Linux soft lockup分析

關鍵詞：watchdog、soft lockup、percpu thread、lockdep等。

 

近日遇到一個soft lockup問題，打印相似「[ 56.032356] NMI watchdog: BUG: soft lockup - CPU#0 stuck for 23s! [cat:153]「。

這是lockup檢測機制在起做用，lockup檢測機制包括soft lockup detector和hard lockup detector。

藉機分析下soft lockup機制以及什麼狀況下致使soft watchdog異常、對watchdog的配置、如何定位異常點。

這裏跳過hard lockup detector的分析。

1. soft lockup機制分析

lockup_detector_init()函數首先獲取sample_period以及watchdog_cpumask，而後根據狀況建立線程，啓動喂狗程序；建立hrtimer啓動看門狗。

而後有兩個重點一個是建立內核線程的API以及struct smp_hotplug_thread結構體。

void __init lockup_detector_init(void) { set_sample_period();----------------------------------------獲取變量sample_period，爲watchdog_thresh*2/5，即4秒喂一次狗。 ... cpumask_copy(&watchdog_cpumask, cpu_possible_mask); 
    if (watchdog_enabled) watchdog_enable_all_cpus(); } static int watchdog_enable_all_cpus(void) { int err = 0; if (!watchdog_running) {----------------------------------若是當前watchdog_running沒有再運行，那麼爲每一個CPU建立一個watchdog/x線程，這些線程每隔sample_period時間喂一次狗。watchdog_threads時watchdog/x線程的主要輸入參數，watchdog_cpumask規定了爲哪些CPU建立線程。 err = smpboot_register_percpu_thread_cpumask(&watchdog_threads, &watchdog_cpumask); if (err) pr_err("Failed to create watchdog threads, disabled\n"); else watchdog_running = 1; } else {  err = update_watchdog_all_cpus(); if (err) { watchdog_disable_all_cpus(); pr_err("Failed to update lockup detectors, disabled\n"); } } if (err) watchdog_enabled = 0; return err; } static void watchdog_disable_all_cpus(void) { if (watchdog_running) { watchdog_running = 0; smpboot_unregister_percpu_thread(&watchdog_threads); } } static int update_watchdog_all_cpus(void) { int ret; ret = watchdog_park_threads(); if (ret) return ret; watchdog_unpark_threads(); return 0; } static int watchdog_park_threads(void) { int cpu, ret = 0; atomic_set(&watchdog_park_in_progress, 1); for_each_watchdog_cpu(cpu) { ret = kthread_park(per_cpu(softlockup_watchdog, cpu));---------------------------設置struct kthread->flags的KTHREAD_SHOULD_PARK位，在watchdog/x線程中會調用unpark成員函數進行處理。 if (ret) break; } atomic_set(&watchdog_park_in_progress, 0); return ret; } static void watchdog_unpark_threads(void) { int cpu; for_each_watchdog_cpu(cpu) kthread_unpark(per_cpu(softlockup_watchdog, cpu));-------------------------------清空struct kthread->flags的KTHREAD_SHOULD_PARK位，在watchdog/x線程中會調用park成員函數。 }

 

1.1 watchdog_threads結構體介紹

在介紹如何建立watchdog/x線程以前，有必要先介紹一些struct smp_hotplug_thread線程。

struct smp_hotplug_thread { struct task_struct __percpu    **store;--------------------------存放percpu strcut task_strcut指針的指針。 struct list_head list; int                (*thread_should_run)(unsigned int cpu);-------檢查是否應該運行watchdog/x線程。 void                (*thread_fn)(unsigned int cpu);--------------watchdog/x線程的主函數。 void                (*create)(unsigned int cpu); void                (*setup)(unsigned int cpu);------------------在運行watchdog/x線程以前的準備工做。 void                (*cleanup)(unsigned int cpu, bool online);---在退出watchdog/x線程以後的清楚工做。 void                (*park)(unsigned int cpu);-------------------當CPU offline時，須要臨時中止。 void                (*unpark)(unsigned int cpu);-----------------當CPU變成online時，進行準備工做。 cpumask_var_t cpumask;--------------------------------容許哪些CPU online。 bool selfparking; const char            *thread_comm;------------------------------watchdog/x線程名稱。 };

 watchdog_threads是soft lockup監控線程的實體，基於此建立 watchdog/x線程。

static struct smp_hotplug_thread watchdog_threads = { .store = &softlockup_watchdog, .thread_should_run = watchdog_should_run, .thread_fn = watchdog, .thread_comm = "watchdog/%u", .setup = watchdog_enable, .cleanup = watchdog_cleanup, .park = watchdog_disable, .unpark = watchdog_enable, }; static void watchdog_enable(unsigned int cpu) { struct hrtimer *hrtimer = raw_cpu_ptr(&watchdog_hrtimer); /* kick off the timer for the hardlockup detector */ hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); hrtimer->function = watchdog_timer_fn;------------------------------------------建立一個hrtimer，超時函數爲watchdog_timer_fn，這裏面會檢查watchdog_touch_ts變量是否超過20秒沒有被更新。若是是，則有soft lockup。 /* Enable the perf event */ watchdog_nmi_enable(cpu); /* done here because hrtimer_start can only pin to smp_processor_id() */ hrtimer_start(hrtimer, ns_to_ktime(sample_period), HRTIMER_MODE_REL_PINNED);---------------------------------------------啓動一個超時爲sample_period(4秒)的hrtimer，HRTIMER_MODE_REL_PINNED表示此hrtimer和當前CPU綁定。 /* initialize timestamp */ watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1);---------------------------------設置當前線程爲實時FIFO，而且優先級爲實時99.這個優先級表示高於全部的非實時線程，可是實時優先級最低的。 __touch_watchdog();-------------------------------------------------------------更新watchdog_touch_ts變量，至關於喂狗操做。 } static void watchdog_set_prio(unsigned int policy, unsigned int prio) { struct sched_param param = { .sched_priority = prio }; sched_setscheduler(current, policy, &param); } /* Commands for resetting the watchdog */
static void __touch_watchdog(void) { __this_cpu_write(watchdog_touch_ts, get_timestamp());----------------------------喂狗的操做就是更新watchdog_touch_ts變量，也即當前時間戳。 } static void watchdog_disable(unsigned int cpu)-------------------------------------至關於watchdog_enable()反操做，將線程恢復爲普通線程；取消hrtimer。 { struct hrtimer *hrtimer = raw_cpu_ptr(&watchdog_hrtimer); watchdog_set_prio(SCHED_NORMAL, 0); hrtimer_cancel(hrtimer); /* disable the perf event */ watchdog_nmi_disable(cpu); } static void watchdog_cleanup(unsigned int cpu, bool online) { watchdog_disable(cpu); }  static int watchdog_should_run(unsigned int cpu) { return __this_cpu_read(hrtimer_interrupts) != __this_cpu_read(soft_lockup_hrtimer_cnt);------------------------------------hrtimer_interrupts記錄了產生hrtimer的次數；在watchdog()中，將hrtimer_interrupts賦給soft_lockup_hrtimer_cnt。二者相等表示沒有hrtimer產生，不須要運行watchdog/x線程；相反不等，則須要watchdog/x線程運行。 } static void watchdog(unsigned int cpu) { __this_cpu_write(soft_lockup_hrtimer_cnt, __this_cpu_read(hrtimer_interrupts));-----------------------------------更新soft_lockup_hrtimer_cnt，在watch_should_run()中就返回false，表示線程不須要運行，即不須要喂狗。 __touch_watchdog();--------------------------------------------------------------雖然就是一句話，可是卻很重要的喂狗操做。     if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED)) watchdog_nmi_disable(cpu); }

 

1.2 建立喂狗線程watchdog/x

在分析了watchdog_threads以後，再來看看如何建立watchdog/x線程。 

int smpboot_register_percpu_thread_cpumask(struct smp_hotplug_thread *plug_thread, const struct cpumask *cpumask) { unsigned int cpu; int ret = 0; if (!alloc_cpumask_var(&plug_thread->cpumask, GFP_KERNEL)) return -ENOMEM; cpumask_copy(plug_thread->cpumask, cpumask); get_online_cpus(); mutex_lock(&smpboot_threads_lock); for_each_online_cpu(cpu) {------------------------------------------------遍歷全部online CPU，爲每一個CPU建立一個percpu的watchdog/x線程。 ret = __smpboot_create_thread(plug_thread, cpu); if (ret) { smpboot_destroy_threads(plug_thread);-----------------------------建立失敗則釋放相關資源。 free_cpumask_var(plug_thread->cpumask); goto out; } if (cpumask_test_cpu(cpu, cpumask)) smpboot_unpark_thread(plug_thread, cpu);--------------------------若是當前CPU不在cpumask中，則清空KTHREAD_SHOULD_PARK，進而調用watchdog_therads的umpark成員函數。 } list_add(&plug_thread->list, &hotplug_threads); out: mutex_unlock(&smpboot_threads_lock); put_online_cpus(); return ret; } static int __smpboot_create_thread(struct smp_hotplug_thread *ht, unsigned int cpu) { struct task_struct *tsk = *per_cpu_ptr(ht->store, cpu); struct smpboot_thread_data *td; if (tsk) return 0; td = kzalloc_node(sizeof(*td), GFP_KERNEL, cpu_to_node(cpu)); if (!td) return -ENOMEM; td->cpu = cpu; td->ht = ht; tsk =kthread_create_on_cpu(smpboot_thread_fn, td, cpu, ht->thread_comm);-----------------------------------------在指定CPU上建立watchdog/x線程，處理函數爲smpboot_thread_fn()。 if (IS_ERR(tsk)) { kfree(td); return PTR_ERR(tsk); } /* * Park the thread so that it could start right on the CPU * when it is available. */ kthread_park(tsk);--------------------------------------------------------在CPU上當即啓動watchdog/x線程。 get_task_struct(tsk);-----------------------------------------------------增長對線程的引用計數。 *per_cpu_ptr(ht->store, cpu) = tsk;---------------------------------------store存放線程結構體指針的指針。 if (ht->create) {         if (!wait_task_inactive(tsk, TASK_PARKED)) WARN_ON(1); else ht->create(cpu); } return 0; }  static int smpboot_thread_fn(void *data) { struct smpboot_thread_data *td = data; struct smp_hotplug_thread *ht = td->ht; while (1) { set_current_state(TASK_INTERRUPTIBLE); preempt_disable(); if (kthread_should_stop()) {----------------------------------------若是能夠終止線程，調用cleanup，退出線程。 __set_current_state(TASK_RUNNING); preempt_enable(); /* cleanup must mirror setup */
            if (ht->cleanup && td->status != HP_THREAD_NONE) ht->cleanup(td->cpu, cpu_online(td->cpu)); kfree(td); return 0; } if (kthread_should_park()) {----------------------------------------若是KTHREAD_SHOULD_PARK置位，調用park()暫停進程執行。 __set_current_state(TASK_RUNNING); preempt_enable(); if (ht->park && td->status == HP_THREAD_ACTIVE) { BUG_ON(td->cpu != smp_processor_id()); ht->park(td->cpu); td->status = HP_THREAD_PARKED; } kthread_parkme(); /* We might have been woken for stop */
            continue; } BUG_ON(td->cpu != smp_processor_id()); /* Check for state change setup */
        switch (td->status) { case HP_THREAD_NONE:-----------------------------------------------至關於第一次運行，調用setup()進行初始化操做。 __set_current_state(TASK_RUNNING); preempt_enable(); if (ht->setup) ht->setup(td->cpu); td->status = HP_THREAD_ACTIVE; continue; case HP_THREAD_PARKED:---------------------------------------------從parked狀態恢復。 __set_current_state(TASK_RUNNING); preempt_enable(); if (ht->unpark) ht->unpark(td->cpu); td->status = HP_THREAD_ACTIVE; continue; } if (!ht->thread_should_run(td->cpu)) {-----------------------------若是不須要進程運行，schedule()主動放棄CPU給其餘線程使用。 preempt_enable_no_resched(); schedule(); } else { __set_current_state(TASK_RUNNING); preempt_enable(); ht->thread_fn(td->cpu);----------------------------------------調用struct smpboot_thread_fn->thread_fn及watchdog()，進行喂狗操做。 } } } void smpboot_unregister_percpu_thread(struct smp_hotplug_thread *plug_thread)----將建立的內核線程移除操做。 { get_online_cpus(); mutex_lock(&smpboot_threads_lock); list_del(&plug_thread->list); smpboot_destroy_threads(plug_thread); mutex_unlock(&smpboot_threads_lock); put_online_cpus(); free_cpumask_var(plug_thread->cpumask); } static void smpboot_destroy_threads(struct smp_hotplug_thread *ht) { unsigned int cpu; /* We need to destroy also the parked threads of offline cpus */ for_each_possible_cpu(cpu) { struct task_struct *tsk = *per_cpu_ptr(ht->store, cpu); if (tsk) { kthread_stop(tsk); put_task_struct(tsk); *per_cpu_ptr(ht->store, cpu) = NULL; } } }

 

1.3 hrtimer看門狗

 在分析了喂狗線程watchdog/x以後，再來分析看門狗是如何實現的？

看門狗是經過啓動一個週期爲4秒的hrtimer來實現的，這個hrtimer和CPU綁定，使用的變量都是percpu的。確保每一個CPU之間不相互干擾。

每次hrtimer超時，都會喚醒watchdog/x線程，並進行一次喂狗操做。

由於hrtimer超時函數在軟中斷中調用，在中斷產生後會比線程優先獲得執行。

因此在watchdog/x線程沒有獲得執行的狀況下，經過is_softlockup()來判斷看門狗是否超過20秒沒有獲得喂狗。

static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer) { unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts); struct pt_regs *regs = get_irq_regs(); int duration; int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace; if (atomic_read(&watchdog_park_in_progress) != 0) return HRTIMER_NORESTART; /* kick the hardlockup detector */ watchdog_interrupt_count();------------------------------------------------------------------沒產生一次中斷，hrtimer_interrupts計數加1.hrtimer_interrupts記錄了產生hrtimer的次數。 /* kick the softlockup detector */ wake_up_process(__this_cpu_read(softlockup_watchdog));---------------------------------------喚醒watchdog/x線程，進行喂狗操做。 /* .. and repeat */ hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));------------------------------------從新設置超時點，造成周期性時鐘。 ...  duration = is_softlockup(touch_ts);----------------------------------------------------------返回非0表示，看門狗超時。 if (unlikely(duration)) {--------------------------------------------------------------------看門狗超時狀況的處理。         if (kvm_check_and_clear_guest_paused()) return HRTIMER_RESTART; /* only warn once */
        if (__this_cpu_read(soft_watchdog_warn) == true) {             if (__this_cpu_read(softlockup_task_ptr_saved) != current) { __this_cpu_write(soft_watchdog_warn, false); __touch_watchdog(); } return HRTIMER_RESTART; } if (softlockup_all_cpu_backtrace) {             if (test_and_set_bit(0, &soft_lockup_nmi_warn)) { /* Someone else will report us. Let's give up */ __this_cpu_write(soft_watchdog_warn, true); return HRTIMER_RESTART; } } pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n", smp_processor_id(), duration, current->comm, task_pid_nr(current));-------------------------------------------------打印哪一個CPU被卡死duration秒，以及死在哪一個進程。 __this_cpu_write(softlockup_task_ptr_saved, current); print_modules(); print_irqtrace_events(current);-----------------------------------------------------------顯示開關中斷、軟中斷信息，禁止中斷和軟中斷也是形成soft lockup的一個緣由。 if (regs)---------------------------------------------------------------------------------有寄存器顯示寄存器信息，同時顯示棧信息。 show_regs(regs); else dump_stack(); if (softlockup_all_cpu_backtrace) {  trigger_allbutself_cpu_backtrace(); clear_bit(0, &soft_lockup_nmi_warn); /* Barrier to sync with other cpus */ smp_mb__after_atomic(); } add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK); if (softlockup_panic)---------------------------------------------------------------------若是定義softlockup_panic則進入panic()。 panic("softlockup: hung tasks"); __this_cpu_write(soft_watchdog_warn, true); } else __this_cpu_write(soft_watchdog_warn, false); return HRTIMER_RESTART; }

  static void watchdog_interrupt_count(void)
  {
      __this_cpu_inc(hrtimer_interrupts);
  }

static int is_softlockup(unsigned long touch_ts) { unsigned long now = get_timestamp(); if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){ /* Warn about unreasonable delays. */
        if (time_after(now, touch_ts + get_softlockup_thresh())) return now - touch_ts; } return 0; }

 

2. 對watchdog的設置

 對watchdog行爲的設置有兩個途徑：經過命令行傳入參數和經過proc設置。

2.1 經過命令行設置

經過命令行傳入參數，能夠對soft lockup進行開關設置、超時事後是否panic等等行爲。

static int __init softlockup_panic_setup(char *str) { softlockup_panic = simple_strtoul(str, NULL, 0); return 1; } __setup("softlockup_panic=", softlockup_panic_setup); static int __init nowatchdog_setup(char *str) { watchdog_enabled = 0; return 1; } __setup("nowatchdog", nowatchdog_setup); static int __init nosoftlockup_setup(char *str) { watchdog_enabled &= ~SOFT_WATCHDOG_ENABLED; return 1; } __setup("nosoftlockup", nosoftlockup_setup); #ifdef CONFIG_SMP static int __init softlockup_all_cpu_backtrace_setup(char *str) { sysctl_softlockup_all_cpu_backtrace =
        !!simple_strtol(str, NULL, 0); return 1; } __setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup); static int __init hardlockup_all_cpu_backtrace_setup(char *str) { sysctl_hardlockup_all_cpu_backtrace =
        !!simple_strtol(str, NULL, 0); return 1; } __setup("hardlockup_all_cpu_backtrace=", hardlockup_all_cpu_backtrace_setup); #endif

 

2.2 經過sysfs節點調節watchdog

 watchdog相關的配置還能夠經過proc文件系統進行配置。

/proc/sys/kernel/nmi_watchdog-------------------------hard lockup開關，proc_nmi_watchdog()。 /proc/sys/kernel/soft_watchdog------------------------soft lockup開關，proc_soft_watchdog()。 /proc/sys/kernel/watchdog-----------------------------watchdog總開關，proc_watchdog()。 /proc/sys/kernel/watchdog_cpumask---------------------watchdog cpumaks，proc_watchdog_cpumask()。 /proc/sys/kernel/watchdog_thresh----------------------watchdog超時閾值設置，proc_watchdog_thresh()。

 

3. 定位soft lockup異常

引發soft lockup的緣由通常是死循環或者死鎖， 死循環能夠經過棧回溯找到問題點；死鎖問題須要打開內核的lockdep功能。

打開內核的lockdep功能能夠參考《Linux死鎖檢測-Lockdep》。

下面看一個while(1)引發的soft lockup異常分析：

[ 5656.032325] NMI watchdog: BUG: soft lockup - CPU#0 stuck for 22s! [cat:157]-----------------------CPU、進程等信息粗略定位。 [ 5656.039314] Modules linked in: [ 5656.042386] [ 5656.042386] CURRENT PROCESS: [ 5656.042386] [ 5656.048229] COMM=cat PID=157 [ 5656.051117] TEXT=00008000-000c5a68 DATA=000c6f1c-000c7175 BSS=000c7175-000c8000 [ 5656.058432] USER-STACK=7fc1ee50  KERNEL-STACK=bd0b7080 [ 5656.058432] [ 5656.065069] PC: 0x8032a1b2 (clk_summary_show+0x62/0xb4)--------------------------------------------PC指向出問題的點，更加精確的定位。 [ 5656.070302] LR: 0x8032a186 (clk_summary_show+0x36/0xb4) [ 5656.075531] SP: 0xbd8b1b74... [ 5656.217622] Call Trace:-----------------------------------------------------------------------------------------經過Call Trace，能夠了解如何作到PC指向的問題點的。前因後果一目瞭然。 [<80155c5e>] seq_read+0xc2/0x46c [<802826ac>] full_proxy_read+0x58/0x98 [<8013239c>] do_readv_writev+0x31c/0x384 [<80132458>] vfs_readv+0x54/0x8c [<80160b52>] default_file_splice_read+0x166/0x2b0 [<801606ee>] do_splice_to+0x76/0xb0 [<801607de>] splice_direct_to_actor+0xb6/0x21c [<801609c2>] do_splice_direct+0x7e/0xa8 [<80132a5a>] do_sendfile+0x21a/0x45c [<80133776>] SyS_sendfile64+0xf6/0xfc [<80046186>] csky_systemcall+0x96/0xe0