[PATCH v2 net-next 2/3] samples/bpf: add samples for more perf event types

[Alexei Starovoitov]

From: Teng Qin <qinteng@xxxxxx>

This commit adds test code to attach BPF to HW_CACHE and RAW type events
and updates clean-up logic to disable the perf events before closing pmu_fd.

This commit also adds test code to read SOFTWARE, HW_CACHE, RAW and dynamic
pmu events from BPF program using bpf_perf_event_read(). Refactored the
existing sample to fork individual task on each CPU, attach kprobe to
more controllable function, and more accurately check if each read on
every CPU returned with good value.

Signed-off-by: Teng Qin <qinteng@xxxxxx>
Signed-off-by: Alexei Starovoitov <ast@xxxxxxxxxx>
---
 samples/bpf/bpf_helpers.h      |   3 +-
 samples/bpf/trace_event_user.c |  46 ++++++++++-
 samples/bpf/tracex6_kern.c     |  28 +++++--
 samples/bpf/tracex6_user.c     | 176 ++++++++++++++++++++++++++++++++---------
 4 files changed, 204 insertions(+), 49 deletions(-)

diff --git a/samples/bpf/bpf_helpers.h b/samples/bpf/bpf_helpers.h
index 9a9c95f2c9fb..51e567bc70fc 100644
--- a/samples/bpf/bpf_helpers.h
+++ b/samples/bpf/bpf_helpers.h
@@ -31,7 +31,8 @@ static unsigned long long (*bpf_get_current_uid_gid)(void) =
 	(void *) BPF_FUNC_get_current_uid_gid;
 static int (*bpf_get_current_comm)(void *buf, int buf_size) =
 	(void *) BPF_FUNC_get_current_comm;
-static int (*bpf_perf_event_read)(void *map, int index) =
+static unsigned long long (*bpf_perf_event_read)(void *map,
+						 unsigned long long flags) =
 	(void *) BPF_FUNC_perf_event_read;
 static int (*bpf_clone_redirect)(void *ctx, int ifindex, int flags) =
 	(void *) BPF_FUNC_clone_redirect;
diff --git a/samples/bpf/trace_event_user.c b/samples/bpf/trace_event_user.c
index fa4336423da5..666761773fda 100644
--- a/samples/bpf/trace_event_user.c
+++ b/samples/bpf/trace_event_user.c
@@ -122,13 +122,14 @@ static void test_perf_event_all_cpu(struct perf_event_attr *attr)
 {
 	int nr_cpus = sysconf(_SC_NPROCESSORS_CONF);
 	int *pmu_fd = malloc(nr_cpus * sizeof(int));
-	int i;
+	int i, error = 0;
 
 	/* open perf_event on all cpus */
 	for (i = 0; i < nr_cpus; i++) {
 		pmu_fd[i] = sys_perf_event_open(attr, -1, i, -1, 0);
 		if (pmu_fd[i] < 0) {
 			printf("sys_perf_event_open failed\n");
+			error = 1;
 			goto all_cpu_err;
 		}
 		assert(ioctl(pmu_fd[i], PERF_EVENT_IOC_SET_BPF, prog_fd[0]) == 0);
@@ -137,9 +138,13 @@ static void test_perf_event_all_cpu(struct perf_event_attr *attr)
 	system("dd if=/dev/zero of=/dev/null count=5000k");
 	print_stacks();
 all_cpu_err:
-	for (i--; i >= 0; i--)
+	for (i--; i >= 0; i--) {
+		ioctl(pmu_fd[i], PERF_EVENT_IOC_DISABLE, 0);
 		close(pmu_fd[i]);
+	}
 	free(pmu_fd);
+	if (error)
+		int_exit(0);
 }
 
 static void test_perf_event_task(struct perf_event_attr *attr)
@@ -150,7 +155,7 @@ static void test_perf_event_task(struct perf_event_attr *attr)
 	pmu_fd = sys_perf_event_open(attr, 0, -1, -1, 0);
 	if (pmu_fd < 0) {
 		printf("sys_perf_event_open failed\n");
-		return;
+		int_exit(0);
 	}
 	assert(ioctl(pmu_fd, PERF_EVENT_IOC_SET_BPF, prog_fd[0]) == 0);
 	assert(ioctl(pmu_fd, PERF_EVENT_IOC_ENABLE, 0) == 0);
@@ -175,11 +180,45 @@ static void test_bpf_perf_event(void)
 		.config = PERF_COUNT_SW_CPU_CLOCK,
 		.inherit = 1,
 	};
+	struct perf_event_attr attr_hw_cache_l1d = {
+		.sample_freq = SAMPLE_FREQ,
+		.freq = 1,
+		.type = PERF_TYPE_HW_CACHE,
+		.config =
+			PERF_COUNT_HW_CACHE_L1D |
+			(PERF_COUNT_HW_CACHE_OP_READ << 8) |
+			(PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16),
+		.inherit = 1,
+	};
+	struct perf_event_attr attr_hw_cache_branch_miss = {
+		.sample_freq = SAMPLE_FREQ,
+		.freq = 1,
+		.type = PERF_TYPE_HW_CACHE,
+		.config =
+			PERF_COUNT_HW_CACHE_BPU |
+			(PERF_COUNT_HW_CACHE_OP_READ << 8) |
+			(PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
+		.inherit = 1,
+	};
+	struct perf_event_attr attr_type_raw = {
+		.sample_freq = SAMPLE_FREQ,
+		.freq = 1,
+		.type = PERF_TYPE_RAW,
+		/* Intel Instruction Retired */
+		.config = 0xc0,
+		.inherit = 1,
+	};
 
 	test_perf_event_all_cpu(&attr_type_hw);
 	test_perf_event_task(&attr_type_hw);
 	test_perf_event_all_cpu(&attr_type_sw);
 	test_perf_event_task(&attr_type_sw);
+	test_perf_event_all_cpu(&attr_hw_cache_l1d);
+	test_perf_event_task(&attr_hw_cache_l1d);
+	test_perf_event_all_cpu(&attr_hw_cache_branch_miss);
+	test_perf_event_task(&attr_hw_cache_branch_miss);
+	test_perf_event_all_cpu(&attr_type_raw);
+	test_perf_event_task(&attr_type_raw);
 }
 
 
@@ -210,6 +249,7 @@ int main(int argc, char **argv)
 	}
 	test_bpf_perf_event();
 
+	printf("Success!\n");
 	int_exit(0);
 	return 0;
 }
diff --git a/samples/bpf/tracex6_kern.c b/samples/bpf/tracex6_kern.c
index be479c4af9e2..646f86426d09 100644
--- a/samples/bpf/tracex6_kern.c
+++ b/samples/bpf/tracex6_kern.c
@@ -3,22 +3,36 @@
 #include <uapi/linux/bpf.h>
 #include "bpf_helpers.h"
 
-struct bpf_map_def SEC("maps") my_map = {
+struct bpf_map_def SEC("maps") counters = {
 	.type = BPF_MAP_TYPE_PERF_EVENT_ARRAY,
 	.key_size = sizeof(int),
 	.value_size = sizeof(u32),
-	.max_entries = 32,
+	.max_entries = 64,
+};
+struct bpf_map_def SEC("maps") values = {
+	.type = BPF_MAP_TYPE_HASH,
+	.key_size = sizeof(int),
+	.value_size = sizeof(u64),
+	.max_entries = 64,
 };
 
-SEC("kprobe/sys_write")
+SEC("kprobe/htab_map_get_next_key")
 int bpf_prog1(struct pt_regs *ctx)
 {
-	u64 count;
+	u64 count, *val;
+	s64 error;
 	u32 key = bpf_get_smp_processor_id();
-	char fmt[] = "CPU-%d   %llu\n";
 
-	count = bpf_perf_event_read(&my_map, key);
-	bpf_trace_printk(fmt, sizeof(fmt), key, count);
+	count = bpf_perf_event_read(&counters, key);
+	error = (s64)count;
+	if (error < 0 && error > -256)
+		return 0;
+
+	val = bpf_map_lookup_elem(&values, &key);
+	if (val)
+		*val = count;
+	else
+		bpf_map_update_elem(&values, &key, &count, BPF_NOEXIST);
 
 	return 0;
 }
diff --git a/samples/bpf/tracex6_user.c b/samples/bpf/tracex6_user.c
index ca7874ed77f4..43743aa8b3e0 100644
--- a/samples/bpf/tracex6_user.c
+++ b/samples/bpf/tracex6_user.c
@@ -1,73 +1,173 @@
-#include <stdio.h>
-#include <unistd.h>
-#include <stdlib.h>
-#include <stdbool.h>
-#include <string.h>
+#define _GNU_SOURCE
+
+#include <assert.h>
 #include <fcntl.h>
-#include <poll.h>
-#include <sys/ioctl.h>
 #include <linux/perf_event.h>
 #include <linux/bpf.h>
-#include "libbpf.h"
+#include <sched.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <sys/ioctl.h>
+#include <sys/resource.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <unistd.h>
+
 #include "bpf_load.h"
+#include "libbpf.h"
 #include "perf-sys.h"
 
 #define SAMPLE_PERIOD  0x7fffffffffffffffULL
 
-static void test_bpf_perf_event(void)
+static void check_on_cpu(int cpu, struct perf_event_attr *attr)
+{
+	cpu_set_t set;
+	int pmu_fd;
+	__u64 value;
+	int error = 0;
+	/* Move to target CPU */
+	CPU_ZERO(&set);
+	CPU_SET(cpu, &set);
+	assert(sched_setaffinity(0, sizeof(set), &set) == 0);
+	/* Open perf event and attach to the perf_event_array */
+	pmu_fd = sys_perf_event_open(attr, -1/*pid*/, cpu/*cpu*/, -1/*group_fd*/, 0);
+	if (pmu_fd < 0) {
+		fprintf(stderr, "sys_perf_event_open failed on CPU %d\n", cpu);
+		error = 1;
+		goto on_exit;
+	}
+	assert(bpf_map_update_elem(map_fd[0], &cpu, &pmu_fd, BPF_ANY) == 0);
+	assert(ioctl(pmu_fd, PERF_EVENT_IOC_ENABLE, 0) == 0);
+	/* Trigger the kprobe */
+	bpf_map_get_next_key(map_fd[1], &cpu, NULL);
+	/* Check the value */
+	if (bpf_map_lookup_elem(map_fd[1], &cpu, &value)) {
+		fprintf(stderr, "Value missing for CPU %d\n", cpu);
+		error = 1;
+		goto on_exit;
+	}
+	fprintf(stderr, "CPU %d: %llu\n", cpu, value);
+
+on_exit:
+	assert(bpf_map_delete_elem(map_fd[0], &cpu) == 0 || error);
+	assert(ioctl(pmu_fd, PERF_EVENT_IOC_DISABLE, 0) == 0);
+	assert(close(pmu_fd) == 0 || error);
+	assert(bpf_map_delete_elem(map_fd[1], &cpu) == 0 || error);
+	exit(error);
+}
+
+static void test_perf_event_array(struct perf_event_attr *attr,
+				  const char *name)
 {
-	int nr_cpus = sysconf(_SC_NPROCESSORS_CONF);
-	int *pmu_fd = malloc(nr_cpus * sizeof(int));
-	int status, i;
+	int i, status, nr_cpus = sysconf(_SC_NPROCESSORS_CONF);
+	pid_t pid[nr_cpus];
+
+	printf("Test reading %s counters\n", name);
+
+	for (i = 0; i < nr_cpus; i++) {
+		pid[i] = fork();
+		assert(pid[i] >= 0);
+		if (pid[i] == 0) {
+			check_on_cpu(i, attr);
+			exit(1);
+		}
+	}
 
-	struct perf_event_attr attr_insn_pmu = {
+	for (i = 0; i < nr_cpus; i++) {
+		assert(waitpid(pid[i], &status, 0) == pid[i]);
+		assert(status == 0);
+	}
+}
+
+static void test_bpf_perf_event(void)
+{
+	struct perf_event_attr attr_cycles = {
 		.freq = 0,
 		.sample_period = SAMPLE_PERIOD,
 		.inherit = 0,
 		.type = PERF_TYPE_HARDWARE,
 		.read_format = 0,
 		.sample_type = 0,
-		.config = 0,/* PMU: cycles */
+		.config = PERF_COUNT_HW_CPU_CYCLES,
+	};
+	struct perf_event_attr attr_clock = {
+		.freq = 0,
+		.sample_period = SAMPLE_PERIOD,
+		.inherit = 0,
+		.type = PERF_TYPE_SOFTWARE,
+		.read_format = 0,
+		.sample_type = 0,
+		.config = PERF_COUNT_SW_CPU_CLOCK,
+	};
+	struct perf_event_attr attr_raw = {
+		.freq = 0,
+		.sample_period = SAMPLE_PERIOD,
+		.inherit = 0,
+		.type = PERF_TYPE_RAW,
+		.read_format = 0,
+		.sample_type = 0,
+		/* Intel Instruction Retired */
+		.config = 0xc0,
+	};
+	struct perf_event_attr attr_l1d_load = {
+		.freq = 0,
+		.sample_period = SAMPLE_PERIOD,
+		.inherit = 0,
+		.type = PERF_TYPE_HW_CACHE,
+		.read_format = 0,
+		.sample_type = 0,
+		.config =
+			PERF_COUNT_HW_CACHE_L1D |
+			(PERF_COUNT_HW_CACHE_OP_READ << 8) |
+			(PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16),
+	};
+	struct perf_event_attr attr_llc_miss = {
+		.freq = 0,
+		.sample_period = SAMPLE_PERIOD,
+		.inherit = 0,
+		.type = PERF_TYPE_HW_CACHE,
+		.read_format = 0,
+		.sample_type = 0,
+		.config =
+			PERF_COUNT_HW_CACHE_LL |
+			(PERF_COUNT_HW_CACHE_OP_READ << 8) |
+			(PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
+	};
+	struct perf_event_attr attr_msr_tsc = {
+		.freq = 0,
+		.sample_period = 0,
+		.inherit = 0,
+		/* From /sys/bus/event_source/devices/msr/ */
+		.type = 7,
+		.read_format = 0,
+		.sample_type = 0,
+		.config = 0,
 	};
 
-	for (i = 0; i < nr_cpus; i++) {
-		pmu_fd[i] = sys_perf_event_open(&attr_insn_pmu, -1/*pid*/, i/*cpu*/, -1/*group_fd*/, 0);
-		if (pmu_fd[i] < 0) {
-			printf("event syscall failed\n");
-			goto exit;
-		}
-
-		bpf_map_update_elem(map_fd[0], &i, &pmu_fd[i], BPF_ANY);
-		ioctl(pmu_fd[i], PERF_EVENT_IOC_ENABLE, 0);
-	}
-
-	status = system("ls > /dev/null");
-	if (status)
-		goto exit;
-	status = system("sleep 2");
-	if (status)
-		goto exit;
-
-exit:
-	for (i = 0; i < nr_cpus; i++)
-		close(pmu_fd[i]);
-	close(map_fd[0]);
-	free(pmu_fd);
+	test_perf_event_array(&attr_cycles, "HARDWARE-cycles");
+	test_perf_event_array(&attr_clock, "SOFTWARE-clock");
+	test_perf_event_array(&attr_raw, "RAW-instruction-retired");
+	test_perf_event_array(&attr_l1d_load, "HW_CACHE-L1D-load");
+	test_perf_event_array(&attr_llc_miss, "HW_CACHE-LLC-miss");
+	test_perf_event_array(&attr_msr_tsc, "Dynamic-msr-tsc");
 }
 
 int main(int argc, char **argv)
 {
+	struct rlimit r = {RLIM_INFINITY, RLIM_INFINITY};
 	char filename[256];
 
 	snprintf(filename, sizeof(filename), "%s_kern.o", argv[0]);
 
+	setrlimit(RLIMIT_MEMLOCK, &r);
 	if (load_bpf_file(filename)) {
 		printf("%s", bpf_log_buf);
 		return 1;
 	}
 
 	test_bpf_perf_event();
-	read_trace_pipe();
 
+	printf("Success!\n");
 	return 0;
 }
-- 
2.9.3