-
Notifications
You must be signed in to change notification settings - Fork 1
/
machine.c
166 lines (143 loc) · 5.31 KB
/
machine.c
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
#include "miniprof.h"
static int msr_count;
static struct msr *available_msrs;
static int **available_msr_usage;
extern int ncpus;
void cpuid(unsigned info, unsigned *eax, unsigned *ebx, unsigned *ecx, unsigned *edx) {
*eax = info;
__asm volatile
("mov %%ebx, %%edi;"
"cpuid;"
"mov %%ebx, %%esi;"
"mov %%edi, %%ebx;"
: "+a"(*eax), "=S"(*ebx), "=c"(*ecx), "=d"(*edx)
: : "edi");
}
unsigned int get_processor_family() {
char vendor[12];
unsigned int family;
unsigned int a, b, c, d;
cpuid(0x0, &a, (unsigned int *)vendor, (unsigned int *)(vendor + 8), (unsigned int *)(vendor + 4));
if(memcmp(vendor, "AuthenticAMD", sizeof(vendor)))
die("Unsupported CPU (expected AuthenticAMD, found %12.12s)\n", vendor);
cpuid(0x1, &a, &b, &c, &d);
family = (a & 0x0ff00f00);
return family;
}
int can_be_used_10h(struct msr *msr, uint64_t evt) {
return 1;
}
int is_per_node(uint64_t evt) {
return ((evt & 0xe0) == 0xe0);
}
int can_be_used_15h(struct msr *msr, uint64_t evt) {
if(is_per_node(evt)) {
return (msr->select & 0x40);
} else {
if(msr->select & 0x40)
return 0;
evt &= 0xF000000FF;
if(evt <= 0x1F) { /* FP */
if(evt == 0x0 || evt == 0x3 || evt == 0x4)
return msr->id == 3;
return msr->id >= 3;
} else if(evt <= 0x3F) { /* LS */
if(evt == 0x23)
return msr->id <= 2;
return 1;
} else if(evt <= 0x5F) { /* DC */
if(evt == 0x43 || evt == 0x45 || evt == 0x46)
return msr->id <= 2;
return 1;
} else if(evt <= 0x7F) { /* CU */
return msr->id <= 2;
} else if(evt <= 0x9F) { /* IC */
return msr->id <= 2;
} else { /* EX, DE */
if((evt >= 0xD0 && evt <= 0xD9) || evt == 0x1000000DD || evt == 0x1000000DE)
return msr->id <= 2;
if(evt == 0x1000000D0)
return msr->id >= 3;
return 1;
}
}
return 0;
}
void get_available_msr(void) {
if(available_msrs)
return;
int i;
unsigned int family = get_processor_family();
switch(family) {
case 0x100f00: /* AMD fam10h, see AMD BKDG 10h, section 2.16.1 */
msr_count = 4;
available_msrs = malloc(msr_count * sizeof(*available_msrs));
available_msr_usage = malloc(msr_count * sizeof(*available_msr_usage));
for(i = 0; i < msr_count; i++) {
available_msrs[i].id = i;
available_msrs[i].select = 0xC0010000 + i;
available_msrs[i].value = 0xC0010000 + i + 4;
available_msrs[i].can_be_used = can_be_used_10h;
available_msr_usage[i] = calloc(ncpus, sizeof(*available_msr_usage[i]));
}
break;
case 0x600f00: /* 15h */
msr_count = 10;
available_msrs = malloc(msr_count * sizeof(*available_msrs));
available_msr_usage = malloc(msr_count * sizeof(*available_msr_usage));
for(i = 0; i < 6; i++) {
available_msrs[i].id = i;
available_msrs[i].select = 0xC0010200 + 2 * i;
available_msrs[i].value = 0xC0010200 + 2 * i + 1;
available_msrs[i].can_be_used = can_be_used_15h;
available_msr_usage[i] = calloc(ncpus, sizeof(*available_msr_usage[i]));
}
for(i = 6; i < msr_count; i++) {
available_msrs[i].id = i;
available_msrs[i].select = 0xC0010240 + 2 * (i - 6);
available_msrs[i].value = 0xC0010240 + 2 * (i - 6) + 1;
available_msrs[i].can_be_used = can_be_used_15h;
available_msr_usage[i] = calloc(ncpus, sizeof(*available_msr_usage[i]));
}
break;
default:
die("Unsupported processor family (%d)\n", family);
}
}
int is_reserved(int msr_id, uint64_t evt, int cpu_filter) {
int i;
int per_node = is_per_node(evt);
for(i = 0; i < ncpus; i++) {
if(available_msr_usage[msr_id][i] // msr has been configured on cpu i
&& ((cpu_filter == -1) // and we want to use it on all cpu
|| (cpu_filter == i) // or on this cpu
|| (per_node && (numa_node_of_cpu(cpu_filter) == numa_node_of_cpu(i))))) // or on the same node (and it is a problem)
return 1;
}
return 0;
}
void reserve_msr(int msr_id, uint64_t evt, int cpu_filter) {
int i;
int per_node = is_per_node(evt);
for(i = 0; i < ncpus; i++) {
if(cpu_filter == -1 || cpu_filter == i
|| (per_node && (numa_node_of_cpu(cpu_filter) == numa_node_of_cpu(i))))
available_msr_usage[msr_id][i] = 1;
}
}
/* Returns a MSR for a performance monitoring counter. */
struct msr *get_msr(uint64_t evt, uint64_t cpu_filter) {
int i;
get_available_msr();
/* Perform search in reverse to increase the chance to use MSR 5-3 on 15h */
/* because these counters can be used on a limited subset of events */
for(i = msr_count - 1; i >= 0; i--) {
if(!is_reserved(i, evt, cpu_filter) && available_msrs[i].can_be_used(&available_msrs[i], evt)) {
struct msr *msr = malloc(sizeof(*msr));
memcpy(msr, &available_msrs[i], sizeof(*msr));
return msr;
}
}
die("No free msr for event %llx", (long long unsigned)evt);
return 0;
}