[patch V2 4/5] posix-cpu-timers: Expiry timers directly when in task work context

[Thomas Gleixner <tglx@linutronix.de>]

If the expiry happens in task context, there is no point in collecting the
expired timers on a list first. Just expire them directly.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 kernel/time/posix-cpu-timers.c |   92 +++++++++++++++++++++++++++++++++++------
 1 file changed, 79 insertions(+), 13 deletions(-)

--- a/kernel/time/posix-cpu-timers.c
+++ b/kernel/time/posix-cpu-timers.c
@@ -753,13 +753,71 @@ static void posix_cpu_timer_get(struct k
 
 #define MAX_COLLECTED	20
 
-static u64 collect_timerqueue(struct timerqueue_head *head,
+#ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
+/*
+ * Expiry in task work context. Access to sighand->siglock is safe here.
+ */
+static void handle_expired_timer(struct cpu_timer *ctmr,
+				 struct list_head *firing)
+{
+	struct k_itimer *timer;
+	int cpu_firing;
+
+	/*
+	 * Unlock sighand lock so the timer can be locked. Keep interrupts
+	 * disabled accross the lock switch.
+	 */
+	spin_unlock(&current->sighand->siglock);
+	timer = container_of(ctmr, struct k_itimer, it.cpu);
+	spin_lock(&timer->it_lock);
+	cpu_firing = timer->it.cpu.firing;
+	timer->it.cpu.firing = 0;
+	/*
+	 * The firing flag is -1 if this raced with a reset of the timer,
+	 * which already reported this almost-firing as an overrun.  So
+	 * don't generate an event.
+	 */
+	if (likely(cpu_firing >= 0))
+		cpu_timer_fire(timer);
+	/*
+	 * Drop timer lock again and reacquire sighand lock. Allow
+	 * interrupts to come in between so this wont block interrupts
+	 * accross the delivery of a gazillion of timers.
+	 */
+	spin_unlock_irq(&timer->it_lock);
+	spin_lock_irq(&current->sighand->siglock);
+}
+#else
+/*
+ * Expiry in interupt context. Just move them to the firing list.
+ */
+static void handle_expired_timer(struct cpu_timer *ctmr,
+				 struct list_head *firing)
+{
+	list_add_tail(&ctmr->elist, firing);
+}
+#endif
+
+static u64 collect_timerqueue(struct posix_cputimer_base *base,
 			      struct list_head *firing, u64 now)
 {
 	struct timerqueue_node *next;
 	int i = 0;
 
-	while ((next = timerqueue_getnext(head))) {
+	/*
+	 * Reset the expiry cache first when expiry context is task work.
+	 * This is required because when sighand lock is dropped new timers
+	 * can be enqueued. That's not a problem for regular posix timers
+	 * as the expiry time would be correct when expire_timerqueue()
+	 * returns, but the expiry cache is also used by itimers which do
+	 * not have a corresponding posix timer and therefore a simple
+	 * update after expire_timerqueue() might overwrite their newly
+	 * written expiry time.
+	 */
+	if (IS_ENABLED(CONFIG_POSIX_CPU_TIMERS_TASK_WORK))
+		base->nextevt = U64_MAX;
+
+	while ((next = timerqueue_getnext(&base->tqhead))) {
 		struct cpu_timer *ctmr;
 		u64 expires;
 
@@ -771,7 +829,7 @@ static u64 collect_timerqueue(struct tim
 
 		ctmr->firing = 1;
 		cpu_timer_dequeue(ctmr);
-		list_add_tail(&ctmr->elist, firing);
+		handle_expired_timer(ctmr, firing);
 	}
 
 	return U64_MAX;
@@ -783,10 +841,8 @@ static void collect_posix_cputimers(stru
 	struct posix_cputimer_base *base = pct->bases;
 	int i;
 
-	for (i = 0; i < CPUCLOCK_MAX; i++, base++) {
-		base->nextevt = collect_timerqueue(&base->tqhead, firing,
-						    samples[i]);
-	}
+	for (i = 0; i < CPUCLOCK_MAX; i++, base++)
+		base->nextevt = collect_timerqueue(base, firing, samples[i]);
 }
 
 static inline void check_dl_overrun(struct task_struct *tsk)
@@ -812,9 +868,10 @@ static bool check_rlimit(u64 time, u64 l
 }
 
 /*
- * Check for any per-thread CPU timers that have fired and move them off
- * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
- * tsk->it_*_expires values to reflect the remaining thread CPU timers.
+ * Check for any per-thread CPU timers that have fired and depending on the
+ * context (task work or interrupt) move them off the tsk->cpu_timers[N]
+ * list onto the firing list or expire them directly.  Update the expiry
+ * cache as well to reflect the remaining thread CPU timers.
  */
 static void check_thread_timers(struct task_struct *tsk,
 				struct list_head *firing)
@@ -889,9 +946,11 @@ static void check_cpu_itimer(struct task
 }
 
 /*
- * 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
- * have already been taken off.
+ * Check for any per-process CPU timers that have fired and depending on
+ * the context (task work or interrupt) move them off the tsk->signal timer
+ * list onto the firing list or expire them directly.  Update the expiry
+ * cache to reflect the resulting state. Per-thread timers have already
+ * been handled.
  */
 static void check_process_timers(struct task_struct *tsk,
 				 struct list_head *firing)
@@ -1115,6 +1174,12 @@ static void handle_posix_cpu_timers(stru
 	unlock_task_sighand(tsk, &flags);
 
 	/*
+	 * If task work delivery is enabled, the timers are already
+	 * expired.
+	 */
+	if (IS_ENABLED(CONFIG_POSIX_CPU_TIMERS_TASK_WORK))
+		goto out;
+	/*
 	 * Now that all the timers on our list have the firing flag,
 	 * no one will touch their list entries but us.  We'll take
 	 * each timer's lock before clearing its firing flag, so no
@@ -1136,6 +1201,7 @@ static void handle_posix_cpu_timers(stru
 			cpu_timer_fire(timer);
 		spin_unlock(&timer->it_lock);
 	}
+out:
 	lockdep_posixtimer_exit();
 }