perf record: Use an eventfd to wakeup when done

The setting and checking of 'done' contains a rare race where the signal
handler setting 'done' is run after checking to break the loop, but
before waiting in evlist__poll(). In this case, the main loop won't wake
up until either another signal is sent, or the perf data fd causes a
wake up.

The following simple script can trigger this condition (but you might
need to run it for several hours):

for ((i = 0; i >= 0; i++)) ; do
  echo "Loop $i"
  delay=$(echo "scale=4; 0.1 * $RANDOM/32768" | bc)
  ./perf record -- sleep 30000000 >/dev/null&
  pid=$!
  sleep $delay
  kill -TERM $pid
  echo "PID $pid"
  wait $pid
done

At some point, the loop will stall. Adding logging, even though perf has
received the SIGTERM and set 'done = 1', perf will remain sleeping until
a second signal is sent.

Committer notes:

Make this dependent on HAVE_EVENTFD_SUPPORT, so that we continue
building on older systems without the eventfd syscall.
Signed-off-by: Anand K Mistry <amistry@google.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lore.kernel.org/lkml/20200513122012.v3.1.I4d7421c6bbb1f83ea58419082481082e19097841@changeidSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools/perf/builtin-record.c

@@ -56,6 +56,9 @@
 #include <unistd.h>
 #include <sched.h>
 #include <signal.h>
+#ifdef HAVE_EVENTFD_SUPPORT
+#include <sys/eventfd.h>
+#endif
 #include <sys/mman.h>
 #include <sys/wait.h>
 #include <sys/types.h>
@@ -538,6 +541,9 @@ static int record__pushfn(struct mmap *map, void *to, void *bf, size_t size)
 
 static volatile int signr = -1;
 static volatile int child_finished;
+#ifdef HAVE_EVENTFD_SUPPORT
+static int done_fd = -1;
+#endif
 
 static void sig_handler(int sig)
 {
@@ -547,6 +553,21 @@ static void sig_handler(int sig)
 		signr = sig;
 
 	done = 1;
+#ifdef HAVE_EVENTFD_SUPPORT
+{
+	u64 tmp = 1;
+	/*
+	 * It is possible for this signal handler to run after done is checked
+	 * in the main loop, but before the perf counter fds are polled. If this
+	 * happens, the poll() will continue to wait even though done is set,
+	 * and will only break out if either another signal is received, or the
+	 * counters are ready for read. To ensure the poll() doesn't sleep when
+	 * done is set, use an eventfd (done_fd) to wake up the poll().
+	 */
+	if (write(done_fd, &tmp, sizeof(tmp)) < 0)
+		pr_err("failed to signal wakeup fd, error: %m\n");
+}
+#endif // HAVE_EVENTFD_SUPPORT
 }
 
 static void sigsegv_handler(int sig)
@@ -1547,6 +1568,20 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
 		pr_err("Compression initialization failed.\n");
 		return -1;
 	}
+#ifdef HAVE_EVENTFD_SUPPORT
+	done_fd = eventfd(0, EFD_NONBLOCK);
+	if (done_fd < 0) {
+		pr_err("Failed to create wakeup eventfd, error: %m\n");
+		status = -1;
+		goto out_delete_session;
+	}
+	err = evlist__add_pollfd(rec->evlist, done_fd);
+	if (err < 0) {
+		pr_err("Failed to add wakeup eventfd to poll list\n");
+		status = err;
+		goto out_delete_session;
+	}
+#endif // HAVE_EVENTFD_SUPPORT
 
 	session->header.env.comp_type  = PERF_COMP_ZSTD;
 	session->header.env.comp_level = rec->opts.comp_level;
@@ -1905,6 +1940,10 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
 	}
 
 out_delete_session:
+#ifdef HAVE_EVENTFD_SUPPORT
+	if (done_fd >= 0)
+		close(done_fd);
+#endif
 	zstd_fini(&session->zstd_data);
 	perf_session__delete(session);