Merge tag 'perf_urgent_for_v6.1_rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull perf fixes from Borislav Petkov:

 - Fix raw data handling when perf events are used in bpf

 - Rework how SIGTRAPs get delivered to events to address a bunch of
   problems with it. Add a selftest for that too

* tag 'perf_urgent_for_v6.1_rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  bpf: Fix sample_flags for bpf_perf_event_output
  selftests/perf_events: Add a SIGTRAP stress test with disables
  perf: Fix missing SIGTRAPs

5 files changed, 163 insertions, 46 deletions

diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 853f64b6c8c2..0031f7b4d9ab 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -756,11 +756,14 @@ struct perf_event {
 	struct fasync_struct		*fasync;
 
 	/* delayed work for NMIs and such */
-	int				pending_wakeup;
-	int				pending_kill;
-	int				pending_disable;
+	unsigned int			pending_wakeup;
+	unsigned int			pending_kill;
+	unsigned int			pending_disable;
+	unsigned int			pending_sigtrap;
 	unsigned long			pending_addr;	/* SIGTRAP */
-	struct irq_work			pending;
+	struct irq_work			pending_irq;
+	struct callback_head		pending_task;
+	unsigned int			pending_work;
 
 	atomic_t			event_limit;
 
@@ -877,6 +880,14 @@ struct perf_event_context {
 #endif
 	void				*task_ctx_data; /* pmu specific data */
 	struct rcu_head			rcu_head;
+
+	/*
+	 * Sum (event->pending_sigtrap + event->pending_work)
+	 *
+	 * The SIGTRAP is targeted at ctx->task, as such it won't do changing
+	 * that until the signal is delivered.
+	 */
+	local_t				nr_pending;
 };
 
 /*
diff --git a/kernel/events/core.c b/kernel/events/core.c
index aefc1e08e015..01933db7629c 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -54,6 +54,7 @@
 #include <linux/highmem.h>
 #include <linux/pgtable.h>
 #include <linux/buildid.h>
+#include <linux/task_work.h>
 
 #include "internal.h"
 
@@ -2276,11 +2277,26 @@ event_sched_out(struct perf_event *event,
 	event->pmu->del(event, 0);
 	event->oncpu = -1;
 
-	if (READ_ONCE(event->pending_disable) >= 0) {
-		WRITE_ONCE(event->pending_disable, -1);
+	if (event->pending_disable) {
+		event->pending_disable = 0;
 		perf_cgroup_event_disable(event, ctx);
 		state = PERF_EVENT_STATE_OFF;
 	}
+
+	if (event->pending_sigtrap) {
+		bool dec = true;
+
+		event->pending_sigtrap = 0;
+		if (state != PERF_EVENT_STATE_OFF &&
+		    !event->pending_work) {
+			event->pending_work = 1;
+			dec = false;
+			task_work_add(current, &event->pending_task, TWA_RESUME);
+		}
+		if (dec)
+			local_dec(&event->ctx->nr_pending);
+	}
+
 	perf_event_set_state(event, state);
 
 	if (!is_software_event(event))
@@ -2432,7 +2448,7 @@ 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_event it's OK because event->ctx
+ * When called from perf_pending_irq 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.
  */
@@ -2471,9 +2487,8 @@ EXPORT_SYMBOL_GPL(perf_event_disable);
 
 void perf_event_disable_inatomic(struct perf_event *event)
 {
-	WRITE_ONCE(event->pending_disable, smp_processor_id());
-	/* can fail, see perf_pending_event_disable() */
-	irq_work_queue(&event->pending);
+	event->pending_disable = 1;
+	irq_work_queue(&event->pending_irq);
 }
 
 #define MAX_INTERRUPTS (~0ULL)
@@ -3428,11 +3443,23 @@ static void perf_event_context_sched_out(struct task_struct *task, int ctxn,
 		raw_spin_lock_nested(&next_ctx->lock, SINGLE_DEPTH_NESTING);
 		if (context_equiv(ctx, next_ctx)) {
 
+			perf_pmu_disable(pmu);
+
+			/* PMIs are disabled; ctx->nr_pending is stable. */
+			if (local_read(&ctx->nr_pending) ||
+			    local_read(&next_ctx->nr_pending)) {
+				/*
+				 * Must not swap out ctx when there's pending
+				 * events that rely on the ctx->task relation.
+				 */
+				raw_spin_unlock(&next_ctx->lock);
+				rcu_read_unlock();
+				goto inside_switch;
+			}
+
 			WRITE_ONCE(ctx->task, next);
 			WRITE_ONCE(next_ctx->task, task);
 
-			perf_pmu_disable(pmu);
-
 			if (cpuctx->sched_cb_usage && pmu->sched_task)
 				pmu->sched_task(ctx, false);
 
@@ -3473,6 +3500,7 @@ unlock:
 		raw_spin_lock(&ctx->lock);
 		perf_pmu_disable(pmu);
 
+inside_switch:
 		if (cpuctx->sched_cb_usage && pmu->sched_task)
 			pmu->sched_task(ctx, false);
 		task_ctx_sched_out(cpuctx, ctx, EVENT_ALL);
@@ -4939,7 +4967,7 @@ static void perf_addr_filters_splice(struct perf_event *event,
 
 static void _free_event(struct perf_event *event)
 {
-	irq_work_sync(&event->pending);
+	irq_work_sync(&event->pending_irq);
 
 	unaccount_event(event);
 
@@ -6439,7 +6467,8 @@ static void perf_sigtrap(struct perf_event *event)
 		return;
 
 	/*
-	 * perf_pending_event() can race with the task exiting.
+	 * Both perf_pending_task() and perf_pending_irq() can race with the
+	 * task exiting.
 	 */
 	if (current->flags & PF_EXITING)
 		return;
@@ -6448,23 +6477,33 @@ static void perf_sigtrap(struct perf_event *event)
 		      event->attr.type, event->attr.sig_data);
 }
 
-static void perf_pending_event_disable(struct perf_event *event)
+/*
+ * Deliver the pending work in-event-context or follow the context.
+ */
+static void __perf_pending_irq(struct perf_event *event)
 {
-	int cpu = READ_ONCE(event->pending_disable);
+	int cpu = READ_ONCE(event->oncpu);
 
+	/*
+	 * If the event isn't running; we done. event_sched_out() will have
+	 * taken care of things.
+	 */
 	if (cpu < 0)
 		return;
 
+	/*
+	 * Yay, we hit home and are in the context of the event.
+	 */
 	if (cpu == smp_processor_id()) {
-		WRITE_ONCE(event->pending_disable, -1);
-
-		if (event->attr.sigtrap) {
+		if (event->pending_sigtrap) {
+			event->pending_sigtrap = 0;
 			perf_sigtrap(event);
-			atomic_set_release(&event->event_limit, 1); /* rearm event */
-			return;
+			local_dec(&event->ctx->nr_pending);
+		}
+		if (event->pending_disable) {
+			event->pending_disable = 0;
+			perf_event_disable_local(event);
 		}
-
-		perf_event_disable_local(event);
 		return;
 	}
 
@@ -6484,35 +6523,62 @@ static void perf_pending_event_disable(struct perf_event *event)
 	 *				  irq_work_queue(); // FAILS
 	 *
 	 *  irq_work_run()
-	 *    perf_pending_event()
+	 *    perf_pending_irq()
 	 *
 	 * But the event runs on CPU-B and wants disabling there.
 	 */
-	irq_work_queue_on(&event->pending, cpu);
+	irq_work_queue_on(&event->pending_irq, cpu);
 }
 
-static void perf_pending_event(struct irq_work *entry)
+static void perf_pending_irq(struct irq_work *entry)
 {
-	struct perf_event *event = container_of(entry, struct perf_event, pending);
+	struct perf_event *event = container_of(entry, struct perf_event, pending_irq);
 	int rctx;
 
-	rctx = perf_swevent_get_recursion_context();
 	/*
 	 * 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_event_disable(event);
-
+	/*
+	 * The wakeup isn't bound to the context of the event -- it can happen
+	 * irrespective of where the event is.
+	 */
 	if (event->pending_wakeup) {
 		event->pending_wakeup = 0;
 		perf_event_wakeup(event);
 	}
 
+	__perf_pending_irq(event);
+
 	if (rctx >= 0)
 		perf_swevent_put_recursion_context(rctx);
 }
 
+static void perf_pending_task(struct callback_head *head)
+{
+	struct perf_event *event = container_of(head, struct perf_event, pending_task);
+	int rctx;
+
+	/*
+	 * If we 'fail' here, that's OK, it means recursion is already disabled
+	 * and we won't recurse 'further'.
+	 */
+	preempt_disable_notrace();
+	rctx = perf_swevent_get_recursion_context();
+
+	if (event->pending_work) {
+		event->pending_work = 0;
+		perf_sigtrap(event);
+		local_dec(&event->ctx->nr_pending);
+	}
+
+	if (rctx >= 0)
+		perf_swevent_put_recursion_context(rctx);
+	preempt_enable_notrace();
+}
+
 #ifdef CONFIG_GUEST_PERF_EVENTS
 struct perf_guest_info_callbacks __rcu *perf_guest_cbs;
 
@@ -9212,8 +9278,8 @@ int perf_event_account_interrupt(struct perf_event *event)
  */
 
 static int __perf_event_overflow(struct perf_event *event,
-				   int throttle, struct perf_sample_data *data,
-				   struct pt_regs *regs)
+				 int throttle, struct perf_sample_data *data,
+				 struct pt_regs *regs)
 {
 	int events = atomic_read(&event->event_limit);
 	int ret = 0;
@@ -9236,24 +9302,36 @@ static int __perf_event_overflow(struct perf_event *event,
 	if (events && atomic_dec_and_test(&event->event_limit)) {
 		ret = 1;
 		event->pending_kill = POLL_HUP;
-		event->pending_addr = data->addr;
-
 		perf_event_disable_inatomic(event);
 	}
 
+	if (event->attr.sigtrap) {
+		/*
+		 * Should not be able to return to user space without processing
+		 * pending_sigtrap (kernel events can overflow multiple times).
+		 */
+		WARN_ON_ONCE(event->pending_sigtrap && event->attr.exclude_kernel);
+		if (!event->pending_sigtrap) {
+			event->pending_sigtrap = 1;
+			local_inc(&event->ctx->nr_pending);
+		}
+		event->pending_addr = data->addr;
+		irq_work_queue(&event->pending_irq);
+	}
+
 	READ_ONCE(event->overflow_handler)(event, data, regs);
 
 	if (*perf_event_fasync(event) && event->pending_kill) {
 		event->pending_wakeup = 1;
-		irq_work_queue(&event->pending);
+		irq_work_queue(&event->pending_irq);
 	}
 
 	return ret;
 }
 
 int perf_event_overflow(struct perf_event *event,
-			  struct perf_sample_data *data,
-			  struct pt_regs *regs)
+			struct perf_sample_data *data,
+			struct pt_regs *regs)
 {
 	return __perf_event_overflow(event, 1, data, regs);
 }
@@ -11570,8 +11648,8 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu,
 
 
 	init_waitqueue_head(&event->waitq);
-	event->pending_disable = -1;
-	init_irq_work(&event->pending, perf_pending_event);
+	init_irq_work(&event->pending_irq, perf_pending_irq);
+	init_task_work(&event->pending_task, perf_pending_task);
 
 	mutex_init(&event->mmap_mutex);
 	raw_spin_lock_init(&event->addr_filters.lock);
@@ -11593,9 +11671,6 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu,
 	if (parent_event)
 		event->event_caps = parent_event->event_caps;
 
-	if (event->attr.sigtrap)
-		atomic_set(&event->event_limit, 1);
-
 	if (task) {
 		event->attach_state = PERF_ATTACH_TASK;
 		/*
diff --git a/kernel/events/ring_buffer.c b/kernel/events/ring_buffer.c
index 726132039c38..273a0fe7910a 100644
--- a/kernel/events/ring_buffer.c
+++ b/kernel/events/ring_buffer.c
@@ -22,7 +22,7 @@ static void perf_output_wakeup(struct perf_output_handle *handle)
 	atomic_set(&handle->rb->poll, EPOLLIN);
 
 	handle->event->pending_wakeup = 1;
-	irq_work_queue(&handle->event->pending);
+	irq_work_queue(&handle->event->pending_irq);
 }
 
 /*
diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
index 49fb9ec8366d..1ed08967fb97 100644
--- a/kernel/trace/bpf_trace.c
+++ b/kernel/trace/bpf_trace.c
@@ -687,6 +687,7 @@ BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
 
 	perf_sample_data_init(sd, 0, 0);
 	sd->raw = &raw;
+	sd->sample_flags |= PERF_SAMPLE_RAW;
 
 	err = __bpf_perf_event_output(regs, map, flags, sd);
 
@@ -745,6 +746,7 @@ u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
 	perf_fetch_caller_regs(regs);
 	perf_sample_data_init(sd, 0, 0);
 	sd->raw = &raw;
+	sd->sample_flags |= PERF_SAMPLE_RAW;
 
 	ret = __bpf_perf_event_output(regs, map, flags, sd);
 out:
diff --git a/tools/testing/selftests/perf_events/sigtrap_threads.c b/tools/testing/selftests/perf_events/sigtrap_threads.c
index 6d849dc2bee0..d1d8483ac628 100644
--- a/tools/testing/selftests/perf_events/sigtrap_threads.c
+++ b/tools/testing/selftests/perf_events/sigtrap_threads.c
@@ -62,6 +62,8 @@ static struct perf_event_attr make_event_attr(bool enabled, volatile void *addr,
 		.remove_on_exec = 1, /* Required by sigtrap. */
 		.sigtrap	= 1, /* Request synchronous SIGTRAP on event. */
 		.sig_data	= TEST_SIG_DATA(addr, id),
+		.exclude_kernel = 1, /* To allow */
+		.exclude_hv     = 1, /* running as !root */
 	};
 	return attr;
 }
@@ -93,9 +95,13 @@ static void *test_thread(void *arg)
 
 	__atomic_fetch_add(&ctx.tids_want_signal, tid, __ATOMIC_RELAXED);
 	iter = ctx.iterate_on; /* read */
-	for (i = 0; i < iter - 1; i++) {
-		__atomic_fetch_add(&ctx.tids_want_signal, tid, __ATOMIC_RELAXED);
-		ctx.iterate_on = iter; /* idempotent write */
+	if (iter >= 0) {
+		for (i = 0; i < iter - 1; i++) {
+			__atomic_fetch_add(&ctx.tids_want_signal, tid, __ATOMIC_RELAXED);
+			ctx.iterate_on = iter; /* idempotent write */
+		}
+	} else {
+		while (ctx.iterate_on);
 	}
 
 	return NULL;
@@ -208,4 +214,27 @@ TEST_F(sigtrap_threads, signal_stress)
 	EXPECT_EQ(ctx.first_siginfo.si_perf_data, TEST_SIG_DATA(&ctx.iterate_on, 0));
 }
 
+TEST_F(sigtrap_threads, signal_stress_with_disable)
+{
+	const int target_count = NUM_THREADS * 3000;
+	int i;
+
+	ctx.iterate_on = -1;
+
+	EXPECT_EQ(ioctl(self->fd, PERF_EVENT_IOC_ENABLE, 0), 0);
+	pthread_barrier_wait(&self->barrier);
+	while (__atomic_load_n(&ctx.signal_count, __ATOMIC_RELAXED) < target_count) {
+		EXPECT_EQ(ioctl(self->fd, PERF_EVENT_IOC_DISABLE, 0), 0);
+		EXPECT_EQ(ioctl(self->fd, PERF_EVENT_IOC_ENABLE, 0), 0);
+	}
+	ctx.iterate_on = 0;
+	for (i = 0; i < NUM_THREADS; i++)
+		ASSERT_EQ(pthread_join(self->threads[i], NULL), 0);
+	EXPECT_EQ(ioctl(self->fd, PERF_EVENT_IOC_DISABLE, 0), 0);
+
+	EXPECT_EQ(ctx.first_siginfo.si_addr, &ctx.iterate_on);
+	EXPECT_EQ(ctx.first_siginfo.si_perf_type, PERF_TYPE_BREAKPOINT);
+	EXPECT_EQ(ctx.first_siginfo.si_perf_data, TEST_SIG_DATA(&ctx.iterate_on, 0));
+}
+
 TEST_HARNESS_MAIN