1 /*
2 * x86_cpu_features.c - feature detection for x86 processors
3 *
4 * Originally public domain; changes after 2016-09-07 are copyrighted.
5 *
6 * Copyright 2016 Eric Biggers
7 *
8 * Permission is hereby granted, free of charge, to any person
9 * obtaining a copy of this software and associated documentation
10 * files (the "Software"), to deal in the Software without
11 * restriction, including without limitation the rights to use,
12 * copy, modify, merge, publish, distribute, sublicense, and/or sell
13 * copies of the Software, and to permit persons to whom the
14 * Software is furnished to do so, subject to the following
15 * conditions:
16 *
17 * The above copyright notice and this permission notice shall be
18 * included in all copies or substantial portions of the Software.
19 *
20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
21 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
22 * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
23 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
24 * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
25 * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
26 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
27 * OTHER DEALINGS IN THE SOFTWARE.
28 */
29
30 #include "x86_cpu_features.h"
31
32 #if X86_CPU_FEATURES_ENABLED
33
34 #define DEBUG 0
35
36 #if DEBUG
37 # include <stdio.h>
38 #endif
39
40 u32 _x86_cpu_features = 0;
41
42 /* With old GCC versions we have to manually save and restore the x86_32 PIC
43 * register (ebx). See: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=47602 */
44 #if defined(__i386__) && defined(__PIC__)
45 # define EBX_CONSTRAINT "=r"
46 #else
47 # define EBX_CONSTRAINT "=b"
48 #endif
49
50 /* Execute the CPUID instruction. */
51 static inline void
cpuid(u32 leaf,u32 subleaf,u32 * a,u32 * b,u32 * c,u32 * d)52 cpuid(u32 leaf, u32 subleaf, u32 *a, u32 *b, u32 *c, u32 *d)
53 {
54 __asm__(".ifnc %%ebx, %1; mov %%ebx, %1; .endif\n"
55 "cpuid \n"
56 ".ifnc %%ebx, %1; xchg %%ebx, %1; .endif\n"
57 : "=a" (*a), EBX_CONSTRAINT (*b), "=c" (*c), "=d" (*d)
58 : "a" (leaf), "c" (subleaf));
59 }
60
61 /* Read an extended control register. */
62 static inline u64
read_xcr(u32 index)63 read_xcr(u32 index)
64 {
65 u32 edx, eax;
66
67 /* Execute the "xgetbv" instruction. Old versions of binutils do not
68 * recognize this instruction, so list the raw bytes instead. */
69 __asm__ (".byte 0x0f, 0x01, 0xd0" : "=d" (edx), "=a" (eax) : "c" (index));
70
71 return ((u64)edx << 32) | eax;
72 }
73
74 #define IS_SET(reg, bit) ((reg) & ((u32)1 << (bit)))
75
76 /* Initialize _x86_cpu_features with bits for interesting processor features. */
77 void
x86_setup_cpu_features(void)78 x86_setup_cpu_features(void)
79 {
80 u32 features = 0;
81 u32 dummy1, dummy2, dummy3, dummy4;
82 u32 max_function;
83 u32 features_1, features_2, features_3, features_4;
84 bool os_saves_ymm_regs = false;
85
86 /* Get maximum supported function */
87 cpuid(0, 0, &max_function, &dummy2, &dummy3, &dummy4);
88 if (max_function < 1)
89 goto out;
90
91 /* Standard feature flags */
92 cpuid(1, 0, &dummy1, &dummy2, &features_2, &features_1);
93
94 if (IS_SET(features_1, 25))
95 features |= X86_CPU_FEATURE_SSE;
96
97 if (IS_SET(features_1, 26))
98 features |= X86_CPU_FEATURE_SSE2;
99
100 if (IS_SET(features_2, 0))
101 features |= X86_CPU_FEATURE_SSE3;
102
103 if (IS_SET(features_2, 1))
104 features |= X86_CPU_FEATURE_PCLMULQDQ;
105
106 if (IS_SET(features_2, 9))
107 features |= X86_CPU_FEATURE_SSSE3;
108
109 if (IS_SET(features_2, 19))
110 features |= X86_CPU_FEATURE_SSE4_1;
111
112 if (IS_SET(features_2, 20))
113 features |= X86_CPU_FEATURE_SSE4_2;
114
115 if (IS_SET(features_2, 27)) /* OSXSAVE set? */
116 if ((read_xcr(0) & 0x6) == 0x6)
117 os_saves_ymm_regs = true;
118
119 if (os_saves_ymm_regs && IS_SET(features_2, 28))
120 features |= X86_CPU_FEATURE_AVX;
121
122 if (max_function < 7)
123 goto out;
124
125 /* Extended feature flags */
126 cpuid(7, 0, &dummy1, &features_3, &features_4, &dummy4);
127
128 if (IS_SET(features_3, 3))
129 features |= X86_CPU_FEATURE_BMI;
130
131 if (os_saves_ymm_regs && IS_SET(features_3, 5))
132 features |= X86_CPU_FEATURE_AVX2;
133
134 if (IS_SET(features_3, 8))
135 features |= X86_CPU_FEATURE_BMI2;
136
137 out:
138
139 #if DEBUG
140 printf("Detected x86 CPU features: ");
141 if (features & X86_CPU_FEATURE_SSE)
142 printf("SSE ");
143 if (features & X86_CPU_FEATURE_SSE2)
144 printf("SSE2 ");
145 if (features & X86_CPU_FEATURE_SSE3)
146 printf("SSE3 ");
147 if (features & X86_CPU_FEATURE_PCLMULQDQ)
148 printf("PCLMULQDQ ");
149 if (features & X86_CPU_FEATURE_SSSE3)
150 printf("SSSE3 ");
151 if (features & X86_CPU_FEATURE_SSE4_1)
152 printf("SSE4.1 ");
153 if (features & X86_CPU_FEATURE_SSE4_2)
154 printf("SSE4.2 ");
155 if (features & X86_CPU_FEATURE_BMI)
156 printf("BMI ");
157 if (features & X86_CPU_FEATURE_AVX)
158 printf("AVX ");
159 if (features & X86_CPU_FEATURE_BMI2)
160 printf("BMI2 ");
161 if (features & X86_CPU_FEATURE_AVX2)
162 printf("AVX2 ");
163 printf("\n");
164 #endif /* DEBUG */
165
166 _x86_cpu_features = features | X86_CPU_FEATURES_KNOWN;
167 }
168
169 #endif /* X86_CPU_FEATURES_ENABLED */
170