1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
|
#if defined(__SUNPRO_C) && defined(__sparcv9)
# define ABI64 /* They've said -xarch=v9 at command line */
#elif defined(__GNUC__) && defined(__arch64__)
# define ABI64 /* They've said -m64 at command line */
#endif
#ifdef ABI64
.register %g2,#scratch
.register %g3,#scratch
# define FRAME -192
# define BIAS 2047
#else
# define FRAME -96
# define BIAS 0
#endif
.text
.align 32
.global OPENSSL_wipe_cpu
.type OPENSSL_wipe_cpu,#function
! Keep in mind that this does not excuse us from wiping the stack!
! This routine wipes registers, but not the backing store [which
! resides on the stack, toward lower addresses]. To facilitate for
! stack wiping I return pointer to the top of stack of the *caller*.
OPENSSL_wipe_cpu:
save %sp,FRAME,%sp
nop
#ifdef __sun
#include <sys/trap.h>
ta ST_CLEAN_WINDOWS
#else
call .walk.reg.wins
#endif
nop
call .PIC.zero.up
mov .zero-(.-4),%o0
ld [%o0],%f0
ld [%o0],%f1
subcc %g0,1,%o0
! Following is V9 "rd %ccr,%o0" instruction. However! V8
! specification says that it ("rd %asr2,%o0" in V8 terms) does
! not cause illegal_instruction trap. It therefore can be used
! to determine if the CPU the code is executing on is V8- or
! V9-compliant, as V9 returns a distinct value of 0x99,
! "negative" and "borrow" bits set in both %icc and %xcc.
.word 0x91408000 !rd %ccr,%o0
cmp %o0,0x99
bne .v8
nop
! Even though we do not use %fp register bank,
! we wipe it as memcpy might have used it...
.word 0xbfa00040 !fmovd %f0,%f62
.word 0xbba00040 !...
.word 0xb7a00040
.word 0xb3a00040
.word 0xafa00040
.word 0xaba00040
.word 0xa7a00040
.word 0xa3a00040
.word 0x9fa00040
.word 0x9ba00040
.word 0x97a00040
.word 0x93a00040
.word 0x8fa00040
.word 0x8ba00040
.word 0x87a00040
.word 0x83a00040 !fmovd %f0,%f32
.v8: fmovs %f1,%f31
clr %o0
fmovs %f0,%f30
clr %o1
fmovs %f1,%f29
clr %o2
fmovs %f0,%f28
clr %o3
fmovs %f1,%f27
clr %o4
fmovs %f0,%f26
clr %o5
fmovs %f1,%f25
clr %o7
fmovs %f0,%f24
clr %l0
fmovs %f1,%f23
clr %l1
fmovs %f0,%f22
clr %l2
fmovs %f1,%f21
clr %l3
fmovs %f0,%f20
clr %l4
fmovs %f1,%f19
clr %l5
fmovs %f0,%f18
clr %l6
fmovs %f1,%f17
clr %l7
fmovs %f0,%f16
clr %i0
fmovs %f1,%f15
clr %i1
fmovs %f0,%f14
clr %i2
fmovs %f1,%f13
clr %i3
fmovs %f0,%f12
clr %i4
fmovs %f1,%f11
clr %i5
fmovs %f0,%f10
clr %g1
fmovs %f1,%f9
clr %g2
fmovs %f0,%f8
clr %g3
fmovs %f1,%f7
clr %g4
fmovs %f0,%f6
clr %g5
fmovs %f1,%f5
fmovs %f0,%f4
fmovs %f1,%f3
fmovs %f0,%f2
add %fp,BIAS,%i0 ! return pointer to caller�s top of stack
ret
restore
.zero: .long 0x0,0x0
.PIC.zero.up:
retl
add %o0,%o7,%o0
#ifdef DEBUG
.global walk_reg_wins
.type walk_reg_wins,#function
walk_reg_wins:
#endif
.walk.reg.wins:
save %sp,FRAME,%sp
cmp %i7,%o7
be 2f
clr %o0
cmp %o7,0 ! compiler never cleans %o7...
be 1f ! could have been a leaf function...
clr %o1
call .walk.reg.wins
nop
1: clr %o2
clr %o3
clr %o4
clr %o5
clr %o7
clr %l0
clr %l1
clr %l2
clr %l3
clr %l4
clr %l5
clr %l6
clr %l7
add %o0,1,%i0 ! used for debugging
2: ret
restore
.size OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
.global OPENSSL_atomic_add
.type OPENSSL_atomic_add,#function
.align 32
OPENSSL_atomic_add:
#ifndef ABI64
subcc %g0,1,%o2
.word 0x95408000 !rd %ccr,%o2, see comment above
cmp %o2,0x99
be .v9
nop
save %sp,FRAME,%sp
ba .enter
nop
#ifdef __sun
! Note that you don't have to link with libthread to call thr_yield,
! as libc provides a stub, which is overloaded the moment you link
! with *either* libpthread or libthread...
#define YIELD_CPU thr_yield
#else
! applies at least to Linux and FreeBSD... Feedback expected...
#define YIELD_CPU sched_yield
#endif
.spin: call YIELD_CPU
nop
.enter: ld [%i0],%i2
cmp %i2,-4096
be .spin
mov -1,%i2
swap [%i0],%i2
cmp %i2,-1
be .spin
add %i2,%i1,%i2
stbar
st %i2,[%i0]
sra %i2,%g0,%i0
ret
restore
.v9:
#endif
ld [%o0],%o2
1: add %o1,%o2,%o3
.word 0xd7e2100a !cas [%o0],%o2,%o3, compare [%o0] with %o2 and swap %o3
cmp %o2,%o3
bne 1b
mov %o3,%o2 ! cas is always fetching to dest. register
add %o1,%o2,%o0 ! OpenSSL expects the new value
retl
sra %o0,%g0,%o0 ! we return signed int, remember?
.size OPENSSL_atomic_add,.-OPENSSL_atomic_add
.global OPENSSL_rdtick
.align 32
OPENSSL_rdtick:
subcc %g0,1,%o0
.word 0x91408000 !rd %ccr,%o0
cmp %o0,0x99
bne .notick
xor %o0,%o0,%o0
.word 0x91410000 !rd %tick,%o0
retl
.word 0x93323020 !srlx %o2,32,%o1
.notick:
retl
xor %o1,%o1,%o1
.type OPENSSL_rdtick,#function
.size OPENSSL_rdtick,.-OPENSSL_rdtick
.section ".init",#alloc,#execinstr
call OPENSSL_cpuid_setup
nop
|