Whamcloud - gitweb
LU-7434 ptlrpc: lost bulk leads to a hang
[fs/lustre-release.git] / libcfs / libcfs / watchdog.c
index 0981a2c..95a0177 100644 (file)
@@ -27,7 +27,7 @@
  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2012, Intel Corporation.
+ * Copyright (c) 2012, 2014, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
 
 #define DEBUG_SUBSYSTEM S_LNET
 
+#include <linux/kthread.h>
 #include <libcfs/libcfs.h>
 #include "tracefile.h"
 
 struct lc_watchdog {
-       spinlock_t  lcw_lock;     /* check or change lcw_list */
-        int             lcw_refcount; /* must hold lcw_pending_timers_lock */
-        cfs_timer_t     lcw_timer;    /* kernel timer */
-        cfs_list_t      lcw_list;     /* chain on pending list */
-        cfs_time_t      lcw_last_touched; /* last touched stamp */
-        cfs_task_t     *lcw_task;     /* owner task */
-        void          (*lcw_callback)(pid_t, void *);
-        void           *lcw_data;
-
-        pid_t           lcw_pid;
-
-        enum {
-                LC_WATCHDOG_DISABLED,
-                LC_WATCHDOG_ENABLED,
-                LC_WATCHDOG_EXPIRED
-        } lcw_state;
+       spinlock_t              lcw_lock;       /* check or change lcw_list */
+       int                     lcw_refcount;   /* must hold lcw_pending_timers_lock */
+       struct timer_list       lcw_timer;      /* kernel timer */
+       struct list_head        lcw_list;       /* chain on pending list */
+       cfs_time_t              lcw_last_touched;/* last touched stamp */
+       struct task_struct     *lcw_task;       /* owner task */
+       void                    (*lcw_callback)(pid_t, void *);
+       void                    *lcw_data;
+
+       pid_t                   lcw_pid;
+
+       enum {
+               LC_WATCHDOG_DISABLED,
+               LC_WATCHDOG_ENABLED,
+               LC_WATCHDOG_EXPIRED
+       } lcw_state;
 };
 
 #ifdef WITH_WATCHDOG
@@ -70,7 +71,7 @@ struct lc_watchdog {
  */
 static struct completion lcw_start_completion;
 static struct completion  lcw_stop_completion;
-static cfs_waitq_t lcw_event_waitq;
+static wait_queue_head_t lcw_event_waitq;
 
 /*
  * Set this and wake lcw_event_waitq to stop the dispatcher.
@@ -94,7 +95,7 @@ static DEFINE_MUTEX(lcw_refcount_mutex);
  */
 /* BH lock! */
 static DEFINE_SPINLOCK(lcw_pending_timers_lock);
-static cfs_list_t lcw_pending_timers = CFS_LIST_HEAD_INIT(lcw_pending_timers);
+static struct list_head lcw_pending_timers = LIST_HEAD_INIT(lcw_pending_timers);
 
 /* Last time a watchdog expired */
 static cfs_time_t lcw_last_watchdog_time;
@@ -104,28 +105,20 @@ static void
 lcw_dump(struct lc_watchdog *lcw)
 {
         ENTRY;
-#if defined(HAVE_TASKLIST_LOCK)
-       read_lock(&tasklist_lock);
-#else
         rcu_read_lock();
-#endif
        if (lcw->lcw_task == NULL) {
-                LCONSOLE_WARN("Process " LPPID " was not found in the task "
+               LCONSOLE_WARN("Process %d was not found in the task "
                               "list; watchdog callback may be incomplete\n",
                               (int)lcw->lcw_pid);
         } else {
                 libcfs_debug_dumpstack(lcw->lcw_task);
         }
 
-#if defined(HAVE_TASKLIST_LOCK)
-       read_unlock(&tasklist_lock);
-#else
         rcu_read_unlock();
-#endif
         EXIT;
 }
 
-static void lcw_cb(ulong_ptr_t data)
+static void lcw_cb(uintptr_t data)
 {
         struct lc_watchdog *lcw = (struct lc_watchdog *)data;
         ENTRY;
@@ -138,12 +131,12 @@ static void lcw_cb(ulong_ptr_t data)
         lcw->lcw_state = LC_WATCHDOG_EXPIRED;
 
        spin_lock_bh(&lcw->lcw_lock);
-       LASSERT(cfs_list_empty(&lcw->lcw_list));
+       LASSERT(list_empty(&lcw->lcw_list));
 
        spin_lock_bh(&lcw_pending_timers_lock);
        lcw->lcw_refcount++; /* +1 for pending list */
-       cfs_list_add(&lcw->lcw_list, &lcw_pending_timers);
-       cfs_waitq_signal(&lcw_event_waitq);
+       list_add(&lcw->lcw_list, &lcw_pending_timers);
+       wake_up(&lcw_event_waitq);
 
        spin_unlock_bh(&lcw_pending_timers_lock);
        spin_unlock_bh(&lcw->lcw_lock);
@@ -158,7 +151,7 @@ static int is_watchdog_fired(void)
                return 1;
 
        spin_lock_bh(&lcw_pending_timers_lock);
-       rc = !cfs_list_empty(&lcw_pending_timers);
+       rc = !list_empty(&lcw_pending_timers);
        spin_unlock_bh(&lcw_pending_timers_lock);
        return rc;
 }
@@ -210,11 +203,20 @@ static void lcw_dump_stack(struct lc_watchdog *lcw)
         }
 }
 
+/*
+ * Provided watchdog handlers
+ */
+
+static void lc_watchdog_dumplog(pid_t pid, void *data)
+{
+       libcfs_debug_dumplog_internal((void *)((uintptr_t)pid));
+}
+
 static int lcw_dispatch_main(void *data)
 {
         int                 rc = 0;
         struct lc_watchdog *lcw;
-        CFS_LIST_HEAD      (zombies);
+       struct list_head zombies = LIST_HEAD_INIT(zombies);
 
         ENTRY;
 
@@ -223,14 +225,14 @@ static int lcw_dispatch_main(void *data)
         while (1) {
                 int dumplog = 1;
 
-                cfs_wait_event_interruptible(lcw_event_waitq,
-                                             is_watchdog_fired(), rc);
+               rc = wait_event_interruptible(lcw_event_waitq,
+                                             is_watchdog_fired());
                 CDEBUG(D_INFO, "Watchdog got woken up...\n");
                if (test_bit(LCW_FLAG_STOP, &lcw_flags)) {
                        CDEBUG(D_INFO, "LCW_FLAG_STOP set, shutting down...\n");
 
                        spin_lock_bh(&lcw_pending_timers_lock);
-                       rc = !cfs_list_empty(&lcw_pending_timers);
+                       rc = !list_empty(&lcw_pending_timers);
                        spin_unlock_bh(&lcw_pending_timers_lock);
                        if (rc) {
                                CERROR("pending timers list was not empty at "
@@ -240,37 +242,37 @@ static int lcw_dispatch_main(void *data)
                }
 
                spin_lock_bh(&lcw_pending_timers_lock);
-                while (!cfs_list_empty(&lcw_pending_timers)) {
-                        int is_dumplog;
-
-                        lcw = cfs_list_entry(lcw_pending_timers.next,
-                                             struct lc_watchdog, lcw_list);
-                        /* +1 ref for callback to make sure lwc wouldn't be
-                         * deleted after releasing lcw_pending_timers_lock */
-                        lcw->lcw_refcount++;
+               while (!list_empty(&lcw_pending_timers)) {
+                       int is_dumplog;
+
+                       lcw = list_entry(lcw_pending_timers.next,
+                                        struct lc_watchdog, lcw_list);
+                       /* +1 ref for callback to make sure lwc wouldn't be
+                        * deleted after releasing lcw_pending_timers_lock */
+                       lcw->lcw_refcount++;
                        spin_unlock_bh(&lcw_pending_timers_lock);
 
                        /* lock ordering */
                        spin_lock_bh(&lcw->lcw_lock);
                        spin_lock_bh(&lcw_pending_timers_lock);
 
-                       if (cfs_list_empty(&lcw->lcw_list)) {
+                       if (list_empty(&lcw->lcw_list)) {
                                /* already removed from pending list */
                                lcw->lcw_refcount--; /* -1 ref for callback */
                                if (lcw->lcw_refcount == 0)
-                                       cfs_list_add(&lcw->lcw_list, &zombies);
+                                       list_add(&lcw->lcw_list, &zombies);
                                spin_unlock_bh(&lcw->lcw_lock);
-                                /* still hold lcw_pending_timers_lock */
-                                continue;
-                        }
+                               /* still hold lcw_pending_timers_lock */
+                               continue;
+                       }
 
-                        cfs_list_del_init(&lcw->lcw_list);
-                        lcw->lcw_refcount--; /* -1 ref for pending list */
+                       list_del_init(&lcw->lcw_list);
+                       lcw->lcw_refcount--; /* -1 ref for pending list */
 
                        spin_unlock_bh(&lcw_pending_timers_lock);
                        spin_unlock_bh(&lcw->lcw_lock);
 
-                        CDEBUG(D_INFO, "found lcw for pid " LPPID "\n",
+                       CDEBUG(D_INFO, "found lcw for pid %d\n",
                                lcw->lcw_pid);
                         lcw_dump_stack(lcw);
 
@@ -285,17 +287,17 @@ static int lcw_dispatch_main(void *data)
                        spin_lock_bh(&lcw_pending_timers_lock);
                        lcw->lcw_refcount--; /* -1 ref for callback */
                        if (lcw->lcw_refcount == 0)
-                               cfs_list_add(&lcw->lcw_list, &zombies);
+                               list_add(&lcw->lcw_list, &zombies);
                }
                spin_unlock_bh(&lcw_pending_timers_lock);
 
-                while (!cfs_list_empty(&zombies)) {
-                        lcw = cfs_list_entry(lcw_pending_timers.next,
-                                         struct lc_watchdog, lcw_list);
-                        cfs_list_del(&lcw->lcw_list);
-                        LIBCFS_FREE(lcw, sizeof(*lcw));
-                }
-        }
+               while (!list_empty(&zombies)) {
+                       lcw = list_entry(zombies.next,
+                                            struct lc_watchdog, lcw_list);
+                       list_del_init(&lcw->lcw_list);
+                       LIBCFS_FREE(lcw, sizeof(*lcw));
+               }
+       }
 
        complete(&lcw_stop_completion);
 
@@ -304,14 +306,14 @@ static int lcw_dispatch_main(void *data)
 
 static void lcw_dispatch_start(void)
 {
-       cfs_task_t *task;
+       struct task_struct *task;
 
        ENTRY;
        LASSERT(lcw_refcount == 1);
 
        init_completion(&lcw_stop_completion);
        init_completion(&lcw_start_completion);
-        cfs_waitq_init(&lcw_event_waitq);
+       init_waitqueue_head(&lcw_event_waitq);
 
        CDEBUG(D_INFO, "starting dispatch thread\n");
        task = kthread_run(lcw_dispatch_main, NULL, "lc_watchdogd");
@@ -335,7 +337,7 @@ static void lcw_dispatch_stop(void)
        CDEBUG(D_INFO, "trying to stop watchdog dispatcher.\n");
 
        set_bit(LCW_FLAG_STOP, &lcw_flags);
-       cfs_waitq_signal(&lcw_event_waitq);
+       wake_up(&lcw_event_waitq);
 
        wait_for_completion(&lcw_stop_completion);
 
@@ -358,15 +360,15 @@ struct lc_watchdog *lc_watchdog_add(int timeout,
         }
 
        spin_lock_init(&lcw->lcw_lock);
-        lcw->lcw_refcount = 1; /* refcount for owner */
-        lcw->lcw_task     = cfs_current();
-        lcw->lcw_pid      = cfs_curproc_pid();
-        lcw->lcw_callback = (callback != NULL) ? callback : lc_watchdog_dumplog;
-        lcw->lcw_data     = data;
-        lcw->lcw_state    = LC_WATCHDOG_DISABLED;
+       lcw->lcw_refcount = 1; /* refcount for owner */
+       lcw->lcw_task     = current;
+       lcw->lcw_pid      = current_pid();
+       lcw->lcw_callback = (callback != NULL) ? callback : lc_watchdog_dumplog;
+       lcw->lcw_data     = data;
+       lcw->lcw_state    = LC_WATCHDOG_DISABLED;
 
-        CFS_INIT_LIST_HEAD(&lcw->lcw_list);
-        cfs_timer_init(&lcw->lcw_timer, lcw_cb, lcw);
+       INIT_LIST_HEAD(&lcw->lcw_list);
+       cfs_timer_init(&lcw->lcw_timer, lcw_cb, lcw);
 
        mutex_lock(&lcw_refcount_mutex);
        if (++lcw_refcount == 1)
@@ -386,7 +388,7 @@ EXPORT_SYMBOL(lc_watchdog_add);
 
 static void lcw_update_time(struct lc_watchdog *lcw, const char *message)
 {
-        cfs_time_t newtime = cfs_time_current();;
+       cfs_time_t newtime = cfs_time_current();
 
         if (lcw->lcw_state == LC_WATCHDOG_EXPIRED) {
                 struct timeval timediff;
@@ -409,9 +411,9 @@ static void lcw_update_time(struct lc_watchdog *lcw, const char *message)
 static void lc_watchdog_del_pending(struct lc_watchdog *lcw)
 {
        spin_lock_bh(&lcw->lcw_lock);
-       if (unlikely(!cfs_list_empty(&lcw->lcw_list))) {
+       if (unlikely(!list_empty(&lcw->lcw_list))) {
                spin_lock_bh(&lcw_pending_timers_lock);
-               cfs_list_del_init(&lcw->lcw_list);
+               list_del_init(&lcw->lcw_list);
                lcw->lcw_refcount--; /* -1 ref for pending list */
                spin_unlock_bh(&lcw_pending_timers_lock);
        }
@@ -427,10 +429,10 @@ void lc_watchdog_touch(struct lc_watchdog *lcw, int timeout)
         lc_watchdog_del_pending(lcw);
 
         lcw_update_time(lcw, "resumed");
-        lcw->lcw_state = LC_WATCHDOG_ENABLED;
 
         cfs_timer_arm(&lcw->lcw_timer, cfs_time_current() +
                       cfs_time_seconds(timeout));
+       lcw->lcw_state = LC_WATCHDOG_ENABLED;
 
         EXIT;
 }
@@ -463,8 +465,8 @@ void lc_watchdog_delete(struct lc_watchdog *lcw)
 
        spin_lock_bh(&lcw->lcw_lock);
        spin_lock_bh(&lcw_pending_timers_lock);
-       if (unlikely(!cfs_list_empty(&lcw->lcw_list))) {
-               cfs_list_del_init(&lcw->lcw_list);
+       if (unlikely(!list_empty(&lcw->lcw_list))) {
+               list_del_init(&lcw->lcw_list);
                lcw->lcw_refcount--; /* -1 ref for pending list */
        }
 
@@ -485,16 +487,6 @@ void lc_watchdog_delete(struct lc_watchdog *lcw)
 }
 EXPORT_SYMBOL(lc_watchdog_delete);
 
-/*
- * Provided watchdog handlers
- */
-
-void lc_watchdog_dumplog(pid_t pid, void *data)
-{
-        libcfs_debug_dumplog_internal((void *)((long_ptr_t)pid));
-}
-EXPORT_SYMBOL(lc_watchdog_dumplog);
-
 #else   /* !defined(WITH_WATCHDOG) */
 
 struct lc_watchdog *lc_watchdog_add(int timeout,