bpftool: Introduce "prog profile" command

With fentry/fexit programs, it is possible to profile BPF program with
hardware counters. Introduce bpftool "prog profile", which measures key
metrics of a BPF program.

bpftool prog profile command creates per-cpu perf events. Then it attaches
fentry/fexit programs to the target BPF program. The fentry program saves
perf event value to a map. The fexit program reads the perf event again,
and calculates the difference, which is the instructions/cycles used by
the target program.

Example input and output:

  ./bpftool prog profile id 337 duration 3 cycles instructions llc_misses

        4228 run_cnt
     3403698 cycles                                              (84.08%)
     3525294 instructions   #  1.04 insn per cycle               (84.05%)
          13 llc_misses     #  3.69 LLC misses per million isns  (83.50%)

This command measures cycles and instructions for BPF program with id
337 for 3 seconds. The program has triggered 4228 times. The rest of the
output is similar to perf-stat. In this example, the counters were only
counting ~84% of the time because of time multiplexing of perf counters.

Note that, this approach measures cycles and instructions in very small
increments. So the fentry/fexit programs introduce noticeable errors to
the measurement results.

The fentry/fexit programs are generated with BPF skeletons. Therefore, we
build bpftool twice. The first time _bpftool is built without skeletons.
Then, _bpftool is used to generate the skeletons. The second time, bpftool
is built with skeletons.

Signed-off-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Quentin Monnet <quentin@isovalent.com>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20200309173218.2739965-2-songliubraving@fb.com

tools/bpf/bpftool/skeleton/profiler.bpf.c

@@ -0,0 +1,119 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (c) 2020 Facebook
+#include "profiler.h"
+#include <linux/bpf.h>
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+
+/* map of perf event fds, num_cpu * num_metric entries */
+struct {
+	__uint(type, BPF_MAP_TYPE_PERF_EVENT_ARRAY);
+	__uint(key_size, sizeof(u32));
+	__uint(value_size, sizeof(int));
+} events SEC(".maps");
+
+/* readings at fentry */
+struct {
+	__uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
+	__uint(key_size, sizeof(u32));
+	__uint(value_size, sizeof(struct bpf_perf_event_value));
+} fentry_readings SEC(".maps");
+
+/* accumulated readings */
+struct {
+	__uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
+	__uint(key_size, sizeof(u32));
+	__uint(value_size, sizeof(struct bpf_perf_event_value));
+} accum_readings SEC(".maps");
+
+/* sample counts, one per cpu */
+struct {
+	__uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
+	__uint(key_size, sizeof(u32));
+	__uint(value_size, sizeof(u64));
+} counts SEC(".maps");
+
+const volatile __u32 num_cpu = 1;
+const volatile __u32 num_metric = 1;
+#define MAX_NUM_MATRICS 4
+
+SEC("fentry/XXX")
+int BPF_PROG(fentry_XXX)
+{
+	struct bpf_perf_event_value *ptrs[MAX_NUM_MATRICS];
+	u32 key = bpf_get_smp_processor_id();
+	u32 i;
+
+	/* look up before reading, to reduce error */
+	for (i = 0; i < num_metric && i < MAX_NUM_MATRICS; i++) {
+		u32 flag = i;
+
+		ptrs[i] = bpf_map_lookup_elem(&fentry_readings, &flag);
+		if (!ptrs[i])
+			return 0;
+	}
+
+	for (i = 0; i < num_metric && i < MAX_NUM_MATRICS; i++) {
+		struct bpf_perf_event_value reading;
+		int err;
+
+		err = bpf_perf_event_read_value(&events, key, &reading,
+						sizeof(reading));
+		if (err)
+			return 0;
+		*(ptrs[i]) = reading;
+		key += num_cpu;
+	}
+
+	return 0;
+}
+
+static inline void
+fexit_update_maps(u32 id, struct bpf_perf_event_value *after)
+{
+	struct bpf_perf_event_value *before, diff, *accum;
+
+	before = bpf_map_lookup_elem(&fentry_readings, &id);
+	/* only account samples with a valid fentry_reading */
+	if (before && before->counter) {
+		struct bpf_perf_event_value *accum;
+
+		diff.counter = after->counter - before->counter;
+		diff.enabled = after->enabled - before->enabled;
+		diff.running = after->running - before->running;
+
+		accum = bpf_map_lookup_elem(&accum_readings, &id);
+		if (accum) {
+			accum->counter += diff.counter;
+			accum->enabled += diff.enabled;
+			accum->running += diff.running;
+		}
+	}
+}
+
+SEC("fexit/XXX")
+int BPF_PROG(fexit_XXX)
+{
+	struct bpf_perf_event_value readings[MAX_NUM_MATRICS];
+	u32 cpu = bpf_get_smp_processor_id();
+	u32 i, one = 1, zero = 0;
+	int err;
+	u64 *count;
+
+	/* read all events before updating the maps, to reduce error */
+	for (i = 0; i < num_metric && i < MAX_NUM_MATRICS; i++) {
+		err = bpf_perf_event_read_value(&events, cpu + i * num_cpu,
+						readings + i, sizeof(*readings));
+		if (err)
+			return 0;
+	}
+	count = bpf_map_lookup_elem(&counts, &zero);
+	if (count) {
+		*count += 1;
+		for (i = 0; i < num_metric && i < MAX_NUM_MATRICS; i++)
+			fexit_update_maps(i, &readings[i]);
+	}
+	return 0;
+}
+
+char LICENSE[] SEC("license") = "GPL";

tools/bpf/bpftool/skeleton/profiler.h

@@ -0,0 +1,47 @@
+/* SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) */
+#ifndef __PROFILER_H
+#define __PROFILER_H
+
+/* useful typedefs from vmlinux.h */
+
+typedef signed char __s8;
+typedef unsigned char __u8;
+typedef short int __s16;
+typedef short unsigned int __u16;
+typedef int __s32;
+typedef unsigned int __u32;
+typedef long long int __s64;
+typedef long long unsigned int __u64;
+
+typedef __s8 s8;
+typedef __u8 u8;
+typedef __s16 s16;
+typedef __u16 u16;
+typedef __s32 s32;
+typedef __u32 u32;
+typedef __s64 s64;
+typedef __u64 u64;
+
+enum {
+	false = 0,
+	true = 1,
+};
+
+#ifdef __CHECKER__
+#define __bitwise__ __attribute__((bitwise))
+#else
+#define __bitwise__
+#endif
+#define __bitwise __bitwise__
+
+typedef __u16 __bitwise __le16;
+typedef __u16 __bitwise __be16;
+typedef __u32 __bitwise __le32;
+typedef __u32 __bitwise __be32;
+typedef __u64 __bitwise __le64;
+typedef __u64 __bitwise __be64;
+
+typedef __u16 __bitwise __sum16;
+typedef __u32 __bitwise __wsum;
+
+#endif /* __PROFILER_H */