context_switch_latency: Allow fork as an alternative to threads

1 files changed, 97 insertions, 71 deletions

diff --git a/context_switch_latency.c b/context_switch_latency.c
index 650f853..7920ea4 100644
--- a/context_switch_latency.c
+++ b/context_switch_latency.c
@@ -3,11 +3,14 @@
 #include <time.h>
 #include <stdio.h>
 #include <stdlib.h>
+#include <stdbool.h>
 #include <string.h>
 #include <unistd.h>
 #include <sched.h>
 #include <pthread.h>
 #include <sys/eventfd.h>
+#include <sys/prctl.h>
+#include <signal.h>
 #include <err.h>
 
 int to_thread, from_thread;
@@ -69,78 +72,101 @@ void killit()
 
 int main(int argc, char **argv)
 {
-  struct timespec start, end;
-  pthread_t thread;
-  uint64_t i, iters = 100000;
-  uint64_t ns;
-  int main_cpu;
-
-  if (argc != 3) {
-    printf("Usage: %s <0|1> <same|different>\n\nSet the parameter to 1 to use xstate\nUse 'same' for same CPU or 'different' for cross-CPU\n",
-	   argv[0]);
-    return 1;
-  } else {
-    if (!strcmp(argv[1], "1"))
-      use_xstate = 1;
-    else if (strcmp(argv[1], "0"))
-      errx(1, "first arg is wrong");
-
-    if (!strcmp(argv[2], "same")) {
-      main_cpu = 0;
-      thread_cpu = 0;
-    } else if (!strcmp(argv[2], "different")) {
-      main_cpu = 0;
-      thread_cpu = 1;
-    } else {
-      errx(1, "second arg is wrong");
-    }
-  }
-
-  printf("use_xstate = %d\n", use_xstate);
-
-  to_thread = eventfd(0, 0);
-  if (to_thread < 0)
-    err(1, "eventfd");
-  from_thread = eventfd(0, 0);
-  if (from_thread < 0)
-    err(1, "eventfd");
-  
-  if (pthread_create(&thread, 0, threadproc, 0) != 0)
-    err(1, "pthread_create");
-
-  cpu_set_t cpuset;
-  CPU_ZERO(&cpuset);
-  CPU_SET(main_cpu, &cpuset);
-  if (sched_setaffinity(0, sizeof(cpuset), &cpuset) != 0)
-    printf("[WARN]\tsched_setaffinity failed\n");
-
-  /* Warm up (and burn the xstate heuristic) */
-  for(i = 0; i < 10000; i++) {
-    bounce();
-    maybe_use_xstate();
-  }
-
-  /* Clear out the xstate hack. */
-  state = 1;
-  bounce();
-
-  clock_gettime(CLOCK_MONOTONIC, &start);
-
-  for (i = 0; i < iters; i++) {
-    bounce();
-    maybe_use_xstate();
-  }
-
-  clock_gettime(CLOCK_MONOTONIC, &end);
-
-  killit();
-
-  ns = 1000000000ULL * (end.tv_sec - start.tv_sec) + (end.tv_nsec - start.tv_nsec);
-
-  printf("%llu iters at %.1f ns/switch\n",
+	struct timespec start, end;
+	pthread_t thread;
+	uint64_t i, iters = 100000;
+	uint64_t ns;
+	int main_cpu;
+	bool use_thread = false;
+
+	if (argc != 4) {
+		printf("Usage: %s <0|1> <thread|process> <same|different>\n\nSet the parameter to 1 to use xstate\nUse 'same' for same CPU or 'different' for cross-CPU\n",
+		       argv[0]);
+		return 1;
+	}
+
+	if (!strcmp(argv[1], "1"))
+		use_xstate = 1;
+	else if (strcmp(argv[1], "0"))
+		errx(1, "first arg is wrong");
+
+	if (!strcmp(argv[3], "same")) {
+		main_cpu = 0;
+		thread_cpu = 0;
+	} else if (!strcmp(argv[3], "different")) {
+		main_cpu = 0;
+		thread_cpu = 1;
+	} else {
+		errx(1, "third arg is wrong");
+	}
+
+	if (!strcmp(argv[2], "thread")) {
+		use_thread = true;
+	} else if (!strcmp(argv[2], "process")) {
+		use_thread = false;
+	} else {
+		errx(1, "second arg is wrong");
+	}
+
+	printf("use_xstate = %d\n", use_xstate);
+
+	to_thread = eventfd(0, 0);
+	if (to_thread < 0)
+		err(1, "eventfd");
+	from_thread = eventfd(0, 0);
+	if (from_thread < 0)
+		err(1, "eventfd");
+
+	if (use_thread) {
+		printf("Using threads\n");
+		if (pthread_create(&thread, 0, threadproc, 0) != 0)
+			err(1, "pthread_create");
+	} else {
+		printf("Using a subprocess\n");
+		pid_t child = fork();
+		if (child < 0)
+			err(1, "fork");
+
+		if (child == 0) {
+			prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0);
+			threadproc(NULL);
+		}
+	}
+
+	cpu_set_t cpuset;
+	CPU_ZERO(&cpuset);
+	CPU_SET(main_cpu, &cpuset);
+	if (sched_setaffinity(0, sizeof(cpuset), &cpuset) != 0)
+		printf("[WARN]\tsched_setaffinity failed\n");
+
+	/* Warm up (and burn the xstate heuristic) */
+	for(i = 0; i < 10000; i++) {
+		bounce();
+		maybe_use_xstate();
+	}
+
+	/* Clear out the xstate hack. */
+	state = 1;
+	bounce();
+
+	clock_gettime(CLOCK_MONOTONIC, &start);
+
+	for (i = 0; i < iters; i++) {
+		bounce();
+		maybe_use_xstate();
+	}
+
+	clock_gettime(CLOCK_MONOTONIC, &end);
+
+	killit();
+
+	ns = 1000000000ULL * (end.tv_sec - start.tv_sec) + (end.tv_nsec - start.tv_nsec);
+
+	printf("%llu iters at %.1f ns/switch\n",
 	 (unsigned long long)iters, (double)ns / (2*iters));
 
-  pthread_join(thread, 0);
+	if (use_thread)
+		pthread_join(thread, 0);
 
-  return 0;
+	return 0;
 }