perf bench: Fix numa report output code
authorJiri Olsa <jolsa@kernel.org>
Wed, 20 Jun 2018 09:40:36 +0000 (11:40 +0200)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Mon, 25 Jun 2018 14:59:37 +0000 (11:59 -0300)
Currently we can hit following assert when running numa bench:

  $ perf bench numa mem -p 3 -t 1 -P 512 -s 100 -zZ0cm --thp 1
  perf: bench/numa.c:1577: __bench_numa: Assertion `!(!(((wait_stat) & 0x7f) == 0))' failed.

The assertion is correct, because we hit the SIGFPE in following line:

  Thread 2.2 "thread 0/0" received signal SIGFPE, Arithmetic exception.
  [Switching to Thread 0x7fffd28c6700 (LWP 11750)]
  0x000.. in worker_thread (__tdata=0x7.. ) at bench/numa.c:1257
  1257 td->speed_gbs = bytes_done / (td->runtime_ns / NSEC_PER_SEC) / 1e9;

We don't check if the runtime is actually bigger than 1 second,
and thus this might end up with zero division within FPU.

Adding the check to prevent this.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180620094036.17278-1-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/bench/numa.c

index 63eb49082774c94dfbabe7a18db73bdc2403fb6a..44195514b19e65a5ee0287b48fa0ab25fa44d66f 100644 (file)
@@ -1098,7 +1098,7 @@ static void *worker_thread(void *__tdata)
        u8 *global_data;
        u8 *process_data;
        u8 *thread_data;
-       u64 bytes_done;
+       u64 bytes_done, secs;
        long work_done;
        u32 l;
        struct rusage rusage;
@@ -1254,7 +1254,8 @@ static void *worker_thread(void *__tdata)
        timersub(&stop, &start0, &diff);
        td->runtime_ns = diff.tv_sec * NSEC_PER_SEC;
        td->runtime_ns += diff.tv_usec * NSEC_PER_USEC;
-       td->speed_gbs = bytes_done / (td->runtime_ns / NSEC_PER_SEC) / 1e9;
+       secs = td->runtime_ns / NSEC_PER_SEC;
+       td->speed_gbs = secs ? bytes_done / secs / 1e9 : 0;
 
        getrusage(RUSAGE_THREAD, &rusage);
        td->system_time_ns = rusage.ru_stime.tv_sec * NSEC_PER_SEC;