builtin-stat.c
/*
* builtin-stat.c
*
* Builtin stat command: Give a precise performance counters summary
* overview about any workload, CPU or specific PID.
*
* Sample output:
$ perf stat ~/hackbench 10
Time: 0.104
Performance counter stats for ‘/home/mingo/hackbench’:
1255.538611 task clock ticks # 10.143 CPU utilization factor
54011 context switches # 0.043 M/sec
385 CPU migrations # 0.000 M/sec
17755 pagefaults # 0.014 M/sec
3808323185 CPU cycles # 3033.219 M/sec
1575111190 instructions # 1254.530 M/sec
17367895 cache references # 13.833 M/sec
7674421 cache misses # 6.112 M/sec
Wall-clock time elapsed: 123.786620 msecs
*
* Copyright (C) 2008, Red Hat Inc, Ingo Molnar
*
* Improvements and fixes by:
*
* Arjan van de Ven
* Yanmin Zhang
* Wu Fengguang
* Mike Galbraith
* Paul Mackerras
*
* Released under the GPL v2. (and only v2, not any later version)
*/
#include «perf.h»
#include «builtin.h»
#include «util/util.h»
#include «util/parse-options.h»
#include «util/parse-events.h»
#include «util/event.h»
#include «util/debug.h»
#include static struct perf_event_attr default_attrs[] = { { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_TASK_CLOCK }, { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_CPU_CYCLES }, }; static int system_wide = 0; static int run_count = 1; static int fd[MAX_NR_CPUS][MAX_COUNTERS]; static int event_scaled[MAX_COUNTERS]; struct stats static void update_stats(struct stats *stats, u64 val) stats->n++; static double avg_stats(struct stats *stats) /* return sqrt(variance_mean); struct stats event_res_stats[MAX_COUNTERS][3]; #define MATCH_EVENT(t, c, counter) \ #define ERR_PERF_OPEN \ static void create_perf_stat_counter(int counter, int pid) if (scale) if (system_wide) { for (cpu = 0; cpu < nr_cpus; cpu++) { fd[0][counter] = sys_perf_event_open(attr, pid, -1, -1, 0); /* return 0; /* count[0] = count[1] = count[2] = 0; nv = scale ? 3 : 1; res = read(fd[cpu][counter], single_count, nv * sizeof(u64)); close(fd[cpu][counter]); count[0] += single_count[0]; scaled = 0; if (count[2] < count[1]) { for (i = 0; i < 3; i++) if (verbose) { /* static int run_perf_stat(int argc __used, const char **argv) if (!system_wide) if (pipe(child_ready_pipe) < 0 || pipe(go_pipe) < 0) { if ((pid = fork()) < 0) if (!pid) { /* /* /* execvp(argv[0], (char **)argv); perror(argv[0]); child_pid = pid; /* for (counter = 0; counter < nr_counters; counter++) /* close(go_pipe[1]); t1 = rdclock(); update_stats(&walltime_nsecs_stats, t1 - t0); for (counter = 0; counter < nr_counters; counter++) return WEXITSTATUS(status); static void print_noise(int counter, double avg) fprintf(stderr, " ( +- %7.3f%% )", static void nsec_printout(int counter, double avg) fprintf(stderr, " %14.6f %-24s", msecs, event_name(counter)); if (MATCH_EVENT(SOFTWARE, SW_TASK_CLOCK, counter)) { static void abs_printout(int counter, double avg) fprintf(stderr, " %14.0f %-24s", avg, event_name(counter)); if (MATCH_EVENT(HARDWARE, HW_INSTRUCTIONS, counter)) { if (total) fprintf(stderr, " # %10.3f IPC ", ratio); if (total) fprintf(stderr, " # %10.3f M/sec", ratio); /* if (scaled == -1) { if (nsec_counter(counter)) print_noise(counter, avg); if (scaled) { avg_enabled = avg_stats(&event_res_stats[counter][1]); fprintf(stderr, » (scaled from %.2f%%)», fprintf(stderr, «\n»); static void print_stat(int argc, const char **argv) fflush(stdout); fprintf(stderr, «\n»); for (i = 1; i < argc; i++) fprintf(stderr, "\'"); for (counter = 0; counter < nr_counters; counter++) fprintf(stderr, "\n"); static volatile int signr = -1; static void skip_signal(int signo) static void sig_atexit(void) if (signr == -1) signal(signr, SIG_DFL); static const char * const stat_usage[] = { static const struct option options[] = { int cmd_stat(int argc, const char **argv, const char *prefix __used) argc = parse_options(argc, argv, options, stat_usage, /* Set attrs and nr_counters if no event is selected and !null_run */ nr_cpus = sysconf(_SC_NPROCESSORS_ONLN); /* status = 0; print_stat(argc, argv); return status;
#include
{ .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_CONTEXT_SWITCHES},
{ .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_CPU_MIGRATIONS },
{ .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_PAGE_FAULTS },
{ .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_INSTRUCTIONS },
{ .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_CACHE_REFERENCES},
{ .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_CACHE_MISSES },
static unsigned int nr_cpus = 0;
static int run_idx = 0;
static int inherit = 1;
static int scale = 1;
static pid_t target_pid = -1;
static pid_t child_pid = -1;
static int null_run = 0;
{
double n, mean, M2;
};
{
double delta;
delta = val – stats->mean;
stats->mean += delta / stats->n;
stats->M2 += delta*(val – stats->mean);
}
{
return stats->mean;
}
* http://en.wikipedia.org/wiki/Algorithms_for_calculating_variance
*
* (\Sum n_i^2) – ((\Sum n_i)^2)/n
* s^2 = ——————————-
* n – 1
*
* http://en.wikipedia.org/wiki/Stddev
*
* The std dev of the mean is related to the std dev by:
*
* s
* s_mean = ——-
* sqrt(n)
*
*/
static double stddev_stats(struct stats *stats)
{
double variance = stats->M2 / (stats->n – 1);
double variance_mean = variance / stats->n;
}
struct stats runtime_nsecs_stats;
struct stats walltime_nsecs_stats;
struct stats runtime_cycles_stats;
(attrs[counter].type == PERF_TYPE_##t && \
attrs[counter].config == PERF_COUNT_##c)
«Error: counter %d, sys_perf_event_open() syscall returned with %d (%s)\n»
{
struct perf_event_attr *attr = attrs + counter;
attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
PERF_FORMAT_TOTAL_TIME_RUNNING;
unsigned int cpu;
fd[cpu][counter] = sys_perf_event_open(attr, -1, cpu, -1, 0);
if (fd[cpu][counter] < 0 && verbose)
fprintf(stderr, ERR_PERF_OPEN, counter,
fd[cpu][counter], strerror(errno));
}
} else {
attr->inherit = inherit;
attr->disabled = 1;
attr->enable_on_exec = 1;
if (fd[0][counter] < 0 && verbose)
fprintf(stderr, ERR_PERF_OPEN, counter,
fd[0][counter], strerror(errno));
}
}
* Does the counter have nsecs as a unit?
*/
static inline int nsec_counter(int counter)
{
if (MATCH_EVENT(SOFTWARE, SW_CPU_CLOCK, counter) ||
MATCH_EVENT(SOFTWARE, SW_TASK_CLOCK, counter))
return 1;
}
* Read out the results of a single counter:
*/
static void read_counter(int counter)
{
u64 count[3], single_count[3];
unsigned int cpu;
size_t res, nv;
int scaled;
int i;
for (cpu = 0; cpu < nr_cpus; cpu++) {
if (fd[cpu][counter] < 0)
continue;
assert(res == nv * sizeof(u64));
fd[cpu][counter] = -1;
if (scale) {
count[1] += single_count[1];
count[2] += single_count[2];
}
}
if (scale) {
if (count[2] == 0) {
event_scaled[counter] = -1;
count[0] = 0;
return;
}
event_scaled[counter] = 1;
count[0] = (unsigned long long)
((double)count[0] * count[1] / count[2] + 0.5);
}
}
update_stats(&event_res_stats[counter][i], count[i]);
fprintf(stderr, "%s: %Ld %Ld %Ld\n", event_name(counter),
count[0], count[1], count[2]);
}
* Save the full runtime - to allow normalization during printout:
*/
if (MATCH_EVENT(SOFTWARE, SW_TASK_CLOCK, counter))
update_stats(&runtime_nsecs_stats, count[0]);
if (MATCH_EVENT(HARDWARE, HW_CPU_CYCLES, counter))
update_stats(&runtime_cycles_stats, count[0]);
}
{
unsigned long long t0, t1;
int status = 0;
int counter;
int pid;
int child_ready_pipe[2], go_pipe[2];
char buf;
nr_cpus = 1;
perror("failed to create pipes");
exit(1);
}
perror("failed to fork");
close(child_ready_pipe[0]);
close(go_pipe[1]);
fcntl(go_pipe[0], F_SETFD, FD_CLOEXEC);
* Do a dummy execvp to get the PLT entry resolved,
* so we avoid the resolver overhead on the real
* execvp call.
*/
execvp("", (char **)argv);
* Tell the parent we're ready to go
*/
close(child_ready_pipe[1]);
* Wait until the parent tells us to go.
*/
if (read(go_pipe[0], &buf, 1) == -1)
perror("unable to read pipe");
exit(-1);
}
* Wait for the child to be ready to exec.
*/
close(child_ready_pipe[1]);
close(go_pipe[0]);
if (read(child_ready_pipe[0], &buf, 1) == -1)
perror("unable to read pipe");
close(child_ready_pipe[0]);
create_perf_stat_counter(counter, pid);
* Enable counters and exec the command:
*/
t0 = rdclock();
wait(&status);
read_counter(counter);
}
{
if (run_count == 1)
return;
100 * stddev_stats(&event_res_stats[counter][0]) / avg);
}
{
double msecs = avg / 1e6;
fprintf(stderr, " # %10.3f CPUs ",
avg / avg_stats(&walltime_nsecs_stats));
}
}
{
double total, ratio = 0.0;
total = avg_stats(&runtime_cycles_stats);
ratio = avg / total;
} else {
total = avg_stats(&runtime_nsecs_stats);
ratio = 1000.0 * avg / total;
}
}
* Print out the results of a single counter:
*/
static void print_counter(int counter)
{
double avg = avg_stats(&event_res_stats[counter][0]);
int scaled = event_scaled[counter];
fprintf(stderr, " %14s %-24s\n",
"
return;
}
nsec_printout(counter, avg);
else
abs_printout(counter, avg);
double avg_enabled, avg_running;
avg_running = avg_stats(&event_res_stats[counter][2]);
100 * avg_running / avg_enabled);
}
}
{
int i, counter;
fprintf(stderr, » Performance counter stats for \’%s», argv[0]);
fprintf(stderr, " %s", argv[i]);
if (run_count > 1)
fprintf(stderr, » (%d runs)», run_count);
fprintf(stderr, «:\n\n»);
print_counter(counter);
fprintf(stderr, " %14.9f seconds time elapsed",
avg_stats(&walltime_nsecs_stats)/1e9);
if (run_count > 1) {
fprintf(stderr, » ( +- %7.3f%% )»,
100*stddev_stats(&walltime_nsecs_stats) /
avg_stats(&walltime_nsecs_stats));
}
fprintf(stderr, «\n\n»);
}
{
signr = signo;
}
{
if (child_pid != -1)
kill(child_pid, SIGTERM);
return;
kill(getpid(), signr);
}
«perf stat [
NULL
};
OPT_CALLBACK(’e', «event», NULL, «event»,
«event selector. use ‘perf list’ to list available events»,
parse_events),
OPT_BOOLEAN(’i', «inherit», &inherit,
«child tasks inherit counters»),
OPT_INTEGER(’p', «pid», &target_pid,
«stat events on existing pid»),
OPT_BOOLEAN(’a', «all-cpus», &system_wide,
«system-wide collection from all CPUs»),
OPT_BOOLEAN(’c', «scale», &scale,
«scale/normalize counters»),
OPT_BOOLEAN(’v', «verbose», &verbose,
«be more verbose (show counter open errors, etc)»),
OPT_INTEGER(’r', «repeat», &run_count,
«repeat command and print average + stddev (max: 100)»),
OPT_BOOLEAN(’n', «null», &null_run,
«null run – dont start any counters»),
OPT_END()
};
{
int status;
PARSE_OPT_STOP_AT_NON_OPTION);
if (!argc)
usage_with_options(stat_usage, options);
if (run_count <= 0)
usage_with_options(stat_usage, options);
if (!null_run && !nr_counters) {
memcpy(attrs, default_attrs, sizeof(default_attrs));
nr_counters = ARRAY_SIZE(default_attrs);
}
assert(nr_cpus <= MAX_NR_CPUS);
assert((int)nr_cpus >= 0);
* We dont want to block the signals – that would cause
* child tasks to inherit that and Ctrl-C would not work.
* What we want is for Ctrl-C to work in the exec()-ed
* task, but being ignored by perf stat itself:
*/
atexit(sig_atexit);
signal(SIGINT, skip_signal);
signal(SIGALRM, skip_signal);
signal(SIGABRT, skip_signal);
for (run_idx = 0; run_idx < run_count; run_idx++) {
if (run_count != 1 && verbose)
fprintf(stderr, «[ perf stat: executing run #%d ... ]\n», run_idx + 1);
status = run_perf_stat(argc, argv);
}
}