perf test: Add test for the evlist mmap routines
authorArnaldo Carvalho de Melo <acme@redhat.com>
Sat, 15 Jan 2011 12:42:46 +0000 (10:42 -0200)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Sat, 22 Jan 2011 21:56:31 +0000 (19:56 -0200)
This test will generate random numbers of calls to some getpid syscalls,
then establish an mmap for a group of events that are created to monitor
these syscalls.

It will receive the events, using mmap, use its PERF_SAMPLE_ID generated
sample.id field to map back to its respective perf_evsel instance.

Then it checks if the number of syscalls reported as perf events by the
kernel corresponds to the number of syscalls made.

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: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/builtin-test.c

index 5a50e47..4fd3453 100644 (file)
@@ -7,7 +7,9 @@
 
 #include "util/cache.h"
 #include "util/debug.h"
+#include "util/evlist.h"
 #include "util/parse-options.h"
+#include "util/parse-events.h"
 #include "util/session.h"
 #include "util/symbol.h"
 #include "util/thread.h"
@@ -238,14 +240,14 @@ out:
 #include "util/evsel.h"
 #include <sys/types.h>
 
-static int trace_event__id(const char *event_name)
+static int trace_event__id(const char *evname)
 {
        char *filename;
        int err = -1, fd;
 
        if (asprintf(&filename,
                     "/sys/kernel/debug/tracing/events/syscalls/%s/id",
-                    event_name) < 0)
+                    evname) < 0)
                return -1;
 
        fd = open(filename, O_RDONLY);
@@ -439,6 +441,167 @@ out_thread_map_delete:
        return err;
 }
 
+/*
+ * This test will generate random numbers of calls to some getpid syscalls,
+ * then establish an mmap for a group of events that are created to monitor
+ * the syscalls.
+ *
+ * It will receive the events, using mmap, use its PERF_SAMPLE_ID generated
+ * sample.id field to map back to its respective perf_evsel instance.
+ *
+ * Then it checks if the number of syscalls reported as perf events by
+ * the kernel corresponds to the number of syscalls made.
+ */
+static int test__basic_mmap(void)
+{
+       int err = -1;
+       event_t *event;
+       struct thread_map *threads;
+       struct perf_session session;
+       struct cpu_map *cpus;
+       struct perf_evlist *evlist;
+       struct perf_event_attr attr = {
+               .type           = PERF_TYPE_TRACEPOINT,
+               .read_format    = PERF_FORMAT_ID,
+               .sample_type    = PERF_SAMPLE_ID,
+               .watermark      = 0,
+       };
+       cpu_set_t cpu_set;
+       const char *syscall_names[] = { "getsid", "getppid", "getpgrp",
+                                       "getpgid", };
+       pid_t (*syscalls[])(void) = { (void *)getsid, getppid, getpgrp,
+                                     (void*)getpgid };
+#define nsyscalls ARRAY_SIZE(syscall_names)
+       int ids[nsyscalls];
+       unsigned int nr_events[nsyscalls],
+                    expected_nr_events[nsyscalls], i, j;
+       struct perf_evsel *evsels[nsyscalls], *evsel;
+
+       for (i = 0; i < nsyscalls; ++i) {
+               char name[64];
+
+               snprintf(name, sizeof(name), "sys_enter_%s", syscall_names[i]);
+               ids[i] = trace_event__id(name);
+               if (ids[i] < 0) {
+                       pr_debug("Is debugfs mounted on /sys/kernel/debug?\n");
+                       return -1;
+               }
+               nr_events[i] = 0;
+               expected_nr_events[i] = random() % 257;
+       }
+
+       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");
+               goto out_free_threads;
+       }
+
+       CPU_ZERO(&cpu_set);
+       CPU_SET(cpus->map[0], &cpu_set);
+       sched_setaffinity(0, sizeof(cpu_set), &cpu_set);
+       if (sched_setaffinity(0, sizeof(cpu_set), &cpu_set) < 0) {
+               pr_debug("sched_setaffinity() failed on CPU %d: %s ",
+                        cpus->map[0], strerror(errno));
+               goto out_free_cpus;
+       }
+
+       evlist = perf_evlist__new();
+       if (threads == NULL) {
+               pr_debug("perf_evlist__new\n");
+               goto out_free_cpus;
+       }
+
+       /* anonymous union fields, can't be initialized above */
+       attr.wakeup_events = 1;
+       attr.sample_period = 1;
+
+       /*
+        * FIXME: use evsel->attr.sample_type in event__parse_sample.
+        *        This will nicely remove the requirement that we have
+        *        all the events with the same sample_type.
+        */
+       session.sample_type = attr.sample_type;
+
+       for (i = 0; i < nsyscalls; ++i) {
+               attr.config = ids[i];
+               evsels[i] = perf_evsel__new(&attr, i);
+               if (evsels[i] == NULL) {
+                       pr_debug("perf_evsel__new\n");
+                       goto out_free_evlist;
+               }
+
+               perf_evlist__add(evlist, evsels[i]);
+
+               if (perf_evsel__open(evsels[i], cpus, threads, false, false) < 0) {
+                       pr_debug("failed to open counter: %s, "
+                                "tweak /proc/sys/kernel/perf_event_paranoid?\n",
+                                strerror(errno));
+                       goto out_close_fd;
+               }
+       }
+
+       if (perf_evlist__mmap(evlist, cpus, threads, 128, true) < 0) {
+               pr_debug("failed to mmap events: %d (%s)\n", errno,
+                        strerror(errno));
+               goto out_close_fd;
+       }
+
+       for (i = 0; i < nsyscalls; ++i)
+               for (j = 0; j < expected_nr_events[i]; ++j) {
+                       int foo = syscalls[i]();
+                       ++foo;
+               }
+
+       while ((event = perf_evlist__read_on_cpu(evlist, 0)) != NULL) {
+               struct sample_data sample;
+
+               if (event->header.type != PERF_RECORD_SAMPLE) {
+                       pr_debug("unexpected %s event\n",
+                                event__get_event_name(event->header.type));
+                       goto out_munmap;
+               }
+
+               event__parse_sample(event, &session, &sample);
+               evsel = perf_evlist__id2evsel(evlist, sample.id);
+               if (evsel == NULL) {
+                       pr_debug("event with id %" PRIu64
+                                " doesn't map to an evsel\n", sample.id);
+                       goto out_munmap;
+               }
+               nr_events[evsel->idx]++;
+       }
+
+       list_for_each_entry(evsel, &evlist->entries, node) {
+               if (nr_events[evsel->idx] != expected_nr_events[evsel->idx]) {
+                       pr_debug("expected %d %s events, got %d\n",
+                                expected_nr_events[evsel->idx],
+                                event_name(evsel), nr_events[evsel->idx]);
+                       goto out_munmap;
+               }
+       }
+
+       err = 0;
+out_munmap:
+       perf_evlist__munmap(evlist, 1);
+out_close_fd:
+       for (i = 0; i < nsyscalls; ++i)
+               perf_evsel__close_fd(evsels[i], 1, threads->nr);
+out_free_evlist:
+       perf_evlist__delete(evlist);
+out_free_cpus:
+       cpu_map__delete(cpus);
+out_free_threads:
+       thread_map__delete(threads);
+       return err;
+#undef nsyscalls
+}
+
 static struct test {
        const char *desc;
        int (*func)(void);
@@ -456,6 +619,10 @@ static struct test {
                .func = test__open_syscall_event_on_all_cpus,
        },
        {
+               .desc = "read samples using the mmap interface",
+               .func = test__basic_mmap,
+       },
+       {
                .func = NULL,
        },
 };