Skip to contentSkip to conversation

Various fixes around undefined behavior

PATCH Applied Ian Rogers 4 participants 9 messages Aug 21, 2025 – Sep 2, 2025 perf events

lore
Share & tools
lore

Ian Rogers

author reviewer
#0 lore raw

v2 cover Various fixes around undefined behavior

Fix various undefined behavior issues, improve tests to make them easier to diagnose and add assertions so that problems don't recur.

v2: Add Namhyung's acked-by. Drop container_of assert that ptr !=
NULL, to simplify the series. The bsearch UB fix was picked up as a patch by CT:
https://lore.kernel.org/r/[email protected]
It seems this patch series fell-through the cracks as v1 was sent/acked 9 months ago.

v1: https://lore.kernel.org/lkml/[email protected]/

Ian Rogers (5):
perf disasm: Avoid undefined behavior in incrementing NULL perf test trace_btf_enum: Skip if permissions are insufficient perf evsel: Avoid container_of on a NULL leader
perf test shell lock_contention: Extra debug diagnostics libperf event: Ensure tracing data is multiple of 8 sized

 tools/lib/perf/include/perf/event.h       |  1 +
 tools/perf/tests/shell/lock_contention.sh |  7 ++++++-
 tools/perf/tests/shell/trace_btf_enum.sh  | 11 +++++++++++
 tools/perf/util/disasm.c                  |  7 +++++--
 tools/perf/util/evsel.c                   |  2 ++
 5 files changed, 25 insertions(+), 3 deletions(-)
Signature · 1 line
2.51.0.rc1.193.gad69d77794-goog
author reviewer
#1 lore raw

PATCH v2 1/5 perf disasm: Avoid undefined behavior in incrementing NULL

Incrementing NULL is undefined behavior and triggers ubsan during the perf annotate test. Split a compound statement over two lines to avoid this.

Fixes 98f69a573c66 ("perf annotate: Split out util/disasm.c")
---
 tools/perf/util/disasm.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
tools/perf/util/disasm.c +5 -2
index b1e4919d016f..e257bd918c89 100644
@@ -390,13 +390,16 @@ static int jump__parse(struct arch *arch, struct ins_operands *ops, struct map_s
 	 * skip over possible up to 2 operands to get to address, e.g.:
 	 * tbnz	 w0, #26, ffff0000083cd190 <security_file_permission+0xd0>
 	 */
-	if (c++ != NULL) {
+	if (c != NULL) {
+		c++;
 		ops->target.addr = strtoull(c, NULL, 16);
 		if (!ops->target.addr) {
 			c = strchr(c, ',');
 			c = validate_comma(c, ops);
-			if (c++ != NULL)
+			if (c != NULL) {
+				c++;
 				ops->target.addr = strtoull(c, NULL, 16);
+			}
 		}
 	} else {
 		ops->target.addr = strtoull(ops->raw, NULL, 16);
Signature · 1 line
2.51.0.rc1.193.gad69d77794-goog
Reviewed-by on v2 1/5
#2 lore raw

Re: [PATCH v2 1/5] perf disasm: Avoid undefined behavior in incrementing NULL

Quoted 24 lines · Ian Rogers

Incrementing NULL is undefined behavior and triggers ubsan during the perf annotate test. Split a compound statement over two lines to avoid this.

Fixes 98f69a573c66 ("perf annotate: Split out util/disasm.c")
---
 tools/perf/util/disasm.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
tools/perf/util/disasm.c +2 -1
index b1e4919d016f..e257bd918c89 100644
@@ -390,13 +390,16 @@ static int jump__parse(struct arch *arch, struct ins_operands *ops, struct map_s
 	 * skip over possible up to 2 operands to get to address, e.g.:
 	 * tbnz	 w0, #26, ffff0000083cd190 <security_file_permission+0xd0>
 	 */
-	if (c++ != NULL) {
+	if (c != NULL) {
+		c++;
 		ops->target.addr = strtoull(c, NULL, 16);
 		if (!ops->target.addr) {

tools/perf/util/disasm.c static int jump__parse(struct arch *arch, struct ins_operands *ops, struct map_s · line 405

Quoted 10 lines
tools/perf/util/disasm.c +3 -1
 			c = strchr(c, ',');
 			c = validate_comma(c, ops);
-			if (c++ != NULL)
+			if (c != NULL) {
+				c++;
 				ops->target.addr = strtoull(c, NULL, 16);
+			}
 		}
 	} else {
 		ops->target.addr = strtoull(ops->raw, NULL, 16);

It is undefined behavior, but works correctly with GCC and Clang. In Gnulib, we allow it and suggest using -fno-sanitize=pointer-overflow instead [1].

But I can understand that is not every projects preference. Therefore, this change looks good to me.

Collin

[1] https://www.gnu.org/software/gnulib/manual/html_node/Unsupported-Platforms.html

author reviewer
#3 lore raw

PATCH v2 2/5 perf test trace_btf_enum: Skip if permissions are insufficient

In reply to Ian Rogers

Modify test behavior to skip if BPF calls fail with "Operation not permitted".

Fixes d66763fed30f ("perf test trace_btf_enum: Add regression test for the BTF augmentation of enums in 'perf trace'")
---
 tools/perf/tests/shell/trace_btf_enum.sh | 11 +++++++++++
 1 file changed, 11 insertions(+)
tools/perf/tests/shell/trace_btf_enum.sh +11 -0
index 572001d75d78..03e9f680a4a6 100755
@@ -23,6 +23,14 @@ check_vmlinux() {
   fi
 }
 
+check_permissions() {
+  if perf trace -e $syscall $TESTPROG 2>&1 | grep -q "Operation not permitted"
+  then
+    echo "trace+enum test [Skipped permissions]"
+    err=2
+  fi
+}
+
 trace_landlock() {
   echo "Tracing syscall ${syscall}"
 
@@ -56,6 +64,9 @@ trace_non_syscall() {
 }
 
 check_vmlinux
+if [ $err = 0 ]; then
+  check_permissions
+fi
 
 if [ $err = 0 ]; then
   trace_landlock
Signature · 1 line
2.51.0.rc1.193.gad69d77794-goog
author reviewer
#4 lore raw

PATCH v2 3/5 perf evsel: Avoid container_of on a NULL leader

In reply to Ian Rogers

An evsel should typically have a leader of itself, however, in tests like 'Sample parsing' a NULL leader may occur and the container_of will return a corrupt pointer. Avoid this with an explicit NULL test.

Fixes fba7c86601e2 ("libperf: Move 'leader' from tools/perf to perf_evsel::leader")
---
 tools/perf/util/evsel.c | 2 ++
 1 file changed, 2 insertions(+)
tools/perf/util/evsel.c +2 -0
index d264c143b592..496f42434327 100644
@@ -3935,6 +3935,8 @@ bool evsel__is_hybrid(const struct evsel *evsel)
 
 struct evsel *evsel__leader(const struct evsel *evsel)
 {
+	if (evsel->core.leader == NULL)
+		return NULL;
 	return container_of(evsel->core.leader, struct evsel, core);
 }
 
Signature · 1 line
2.51.0.rc1.193.gad69d77794-goog
author reviewer
#5 lore raw

PATCH v2 4/5 perf test shell lock_contention: Extra debug diagnostics

In reply to Ian Rogers

In test_record_concurrent, as stderr is sent to /dev/null, error messages are hidden. Change this to gather the error messages and dump them on failure.

Some minor sh->bash changes to add some more diagnostics in trap_cleanup.

---
 tools/perf/tests/shell/lock_contention.sh | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
tools/perf/tests/shell/lock_contention.sh +6 -1
index d33d9e4392b0..7248a74ca2a3 100755
@@ -7,14 +7,17 @@ set -e
 err=0
 perfdata=$(mktemp /tmp/__perf_test.perf.data.XXXXX)
 result=$(mktemp /tmp/__perf_test.result.XXXXX)
+errout=$(mktemp /tmp/__perf_test.errout.XXXXX)
 
 cleanup() {
 	rm -f ${perfdata}
 	rm -f ${result}
+	rm -f ${errout}
 	trap - EXIT TERM INT
 }
 
 trap_cleanup() {
+	echo "Unexpected signal in ${FUNCNAME[1]}"
 	cleanup
 	exit ${err}
 }
@@ -75,10 +78,12 @@ test_bpf()
 test_record_concurrent()
 {
 	echo "Testing perf lock record and perf lock contention at the same time"
-	perf lock record -o- -- perf bench sched messaging -p 2> /dev/null | \
+	perf lock record -o- -- perf bench sched messaging -p 2> ${errout} | \
 	perf lock contention -i- -E 1 -q 2> ${result}
 	if [ "$(cat "${result}" | wc -l)" != "1" ]; then
 		echo "[Fail] Recorded result count is not 1:" "$(cat "${result}" | wc -l)"
+		cat ${errout}
+		cat ${result}
 		err=1
 		exit
 	fi
Signature · 1 line
2.51.0.rc1.193.gad69d77794-goog
author reviewer
#6 lore raw

PATCH v2 5/5 libperf event: Ensure tracing data is multiple of 8 sized

In reply to Ian Rogers

Perf's synthetic-events.c will ensure 8-byte alignment of tracing data, writing it after a perf_record_header_tracing_data event. Add padding to struct perf_record_header_tracing_data to make it 16-byte rather than 12-byte sized.

Fixes 055c67ed3988 ("perf tools: Move event synthesizing routines to separate .c file")
---
 tools/lib/perf/include/perf/event.h | 1 +
 1 file changed, 1 insertion(+)
tools/lib/perf/include/perf/event.h +1 -0
index 6608f1e3701b..aa1e91c97a22 100644
@@ -291,6 +291,7 @@ struct perf_record_header_event_type {
 struct perf_record_header_tracing_data {
 	struct perf_event_header header;
 	__u32			 size;
+	__u32			 pad;
 };
 
 #define PERF_RECORD_MISC_BUILD_ID_SIZE (1 << 15)
Signature · 1 line
2.51.0.rc1.193.gad69d77794-goog
reviewer Reviewed-by on the v2 series
#7 lore raw
In reply to Ian Rogers
Quoted 19 lines · Ian Rogers

Fix various undefined behavior issues, improve tests to make them easier to diagnose and add assertions so that problems don't recur.

v2: Add Namhyung's acked-by. Drop container_of assert that ptr !=
NULL, to simplify the series. The bsearch UB fix was picked up as a patch by CT:
https://lore.kernel.org/r/[email protected]
It seems this patch series fell-through the cracks as v1 was sent/acked 9 months ago.

v1: https://lore.kernel.org/lkml/[email protected]/

Ian Rogers (5):
perf disasm: Avoid undefined behavior in incrementing NULL perf test trace_btf_enum: Skip if permissions are insufficient perf evsel: Avoid container_of on a NULL leader
perf test shell lock_contention: Extra debug diagnostics libperf event: Ensure tracing data is multiple of 8 sized

Quoted 7 lines
  tools/lib/perf/include/perf/event.h       |  1 +
  tools/perf/tests/shell/lock_contention.sh |  7 ++++++-
  tools/perf/tests/shell/trace_btf_enum.sh  | 11 +++++++++++
  tools/perf/util/disasm.c                  |  7 +++++--
  tools/perf/util/evsel.c                   |  2 ++
  5 files changed, 25 insertions(+), 3 deletions(-)
maintainer Applied
#8 lore raw
Quoted 5 lines · James Clark
Quoted 2 lines · Ian Rogers

Fix various undefined behavior issues, improve tests to make them easier to diagnose and add assertions so that problems don't recur.

Thanks, applied to perf-tools-next,

- Arnaldo