perf: Move swevent_htable::recursion into task_struct.

The swevent_htable::recursion counter is used to avoid creating an
swevent while an event is processed to avoid recursion. The counter is
per-CPU and preemption must be disabled to have a stable counter.
perf_pending_task() disables preemption to access the counter and then
signal. This is problematic on PREEMPT_RT because sending a signal uses
a spinlock_t which must not be acquired in atomic on PREEMPT_RT because
it becomes a sleeping lock.

The atomic context can be avoided by moving the counter into the
task_struct. There is a 4 byte hole between futex_state (usually always
on) and the following perf pointer (perf_event_ctxp). After the
recursion lost some weight it fits perfectly.

Move swevent_htable::recursion into task_struct.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Marco Elver <elver@google.com>
Link: https://lore.kernel.org/r/20240704170424.1466941-6-bigeasy@linutronix.de

Author: Sebastian Andrzej Siewior

include/linux/perf_event.h

@@ -970,12 +970,6 @@ struct perf_event_context {
 	local_t				nr_pending;
 };
 
-/*
- * Number of contexts where an event can trigger:
- *	task, softirq, hardirq, nmi.
- */
-#define PERF_NR_CONTEXTS	4
-
 struct perf_cpu_pmu_context {
 	struct perf_event_pmu_context	epc;
 	struct perf_event_pmu_context	*task_epc;

include/linux/sched.h

@@ -734,6 +734,12 @@ enum perf_event_task_context {
 	perf_nr_task_contexts,
 };
 
+/*
+ * Number of contexts where an event can trigger:
+ *      task, softirq, hardirq, nmi.
+ */
+#define PERF_NR_CONTEXTS	4
+
 struct wake_q_node {
 	struct wake_q_node *next;
 };
@@ -1256,6 +1262,7 @@ struct task_struct {
 	unsigned int			futex_state;
 #endif
 #ifdef CONFIG_PERF_EVENTS
+	u8				perf_recursion[PERF_NR_CONTEXTS];
 	struct perf_event_context	*perf_event_ctxp;
 	struct mutex			perf_event_mutex;
 	struct list_head		perf_event_list;

kernel/events/core.c

@@ -9763,11 +9763,7 @@ struct swevent_htable {
 	struct swevent_hlist		*swevent_hlist;
 	struct mutex			hlist_mutex;
 	int				hlist_refcount;
-
-	/* Recursion avoidance in each contexts */
-	u8				recursion[PERF_NR_CONTEXTS];
 };
-
 static DEFINE_PER_CPU(struct swevent_htable, swevent_htable);
 
 /*
@@ -9965,17 +9961,13 @@ DEFINE_PER_CPU(struct pt_regs, __perf_regs[4]);
 
 int perf_swevent_get_recursion_context(void)
 {
-	struct swevent_htable *swhash = this_cpu_ptr(&swevent_htable);
-
-	return get_recursion_context(swhash->recursion);
+	return get_recursion_context(current->perf_recursion);
 }
 EXPORT_SYMBOL_GPL(perf_swevent_get_recursion_context);
 
 void perf_swevent_put_recursion_context(int rctx)
 {
-	struct swevent_htable *swhash = this_cpu_ptr(&swevent_htable);
-
-	put_recursion_context(swhash->recursion, rctx);
+	put_recursion_context(current->perf_recursion, rctx);
 }
 
 void ___perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)
@@ -13642,6 +13634,7 @@ int perf_event_init_task(struct task_struct *child, u64 clone_flags)
 {
 	int ret;
 
+	memset(child->perf_recursion, 0, sizeof(child->perf_recursion));
 	child->perf_event_ctxp = NULL;
 	mutex_init(&child->perf_event_mutex);
 	INIT_LIST_HEAD(&child->perf_event_list);

kernel/events/internal.h

@@ -221,7 +221,7 @@ static inline int get_recursion_context(u8 *recursion)
 	return rctx;
 }
 
-static inline void put_recursion_context(u8 *recursion, int rctx)
+static inline void put_recursion_context(u8 *recursion, unsigned char rctx)
 {
 	barrier();
 	recursion[rctx]--;