Re: rcutorture: Add RCU grace-period performance tests

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Wed, Feb 03, 2016 at 07:34:08AM +0300, Dan Carpenter wrote:
> Hello Paul E. McKenney,
> 
> This is a semi-automatic email about new static checker warnings.
> 
> The patch 7f68f317a3d9: "rcutorture: Add RCU grace-period performance 
> tests" from Dec 31, 2015, leads to the following Smatch complaint:
> 
> kernel/rcu/rcuperf.c:486 rcu_perf_cleanup()
> 	 error: we previously assumed 'writer_n_durations' could be null (see line 465)
> 
> kernel/rcu/rcuperf.c
>    442  static void
>    443  rcu_perf_cleanup(void)
>    444  {
>    445          int i;
>    446          int j;
>    447          int ngps = 0;
>    448          u64 *wdp;
>    449          u64 *wdpp;
>    450  
>    451          if (torture_cleanup_begin())
>    452                  return;
>    453  
>    454          if (reader_tasks) {
>    455                  for (i = 0; i < nrealreaders; i++)
>    456                          torture_stop_kthread(rcu_perf_reader,
>    457                                               reader_tasks[i]);
>    458                  kfree(reader_tasks);
>    459          }
>    460  
>    461          if (writer_tasks) {
>    462                  for (i = 0; i < nrealwriters; i++) {
>    463                          torture_stop_kthread(rcu_perf_writer,
>    464						     writer_tasks[i]);
>    465				if (!writer_n_durations)
>                                      ^^^^^^^^^^^^^^^^^^
> Can be NULL here.
> 
>    466					continue;
>    467				j = writer_n_durations[i];
>    468				pr_alert("%s%s writer %d gps: %d\n",
>    469					 perf_type, PERF_FLAG, i, j);
>    470				ngps += j;
>    471			}
>    472			pr_alert("%s%s start: %llu end: %llu duration: %llu gps: %d batches: %ld\n",
>    473				 perf_type, PERF_FLAG,
>    474				 t_rcu_perf_writer_started, t_rcu_perf_writer_finished,
>    475				 t_rcu_perf_writer_finished -
>    476				 t_rcu_perf_writer_started,
>    477				 ngps,
>    478				 b_rcu_perf_writer_finished -
>    479				 b_rcu_perf_writer_started);
>    480			for (i = 0; i < nrealwriters; i++) {
>    481				if (!writer_durations)
>    482					break;
>    483				wdpp = writer_durations[i];
>    484				if (!wdpp)
>    485					continue;
>    486				for (j = 0; j <= writer_n_durations[i]; j++) {
>                                                  ^^^^^^^^^^^^^^^^^^^^^
> Unchecked dereference.
> 
>    487					wdp = &wdpp[j];
>    488					pr_alert("%s%s %4d writer-duration: %5d %llu\n",
> 
> regards,
> dan carpenter

Good catch!  Would the following cover it?

							Thanx, Paul

------------------------------------------------------------------------
 
diff --git a/kernel/rcu/rcuperf.c b/kernel/rcu/rcuperf.c
index c110298b76e6..1d88f1347301 100644
--- a/kernel/rcu/rcuperf.c
+++ b/kernel/rcu/rcuperf.c
@@ -489,6 +489,8 @@ rcu_perf_cleanup(void)
 		for (i = 0; i < nrealwriters; i++) {
 			if (!writer_durations)
 				break;
+			if (!writer_n_durations)
+				continue;
 			wdpp = writer_durations[i];
 			if (!wdpp)
 				continue;