[PATCH 0/2] perf bench: Add epoll-wait-mp1c benchmark calling epoll_wait(2)

[PATCH 0/2] perf bench: Add epoll-wait-mp1c benchmark calling epoll_wait(2)

[Roman Penyaev]

The intention of a new benchmark is to measure events delivery
bandwidth: N threads produce events and 1 thread consumes events
calling epoll_wait(2).

Benchmark does measurements for 8, 16, 32, 64 and 128 threads in
a loop.

This one differs from epoll-wait in that it produces events from
many threads and consumes from one, thus mp1c (many producers
1 consumer).

Roman Penyaev (2):
  perf bench: rename epoll-wait bench to epoll-wait-1pmc
  perf bench: Add epoll-wait-mp1c benchmark calling epoll_wait(2)

 tools/perf/bench/Build                        |   3 +-
 tools/perf/bench/bench.h                      |   3 +-
 .../bench/{epoll-wait.c => epoll-wait-1pmc.c} |   6 +-
 tools/perf/bench/epoll-wait-mp1c.c            | 175 ++++++++++++++++++
 tools/perf/builtin-bench.c                    |   3 +-
 5 files changed, 186 insertions(+), 4 deletions(-)
 rename tools/perf/bench/{epoll-wait.c => epoll-wait-1pmc.c} (98%)
 create mode 100644 tools/perf/bench/epoll-wait-mp1c.c

Signed-off-by: Roman Penyaev <rpenyaev@suse.de>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Davidlohr Bueso <dbueso@suse.de>
Cc: Jason Baron <jbaron@akamai.com>
Cc: linux-kernel@vger.kernel.org
-- 
2.19.1

[PATCH 1/2] perf bench: rename epoll-wait bench to epoll-wait-1pmc

[Roman Penyaev]

I want to rename this in order to add another epoll-wait bench in next
patch and to stress in the name, that the bench corresponds to 1 producer
and many consumers.

In the following patch I plan to add epoll-wait-mp1c bench.

Signed-off-by: Roman Penyaev <rpenyaev@suse.de>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Davidlohr Bueso <dbueso@suse.de>
Cc: Jason Baron <jbaron@akamai.com>
Cc: linux-kernel@vger.kernel.org
---
 tools/perf/bench/Build                               | 2 +-
 tools/perf/bench/bench.h                             | 2 +-
 tools/perf/bench/{epoll-wait.c => epoll-wait-1pmc.c} | 6 +++++-
 tools/perf/builtin-bench.c                           | 2 +-
 4 files changed, 8 insertions(+), 4 deletions(-)
 rename tools/perf/bench/{epoll-wait.c => epoll-wait-1pmc.c} (98%)

diff --git a/tools/perf/bench/Build b/tools/perf/bench/Build
index e4e321b6f883..ef07fc40bc35 100644
--- a/tools/perf/bench/Build
+++ b/tools/perf/bench/Build
@@ -7,7 +7,7 @@ perf-y += futex-wake-parallel.o
 perf-y += futex-requeue.o
 perf-y += futex-lock-pi.o
 
-perf-y += epoll-wait.o
+perf-y += epoll-wait-1pmc.o
 perf-y += epoll-ctl.o
 
 perf-$(CONFIG_X86_64) += mem-memcpy-x86-64-lib.o
diff --git a/tools/perf/bench/bench.h b/tools/perf/bench/bench.h
index fddb3ced9db6..fb9782624644 100644
--- a/tools/perf/bench/bench.h
+++ b/tools/perf/bench/bench.h
@@ -38,7 +38,7 @@ int bench_futex_requeue(int argc, const char **argv);
 /* pi futexes */
 int bench_futex_lock_pi(int argc, const char **argv);
 
-int bench_epoll_wait(int argc, const char **argv);
+int bench_epoll_wait_1pmc(int argc, const char **argv);
 int bench_epoll_ctl(int argc, const char **argv);
 
 #define BENCH_FORMAT_DEFAULT_STR	"default"
diff --git a/tools/perf/bench/epoll-wait.c b/tools/perf/bench/epoll-wait-1pmc.c
similarity index 98%
rename from tools/perf/bench/epoll-wait.c
rename to tools/perf/bench/epoll-wait-1pmc.c
index 5a11534e96a0..1b88cdd2ef78 100644
--- a/tools/perf/bench/epoll-wait.c
+++ b/tools/perf/bench/epoll-wait-1pmc.c
@@ -26,6 +26,10 @@
  *             [queue D]  ---> [worker]
  *             [queue E]  ---> [worker]
  *
+ * Both models correspond to 1 producer and many consumers, where a single
+ * producer is a writer thread and consumers are many workers threads
+ * waiting for events.
+ *
  * Naturally, the single queue will enforce more concurrency on the epoll
  * instance, and can therefore scale poorly compared to multiple queues.
  * However, this is a benchmark raw data and must be taken with a grain of
@@ -409,7 +413,7 @@ static int cmpworker(const void *p1, const void *p2)
 	return w1->tid > w2->tid;
 }
 
-int bench_epoll_wait(int argc, const char **argv)
+int bench_epoll_wait_1pmc(int argc, const char **argv)
 {
 	int ret = 0;
 	struct sigaction act;
diff --git a/tools/perf/builtin-bench.c b/tools/perf/builtin-bench.c
index 334c77ffc1d9..6926b5a5eebf 100644
--- a/tools/perf/builtin-bench.c
+++ b/tools/perf/builtin-bench.c
@@ -70,7 +70,7 @@ static struct bench futex_benchmarks[] = {
 
 #ifdef HAVE_EVENTFD
 static struct bench epoll_benchmarks[] = {
-	{ "wait",	"Benchmark epoll concurrent epoll_waits",       bench_epoll_wait	},
+	{ "wait-1pmc",	"Benchmark epoll concurrent epoll_waits",       bench_epoll_wait_1pmc	},
 	{ "ctl",	"Benchmark epoll concurrent epoll_ctls",        bench_epoll_ctl		},
 	{ "all",	"Run all futex benchmarks",			NULL			},
 	{ NULL,		NULL,						NULL			}
-- 
2.19.1

[PATCH 2/2] perf bench: Add epoll-wait-mp1c benchmark calling epoll_wait(2)

[Roman Penyaev]

The intention of this benchmark is to measure events delivery
bandwidth: N threads produce events and 1 thread consumes events
calling epoll_wait(2).

Benchmark does measurements for 8, 16, 32, 64 and 128 threads in
a loop.

This one differs from epoll-wait-1pmc in that it produces events
from many threads and consumes from one, thus mp1c (many producers
1 consumer).

Signed-off-by: Roman Penyaev <rpenyaev@suse.de>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Davidlohr Bueso <dbueso@suse.de>
Cc: Jason Baron <jbaron@akamai.com>
Cc: linux-kernel@vger.kernel.org
---
 tools/perf/bench/Build             |   1 +
 tools/perf/bench/bench.h           |   1 +
 tools/perf/bench/epoll-wait-mp1c.c | 175 +++++++++++++++++++++++++++++
 tools/perf/builtin-bench.c         |   1 +
 4 files changed, 178 insertions(+)
 create mode 100644 tools/perf/bench/epoll-wait-mp1c.c

diff --git a/tools/perf/bench/Build b/tools/perf/bench/Build
index ef07fc40bc35..570df3f475b8 100644
--- a/tools/perf/bench/Build
+++ b/tools/perf/bench/Build
@@ -8,6 +8,7 @@ perf-y += futex-requeue.o
 perf-y += futex-lock-pi.o
 
 perf-y += epoll-wait-1pmc.o
+perf-y += epoll-wait-mp1c.o
 perf-y += epoll-ctl.o
 
 perf-$(CONFIG_X86_64) += mem-memcpy-x86-64-lib.o
diff --git a/tools/perf/bench/bench.h b/tools/perf/bench/bench.h
index fb9782624644..2ee7e7256e23 100644
--- a/tools/perf/bench/bench.h
+++ b/tools/perf/bench/bench.h
@@ -39,6 +39,7 @@ int bench_futex_requeue(int argc, const char **argv);
 int bench_futex_lock_pi(int argc, const char **argv);
 
 int bench_epoll_wait_1pmc(int argc, const char **argv);
+int bench_epoll_wait_mp1c(int argc, const char **argv);
 int bench_epoll_ctl(int argc, const char **argv);
 
 #define BENCH_FORMAT_DEFAULT_STR	"default"
diff --git a/tools/perf/bench/epoll-wait-mp1c.c b/tools/perf/bench/epoll-wait-mp1c.c
new file mode 100644
index 000000000000..44b06ae86e5b
--- /dev/null
+++ b/tools/perf/bench/epoll-wait-mp1c.c
@@ -0,0 +1,175 @@
+// SPDX-License-Identifier: GPL-2.0
+#ifdef HAVE_EVENTFD
+/*
+ * Copyright (C) 2019 Roman Penyaev
+ *
+ * This program benchmarks bandwidth of events delivered from many threads
+ * (many producers) to a single consumer, which monitors for events calling
+ * epoll_wait(2).
+ */
+
+#include <string.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <time.h>
+#include <assert.h>
+#include <sys/epoll.h>
+#include <sys/eventfd.h>
+#include <unistd.h>
+#include <pthread.h>
+#include <errno.h>
+#include <err.h>
+#include <linux/kernel.h> /* For ARRAY_SIZE only */
+#include "bench.h"
+#include "cpumap.h"
+
+#define ITERS     1000000ull
+
+struct thread_ctx {
+	pthread_t thread;
+	int efd;
+};
+
+static volatile unsigned int thr_ready;
+static volatile unsigned int start;
+
+static inline unsigned long long nsecs(void)
+{
+	struct timespec ts = {0, 0};
+
+	clock_gettime(CLOCK_MONOTONIC, &ts);
+	return ((unsigned long long)ts.tv_sec * 1000000000ull) + ts.tv_nsec;
+}
+
+static void *thread_work(void *arg)
+{
+	struct thread_ctx *ctx = arg;
+	uint64_t ucnt = 1;
+	unsigned int i;
+	int rc;
+
+	__atomic_add_fetch(&thr_ready, 1, __ATOMIC_RELAXED);
+
+	while (!start)
+		;
+
+	for (i = 0; i < ITERS; i++) {
+		rc = write(ctx->efd, &ucnt, sizeof(ucnt));
+		assert(rc == sizeof(ucnt));
+	}
+
+	return NULL;
+}
+
+static int do_bench(struct cpu_map *cpu, unsigned int nthreads)
+{
+	struct epoll_event ev, events[nthreads];
+	struct thread_ctx threads[nthreads];
+	pthread_attr_t thrattr;
+	struct thread_ctx *ctx;
+	int rc, epfd, nfds;
+	cpu_set_t cpuset;
+	unsigned int i;
+
+	unsigned long long epoll_calls = 0, epoll_nsecs;
+	unsigned long long ucnt, ucnt_sum = 0;
+
+	epfd = epoll_create1(0);
+	if (epfd < 0)
+		err(EXIT_FAILURE, "epoll_create1");
+
+	for (i = 0; i < nthreads; i++) {
+		ctx = &threads[i];
+
+		ctx->efd = eventfd(0, EFD_NONBLOCK);
+		if (ctx->efd < 0)
+			err(EXIT_FAILURE, "eventfd");
+
+		ev.events = EPOLLIN;
+		ev.data.ptr = ctx;
+		rc = epoll_ctl(epfd, EPOLL_CTL_ADD, ctx->efd, &ev);
+		if (rc)
+			err(EXIT_FAILURE, "epoll_ctl");
+
+		CPU_ZERO(&cpuset);
+		CPU_SET(cpu->map[i % cpu->nr], &cpuset);
+
+		pthread_attr_init(&thrattr);
+		rc = pthread_attr_setaffinity_np(&thrattr, sizeof(cpu_set_t),
+						 &cpuset);
+		if (rc) {
+			errno = rc;
+			err(EXIT_FAILURE, "pthread_attr_setaffinity_np");
+		}
+
+		rc = pthread_create(&ctx->thread, &thrattr, thread_work, ctx);
+		if (rc) {
+			errno = rc;
+			err(EXIT_FAILURE, "pthread_create");
+		}
+	}
+
+	while (thr_ready == nthreads)
+		;
+
+	/* Signal start for all threads */
+	start = 1;
+
+	epoll_nsecs = nsecs();
+	while (1) {
+		nfds = epoll_wait(epfd, events, nthreads, -1);
+		if (nfds < 0)
+			err(EXIT_FAILURE, "epoll_wait");
+
+		epoll_calls++;
+
+		for (i = 0; i < (unsigned int)nfds; ++i) {
+			ctx = events[i].data.ptr;
+			rc = read(ctx->efd, &ucnt, sizeof(ucnt));
+			assert(rc == sizeof(ucnt));
+			ucnt_sum += ucnt;
+			if (ucnt_sum == nthreads * ITERS)
+				goto end;
+		}
+	}
+end:
+	epoll_nsecs = nsecs() - epoll_nsecs;
+
+	for (i = 0; i < nthreads; i++) {
+		ctx = &threads[i];
+		pthread_join(ctx->thread, NULL);
+		close(ctx->efd);
+	}
+	close(epfd);
+
+	printf("%7d   %8lld     %8lld\n",
+		   nthreads,
+		   ITERS*nthreads/(epoll_nsecs/1000/1000),
+		   epoll_nsecs/1000/1000);
+
+	return 0;
+}
+
+int bench_epoll_wait_mp1c(int argc, const char **argv)
+{
+	unsigned int i, nthreads_arr[] = {8, 16, 32, 64, 128};
+	struct cpu_map *cpu;
+
+	(void)argc; (void)argv;
+
+	cpu = cpu_map__new(NULL);
+	if (!cpu) {
+		errno = ENOMEM;
+		err(EXIT_FAILURE, "cpu_map__new");
+	}
+
+	printf("threads  events/ms  run-time ms\n");
+	for (i = 0; i < ARRAY_SIZE(nthreads_arr); i++)
+		do_bench(cpu, nthreads_arr[i]);
+
+	cpu_map__put(cpu);
+
+	return 0;
+}
+
+#endif // HAVE_EVENTFD
diff --git a/tools/perf/builtin-bench.c b/tools/perf/builtin-bench.c
index 6926b5a5eebf..4426b7afadf3 100644
--- a/tools/perf/builtin-bench.c
+++ b/tools/perf/builtin-bench.c
@@ -71,6 +71,7 @@ static struct bench futex_benchmarks[] = {
 #ifdef HAVE_EVENTFD
 static struct bench epoll_benchmarks[] = {
 	{ "wait-1pmc",	"Benchmark epoll concurrent epoll_waits",       bench_epoll_wait_1pmc	},
+	{ "wait-mp1c",	"Benchmark epoll events delivery bandwidth",    bench_epoll_wait_mp1c	},
 	{ "ctl",	"Benchmark epoll concurrent epoll_ctls",        bench_epoll_ctl		},
 	{ "all",	"Run all futex benchmarks",			NULL			},
 	{ NULL,		NULL,						NULL			}
-- 
2.19.1