nfsd: clean up readdirplus encoding
[safe/jmp/linux-2.6] / kernel / posix-cpu-timers.c
index dba1c33..e33a21c 100644 (file)
@@ -7,83 +7,7 @@
 #include <linux/errno.h>
 #include <linux/math64.h>
 #include <asm/uaccess.h>
-
-#ifdef CONFIG_SMP
-/*
- * Allocate the thread_group_cputime structure appropriately for SMP kernels
- * and fill in the current values of the fields.  Called from copy_signal()
- * via thread_group_cputime_clone_thread() when adding a second or subsequent
- * thread to a thread group.  Assumes interrupts are enabled when called.
- */
-int thread_group_cputime_alloc_smp(struct task_struct *tsk)
-{
-       struct signal_struct *sig = tsk->signal;
-       struct task_cputime *cputime;
-
-       /*
-        * If we have multiple threads and we don't already have a
-        * per-CPU task_cputime struct, allocate one and fill it in with
-        * the times accumulated so far.
-        */
-       if (sig->cputime.totals)
-               return 0;
-       cputime = alloc_percpu(struct task_cputime);
-       if (cputime == NULL)
-               return -ENOMEM;
-       read_lock(&tasklist_lock);
-       spin_lock_irq(&tsk->sighand->siglock);
-       if (sig->cputime.totals) {
-               spin_unlock_irq(&tsk->sighand->siglock);
-               read_unlock(&tasklist_lock);
-               free_percpu(cputime);
-               return 0;
-       }
-       sig->cputime.totals = cputime;
-       cputime = per_cpu_ptr(sig->cputime.totals, get_cpu());
-       cputime->utime = tsk->utime;
-       cputime->stime = tsk->stime;
-       cputime->sum_exec_runtime = tsk->se.sum_exec_runtime;
-       put_cpu_no_resched();
-       spin_unlock_irq(&tsk->sighand->siglock);
-       read_unlock(&tasklist_lock);
-       return 0;
-}
-
-/**
- * thread_group_cputime_smp - Sum the thread group time fields across all CPUs.
- *
- * @tsk:       The task we use to identify the thread group.
- * @times:     task_cputime structure in which we return the summed fields.
- *
- * Walk the list of CPUs to sum the per-CPU time fields in the thread group
- * time structure.
- */
-void thread_group_cputime_smp(
-       struct task_struct *tsk,
-       struct task_cputime *times)
-{
-       struct signal_struct *sig;
-       int i;
-       struct task_cputime *tot;
-
-       sig = tsk->signal;
-       if (unlikely(!sig) || !sig->cputime.totals) {
-               times->utime = tsk->utime;
-               times->stime = tsk->stime;
-               times->sum_exec_runtime = tsk->se.sum_exec_runtime;
-               return;
-       }
-       times->stime = times->utime = cputime_zero;
-       times->sum_exec_runtime = 0;
-       for_each_possible_cpu(i) {
-               tot = per_cpu_ptr(tsk->signal->cputime.totals, i);
-               times->utime = cputime_add(times->utime, tot->utime);
-               times->stime = cputime_add(times->stime, tot->stime);
-               times->sum_exec_runtime += tot->sum_exec_runtime;
-       }
-}
-
-#endif /* CONFIG_SMP */
+#include <linux/kernel_stat.h>
 
 /*
  * Called after updating RLIMIT_CPU to set timer expiration if necessary.
@@ -94,7 +18,7 @@ void update_rlimit_cpu(unsigned long rlim_new)
 
        cputime = secs_to_cputime(rlim_new);
        if (cputime_eq(current->signal->it_prof_expires, cputime_zero) ||
-            cputime_lt(current->signal->it_prof_expires, cputime)) {
+           cputime_gt(current->signal->it_prof_expires, cputime)) {
                spin_lock_irq(&current->sighand->siglock);
                set_process_cpu_timer(current, CPUCLOCK_PROF, &cputime, NULL);
                spin_unlock_irq(&current->sighand->siglock);
@@ -306,25 +230,90 @@ static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
        return 0;
 }
 
+void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times)
+{
+       struct sighand_struct *sighand;
+       struct signal_struct *sig;
+       struct task_struct *t;
+
+       *times = INIT_CPUTIME;
+
+       rcu_read_lock();
+       sighand = rcu_dereference(tsk->sighand);
+       if (!sighand)
+               goto out;
+
+       sig = tsk->signal;
+
+       t = tsk;
+       do {
+               times->utime = cputime_add(times->utime, t->utime);
+               times->stime = cputime_add(times->stime, t->stime);
+               times->sum_exec_runtime += t->se.sum_exec_runtime;
+
+               t = next_thread(t);
+       } while (t != tsk);
+
+       times->utime = cputime_add(times->utime, sig->utime);
+       times->stime = cputime_add(times->stime, sig->stime);
+       times->sum_exec_runtime += sig->sum_sched_runtime;
+out:
+       rcu_read_unlock();
+}
+
+static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b)
+{
+       if (cputime_gt(b->utime, a->utime))
+               a->utime = b->utime;
+
+       if (cputime_gt(b->stime, a->stime))
+               a->stime = b->stime;
+
+       if (b->sum_exec_runtime > a->sum_exec_runtime)
+               a->sum_exec_runtime = b->sum_exec_runtime;
+}
+
+void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
+{
+       struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
+       struct task_cputime sum;
+       unsigned long flags;
+
+       spin_lock_irqsave(&cputimer->lock, flags);
+       if (!cputimer->running) {
+               cputimer->running = 1;
+               /*
+                * The POSIX timer interface allows for absolute time expiry
+                * values through the TIMER_ABSTIME flag, therefore we have
+                * to synchronize the timer to the clock every time we start
+                * it.
+                */
+               thread_group_cputime(tsk, &sum);
+               update_gt_cputime(&cputimer->cputime, &sum);
+       }
+       *times = cputimer->cputime;
+       spin_unlock_irqrestore(&cputimer->lock, flags);
+}
+
 /*
  * Sample a process (thread group) clock for the given group_leader task.
  * Must be called with tasklist_lock held for reading.
- * Must be called with tasklist_lock held for reading, and p->sighand->siglock.
  */
-static int cpu_clock_sample_group_locked(unsigned int clock_idx,
-                                        struct task_struct *p,
-                                        union cpu_time_count *cpu)
+static int cpu_clock_sample_group(const clockid_t which_clock,
+                                 struct task_struct *p,
+                                 union cpu_time_count *cpu)
 {
        struct task_cputime cputime;
 
-       thread_group_cputime(p, &cputime);
-       switch (clock_idx) {
+       switch (CPUCLOCK_WHICH(which_clock)) {
        default:
                return -EINVAL;
        case CPUCLOCK_PROF:
+               thread_group_cputime(p, &cputime);
                cpu->cpu = cputime_add(cputime.utime, cputime.stime);
                break;
        case CPUCLOCK_VIRT:
+               thread_group_cputime(p, &cputime);
                cpu->cpu = cputime.utime;
                break;
        case CPUCLOCK_SCHED:
@@ -334,23 +323,6 @@ static int cpu_clock_sample_group_locked(unsigned int clock_idx,
        return 0;
 }
 
-/*
- * Sample a process (thread group) clock for the given group_leader task.
- * Must be called with tasklist_lock held for reading.
- */
-static int cpu_clock_sample_group(const clockid_t which_clock,
-                                 struct task_struct *p,
-                                 union cpu_time_count *cpu)
-{
-       int ret;
-       unsigned long flags;
-       spin_lock_irqsave(&p->sighand->siglock, flags);
-       ret = cpu_clock_sample_group_locked(CPUCLOCK_WHICH(which_clock), p,
-                                           cpu);
-       spin_unlock_irqrestore(&p->sighand->siglock, flags);
-       return ret;
-}
-
 
 int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
 {
@@ -549,11 +521,12 @@ void posix_cpu_timers_exit(struct task_struct *tsk)
 }
 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 {
-       struct task_cputime cputime;
+       struct signal_struct *const sig = tsk->signal;
 
-       thread_group_cputime(tsk, &cputime);
        cleanup_timers(tsk->signal->cpu_timers,
-                      cputime.utime, cputime.stime, cputime.sum_exec_runtime);
+                      cputime_add(tsk->utime, sig->utime),
+                      cputime_add(tsk->stime, sig->stime),
+                      tsk->se.sum_exec_runtime + sig->sum_sched_runtime);
 }
 
 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
@@ -710,6 +683,33 @@ static void cpu_timer_fire(struct k_itimer *timer)
 }
 
 /*
+ * Sample a process (thread group) timer for the given group_leader task.
+ * Must be called with tasklist_lock held for reading.
+ */
+static int cpu_timer_sample_group(const clockid_t which_clock,
+                                 struct task_struct *p,
+                                 union cpu_time_count *cpu)
+{
+       struct task_cputime cputime;
+
+       thread_group_cputimer(p, &cputime);
+       switch (CPUCLOCK_WHICH(which_clock)) {
+       default:
+               return -EINVAL;
+       case CPUCLOCK_PROF:
+               cpu->cpu = cputime_add(cputime.utime, cputime.stime);
+               break;
+       case CPUCLOCK_VIRT:
+               cpu->cpu = cputime.utime;
+               break;
+       case CPUCLOCK_SCHED:
+               cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
+               break;
+       }
+       return 0;
+}
+
+/*
  * Guts of sys_timer_settime for CPU timers.
  * This is called with the timer locked and interrupts disabled.
  * If we return TIMER_RETRY, it's necessary to release the timer's lock
@@ -770,7 +770,7 @@ int posix_cpu_timer_set(struct k_itimer *timer, int flags,
        if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
                cpu_clock_sample(timer->it_clock, p, &val);
        } else {
-               cpu_clock_sample_group(timer->it_clock, p, &val);
+               cpu_timer_sample_group(timer->it_clock, p, &val);
        }
 
        if (old) {
@@ -918,7 +918,7 @@ void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
                        read_unlock(&tasklist_lock);
                        goto dead;
                } else {
-                       cpu_clock_sample_group(timer->it_clock, p, &now);
+                       cpu_timer_sample_group(timer->it_clock, p, &now);
                        clear_dead = (unlikely(p->exit_state) &&
                                      thread_group_empty(p));
                }
@@ -1058,6 +1058,19 @@ static void check_thread_timers(struct task_struct *tsk,
        }
 }
 
+static void stop_process_timers(struct task_struct *tsk)
+{
+       struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
+       unsigned long flags;
+
+       if (!cputimer->running)
+               return;
+
+       spin_lock_irqsave(&cputimer->lock, flags);
+       cputimer->running = 0;
+       spin_unlock_irqrestore(&cputimer->lock, flags);
+}
+
 /*
  * Check for any per-thread CPU timers that have fired and move them
  * off the tsk->*_timers list onto the firing list.  Per-thread timers
@@ -1081,13 +1094,15 @@ static void check_process_timers(struct task_struct *tsk,
            sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
            list_empty(&timers[CPUCLOCK_VIRT]) &&
            cputime_eq(sig->it_virt_expires, cputime_zero) &&
-           list_empty(&timers[CPUCLOCK_SCHED]))
+           list_empty(&timers[CPUCLOCK_SCHED])) {
+               stop_process_timers(tsk);
                return;
+       }
 
        /*
         * Collect the current process totals.
         */
-       thread_group_cputime(tsk, &cputime);
+       thread_group_cputimer(tsk, &cputime);
        utime = cputime.utime;
        ptime = cputime_add(utime, cputime.stime);
        sum_sched_runtime = cputime.sum_exec_runtime;
@@ -1258,7 +1273,7 @@ void posix_cpu_timer_schedule(struct k_itimer *timer)
                        clear_dead_task(timer, now);
                        goto out_unlock;
                }
-               cpu_clock_sample_group(timer->it_clock, p, &now);
+               cpu_timer_sample_group(timer->it_clock, p, &now);
                bump_cpu_timer(timer, now);
                /* Leave the tasklist_lock locked for the call below.  */
        }
@@ -1324,29 +1339,41 @@ static inline int task_cputime_expired(const struct task_cputime *sample,
  * fastpath_timer_check - POSIX CPU timers fast path.
  *
  * @tsk:       The task (thread) being checked.
- * @sig:       The signal pointer for that task.
  *
- * If there are no timers set return false.  Otherwise snapshot the task and
- * thread group timers, then compare them with the corresponding expiration
- # times.  Returns true if a timer has expired, else returns false.
+ * Check the task and thread group timers.  If both are zero (there are no
+ * timers set) return false.  Otherwise snapshot the task and thread group
+ * timers and compare them with the corresponding expiration times.  Return
+ * true if a timer has expired, else return false.
  */
-static inline int fastpath_timer_check(struct task_struct *tsk,
-                                       struct signal_struct *sig)
+static inline int fastpath_timer_check(struct task_struct *tsk)
 {
-       struct task_cputime task_sample = {
-               .utime = tsk->utime,
-               .stime = tsk->stime,
-               .sum_exec_runtime = tsk->se.sum_exec_runtime
-       };
-       struct task_cputime group_sample;
+       struct signal_struct *sig;
 
-       if (task_cputime_zero(&tsk->cputime_expires) &&
-           task_cputime_zero(&sig->cputime_expires))
+       /* tsk == current, ensure it is safe to use ->signal/sighand */
+       if (unlikely(tsk->exit_state))
                return 0;
-       if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
-               return 1;
-       thread_group_cputime(tsk, &group_sample);
-       return task_cputime_expired(&group_sample, &sig->cputime_expires);
+
+       if (!task_cputime_zero(&tsk->cputime_expires)) {
+               struct task_cputime task_sample = {
+                       .utime = tsk->utime,
+                       .stime = tsk->stime,
+                       .sum_exec_runtime = tsk->se.sum_exec_runtime
+               };
+
+               if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
+                       return 1;
+       }
+
+       sig = tsk->signal;
+       if (!task_cputime_zero(&sig->cputime_expires)) {
+               struct task_cputime group_sample;
+
+               thread_group_cputimer(tsk, &group_sample);
+               if (task_cputime_expired(&group_sample, &sig->cputime_expires))
+                       return 1;
+       }
+
+       return sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY;
 }
 
 /*
@@ -1358,43 +1385,34 @@ void run_posix_cpu_timers(struct task_struct *tsk)
 {
        LIST_HEAD(firing);
        struct k_itimer *timer, *next;
-       struct signal_struct *sig;
-       struct sighand_struct *sighand;
-       unsigned long flags;
 
        BUG_ON(!irqs_disabled());
 
-       /* Pick up tsk->signal and make sure it's valid. */
-       sig = tsk->signal;
        /*
         * The fast path checks that there are no expired thread or thread
-        * group timers.  If that's so, just return.  Also check that
-        * tsk->signal is non-NULL; this probably can't happen but cover the
-        * possibility anyway.
+        * group timers.  If that's so, just return.
         */
-       if (unlikely(!sig) || !fastpath_timer_check(tsk, sig)) {
+       if (!fastpath_timer_check(tsk))
                return;
-       }
-       sighand = lock_task_sighand(tsk, &flags);
-       if (likely(sighand)) {
-               /*
-                * Here we take off tsk->signal->cpu_timers[N] and
-                * tsk->cpu_timers[N] all the timers that are firing, and
-                * put them on the firing list.
-                */
-               check_thread_timers(tsk, &firing);
-               check_process_timers(tsk, &firing);
 
-               /*
-                * We must release these locks before taking any timer's lock.
-                * There is a potential race with timer deletion here, as the
-                * siglock now protects our private firing list.  We have set
-                * the firing flag in each timer, so that a deletion attempt
-                * that gets the timer lock before we do will give it up and
-                * spin until we've taken care of that timer below.
-                */
-       }
-       unlock_task_sighand(tsk, &flags);
+       spin_lock(&tsk->sighand->siglock);
+       /*
+        * Here we take off tsk->signal->cpu_timers[N] and
+        * tsk->cpu_timers[N] all the timers that are firing, and
+        * put them on the firing list.
+        */
+       check_thread_timers(tsk, &firing);
+       check_process_timers(tsk, &firing);
+
+       /*
+        * We must release these locks before taking any timer's lock.
+        * There is a potential race with timer deletion here, as the
+        * siglock now protects our private firing list.  We have set
+        * the firing flag in each timer, so that a deletion attempt
+        * that gets the timer lock before we do will give it up and
+        * spin until we've taken care of that timer below.
+        */
+       spin_unlock(&tsk->sighand->siglock);
 
        /*
         * Now that all the timers on our list have the firing flag,
@@ -1403,19 +1421,19 @@ void run_posix_cpu_timers(struct task_struct *tsk)
         * timer call will interfere.
         */
        list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
-               int firing;
+               int cpu_firing;
+
                spin_lock(&timer->it_lock);
                list_del_init(&timer->it.cpu.entry);
-               firing = timer->it.cpu.firing;
+               cpu_firing = timer->it.cpu.firing;
                timer->it.cpu.firing = 0;
                /*
                 * The firing flag is -1 if we collided with a reset
                 * of the timer, which already reported this
                 * almost-firing as an overrun.  So don't generate an event.
                 */
-               if (likely(firing >= 0)) {
+               if (likely(cpu_firing >= 0))
                        cpu_timer_fire(timer);
-               }
                spin_unlock(&timer->it_lock);
        }
 }
@@ -1433,7 +1451,7 @@ void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
        struct list_head *head;
 
        BUG_ON(clock_idx == CPUCLOCK_SCHED);
-       cpu_clock_sample_group_locked(clock_idx, tsk, &now);
+       cpu_timer_sample_group(clock_idx, tsk, &now);
 
        if (oldval) {
                if (!cputime_eq(*oldval, cputime_zero)) {