perf: Reduce stack usage of perf_output_begin()

__perf_output_begin() has an on-stack struct perf_sample_data in the
unlikely case it needs to generate a LOST record. However, every call
to perf_output_begin() must already have a perf_sample_data on-stack.

Reported-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20201030151954.985416146@infradead.org

arch/powerpc/perf/imc-pmu.c

@@ -1336,7 +1336,7 @@ static void dump_trace_imc_data(struct perf_event *event)
 			/* If this is a valid record, create the sample */
 			struct perf_output_handle handle;
 
-			if (perf_output_begin(&handle, event, header.size))
+			if (perf_output_begin(&handle, &data, event, header.size))
 				return;
 
 			perf_output_sample(&handle, &header, &data, event);

arch/s390/kernel/perf_cpum_sf.c

@@ -672,7 +672,7 @@ static void cpumsf_output_event_pid(struct perf_event *event,
 	rcu_read_lock();
 
 	perf_prepare_sample(&header, data, event, regs);
-	if (perf_output_begin(&handle, event, header.size))
+	if (perf_output_begin(&handle, data, event, header.size))
 		goto out;
 
 	/* Update the process ID (see also kernel/events/core.c) */

arch/x86/events/intel/ds.c

@@ -642,8 +642,8 @@ int intel_pmu_drain_bts_buffer(void)
 	rcu_read_lock();
 	perf_prepare_sample(&header, &data, event, &regs);
 
-	if (perf_output_begin(&handle, event, header.size *
+	if (perf_output_begin(&handle, &data, event,
-			      (top - base - skip)))
+			      header.size * (top - base - skip)))
 		goto unlock;
 
 	for (at = base; at < top; at++) {

include/linux/perf_event.h

@@ -1400,11 +1400,14 @@ perf_event_addr_filters(struct perf_event *event)
 extern void perf_event_addr_filters_sync(struct perf_event *event);
 
 extern int perf_output_begin(struct perf_output_handle *handle,
+			     struct perf_sample_data *data,
 			     struct perf_event *event, unsigned int size);
 extern int perf_output_begin_forward(struct perf_output_handle *handle,
-				    struct perf_event *event,
+				     struct perf_sample_data *data,
-				    unsigned int size);
+				     struct perf_event *event,
+				     unsigned int size);
 extern int perf_output_begin_backward(struct perf_output_handle *handle,
+				      struct perf_sample_data *data,
 				      struct perf_event *event,
 				      unsigned int size);
 

kernel/events/core.c

@@ -7186,6 +7186,7 @@ __perf_event_output(struct perf_event *event,
 		    struct perf_sample_data *data,
 		    struct pt_regs *regs,
 		    int (*output_begin)(struct perf_output_handle *,
+					struct perf_sample_data *,
 					struct perf_event *,
 					unsigned int))
 {
@@ -7198,7 +7199,7 @@ __perf_event_output(struct perf_event *event,
 
 	perf_prepare_sample(&header, data, event, regs);
 
-	err = output_begin(&handle, event, header.size);
+	err = output_begin(&handle, data, event, header.size);
 	if (err)
 		goto exit;
 
@@ -7264,7 +7265,7 @@ perf_event_read_event(struct perf_event *event,
 	int ret;
 
 	perf_event_header__init_id(&read_event.header, &sample, event);
-	ret = perf_output_begin(&handle, event, read_event.header.size);
+	ret = perf_output_begin(&handle, &sample, event, read_event.header.size);
 	if (ret)
 		return;
 
@@ -7533,7 +7534,7 @@ static void perf_event_task_output(struct perf_event *event,
 
 	perf_event_header__init_id(&task_event->event_id.header, &sample, event);
 
-	ret = perf_output_begin(&handle, event,
+	ret = perf_output_begin(&handle, &sample, event,
 				task_event->event_id.header.size);
 	if (ret)
 		goto out;
@@ -7636,7 +7637,7 @@ static void perf_event_comm_output(struct perf_event *event,
 		return;
 
 	perf_event_header__init_id(&comm_event->event_id.header, &sample, event);
-	ret = perf_output_begin(&handle, event,
+	ret = perf_output_begin(&handle, &sample, event,
 				comm_event->event_id.header.size);
 
 	if (ret)
@@ -7736,7 +7737,7 @@ static void perf_event_namespaces_output(struct perf_event *event,
 
 	perf_event_header__init_id(&namespaces_event->event_id.header,
 				   &sample, event);
-	ret = perf_output_begin(&handle, event,
+	ret = perf_output_begin(&handle, &sample, event,
 				namespaces_event->event_id.header.size);
 	if (ret)
 		goto out;
@@ -7863,7 +7864,7 @@ static void perf_event_cgroup_output(struct perf_event *event, void *data)
 
 	perf_event_header__init_id(&cgroup_event->event_id.header,
 				   &sample, event);
-	ret = perf_output_begin(&handle, event,
+	ret = perf_output_begin(&handle, &sample, event,
 				cgroup_event->event_id.header.size);
 	if (ret)
 		goto out;
@@ -7989,7 +7990,7 @@ static void perf_event_mmap_output(struct perf_event *event,
 	}
 
 	perf_event_header__init_id(&mmap_event->event_id.header, &sample, event);
-	ret = perf_output_begin(&handle, event,
+	ret = perf_output_begin(&handle, &sample, event,
 				mmap_event->event_id.header.size);
 	if (ret)
 		goto out;
@@ -8299,7 +8300,7 @@ void perf_event_aux_event(struct perf_event *event, unsigned long head,
 	int ret;
 
 	perf_event_header__init_id(&rec.header, &sample, event);
-	ret = perf_output_begin(&handle, event, rec.header.size);
+	ret = perf_output_begin(&handle, &sample, event, rec.header.size);
 
 	if (ret)
 		return;
@@ -8333,7 +8334,7 @@ void perf_log_lost_samples(struct perf_event *event, u64 lost)
 
 	perf_event_header__init_id(&lost_samples_event.header, &sample, event);
 
-	ret = perf_output_begin(&handle, event,
+	ret = perf_output_begin(&handle, &sample, event,
 				lost_samples_event.header.size);
 	if (ret)
 		return;
@@ -8388,7 +8389,7 @@ static void perf_event_switch_output(struct perf_event *event, void *data)
 
 	perf_event_header__init_id(&se->event_id.header, &sample, event);
 
-	ret = perf_output_begin(&handle, event, se->event_id.header.size);
+	ret = perf_output_begin(&handle, &sample, event, se->event_id.header.size);
 	if (ret)
 		return;
 
@@ -8463,7 +8464,7 @@ static void perf_log_throttle(struct perf_event *event, int enable)
 
 	perf_event_header__init_id(&throttle_event.header, &sample, event);
 
-	ret = perf_output_begin(&handle, event,
+	ret = perf_output_begin(&handle, &sample, event,
 				throttle_event.header.size);
 	if (ret)
 		return;
@@ -8506,7 +8507,7 @@ static void perf_event_ksymbol_output(struct perf_event *event, void *data)
 
 	perf_event_header__init_id(&ksymbol_event->event_id.header,
 				   &sample, event);
-	ret = perf_output_begin(&handle, event,
+	ret = perf_output_begin(&handle, &sample, event,
 				ksymbol_event->event_id.header.size);
 	if (ret)
 		return;
@@ -8596,7 +8597,7 @@ static void perf_event_bpf_output(struct perf_event *event, void *data)
 
 	perf_event_header__init_id(&bpf_event->event_id.header,
 				   &sample, event);
-	ret = perf_output_begin(&handle, event,
+	ret = perf_output_begin(&handle, data, event,
 				bpf_event->event_id.header.size);
 	if (ret)
 		return;
@@ -8705,7 +8706,8 @@ static void perf_event_text_poke_output(struct perf_event *event, void *data)
 
 	perf_event_header__init_id(&text_poke_event->event_id.header, &sample, event);
 
-	ret = perf_output_begin(&handle, event, text_poke_event->event_id.header.size);
+	ret = perf_output_begin(&handle, &sample, event,
+				text_poke_event->event_id.header.size);
 	if (ret)
 		return;
 
@@ -8786,7 +8788,7 @@ static void perf_log_itrace_start(struct perf_event *event)
 	rec.tid	= perf_event_tid(event, current);
 
 	perf_event_header__init_id(&rec.header, &sample, event);
-	ret = perf_output_begin(&handle, event, rec.header.size);
+	ret = perf_output_begin(&handle, &sample, event, rec.header.size);
 
 	if (ret)
 		return;

kernel/events/ring_buffer.c

@@ -147,6 +147,7 @@ ring_buffer_has_space(unsigned long head, unsigned long tail,
 
 static __always_inline int
 __perf_output_begin(struct perf_output_handle *handle,
+		    struct perf_sample_data *data,
 		    struct perf_event *event, unsigned int size,
 		    bool backward)
 {
@@ -237,18 +238,16 @@ __perf_output_begin(struct perf_output_handle *handle,
 	handle->size = (1UL << page_shift) - offset;
 
 	if (unlikely(have_lost)) {
-		struct perf_sample_data sample_data;
-
 		lost_event.header.size = sizeof(lost_event);
 		lost_event.header.type = PERF_RECORD_LOST;
 		lost_event.header.misc = 0;
 		lost_event.id          = event->id;
 		lost_event.lost        = local_xchg(&rb->lost, 0);
 
-		perf_event_header__init_id(&lost_event.header,
+		/* XXX mostly redundant; @data is already fully initializes */
-					   &sample_data, event);
+		perf_event_header__init_id(&lost_event.header, data, event);
 		perf_output_put(handle, lost_event);
-		perf_event__output_id_sample(event, handle, &sample_data);
+		perf_event__output_id_sample(event, handle, data);
 	}
 
 	return 0;
@@ -263,22 +262,25 @@ out:
 }
 
 int perf_output_begin_forward(struct perf_output_handle *handle,
-			     struct perf_event *event, unsigned int size)
+			      struct perf_sample_data *data,
+			      struct perf_event *event, unsigned int size)
 {
-	return __perf_output_begin(handle, event, size, false);
+	return __perf_output_begin(handle, data, event, size, false);
 }
 
 int perf_output_begin_backward(struct perf_output_handle *handle,
+			       struct perf_sample_data *data,
 			       struct perf_event *event, unsigned int size)
 {
-	return __perf_output_begin(handle, event, size, true);
+	return __perf_output_begin(handle, data, event, size, true);
 }
 
 int perf_output_begin(struct perf_output_handle *handle,
+		      struct perf_sample_data *data,
 		      struct perf_event *event, unsigned int size)
 {
 
-	return __perf_output_begin(handle, event, size,
+	return __perf_output_begin(handle, data, event, size,
 				   unlikely(is_write_backward(event)));
 }