tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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 "bpf_tracing.h"
10#include <uapi/linux/ptrace.h>
11#include <uapi/linux/perf_event.h>
12#include <linux/version.h>
13#include <linux/sched.h>
14
15#define _(P) ({typeof(P) val; bpf_probe_read(&val, sizeof(val), &P); val;})
16
17#define MINBLOCK_US 1
18
19struct key_t {
20 char waker[TASK_COMM_LEN];
21 char target[TASK_COMM_LEN];
22 u32 wret;
23 u32 tret;
24};
25
26struct bpf_map_def SEC("maps") counts = {
27 .type = BPF_MAP_TYPE_HASH,
28 .key_size = sizeof(struct key_t),
29 .value_size = sizeof(u64),
30 .max_entries = 10000,
31};
32
33struct bpf_map_def SEC("maps") start = {
34 .type = BPF_MAP_TYPE_HASH,
35 .key_size = sizeof(u32),
36 .value_size = sizeof(u64),
37 .max_entries = 10000,
38};
39
40struct wokeby_t {
41 char name[TASK_COMM_LEN];
42 u32 ret;
43};
44
45struct bpf_map_def SEC("maps") wokeby = {
46 .type = BPF_MAP_TYPE_HASH,
47 .key_size = sizeof(u32),
48 .value_size = sizeof(struct wokeby_t),
49 .max_entries = 10000,
50};
51
52struct bpf_map_def SEC("maps") stackmap = {
53 .type = BPF_MAP_TYPE_STACK_TRACE,
54 .key_size = sizeof(u32),
55 .value_size = PERF_MAX_STACK_DEPTH * sizeof(u64),
56 .max_entries = 10000,
57};
58
59#define STACKID_FLAGS (0 | BPF_F_FAST_STACK_CMP)
60
61SEC("kprobe/try_to_wake_up")
62int waker(struct pt_regs *ctx)
63{
64 struct task_struct *p = (void *) PT_REGS_PARM1(ctx);
65 struct wokeby_t woke;
66 u32 pid;
67
68 pid = _(p->pid);
69
70 bpf_get_current_comm(&woke.name, sizeof(woke.name));
71 woke.ret = bpf_get_stackid(ctx, &stackmap, STACKID_FLAGS);
72
73 bpf_map_update_elem(&wokeby, &pid, &woke, BPF_ANY);
74 return 0;
75}
76
77static inline int update_counts(void *ctx, u32 pid, u64 delta)
78{
79 struct wokeby_t *woke;
80 u64 zero = 0, *val;
81 struct key_t key;
82
83 __builtin_memset(&key.waker, 0, sizeof(key.waker));
84 bpf_get_current_comm(&key.target, sizeof(key.target));
85 key.tret = bpf_get_stackid(ctx, &stackmap, STACKID_FLAGS);
86 key.wret = 0;
87
88 woke = bpf_map_lookup_elem(&wokeby, &pid);
89 if (woke) {
90 key.wret = woke->ret;
91 __builtin_memcpy(&key.waker, woke->name, sizeof(key.waker));
92 bpf_map_delete_elem(&wokeby, &pid);
93 }
94
95 val = bpf_map_lookup_elem(&counts, &key);
96 if (!val) {
97 bpf_map_update_elem(&counts, &key, &zero, BPF_NOEXIST);
98 val = bpf_map_lookup_elem(&counts, &key);
99 if (!val)
100 return 0;
101 }
102 (*val) += delta;
103 return 0;
104}
105
106#if 1
107/* taken from /sys/kernel/debug/tracing/events/sched/sched_switch/format */
108struct sched_switch_args {
109 unsigned long long pad;
110 char prev_comm[16];
111 int prev_pid;
112 int prev_prio;
113 long long prev_state;
114 char next_comm[16];
115 int next_pid;
116 int next_prio;
117};
118SEC("tracepoint/sched/sched_switch")
119int oncpu(struct sched_switch_args *ctx)
120{
121 /* record previous thread sleep time */
122 u32 pid = ctx->prev_pid;
123#else
124SEC("kprobe/finish_task_switch")
125int oncpu(struct pt_regs *ctx)
126{
127 struct task_struct *p = (void *) PT_REGS_PARM1(ctx);
128 /* record previous thread sleep time */
129 u32 pid = _(p->pid);
130#endif
131 u64 delta, ts, *tsp;
132
133 ts = bpf_ktime_get_ns();
134 bpf_map_update_elem(&start, &pid, &ts, BPF_ANY);
135
136 /* calculate current thread's delta time */
137 pid = bpf_get_current_pid_tgid();
138 tsp = bpf_map_lookup_elem(&start, &pid);
139 if (!tsp)
140 /* missed start or filtered */
141 return 0;
142
143 delta = bpf_ktime_get_ns() - *tsp;
144 bpf_map_delete_elem(&start, &pid);
145 delta = delta / 1000;
146 if (delta < MINBLOCK_US)
147 return 0;
148
149 return update_counts(ctx, pid, delta);
150}
151char _license[] SEC("license") = "GPL";
152u32 _version SEC("version") = LINUX_VERSION_CODE;