调度器59—CONFIG_SCHED_DEBUG 的作用

一、使能 CONFIG_SCHED_DEBUG 提供的文件

1. /proc/pid/sched 文件

cat显示统计数据，echo清0，无论echo的是啥。

2. /proc/sys/kernel 下的

sched_min_granularity_ns
sched_latency_ns
sched_sync_hint_enable
sched_cstate_aware
sched_wakeup_granularity_ns
sched_tunable_scaling
sched_migration_cost_ns
sched_nr_migrate
sched_time_avg_ms
sched_schedstats
//若使能 CONFIG_IO_SCHED_OPTIMIZE 还有：
io_sched_optimize
//若使能 CONFIG_NUMA_BALANCING 还有：
numa_balancing_scan_delay_ms
numa_balancing_scan_period_min_ms
numa_balancing_scan_period_max_ms
numa_balancing_scan_size_mb
numa_balancing

3. /proc/sched_debug 文件

二、提供启动参数 sched_debug

1. 作用

static int __init sched_debug_setup(char *str) //topology.c
{
    sched_debug_enabled = true;

    return 0;
}
early_param("sched_debug", sched_debug_setup); //topology.c

sched_init_domains //topology.c
partition_sched_domains //topology.c
    build_sched_domains
        cpu_attach_domain //sched/topology.c
            sched_domain_debug //sched/topology.c
                if (sched_debug_enabled) {
                    printk(KERN_DEBUG "CPU%d attaching sched-domain(s):\n", cpu);
                    for (;;) {
                        if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask))
                            break;
                        level++;
                        sd = sd->parent;
                        if (!sd)
                            break;
                    }
                }

 

三、Feature使能与关闭

1. 对于 default y 的，需要使用"CONFIG_XX is not set" 来关闭

/lib/Kconfig.debug
config SCHED_DEBUG
    bool "Collect scheduler debugging info"
    depends on DEBUG_KERNEL && PROC_FS
    default y

//user-perf_defconfig
# CONFIG_SCHED_DEBUG is not set

四、代码分析

1. 使能后编译住sched/debug.c

msm-4.14/kernel/sched$ cat Makefile
obj-$(CONFIG_SCHED_DEBUG) += debug.o //代码编译上会编译住debug.c

2. deadlne.c中的使用

/proc/sched_debug 
    sched_debug_show //sched/debug.c
        print_cpu //sched/debug.c
            print_dl_stats(*m, cpu) //deadline.c
                print_dl_rq(*m, cpu, *dl_rq)//sched/debug.c

3. sched/core.c 中的使用

sysrq_sched_debug_show() //sched/core.c

//调用路径：
    echo t > /proc/sysrq-trigger
        sysrq_showstate_op.handler //"show-task-states(t)" 对应的handler回调
            sysrq_handle_showstate //sysrq.c
            fn_show_state //tty/vt/keyboard.c
                show_state() //sched/debug.h 传参0
                sysrq_handle_showstate_blocked //sysrq.c 传参 TASK_UNINTERRUPTIBLE=2，无效
                    show_state_filter(ustate_filter) //sched/core.c 参数为0才执行
                        sysrq_sched_debug_show //sched/debug.c 使能 CONFIG_SCHED_DEBUG 才存在
                            for_each_online_cpu(cpu)
                                print_cpu(NULL, cpu);

echo t > /proc/sysrq-trigger 会触发打印每一个进程的内核栈，如下：

[  565.347148] (7)[7110:sh]fxp.monlolo S     0   838  1073  4078    623 0x00000000
[  565.347155] (7)[7110:sh]Call trace:
[  565.347159] (7)[7110:sh] __switch_to+0x118/0x134
[  565.347161] (7)[7110:sh] __schedule+0x7c0/0xa78
[  565.347167] (7)[7110:sh] schedule+0x70/0x90
[  565.347172] (7)[7110:sh] schedule_hrtimeout_range_clock+0xc0/0xe8
[  565.347174] (7)[7110:sh] schedule_hrtimeout_range+0x10/0x18
[  565.347180] (7)[7110:sh] SyS_epoll_wait+0x2ac/0x3c4
[  565.347184] (7)[7110:sh] SyS_epoll_pwait+0x108/0x168
[  565.347186] (7)[7110:sh] el0_svc_naked+0x34/0x38

若正在运行，会多打印出"running task"，打印如下：

[  565.299661] (4)[7110:sh]UnityMain       R   running task        0 32556    28  3430    623 0x00000000
[  565.299664] (4)[7110:sh]Call trace:
[  565.299667] (4)[7110:sh] __switch_to+0x118/0x134
[  565.299670] (4)[7110:sh] __schedule+0x7c0/0xa78
[  565.299672] (4)[7110:sh] schedule+0x70/0x90
[  565.299675] (4)[7110:sh] futex_wait_queue_me+0xbc/0x108
[  565.299678] (4)[7110:sh] futex_wait+0x104/0x2d0
[  565.299681] (4)[7110:sh] do_futex+0x114/0x18e4
[  565.299684] (4)[7110:sh] SyS_futex+0x140/0x1ac
[  565.299687] (4)[7110:sh] el0_svc_naked+0x34/0x38

第一行各字段分别为：p->comm, task_state_to_char(p), "running task"(正在runing的有，sleep的无)，0, p->nvcsw, p->nivcsw, p->pid, p->real_parent->pid, task_thread_info(p)->flags

# ps -e| grep fxp.monlolo
USER           PID  PPID     VSZ    RSS WCHAN            ADDR S NAME
system        4078   623 7980716 237860 SyS_epoll_wait      0 S com.fxp.monlolo

3. rt.c中的使用

print_rt_stats //rt.c

//调用路径：
            print_cpu //sched/debug.c 同上
                print_rt_stats(*m, cpu) //rt.c

 

五、补充

1. 上面说到了打印内核栈，这里增量两个打印每CPU调用栈和进程用户栈的方法

# echo l > /proc/sysrq-trigger 打印每个CPU当前调用栈

thaw-filesystems(j) show-backtrace-all-active-cpus(l) show-task-states(t) show-blocked-tasks(w)

# debuggerd -b <pid> 打印pid进程所有线程的用户栈