forked from OpenCloudOS/perf-prof
-
Notifications
You must be signed in to change notification settings - Fork 0
/
percpu-stat.c
308 lines (273 loc) · 9.15 KB
/
percpu-stat.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
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
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
#include <stdio.h>
#include <stdlib.h>
#include <linux/rblist.h>
#include "monitor.h"
#include "trace_helpers.h"
#include "tep.h"
static profiler percpu_stat;
struct swevent_stat {
uint64_t count;
uint64_t diff;
};
struct evsel_node {
struct rb_node rbnode;
struct evsel_node *next;
struct perf_evsel *evsel;
const char *name;
int name_len;
bool cpu_idle;
struct swevent_stat *perins_stats;
struct swevent_stat *total_stats;
};
static struct monitor_ctx {
int nr_ins;
struct evsel_node *first;
struct evsel_node **p_next;
struct rblist evsel_list;
struct tp_list *tp_list;
struct env *env;
} ctx;
static int evsel_node_cmp(struct rb_node *rbn, const void *entry)
{
struct evsel_node *e = container_of(rbn, struct evsel_node, rbnode);
const struct evsel_node *n = entry;
if (e->evsel > n->evsel)
return 1;
else if (e->evsel < n->evsel)
return -1;
else
return 0;
}
static struct rb_node *evsel_node_new(struct rblist *rlist, const void *new_entry)
{
const struct evsel_node *n = new_entry;
struct evsel_node *e = malloc(sizeof(*e));
if (e) {
e->next = NULL;
e->evsel = n->evsel;
e->name = n->name;
e->name_len = (int)strlen(e->name);
e->cpu_idle = n->cpu_idle;
e->perins_stats = calloc(ctx.nr_ins + 1, sizeof(*e->perins_stats));
if (!e->perins_stats) {
free(e);
return NULL;
}
e->total_stats = e->perins_stats + ctx.nr_ins;
*ctx.p_next = e;
ctx.p_next = &e->next;
RB_CLEAR_NODE(&e->rbnode);
return &e->rbnode;
} else
return NULL;
}
static void evsel_node_delete(struct rblist *rblist, struct rb_node *rb_node)
{
struct evsel_node *e = container_of(rb_node, struct evsel_node, rbnode);
free(e->perins_stats);
free(e);
}
static int monitor_ctx_init(struct env *env)
{
ctx.nr_ins = monitor_nr_instance();
ctx.first = NULL;
ctx.p_next = &ctx.first;
rblist__init(&ctx.evsel_list);
ctx.evsel_list.node_cmp = evsel_node_cmp;
ctx.evsel_list.node_new = evsel_node_new;
ctx.evsel_list.node_delete = evsel_node_delete;
ctx.env = env;
tep__ref();
return 0;
}
static void monitor_ctx_exit(void)
{
tep__unref();
rblist__exit(&ctx.evsel_list);
}
static struct perf_evsel *perf_tp_event(struct perf_evlist *evlist, const char *sys, const char *name)
{
struct perf_event_attr attr = {
.type = PERF_TYPE_TRACEPOINT,
.config = 0,
.size = sizeof(struct perf_event_attr),
.sample_period = 0,
.freq = 0,
.sample_type = 0,
.read_format = 0,
.pinned = 0,
.disabled = 0,
};
struct perf_evsel *evsel;
int id;
id = tep__event_id(sys, name);
if (id < 0)
return NULL;
attr.config = id;
evsel = perf_evsel__new(&attr);
if (!evsel) {
return NULL;
}
perf_evlist__add(evlist, evsel);
return evsel;
}
static struct perf_evsel *perf_sw_event(struct perf_evlist *evlist, int config)
{
static bool leader = true;
struct perf_event_attr attr = {
.type = PERF_TYPE_SOFTWARE,
.config = 0,
.size = sizeof(struct perf_event_attr),
.sample_period = 0,
.freq = 0,
.sample_type = 0,
.read_format = 0,
.pinned = 0,
.disabled = leader ? 1 : 0,
};
struct perf_evsel *evsel;
attr.config = config;
evsel = perf_evsel__new(&attr);
if (!evsel) {
return NULL;
}
perf_evlist__add(evlist, evsel);
leader = false;
return evsel;
}
static void __evsel_name(struct perf_evsel *evsel, const char *name, bool cpu_idle)
{
struct evsel_node n;
if (evsel) {
n.evsel = evsel;
n.name = name;
n.cpu_idle = cpu_idle;
rblist__add_node(&ctx.evsel_list, &n);
}
}
#define evsel_name(evsel, name) __evsel_name((evsel), (name), false)
static int percpu_stat_init(struct perf_evlist *evlist, struct env *env)
{
if (monitor_ctx_init(env) < 0)
return -1;
if (env->interval == 0)
env->interval = 1000;
env->perins = true;
//software event
evsel_name(perf_sw_event(evlist, PERF_COUNT_SW_CONTEXT_SWITCHES), " SOFT csw");
evsel_name(perf_sw_event(evlist, PERF_COUNT_SW_CPU_MIGRATIONS), "cpu-mig");
evsel_name(perf_sw_event(evlist, PERF_COUNT_SW_PAGE_FAULTS_MIN), "minflt");
evsel_name(perf_sw_event(evlist, PERF_COUNT_SW_PAGE_FAULTS_MAJ), "majflt");
//syscalls
if (env->syscalls)
evsel_name(perf_tp_event(evlist, "raw_syscalls", "sys_enter"), " syscalls");
//irq, softirq, workqueue
evsel_name(perf_tp_event(evlist, "irq", "irq_handler_entry"), " hardirq");
evsel_name(perf_tp_event(evlist, "irq", "softirq_entry"), "softirq");
evsel_name(perf_tp_event(evlist, "timer", "hrtimer_expire_entry"), "hrtimer");
evsel_name(perf_tp_event(evlist, "workqueue", "workqueue_execute_start"), "workqueue");
//kvm
evsel_name(perf_tp_event(evlist, "kvm", "kvm_exit"), " KVM exit");
//net
evsel_name(perf_tp_event(evlist, "net", "netif_receive_skb"), " NET recv");
evsel_name(perf_tp_event(evlist, "net", "net_dev_xmit"), " xmit");
evsel_name(perf_tp_event(evlist, "net", "napi_gro_receive_entry"), "gro");
//page alloc
evsel_name(perf_tp_event(evlist, "kmem", "mm_page_alloc"), " MM alloc");
evsel_name(perf_tp_event(evlist, "compaction", "mm_compaction_migratepages"), "compact");
evsel_name(perf_tp_event(evlist, "vmscan", "mm_vmscan_direct_reclaim_begin"), "reclaim");
evsel_name(perf_tp_event(evlist, "migrate", "mm_migrate_pages"), "migrate");
//page cache
evsel_name(perf_tp_event(evlist, "filemap", "mm_filemap_add_to_page_cache"), " PAGE cache");
evsel_name(perf_tp_event(evlist, "writeback", "wbc_writepage"), " WB pages");
//cpu_idle
__evsel_name(perf_tp_event(evlist, "power", "cpu_idle"), " idle", true);
perf_evlist__set_leader(evlist);
return 0;
}
static void percpu_stat_exit(struct perf_evlist *evlist)
{
monitor_ctx_exit();
}
static int percpu_stat_read(struct perf_evsel *evsel, struct perf_counts_values *count, int instance)
{
struct evsel_node n = {.evsel = evsel};
struct rb_node *rbn = rblist__find(&ctx.evsel_list, &n);
struct evsel_node *e = rbn ? container_of(rbn, struct evsel_node, rbnode) : NULL;
if (e == NULL)
return 0;
e->perins_stats[instance].diff = 0;
if (count->val > e->perins_stats[instance].count) {
e->perins_stats[instance].diff = count->val - e->perins_stats[instance].count;
e->perins_stats[instance].count = count->val;
if (e->cpu_idle) {
//cpu_idle, contains enter and exit, must be divided by 2
e->perins_stats[instance].diff /= 2;
}
}
e->total_stats->diff += e->perins_stats[instance].diff;
return 0;
}
static void percpu_stat_interval(void)
{
struct evsel_node *next = ctx.first;
int ins;
print_time(stdout);
printf("\n[CPU] ");
while (next) {
printf("%s ", next->name);
next = next->next;
}
if (ctx.env->perins)
for (ins = 0; ins < ctx.nr_ins; ins ++) {
printf("\n[%03d] ", monitor_instance_cpu(ins));
next = ctx.first;
while (next) {
printf("%*lu ", next->name_len, next->perins_stats[ins].diff);
next = next->next;
}
}
printf("\n[ALL] ");
next = ctx.first;
while (next) {
printf("%*lu ", next->name_len, next->total_stats->diff);
next->total_stats->diff = 0;
next = next->next;
}
printf("\n");
}
static void percpu_stat_sample(union perf_event *event, int instance)
{
}
static const char *percpu_stat_desc[] = PROFILER_DESC("percpu-stat",
"[OPTION...] [--syscalls]",
"Handpicked event statistics.", "",
"TRACEPOINT", "",
" PERF_COUNT_SW_CONTEXT_SWITCHES PERF_COUNT_SW_CPU_MIGRATIONS",
" PERF_COUNT_SW_PAGE_FAULTS_MIN PERF_COUNT_SW_PAGE_FAULTS_MAJ",
" irq:irq_handler_entry irq:softirq_entry",
" timer:hrtimer_expire_entry workqueue:workqueue_execute_start",
" kvm:kvm_exit net:netif_receive_skb",
" net:net_dev_xmit net:napi_gro_receive_entry ",
" kmem:mm_page_alloc compaction:mm_compaction_migratepages",
" migrate:mm_migrate_pages vmscan:mm_vmscan_direct_reclaim_begin",
" writeback:wbc_writepage filemap:mm_filemap_add_to_page_cache",
" raw_syscalls:sys_enter power:cpu_idle", "",
"EXAMPLES", "",
" "PROGRAME" percpu-stat -C 0",
" "PROGRAME" percpu-stat --syscalls -i 2000");
static const char *percpu_stat_argv[] = PROFILER_ARGV("percpu-stat",
PROFILER_ARGV_OPTION,
PROFILER_ARGV_PROFILER, "syscalls");
static profiler percpu_stat = {
.name = "percpu-stat",
.desc = percpu_stat_desc,
.argv = percpu_stat_argv,
.pages = 0,
.init = percpu_stat_init,
.deinit = percpu_stat_exit,
.interval = percpu_stat_interval,
.read = percpu_stat_read,
.sample = percpu_stat_sample,
};
PROFILER_REGISTER(percpu_stat);