perf/core: Save raw sample data conditionally based on sample type

Currently, space for raw sample data is always allocated within sample
records for both BPF output and tracepoint events. This leads to unused
space in sample records when raw sample data is not requested.

This patch enforces checking sample type of an event in
perf_sample_save_raw_data(). So raw sample data will only be saved if
explicitly requested, reducing overhead when it is not needed.

Fixes: 0a9081cf0a11 ("perf/core: Add perf_sample_save_raw_data() helper")
Signed-off-by: Yabin Cui <yabinc@google.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20240515193610.2350456-2-yabinc@google.com

7 files changed, 34 insertions, 26 deletions

diff --git a/arch/s390/kernel/perf_cpum_cf.c b/arch/s390/kernel/perf_cpum_cf.c
index e2e0aa463fbd..c3075e4a8efc 100644
--- a/arch/s390/kernel/perf_cpum_cf.c
+++ b/arch/s390/kernel/perf_cpum_cf.c
@@ -981,7 +981,7 @@ static int cfdiag_push_sample(struct perf_event *event,
 	if (event->attr.sample_type & PERF_SAMPLE_RAW) {
 		raw.frag.size = cpuhw->usedss;
 		raw.frag.data = cpuhw->stop;
-		perf_sample_save_raw_data(&data, &raw);
+		perf_sample_save_raw_data(&data, event, &raw);
 	}
 
 	overflow = perf_event_overflow(event, &data, &regs);
diff --git a/arch/s390/kernel/perf_pai_crypto.c b/arch/s390/kernel/perf_pai_crypto.c
index fa7325454266..10725f5a6f0f 100644
--- a/arch/s390/kernel/perf_pai_crypto.c
+++ b/arch/s390/kernel/perf_pai_crypto.c
@@ -478,7 +478,7 @@ static int paicrypt_push_sample(size_t rawsize, struct paicrypt_map *cpump,
 	if (event->attr.sample_type & PERF_SAMPLE_RAW) {
 		raw.frag.size = rawsize;
 		raw.frag.data = cpump->save;
-		perf_sample_save_raw_data(&data, &raw);
+		perf_sample_save_raw_data(&data, event, &raw);
 	}
 
 	overflow = perf_event_overflow(event, &data, &regs);
diff --git a/arch/s390/kernel/perf_pai_ext.c b/arch/s390/kernel/perf_pai_ext.c
index 7f462bef1fc0..a8f0bad99cf0 100644
--- a/arch/s390/kernel/perf_pai_ext.c
+++ b/arch/s390/kernel/perf_pai_ext.c
@@ -503,7 +503,7 @@ static int paiext_push_sample(size_t rawsize, struct paiext_map *cpump,
 	if (event->attr.sample_type & PERF_SAMPLE_RAW) {
 		raw.frag.size = rawsize;
 		raw.frag.data = cpump->save;
-		perf_sample_save_raw_data(&data, &raw);
+		perf_sample_save_raw_data(&data, event, &raw);
 	}
 
 	overflow = perf_event_overflow(event, &data, &regs);
diff --git a/arch/x86/events/amd/ibs.c b/arch/x86/events/amd/ibs.c
index e91970b01d62..c3a2f6f57770 100644
--- a/arch/x86/events/amd/ibs.c
+++ b/arch/x86/events/amd/ibs.c
@@ -1118,7 +1118,7 @@ fail:
 				.data = ibs_data.data,
 			},
 		};
-		perf_sample_save_raw_data(&data, &raw);
+		perf_sample_save_raw_data(&data, event, &raw);
 	}
 
 	if (perf_ibs == &perf_ibs_op)
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index cb99ec8c9e96..f7c0a3f2f502 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -1287,12 +1287,18 @@ static inline void perf_sample_save_callchain(struct perf_sample_data *data,
 }
 
 static inline void perf_sample_save_raw_data(struct perf_sample_data *data,
+					     struct perf_event *event,
 					     struct perf_raw_record *raw)
 {
 	struct perf_raw_frag *frag = &raw->frag;
 	u32 sum = 0;
 	int size;
 
+	if (!(event->attr.sample_type & PERF_SAMPLE_RAW))
+		return;
+	if (WARN_ON_ONCE(data->sample_flags & PERF_SAMPLE_RAW))
+		return;
+
 	do {
 		sum += frag->size;
 		if (perf_raw_frag_last(frag))
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 1869164a4e99..3670b91f182f 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -10451,9 +10451,9 @@ static struct pmu perf_tracepoint = {
 };
 
 static int perf_tp_filter_match(struct perf_event *event,
-				struct perf_sample_data *data)
+				struct perf_raw_record *raw)
 {
-	void *record = data->raw->frag.data;
+	void *record = raw->frag.data;
 
 	/* only top level events have filters set */
 	if (event->parent)
@@ -10465,7 +10465,7 @@ static int perf_tp_filter_match(struct perf_event *event,
 }
 
 static int perf_tp_event_match(struct perf_event *event,
-				struct perf_sample_data *data,
+				struct perf_raw_record *raw,
 				struct pt_regs *regs)
 {
 	if (event->hw.state & PERF_HES_STOPPED)
@@ -10476,7 +10476,7 @@ static int perf_tp_event_match(struct perf_event *event,
 	if (event->attr.exclude_kernel && !user_mode(regs))
 		return 0;
 
-	if (!perf_tp_filter_match(event, data))
+	if (!perf_tp_filter_match(event, raw))
 		return 0;
 
 	return 1;
@@ -10502,6 +10502,7 @@ EXPORT_SYMBOL_GPL(perf_trace_run_bpf_submit);
 static void __perf_tp_event_target_task(u64 count, void *record,
 					struct pt_regs *regs,
 					struct perf_sample_data *data,
+					struct perf_raw_record *raw,
 					struct perf_event *event)
 {
 	struct trace_entry *entry = record;
@@ -10511,13 +10512,17 @@ static void __perf_tp_event_target_task(u64 count, void *record,
 	/* Cannot deliver synchronous signal to other task. */
 	if (event->attr.sigtrap)
 		return;
-	if (perf_tp_event_match(event, data, regs))
+	if (perf_tp_event_match(event, raw, regs)) {
+		perf_sample_data_init(data, 0, 0);
+		perf_sample_save_raw_data(data, event, raw);
 		perf_swevent_event(event, count, data, regs);
+	}
 }
 
 static void perf_tp_event_target_task(u64 count, void *record,
 				      struct pt_regs *regs,
 				      struct perf_sample_data *data,
+				      struct perf_raw_record *raw,
 				      struct perf_event_context *ctx)
 {
 	unsigned int cpu = smp_processor_id();
@@ -10525,15 +10530,15 @@ static void perf_tp_event_target_task(u64 count, void *record,
 	struct perf_event *event, *sibling;
 
 	perf_event_groups_for_cpu_pmu(event, &ctx->pinned_groups, cpu, pmu) {
-		__perf_tp_event_target_task(count, record, regs, data, event);
+		__perf_tp_event_target_task(count, record, regs, data, raw, event);
 		for_each_sibling_event(sibling, event)
-			__perf_tp_event_target_task(count, record, regs, data, sibling);
+			__perf_tp_event_target_task(count, record, regs, data, raw, sibling);
 	}
 
 	perf_event_groups_for_cpu_pmu(event, &ctx->flexible_groups, cpu, pmu) {
-		__perf_tp_event_target_task(count, record, regs, data, event);
+		__perf_tp_event_target_task(count, record, regs, data, raw, event);
 		for_each_sibling_event(sibling, event)
-			__perf_tp_event_target_task(count, record, regs, data, sibling);
+			__perf_tp_event_target_task(count, record, regs, data, raw, sibling);
 	}
 }
 
@@ -10551,15 +10556,10 @@ void perf_tp_event(u16 event_type, u64 count, void *record, int entry_size,
 		},
 	};
 
-	perf_sample_data_init(&data, 0, 0);
-	perf_sample_save_raw_data(&data, &raw);
-
 	perf_trace_buf_update(record, event_type);
 
 	hlist_for_each_entry_rcu(event, head, hlist_entry) {
-		if (perf_tp_event_match(event, &data, regs)) {
-			perf_swevent_event(event, count, &data, regs);
-
+		if (perf_tp_event_match(event, &raw, regs)) {
 			/*
 			 * Here use the same on-stack perf_sample_data,
 			 * some members in data are event-specific and
@@ -10569,7 +10569,8 @@ void perf_tp_event(u16 event_type, u64 count, void *record, int entry_size,
 			 * because data->sample_flags is set.
 			 */
 			perf_sample_data_init(&data, 0, 0);
-			perf_sample_save_raw_data(&data, &raw);
+			perf_sample_save_raw_data(&data, event, &raw);
+			perf_swevent_event(event, count, &data, regs);
 		}
 	}
 
@@ -10586,7 +10587,7 @@ void perf_tp_event(u16 event_type, u64 count, void *record, int entry_size,
 			goto unlock;
 
 		raw_spin_lock(&ctx->lock);
-		perf_tp_event_target_task(count, record, regs, &data, ctx);
+		perf_tp_event_target_task(count, record, regs, &data, &raw, ctx);
 		raw_spin_unlock(&ctx->lock);
 unlock:
 		rcu_read_unlock();
diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
index fdab7ecd8dfa..162bacf8aa5d 100644
--- a/kernel/trace/bpf_trace.c
+++ b/kernel/trace/bpf_trace.c
@@ -619,7 +619,8 @@ static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
 
 static __always_inline u64
 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
-			u64 flags, struct perf_sample_data *sd)
+			u64 flags, struct perf_raw_record *raw,
+			struct perf_sample_data *sd)
 {
 	struct bpf_array *array = container_of(map, struct bpf_array, map);
 	unsigned int cpu = smp_processor_id();
@@ -644,6 +645,8 @@ __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
 	if (unlikely(event->oncpu != cpu))
 		return -EOPNOTSUPP;
 
+	perf_sample_save_raw_data(sd, event, raw);
+
 	return perf_event_output(event, sd, regs);
 }
 
@@ -687,9 +690,8 @@ BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
 	}
 
 	perf_sample_data_init(sd, 0, 0);
-	perf_sample_save_raw_data(sd, &raw);
 
-	err = __bpf_perf_event_output(regs, map, flags, sd);
+	err = __bpf_perf_event_output(regs, map, flags, &raw, sd);
 out:
 	this_cpu_dec(bpf_trace_nest_level);
 	preempt_enable();
@@ -748,9 +750,8 @@ 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);
-	perf_sample_save_raw_data(sd, &raw);
 
-	ret = __bpf_perf_event_output(regs, map, flags, sd);
+	ret = __bpf_perf_event_output(regs, map, flags, &raw, sd);
 out:
 	this_cpu_dec(bpf_event_output_nest_level);
 	preempt_enable();