linux/tools/perf/util/perf_api_probe.c
Namhyung Kim 4f2abe9192 perf record: Move probing cgroup sampling support
I found that checking cgroup sampling support using the missing features
doesn't work on old kernels.  Because it added both attr.cgroup bit and
PERF_SAMPLE_CGROUP bit, it needs to check whichever comes first (usually
the actual event, not dummy).

But it only checks the attr.cgroup bit which is set only in the dummy
event so cannot detect failtures due the sample bits.  Also we don't
ignore the missing feature and retry, it'd be better checking it with
the API probing logic.

Committer notes:

Extracted the minimal part to check using the new cgroup API probe
routine, the part that removes the cgroup member can be left for further
discussion.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lore.kernel.org/lkml/20210527182835.1634339-1-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2021-06-01 10:32:00 -03:00

195 lines
3.8 KiB
C

/* SPDX-License-Identifier: GPL-2.0 */
#include "perf-sys.h"
#include "util/cloexec.h"
#include "util/evlist.h"
#include "util/evsel.h"
#include "util/parse-events.h"
#include "util/perf_api_probe.h"
#include <perf/cpumap.h>
#include <errno.h>
typedef void (*setup_probe_fn_t)(struct evsel *evsel);
static int perf_do_probe_api(setup_probe_fn_t fn, int cpu, const char *str)
{
struct evlist *evlist;
struct evsel *evsel;
unsigned long flags = perf_event_open_cloexec_flag();
int err = -EAGAIN, fd;
static pid_t pid = -1;
evlist = evlist__new();
if (!evlist)
return -ENOMEM;
if (parse_events(evlist, str, NULL))
goto out_delete;
evsel = evlist__first(evlist);
while (1) {
fd = sys_perf_event_open(&evsel->core.attr, pid, cpu, -1, flags);
if (fd < 0) {
if (pid == -1 && errno == EACCES) {
pid = 0;
continue;
}
goto out_delete;
}
break;
}
close(fd);
fn(evsel);
fd = sys_perf_event_open(&evsel->core.attr, pid, cpu, -1, flags);
if (fd < 0) {
if (errno == EINVAL)
err = -EINVAL;
goto out_delete;
}
close(fd);
err = 0;
out_delete:
evlist__delete(evlist);
return err;
}
static bool perf_probe_api(setup_probe_fn_t fn)
{
const char *try[] = {"cycles:u", "instructions:u", "cpu-clock:u", NULL};
struct perf_cpu_map *cpus;
int cpu, ret, i = 0;
cpus = perf_cpu_map__new(NULL);
if (!cpus)
return false;
cpu = cpus->map[0];
perf_cpu_map__put(cpus);
do {
ret = perf_do_probe_api(fn, cpu, try[i++]);
if (!ret)
return true;
} while (ret == -EAGAIN && try[i]);
return false;
}
static void perf_probe_sample_identifier(struct evsel *evsel)
{
evsel->core.attr.sample_type |= PERF_SAMPLE_IDENTIFIER;
}
static void perf_probe_comm_exec(struct evsel *evsel)
{
evsel->core.attr.comm_exec = 1;
}
static void perf_probe_context_switch(struct evsel *evsel)
{
evsel->core.attr.context_switch = 1;
}
static void perf_probe_text_poke(struct evsel *evsel)
{
evsel->core.attr.text_poke = 1;
}
static void perf_probe_build_id(struct evsel *evsel)
{
evsel->core.attr.build_id = 1;
}
static void perf_probe_cgroup(struct evsel *evsel)
{
evsel->core.attr.cgroup = 1;
}
bool perf_can_sample_identifier(void)
{
return perf_probe_api(perf_probe_sample_identifier);
}
bool perf_can_comm_exec(void)
{
return perf_probe_api(perf_probe_comm_exec);
}
bool perf_can_record_switch_events(void)
{
return perf_probe_api(perf_probe_context_switch);
}
bool perf_can_record_text_poke_events(void)
{
return perf_probe_api(perf_probe_text_poke);
}
bool perf_can_record_cpu_wide(void)
{
struct perf_event_attr attr = {
.type = PERF_TYPE_SOFTWARE,
.config = PERF_COUNT_SW_CPU_CLOCK,
.exclude_kernel = 1,
};
struct perf_cpu_map *cpus;
int cpu, fd;
cpus = perf_cpu_map__new(NULL);
if (!cpus)
return false;
cpu = cpus->map[0];
perf_cpu_map__put(cpus);
fd = sys_perf_event_open(&attr, -1, cpu, -1, 0);
if (fd < 0)
return false;
close(fd);
return true;
}
/*
* Architectures are expected to know if AUX area sampling is supported by the
* hardware. Here we check for kernel support.
*/
bool perf_can_aux_sample(void)
{
struct perf_event_attr attr = {
.size = sizeof(struct perf_event_attr),
.exclude_kernel = 1,
/*
* Non-zero value causes the kernel to calculate the effective
* attribute size up to that byte.
*/
.aux_sample_size = 1,
};
int fd;
fd = sys_perf_event_open(&attr, -1, 0, -1, 0);
/*
* If the kernel attribute is big enough to contain aux_sample_size
* then we assume that it is supported. We are relying on the kernel to
* validate the attribute size before anything else that could be wrong.
*/
if (fd < 0 && errno == E2BIG)
return false;
if (fd >= 0)
close(fd);
return true;
}
bool perf_can_record_build_id(void)
{
return perf_probe_api(perf_probe_build_id);
}
bool perf_can_record_cgroup(void)
{
return perf_probe_api(perf_probe_cgroup);
}