sched: fix process time monotonicity
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / kernel / exit.c
index c3691cbc220ae4fe77e37b58a86577bad0ddb3e2..16395644a98ff8c060b2f5fd776fe7abadd61c6a 100644 (file)
@@ -112,27 +112,16 @@ 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->utime = cputime_add(sig->utime, task_utime(tsk));
+               sig->stime = cputime_add(sig->stime, task_stime(tsk));
+               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;
                sig->nivcsw += tsk->nivcsw;
                sig->inblock += task_io_get_inblock(tsk);
                sig->oublock += task_io_get_oublock(tsk);
-#ifdef CONFIG_TASK_XACCT
-               sig->rchar += tsk->rchar;
-               sig->wchar += tsk->wchar;
-               sig->syscr += tsk->syscr;
-               sig->syscw += tsk->syscw;
-#endif /* CONFIG_TASK_XACCT */
-#ifdef CONFIG_TASK_IO_ACCOUNTING
-               sig->ioac.read_bytes += tsk->ioac.read_bytes;
-               sig->ioac.write_bytes += tsk->ioac.write_bytes;
-               sig->ioac.cancelled_write_bytes +=
-                                       tsk->ioac.cancelled_write_bytes;
-#endif /* CONFIG_TASK_IO_ACCOUNTING */
+               task_io_accounting_add(&sig->ioac, &tsk->ioac);
                sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
                sig = NULL; /* Marker for below. */
        }
@@ -163,27 +152,17 @@ static void delayed_put_task_struct(struct rcu_head *rhp)
        put_task_struct(container_of(rhp, struct task_struct, rcu));
 }
 
-/*
- * Do final ptrace-related cleanup of a zombie being reaped.
- *
- * Called with write_lock(&tasklist_lock) held.
- */
-static void ptrace_release_task(struct task_struct *p)
-{
-       BUG_ON(!list_empty(&p->ptraced));
-       ptrace_unlink(p);
-       BUG_ON(!list_empty(&p->ptrace_entry));
-}
 
 void release_task(struct task_struct * p)
 {
        struct task_struct *leader;
        int zap_leader;
 repeat:
+       tracehook_prepare_release_task(p);
        atomic_dec(&p->user->processes);
        proc_flush_task(p);
        write_lock_irq(&tasklist_lock);
-       ptrace_release_task(p);
+       tracehook_finish_release_task(p);
        __exit_signal(p);
 
        /*
@@ -205,6 +184,13 @@ repeat:
                 * that case.
                 */
                zap_leader = task_detached(leader);
+
+               /*
+                * This maintains the invariant that release_task()
+                * only runs on a task in EXIT_DEAD, just for sanity.
+                */
+               if (zap_leader)
+                       leader->exit_state = EXIT_DEAD;
        }
 
        write_unlock_irq(&tasklist_lock);
@@ -568,8 +554,6 @@ void put_fs_struct(struct fs_struct *fs)
        if (atomic_dec_and_test(&fs->count)) {
                path_put(&fs->root);
                path_put(&fs->pwd);
-               if (fs->altroot.dentry)
-                       path_put(&fs->altroot);
                kmem_cache_free(fs_cachep, fs);
        }
 }
@@ -847,26 +831,50 @@ 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) {
@@ -888,7 +896,8 @@ static void forget_original_parent(struct task_struct *father)
  */
 static void exit_notify(struct task_struct *tsk, int group_dead)
 {
-       int state;
+       int signal;
+       void *cookie;
 
        /*
         * This does two things:
@@ -925,33 +934,24 @@ static void exit_notify(struct task_struct *tsk, int group_dead)
            !capable(CAP_KILL))
                tsk->exit_signal = SIGCHLD;
 
-       /* If something other than our normal parent is ptracing us, then
-        * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
-        * only has special meaning to our real parent.
-        */
-       if (!task_detached(tsk) && thread_group_empty(tsk)) {
-               int signal = ptrace_reparented(tsk) ?
-                               SIGCHLD : tsk->exit_signal;
-               do_notify_parent(tsk, signal);
-       } else if (tsk->ptrace) {
-               do_notify_parent(tsk, SIGCHLD);
-       }
+       signal = tracehook_notify_death(tsk, &cookie, group_dead);
+       if (signal >= 0)
+               signal = do_notify_parent(tsk, signal);
 
-       state = EXIT_ZOMBIE;
-       if (task_detached(tsk) && likely(!tsk->ptrace))
-               state = EXIT_DEAD;
-       tsk->exit_state = state;
+       tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
 
        /* 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);
 
+       tracehook_report_death(tsk, signal, cookie, group_dead);
+
        /* If the process is dead, release it - nobody will wait for it */
-       if (state == EXIT_DEAD)
+       if (signal == DEATH_REAP)
                release_task(tsk);
 }
 
@@ -983,39 +983,6 @@ 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;
@@ -1075,7 +1042,6 @@ 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);
        }
@@ -1376,21 +1342,8 @@ static int wait_task_zombie(struct task_struct *p, int options,
                psig->coublock +=
                        task_io_get_oublock(p) +
                        sig->oublock + sig->coublock;
-#ifdef CONFIG_TASK_XACCT
-               psig->rchar += p->rchar + sig->rchar;
-               psig->wchar += p->wchar + sig->wchar;
-               psig->syscr += p->syscr + sig->syscr;
-               psig->syscw += p->syscw + sig->syscw;
-#endif /* CONFIG_TASK_XACCT */
-#ifdef CONFIG_TASK_IO_ACCOUNTING
-               psig->ioac.read_bytes +=
-                       p->ioac.read_bytes + sig->ioac.read_bytes;
-               psig->ioac.write_bytes +=
-                       p->ioac.write_bytes + sig->ioac.write_bytes;
-               psig->ioac.cancelled_write_bytes +=
-                               p->ioac.cancelled_write_bytes +
-                               sig->ioac.cancelled_write_bytes;
-#endif /* CONFIG_TASK_IO_ACCOUNTING */
+               task_io_accounting_add(&psig->ioac, &p->ioac);
+               task_io_accounting_add(&psig->ioac, &sig->ioac);
                spin_unlock_irq(&p->parent->sighand->siglock);
        }