perf sample: Make user_regs and intr_regs optional

The struct dump_regs contains 512 bytes of cache_regs, meaning the two
values in perf_sample contribute 1088 bytes of its total 1384 bytes
size. Initializing this much memory has a cost reported by Tavian
Barnes <tavianator@tavianator.com> as about 2.5% when running `perf
script --itrace=i0`:
https://lore.kernel.org/lkml/d841b97b3ad2ca8bcab07e4293375fb7c32dfce7.1736618095.git.tavianator@tavianator.com/

Adrian Hunter <adrian.hunter@intel.com> replied that the zero
initialization was necessary and couldn't simply be removed.

This patch aims to strike a middle ground of still zeroing the
perf_sample, but removing 79% of its size by make user_regs and
intr_regs optional pointers to zalloc-ed memory. To support the
allocation accessors are created for user_regs and intr_regs. To
support correct cleanup perf_sample__init and perf_sample__exit
functions are created and added throughout the code base.

Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250113194345.1537821-1-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

tools/perf/arch/x86/tests/dwarf-unwind.c

@@ -53,7 +53,7 @@ static int sample_ustack(struct perf_sample *sample,
 int test__arch_unwind_sample(struct perf_sample *sample,
 			     struct thread *thread)
 {
-	struct regs_dump *regs = &sample->user_regs;
+	struct regs_dump *regs = perf_sample__user_regs(sample);
 	u64 *buf;
 
 	buf = malloc(sizeof(u64) * PERF_REGS_MAX);

tools/perf/arch/x86/util/unwind-libdw.c

@@ -8,7 +8,7 @@
 bool libdw__arch_set_initial_registers(Dwfl_Thread *thread, void *arg)
 {
 	struct unwind_info *ui = arg;
-	struct regs_dump *user_regs = &ui->sample->user_regs;
+	struct regs_dump *user_regs = perf_sample__user_regs(ui->sample);
 	Dwarf_Word dwarf_regs[17];
 	unsigned nregs;
 

tools/perf/builtin-record.c

@@ -1917,9 +1917,10 @@ static void __record__save_lost_samples(struct record *rec, struct evsel *evsel,
 					u16 misc_flag)
 {
 	struct perf_sample_id *sid;
-	struct perf_sample sample = {};
+	struct perf_sample sample;
 	int id_hdr_size;
 
+	perf_sample__init(&sample, /*all=*/true);
 	lost->lost = lost_count;
 	if (evsel->core.ids) {
 		sid = xyarray__entry(evsel->core.sample_id, cpu_idx, thread_idx);
@@ -1931,6 +1932,7 @@ static void __record__save_lost_samples(struct record *rec, struct evsel *evsel,
 	lost->header.size = sizeof(*lost) + id_hdr_size;
 	lost->header.misc = misc_flag;
 	record__write(rec, NULL, lost, lost->header.size);
+	perf_sample__exit(&sample);
 }
 
 static void record__read_lost_samples(struct record *rec)

tools/perf/builtin-script.c

@@ -783,14 +783,20 @@ tod_scnprintf(struct perf_script *script, char *buf, int buflen,
 static int perf_sample__fprintf_iregs(struct perf_sample *sample,
 				      struct perf_event_attr *attr, const char *arch, FILE *fp)
 {
-	return perf_sample__fprintf_regs(&sample->intr_regs,
+	if (!sample->intr_regs)
+		return 0;
+
+	return perf_sample__fprintf_regs(perf_sample__intr_regs(sample),
 					 attr->sample_regs_intr, arch, fp);
 }
 
 static int perf_sample__fprintf_uregs(struct perf_sample *sample,
 				      struct perf_event_attr *attr, const char *arch, FILE *fp)
 {
-	return perf_sample__fprintf_regs(&sample->user_regs,
+	if (!sample->user_regs)
+		return 0;
+
+	return perf_sample__fprintf_regs(perf_sample__user_regs(sample),
 					 attr->sample_regs_user, arch, fp);
 }
 

tools/perf/builtin-top.c

@@ -1157,6 +1157,7 @@ static int deliver_event(struct ordered_events *qe,
 		return 0;
 	}
 
+	perf_sample__init(&sample, /*all=*/false);
 	ret = evlist__parse_sample(evlist, event, &sample);
 	if (ret) {
 		pr_err("Can't parse sample, err = %d\n", ret);
@@ -1167,8 +1168,10 @@ static int deliver_event(struct ordered_events *qe,
 	assert(evsel != NULL);
 
 	if (event->header.type == PERF_RECORD_SAMPLE) {
-		if (evswitch__discard(&top->evswitch, evsel))
-			return 0;
+		if (evswitch__discard(&top->evswitch, evsel)) {
+			ret = 0;
+			goto next_event;
+		}
 		++top->samples;
 	}
 
@@ -1219,6 +1222,7 @@ static int deliver_event(struct ordered_events *qe,
 
 	ret = 0;
 next_event:
+	perf_sample__exit(&sample);
 	return ret;
 }
 

tools/perf/builtin-trace.c

@@ -4066,13 +4066,16 @@ static int __trace__deliver_event(struct trace *trace, union perf_event *event)
 {
 	struct evlist *evlist = trace->evlist;
 	struct perf_sample sample;
-	int err = evlist__parse_sample(evlist, event, &sample);
+	int err;
 
+	perf_sample__init(&sample, /*all=*/false);
+	err = evlist__parse_sample(evlist, event, &sample);
 	if (err)
 		fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
 	else
 		trace__handle_event(trace, event, &sample);
 
+	perf_sample__exit(&sample);
 	return 0;
 }
 

tools/perf/tests/code-reading.c

@@ -479,19 +479,25 @@ static int process_sample_event(struct machine *machine,
 	struct thread *thread;
 	int ret;
 
-	if (evlist__parse_sample(evlist, event, &sample)) {
+	perf_sample__init(&sample, /*all=*/false);
+	ret = evlist__parse_sample(evlist, event, &sample);
+	if (ret) {
 		pr_debug("evlist__parse_sample failed\n");
-		return -1;
+		ret = -1;
+		goto out;
 	}
 
 	thread = machine__findnew_thread(machine, sample.pid, sample.tid);
 	if (!thread) {
 		pr_debug("machine__findnew_thread failed\n");
-		return -1;
+		ret = -1;
+		goto out;
 	}
 
 	ret = read_object_code(sample.ip, READLEN, sample.cpumode, thread, state);
 	thread__put(thread);
+out:
+	perf_sample__exit(&sample);
 	return ret;
 }
 

tools/perf/tests/dwarf-unwind.c

@@ -115,8 +115,7 @@ NO_TAIL_CALL_ATTRIBUTE noinline int test_dwarf_unwind__thread(struct thread *thr
 	unsigned long cnt = 0;
 	int err = -1;
 
-	memset(&sample, 0, sizeof(sample));
-
+	perf_sample__init(&sample, /*all=*/true);
 	if (test__arch_unwind_sample(&sample, thread)) {
 		pr_debug("failed to get unwind sample\n");
 		goto out;
@@ -134,7 +133,8 @@ NO_TAIL_CALL_ATTRIBUTE noinline int test_dwarf_unwind__thread(struct thread *thr
 
  out:
 	zfree(&sample.user_stack.data);
-	zfree(&sample.user_regs.regs);
+	zfree(&sample.user_regs->regs);
+	perf_sample__exit(&sample);
 	return err;
 }
 

tools/perf/tests/mmap-basic.c

@@ -130,14 +130,17 @@ static int test__basic_mmap(struct test_suite *test __maybe_unused, int subtest
 			goto out_delete_evlist;
 		}
 
+		perf_sample__init(&sample, /*all=*/false);
 		err = evlist__parse_sample(evlist, event, &sample);
 		if (err) {
 			pr_err("Can't parse sample, err = %d\n", err);
+			perf_sample__exit(&sample);
 			goto out_delete_evlist;
 		}
 
 		err = -1;
 		evsel = evlist__id2evsel(evlist, sample.id);
+		perf_sample__exit(&sample);
 		if (evsel == NULL) {
 			pr_debug("event with id %" PRIu64
 				 " doesn't map to an evsel\n", sample.id);

tools/perf/tests/openat-syscall-tp-fields.c

@@ -111,14 +111,16 @@ static int test__syscall_openat_tp_fields(struct test_suite *test __maybe_unused
 					continue;
 				}
 
+				perf_sample__init(&sample, /*all=*/false);
 				err = evsel__parse_sample(evsel, event, &sample);
 				if (err) {
 					pr_debug("Can't parse sample, err = %d\n", err);
+					perf_sample__exit(&sample);
 					goto out_delete_evlist;
 				}
 
 				tp_flags = evsel__intval(evsel, &sample, "flags");
-
+				perf_sample__exit(&sample);
 				if (flags != tp_flags) {
 					pr_debug("%s: Expected flags=%#x, got %#x\n",
 						 __func__, flags, tp_flags);

tools/perf/tests/parse-no-sample-id-all.c

@@ -13,6 +13,7 @@
 static int process_event(struct evlist **pevlist, union perf_event *event)
 {
 	struct perf_sample sample;
+	int ret;
 
 	if (event->header.type == PERF_RECORD_HEADER_ATTR) {
 		if (perf_event__process_attr(NULL, event, pevlist)) {
@@ -28,7 +29,10 @@ static int process_event(struct evlist **pevlist, union perf_event *event)
 	if (!*pevlist)
 		return -1;
 
-	if (evlist__parse_sample(*pevlist, event, &sample)) {
+	perf_sample__init(&sample, /*all=*/false);
+	ret = evlist__parse_sample(*pevlist, event, &sample);
+	perf_sample__exit(&sample);
+	if (ret) {
 		pr_debug("evlist__parse_sample failed\n");
 		return -1;
 	}

tools/perf/tests/perf-record.c

@@ -70,6 +70,7 @@ static int test__PERF_RECORD(struct test_suite *test __maybe_unused, int subtest
 	int total_events = 0, nr_events[PERF_RECORD_MAX] = { 0, };
 	char sbuf[STRERR_BUFSIZE];
 
+	perf_sample__init(&sample, /*all=*/false);
 	if (evlist == NULL) /* Fallback for kernels lacking PERF_COUNT_SW_DUMMY */
 		evlist = evlist__new_default();
 
@@ -330,6 +331,7 @@ static int test__PERF_RECORD(struct test_suite *test __maybe_unused, int subtest
 out_delete_evlist:
 	evlist__delete(evlist);
 out:
+	perf_sample__exit(&sample);
 	if (err == -EACCES)
 		return TEST_SKIP;
 	if (err < 0 || errs != 0)

tools/perf/tests/perf-time-to-tsc.c

@@ -153,6 +153,7 @@ static int test__perf_time_to_tsc(struct test_suite *test __maybe_unused, int su
 		while ((event = perf_mmap__read_event(&md->core)) != NULL) {
 			struct perf_sample sample;
 
+			perf_sample__init(&sample, /*all=*/false);
 			if (event->header.type != PERF_RECORD_COMM ||
 			    (pid_t)event->comm.pid != getpid() ||
 			    (pid_t)event->comm.tid != getpid())
@@ -170,6 +171,7 @@ static int test__perf_time_to_tsc(struct test_suite *test __maybe_unused, int su
 			}
 next_event:
 			perf_mmap__consume(&md->core);
+			perf_sample__exit(&sample);
 		}
 		perf_mmap__read_done(&md->core);
 	}

tools/perf/tests/sample-parsing.c

@@ -40,8 +40,8 @@
 #define BS_EXPECTED_LE	0x1aa00000000
 #define FLAG(s)	s->branch_stack->entries[i].flags
 
-static bool samples_same(const struct perf_sample *s1,
-			 const struct perf_sample *s2,
+static bool samples_same(struct perf_sample *s1,
+			 struct perf_sample *s2,
 			 u64 type, u64 read_format, bool needs_swap)
 {
 	size_t i;
@@ -126,13 +126,15 @@ static bool samples_same(const struct perf_sample *s1,
 	}
 
 	if (type & PERF_SAMPLE_REGS_USER) {
-		size_t sz = hweight_long(s1->user_regs.mask) * sizeof(u64);
-
-		COMP(user_regs.mask);
-		COMP(user_regs.abi);
-		if (s1->user_regs.abi &&
-		    (!s1->user_regs.regs || !s2->user_regs.regs ||
-		     memcmp(s1->user_regs.regs, s2->user_regs.regs, sz))) {
+		struct regs_dump *s1_regs = perf_sample__user_regs(s1);
+		struct regs_dump *s2_regs = perf_sample__user_regs(s2);
+		size_t sz = hweight_long(s1_regs->mask) * sizeof(u64);
+
+		COMP(user_regs->mask);
+		COMP(user_regs->abi);
+		if (s1_regs->abi &&
+		    (!s1_regs->regs || !s2_regs->regs ||
+		     memcmp(s1_regs->regs, s2_regs->regs, sz))) {
 			pr_debug("Samples differ at 'user_regs'\n");
 			return false;
 		}
@@ -157,13 +159,15 @@ static bool samples_same(const struct perf_sample *s1,
 		COMP(transaction);
 
 	if (type & PERF_SAMPLE_REGS_INTR) {
-		size_t sz = hweight_long(s1->intr_regs.mask) * sizeof(u64);
-
-		COMP(intr_regs.mask);
-		COMP(intr_regs.abi);
-		if (s1->intr_regs.abi &&
-		    (!s1->intr_regs.regs || !s2->intr_regs.regs ||
-		     memcmp(s1->intr_regs.regs, s2->intr_regs.regs, sz))) {
+		struct regs_dump *s1_regs = perf_sample__intr_regs(s1);
+		struct regs_dump *s2_regs = perf_sample__intr_regs(s2);
+		size_t sz = hweight_long(s1_regs->mask) * sizeof(u64);
+
+		COMP(intr_regs->mask);
+		COMP(intr_regs->abi);
+		if (s1_regs->abi &&
+		    (!s1_regs->regs || !s2_regs->regs ||
+		     memcmp(s1_regs->regs, s2_regs->regs, sz))) {
 			pr_debug("Samples differ at 'intr_regs'\n");
 			return false;
 		}
@@ -223,6 +227,16 @@ static int do_test(u64 sample_type, u64 sample_regs, u64 read_format)
 	const u32 raw_data[] = {0x12345678, 0x0a0b0c0d, 0x11020304, 0x05060708, 0 };
 	const u64 data[] = {0x2211443366558877ULL, 0, 0xaabbccddeeff4321ULL};
 	const u64 aux_data[] = {0xa55a, 0, 0xeeddee, 0x0282028202820282};
+	struct regs_dump user_regs = {
+		.abi	= PERF_SAMPLE_REGS_ABI_64,
+		.mask	= sample_regs,
+		.regs	= regs,
+	};
+	struct regs_dump intr_regs = {
+		.abi	= PERF_SAMPLE_REGS_ABI_64,
+		.mask	= sample_regs,
+		.regs	= regs,
+	};
 	struct perf_sample sample = {
 		.ip		= 101,
 		.pid		= 102,
@@ -241,11 +255,7 @@ static int do_test(u64 sample_type, u64 sample_regs, u64 read_format)
 		.callchain	= &callchain.callchain,
 		.no_hw_idx      = false,
 		.branch_stack	= &branch_stack.branch_stack,
-		.user_regs	= {
-			.abi	= PERF_SAMPLE_REGS_ABI_64,
-			.mask	= sample_regs,
-			.regs	= regs,
-		},
+		.user_regs	= &user_regs,
 		.user_stack	= {
 			.size	= sizeof(data),
 			.data	= (void *)data,
@@ -254,11 +264,7 @@ static int do_test(u64 sample_type, u64 sample_regs, u64 read_format)
 			.time_enabled = 0x030a59d664fca7deULL,
 			.time_running = 0x011b6ae553eb98edULL,
 		},
-		.intr_regs	= {
-			.abi	= PERF_SAMPLE_REGS_ABI_64,
-			.mask	= sample_regs,
-			.regs	= regs,
-		},
+		.intr_regs	= &intr_regs,
 		.phys_addr	= 113,
 		.cgroup		= 114,
 		.data_page_size = 115,
@@ -273,6 +279,8 @@ static int do_test(u64 sample_type, u64 sample_regs, u64 read_format)
 	size_t i, sz, bufsz;
 	int err, ret = -1;
 
+	perf_sample__init(&sample_out, /*all=*/false);
+	perf_sample__init(&sample_out_endian, /*all=*/false);
 	if (sample_type & PERF_SAMPLE_REGS_USER)
 		evsel.core.attr.sample_regs_user = sample_regs;
 
@@ -361,6 +369,8 @@ static int do_test(u64 sample_type, u64 sample_regs, u64 read_format)
 	ret = 0;
 out_free:
 	free(event);
+	perf_sample__exit(&sample_out_endian);
+	perf_sample__exit(&sample_out);
 	if (ret && read_format)
 		pr_debug("read_format %#"PRIx64"\n", read_format);
 	return ret;

tools/perf/tests/sw-clock.c

@@ -104,19 +104,22 @@ static int __test__sw_clock_freq(enum perf_sw_ids clock_id)
 	while ((event = perf_mmap__read_event(&md->core)) != NULL) {
 		struct perf_sample sample;
 
+		perf_sample__init(&sample, /*all=*/false);
 		if (event->header.type != PERF_RECORD_SAMPLE)
 			goto next_event;
 
 		err = evlist__parse_sample(evlist, event, &sample);
 		if (err < 0) {
 			pr_debug("Error during parse sample\n");
+			perf_sample__exit(&sample);
 			goto out_delete_evlist;
 		}
 
 		total_periods += sample.period;
 		nr_samples++;
 next_event:
 		perf_mmap__consume(&md->core);
+		perf_sample__exit(&sample);
 	}
 	perf_mmap__read_done(&md->core);