perf/core: Add a new PERF_EV_CAP_SIBLING event capability
Current perf assumes that events in a group are independent. Close an
event doesn't impact the value of the other events in the same group.
If the closed event is a member, after the event closure, other events
are still running like a group. If the closed event is a leader, other
events are running as singleton events.
Add PERF_EV_CAP_SIBLING to allow events to indicate they require being
part of a group, and when the leader dies they cannot exist
independently.
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200723171117.9918-8-kan.liang@linux.intel.com
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 5bfe8e3..57efe3b 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -2133,8 +2133,24 @@ static inline struct list_head *get_event_list(struct perf_event *event)
return event->attr.pinned ? &ctx->pinned_active : &ctx->flexible_active;
}
+/*
+ * Events that have PERF_EV_CAP_SIBLING require being part of a group and
+ * cannot exist on their own, schedule them out and move them into the ERROR
+ * state. Also see _perf_event_enable(), it will not be able to recover
+ * this ERROR state.
+ */
+static inline void perf_remove_sibling_event(struct perf_event *event)
+{
+ struct perf_event_context *ctx = event->ctx;
+ struct perf_cpu_context *cpuctx = __get_cpu_context(ctx);
+
+ event_sched_out(event, cpuctx, ctx);
+ perf_event_set_state(event, PERF_EVENT_STATE_ERROR);
+}
+
static void perf_group_detach(struct perf_event *event)
{
+ struct perf_event *leader = event->group_leader;
struct perf_event *sibling, *tmp;
struct perf_event_context *ctx = event->ctx;
@@ -2153,7 +2169,7 @@ static void perf_group_detach(struct perf_event *event)
/*
* If this is a sibling, remove it from its group.
*/
- if (event->group_leader != event) {
+ if (leader != event) {
list_del_init(&event->sibling_list);
event->group_leader->nr_siblings--;
goto out;
@@ -2166,6 +2182,9 @@ static void perf_group_detach(struct perf_event *event)
*/
list_for_each_entry_safe(sibling, tmp, &event->sibling_list, sibling_list) {
+ if (sibling->event_caps & PERF_EV_CAP_SIBLING)
+ perf_remove_sibling_event(sibling);
+
sibling->group_leader = sibling;
list_del_init(&sibling->sibling_list);
@@ -2183,10 +2202,10 @@ static void perf_group_detach(struct perf_event *event)
}
out:
- perf_event__header_size(event->group_leader);
-
- for_each_sibling_event(tmp, event->group_leader)
+ for_each_sibling_event(tmp, leader)
perf_event__header_size(tmp);
+
+ perf_event__header_size(leader);
}
static bool is_orphaned_event(struct perf_event *event)
@@ -2979,6 +2998,7 @@ static void _perf_event_enable(struct perf_event *event)
raw_spin_lock_irq(&ctx->lock);
if (event->state >= PERF_EVENT_STATE_INACTIVE ||
event->state < PERF_EVENT_STATE_ERROR) {
+out:
raw_spin_unlock_irq(&ctx->lock);
return;
}
@@ -2990,8 +3010,16 @@ static void _perf_event_enable(struct perf_event *event)
* has gone back into error state, as distinct from the task having
* been scheduled away before the cross-call arrived.
*/
- if (event->state == PERF_EVENT_STATE_ERROR)
+ if (event->state == PERF_EVENT_STATE_ERROR) {
+ /*
+ * Detached SIBLING events cannot leave ERROR state.
+ */
+ if (event->event_caps & PERF_EV_CAP_SIBLING &&
+ event->group_leader == event)
+ goto out;
+
event->state = PERF_EVENT_STATE_OFF;
+ }
raw_spin_unlock_irq(&ctx->lock);
event_function_call(event, __perf_event_enable, NULL);