHA512_Round t-1
336		.endif
337		t = t+2
338	.endr
339
340	# Update digest
341	add	a_64, DIGEST(0)
342	add	b_64, DIGEST(1)
343	add	c_64, DIGEST(2)
344	add	d_64, DIGEST(3)
345	add	e_64, DIGEST(4)
346	add	f_64, DIGEST(5)
347	add	g_64, DIGEST(6)
348	add	h_64, DIGEST(7)
349
350	# Advance to next message block
351	add	$16*8, msg
352	dec	msglen
353	jnz	updateblock
354
355	# Restore GPRs
356	mov	frame_GPRSAVE(%rsp),      %rbx
357	mov	frame_GPRSAVE +8*1(%rsp), %r12
358	mov	frame_GPRSAVE +8*2(%rsp), %r13
359	mov	frame_GPRSAVE +8*3(%rsp), %r14
360	mov	frame_GPRSAVE +8*4(%rsp), %r15
361
362	# Restore Stack Pointer
363	mov	frame_RSPSAVE(%rsp), %rsp
364
365nowork:
366	ret
367ENDPROC(sha512_transform_ssse3)
368
369########################################################################
370### Binary Data
371
372.data
373
374.align 16
375
376# Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
377XMM_QWORD_BSWAP:
378	.octa 0x08090a0b0c0d0e0f0001020304050607
379
380# K[t] used in SHA512 hashing
381K512:
382	.quad 0x428a2f98d728ae22,0x7137449123ef65cd
383	.quad 0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
384	.quad 0x3956c25bf348b538,0x59f111f1b605d019
385	.quad 0x923f82a4af194f9b,0xab1c5ed5da6d8118
386	.quad 0xd807aa98a3030242,0x12835b0145706fbe
387	.quad 0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
388	.quad 0x72be5d74f27b896f,0x80deb1fe3b1696b1
389	.quad 0x9bdc06a725c71235,0xc19bf174cf692694
390	.quad 0xe49b69c19ef14ad2,0xefbe4786384f25e3
391	.quad 0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
392	.quad 0x2de92c6f592b0275,0x4a7484aa6ea6e483
393	.quad 0x5cb0a9dcbd41fbd4,0x76f988da831153b5
394	.quad 0x983e5152ee66dfab,0xa831c66d2db43210
395	.quad 0xb00327c898fb213f,0xbf597fc7beef0ee4
396	.quad 0xc6e00bf33da88fc2,0xd5a79147930aa725
397	.quad 0x06ca6351e003826f,0x142929670a0e6e70
398	.quad 0x27b70a8546d22ffc,0x2e1b21385c26c926
399	.quad 0x4d2c6dfc5ac42aed,0x53380d139d95b3df
400	.quad 0x650a73548baf63de,0x766a0abb3c77b2a8
401	.quad 0x81c2c92e47edaee6,0x92722c851482353b
402	.quad 0xa2bfe8a14cf10364,0xa81a664bbc423001
403	.quad 0xc24b8b70d0f89791,0xc76c51a30654be30
404	.quad 0xd192e819d6ef5218,0xd69906245565a910
405	.quad 0xf40e35855771202a,0x106aa07032bbd1b8
406	.quad 0x19a4c116b8d2d0c8,0x1e376c085141ab53
407	.quad 0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
408	.quad 0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
409	.quad 0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
410	.quad 0x748f82ee5defb2fc,0x78a5636f43172f60
411	.quad 0x84c87814a1f0ab72,0x8cc702081a6439ec
412	.quad 0x90befffa23631e28,0xa4506cebde82bde9
413	.quad 0xbef9a3f7b2c67915,0xc67178f2e372532b
414	.quad 0xca273eceea26619c,0xd186b8c721c0c207
415	.quad 0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
416	.quad 0x06f067aa72176fba,0x0a637dc5a2c898a6
417	.quad 0x113f9804bef90dae,0x1b710b35131c471b
418	.quad 0x28db77f523047d84,0x32caab7b40c72493
419	.quad 0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
420	.quad 0x4cc5d4becb3e42b6,0x597f299cfc657e2a
421	.quad 0x5fcb6fab3ad6faec,0x6c44198c4a475817
422