1#!/usr/bin/env perl
2
3# ====================================================================
4# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
5# project. The module is, however, dual licensed under OpenSSL and
6# CRYPTOGAMS licenses depending on where you obtain it. For further
7# details see http://www.openssl.org/~appro/cryptogams/.
8# ====================================================================
9
10# September 2010.
11#
12# The module implements "4-bit" GCM GHASH function and underlying
13# single multiplication operation in GF(2^128). "4-bit" means that it
14# uses 256 bytes per-key table [+128 bytes shared table]. Performance
15# was measured to be ~18 cycles per processed byte on z10, which is
16# almost 40% better than gcc-generated code. It should be noted that
17# 18 cycles is worse result than expected: loop is scheduled for 12
18# and the result should be close to 12. In the lack of instruction-
19# level profiling data it's impossible to tell why...
20
21# November 2010.
22#
23# Adapt for -m31 build. If kernel supports what's called "highgprs"
24# feature on Linux [see /proc/cpuinfo], it's possible to use 64-bit
25# instructions and achieve "64-bit" performance even in 31-bit legacy
26# application context. The feature is not specific to any particular
27# processor, as long as it's "z-CPU". Latter implies that the code
28# remains z/Architecture specific. On z990 it was measured to perform
29# 2.8x better than 32-bit code generated by gcc 4.3.
30
31# March 2011.
32#
33# Support for hardware KIMD-GHASH is verified to produce correct
34# result and therefore is engaged. On z196 it was measured to process
35# 8KB buffer ~7 faster than software implementation. It's not as
36# impressive for smaller buffer sizes and for smallest 16-bytes buffer
37# it's actually almost 2 times slower. Which is the reason why
38# KIMD-GHASH is not used in gcm_gmult_4bit.
39
40$flavour = shift;
41
42if ($flavour =~ /3[12]/) {
43	$SIZE_T=4;
44	$g="";
45} else {
46	$SIZE_T=8;
47	$g="g";
48}
49
50while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
51open STDOUT,">$output";
52
53$softonly=0;
54
55$Zhi="%r0";
56$Zlo="%r1";
57
58$Xi="%r2";	# argument block
59$Htbl="%r3";
60$inp="%r4";
61$len="%r5";
62
63$rem0="%r6";	# variables
64$rem1="%r7";
65$nlo="%r8";
66$nhi="%r9";
67$xi="%r10";
68$cnt="%r11";
69$tmp="%r12";
70$x78="%r13";
71$rem_4bit="%r14";
72
73$sp="%r15";
74
75$code.=<<___;
76.text
77
78.globl	gcm_gmult_4bit
79.align	32
80gcm_gmult_4bit:
81___
82$code.=<<___ if(!$softonly && 0);	# hardware is slow for single block...
83	larl	%r1,OPENSSL_s390xcap_P
84	lg	%r0,0(%r1)
85	tmhl	%r0,0x4000	# check for message-security-assist
86	jz	.Lsoft_gmult
87	lghi	%r0,0
88	la	%r1,16($sp)
89	.long	0xb93e0004	# kimd %r0,%r4
90	lg	%r1,24($sp)
91	tmhh	%r1,0x4000	# check for function 65
92	jz	.Lsoft_gmult
93	stg	%r0,16($sp)	# arrange 16 bytes of zero input
94	stg	%r0,24($sp)
95	lghi	%r0,65		# function 65
96	la	%r1,0($Xi)	# H lies right after Xi in gcm128_context
97	la	$inp,16($sp)
98	lghi	$len,16
99	.long	0xb93e0004	# kimd %r0,$inp
100	brc	1,.-4		# pay attention to "partial completion"
101	br	%r14
102.align	32
103.Lsoft_gmult:
104___
105$code.=<<___;
106	stm${g}	%r6,%r14,6*$SIZE_T($sp)
107
108	aghi	$Xi,-1
109	lghi	$len,1
110	lghi	$x78,`0xf<<3`
111	larl	$rem_4bit,rem_4bit
112
113	lg	$Zlo,8+1($Xi)		# Xi
114	j	.Lgmult_shortcut
115.type	gcm_gmult_4bit,\@function
116.size	gcm_gmult_4bit,(.-gcm_gmult_4bit)
117
118.globl	gcm_ghash_4bit
119.align	32
120gcm_ghash_4bit:
121___
122$code.=<<___ if(!$softonly);
123	larl	%r1,OPENSSL_s390xcap_P
124	lg	%r0,0(%r1)
125	tmhl	%r0,0x4000	# check for message-security-assist
126	jz	.Lsoft_ghash
127	lghi	%r0,0
128	la	%r1,16($sp)
129	.long	0xb93e0004	# kimd %r0,%r4
130	lg	%r1,24($sp)
131	tmhh	%r1,0x4000	# check for function 65
132	jz	.Lsoft_ghash
133	lghi	%r0,65		# function 65
134	la	%r1,0($Xi)	# H lies right after Xi in gcm128_context
135	.long	0xb93e0004	# kimd %r0,$inp
136	brc	1,.-4		# pay attention to "partial completion"
137	br	%r14
138.align	32
139.Lsoft_ghash:
140___
141$code.=<<___ if ($flavour =~ /3[12]/);
142	llgfr	$len,$len
143___
144$code.=<<___;
145	stm${g}	%r6,%r14,6*$SIZE_T($sp)
146
147	aghi	$Xi,-1
148	srlg	$len,$len,4
149	lghi	$x78,`0xf<<3`
150	larl	$rem_4bit,rem_4bit
151
152	lg	$Zlo,8+1($Xi)		# Xi
153	lg	$Zhi,0+1($Xi)
154	lghi	$tmp,0
155.Louter:
156	xg	$Zhi,0($inp)		# Xi ^= inp
157	xg	$Zlo,8($inp)
158	xgr	$Zhi,$tmp
159	stg	$Zlo,8+1($Xi)
160	stg	$Zhi,0+1($Xi)
161
162.Lgmult_shortcut:
163	lghi	$tmp,0xf0
164	sllg	$nlo,$Zlo,4
165	srlg	$xi,$Zlo,8		# extract second byte
166	ngr	$nlo,$tmp
167	lgr	$nhi,$Zlo
168	lghi	$cnt,14
169	ngr	$nhi,$tmp
170
171	lg	$Zlo,8($nlo,$Htbl)
172	lg	$Zhi,0($nlo,$Htbl)
173
174	sllg	$nlo,$xi,4
175	sllg	$rem0,$Zlo,3
176	ngr	$nlo,$tmp
177	ngr	$rem0,$x78
178	ngr	$xi,$tmp
179
180	sllg	$tmp,$Zhi,60
181	srlg	$Zlo,$Zlo,4
182	srlg	$Zhi,$Zhi,4
183	xg	$Zlo,8($nhi,$Htbl)
184	xg	$Zhi,0($nhi,$Htbl)
185	lgr	$nhi,$xi
186	sllg	$rem1,$Zlo,3
187	xgr	$Zlo,$tmp
188	ngr	$rem1,$x78
189	j	.Lghash_inner
190.align	16
191.Lghash_inner:
192	srlg	$Zlo,$Zlo,4
193	sllg	$tmp,$Zhi,60
194	xg	$Zlo,8($nlo,$Htbl)
195	srlg	$Zhi,$Zhi,4
196	llgc	$xi,0($cnt,$Xi)
197	xg	$Zhi,0($nlo,$Htbl)
198	sllg	$nlo,$xi,4
199	xg	$Zhi,0($rem0,$rem_4bit)
200	nill	$nlo,0xf0
201	sllg	$rem0,$Zlo,3
202	xgr	$Zlo,$tmp
203	ngr	$rem0,$x78
204	nill	$xi,0xf0
205
206	sllg	$tmp,$Zhi,60
207	srlg	$Zlo,$Zlo,4
208	srlg	$Zhi,$Zhi,4
209	xg	$Zlo,8($nhi,$Htbl)
210	xg	$Zhi,0($nhi,$Htbl)
211	lgr	$nhi,$xi
212	xg	$Zhi,0($rem1,$rem_4bit)
213	sllg	$rem1,$Zlo,3
214	xgr	$Zlo,$tmp
215	ngr	$rem1,$x78
216	brct	$cnt,.Lghash_inner
217
218	sllg	$tmp,$Zhi,60
219	srlg	$Zlo,$Zlo,4
220	srlg	$Zhi,$Zhi,4
221	xg	$Zlo,8($nlo,$Htbl)
222	xg	$Zhi,0($nlo,$Htbl)
223	sllg	$xi,$Zlo,3
224	xg	$Zhi,0($rem0,$rem_4bit)
225	xgr	$Zlo,$tmp
226	ngr	$xi,$x78
227
228	sllg	$tmp,$Zhi,60
229	srlg	$Zlo,$Zlo,4
230	srlg	$Zhi,$Zhi,4
231	xg	$Zlo,8($nhi,$Htbl)
232	xg	$Zhi,0($nhi,$Htbl)
233	xgr	$Zlo,$tmp
234	xg	$Zhi,0($rem1,$rem_4bit)
235
236	lg	$tmp,0($xi,$rem_4bit)
237	la	$inp,16($inp)
238	sllg	$tmp,$tmp,4		# correct last rem_4bit[rem]
239	brctg	$len,.Louter
240
241	xgr	$Zhi,$tmp
242	stg	$Zlo,8+1($Xi)
243	stg	$Zhi,0+1($Xi)
244	lm${g}	%r6,%r14,6*$SIZE_T($sp)
245	br	%r14
246.type	gcm_ghash_4bit,\@function
247.size	gcm_ghash_4bit,(.-gcm_ghash_4bit)
248
249.align	64
250rem_4bit:
251	.long	`0x0000<<12`,0,`0x1C20<<12`,0,`0x3840<<12`,0,`0x2460<<12`,0
252	.long	`0x7080<<12`,0,`0x6CA0<<12`,0,`0x48C0<<12`,0,`0x54E0<<12`,0
253	.long	`0xE100<<12`,0,`0xFD20<<12`,0,`0xD940<<12`,0,`0xC560<<12`,0
254	.long	`0x9180<<12`,0,`0x8DA0<<12`,0,`0xA9C0<<12`,0,`0xB5E0<<12`,0
255.type	rem_4bit,\@object
256.size	rem_4bit,(.-rem_4bit)
257.string	"GHASH for s390x, CRYPTOGAMS by <appro\@openssl.org>"
258___
259
260$code =~ s/\`([^\`]*)\`/eval $1/gem;
261print $code;
262close STDOUT;
263