Merge tag 'perf_urgent_for_v6.2_rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull perf fix from Borislav Petkov:

 - Lock the proper critical section when dealing with perf event context

* tag 'perf_urgent_for_v6.2_rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf: Fix perf_event_pmu_context serialization

include/linux/spinlock.h

@@ -476,6 +476,15 @@ extern int _atomic_dec_and_lock_irqsave(atomic_t *atomic, spinlock_t *lock,
 #define atomic_dec_and_lock_irqsave(atomic, lock, flags) \
 		__cond_lock(lock, _atomic_dec_and_lock_irqsave(atomic, lock, &(flags)))
 
+extern int _atomic_dec_and_raw_lock(atomic_t *atomic, raw_spinlock_t *lock);
+#define atomic_dec_and_raw_lock(atomic, lock) \
+		__cond_lock(lock, _atomic_dec_and_raw_lock(atomic, lock))
+
+extern int _atomic_dec_and_raw_lock_irqsave(atomic_t *atomic, raw_spinlock_t *lock,
+					unsigned long *flags);
+#define atomic_dec_and_raw_lock_irqsave(atomic, lock, flags) \
+		__cond_lock(lock, _atomic_dec_and_raw_lock_irqsave(atomic, lock, &(flags)))
+
 int __alloc_bucket_spinlocks(spinlock_t **locks, unsigned int *lock_mask,
 			     size_t max_size, unsigned int cpu_mult,
 			     gfp_t gfp, const char *name,

kernel/events/core.c

@@ -4813,19 +4813,17 @@ find_get_pmu_context(struct pmu *pmu, struct perf_event_context *ctx,
 
 		cpc = per_cpu_ptr(pmu->cpu_pmu_context, event->cpu);
 		epc = &cpc->epc;
-
+		raw_spin_lock_irq(&ctx->lock);
 		if (!epc->ctx) {
 			atomic_set(&epc->refcount, 1);
 			epc->embedded = 1;
-			raw_spin_lock_irq(&ctx->lock);
 			list_add(&epc->pmu_ctx_entry, &ctx->pmu_ctx_list);
 			epc->ctx = ctx;
-			raw_spin_unlock_irq(&ctx->lock);
 		} else {
 			WARN_ON_ONCE(epc->ctx != ctx);
 			atomic_inc(&epc->refcount);
 		}
-
+		raw_spin_unlock_irq(&ctx->lock);
 		return epc;
 	}
 
@@ -4896,13 +4894,8 @@ static void free_epc_rcu(struct rcu_head *head)
 
 static void put_pmu_ctx(struct perf_event_pmu_context *epc)
 {
-	unsigned long flags;
-
-	if (!atomic_dec_and_test(&epc->refcount))
-		return;
-
-	if (epc->ctx) {
 	struct perf_event_context *ctx = epc->ctx;
+	unsigned long flags;
 
 	/*
 	 * XXX
@@ -4912,17 +4905,19 @@ static void put_pmu_ctx(struct perf_event_pmu_context *epc)
 	 * can't because of the call-site in _free_event()/put_event()
 	 * which isn't always called under ctx->mutex.
 	 */
+	if (!atomic_dec_and_raw_lock_irqsave(&epc->refcount, &ctx->lock, flags))
+		return;
 
 	WARN_ON_ONCE(list_empty(&epc->pmu_ctx_entry));
-		raw_spin_lock_irqsave(&ctx->lock, flags);
+
 	list_del_init(&epc->pmu_ctx_entry);
 	epc->ctx = NULL;
-		raw_spin_unlock_irqrestore(&ctx->lock, flags);
-	}
 
 	WARN_ON_ONCE(!list_empty(&epc->pinned_active));
 	WARN_ON_ONCE(!list_empty(&epc->flexible_active));
 
+	raw_spin_unlock_irqrestore(&ctx->lock, flags);
+
 	if (epc->embedded)
 		return;
 

lib/dec_and_lock.c

@@ -49,3 +49,34 @@ int _atomic_dec_and_lock_irqsave(atomic_t *atomic, spinlock_t *lock,
 	return 0;
 }
 EXPORT_SYMBOL(_atomic_dec_and_lock_irqsave);
+
+int _atomic_dec_and_raw_lock(atomic_t *atomic, raw_spinlock_t *lock)
+{
+	/* Subtract 1 from counter unless that drops it to 0 (ie. it was 1) */
+	if (atomic_add_unless(atomic, -1, 1))
+		return 0;
+
+	/* Otherwise do it the slow way */
+	raw_spin_lock(lock);
+	if (atomic_dec_and_test(atomic))
+		return 1;
+	raw_spin_unlock(lock);
+	return 0;
+}
+EXPORT_SYMBOL(_atomic_dec_and_raw_lock);
+
+int _atomic_dec_and_raw_lock_irqsave(atomic_t *atomic, raw_spinlock_t *lock,
+				     unsigned long *flags)
+{
+	/* Subtract 1 from counter unless that drops it to 0 (ie. it was 1) */
+	if (atomic_add_unless(atomic, -1, 1))
+		return 0;
+
+	/* Otherwise do it the slow way */
+	raw_spin_lock_irqsave(lock, *flags);
+	if (atomic_dec_and_test(atomic))
+		return 1;
+	raw_spin_unlock_irqrestore(lock, *flags);
+	return 0;
+}
+EXPORT_SYMBOL(_atomic_dec_and_raw_lock_irqsave);