perf bench futex, requeue: Add --pi parameter

This extends the program to measure WAIT_REQUEUE_PI+CMP_REQUEUE_PI
pairs, which are the underlying machinery behind priority-inheritance
aware condition variables. The defaults are the same as with the regular
non-pi version, requeueing one task at a time, with the exception that
PI will always wakeup the first waiter.

Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Cc: Davidlohr Bueso <dbueso@suse.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lore.kernel.org/lkml/20210809043301.66002-8-dave@stgolabs.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

1 files changed, 75 insertions, 25 deletions

diff --git a/tools/perf/bench/futex-requeue.c b/tools/perf/bench/futex-requeue.c
index e23a08037de2..97fe31fd3a23 100644
--- a/tools/perf/bench/futex-requeue.c
+++ b/tools/perf/bench/futex-requeue.c
@@ -6,7 +6,8 @@
  *                on futex2, N at a time.
  *
  * This program is particularly useful to measure the latency of nthread
- * requeues without waking up any tasks -- thus mimicking a regular futex_wait.
+ * requeues without waking up any tasks (in the non-pi case) -- thus
+ * mimicking a regular futex_wait.
  */
 
 /* For the CLR_() macros */
@@ -54,6 +55,8 @@ static const struct option options[] = {
 	OPT_BOOLEAN( 'S', "shared",   &params.fshared, "Use shared futexes instead of private ones"),
 	OPT_BOOLEAN( 'm', "mlockall", &params.mlockall, "Lock all current and future memory"),
 	OPT_BOOLEAN( 'B', "broadcast", &params.broadcast, "Requeue all threads at once"),
+	OPT_BOOLEAN( 'p', "pi", &params.pi, "Use PI-aware variants of FUTEX_CMP_REQUEUE"),
+
 	OPT_END()
 };
 
@@ -87,14 +90,30 @@ static void *workerfn(void *arg __maybe_unused)
 	pthread_mutex_unlock(&thread_lock);
 
 	while (1) {
-		ret = futex_wait(&futex1, 0, NULL, futex_flag);
-		if (!ret)
-			break;
-
-		if (ret && errno != EAGAIN) {
-			if (!params.silent)
-				warn("futex_wait");
-			break;
+		if (!params.pi) {
+			ret = futex_wait(&futex1, 0, NULL, futex_flag);
+			if (!ret)
+				break;
+
+			if (ret && errno != EAGAIN) {
+				if (!params.silent)
+					warnx("futex_wait");
+				break;
+			}
+		} else {
+			ret = futex_wait_requeue_pi(&futex1, 0, &futex2,
+						    NULL, futex_flag);
+			if (!ret) {
+				/* got the lock at futex2 */
+				futex_unlock_pi(&futex2, futex_flag);
+				break;
+			}
+
+			if (ret && errno != EAGAIN) {
+				if (!params.silent)
+					warnx("futex_wait_requeue_pi");
+				break;
+			}
 		}
 	}
 
@@ -171,9 +190,10 @@ int bench_futex_requeue(int argc, const char **argv)
 	if (params.broadcast)
 		params.nrequeue = params.nthreads;
 
-	printf("Run summary [PID %d]: Requeuing %d threads (from [%s] %p to %p), "
+	printf("Run summary [PID %d]: Requeuing %d threads (from [%s] %p to %s%p), "
 	       "%d at a time.\n\n",  getpid(), params.nthreads,
-	       params.fshared ? "shared":"private", &futex1, &futex2, params.nrequeue);
+	       params.fshared ? "shared":"private", &futex1,
+	       params.pi ? "PI ": "", &futex2, params.nrequeue);
 
 	init_stats(&requeued_stats);
 	init_stats(&requeuetime_stats);
@@ -183,7 +203,7 @@ int bench_futex_requeue(int argc, const char **argv)
 	pthread_cond_init(&thread_worker, NULL);
 
 	for (j = 0; j < bench_repeat && !done; j++) {
-		unsigned int nrequeued = 0;
+		unsigned int nrequeued = 0, wakeups = 0;
 		struct timeval start, end, runtime;
 
 		/* create, launch & block all threads */
@@ -201,13 +221,30 @@ int bench_futex_requeue(int argc, const char **argv)
 		/* Ok, all threads are patiently blocked, start requeueing */
 		gettimeofday(&start, NULL);
 		while (nrequeued < params.nthreads) {
+			int r;
+
 			/*
-			 * Do not wakeup any tasks blocked on futex1, allowing
-			 * us to really measure futex_wait functionality.
+			 * For the regular non-pi case, do not wakeup any tasks
+			 * blocked on futex1, allowing us to really measure
+			 * futex_wait functionality. For the PI case the first
+			 * waiter is always awoken.
 			 */
-			nrequeued += futex_cmp_requeue(&futex1, 0, &futex2, 0,
-						       params.nrequeue,
-						       futex_flag);
+			if (!params.pi) {
+				r = futex_cmp_requeue(&futex1, 0, &futex2, 0,
+						      params.nrequeue,
+						      futex_flag);
+			} else {
+				r = futex_cmp_requeue_pi(&futex1, 0, &futex2,
+							 params.nrequeue,
+							 futex_flag);
+				wakeups++; /* assume no error */
+			}
+
+			if (r < 0)
+				err(EXIT_FAILURE, "couldn't requeue from %p to %p",
+				    &futex1, &futex2);
+
+			nrequeued += r;
 		}
 
 		gettimeofday(&end, NULL);
@@ -217,16 +254,29 @@ int bench_futex_requeue(int argc, const char **argv)
 		update_stats(&requeuetime_stats, runtime.tv_usec);
 
 		if (!params.silent) {
-			printf("[Run %d]: Requeued %d of %d threads in %.4f ms\n",
-			       j + 1, nrequeued, params.nthreads,
-			       runtime.tv_usec / (double)USEC_PER_MSEC);
+			if (!params.pi)
+				printf("[Run %d]: Requeued %d of %d threads in "
+				       "%.4f ms\n", j + 1, nrequeued,
+				       params.nthreads,
+				       runtime.tv_usec / (double)USEC_PER_MSEC);
+			else {
+				nrequeued -= wakeups;
+				printf("[Run %d]: Awoke and Requeued (%d+%d) of "
+				       "%d threads in %.4f ms\n",
+				       j + 1, wakeups, nrequeued,
+				       params.nthreads,
+				       runtime.tv_usec / (double)USEC_PER_MSEC);
+			}
+
 		}
 
-		/* everybody should be blocked on futex2, wake'em up */
-		nrequeued = futex_wake(&futex2, nrequeued, futex_flag);
-		if (params.nthreads != nrequeued)
-			warnx("couldn't wakeup all tasks (%d/%d)",
-			      nrequeued, params.nthreads);
+		if (!params.pi) {
+			/* everybody should be blocked on futex2, wake'em up */
+			nrequeued = futex_wake(&futex2, nrequeued, futex_flag);
+			if (params.nthreads != nrequeued)
+				warnx("couldn't wakeup all tasks (%d/%d)",
+				      nrequeued, params.nthreads);
+		}
 
 		for (i = 0; i < params.nthreads; i++) {
 			ret = pthread_join(worker[i], NULL);