xref: /openssl/doc/man3/OPENSSL_ia32cap.pod (revision acc26552)
1=pod
2
3=head1 NAME
4
5OPENSSL_ia32cap - the x86[_64] processor capabilities vector
6
7=head1 SYNOPSIS
8
9 env OPENSSL_ia32cap=... <application>
10
11=head1 DESCRIPTION
12
13OpenSSL supports a range of x86[_64] instruction set extensions and
14features. These extensions are denoted by individual bits or groups of bits
15stored internally as ten 32-bit capability vectors and for simplicity
16represented logically below as five 64-bit vectors. This logical
17vector (LV) representation is used to streamline the definition of the
18OPENSSL_ia32cap environment variable.
19
20Upon toolkit initialization, the capability vectors are populated through
21successive executions of the CPUID instruction, after which any OPENSSL_ia32cap
22environment variable capability bit modifications are applied. After toolkit
23initialization is complete, populated vectors are then used to choose
24between different code paths to provide optimal performance across a wide
25range of x86[_64] based processors.
26
27Further CPUID information can be found in the Intel(R) Architecture
28Instruction Set Extensions Programming Reference, and the AMD64 Architecture
29Programmer's Manual (Volume 3).
30
31=head2 Notable Capability Bits for LV0
32
33The following are notable capability bits from logical vector 0 (LV0)
34resulting from the following execution of CPUID.(EAX=01H).EDX and
35CPUID.(EAX=01H).ECX:
36
37=over 4
38
39=item bit #0+4 denoting presence of Time-Stamp Counter;
40
41=item bit #0+19 denoting availability of CLFLUSH instruction;
42
43=item bit #0+20, reserved by Intel, is used to choose among RC4 code paths;
44
45=item bit #0+23 denoting MMX support;
46
47=item bit #0+24, FXSR bit, denoting availability of XMM registers;
48
49=item bit #0+25 denoting SSE support;
50
51=item bit #0+26 denoting SSE2 support;
52
53=item bit #0+28 denoting Hyperthreading, which is used to distinguish
54cores with shared cache;
55
56=item bit #0+30, reserved by Intel, denotes specifically Intel CPUs;
57
58=item bit #0+33 denoting availability of PCLMULQDQ instruction;
59
60=item bit #0+41 denoting SSSE3, Supplemental SSE3, support;
61
62=item bit #0+43 denoting AMD XOP support (forced to zero on non-AMD CPUs);
63
64=item bit #0+54 denoting availability of MOVBE instruction;
65
66=item bit #0+57 denoting AES-NI instruction set extension;
67
68=item bit #0+58, XSAVE bit, lack of which in combination with MOVBE is used
69to identify Atom Silvermont core;
70
71=item bit #0+59, OSXSAVE bit, denoting availability of YMM registers;
72
73=item bit #0+60 denoting AVX extension;
74
75=item bit #0+62 denoting availability of RDRAND instruction;
76
77=back
78
79=head2 Notable Capability Bits for LV1
80
81The following are notable capability bits from logical vector 1 (LV1)
82resulting from the following execution of CPUID.(EAX=07H,ECX=0H).EBX and
83CPUID.(EAX=07H,ECX=0H).ECX:
84
85=over 4
86
87=item bit #64+3 denoting availability of BMI1 instructions, e.g. ANDN;
88
89=item bit #64+5 denoting availability of AVX2 instructions;
90
91=item bit #64+8 denoting availability of BMI2 instructions, e.g. MULX
92and RORX;
93
94=item bit #64+16 denoting availability of AVX512F extension;
95
96=item bit #64+17 denoting availability of AVX512DQ extension;
97
98=item bit #64+18 denoting availability of RDSEED instruction;
99
100=item bit #64+19 denoting availability of ADCX and ADOX instructions;
101
102=item bit #64+21 denoting availability of AVX512IFMA extension;
103
104=item bit #64+29 denoting availability of SHA extension;
105
106=item bit #64+30 denoting availability of AVX512BW extension;
107
108=item bit #64+31 denoting availability of AVX512VL extension;
109
110=item bit #64+41 denoting availability of VAES extension;
111
112=item bit #64+42 denoting availability of VPCLMULQDQ extension;
113
114=back
115
116=head2 Notable Capability Bits for LV2
117
118The following are notable capability bits from logical vector 2 (LV2)
119resulting from the following execution of CPUID.(EAX=07H,ECX=0H).EDX and
120CPUID.(EAX=07H,ECX=1H).EAX:
121
122=over 4
123
124=item bit #128+15 denoting availability of Hybrid CPU;
125
126=item bit #128+29 denoting support for IA32_ARCH_CAPABILITIES MSR;
127
128=item bit #128+32 denoting availability of SHA512 extension;
129
130=item bit #128+33 denoting availability of SM3 extension;
131
132=item bit #128+34 denoting availability of SM4 extension;
133
134=item bit #128+55 denoting availability of AVX-IFMA extension;
135
136=back
137
138=head2 Notable Capability Bits for LV3
139
140The following are notable capability bits from logical vector 3 (LV3)
141resulting from the following execution of CPUID.(EAX=07H,ECX=1H).EDX and
142CPUID.(EAX=07H,ECX=1H).EBX:
143
144=over 4
145
146=item bit #192+19 denoting availability of AVX10 Converged Vector ISA extension;
147
148=item bit #192+21 denoting availability of APX_F extension;
149
150=back
151
152=head2 Notable Capability Bits for LV4
153
154The following are notable capability bits from logical vector 4 (LV4)
155resulting from the following execution of CPUID.(EAX=07H,ECX=1H).ECX and
156CPUID.(EAX=24H,ECX=0H).EBX:
157
158=over 4
159
160=item bits #256+32+[0:7] denoting AVX10 Converged Vector ISA Version (8 bits);
161
162=item bit #256+48 denoting AVX10 XMM support;
163
164=item bit #256+49 denoting AVX10 YMM support;
165
166=item bit #256+50 denoting AVX10 ZMM support;
167
168=back
169
170=head2 OPENSSL_ia32cap environment variable
171
172The B<OPENSSL_ia32cap> environment variable provides a mechanism to override
173the default capability vector values at library initialization time.
174The variable consists of a series of 64-bit numbers representing each
175of the logical vectors (LV) described above. Each value is delimited by a 'B<:>'.
176Decimal/Octal/Hexadecimal values representations are supported.
177
178C<env OPENSSL_ia32cap=LV0:LV1:LV2:LV3:LV4>
179
180Used in this form, each non-null logical vector will *overwrite* the entire corresponding
181capability vector pair with the provided value. To keep compatibility with the
182behaviour of the original OPENSSL_ia32cap environment variable
183<env OPENSSL_ia32cap=LV0:LV1>, the next capability vector pairs will be set to zero.
184
185To illustrate, the following will zero all capability bits in logical vectors 1 and further
186(disable all post-AVX extensions):
187
188C<env OPENSSL_ia32cap=:0>
189
190The following will zero all capability bits in logical vectors 2 and further:
191
192C<env OPENSSL_ia32cap=::0>
193
194The following will zero all capability bits only in logical vector 1:
195C<env OPENSSL_ia32cap=:0::::>
196
197A more likely usage scenario would be to disable specific instruction set extensions.
198The 'B<~>' character is used to specify a bit mask of the extensions to be disabled for
199a particular logical vector.
200
201To illustrate, the following will disable AVX2 code paths and further extensions:
202
203C<env OPENSSL_ia32cap=:~0x20000000000>
204
205The following will disable AESNI (LV0 bit 57) and VAES (LV1 bit 41)
206extensions and therefore any code paths using those extensions but leave
207the rest of the logical vectors unchanged:
208
209C<env OPENSSL_ia32cap=~0x200000000000000:~0x20000000000:~0x0:~0x0:~0x0>
210
211=head1 NOTES
212
213Not all capability bits are copied from CPUID output verbatim. An example
214of this is the somewhat less intuitive clearing of LV0 bit #28, or ~0x10000000
215in the "environment variable" terms. It has been adjusted to reflect whether or
216not the data cache is actually shared between logical cores. This in turn affects
217the decision on whether or not expensive countermeasures against cache-timing attacks
218are applied, most notably in AES assembler module.
219
220=head1 RETURN VALUES
221
222Not available.
223
224=head1 COPYRIGHT
225
226Copyright 2004-2021 The OpenSSL Project Authors. All Rights Reserved.
227
228Licensed under the Apache License 2.0 (the "License").  You may not use
229this file except in compliance with the License.  You can obtain a copy
230in the file LICENSE in the source distribution or at
231L<https://www.openssl.org/source/license.html>.
232
233=cut
234