Commit 69aad6f1 authored by Arnaldo Carvalho de Melo's avatar Arnaldo Carvalho de Melo
Browse files

perf tools: Introduce event selectors



Out of ad-hoc code and global arrays with hard coded sizes.

This is the first step on having a library that will be first
used on regression tests in the 'perf test' tool.

[acme@felicio linux]$ size /tmp/perf.before
   text	   data	    bss	    dec	    hex	filename
1273776	  97384	5104416	6475576	 62cf38	/tmp/perf.before
[acme@felicio linux]$ size /tmp/perf.new
   text	   data	    bss	    dec	    hex	filename
1275422	  97416	1392416	2765254	 2a31c6	/tmp/perf.new

Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
parent 56f4c400
......@@ -396,6 +396,7 @@ LIB_H += util/build-id.h
LIB_H += util/debug.h
LIB_H += util/debugfs.h
LIB_H += util/event.h
LIB_H += util/evsel.h
LIB_H += util/exec_cmd.h
LIB_H += util/types.h
LIB_H += util/levenshtein.h
......@@ -404,6 +405,7 @@ LIB_H += util/parse-options.h
LIB_H += util/parse-events.h
LIB_H += util/quote.h
LIB_H += util/util.h
LIB_H += util/xyarray.h
LIB_H += util/header.h
LIB_H += util/help.h
LIB_H += util/session.h
......@@ -433,6 +435,7 @@ LIB_OBJS += $(OUTPUT)util/ctype.o
LIB_OBJS += $(OUTPUT)util/debugfs.o
LIB_OBJS += $(OUTPUT)util/environment.o
LIB_OBJS += $(OUTPUT)util/event.o
LIB_OBJS += $(OUTPUT)util/evsel.o
LIB_OBJS += $(OUTPUT)util/exec_cmd.o
LIB_OBJS += $(OUTPUT)util/help.o
LIB_OBJS += $(OUTPUT)util/levenshtein.o
......@@ -470,6 +473,7 @@ LIB_OBJS += $(OUTPUT)util/sort.o
LIB_OBJS += $(OUTPUT)util/hist.o
LIB_OBJS += $(OUTPUT)util/probe-event.o
LIB_OBJS += $(OUTPUT)util/util.o
LIB_OBJS += $(OUTPUT)util/xyarray.o
LIB_OBJS += $(OUTPUT)util/cpumap.o
BUILTIN_OBJS += $(OUTPUT)builtin-annotate.o
......
......@@ -18,6 +18,7 @@
#include "util/header.h"
#include "util/event.h"
#include "util/evsel.h"
#include "util/debug.h"
#include "util/session.h"
#include "util/symbol.h"
......@@ -27,13 +28,13 @@
#include <sched.h>
#include <sys/mman.h>
#define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y))
enum write_mode_t {
WRITE_FORCE,
WRITE_APPEND
};
static int *fd[MAX_NR_CPUS][MAX_COUNTERS];
static u64 user_interval = ULLONG_MAX;
static u64 default_interval = 0;
static u64 sample_type;
......@@ -81,7 +82,6 @@ static struct perf_session *session;
static const char *cpu_list;
struct mmap_data {
int counter;
void *base;
unsigned int mask;
unsigned int prev;
......@@ -229,12 +229,12 @@ static struct perf_header_attr *get_header_attr(struct perf_event_attr *a, int n
return h_attr;
}
static void create_counter(int counter, int cpu)
static void create_counter(struct perf_evsel *evsel, int cpu)
{
char *filter = filters[counter];
struct perf_event_attr *attr = attrs + counter;
char *filter = evsel->filter;
struct perf_event_attr *attr = &evsel->attr;
struct perf_header_attr *h_attr;
int track = !counter; /* only the first counter needs these */
int track = !evsel->idx; /* only the first counter needs these */
int thread_index;
int ret;
struct {
......@@ -320,10 +320,9 @@ retry_sample_id:
for (thread_index = 0; thread_index < thread_num; thread_index++) {
try_again:
fd[nr_cpu][counter][thread_index] = sys_perf_event_open(attr,
all_tids[thread_index], cpu, group_fd, 0);
FD(evsel, nr_cpu, thread_index) = sys_perf_event_open(attr, all_tids[thread_index], cpu, group_fd, 0);
if (fd[nr_cpu][counter][thread_index] < 0) {
if (FD(evsel, nr_cpu, thread_index) < 0) {
int err = errno;
if (err == EPERM || err == EACCES)
......@@ -360,7 +359,7 @@ try_again:
}
printf("\n");
error("sys_perf_event_open() syscall returned with %d (%s). /bin/dmesg may provide additional information.\n",
fd[nr_cpu][counter][thread_index], strerror(err));
FD(evsel, nr_cpu, thread_index), strerror(err));
#if defined(__i386__) || defined(__x86_64__)
if (attr->type == PERF_TYPE_HARDWARE && err == EOPNOTSUPP)
......@@ -374,7 +373,7 @@ try_again:
exit(-1);
}
h_attr = get_header_attr(attr, counter);
h_attr = get_header_attr(attr, evsel->idx);
if (h_attr == NULL)
die("nomem\n");
......@@ -385,7 +384,7 @@ try_again:
}
}
if (read(fd[nr_cpu][counter][thread_index], &read_data, sizeof(read_data)) == -1) {
if (read(FD(evsel, nr_cpu, thread_index), &read_data, sizeof(read_data)) == -1) {
perror("Unable to read perf file descriptor");
exit(-1);
}
......@@ -395,43 +394,44 @@ try_again:
exit(-1);
}
assert(fd[nr_cpu][counter][thread_index] >= 0);
fcntl(fd[nr_cpu][counter][thread_index], F_SETFL, O_NONBLOCK);
assert(FD(evsel, nr_cpu, thread_index) >= 0);
fcntl(FD(evsel, nr_cpu, thread_index), F_SETFL, O_NONBLOCK);
/*
* First counter acts as the group leader:
*/
if (group && group_fd == -1)
group_fd = fd[nr_cpu][counter][thread_index];
if (counter || thread_index) {
ret = ioctl(fd[nr_cpu][counter][thread_index],
PERF_EVENT_IOC_SET_OUTPUT,
fd[nr_cpu][0][0]);
group_fd = FD(evsel, nr_cpu, thread_index);
if (evsel->idx || thread_index) {
struct perf_evsel *first;
first = list_entry(evsel_list.next, struct perf_evsel, node);
ret = ioctl(FD(evsel, nr_cpu, thread_index),
PERF_EVENT_IOC_SET_OUTPUT,
FD(first, nr_cpu, 0));
if (ret) {
error("failed to set output: %d (%s)\n", errno,
strerror(errno));
exit(-1);
}
} else {
mmap_array[nr_cpu].counter = counter;
mmap_array[nr_cpu].prev = 0;
mmap_array[nr_cpu].mask = mmap_pages*page_size - 1;
mmap_array[nr_cpu].base = mmap(NULL, (mmap_pages+1)*page_size,
PROT_READ|PROT_WRITE, MAP_SHARED, fd[nr_cpu][counter][thread_index], 0);
PROT_READ | PROT_WRITE, MAP_SHARED, FD(evsel, nr_cpu, thread_index), 0);
if (mmap_array[nr_cpu].base == MAP_FAILED) {
error("failed to mmap with %d (%s)\n", errno, strerror(errno));
exit(-1);
}
event_array[nr_poll].fd = fd[nr_cpu][counter][thread_index];
event_array[nr_poll].fd = FD(evsel, nr_cpu, thread_index);
event_array[nr_poll].events = POLLIN;
nr_poll++;
}
if (filter != NULL) {
ret = ioctl(fd[nr_cpu][counter][thread_index],
PERF_EVENT_IOC_SET_FILTER, filter);
ret = ioctl(FD(evsel, nr_cpu, thread_index),
PERF_EVENT_IOC_SET_FILTER, filter);
if (ret) {
error("failed to set filter with %d (%s)\n", errno,
strerror(errno));
......@@ -446,11 +446,12 @@ try_again:
static void open_counters(int cpu)
{
int counter;
struct perf_evsel *pos;
group_fd = -1;
for (counter = 0; counter < nr_counters; counter++)
create_counter(counter, cpu);
list_for_each_entry(pos, &evsel_list, node)
create_counter(pos, cpu);
nr_cpu++;
}
......@@ -537,7 +538,7 @@ static void mmap_read_all(void)
static int __cmd_record(int argc, const char **argv)
{
int i, counter;
int i;
struct stat st;
int flags;
int err;
......@@ -604,7 +605,7 @@ static int __cmd_record(int argc, const char **argv)
goto out_delete_session;
}
if (have_tracepoints(attrs, nr_counters))
if (have_tracepoints(&evsel_list))
perf_header__set_feat(&session->header, HEADER_TRACE_INFO);
/*
......@@ -666,12 +667,6 @@ static int __cmd_record(int argc, const char **argv)
close(child_ready_pipe[0]);
}
nr_cpus = read_cpu_map(cpu_list);
if (nr_cpus < 1) {
perror("failed to collect number of CPUs");
return -1;
}
if (!system_wide && no_inherit && !cpu_list) {
open_counters(-1);
} else {
......@@ -711,7 +706,7 @@ static int __cmd_record(int argc, const char **argv)
return err;
}
if (have_tracepoints(attrs, nr_counters)) {
if (have_tracepoints(&evsel_list)) {
/*
* FIXME err <= 0 here actually means that
* there were no tracepoints so its not really
......@@ -720,8 +715,7 @@ static int __cmd_record(int argc, const char **argv)
* return this more properly and also
* propagate errors that now are calling die()
*/
err = event__synthesize_tracing_data(output, attrs,
nr_counters,
err = event__synthesize_tracing_data(output, &evsel_list,
process_synthesized_event,
session);
if (err <= 0) {
......@@ -795,13 +789,13 @@ static int __cmd_record(int argc, const char **argv)
if (done) {
for (i = 0; i < nr_cpu; i++) {
for (counter = 0;
counter < nr_counters;
counter++) {
struct perf_evsel *pos;
list_for_each_entry(pos, &evsel_list, node) {
for (thread = 0;
thread < thread_num;
thread++)
ioctl(fd[i][counter][thread],
ioctl(FD(pos, i, thread),
PERF_EVENT_IOC_DISABLE);
}
}
......@@ -887,7 +881,8 @@ const struct option record_options[] = {
int cmd_record(int argc, const char **argv, const char *prefix __used)
{
int i, j, err = -ENOMEM;
int err = -ENOMEM;
struct perf_evsel *pos;
argc = parse_options(argc, argv, record_options, record_usage,
PARSE_OPT_STOP_AT_NON_OPTION);
......@@ -910,10 +905,9 @@ int cmd_record(int argc, const char **argv, const char *prefix __used)
if (no_buildid_cache || no_buildid)
disable_buildid_cache();
if (!nr_counters) {
nr_counters = 1;
attrs[0].type = PERF_TYPE_HARDWARE;
attrs[0].config = PERF_COUNT_HW_CPU_CYCLES;
if (list_empty(&evsel_list) && perf_evsel_list__create_default() < 0) {
pr_err("Not enough memory for event selector list\n");
goto out_symbol_exit;
}
if (target_pid != -1) {
......@@ -933,12 +927,15 @@ int cmd_record(int argc, const char **argv, const char *prefix __used)
thread_num = 1;
}
for (i = 0; i < MAX_NR_CPUS; i++) {
for (j = 0; j < MAX_COUNTERS; j++) {
fd[i][j] = malloc(sizeof(int)*thread_num);
if (!fd[i][j])
goto out_free_fd;
}
nr_cpus = read_cpu_map(cpu_list);
if (nr_cpus < 1) {
perror("failed to collect number of CPUs");
return -1;
}
list_for_each_entry(pos, &evsel_list, node) {
if (perf_evsel__alloc_fd(pos, nr_cpus, thread_num) < 0)
goto out_free_fd;
}
event_array = malloc(
sizeof(struct pollfd)*MAX_NR_CPUS*MAX_COUNTERS*thread_num);
......@@ -968,10 +965,8 @@ int cmd_record(int argc, const char **argv, const char *prefix __used)
out_free_event_array:
free(event_array);
out_free_fd:
for (i = 0; i < MAX_NR_CPUS; i++) {
for (j = 0; j < MAX_COUNTERS; j++)
free(fd[i][j]);
}
list_for_each_entry(pos, &evsel_list, node)
perf_evsel__free_fd(pos);
free(all_tids);
all_tids = NULL;
out_symbol_exit:
......
<
......@@ -43,6 +43,7 @@
#include "util/parse-options.h"
#include "util/parse-events.h"
#include "util/event.h"
#include "util/evsel.h"
#include "util/debug.h"
#include "util/header.h"
#include "util/cpumap.h"
......@@ -52,6 +53,8 @@
#include <math.h>
#include <locale.h>
#define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y))
#define DEFAULT_SEPARATOR " "
static struct perf_event_attr default_attrs[] = {
......@@ -90,16 +93,11 @@ static const char *cpu_list;
static const char *csv_sep = NULL;
static bool csv_output = false;
static int *fd[MAX_NR_CPUS][MAX_COUNTERS];
static int event_scaled[MAX_COUNTERS];
static struct {
struct cpu_counts {
u64 val;
u64 ena;
u64 run;
} cpu_counts[MAX_NR_CPUS][MAX_COUNTERS];
};
static volatile int done = 0;
......@@ -108,6 +106,26 @@ struct stats
double n, mean, M2;
};
struct perf_stat {
struct stats res_stats[3];
int scaled;
struct cpu_counts cpu_counts[];
};
static int perf_evsel__alloc_stat_priv(struct perf_evsel *evsel, int ncpus)
{
size_t priv_size = (sizeof(struct perf_stat) +
(ncpus * sizeof(struct cpu_counts)));
evsel->priv = zalloc(priv_size);
return evsel->priv == NULL ? -ENOMEM : 0;
}
static void perf_evsel__free_stat_priv(struct perf_evsel *evsel)
{
free(evsel->priv);
evsel->priv = NULL;
}
static void update_stats(struct stats *stats, u64 val)
{
double delta;
......@@ -147,22 +165,21 @@ static double stddev_stats(struct stats *stats)
return sqrt(variance_mean);
}
struct stats event_res_stats[MAX_COUNTERS][3];
struct stats runtime_nsecs_stats[MAX_NR_CPUS];
struct stats runtime_cycles_stats[MAX_NR_CPUS];
struct stats runtime_branches_stats[MAX_NR_CPUS];
struct stats walltime_nsecs_stats;
#define MATCH_EVENT(t, c, counter) \
(attrs[counter].type == PERF_TYPE_##t && \
attrs[counter].config == PERF_COUNT_##c)
#define MATCH_EVENT(t, c, evsel) \
(evsel->attr.type == PERF_TYPE_##t && \
evsel->attr.config == PERF_COUNT_##c)
#define ERR_PERF_OPEN \
"counter %d, sys_perf_event_open() syscall returned with %d (%s). /bin/dmesg may provide additional information."
static int create_perf_stat_counter(int counter, bool *perm_err)
static int create_perf_stat_counter(struct perf_evsel *evsel, bool *perm_err)
{
struct perf_event_attr *attr = attrs + counter;
struct perf_event_attr *attr = &evsel->attr;
int thread;
int ncreated = 0;
......@@ -174,13 +191,13 @@ static int create_perf_stat_counter(int counter, bool *perm_err)
int cpu;
for (cpu = 0; cpu < nr_cpus; cpu++) {
fd[cpu][counter][0] = sys_perf_event_open(attr,
FD(evsel, cpu, 0) = sys_perf_event_open(attr,
-1, cpumap[cpu], -1, 0);
if (fd[cpu][counter][0] < 0) {
if (FD(evsel, cpu, 0) < 0) {
if (errno == EPERM || errno == EACCES)
*perm_err = true;
error(ERR_PERF_OPEN, counter,
fd[cpu][counter][0], strerror(errno));
error(ERR_PERF_OPEN, evsel->idx,
FD(evsel, cpu, 0), strerror(errno));
} else {
++ncreated;
}
......@@ -192,13 +209,13 @@ static int create_perf_stat_counter(int counter, bool *perm_err)
attr->enable_on_exec = 1;
}
for (thread = 0; thread < thread_num; thread++) {
fd[0][counter][thread] = sys_perf_event_open(attr,
FD(evsel, 0, thread) = sys_perf_event_open(attr,
all_tids[thread], -1, -1, 0);
if (fd[0][counter][thread] < 0) {
if (FD(evsel, 0, thread) < 0) {
if (errno == EPERM || errno == EACCES)
*perm_err = true;
error(ERR_PERF_OPEN, counter,
fd[0][counter][thread],
error(ERR_PERF_OPEN, evsel->idx,
FD(evsel, 0, thread),
strerror(errno));
} else {
++ncreated;
......@@ -212,7 +229,7 @@ static int create_perf_stat_counter(int counter, bool *perm_err)
/*
* Does the counter have nsecs as a unit?
*/
static inline int nsec_counter(int counter)
static inline int nsec_counter(struct perf_evsel *counter)
{
if (MATCH_EVENT(SOFTWARE, SW_CPU_CLOCK, counter) ||
MATCH_EVENT(SOFTWARE, SW_TASK_CLOCK, counter))
......@@ -225,8 +242,9 @@ static inline int nsec_counter(int counter)
* Read out the results of a single counter:
* aggregate counts across CPUs in system-wide mode
*/
static void read_counter_aggr(int counter)
static void read_counter_aggr(struct perf_evsel *counter)
{
struct perf_stat *ps = counter->priv;
u64 count[3], single_count[3];
int cpu;
size_t res, nv;
......@@ -238,15 +256,15 @@ static void read_counter_aggr(int counter)
nv = scale ? 3 : 1;
for (cpu = 0; cpu < nr_cpus; cpu++) {
for (thread = 0; thread < thread_num; thread++) {
if (fd[cpu][counter][thread] < 0)
if (FD(counter, cpu, thread) < 0)
continue;
res = read(fd[cpu][counter][thread],
res = read(FD(counter, cpu, thread),
single_count, nv * sizeof(u64));
assert(res == nv * sizeof(u64));
close(fd[cpu][counter][thread]);
fd[cpu][counter][thread] = -1;
close(FD(counter, cpu, thread));
FD(counter, cpu, thread) = -1;
count[0] += single_count[0];
if (scale) {
......@@ -259,20 +277,20 @@ static void read_counter_aggr(int counter)
scaled = 0;
if (scale) {
if (count[2] == 0) {
event_scaled[counter] = -1;
ps->scaled = -1;
count[0] = 0;
return;
}
if (count[2] < count[1]) {
event_scaled[counter] = 1;
ps->scaled = 1;
count[0] = (unsigned long long)
((double)count[0] * count[1] / count[2] + 0.5);
}
}
for (i = 0; i < 3; i++)
update_stats(&event_res_stats[counter][i], count[i]);
update_stats(&ps->res_stats[i], count[i]);
if (verbose) {
fprintf(stderr, "%s: %Ld %Ld %Ld\n", event_name(counter),
......@@ -294,8 +312,9 @@ static void read_counter_aggr(int counter)
* Read out the results of a single counter:
* do not aggregate counts across CPUs in system-wide mode
*/
static void read_counter(int counter)
static void read_counter(struct perf_evsel *counter)
{
struct cpu_counts *cpu_counts = counter->priv;
u64 count[3];
int cpu;
size_t res, nv;
......@@ -306,15 +325,15 @@ static void read_counter(int counter)
for (cpu = 0; cpu < nr_cpus; cpu++) {
if (fd[cpu][counter][0] < 0)
if (FD(counter, cpu, 0) < 0)
continue;
res = read(fd[cpu][counter][0], count, nv * sizeof(u64));
res = read(FD(counter, cpu, 0), count, nv * sizeof(u64));
assert(res == nv * sizeof(u64));
close(fd[cpu][counter][0]);
fd[cpu][counter][0] = -1;
close(FD(counter, cpu, 0));
FD(counter, cpu, 0) = -1;
if (scale) {
if (count[2] == 0) {
......@@ -324,9 +343,9 @@ static void read_counter(int counter)
((double)count[0] * count[1] / count[2] + 0.5);
}
}
cpu_counts[cpu][counter].val = count[0]; /* scaled count */
cpu_counts[cpu][counter].ena = count[1];
cpu_counts[cpu][counter].run = count[2];
cpu_counts[cpu].val = count[0]; /* scaled count */
cpu_counts[cpu].ena = count[1];
cpu_counts[cpu].run = count[2];
if (MATCH_EVENT(SOFTWARE, SW_TASK_CLOCK, counter))
update_stats(&runtime_nsecs_stats[cpu], count[0]);
......@@ -340,8 +359,9 @@ static void read_counter(int counter)
static int run_perf_stat(int argc __used, const char **argv)
{
unsigned long long t0, t1;
struct perf_evsel *counter;
int status = 0;
int counter, ncreated = 0;
int ncreated = 0;
int child_ready_pipe[2], go_pipe[2];
bool perm_err = false;
const bool forks = (argc > 0);
......@@ -401,7 +421,7 @@ static int run_perf_stat(int argc __used, const char **argv)
close(child_ready_pipe[0]);
}
for (counter = 0; counter < nr_counters; counter++)
list_for_each_entry(counter, &evsel_list, node)
ncreated += create_perf_stat_counter(counter, &perm_err);
if (ncreated < nr_counters) {
......@@ -433,25 +453,28 @@ static int run_perf_stat(int argc __used, const char **argv)
update_stats(&walltime_nsecs_stats, t1 - t0);
if (no_aggr) {
for (counter = 0; counter < nr_counters; counter++)
list_for_each_entry(counter, &evsel_list, node)
read_counter(counter);
} else {
for (counter = 0; counter < nr_counters; counter++)
list_for_each_entry(counter, &evsel_list, node)
read_counter_aggr(counter);
}
return WEXITSTATUS(status);
}
static void print_noise(int counter, double avg)
static void print_noise(struct perf_evsel *evsel, double avg)
{
struct perf_stat *ps;
if (run_count == 1)
return;
ps = evsel->priv;
fprintf(stderr, " ( +- %7.3f%% )",
100 * stddev_stats(&event_res_stats[counter][0]) / avg);
100 * stddev_stats(&ps->res_stats[0]) / avg);
}
static void nsec_printout(int cpu, int counter, double avg)
static void nsec_printout(int cpu, struct perf_evsel *counter, double avg)
{
double msecs = avg / 1e6;
char cpustr[16] = { '\0', };
......@@ -473,7 +496,7 @@ static void nsec_printout(int cpu, int counter, double avg)
}
}
static void abs_printout(int cpu, int counter, double avg)
static void abs_printout(int cpu, struct perf_evsel *counter, double avg)
{
double total, ratio = 0.0;
char cpustr[16] = { '\0', };
......@@ -528,10 +551,11 @@ static void abs_printout(int cpu, int counter, double avg)
* Print out the results of a single counter:
* aggregated counts in system-wide mode
*/