]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/blobdiff - kernel/exit.c
Merge branches 'timers/clocksource', 'timers/hrtimers', 'timers/nohz', 'timers/ntp...
[linux-2.6-omap-h63xx.git] / kernel / exit.c
index 38ec40630149bd07b97f06d406d9398763a6de8c..059b38cae3848ea62fc7f03fb5a08c3dfbb53f10 100644 (file)
@@ -112,9 +112,7 @@ static void __exit_signal(struct task_struct *tsk)
                 * We won't ever get here for the group leader, since it
                 * will have been the last reference on the signal_struct.
                 */
-               sig->utime = cputime_add(sig->utime, tsk->utime);
-               sig->stime = cputime_add(sig->stime, tsk->stime);
-               sig->gtime = cputime_add(sig->gtime, tsk->gtime);
+               sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
                sig->min_flt += tsk->min_flt;
                sig->maj_flt += tsk->maj_flt;
                sig->nvcsw += tsk->nvcsw;
@@ -122,7 +120,6 @@ static void __exit_signal(struct task_struct *tsk)
                sig->inblock += task_io_get_inblock(tsk);
                sig->oublock += task_io_get_oublock(tsk);
                task_io_accounting_add(&sig->ioac, &tsk->ioac);
-               sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
                sig = NULL; /* Marker for below. */
        }
 
@@ -583,8 +580,6 @@ mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
         * If there are other users of the mm and the owner (us) is exiting
         * we need to find a new owner to take on the responsibility.
         */
-       if (!mm)
-               return 0;
        if (atomic_read(&mm->mm_users) <= 1)
                return 0;
        if (mm->owner != p)
@@ -627,29 +622,38 @@ retry:
        } while_each_thread(g, c);
 
        read_unlock(&tasklist_lock);
+       /*
+        * We found no owner yet mm_users > 1: this implies that we are
+        * most likely racing with swapoff (try_to_unuse()) or /proc or
+        * ptrace or page migration (get_task_mm()).  Mark owner as NULL,
+        * so that subsystems can understand the callback and take action.
+        */
+       down_write(&mm->mmap_sem);
+       cgroup_mm_owner_callbacks(mm->owner, NULL);
+       mm->owner = NULL;
+       up_write(&mm->mmap_sem);
        return;
 
 assign_new_owner:
        BUG_ON(c == p);
        get_task_struct(c);
+       read_unlock(&tasklist_lock);
+       down_write(&mm->mmap_sem);
        /*
         * The task_lock protects c->mm from changing.
         * We always want mm->owner->mm == mm
         */
        task_lock(c);
-       /*
-        * Delay read_unlock() till we have the task_lock()
-        * to ensure that c does not slip away underneath us
-        */
-       read_unlock(&tasklist_lock);
        if (c->mm != mm) {
                task_unlock(c);
+               up_write(&mm->mmap_sem);
                put_task_struct(c);
                goto retry;
        }
        cgroup_mm_owner_callbacks(mm->owner, c);
        mm->owner = c;
        task_unlock(c);
+       up_write(&mm->mmap_sem);
        put_task_struct(c);
 }
 #endif /* CONFIG_MM_OWNER */
@@ -831,26 +835,50 @@ static void reparent_thread(struct task_struct *p, struct task_struct *father)
  * the child reaper process (ie "init") in our pid
  * space.
  */
+static struct task_struct *find_new_reaper(struct task_struct *father)
+{
+       struct pid_namespace *pid_ns = task_active_pid_ns(father);
+       struct task_struct *thread;
+
+       thread = father;
+       while_each_thread(father, thread) {
+               if (thread->flags & PF_EXITING)
+                       continue;
+               if (unlikely(pid_ns->child_reaper == father))
+                       pid_ns->child_reaper = thread;
+               return thread;
+       }
+
+       if (unlikely(pid_ns->child_reaper == father)) {
+               write_unlock_irq(&tasklist_lock);
+               if (unlikely(pid_ns == &init_pid_ns))
+                       panic("Attempted to kill init!");
+
+               zap_pid_ns_processes(pid_ns);
+               write_lock_irq(&tasklist_lock);
+               /*
+                * We can not clear ->child_reaper or leave it alone.
+                * There may by stealth EXIT_DEAD tasks on ->children,
+                * forget_original_parent() must move them somewhere.
+                */
+               pid_ns->child_reaper = init_pid_ns.child_reaper;
+       }
+
+       return pid_ns->child_reaper;
+}
+
 static void forget_original_parent(struct task_struct *father)
 {
-       struct task_struct *p, *n, *reaper = father;
+       struct task_struct *p, *n, *reaper;
        LIST_HEAD(ptrace_dead);
 
        write_lock_irq(&tasklist_lock);
-
+       reaper = find_new_reaper(father);
        /*
         * First clean up ptrace if we were using it.
         */
        ptrace_exit(father, &ptrace_dead);
 
-       do {
-               reaper = next_thread(reaper);
-               if (reaper == father) {
-                       reaper = task_child_reaper(father);
-                       break;
-               }
-       } while (reaper->flags & PF_EXITING);
-
        list_for_each_entry_safe(p, n, &father->children, sibling) {
                p->real_parent = reaper;
                if (p->parent == father) {
@@ -918,8 +946,8 @@ static void exit_notify(struct task_struct *tsk, int group_dead)
 
        /* mt-exec, de_thread() is waiting for us */
        if (thread_group_leader(tsk) &&
-           tsk->signal->notify_count < 0 &&
-           tsk->signal->group_exit_task)
+           tsk->signal->group_exit_task &&
+           tsk->signal->notify_count < 0)
                wake_up_process(tsk->signal->group_exit_task);
 
        write_unlock_irq(&tasklist_lock);
@@ -959,39 +987,6 @@ static void check_stack_usage(void)
 static inline void check_stack_usage(void) {}
 #endif
 
-static inline void exit_child_reaper(struct task_struct *tsk)
-{
-       if (likely(tsk->group_leader != task_child_reaper(tsk)))
-               return;
-
-       if (tsk->nsproxy->pid_ns == &init_pid_ns)
-               panic("Attempted to kill init!");
-
-       /*
-        * @tsk is the last thread in the 'cgroup-init' and is exiting.
-        * Terminate all remaining processes in the namespace and reap them
-        * before exiting @tsk.
-        *
-        * Note that @tsk (last thread of cgroup-init) may not necessarily
-        * be the child-reaper (i.e main thread of cgroup-init) of the
-        * namespace i.e the child_reaper may have already exited.
-        *
-        * Even after a child_reaper exits, we let it inherit orphaned children,
-        * because, pid_ns->child_reaper remains valid as long as there is
-        * at least one living sub-thread in the cgroup init.
-
-        * This living sub-thread of the cgroup-init will be notified when
-        * a child inherited by the 'child-reaper' exits (do_notify_parent()
-        * uses __group_send_sig_info()). Further, when reaping child processes,
-        * do_wait() iterates over children of all living sub threads.
-
-        * i.e even though 'child_reaper' thread is listed as the parent of the
-        * orphaned children, any living sub-thread in the cgroup-init can
-        * perform the role of the child_reaper.
-        */
-       zap_pid_ns_processes(tsk->nsproxy->pid_ns);
-}
-
 NORET_TYPE void do_exit(long code)
 {
        struct task_struct *tsk = current;
@@ -1051,7 +1046,6 @@ NORET_TYPE void do_exit(long code)
        }
        group_dead = atomic_dec_and_test(&tsk->signal->live);
        if (group_dead) {
-               exit_child_reaper(tsk);
                hrtimer_cancel(&tsk->signal->real_timer);
                exit_itimers(tsk->signal);
        }
@@ -1304,6 +1298,7 @@ static int wait_task_zombie(struct task_struct *p, int options,
        if (likely(!traced)) {
                struct signal_struct *psig;
                struct signal_struct *sig;
+               struct task_cputime cputime;
 
                /*
                 * The resource counters for the group leader are in its
@@ -1319,20 +1314,23 @@ static int wait_task_zombie(struct task_struct *p, int options,
                 * need to protect the access to p->parent->signal fields,
                 * as other threads in the parent group can be right
                 * here reaping other children at the same time.
+                *
+                * We use thread_group_cputime() to get times for the thread
+                * group, which consolidates times for all threads in the
+                * group including the group leader.
                 */
                spin_lock_irq(&p->parent->sighand->siglock);
                psig = p->parent->signal;
                sig = p->signal;
+               thread_group_cputime(p, &cputime);
                psig->cutime =
                        cputime_add(psig->cutime,
-                       cputime_add(p->utime,
-                       cputime_add(sig->utime,
-                                   sig->cutime)));
+                       cputime_add(cputime.utime,
+                                   sig->cutime));
                psig->cstime =
                        cputime_add(psig->cstime,
-                       cputime_add(p->stime,
-                       cputime_add(sig->stime,
-                                   sig->cstime)));
+                       cputime_add(cputime.stime,
+                                   sig->cstime));
                psig->cgtime =
                        cputime_add(psig->cgtime,
                        cputime_add(p->gtime,