int do_finish_task_switch()

in bpf/profiling/offcpu.c [24:82]

• 44 lines of code
• 7 McCabe index (conditional complexity)

int do_finish_task_switch(struct pt_regs *ctx) {
    int monitor_pid;
    asm("%0 = MONITOR_PID ll" : "=r"(monitor_pid));

    __u32 pid, tgid;
    __u64 ts, *tsp;

    struct task_struct *prev = (void *) PT_REGS_PARM1(ctx);
    pid = _KERNEL(prev->pid);
    tgid = _KERNEL(prev->tgid);

    // in kernel, tgid means the process id
    // monitor_pid is the same concept with process id
    // only the monitor pid is same with previous process id could record the start time
    if (tgid == monitor_pid) {
        ts = bpf_ktime_get_ns();
        bpf_map_update_elem(&starts, &pid, &ts, BPF_ANY);
    }

    // if current process have record start time in the map
    // means the monitored process switch to the on-cpu
    __u64 pid_tgid = bpf_get_current_pid_tgid();
    pid = pid_tgid;
    tgid = pid_tgid >> 32;
    if (tgid != monitor_pid) {
        return 0;
    }
    tsp = bpf_map_lookup_elem(&starts, &pid);
    if (tsp == 0) {
        return 0;        // missed start or filtered
    }

    // calculate the duration(on->off to off->on)
    __u64 t_start = *tsp;
    __u64 t_end = bpf_ktime_get_ns();
    bpf_map_delete_elem(&starts, &pid);
    if (t_start > t_end) {
        return 0;
    }

    // create map key
    struct key_t key = {};
    key.kernel_stack_id = bpf_get_stackid(ctx, &stacks, 0);
    key.user_stack_id = bpf_get_stackid(ctx, &stacks, (1ULL << 8));

    // add counters
    struct value_t *val;
    val = bpf_map_lookup_elem(&counts, &key);
    if (!val) {
        struct value_t value = {};
         bpf_map_update_elem(&counts, &key, &value, BPF_NOEXIST);
         val = bpf_map_lookup_elem(&counts, &key);
         if (!val)
             return 0;
    }
    (*val).counts += 1;
    (*val).deltas += t_end - t_start;
    return 0;
}