summaryrefslogtreecommitdiff
path: root/include
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2025-06-22 20:11:45 +0300
committerLinus Torvalds <torvalds@linux-foundation.org>2025-06-22 20:11:45 +0300
commit17ef32ae66b1afc9fa6dbea40eb18a13edba9c31 (patch)
tree3a82bfa0b1167dfe3e57c9b09d037a3b32b1433e /include
parentaff2a7e23f23738ca3cd62e4ce5be2d62a3d52ad (diff)
parentb0823d5fbacb1c551d793cbfe7af24e0d1fa45ed (diff)
downloadlinux-17ef32ae66b1afc9fa6dbea40eb18a13edba9c31.tar.xz
Merge tag 'perf_urgent_for_v6.16_rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull perf fixes from Borislav Petkov: - Avoid a crash on a heterogeneous machine where not all cores support the same hw events features - Avoid a deadlock when throttling events - Document the perf event states more - Make sure a number of perf paths switching off or rescheduling events call perf_cgroup_event_disable() - Make sure perf does task sampling before its userspace mapping is torn down, and not after * tag 'perf_urgent_for_v6.16_rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: perf/x86/intel: Fix crash in icl_update_topdown_event() perf: Fix the throttle error of some clock events perf: Add comment to enum perf_event_state perf/core: Fix WARN in perf_cgroup_switch() perf: Fix dangling cgroup pointer in cpuctx perf: Fix cgroup state vs ERROR perf: Fix sample vs do_exit()
Diffstat (limited to 'include')
-rw-r--r--include/linux/perf_event.h42
1 files changed, 40 insertions, 2 deletions
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 52dc7cfab0e0..ec9d96025683 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -635,8 +635,46 @@ struct perf_addr_filter_range {
unsigned long size;
};
-/**
- * enum perf_event_state - the states of an event:
+/*
+ * The normal states are:
+ *
+ * ACTIVE --.
+ * ^ |
+ * | |
+ * sched_{in,out}() |
+ * | |
+ * v |
+ * ,---> INACTIVE --+ <-.
+ * | | |
+ * | {dis,en}able()
+ * sched_in() | |
+ * | OFF <--' --+
+ * | |
+ * `---> ERROR ------'
+ *
+ * That is:
+ *
+ * sched_in: INACTIVE -> {ACTIVE,ERROR}
+ * sched_out: ACTIVE -> INACTIVE
+ * disable: {ACTIVE,INACTIVE} -> OFF
+ * enable: {OFF,ERROR} -> INACTIVE
+ *
+ * Where {OFF,ERROR} are disabled states.
+ *
+ * Then we have the {EXIT,REVOKED,DEAD} states which are various shades of
+ * defunct events:
+ *
+ * - EXIT means task that the even was assigned to died, but child events
+ * still live, and further children can still be created. But the event
+ * itself will never be active again. It can only transition to
+ * {REVOKED,DEAD};
+ *
+ * - REVOKED means the PMU the event was associated with is gone; all
+ * functionality is stopped but the event is still alive. Can only
+ * transition to DEAD;
+ *
+ * - DEAD event really is DYING tearing down state and freeing bits.
+ *
*/
enum perf_event_state {
PERF_EVENT_STATE_DEAD = -5,