1/* Copyright (c) 2016 Facebook
  2 *
  3 * This program is free software; you can redistribute it and/or
  4 * modify it under the terms of version 2 of the GNU General Public
  5 * License as published by the Free Software Foundation.
  6 */
  7#include <uapi/linux/bpf.h>
  8#include "bpf_helpers.h"
  9#include <uapi/linux/ptrace.h>
 10#include <uapi/linux/perf_event.h>
 11#include <linux/version.h>
 12#include <linux/sched.h>
 13
 14#define _(P) ({typeof(P) val = 0; bpf_probe_read(&val, sizeof(val), &P); val;})
 15
 16#define MINBLOCK_US	1
 17
 18struct key_t {
 19	char waker[TASK_COMM_LEN];
 20	char target[TASK_COMM_LEN];
 21	u32 wret;
 22	u32 tret;
 23};
 24
 25struct bpf_map_def SEC("maps") counts = {
 26	.type = BPF_MAP_TYPE_HASH,
 27	.key_size = sizeof(struct key_t),
 28	.value_size = sizeof(u64),
 29	.max_entries = 10000,
 30};
 31
 32struct bpf_map_def SEC("maps") start = {
 33	.type = BPF_MAP_TYPE_HASH,
 34	.key_size = sizeof(u32),
 35	.value_size = sizeof(u64),
 36	.max_entries = 10000,
 37};
 38
 39struct wokeby_t {
 40	char name[TASK_COMM_LEN];
 41	u32 ret;
 42};
 43
 44struct bpf_map_def SEC("maps") wokeby = {
 45	.type = BPF_MAP_TYPE_HASH,
 46	.key_size = sizeof(u32),
 47	.value_size = sizeof(struct wokeby_t),
 48	.max_entries = 10000,
 49};
 50
 51struct bpf_map_def SEC("maps") stackmap = {
 52	.type = BPF_MAP_TYPE_STACK_TRACE,
 53	.key_size = sizeof(u32),
 54	.value_size = PERF_MAX_STACK_DEPTH * sizeof(u64),
 55	.max_entries = 10000,
 56};
 57
 58#define STACKID_FLAGS (0 | BPF_F_FAST_STACK_CMP)
 59
 60SEC("kprobe/try_to_wake_up")
 61int waker(struct pt_regs *ctx)
 62{
 63	struct task_struct *p = (void *) PT_REGS_PARM1(ctx);
 64	struct wokeby_t woke = {};
 65	u32 pid;
 66
 67	pid = _(p->pid);
 68
 69	bpf_get_current_comm(&woke.name, sizeof(woke.name));
 70	woke.ret = bpf_get_stackid(ctx, &stackmap, STACKID_FLAGS);
 71
 72	bpf_map_update_elem(&wokeby, &pid, &woke, BPF_ANY);
 73	return 0;
 74}
 75
 76static inline int update_counts(struct pt_regs *ctx, u32 pid, u64 delta)
 77{
 78	struct key_t key = {};
 79	struct wokeby_t *woke;
 80	u64 zero = 0, *val;
 81
 82	bpf_get_current_comm(&key.target, sizeof(key.target));
 83	key.tret = bpf_get_stackid(ctx, &stackmap, STACKID_FLAGS);
 84
 85	woke = bpf_map_lookup_elem(&wokeby, &pid);
 86	if (woke) {
 87		key.wret = woke->ret;
 88		__builtin_memcpy(&key.waker, woke->name, TASK_COMM_LEN);
 89		bpf_map_delete_elem(&wokeby, &pid);
 90	}
 91
 92	val = bpf_map_lookup_elem(&counts, &key);
 93	if (!val) {
 94		bpf_map_update_elem(&counts, &key, &zero, BPF_NOEXIST);
 95		val = bpf_map_lookup_elem(&counts, &key);
 96		if (!val)
 97			return 0;
 98	}
 99	(*val) += delta;
100	return 0;
101}
102
103SEC("kprobe/finish_task_switch")
104int oncpu(struct pt_regs *ctx)
105{
106	struct task_struct *p = (void *) PT_REGS_PARM1(ctx);
107	u64 delta, ts, *tsp;
108	u32 pid;
109
110	/* record previous thread sleep time */
111	pid = _(p->pid);
112	ts = bpf_ktime_get_ns();
113	bpf_map_update_elem(&start, &pid, &ts, BPF_ANY);
114
115	/* calculate current thread's delta time */
116	pid = bpf_get_current_pid_tgid();
117	tsp = bpf_map_lookup_elem(&start, &pid);
118	if (!tsp)
119		/* missed start or filtered */
120		return 0;
121
122	delta = bpf_ktime_get_ns() - *tsp;
123	bpf_map_delete_elem(&start, &pid);
124	delta = delta / 1000;
125	if (delta < MINBLOCK_US)
126		return 0;
127
128	return update_counts(ctx, pid, delta);
129}
130char _license[] SEC("license") = "GPL";
131u32 _version SEC("version") = LINUX_VERSION_CODE;