perf evsel: Support perf_evsel__open(cpus > 1 && threads > 1)

And a test for it:

[acme@felicio linux]$ perf test
 1: vmlinux symtab matches kallsyms: Ok
 2: detect open syscall event: Ok
 3: detect open syscall event on all cpus: Ok
[acme@felicio linux]$

Translating C the test does:

1. generates different number of open syscalls on each CPU
   by using sched_setaffinity
2. Verifies that the expected number of events is generated
   on each CPU

It works as expected.

LKML-Reference: <new-submission>
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>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

diff --git a/tools/perf/builtin-test.c b/tools/perf/builtin-test.c
index e12753f..ed56961 100644 (file)
--- a/tools/perf/builtin-test.c
+++ b/tools/perf/builtin-test.c
@@ -234,6 +234,7 @@ out:
        return err;
 }
 
+#include "util/cpumap.h"
 #include "util/evsel.h"
 #include <sys/types.h>
 
@@ -321,6 +322,111 @@ out_thread_map_delete:
        return err;
 }
 
+#include <sched.h>
+
+static int test__open_syscall_event_on_all_cpus(void)
+{
+       int err = -1, fd, cpu;
+       struct thread_map *threads;
+       struct cpu_map *cpus;
+       struct perf_evsel *evsel;
+       struct perf_event_attr attr;
+       unsigned int nr_open_calls = 111, i;
+       cpu_set_t *cpu_set;
+       size_t cpu_set_size;
+       int id = trace_event__id("sys_enter_open");
+
+       if (id < 0) {
+               pr_debug("is debugfs mounted on /sys/kernel/debug?\n");
+               return -1;
+       }
+
+       threads = thread_map__new(-1, getpid());
+       if (threads == NULL) {
+               pr_debug("thread_map__new\n");
+               return -1;
+       }
+
+       cpus = cpu_map__new(NULL);
+       if (threads == NULL) {
+               pr_debug("thread_map__new\n");
+               return -1;
+       }
+
+       cpu_set = CPU_ALLOC(cpus->nr);
+
+       if (cpu_set == NULL)
+               goto out_thread_map_delete;
+
+       cpu_set_size = CPU_ALLOC_SIZE(cpus->nr);
+       CPU_ZERO_S(cpu_set_size, cpu_set);
+
+       memset(&attr, 0, sizeof(attr));
+       attr.type = PERF_TYPE_TRACEPOINT;
+       attr.config = id;
+       evsel = perf_evsel__new(&attr, 0);
+       if (evsel == NULL) {
+               pr_debug("perf_evsel__new\n");
+               goto out_cpu_free;
+       }
+
+       if (perf_evsel__open(evsel, cpus, threads) < 0) {
+               pr_debug("failed to open counter: %s, "
+                        "tweak /proc/sys/kernel/perf_event_paranoid?\n",
+                        strerror(errno));
+               goto out_evsel_delete;
+       }
+
+       for (cpu = 0; cpu < cpus->nr; ++cpu) {
+               unsigned int ncalls = nr_open_calls + cpu;
+
+               CPU_SET(cpu, cpu_set);
+               sched_setaffinity(0, cpu_set_size, cpu_set);
+               for (i = 0; i < ncalls; ++i) {
+                       fd = open("/etc/passwd", O_RDONLY);
+                       close(fd);
+               }
+               CPU_CLR(cpu, cpu_set);
+       }
+
+       /*
+        * Here we need to explicitely preallocate the counts, as if
+        * we use the auto allocation it will allocate just for 1 cpu,
+        * as we start by cpu 0.
+        */
+       if (perf_evsel__alloc_counts(evsel, cpus->nr) < 0) {
+               pr_debug("perf_evsel__alloc_counts(ncpus=%d)\n", cpus->nr);
+               goto out_close_fd;
+       }
+
+       for (cpu = 0; cpu < cpus->nr; ++cpu) {
+               unsigned int expected;
+
+               if (perf_evsel__read_on_cpu(evsel, cpu, 0) < 0) {
+                       pr_debug("perf_evsel__open_read_on_cpu\n");
+                       goto out_close_fd;
+               }
+
+               expected = nr_open_calls + cpu;
+               if (evsel->counts->cpu[cpu].val != expected) {
+                       pr_debug("perf_evsel__read_on_cpu: expected to intercept %d calls on cpu %d, got %Ld\n",
+                                expected, cpu, evsel->counts->cpu[cpu].val);
+                       goto out_close_fd;
+               }
+       }
+
+       err = 0;
+out_close_fd:
+       perf_evsel__close_fd(evsel, 1, threads->nr);
+out_evsel_delete:
+       perf_evsel__delete(evsel);
+out_cpu_free:
+       CPU_FREE(cpu_set);
+out_thread_map_delete:
+       thread_map__delete(threads);
+       return err;
+}
+
 static struct test {
        const char *desc;
        int (*func)(void);
@@ -334,6 +440,10 @@ static struct test {
                .func = test__open_syscall_event,
        },
        {
+               .desc = "detect open syscall event on all cpus",
+               .func = test__open_syscall_event_on_all_cpus,
+       },
+       {
                .func = NULL,
        },
 };

diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 1a5591d..f5cfed6 100644 (file)
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -127,59 +127,75 @@ int __perf_evsel__read(struct perf_evsel *evsel,
        return 0;
 }
 
-int perf_evsel__open_per_cpu(struct perf_evsel *evsel, struct cpu_map *cpus)
+static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
+                             struct thread_map *threads)
 {
-       int cpu;
+       int cpu, thread;
 
-       if (evsel->fd == NULL && perf_evsel__alloc_fd(evsel, cpus->nr, 1) < 0)
+       if (evsel->fd == NULL &&
+           perf_evsel__alloc_fd(evsel, cpus->nr, threads->nr) < 0)
                return -1;
 
        for (cpu = 0; cpu < cpus->nr; cpu++) {
-               FD(evsel, cpu, 0) = sys_perf_event_open(&evsel->attr, -1,
-                                                       cpus->map[cpu], -1, 0);
-               if (FD(evsel, cpu, 0) < 0)
-                       goto out_close;
+               for (thread = 0; thread < threads->nr; thread++) {
+                       FD(evsel, cpu, thread) = sys_perf_event_open(&evsel->attr,
+                                                                    threads->map[thread],
+                                                                    cpus->map[cpu], -1, 0);
+                       if (FD(evsel, cpu, thread) < 0)
+                               goto out_close;
+               }
        }
 
        return 0;
 
 out_close:
-       while (--cpu >= 0) {
-               close(FD(evsel, cpu, 0));
-               FD(evsel, cpu, 0) = -1;
-       }
+       do {
+               while (--thread >= 0) {
+                       close(FD(evsel, cpu, thread));
+                       FD(evsel, cpu, thread) = -1;
+               }
+               thread = threads->nr;
+       } while (--cpu >= 0);
        return -1;
 }
 
-int perf_evsel__open_per_thread(struct perf_evsel *evsel, struct thread_map *threads)
+static struct {
+       struct cpu_map map;
+       int cpus[1];
+} empty_cpu_map = {
+       .map.nr = 1,
+       .cpus   = { -1, },
+};
+
+static struct {
+       struct thread_map map;
+       int threads[1];
+} empty_thread_map = {
+       .map.nr  = 1,
+       .threads = { -1, },
+};
+
+int perf_evsel__open(struct perf_evsel *evsel,
+                    struct cpu_map *cpus, struct thread_map *threads)
 {
-       int thread;
-
-       if (evsel->fd == NULL && perf_evsel__alloc_fd(evsel, 1, threads->nr))
-               return -1;
 
-       for (thread = 0; thread < threads->nr; thread++) {
-               FD(evsel, 0, thread) = sys_perf_event_open(&evsel->attr,
-                                                          threads->map[thread], -1, -1, 0);
-               if (FD(evsel, 0, thread) < 0)
-                       goto out_close;
+       if (cpus == NULL) {
+               /* Work around old compiler warnings about strict aliasing */
+               cpus = &empty_cpu_map.map;
        }
 
-       return 0;
+       if (threads == NULL)
+               threads = &empty_thread_map.map;
 
-out_close:
-       while (--thread >= 0) {
-               close(FD(evsel, 0, thread));
-               FD(evsel, 0, thread) = -1;
-       }
-       return -1;
+       return __perf_evsel__open(evsel, cpus, threads);
 }
 
-int perf_evsel__open(struct perf_evsel *evsel, 
-                    struct cpu_map *cpus, struct thread_map *threads)
+int perf_evsel__open_per_cpu(struct perf_evsel *evsel, struct cpu_map *cpus)
 {
-       if (threads == NULL)
-               return perf_evsel__open_per_cpu(evsel, cpus);
+       return __perf_evsel__open(evsel, cpus, &empty_thread_map.map);
+}
 
-       return perf_evsel__open_per_thread(evsel, threads);
+int perf_evsel__open_per_thread(struct perf_evsel *evsel, struct thread_map *threads)
+{
+       return __perf_evsel__open(evsel, &empty_cpu_map.map, threads);
 }