[PATCH v3 4/4] perf: Split __perf_pending_irq() out of perf_pending_irq()

[Sebastian Andrzej Siewior]

perf_pending_irq() invokes perf_event_wakeup() and __perf_pending_irq().
The former is in charge of waking any tasks which wait to be woken up
while the latter disables perf-events.

The irq_work perf_pending_irq(), while this an irq_work, the callback
is invoked in thread context on PREEMPT_RT. This is needed because all
the waking functions (wake_up_all(), kill_fasync()) acquire sleep locks
which must not be used with disabled interrupts.
Disabling events, as done by __perf_pending_irq(), expects a hardirq
context and disabled interrupts. This requirement is not fulfilled on
PREEMPT_RT.

Split functionality based on perf_event::pending_disable into irq_work
named `pending_disable_irq' and invoke it in hardirq context on
PREEMPT_RT. Rename the split out callback to perf_pending_disable().

Tested-by: Marco Elver <elver@xxxxxxxxxx>
Tested-by: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>
---
 include/linux/perf_event.h |  1 +
 kernel/events/core.c       | 31 +++++++++++++++++++++++--------
 2 files changed, 24 insertions(+), 8 deletions(-)

diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 24ac6765146c7..c1c6600541657 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -783,6 +783,7 @@ struct perf_event {
 	unsigned int			pending_disable;
 	unsigned long			pending_addr;	/* SIGTRAP */
 	struct irq_work			pending_irq;
+	struct irq_work			pending_disable_irq;
 	struct callback_head		pending_task;
 	unsigned int			pending_work;
 
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 5400f7ed2f98b..7266265ed8cc3 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -2449,7 +2449,7 @@ static void __perf_event_disable(struct perf_event *event,
  * hold the top-level event's child_mutex, so any descendant that
  * goes to exit will block in perf_event_exit_event().
  *
- * When called from perf_pending_irq it's OK because event->ctx
+ * When called from perf_pending_disable it's OK because event->ctx
  * is the current context on this CPU and preemption is disabled,
  * hence we can't get into perf_event_task_sched_out for this context.
  */
@@ -2489,7 +2489,7 @@ EXPORT_SYMBOL_GPL(perf_event_disable);
 void perf_event_disable_inatomic(struct perf_event *event)
 {
 	event->pending_disable = 1;
-	irq_work_queue(&event->pending_irq);
+	irq_work_queue(&event->pending_disable_irq);
 }
 
 #define MAX_INTERRUPTS (~0ULL)
@@ -5175,6 +5175,7 @@ static void perf_addr_filters_splice(struct perf_event *event,
 static void _free_event(struct perf_event *event)
 {
 	irq_work_sync(&event->pending_irq);
+	irq_work_sync(&event->pending_disable_irq);
 
 	unaccount_event(event);
 
@@ -6711,7 +6712,7 @@ static void perf_sigtrap(struct perf_event *event)
 /*
  * Deliver the pending work in-event-context or follow the context.
  */
-static void __perf_pending_irq(struct perf_event *event)
+static void __perf_pending_disable(struct perf_event *event)
 {
 	int cpu = READ_ONCE(event->oncpu);
 
@@ -6749,11 +6750,26 @@ static void __perf_pending_irq(struct perf_event *event)
 	 *				  irq_work_queue(); // FAILS
 	 *
 	 *  irq_work_run()
-	 *    perf_pending_irq()
+	 *    perf_pending_disable()
 	 *
 	 * But the event runs on CPU-B and wants disabling there.
 	 */
-	irq_work_queue_on(&event->pending_irq, cpu);
+	irq_work_queue_on(&event->pending_disable_irq, cpu);
+}
+
+static void perf_pending_disable(struct irq_work *entry)
+{
+	struct perf_event *event = container_of(entry, struct perf_event, pending_disable_irq);
+	int rctx;
+
+	/*
+	 * If we 'fail' here, that's OK, it means recursion is already disabled
+	 * and we won't recurse 'further'.
+	 */
+	rctx = perf_swevent_get_recursion_context();
+	__perf_pending_disable(event);
+	if (rctx >= 0)
+		perf_swevent_put_recursion_context(rctx);
 }
 
 static void perf_pending_irq(struct irq_work *entry)
@@ -6776,8 +6792,6 @@ static void perf_pending_irq(struct irq_work *entry)
 		perf_event_wakeup(event);
 	}
 
-	__perf_pending_irq(event);
-
 	if (rctx >= 0)
 		perf_swevent_put_recursion_context(rctx);
 }
@@ -9572,7 +9586,7 @@ static int __perf_event_overflow(struct perf_event *event,
 			 * is processed.
 			 */
 			if (in_nmi())
-				irq_work_queue(&event->pending_irq);
+				irq_work_queue(&event->pending_disable_irq);
 		} else if (event->attr.exclude_kernel && valid_sample) {
 			/*
 			 * Should not be able to return to user space without
@@ -11912,6 +11926,7 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu,
 
 	init_waitqueue_head(&event->waitq);
 	init_irq_work(&event->pending_irq, perf_pending_irq);
+	event->pending_disable_irq = IRQ_WORK_INIT_HARD(perf_pending_disable);
 	init_task_work(&event->pending_task, perf_pending_task);
 
 	mutex_init(&event->mmap_mutex);
-- 
2.43.0