Skip to contentSkip to conversation

perf python: Fix python extension bugs and memory leaks

PATCH Superseded Ian Rogers 3 participants 75 messages Jun 23 – Aug 12 perf events

lore
Share & tools
lore

Ian Rogers

author reviewer
#0 lore raw

cover perf python: Fix python extension bugs and memory leaks

This patch series addresses a number of leaks and python extension bugs discovered during the review of the python scripting phase 1 refactoring series.

These fixes have been separated from the scripting changes into their own dedicated patch series to ease review and integration, as they address regressions or existing bugs rather than adding new features.

The series breaks down the monolithic python extension fixes into discrete logical patches addressing validation, initialization, and resource leaks.

Ian Rogers (13):
perf script: Fix metric_evlist leak in script_find_metrics perf stat: Fix evsel_list leak in cmd_stat
perf top: Fix sb_evlist leak
perf python: Fix memory leak in pyrf_evlist__get_pollfd perf synthetic-events: Fix uninitialized pthread_join perf data: Fix directory file descriptor leak in perf_data__switch perf test: Fix skiplist leak in cmd_test
perf python: Check counts_values size in set_values perf python: Validate CPU and thread maps in pyrf_evsel__open perf python: Validate attribute setters in pyrf_evsel perf python: Zero initialize perf_data in pyrf_data__init perf python: Add thread uninitialized checks
perf python: Fix MetricGroup return type in perf.pyi

 tools/perf/builtin-script.c        |   2 +-
 tools/perf/builtin-stat.c          |   9 ++-
 tools/perf/builtin-top.c           |   1 +
 tools/perf/python/perf.pyi         |   4 +-
 tools/perf/tests/builtin-test.c    |   5 +-
 tools/perf/util/data.c             |   3 +-
 tools/perf/util/python.c           | 114 +++++++++++++++++++++++++----
 tools/perf/util/synthetic-events.c |   4 +-
 8 files changed, 116 insertions(+), 26 deletions(-)
Signature · 1 line
2.55.0.rc0.786.g65d90a0328-goog
author reviewer
#1 lore raw

PATCH 1/13 perf script: Fix metric_evlist leak in script_find_metrics

Fix a memory leak in script_find_metrics() where metric_evlist is leaked when returning early on error paths.

Assisted-by Antigravity:gemini-3.1-pro
Fixes 3622990efaab ("perf script: Change metric format to use json metrics")
---
 tools/perf/builtin-script.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/builtin-script.c +1 -1
index 47afd8cdc2b7..6b69e982ba62 100644
@@ -2275,8 +2275,8 @@ static int script_find_metrics(const struct pmu_metric *pm,
 	}
 	pr_debug("Found metric '%s' whose evsels match those of in the perf data\n",
 		 pm->metric_name);
-	evlist__put(metric_evlist);
 out:
+	evlist__put(metric_evlist);
 	return 0;
 }
 
Signature · 1 line
2.55.0.rc0.786.g65d90a0328-goog
author reviewer
#2 lore raw

PATCH 2/13 perf stat: Fix evsel_list leak in cmd_stat

In reply to Ian Rogers

Fix a memory leak in cmd_stat() where evsel_list is leaked if an error occurs while opening the output file.

Assisted-by Antigravity:gemini-3.1-pro
Fixes 361c99a661a7 ("perf evsel: Introduce perf_evlist")
---
 tools/perf/builtin-stat.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
tools/perf/builtin-stat.c +6 -3
index 3f897b2e8638..7cb6ea937e87 100644
@@ -2718,7 +2718,8 @@ int cmd_stat(int argc, const char **argv)
 		output = fopen(output_name, mode);
 		if (!output) {
 			perror("failed to create output file");
-			return -1;
+			status = -1;
+			goto out;
 		}
 		if (!stat_config.json_output) {
 			clock_gettime(CLOCK_REALTIME, &tm);
@@ -2729,7 +2730,8 @@ int cmd_stat(int argc, const char **argv)
 		output = fdopen(output_fd, mode);
 		if (!output) {
 			perror("Failed opening logfd");
-			return -errno;
+			status = -errno;
+			goto out;
 		}
 	}
 
@@ -2738,7 +2740,8 @@ int cmd_stat(int argc, const char **argv)
 		parse_options_usage(stat_usage, stat_options, "o", 1);
 		parse_options_usage(NULL, stat_options, "log-fd", 0);
 		parse_options_usage(NULL, stat_options, "interval-clear", 0);
-		return -1;
+		status = -1;
+		goto out;
 	}
 
 	stat_config.output = output;
Signature · 1 line
2.55.0.rc0.786.g65d90a0328-goog
author reviewer
#3 lore raw

PATCH 3/13 perf top: Fix sb_evlist leak

In reply to Ian Rogers

Fix a memory leak in cmd_top() where top.sb_evlist was not freed if evlist__add_bpf_sb_event() fails. Note that evlist__start_sb_thread() and evlist__stop_sb_thread() take ownership of the evlist and free it, so we must only free it if we fail before starting the thread.

Assisted-by Antigravity:gemini-3.1-pro
Fixes b38d85ef49cf ("perf bpf: Decouple creating the evlist from adding the SB event")
---
 tools/perf/builtin-top.c | 1 +
 1 file changed, 1 insertion(+)
tools/perf/builtin-top.c +1 -0
index 5933c46ee137..abc74b8ab759 100644
@@ -1881,6 +1881,7 @@ int cmd_top(int argc, const char **argv)
 		if (evlist__add_bpf_sb_event(top.sb_evlist, &host_env)) {
 			pr_err("Couldn't ask for PERF_RECORD_BPF_EVENT side band events.\n.");
 			status = -EINVAL;
+			evlist__put(top.sb_evlist);
 			goto out_put_evlist;
 		}
 	}
Signature · 1 line
2.55.0.rc0.786.g65d90a0328-goog
#4 lore raw

Re: [PATCH v1 03/13] perf top: Fix sb_evlist leak

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Pre-existing issues:
- [High] Error path in record__setup_sb_evlist() bypasses state update, leading to a segmentation fault from an uninitialized pthread_join().
- [High] Sideband thread and sb_evlist are leaked when using --switch-output-event together with --no-bpf-event.
--

commit 9bc374ffd1b7ae00f5342c051ca479b32bc92942
Author: Ian Rogers <[email protected]>

perf top: Fix sb_evlist leak

Fix a memory leak in cmd_top() where top.sb_evlist was not freed if evlist__add_bpf_sb_event() fails. Note that evlist__start_sb_thread() and evlist__stop_sb_thread() take ownership of the evlist and free it, so we must only free it if we fail before starting the thread.

tools/perf/builtin-top.c int cmd_top(int argc, const char **argv) · line 1887

Quoted 12 lines
tools/perf/builtin-top.c +1 -0
index 5933c46ee137e..abc74b8ab7599 100644
@@ -1881,6 +1881,7 @@ int cmd_top(int argc, const char **argv)
 		if (evlist__add_bpf_sb_event(top.sb_evlist, &host_env)) {
 			pr_err("Couldn't ask for PERF_RECORD_BPF_EVENT side band events.\n.");
 			status = -EINVAL;
+			evlist__put(top.sb_evlist);
 			goto out_put_evlist;
 		}
 	}

[Severity: High]
This is a pre-existing issue, but does tools/perf/builtin-record.c have a similar problem where the error path in record__setup_sb_evlist() bypasses updating the state?

If evlist__add_bpf_sb_event() fails in record__setup_sb_evlist(), it returns -1 without updating the opts->no_bpf_event flag:

tools/perf/builtin-record.c:record__setup_sb_evlist() {
    ...
	if (evlist__add_bpf_sb_event(rec->sb_evlist, perf_session__env(rec->session))) {
		pr_err("Couldn't ask for PERF_RECORD_BPF_EVENT side band events.\n.");
		return -1;
	}
    ...
}

During cleanup in __cmd_record(), the unmodified no_bpf_event flag causes evlist__stop_sb_thread() to be unconditionally invoked:

tools/perf/builtin-record.c:__cmd_record() {
    ...
	if (!opts->no_bpf_event)
		evlist__stop_sb_thread(rec->sb_evlist);
    ...
}

This could blindly call pthread_join() on the uninitialized thread ID in evlist__stop_sb_thread():

tools/perf/util/sideband_evlist.c:evlist__stop_sb_thread() {
    ...
	pthread_join(*evlist__sb_thread_th(evlist), NULL);
    ...
}

Could this trigger undefined behavior and a crash?

[Severity: High]
This is also a pre-existing issue, but is there a leak of the sideband thread and sb_evlist in tools/perf/builtin-record.c when using --switch-output-event together with --no-bpf-event?

In record__setup_sb_evlist(), the sideband thread is started when using --switch-output-event, even if BPF events are disabled:

tools/perf/builtin-record.c:record__setup_sb_evlist() {
    ...
	if (evlist__start_sb_thread(rec->sb_evlist, &rec->opts.target)) {
    ...
}

However, cleanup in __cmd_record() is bypassed because opts->no_bpf_event evaluates to true:

tools/perf/builtin-record.c:__cmd_record() {
    ...
	if (!opts->no_bpf_event)
		evlist__stop_sb_thread(rec->sb_evlist);
    ...
}

Does this permanently strand the background poll thread and leak the allocated evlist during process teardown?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=3
author reviewer
#5 lore raw

PATCH 4/13 perf python: Fix memory leak in pyrf_evlist__get_pollfd

In reply to Ian Rogers

Fix a Python list object leak in pyrf_evlist__get_pollfd() by adding a missing Py_DECREF on the error exit path.

Assisted-by Antigravity:gemini-3.1-pro
Fixes 877108e42b1b ("perf tools: Initial python binding")
---
 tools/perf/util/python.c | 1 +
 1 file changed, 1 insertion(+)
tools/perf/util/python.c +1 -0
index 11e1f39c1bea..21e8b5a23a1c 100644
@@ -2843,6 +2843,7 @@ static PyObject *pyrf_evlist__get_pollfd(struct pyrf_evlist *pevlist,
 
 	return list;
 free_list:
+	Py_DECREF(list);
 	return PyErr_NoMemory();
 }
 
Signature · 1 line
2.55.0.rc0.786.g65d90a0328-goog
#6 lore raw

Re: [PATCH v1 04/13] perf python: Fix memory leak in pyrf_evlist__get_pollfd

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [High] The newly added `Py_DECREF(list)` introduces a segmentation fault on the error path if the initial…

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] The newly added Py_DECREF(list) introduces a segmentation fault on the error path if the initial PyList_New(0) allocation fails. --

commit dfef1d27b0ba6fba485792f67daeb68ccdaac6cd
Author: Ian Rogers <[email protected]>

perf python: Fix memory leak in pyrf_evlist__get_pollfd

Fix a Python list object leak in pyrf_evlist__get_pollfd() by adding a missing Py_DECREF on the error exit path.

tools/perf/util/python.c static PyObject *pyrf_evlist__get_pollfd(struct pyrf_evlist *pevlist, · line 2846

Quoted 9 lines
tools/perf/util/python.c +1 -0
index 11e1f39c1bea9..21e8b5a23a1c9 100644
@@ -2843,6 +2843,7 @@ static PyObject *pyrf_evlist__get_pollfd(struct pyrf_evlist *pevlist,
 
 	return list;
 free_list:
+	Py_DECREF(list);

[Severity: High]
If PyList_New(0) fails and returns NULL, this will attempt to decrement the reference count of a NULL pointer.

Because Py_DECREF unconditionally dereferences its argument, will this result in a segmentation fault on the error path?

Should we check list for NULL earlier, or use Py_XDECREF(list) instead?

Quoted 2 lines
tools/perf/util/python.c +0 -0
 	return PyErr_NoMemory();
 }
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=4
author reviewer
#7 lore raw

PATCH 5/13 perf synthetic-events: Fix uninitialized pthread_join

In reply to Ian Rogers

In perf_event__synthesize_threads(), fix an uninitialized pthread_join() call when thread creation fails by only joining the successfully created threads.

Assisted-by Antigravity:gemini-3.1-pro
Fixes 340b47f510bb ("perf top: Implement multithreading for perf_event__synthesize_threads")
---
 tools/perf/util/synthetic-events.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tools/perf/util/synthetic-events.c +2 -2
index b75f9dcf4dbf..35de41ad14d3 100644
@@ -1104,8 +1104,8 @@ int perf_event__synthesize_threads(const struct perf_tool *tool,
 	}
 	err = 0;
 out_join:
-	for (i = 0; i < thread_nr; i++)
-		pthread_join(synthesize_threads[i], NULL);
+	for (j = 0; j < i; j++)
+		pthread_join(synthesize_threads[j], NULL);
 	free(args);
 free_threads:
 	free(synthesize_threads);
Signature · 1 line
2.55.0.rc0.786.g65d90a0328-goog
#8 lore raw

Re: [PATCH v1 05/13] perf synthetic-events: Fix uninitialized pthread_join

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] Divide by zero crash in perf_event__synthesize_threads when no tasks are found. --

commit 30888fa2ce27193736ae38293c89b11a5e47a514
Author: Ian Rogers <[email protected]>

perf synthetic-events: Fix uninitialized pthread_join

This commit fixes an uninitialized pthread_join() call when thread creation fails in perf_event__synthesize_threads() by ensuring only successfully created threads are joined.

tools/perf/util/synthetic-events.c int perf_event__synthesize_threads(const struct perf_tool *tool,

Quoted 5 lines
tools/perf/util/synthetic-events.c +0 -0
index b75f9dcf4dbfe..35de41ad14d33 100644
@@ -1104,8 +1104,8 @@ int perf_event__synthesize_threads(const struct perf_tool *tool,

[Severity: High]
This is a pre-existing issue, but does this code suffer from a fatal division by zero when no tasks are found?

If scandir() finds no matching tasks in /proc (which can happen in an empty chroot or isolated container), n is 0. If the system has multiple CPU cores (thread_nr > 1), we bypass the single-thread fast path.

perf_event__synthesize_threads() then clamps thread_nr:

if (thread_nr > n)
thread_nr = n;

And shortly after evaluates:

    num_per_thread = n / thread_nr;

Does this divide by zero (0 / 0) and cause the perf tool to crash with a fatal SIGFPE? Could we handle the n == 0 case earlier to prevent this regression?

Quoted 10 lines
tools/perf/util/synthetic-events.c +2 -2
 	}
 	err = 0;
 out_join:
-	for (i = 0; i < thread_nr; i++)
-		pthread_join(synthesize_threads[i], NULL);
+	for (j = 0; j < i; j++)
+		pthread_join(synthesize_threads[j], NULL);
 	free(args);
 free_threads:
 	free(synthesize_threads);
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=5
author reviewer
#9 lore raw

PATCH 6/13 perf data: Fix directory file descriptor leak in perf_data__switch

In reply to Ian Rogers

Fix a file descriptor leak in perf_data__switch() when running in directory mode. By substituting perf_data__close() for perf_data_file__close(), any thread-local directory files are properly closed during file rotation.

Assisted-by Antigravity:gemini-3.1-pro
Fixes 040f9915e99e ("perf data: Add perf_data_file__switch() helper")
---
 tools/perf/util/data.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
tools/perf/util/data.c +1 -2
index 33fd1b82019e..32d4d908ad38 100644
@@ -472,8 +472,7 @@ int perf_data__switch(struct perf_data *data,
 		pr_warning("Failed to rename %s to %s\n", data->path, *new_filepath);
 
 	if (!at_exit) {
-		perf_data_file__close(&data->file);
-		data->open = false;
+		perf_data__close(data);
 		ret = perf_data__open(data);
 		if (ret < 0)
 			goto out;
Signature · 1 line
2.55.0.rc0.786.g65d90a0328-goog
author reviewer
#10 lore raw

PATCH 7/13 perf test: Fix skiplist leak in cmd_test

In reply to Ian Rogers

Fix a memory leak in cmd_test() where skiplist was not freed on exit paths.

Assisted-by Antigravity:gemini-3.1-pro
Fixes 2ae828786c65 ("perf test: Allow skipping tests")
---
 tools/perf/tests/builtin-test.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
tools/perf/tests/builtin-test.c +4 -1
index 7e75f590f225..cf5625cf4f11 100644
@@ -1781,9 +1781,12 @@ int cmd_test(int argc, const char **argv)
 	rlimit__bump_memlock();
 
 	suites = build_suites();
-	if (!suites)
+	if (!suites) {
+		intlist__delete(skiplist);
 		return errno ? -errno : -ENOMEM;
+	}
 	ret = __cmd_test(suites, argc, argv, skiplist);
 	free(suites);
+	intlist__delete(skiplist);
 	return ret;
 }
Signature · 1 line
2.55.0.rc0.786.g65d90a0328-goog
author reviewer
#11 lore raw

PATCH 8/13 perf python: Check counts_values size in set_values

In reply to Ian Rogers

The set_values function incorrectly assumed the list contained exactly 5 elements. Add a check to prevent out-of-bounds access.

Fixes 877108e42b1b ("perf tools: Initial python binding")
---
 tools/perf/util/python.c | 5 +++++
 1 file changed, 5 insertions(+)
tools/perf/util/python.c +5 -0
index 21e8b5a23a1c..f9398bc30a80 100644
@@ -1913,6 +1913,11 @@ static int pyrf_counts_values_set_values(struct pyrf_counts_values *self, PyObje
 	}
 
 	size = PyList_Size(list);
+	if (size != 5) {
+		PyErr_SetString(PyExc_ValueError, "List must have exactly 5 entries");
+		return -1;
+	}
+
 	for (Py_ssize_t i = 0; i < size; i++) {
 		item = PyList_GetItem(list, i);
 		if (!PyLong_Check(item)) {
Signature · 1 line
2.55.0.rc0.786.g65d90a0328-goog
#12 lore raw

Re: [PATCH v1 08/13] perf python: Check counts_values size in set_values

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Pre-existing issues:
- [High] The Python bindings for perf_counts_values incorrectly use 32-bit-limited C long functions and types for uint64_t data, leading to data truncation on 32-bit platforms, incorrect negative values, and unhandled OverflowError exceptions causing fatal SystemError crashes.
- [High] Deleting the .values attribute from Python causes a segmentation fault due to a NULL pointer dereference in the property setter.
--

commit a4cb87d6d79ff9be6c884154fa1d1f913fe545d5
Author: Ian Rogers <[email protected]>

perf python: Check counts_values size in set_values

The set_values function incorrectly assumed the list contained exactly 5 elements. Add a check to prevent out-of-bounds access.

tools/perf/util/python.c static int pyrf_counts_values_set_values(struct pyrf_counts_values *self, PyObje · line 1913

Quoted 6 lines
tools/perf/util/python.c +0 -0
index 21e8b5a23a1c9..f9398bc30a80b 100644
@@ -1913,6 +1913,11 @@ static int pyrf_counts_values_set_values(struct pyrf_counts_values *self, PyObje
 	}

[Severity: High]
This is a pre-existing issue, but does this function cause a NULL pointer dereference when the attribute is deleted?

Just above this context in pyrf_counts_values_set_values():

	if (!PyList_Check(list)) {
		PyErr_SetString(PyExc_TypeError, "Value assigned must be a list");
		return -1;
	}

If a user executes del obj.values from Python code, the Python C API calls the property setter with list set to NULL. Passing a NULL list directly into PyList_Check(list) will dereference the object to check its type, which causes an immediate segmentation fault.

tools/perf/util/python.c static int pyrf_counts_values_set_values(struct pyrf_counts_values *self, PyObje · line 1923

Quoted 10 lines
tools/perf/util/python.c +5 -0
 	size = PyList_Size(list);
+	if (size != 5) {
+		PyErr_SetString(PyExc_ValueError, "List must have exactly 5 entries");
+		return -1;
+	}
+
 	for (Py_ssize_t i = 0; i < size; i++) {
 		item = PyList_GetItem(list, i);
 		if (!PyLong_Check(item)) {

[Severity: High]
This is a pre-existing issue, but can this loop lead to data truncation or crash the interpreter?

Just below this context in pyrf_counts_values_set_values():

		self->values.values[i] = PyLong_AsLong(item);
	}

Since self->values.values stores uint64_t data, assigning the result of PyLong_AsLong (which returns a signed C long) can truncate data on 32-bit platforms or produce incorrect negative values.

Additionally, if PyLong_AsLong raises an OverflowError for very large integers, the error is silently ignored and 0 is returned to the interpreter, which triggers a fatal SystemError crash.

The same underlying issue appears in other places where we expose uint64_t data:

tools/perf/util/python.c:pyrf_counts_values_get_values()

	for (int i = 0; i < 5; i++)
		PyList_SetItem(vals, i, PyLong_FromLong(self->values.values[i]));

tools/perf/util/python.c:pyrf_counts_values_members
counts_values_member_def(val, T_ULONG, "Value of event"),

Should we be using 64-bit functions like PyLong_AsUnsignedLongLong() and PyLong_FromUnsignedLongLong(), and ensuring exceptions are caught?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=8
author reviewer
#13 lore raw

PATCH 9/13 perf python: Validate CPU and thread maps in pyrf_evsel__open

In reply to Ian Rogers

Add explicit Py_TYPE checks to ensure the arguments passed are actually of the correct pyrf_thread_map and pyrf_cpu_map types.

Fixes 877108e42b1b ("perf tools: Initial python binding")
---
 tools/perf/util/python.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)
tools/perf/util/python.c +12 -2
index f9398bc30a80..62f404c40d8d 100644
@@ -2096,11 +2096,21 @@ static PyObject *pyrf_evsel__open(struct pyrf_evsel *pevsel,
 					 &pcpus, &pthreads, &group, &inherit))
 		return NULL;
 
-	if (pthreads != NULL && pthreads != Py_None)
+	if (pthreads != NULL && pthreads != Py_None) {
+		if (Py_TYPE(pthreads) != &pyrf_thread_map__type) {
+			PyErr_SetString(PyExc_TypeError, "threads must be a thread_map");
+			return NULL;
+		}
 		threads = ((struct pyrf_thread_map *)pthreads)->threads;
+	}
 
-	if (pcpus != NULL && pcpus != Py_None)
+	if (pcpus != NULL && pcpus != Py_None) {
+		if (Py_TYPE(pcpus) != &pyrf_cpu_map__type) {
+			PyErr_SetString(PyExc_TypeError, "cpus must be a cpu_map");
+			return NULL;
+		}
 		cpus = ((struct pyrf_cpu_map *)pcpus)->cpus;
+	}
 
 	evsel->core.attr.inherit = inherit;
 	/*
Signature · 1 line
2.55.0.rc0.786.g65d90a0328-goog
#14 lore raw

Re: [PATCH v1 09/13] perf python: Validate CPU and thread maps in pyrf_evsel__open

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Medium] The explicit `Py_TYPE()` checks added to `pyrf_evsel__open` break Python subclassing for…

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] The explicit Py_TYPE() checks added to pyrf_evsel__open break Python subclassing for perf.thread_map and perf.cpu_map objects. --

commit 909b91c5a1d04af18afad351b78395740e856d0b
Author: Ian Rogers <[email protected]>

perf python: Validate CPU and thread maps in pyrf_evsel__open

This commit adds explicit Py_TYPE checks to ensure the cpu and thread map arguments passed to pyrf_evsel__open are of the correct types.

Quoted 4 lines
tools/perf/util/python.c +0 -0
index f9398bc30a80b..62f404c40d8dc 100644

[ ... ]

tools/perf/util/python.c static PyObject *pyrf_evsel__open(struct pyrf_evsel *pevsel, · line 2100

Quoted 7 lines
tools/perf/util/python.c +2 -1
@@ -2096,11 +2096,21 @@ static PyObject *pyrf_evsel__open(struct pyrf_evsel *pevsel,
 					 &pcpus, &pthreads, &group, &inherit))
 		return NULL;
 
-	if (pthreads != NULL && pthreads != Py_None)
+	if (pthreads != NULL && pthreads != Py_None) {
+		if (Py_TYPE(pthreads) != &pyrf_thread_map__type) {

[Severity: Medium]
Will this strict equality check break valid Python subclasses of thread_map?

Because pyrf_thread_map__type sets Py_TPFLAGS_BASETYPE, Python scripts are permitted to subclass it, but Py_TYPE() will point to the subclass type and fail this exact-type check.

Should this use PyObject_TypeCheck() instead to safely support subclassing?

tools/perf/util/python.c static PyObject *pyrf_evsel__open(struct pyrf_evsel *pevsel, · line 2108

Quoted 9 lines
tools/perf/util/python.c +6 -1
+			PyErr_SetString(PyExc_TypeError, "threads must be a thread_map");
+			return NULL;
+		}
 		threads = ((struct pyrf_thread_map *)pthreads)->threads;
+	}
 
-	if (pcpus != NULL && pcpus != Py_None)
+	if (pcpus != NULL && pcpus != Py_None) {
+		if (Py_TYPE(pcpus) != &pyrf_cpu_map__type) {

[Severity: Medium]
Similarly, does this strict equality block valid Python subclasses of cpu_map?

Quoted 5 lines
tools/perf/util/python.c +4 -0
+			PyErr_SetString(PyExc_TypeError, "cpus must be a cpu_map");
+			return NULL;
+		}
 		cpus = ((struct pyrf_cpu_map *)pcpus)->cpus;
+	}
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=9
author reviewer
#15 lore raw

PATCH 10/13 perf python: Validate attribute setters in pyrf_evsel

In reply to Ian Rogers

If val is NULL when setting an attribute, PyErr_SetString should be called as deleting the attribute isn't supported. In addition, ensure PyErr_Occurred is checked before setting the attribute to avoid setting a garbage value.

Fixes 877108e42b1b ("perf tools: Initial python binding")
---
 tools/perf/util/python.c | 89 ++++++++++++++++++++++++++++++++++------
 1 file changed, 77 insertions(+), 12 deletions(-)
tools/perf/util/python.c +77 -12
index 62f404c40d8d..e277c89b8347 100644
@@ -2284,6 +2284,11 @@ static int pyrf_evsel__set_tracking(PyObject *self, PyObject *val, void *closure
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
 	is_true = PyObject_IsTrue(val);
 	if (is_true < 0)
 		return -1;
@@ -2295,11 +2300,21 @@ static int pyrf_evsel__set_tracking(PyObject *self, PyObject *val, void *closure
 static int pyrf_evsel__set_attr_config(PyObject *self, PyObject *val, void *closure __maybe_unused)
 {
 	struct pyrf_evsel *pevsel = (void *)self;
+	unsigned long long new_val;
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
-	pevsel->evsel->core.attr.config = PyLong_AsUnsignedLongLong(val);
-	return PyErr_Occurred() ? -1 : 0;
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
+	new_val = PyLong_AsUnsignedLongLong(val);
+	if (PyErr_Occurred())
+		return -1;
+
+	pevsel->evsel->core.attr.config = new_val;
+	return 0;
 }
 
 static PyObject *pyrf_evsel__get_attr_config(PyObject *self, void *closure __maybe_unused)
@@ -2314,11 +2329,21 @@ static PyObject *pyrf_evsel__get_attr_config(PyObject *self, void *closure __may
 static int pyrf_evsel__set_attr_read_format(PyObject *self, PyObject *val, void *closure __maybe_unused)
 {
 	struct pyrf_evsel *pevsel = (void *)self;
+	unsigned long long new_val;
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
-	pevsel->evsel->core.attr.read_format = PyLong_AsUnsignedLongLong(val);
-	return PyErr_Occurred() ? -1 : 0;
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
+	new_val = PyLong_AsUnsignedLongLong(val);
+	if (PyErr_Occurred())
+		return -1;
+
+	pevsel->evsel->core.attr.read_format = new_val;
+	return 0;
 }
 
 static PyObject *pyrf_evsel__get_attr_read_format(PyObject *self, void *closure __maybe_unused)
@@ -2333,11 +2358,21 @@ static PyObject *pyrf_evsel__get_attr_read_format(PyObject *self, void *closure
 static int pyrf_evsel__set_attr_sample_period(PyObject *self, PyObject *val, void *closure __maybe_unused)
 {
 	struct pyrf_evsel *pevsel = (void *)self;
+	unsigned long long new_val;
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
-	pevsel->evsel->core.attr.sample_period = PyLong_AsUnsignedLongLong(val);
-	return PyErr_Occurred() ? -1 : 0;
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
+	new_val = PyLong_AsUnsignedLongLong(val);
+	if (PyErr_Occurred())
+		return -1;
+
+	pevsel->evsel->core.attr.sample_period = new_val;
+	return 0;
 }
 
 static PyObject *pyrf_evsel__get_attr_sample_period(PyObject *self, void *closure __maybe_unused)
@@ -2352,11 +2387,21 @@ static PyObject *pyrf_evsel__get_attr_sample_period(PyObject *self, void *closur
 static int pyrf_evsel__set_attr_sample_type(PyObject *self, PyObject *val, void *closure __maybe_unused)
 {
 	struct pyrf_evsel *pevsel = (void *)self;
+	unsigned long long new_val;
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
-	pevsel->evsel->core.attr.sample_type = PyLong_AsUnsignedLongLong(val);
-	return PyErr_Occurred() ? -1 : 0;
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
+	new_val = PyLong_AsUnsignedLongLong(val);
+	if (PyErr_Occurred())
+		return -1;
+
+	pevsel->evsel->core.attr.sample_type = new_val;
+	return 0;
 }
 
 static PyObject *pyrf_evsel__get_attr_sample_type(PyObject *self, void *closure __maybe_unused)
@@ -2380,11 +2425,21 @@ static PyObject *pyrf_evsel__get_attr_size(PyObject *self, void *closure __maybe
 static int pyrf_evsel__set_attr_type(PyObject *self, PyObject *val, void *closure __maybe_unused)
 {
 	struct pyrf_evsel *pevsel = (void *)self;
+	unsigned long new_val;
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
-	pevsel->evsel->core.attr.type = PyLong_AsUnsignedLong(val);
-	return PyErr_Occurred() ? -1 : 0;
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
+	new_val = PyLong_AsUnsignedLong(val);
+	if (PyErr_Occurred())
+		return -1;
+
+	pevsel->evsel->core.attr.type = new_val;
+	return 0;
 }
 
 static PyObject *pyrf_evsel__get_attr_type(PyObject *self, void *closure __maybe_unused)
@@ -2399,11 +2454,21 @@ static PyObject *pyrf_evsel__get_attr_type(PyObject *self, void *closure __maybe
 static int pyrf_evsel__set_attr_wakeup_events(PyObject *self, PyObject *val, void *closure __maybe_unused)
 {
 	struct pyrf_evsel *pevsel = (void *)self;
+	unsigned long new_val;
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
-	pevsel->evsel->core.attr.wakeup_events = PyLong_AsUnsignedLong(val);
-	return PyErr_Occurred() ? -1 : 0;
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
+	new_val = PyLong_AsUnsignedLong(val);
+	if (PyErr_Occurred())
+		return -1;
+
+	pevsel->evsel->core.attr.wakeup_events = new_val;
+	return 0;
 }
 
 static PyObject *pyrf_evsel__get_attr_wakeup_events(PyObject *self, void *closure __maybe_unused)
Signature · 1 line
2.55.0.rc0.786.g65d90a0328-goog
author reviewer
#16 lore raw

PATCH 11/13 perf python: Zero initialize perf_data in pyrf_data__init

In reply to Ian Rogers

Replace path clearing with memset so the entire struct is zeroed, preventing uninitialized fields from causing errors later.

Fixes 4cd0142f7dec ("perf python: Add wrapper for perf_data file abstraction")
---
 tools/perf/util/python.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/python.c +1 -1
index e277c89b8347..03b4e8809107 100644
@@ -3607,7 +3607,7 @@ static int pyrf_data__init(struct pyrf_data *pdata, PyObject *args, PyObject *kw
 	if (pdata->data.open)
 		perf_data__close(&pdata->data);
 	free((char *)pdata->data.path);
-	pdata->data.path = NULL;
+	memset(&pdata->data, 0, sizeof(pdata->data));
 
 	if (fd != -1) {
 		struct stat st;
Signature · 1 line
2.55.0.rc0.786.g65d90a0328-goog
#17 lore raw

Re: [PATCH v1 11/13] perf python: Zero initialize perf_data in pyrf_data__init

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Medium] Missing explicit inclusion of POSIX standard header `<string.h>` for newly added `memset`. --

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] Missing explicit inclusion of POSIX standard header <string.h> for newly added memset. --

commit 84d9f7832c422ee6951987ed0cdd0d5717fcc7d2
Author: Ian Rogers <[email protected]>

perf python: Zero initialize perf_data in pyrf_data__init

This commit replaces clearing the path with memset to zero out the entire perf_data struct in pyrf_data__init.

tools/perf/util/python.c static int pyrf_data__init(struct pyrf_data *pdata, PyObject *args, PyObject *kw · line 3610

Quoted 10 lines
tools/perf/util/python.c +1 -1
index e277c89b83477..03b4e88091074 100644
@@ -3607,7 +3607,7 @@ static int pyrf_data__init(struct pyrf_data *pdata, PyObject *args, PyObject *kw
 	if (pdata->data.open)
 		perf_data__close(&pdata->data);
 	free((char *)pdata->data.path);
-	pdata->data.path = NULL;
+	memset(&pdata->data, 0, sizeof(pdata->data));

[Severity: Medium]
Will this cause a build failure on musl libc systems? It looks like memset() is added here but <string.h> is not explicitly included in tools/perf/util/python.c.

Musl libc enforces strict POSIX namespace separation and does not implicitly include headers. Could we add an explicit #include <string.h> at the top of the file to ensure the build succeeds in environments using musl?

Quoted 3 lines
tools/perf/util/python.c +0 -0
 	if (fd != -1) {
 		struct stat st;
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=11
author reviewer
#18 lore raw

PATCH 12/13 perf python: Add thread uninitialized checks

In reply to Ian Rogers

Add CHECK_INITIALIZED checks to the thread attribute getters (get_pid, get_tid, get_ppid) to prevent crashes if they are accessed before being properly initialized.

Fixes 3b96bf7af60d ("perf python: Add python session abstraction wrapping perf's session")
---
 tools/perf/util/python.c | 3 +++
 1 file changed, 3 insertions(+)
tools/perf/util/python.c +3 -0
index 03b4e8809107..70f985ad3dd2 100644
@@ -3743,16 +3743,19 @@ static PyMethodDef pyrf_thread__methods[] = {
 
 static PyObject *pyrf_thread__get_pid(struct pyrf_thread *pthread, void *closure __maybe_unused)
 {
+	CHECK_INITIALIZED(pthread->thread, "thread");
 	return PyLong_FromLong(thread__pid(pthread->thread));
 }
 
 static PyObject *pyrf_thread__get_tid(struct pyrf_thread *pthread, void *closure __maybe_unused)
 {
+	CHECK_INITIALIZED(pthread->thread, "thread");
 	return PyLong_FromLong(thread__tid(pthread->thread));
 }
 
 static PyObject *pyrf_thread__get_ppid(struct pyrf_thread *pthread, void *closure __maybe_unused)
 {
+	CHECK_INITIALIZED(pthread->thread, "thread");
 	return PyLong_FromLong(thread__ppid(pthread->thread));
 }
 
Signature · 1 line
2.55.0.rc0.786.g65d90a0328-goog
#19 lore raw

Re: [PATCH v1 12/13] perf python: Add thread uninitialized checks

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:

Pre-existing issues:
- [High] The patch incompletely fixes the missing initialization checks by omitting pyrf_thread__get_cpu, leaving a NULL pointer dereference vulnerability.
- [High] The perf.pmu type methods lack initialization guards, allowing uninitialized objects to cause NULL pointer dereferences.
- [Medium] The Python object count_values is leaked in multiple error paths within pyrf_evsel__read.
--

commit 95a2c7473298664fd72c7086443470ff19559438
Author: Ian Rogers <[email protected]>

perf python: Add thread uninitialized checks

The patch adds CHECK_INITIALIZED checks to the thread attribute getters (get_pid, get_tid, get_ppid) in the perf python module to prevent crashes when accessing uninitialized thread objects.

Quoted 14 lines
tools/perf/util/python.c +1 -0
index 03b4e88091074..70f985ad3dd20 100644
@@ -3743,16 +3743,19 @@ static PyMethodDef pyrf_thread__methods[] = {
 
 static PyObject *pyrf_thread__get_pid(struct pyrf_thread *pthread, void *closure __maybe_unused)
 {
+	CHECK_INITIALIZED(pthread->thread, "thread");
 	return PyLong_FromLong(thread__pid(pthread->thread));
 }
 
 static PyObject *pyrf_thread__get_tid(struct pyrf_thread *pthread, void *closure __maybe_unused)
 {

tools/perf/util/python.c static PyMethodDef pyrf_thread__methods[] = { · line 3761

Quoted 10 lines
tools/perf/util/python.c +2 -0
+	CHECK_INITIALIZED(pthread->thread, "thread");
 	return PyLong_FromLong(thread__tid(pthread->thread));
 }
 
 static PyObject *pyrf_thread__get_ppid(struct pyrf_thread *pthread, void *closure __maybe_unused)
 {
+	CHECK_INITIALIZED(pthread->thread, "thread");
 	return PyLong_FromLong(thread__ppid(pthread->thread));
 }
 

[Severity: High]
This isn't a bug introduced by this patch, but pyrf_thread__get_cpu() is missing a similar check:

tools/perf/util/python.c:pyrf_thread__get_cpu() {
    return PyLong_FromLong(thread__cpu(pthread->thread));
}

Could accessing the .cpu property on an uninitialized thread object cause a NULL pointer dereference? Does it make sense to add a CHECK_INITIALIZED guard here as well?

[Severity: High]
This isn't a bug introduced by this patch, but looking at the perf.pmu methods, they also appear to lack initialization checks:

tools/perf/util/python.c:pyrf_pmu__name() {
    return PyUnicode_FromString(ppmu->pmu->name);
}
tools/perf/util/python.c:pyrf_pmu__events() {
    ...
    ret = perf_pmu__for_each_event(ppmu->pmu,
                                   false,
                                   py_list,
                                   pyrf_pmu__events_cb);
    ...
}

Since perf.pmu uses PyType_GenericNew, could a script instantiate an uninitialized PMU and trigger a NULL pointer dereference when calling .name() or .events()?

[Severity: Medium]
This isn't a bug introduced by this patch, but I noticed a leak in pyrf_evsel__read() on its error paths:

tools/perf/util/python.c:pyrf_evsel__read() {
    count_values = PyObject_New(struct pyrf_counts_values,
                                &pyrf_counts_values__type);
    if (!count_values)
        return NULL;

    if (!PyArg_ParseTuple(args, "ii", &cpu, &thread))
        return NULL;

    cpu_idx = perf_cpu_map__idx(evsel->core.cpus, (struct perf_cpu){.cpu = cpu});
    if (cpu_idx < 0) {
        PyErr_Format(PyExc_TypeError, "CPU %d is not part of evsel's CPUs", cpu);
        return NULL;
    }
    ...
}

Does returning NULL directly on argument parsing or bounds checking failures leak the newly allocated count_values object? Should we call Py_DECREF(count_values) before returning?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=12
author reviewer
#20 lore raw

PATCH 13/13 perf python: Fix MetricGroup return type in perf.pyi

In reply to Ian Rogers

The metrics() function can return a dictionary where the value is either a string or a list of strings, so the type signature in the stub file should be Union[str, List[str]].

---
 tools/perf/python/perf.pyi | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tools/perf/python/perf.pyi +2 -2
index f266c9baffd0..58b0b3ed819d 100644
@@ -1,5 +1,5 @@
 """Type stubs for the perf Python module."""
-from typing import Callable, Dict, List, Optional, Any, Iterator
+from typing import Callable, Dict, List, Optional, Any, Iterator, Union
 
 def config_get(name: str) -> Optional[str]:
     """Get a configuration value from perf config.
@@ -12,7 +12,7 @@ def config_get(name: str) -> Optional[str]:
     """
     ...
 
-def metrics() -> List[Dict[str, str]]:
+def metrics() -> List[Dict[str, Union[str, List[str]]]]:
     """Get a list of available metrics.
 
     Returns:
Signature · 1 line
2.55.0.rc0.786.g65d90a0328-goog
#21 lore raw

Re: [PATCH v1 13/13] perf python: Fix MetricGroup return type in perf.pyi

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] This problem wasn't introduced by this patch, but local references to the Python objects 'key' and 'value' in pyrf__metrics_cb are leaked on the success path. --

commit 0b60b233b211850d4730f46ab4faf2da85366d5f
Author: Ian Rogers <[email protected]>

perf python: Fix MetricGroup return type in perf.pyi

The metrics() function can return a dictionary where the value is either a string or a list of strings, so the type signature in the stub file should be Union[str, List[str]].

Quoted 4 lines
tools/perf/python/perf.pyi +0 -0
index f266c9baffd01..58b0b3ed819d8 100644

[ ... ]

tools/perf/python/perf.pyi def config_get(name: str) -> Optional[str]: · line 15

Quoted 6 lines
tools/perf/python/perf.pyi +1 -1
@@ -12,7 +12,7 @@ def config_get(name: str) -> Optional[str]:
     """
...
 
-def metrics() -> List[Dict[str, str]]:
+def metrics() -> List[Dict[str, Union[str, List[str]]]]:

[Severity: High]
This isn't a bug introduced by this patch, but while reviewing the return type update for metrics(), I noticed a potential resource leak in its C implementation.

Looking at tools/perf/util/python.c:pyrf__metrics_cb():

	if (!value || PyDict_SetItem(dict, key, value) != 0) {
		Py_XDECREF(key);
		Py_XDECREF(value);
		Py_XDECREF(dict);
		return -ENOMEM;
	}

	if (!add_to_dict(dict, "MetricName", pm->metric_name) ||

Does this leak the local references to the Python objects 'key' and 'value' on the success path?

Since PyDict_SetItem() does not steal references, it appears that 'key' and 'value' might need a Py_DECREF() after a successful insertion. Without it, a script repeatedly calling perf.metrics() could continually leak memory.

Quoted 3 lines

"""Get a list of available metrics.

Returns:

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=13
author reviewer
#22 lore raw

v2 cover perf python: Fix python extension bugs from v19 review

In reply to Ian Rogers

Phase 1 of splitting the v19 review patches into smaller fixes. These patches address issues identified in the python extension during the review of the v19 script porting series, as well as fixing several pre-existing bugs and leaks.

v2:
- Addressed sashiko review feedback:
- Fixed 64-bit casting for counts_values by using T_ULONGLONG and PyLong_FromUnsignedLongLong.
- Replaced strict Py_TYPE checks with PyObject_TypeCheck for thread/cpu maps.
- Fixed a missing CHECK_INITIALIZED for pyrf_thread__get_cpu, pyrf_pmu__name, pyrf_pmu__events.
- Included <string.h> where memset was used.
- Fixed line splitting formatting on PyObject_New in pyrf_evsel__read.
- Added fixes for memory leaks and issues discovered during review:
- Fix memory leak of key and value in pyrf__metrics_cb
- Fix count_values memory leak in pyrf_evsel__read
- Fix sb_evlist leaks in both top and record paths.
- Fix potential divide by zero in perf_event__synthesize_threads

v1:
- Decomposed the single large "Fix python extension bugs" patch into 13 individual fixes.
- Added Fixes: tags for issues that date back to earlier commits.
- Addressed checkpatch warnings (line lengths, Signed-off-by, etc).

Ian Rogers (16):
perf script: Fix metric_evlist leak in script_find_metrics perf stat: Fix evsel_list leak in cmd_stat
perf tools: Fix sb_evlist leaks in top and record
perf python: Fix memory leak in pyrf_evlist__get_pollfd perf synthetic-events: Fix uninitialized pthread_join perf data: Fix directory file descriptor leak in perf_data__switch perf test: Fix skiplist leak in cmd_test
perf python: Check counts_values size in set_values perf python: Validate CPU and thread maps in pyrf_evsel__open perf python: Validate attribute setters in pyrf_evsel perf python: Zero initialize perf_data in pyrf_data__init perf python: Add thread and PMU uninitialized checks perf python: Fix MetricGroup return type in perf.pyi perf python: Fix count_values memory leak in pyrf_evsel__read perf python: Fix memory leak in pyrf__metrics_cb
perf synthetic-events: Fix divide by zero in
perf_event__synthesize_threads

 tools/perf/builtin-record.c        |   2 +
 tools/perf/builtin-script.c        |   2 +-
 tools/perf/builtin-stat.c          |   9 +-
 tools/perf/builtin-top.c           |   2 +
 tools/perf/python/perf.pyi         |   4 +-
 tools/perf/tests/builtin-test.c    |   5 +-
 tools/perf/util/data.c             |   3 +-
 tools/perf/util/python.c           | 160 ++++++++++++++++++++++++-----
 tools/perf/util/synthetic-events.c |   6 +-
 9 files changed, 153 insertions(+), 40 deletions(-)
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
author reviewer
#23 lore raw

PATCH v2 1/16 perf script: Fix metric_evlist leak in script_find_metrics

Fix a memory leak in script_find_metrics() where metric_evlist is leaked when returning early on error paths.

Assisted-by Antigravity:gemini-3.1-pro
Fixes 3622990efaab ("perf script: Change metric format to use json metrics")
---
 tools/perf/builtin-script.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/builtin-script.c +1 -1
index 47afd8cdc2b7..6b69e982ba62 100644
@@ -2275,8 +2275,8 @@ static int script_find_metrics(const struct pmu_metric *pm,
 	}
 	pr_debug("Found metric '%s' whose evsels match those of in the perf data\n",
 		 pm->metric_name);
-	evlist__put(metric_evlist);
 out:
+	evlist__put(metric_evlist);
 	return 0;
 }
 
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
author reviewer
#24 lore raw

PATCH v2 2/16 perf stat: Fix evsel_list leak in cmd_stat

In reply to Ian Rogers

Fix a memory leak in cmd_stat() where evsel_list is leaked if an error occurs while opening the output file.

Assisted-by Antigravity:gemini-3.1-pro
Fixes 361c99a661a7 ("perf evsel: Introduce perf_evlist")
---
 tools/perf/builtin-stat.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
tools/perf/builtin-stat.c +6 -3
index 3f897b2e8638..7cb6ea937e87 100644
@@ -2718,7 +2718,8 @@ int cmd_stat(int argc, const char **argv)
 		output = fopen(output_name, mode);
 		if (!output) {
 			perror("failed to create output file");
-			return -1;
+			status = -1;
+			goto out;
 		}
 		if (!stat_config.json_output) {
 			clock_gettime(CLOCK_REALTIME, &tm);
@@ -2729,7 +2730,8 @@ int cmd_stat(int argc, const char **argv)
 		output = fdopen(output_fd, mode);
 		if (!output) {
 			perror("Failed opening logfd");
-			return -errno;
+			status = -errno;
+			goto out;
 		}
 	}
 
@@ -2738,7 +2740,8 @@ int cmd_stat(int argc, const char **argv)
 		parse_options_usage(stat_usage, stat_options, "o", 1);
 		parse_options_usage(NULL, stat_options, "log-fd", 0);
 		parse_options_usage(NULL, stat_options, "interval-clear", 0);
-		return -1;
+		status = -1;
+		goto out;
 	}
 
 	stat_config.output = output;
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
author reviewer
#25 lore raw

PATCH v2 3/16 perf tools: Fix sb_evlist leaks in top and record

In reply to Ian Rogers

Fix a memory leak in cmd_top() where top.sb_evlist was not freed if evlist__add_bpf_sb_event() fails. Note that evlist__start_sb_thread() and evlist__stop_sb_thread() take ownership of the evlist and free it, so we must only free it if we fail before starting the thread. Also set top.sb_evlist to NULL to prevent use-after-free bugs.

Apply the same fix to builtin-record.c to avoid leaking rec->sb_evlist and calling pthread_join on uninitialized threads in the error path.

Assisted-by Antigravity:gemini-3.1-pro
Fixes b38d85ef49cf ("perf bpf: Decouple creating the evlist from adding the SB event")
---
 tools/perf/builtin-record.c | 2 ++
 tools/perf/builtin-top.c    | 2 ++
 2 files changed, 4 insertions(+)
tools/perf/builtin-record.c +2 -0
index ebd3ed0c9b3e..b106d50f6d97 100644
@@ -2247,6 +2247,8 @@ static int record__setup_sb_evlist(struct record *rec)
 
 		if (evlist__add_bpf_sb_event(rec->sb_evlist, perf_session__env(rec->session))) {
 			pr_err("Couldn't ask for PERF_RECORD_BPF_EVENT side band events.\n.");
+			evlist__put(rec->sb_evlist);
+			rec->sb_evlist = NULL;
 			return -1;
 		}
 	}
tools/perf/builtin-top.c +2 -0
index 5933c46ee137..570410599f1b 100644
@@ -1881,6 +1881,8 @@ int cmd_top(int argc, const char **argv)
 		if (evlist__add_bpf_sb_event(top.sb_evlist, &host_env)) {
 			pr_err("Couldn't ask for PERF_RECORD_BPF_EVENT side band events.\n.");
 			status = -EINVAL;
+			evlist__put(top.sb_evlist);
+			top.sb_evlist = NULL;
 			goto out_put_evlist;
 		}
 	}
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
author reviewer
#26 lore raw

PATCH v2 3/16 perf top: Fix sb_evlist leak

In reply to Ian Rogers

Fix a memory leak in cmd_top() where top.sb_evlist was not freed if evlist__add_bpf_sb_event() fails. Note that evlist__start_sb_thread() and evlist__stop_sb_thread() take ownership of the evlist and free it, so we must only free it if we fail before starting the thread.

Assisted-by Antigravity:gemini-3.1-pro
Fixes b38d85ef49cf ("perf bpf: Decouple creating the evlist from adding the SB event")
---
 tools/perf/builtin-top.c | 1 +
 1 file changed, 1 insertion(+)
tools/perf/builtin-top.c +1 -0
index 5933c46ee137..abc74b8ab759 100644
@@ -1881,6 +1881,7 @@ int cmd_top(int argc, const char **argv)
 		if (evlist__add_bpf_sb_event(top.sb_evlist, &host_env)) {
 			pr_err("Couldn't ask for PERF_RECORD_BPF_EVENT side band events.\n.");
 			status = -EINVAL;
+			evlist__put(top.sb_evlist);
 			goto out_put_evlist;
 		}
 	}
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
maintainer
#27 lore raw

Re: [PATCH v2 03/16] perf top: Fix sb_evlist leak

Quoted 8 lines · Ian Rogers

Fix a memory leak in cmd_top() where top.sb_evlist was not freed if evlist__add_bpf_sb_event() fails. Note that evlist__start_sb_thread() and evlist__stop_sb_thread() take ownership of the evlist and free it, so we must only free it if we fail before starting the thread.

Assisted-by Antigravity:gemini-3.1-pro
Fixes b38d85ef49cf ("perf bpf: Decouple creating the evlist from adding the SB event")

Looks like a leftover from v1. There are two 03/16 patches. :)

Thanks,
Namhyung

author reviewer
#28 lore raw

PATCH v2 4/16 perf python: Fix memory leak in pyrf_evlist__get_pollfd

In reply to Ian Rogers

Fix a Python list object leak in pyrf_evlist__get_pollfd() by adding a missing Py_DECREF on the error exit path.

Assisted-by Antigravity:gemini-3.1-pro
Fixes 877108e42b1b ("perf tools: Initial python binding")
---
 tools/perf/util/python.c | 3 +++
 1 file changed, 3 insertions(+)
tools/perf/util/python.c +3 -0
index 11e1f39c1bea..550f1e7d1449 100644
@@ -2825,6 +2825,8 @@ static PyObject *pyrf_evlist__get_pollfd(struct pyrf_evlist *pevlist,
 
 	evlist = pevlist->evlist;
 	list = PyList_New(0);
+	if (!list)
+		return NULL;
 
 	for (i = 0; i < evlist__core(evlist)->pollfd.nr; ++i) {
 		PyObject *file;
@@ -2843,6 +2845,7 @@ static PyObject *pyrf_evlist__get_pollfd(struct pyrf_evlist *pevlist,
 
 	return list;
 free_list:
+	Py_XDECREF(list);
 	return PyErr_NoMemory();
 }
 
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
#29 lore raw

Re: [PATCH v2 04/16] perf python: Fix memory leak in pyrf_evlist__get_pollfd

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [Low] Missing explicit POSIX libc header inclusions for <stdlib.h> and <stdio.h>. --

commit b37f20df0f07ef7454731749a918a55a91b0c7db
Author: Ian Rogers <[email protected]>

perf python: Fix memory leak in pyrf_evlist__get_pollfd

This commit fixes a Python list object leak in pyrf_evlist__get_pollfd() by adding a missing Py_DECREF on the error exit path.

tools/perf/util/python.c

Quoted 4 lines
tools/perf/util/python.c +0 -0
index 11e1f39c1bea9..550f1e7d14499 100644

[Severity: Low]
This isn't a bug introduced by this patch, but there is a pre-existing issue where the top-level includes section is missing explicit POSIX libc headers:

#include <inttypes.h>
#include <string.h>

#include <linux/err.h>

The file relies on POSIX standard functions like calloc() and free() which require <stdlib.h>, as well as asprintf() which requires <stdio.h>. Could these be explicitly added to ensure musl compilation compatibility?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=4
author reviewer
#30 lore raw

PATCH v2 5/16 perf synthetic-events: Fix uninitialized pthread_join

In reply to Ian Rogers

In perf_event__synthesize_threads(), fix an uninitialized pthread_join() call when thread creation fails by only joining the successfully created threads.

Assisted-by Antigravity:gemini-3.1-pro
Fixes 340b47f510bb ("perf top: Implement multithreading for perf_event__synthesize_threads")
---
 tools/perf/util/synthetic-events.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tools/perf/util/synthetic-events.c +2 -2
index b75f9dcf4dbf..35de41ad14d3 100644
@@ -1104,8 +1104,8 @@ int perf_event__synthesize_threads(const struct perf_tool *tool,
 	}
 	err = 0;
 out_join:
-	for (i = 0; i < thread_nr; i++)
-		pthread_join(synthesize_threads[i], NULL);
+	for (j = 0; j < i; j++)
+		pthread_join(synthesize_threads[j], NULL);
 	free(args);
 free_threads:
 	free(synthesize_threads);
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
#31 lore raw

Re: [PATCH v2 05/16] perf synthetic-events: Fix uninitialized pthread_join

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [Low] Missing explicit POSIX inclusions for <stdlib.h> and <pthread.h> in synthetic-events.c. --

commit 9ead28d76c34623188955779076a3c283fe73884
Author: Ian Rogers <[email protected]>

perf synthetic-events: Fix uninitialized pthread_join

This commit fixes an uninitialized pthread_join() call in perf_event__synthesize_threads() when thread creation fails. It updates the cleanup loop to only join the threads that were successfully created.

tools/perf/util/synthetic-events.c int perf_event__synthesize_threads(const struct perf_tool *tool, · line 1108

Quoted 11 lines
tools/perf/util/synthetic-events.c +2 -2
@@ -1104,8 +1104,8 @@ int perf_event__synthesize_threads(const struct perf_tool *tool,
 	}
 	err = 0;
 out_join:
-	for (i = 0; i < thread_nr; i++)
-		pthread_join(synthesize_threads[i], NULL);
+	for (j = 0; j < i; j++)
+		pthread_join(synthesize_threads[j], NULL);

[Severity: Low]
This is a pre-existing issue, but should this file have explicit POSIX header inclusions for <stdlib.h> and <pthread.h> to ensure compatibility with musl libc? The code here relies on pthread_join() and free(), but appears to depend on implicit header inclusions to resolve them.

Quoted 3 lines
tools/perf/util/synthetic-events.c +0 -0
 	free(args);
 free_threads:
 	free(synthesize_threads);
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=5
author reviewer
#32 lore raw

PATCH v2 6/16 perf data: Fix directory file descriptor leak in perf_data__switch

In reply to Ian Rogers

Fix a file descriptor leak in perf_data__switch() when running in directory mode. By substituting perf_data__close() for perf_data_file__close(), any thread-local directory files are properly closed during file rotation.

Assisted-by Antigravity:gemini-3.1-pro
Fixes 040f9915e99e ("perf data: Add perf_data_file__switch() helper")
---
 tools/perf/util/data.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
tools/perf/util/data.c +1 -2
index 33fd1b82019e..32d4d908ad38 100644
@@ -472,8 +472,7 @@ int perf_data__switch(struct perf_data *data,
 		pr_warning("Failed to rename %s to %s\n", data->path, *new_filepath);
 
 	if (!at_exit) {
-		perf_data_file__close(&data->file);
-		data->open = false;
+		perf_data__close(data);
 		ret = perf_data__open(data);
 		if (ret < 0)
 			goto out;
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
#33 lore raw

Re: [PATCH v2 06/16] perf data: Fix directory file descriptor leak in perf_data__switch

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Critical] The patch introduces a Use-After-Free during file rotation in directory mode. By changing…

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Critical] The patch introduces a Use-After-Free during file rotation in directory mode. By changing perf_data__switch() to call perf_data__close(), the data->dir.files array is freed. However, thread mmaps (map->file) still retain pointers to this freed array and are never updated. Subsequent writes dereference the freed memory. --

commit d39defbf0b046135d95bd01181c56fe40f6a6430
Author: Ian Rogers <[email protected]>

perf data: Fix directory file descriptor leak in perf_data__switch

This commit fixes a file descriptor leak in perf_data__switch() when running in directory mode. It replaces perf_data_file__close() with perf_data__close() to properly close thread-local directory files during file rotation.

tools/perf/util/data.c int perf_data__switch(struct perf_data *data, · line 475

Quoted 10 lines
tools/perf/util/data.c +1 -2
@@ -472,8 +472,7 @@ int perf_data__switch(struct perf_data *data,
 		pr_warning("Failed to rename %s to %s\n", data->path, *new_filepath);
 
 	if (!at_exit) {
-		perf_data_file__close(&data->file);
-		data->open = false;
+		perf_data__close(data);

[Severity: Critical]
Could this cause a use-after-free regression during file rotation?

Calling perf_data__close(data) here frees the data->dir.files array. However, thread mmaps (like map->file) appear to still retain pointers to this freed memory and are not updated.

For example, in tools/perf/builtin-record.c:record__write():

    if (perf_data_file__write(file, bf, size) < 0) {

The file pointer here is aliased from map->file, which now points to the freed memory. If someone runs perf record --switch-output with threads enabled, the subsequent writes dereference these dangling pointers.

Quoted 3 lines
tools/perf/util/data.c +0 -0
 		ret = perf_data__open(data);
 		if (ret < 0)
 			goto out;
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=6
author reviewer
#34 lore raw

PATCH v2 7/16 perf test: Fix skiplist leak in cmd_test

In reply to Ian Rogers

Fix a memory leak in cmd_test() where skiplist was not freed on exit paths.

Assisted-by Antigravity:gemini-3.1-pro
Fixes 2ae828786c65 ("perf test: Allow skipping tests")
---
 tools/perf/tests/builtin-test.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
tools/perf/tests/builtin-test.c +4 -1
index 7e75f590f225..cf5625cf4f11 100644
@@ -1781,9 +1781,12 @@ int cmd_test(int argc, const char **argv)
 	rlimit__bump_memlock();
 
 	suites = build_suites();
-	if (!suites)
+	if (!suites) {
+		intlist__delete(skiplist);
 		return errno ? -errno : -ENOMEM;
+	}
 	ret = __cmd_test(suites, argc, argv, skiplist);
 	free(suites);
+	intlist__delete(skiplist);
 	return ret;
 }
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
#35 lore raw

Re: [PATCH v2 07/16] perf test: Fix skiplist leak in cmd_test

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Medium] Unsafe error handling: `errno` may be clobbered by cleanup functions before being evaluated. --

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] Unsafe error handling: errno may be clobbered by cleanup functions before being evaluated. --

commit c1a33a510193529b900f698c55a9892c973edeb2
Author: Ian Rogers <[email protected]>

perf test: Fix skiplist leak in cmd_test

This commit fixes a memory leak in cmd_test() by ensuring the skiplist is properly freed on the exit paths.

Quoted 4 lines
tools/perf/tests/builtin-test.c +0 -0
index 7e75f590f225e..cf5625cf4f110 100644

[ ... ]

tools/perf/tests/builtin-test.c int cmd_test(int argc, const char **argv) · line 1786

Quoted 8 lines
tools/perf/tests/builtin-test.c +2 -1
@@ -1781,9 +1781,12 @@ int cmd_test(int argc, const char **argv)
 	rlimit__bump_memlock();
 
 	suites = build_suites();
-	if (!suites)
+	if (!suites) {
+		intlist__delete(skiplist);
 		return errno ? -errno : -ENOMEM;

[Severity: Medium]
Could the call to intlist__delete() clobber errno before it is evaluated in the return statement?

intlist__delete() cascades to rblist__delete(), which ultimately issues free() calls. On some libc implementations or under memory sanitizers, free() and related cleanup functions do not guarantee that errno is preserved. This could result in returning an incorrect error code from build_suites().

Quoted 6 lines
tools/perf/tests/builtin-test.c +2 -0
+	}
 	ret = __cmd_test(suites, argc, argv, skiplist);
 	free(suites);
+	intlist__delete(skiplist);
 	return ret;
 }
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=7
author reviewer
#36 lore raw

PATCH v2 8/16 perf python: Check counts_values size in set_values

In reply to Ian Rogers

The set_values function incorrectly assumed the list contained exactly 5 elements. Add a check to prevent out-of-bounds access.

Fixes 877108e42b1b ("perf tools: Initial python binding")
---
 tools/perf/util/python.c | 33 +++++++++++++++++++++++++--------
 1 file changed, 25 insertions(+), 8 deletions(-)
tools/perf/util/python.c +25 -8
index 550f1e7d1449..c6c729808a0f 100644
@@ -1881,11 +1881,11 @@ static void pyrf_counts_values__delete(struct pyrf_counts_values *pcounts_values
 	  0, help }
 
 static PyMemberDef pyrf_counts_values_members[] = {
-	counts_values_member_def(val, T_ULONG, "Value of event"),
-	counts_values_member_def(ena, T_ULONG, "Time for which enabled"),
-	counts_values_member_def(run, T_ULONG, "Time for which running"),
-	counts_values_member_def(id, T_ULONG, "Unique ID for an event"),
-	counts_values_member_def(lost, T_ULONG, "Num of lost samples"),
+	counts_values_member_def(val, T_ULONGLONG, "Value of event"),
+	counts_values_member_def(ena, T_ULONGLONG, "Time for which enabled"),
+	counts_values_member_def(run, T_ULONGLONG, "Time for which running"),
+	counts_values_member_def(id, T_ULONGLONG, "Unique ID for an event"),
+	counts_values_member_def(lost, T_ULONGLONG, "Num of lost samples"),
 	{ .name = NULL, },
 };
 
@@ -1895,8 +1895,15 @@ static PyObject *pyrf_counts_values_get_values(struct pyrf_counts_values *self,
 
 	if (!vals)
 		return NULL;
-	for (int i = 0; i < 5; i++)
-		PyList_SetItem(vals, i, PyLong_FromLong(self->values.values[i]));
+	for (int i = 0; i < 5; i++) {
+		PyObject *val = PyLong_FromUnsignedLongLong(self->values.values[i]);
+
+		if (!val) {
+			Py_DECREF(vals);
+			return NULL;
+		}
+		PyList_SetItem(vals, i, val);
+	}
 
 	return vals;
 }
@@ -1913,13 +1920,23 @@ static int pyrf_counts_values_set_values(struct pyrf_counts_values *self, PyObje
 	}
 
 	size = PyList_Size(list);
+	if (size != 5) {
+		PyErr_SetString(PyExc_ValueError, "List must have exactly 5 entries");
+		return -1;
+	}
+
 	for (Py_ssize_t i = 0; i < size; i++) {
+		unsigned long long val;
+
 		item = PyList_GetItem(list, i);
 		if (!PyLong_Check(item)) {
 			PyErr_SetString(PyExc_TypeError, "List members should be numbers");
 			return -1;
 		}
-		self->values.values[i] = PyLong_AsLong(item);
+		val = PyLong_AsUnsignedLongLong(item);
+		if (val == (unsigned long long)-1 && PyErr_Occurred())
+			return -1;
+		self->values.values[i] = val;
 	}
 
 	return 0;
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
#37 lore raw

Re: [PATCH v2 08/16] perf python: Check counts_values size in set_values

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Pre-existing issues:
- [High] Deleting the values attribute from a perf.counts_values object in Python causes a segmentation fault.
- [Medium] The function tracepoint_field truncates 64-bit tracepoint fields on 32-bit platforms by casting them to unsigned long or long. --

commit d69dc2f4b8c6d442757f85756430758b037c2521
Author: Ian Rogers <[email protected]>

perf python: Check counts_values size in set_values

The set_values function incorrectly assumed the list contained exactly 5 elements. Add a check to prevent out-of-bounds access.

Quoted 4 lines
tools/perf/util/python.c +0 -0
index 550f1e7d14499..c6c729808a0f2 100644

[ ... ]

tools/perf/util/python.c static int pyrf_counts_values_set_values(struct pyrf_counts_values *self, PyObje · line 1922

Quoted 4 lines
tools/perf/util/python.c +0 -0
@@ -1913,13 +1920,23 @@ static int pyrf_counts_values_set_values(struct pyrf_counts_values *self, PyObje
 	}
 
 	size = PyList_Size(list);

[Severity: High]
This isn't a bug introduced by this patch, but does pyrf_counts_values_set_values() crash if the list parameter is NULL?

If a user deletes the attribute in Python (e.g., del counts_values.values), the list parameter will be passed as NULL. The PyList_Check(list) macro called just before this line unconditionally dereferences the pointer, which would trigger a segmentation fault.

tools/perf/util/python.c static int pyrf_counts_values_set_values(struct pyrf_counts_values *self, PyObje · line 1928

Quoted 6 lines
tools/perf/util/python.c +5 -0
+	if (size != 5) {
+		PyErr_SetString(PyExc_ValueError, "List must have exactly 5 entries");
+		return -1;
+	}
+
 	for (Py_ssize_t i = 0; i < size; i++) {

[ ... ]

[Severity: Medium]
This is also a pre-existing issue, but could tracepoint_field() truncate 64-bit values on 32-bit platforms?

In tools/perf/util/python.c:tracepoint_field():

		val = tep_read_number(pevent, data + field->offset,
				      field->size);
		if (field->flags & TEP_FIELD_IS_POINTER)
			ret = PyLong_FromUnsignedLong((unsigned long) val);
		else if (field->flags & TEP_FIELD_IS_SIGNED)
			ret = PyLong_FromLong((long) val);
		else
			ret = PyLong_FromUnsignedLong((unsigned long) val);

Since val is returned by tep_read_number(), which provides a 64-bit value, casting it to unsigned long or long drops the top 32 bits on 32-bit architectures. Does this lead to silent data corruption for 64-bit tracepoint fields like timestamps or counters?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=8
author reviewer
#38 lore raw

PATCH v2 9/16 perf python: Validate CPU and thread maps in pyrf_evsel__open

In reply to Ian Rogers

Add explicit Py_TYPE checks to ensure the arguments passed are actually of the correct pyrf_thread_map and pyrf_cpu_map types.

Fixes 877108e42b1b ("perf tools: Initial python binding")
---
 tools/perf/util/python.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)
tools/perf/util/python.c +12 -2
index c6c729808a0f..4bf37e2afb18 100644
@@ -2108,11 +2108,21 @@ static PyObject *pyrf_evsel__open(struct pyrf_evsel *pevsel,
 					 &pcpus, &pthreads, &group, &inherit))
 		return NULL;
 
-	if (pthreads != NULL && pthreads != Py_None)
+	if (pthreads != NULL && pthreads != Py_None) {
+		if (!PyObject_TypeCheck(pthreads, &pyrf_thread_map__type)) {
+			PyErr_SetString(PyExc_TypeError, "threads must be a thread_map");
+			return NULL;
+		}
 		threads = ((struct pyrf_thread_map *)pthreads)->threads;
+	}
 
-	if (pcpus != NULL && pcpus != Py_None)
+	if (pcpus != NULL && pcpus != Py_None) {
+		if (!PyObject_TypeCheck(pcpus, &pyrf_cpu_map__type)) {
+			PyErr_SetString(PyExc_TypeError, "cpus must be a cpu_map");
+			return NULL;
+		}
 		cpus = ((struct pyrf_cpu_map *)pcpus)->cpus;
+	}
 
 	evsel->core.attr.inherit = inherit;
 	/*
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
author reviewer
#39 lore raw

PATCH v2 10/16 perf python: Validate attribute setters in pyrf_evsel

In reply to Ian Rogers

If val is NULL when setting an attribute, PyErr_SetString should be called as deleting the attribute isn't supported. In addition, ensure PyErr_Occurred is checked before setting the attribute to avoid setting a garbage value.

Fixes 877108e42b1b ("perf tools: Initial python binding")
---
 tools/perf/util/python.c | 89 ++++++++++++++++++++++++++++++++++------
 1 file changed, 77 insertions(+), 12 deletions(-)
tools/perf/util/python.c +77 -12
index 4bf37e2afb18..657b054f5e51 100644
@@ -2296,6 +2296,11 @@ static int pyrf_evsel__set_tracking(PyObject *self, PyObject *val, void *closure
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
 	is_true = PyObject_IsTrue(val);
 	if (is_true < 0)
 		return -1;
@@ -2307,11 +2312,21 @@ static int pyrf_evsel__set_tracking(PyObject *self, PyObject *val, void *closure
 static int pyrf_evsel__set_attr_config(PyObject *self, PyObject *val, void *closure __maybe_unused)
 {
 	struct pyrf_evsel *pevsel = (void *)self;
+	unsigned long long new_val;
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
-	pevsel->evsel->core.attr.config = PyLong_AsUnsignedLongLong(val);
-	return PyErr_Occurred() ? -1 : 0;
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
+	new_val = PyLong_AsUnsignedLongLong(val);
+	if (PyErr_Occurred())
+		return -1;
+
+	pevsel->evsel->core.attr.config = new_val;
+	return 0;
 }
 
 static PyObject *pyrf_evsel__get_attr_config(PyObject *self, void *closure __maybe_unused)
@@ -2326,11 +2341,21 @@ static PyObject *pyrf_evsel__get_attr_config(PyObject *self, void *closure __may
 static int pyrf_evsel__set_attr_read_format(PyObject *self, PyObject *val, void *closure __maybe_unused)
 {
 	struct pyrf_evsel *pevsel = (void *)self;
+	unsigned long long new_val;
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
-	pevsel->evsel->core.attr.read_format = PyLong_AsUnsignedLongLong(val);
-	return PyErr_Occurred() ? -1 : 0;
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
+	new_val = PyLong_AsUnsignedLongLong(val);
+	if (PyErr_Occurred())
+		return -1;
+
+	pevsel->evsel->core.attr.read_format = new_val;
+	return 0;
 }
 
 static PyObject *pyrf_evsel__get_attr_read_format(PyObject *self, void *closure __maybe_unused)
@@ -2345,11 +2370,21 @@ static PyObject *pyrf_evsel__get_attr_read_format(PyObject *self, void *closure
 static int pyrf_evsel__set_attr_sample_period(PyObject *self, PyObject *val, void *closure __maybe_unused)
 {
 	struct pyrf_evsel *pevsel = (void *)self;
+	unsigned long long new_val;
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
-	pevsel->evsel->core.attr.sample_period = PyLong_AsUnsignedLongLong(val);
-	return PyErr_Occurred() ? -1 : 0;
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
+	new_val = PyLong_AsUnsignedLongLong(val);
+	if (PyErr_Occurred())
+		return -1;
+
+	pevsel->evsel->core.attr.sample_period = new_val;
+	return 0;
 }
 
 static PyObject *pyrf_evsel__get_attr_sample_period(PyObject *self, void *closure __maybe_unused)
@@ -2364,11 +2399,21 @@ static PyObject *pyrf_evsel__get_attr_sample_period(PyObject *self, void *closur
 static int pyrf_evsel__set_attr_sample_type(PyObject *self, PyObject *val, void *closure __maybe_unused)
 {
 	struct pyrf_evsel *pevsel = (void *)self;
+	unsigned long long new_val;
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
-	pevsel->evsel->core.attr.sample_type = PyLong_AsUnsignedLongLong(val);
-	return PyErr_Occurred() ? -1 : 0;
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
+	new_val = PyLong_AsUnsignedLongLong(val);
+	if (PyErr_Occurred())
+		return -1;
+
+	pevsel->evsel->core.attr.sample_type = new_val;
+	return 0;
 }
 
 static PyObject *pyrf_evsel__get_attr_sample_type(PyObject *self, void *closure __maybe_unused)
@@ -2392,11 +2437,21 @@ static PyObject *pyrf_evsel__get_attr_size(PyObject *self, void *closure __maybe
 static int pyrf_evsel__set_attr_type(PyObject *self, PyObject *val, void *closure __maybe_unused)
 {
 	struct pyrf_evsel *pevsel = (void *)self;
+	unsigned long new_val;
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
-	pevsel->evsel->core.attr.type = PyLong_AsUnsignedLong(val);
-	return PyErr_Occurred() ? -1 : 0;
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
+	new_val = PyLong_AsUnsignedLong(val);
+	if (PyErr_Occurred())
+		return -1;
+
+	pevsel->evsel->core.attr.type = new_val;
+	return 0;
 }
 
 static PyObject *pyrf_evsel__get_attr_type(PyObject *self, void *closure __maybe_unused)
@@ -2411,11 +2466,21 @@ static PyObject *pyrf_evsel__get_attr_type(PyObject *self, void *closure __maybe
 static int pyrf_evsel__set_attr_wakeup_events(PyObject *self, PyObject *val, void *closure __maybe_unused)
 {
 	struct pyrf_evsel *pevsel = (void *)self;
+	unsigned long new_val;
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
-	pevsel->evsel->core.attr.wakeup_events = PyLong_AsUnsignedLong(val);
-	return PyErr_Occurred() ? -1 : 0;
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
+	new_val = PyLong_AsUnsignedLong(val);
+	if (PyErr_Occurred())
+		return -1;
+
+	pevsel->evsel->core.attr.wakeup_events = new_val;
+	return 0;
 }
 
 static PyObject *pyrf_evsel__get_attr_wakeup_events(PyObject *self, void *closure __maybe_unused)
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
#40 lore raw

Re: [PATCH v2 10/16] perf python: Validate attribute setters in pyrf_evsel

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [Medium] Missing NULL check for attribute deletion in pyrf_counts_values_set_values causes a segmentation fault. --

commit 921733bf92d5e6634715543526c3d25b8346f3c3
Author: Ian Rogers <[email protected]>

perf python: Validate attribute setters in pyrf_evsel

This commit adds NULL checks to attribute setters in pyrf_evsel to ensure an error is correctly reported instead of setting a garbage value when attempting to delete an attribute.

Quoted 4 lines
tools/perf/util/python.c +0 -0
index 4bf37e2afb18d..657b054f5e515 100644

[ ... ]

tools/perf/util/python.c static int pyrf_evsel__set_tracking(PyObject *self, PyObject *val, void *closure · line 2306

Quoted 12 lines
tools/perf/util/python.c +5 -0
@@ -2296,6 +2296,11 @@ static int pyrf_evsel__set_tracking(PyObject *self, PyObject *val, void *closure
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
 	is_true = PyObject_IsTrue(val);
 	if (is_true < 0)
 		return -1;

[Severity: Medium]
This is a pre-existing issue, but since you are fixing these attribute deletion issues in pyrf_evsel, should we also fix pyrf_counts_values_set_values in the same file?

Looking at pyrf_counts_values_set_values, the list parameter is passed directly to PyList_Check without verifying if it is NULL:

static int pyrf_counts_values_set_values(struct pyrf_counts_values *self,
PyObject *list, void *closure)

{
	Py_ssize_t size;
	PyObject *item = NULL;

	if (!PyList_Check(list)) {
		PyErr_SetString(PyExc_TypeError, "Value assigned must be a list");
		return -1;
	}

Could deleting the counts_values.values attribute from a Python script trigger a segmentation fault here?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=10
author reviewer
#41 lore raw

PATCH v2 11/16 perf python: Zero initialize perf_data in pyrf_data__init

In reply to Ian Rogers

Replace path clearing with memset so the entire struct is zeroed, preventing uninitialized fields from causing errors later.

Fixes 4cd0142f7dec ("perf python: Add wrapper for perf_data file abstraction")
---
 tools/perf/util/python.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
tools/perf/util/python.c +2 -1
index 657b054f5e51..71be130b38c5 100644
@@ -3,6 +3,7 @@
 #include <Python.h>
 
 #include <inttypes.h>
+#include <string.h>
 
 #include <linux/err.h>
 #include <poll.h>
@@ -3621,7 +3622,7 @@ static int pyrf_data__init(struct pyrf_data *pdata, PyObject *args, PyObject *kw
 	if (pdata->data.open)
 		perf_data__close(&pdata->data);
 	free((char *)pdata->data.path);
-	pdata->data.path = NULL;
+	memset(&pdata->data, 0, sizeof(pdata->data));
 
 	if (fd != -1) {
 		struct stat st;
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
author reviewer
#42 lore raw

PATCH v2 12/16 perf python: Add thread and PMU uninitialized checks

In reply to Ian Rogers

Add CHECK_INITIALIZED checks to the thread attribute getters (get_pid, get_tid, get_ppid) to prevent crashes if they are accessed before being properly initialized.

Fixes 3b96bf7af60d ("perf python: Add python session abstraction wrapping perf's session")
---
 tools/perf/util/python.c | 7 +++++++
 1 file changed, 7 insertions(+)
tools/perf/util/python.c +7 -0
index 71be130b38c5..b17b191b1625 100644
@@ -1678,6 +1678,7 @@ static PyObject *pyrf_pmu__name(PyObject *self)
 {
 	struct pyrf_pmu *ppmu = (void *)self;
 
+	CHECK_INITIALIZED(ppmu->pmu, "pmu");
 	return PyUnicode_FromString(ppmu->pmu->name);
 }
 
@@ -1730,6 +1731,8 @@ static PyObject *pyrf_pmu__events(PyObject *self)
 	PyObject *py_list = PyList_New(0);
 	int ret;
 
+	CHECK_INITIALIZED(ppmu->pmu, "pmu");
+
 	if (!py_list)
 		return NULL;
 
@@ -3758,21 +3761,25 @@ static PyMethodDef pyrf_thread__methods[] = {
 
 static PyObject *pyrf_thread__get_pid(struct pyrf_thread *pthread, void *closure __maybe_unused)
 {
+	CHECK_INITIALIZED(pthread->thread, "thread");
 	return PyLong_FromLong(thread__pid(pthread->thread));
 }
 
 static PyObject *pyrf_thread__get_tid(struct pyrf_thread *pthread, void *closure __maybe_unused)
 {
+	CHECK_INITIALIZED(pthread->thread, "thread");
 	return PyLong_FromLong(thread__tid(pthread->thread));
 }
 
 static PyObject *pyrf_thread__get_ppid(struct pyrf_thread *pthread, void *closure __maybe_unused)
 {
+	CHECK_INITIALIZED(pthread->thread, "thread");
 	return PyLong_FromLong(thread__ppid(pthread->thread));
 }
 
 static PyObject *pyrf_thread__get_cpu(struct pyrf_thread *pthread, void *closure __maybe_unused)
 {
+	CHECK_INITIALIZED(pthread->thread, "thread");
 	return PyLong_FromLong(thread__cpu(pthread->thread));
 }
 
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
#43 lore raw

Re: [PATCH v2 12/16] perf python: Add thread and PMU uninitialized checks

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Medium] The newly added CHECK_INITIALIZED macro in pyrf_pmu__events() leaks the py_list object when the…

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] The newly added CHECK_INITIALIZED macro in pyrf_pmu__events() leaks the py_list object when the pmu is not initialized. --

commit 579d96b72f035d8f31919edd0270e2cfa16975d3
Author: Ian Rogers <[email protected]>

perf python: Add thread and PMU uninitialized checks

Add CHECK_INITIALIZED checks to the thread attribute getters (get_pid, get_tid, get_ppid) to prevent crashes if they are accessed before being properly initialized.

Quoted 4 lines
tools/perf/util/python.c +0 -0
index 71be130b38c53..b17b191b16255 100644

[ ... ]

tools/perf/util/python.c static PyObject *pyrf_pmu__events(PyObject *self) · line 1735

Quoted 6 lines
tools/perf/util/python.c +2 -0
@@ -1730,6 +1731,8 @@ static PyObject *pyrf_pmu__events(PyObject *self)
 	PyObject *py_list = PyList_New(0);
 	int ret;
 
+	CHECK_INITIALIZED(ppmu->pmu, "pmu");
+

[Severity: Medium]
If ppmu->pmu is uninitialized, won't this newly added check return early and leak the py_list object allocated just above it?

Would it make sense to move the CHECK_INITIALIZED() macro before the PyList_New(0) allocation?

Quoted 2 lines
tools/perf/util/python.c +0 -0
 	if (!py_list)
 		return NULL;
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=12
author reviewer
#44 lore raw

PATCH v2 13/16 perf python: Fix MetricGroup return type in perf.pyi

In reply to Ian Rogers

The metrics() function can return a dictionary where the value is either a string or a list of strings, so the type signature in the stub file should be Union[str, List[str]].

---
 tools/perf/python/perf.pyi | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tools/perf/python/perf.pyi +2 -2
index f266c9baffd0..58b0b3ed819d 100644
@@ -1,5 +1,5 @@
 """Type stubs for the perf Python module."""
-from typing import Callable, Dict, List, Optional, Any, Iterator
+from typing import Callable, Dict, List, Optional, Any, Iterator, Union
 
 def config_get(name: str) -> Optional[str]:
     """Get a configuration value from perf config.
@@ -12,7 +12,7 @@ def config_get(name: str) -> Optional[str]:
     """
     ...
 
-def metrics() -> List[Dict[str, str]]:
+def metrics() -> List[Dict[str, Union[str, List[str]]]]:
     """Get a list of available metrics.
 
     Returns:
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
author reviewer
#45 lore raw

PATCH v2 14/16 perf python: Fix count_values memory leak in pyrf_evsel__read

In reply to Ian Rogers

In pyrf_evsel__read, if PyArg_ParseTuple fails, the allocated count_values is leaked. Move the allocation of count_values after the PyArg_ParseTuple call to prevent the memory leak.

---
 tools/perf/util/python.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)
tools/perf/util/python.c +4 -5
index b17b191b1625..e76a407cb2bb 100644
@@ -2204,11 +2204,6 @@ static PyObject *pyrf_evsel__read(struct pyrf_evsel *pevsel,
 
 	CHECK_INITIALIZED(evsel, "evsel");
 
-	count_values = PyObject_New(struct pyrf_counts_values,
-							       &pyrf_counts_values__type);
-	if (!count_values)
-		return NULL;
-
 	if (!PyArg_ParseTuple(args, "ii", &cpu, &thread))
 		return NULL;
 
@@ -2227,6 +2222,10 @@ static PyObject *pyrf_evsel__read(struct pyrf_evsel *pevsel,
 	if (evsel__ensure_counts(evsel))
 		return PyErr_NoMemory();
 
+	count_values = PyObject_New(struct pyrf_counts_values, &pyrf_counts_values__type);
+	if (!count_values)
+		return NULL;
+
 	/* Set up pointers to the old and newly read counter values. */
 	old_count = perf_counts(evsel->prev_raw_counts, cpu_idx, thread_idx);
 	new_count = perf_counts(evsel->counts, cpu_idx, thread_idx);
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
#46 lore raw

Re: [PATCH v2 14/16] perf python: Fix count_values memory leak in pyrf_evsel__read

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:

Pre-existing issues:
- [High] A partially failed allocation in evsel__ensure_counts() results in a NULL pointer dereference when retrying pyrf_evsel__read().
- [High] The evsel->prev_raw_counts array allocated for Python perf.evsel objects is unconditionally leaked when the object is destroyed.
- [Medium] String duplication leak in prepare_metric on error paths.
--

commit 0bdfb7eb04b252f781d1daa5947d46bec1d9df82
Author: Ian Rogers <[email protected]>

perf python: Fix count_values memory leak in pyrf_evsel__read

In pyrf_evsel__read, if PyArg_ParseTuple fails, the allocated count_values is leaked. Move the allocation of count_values after the PyArg_ParseTuple call to prevent the memory leak.

Quoted 17 lines
tools/perf/util/python.c +0 -5
index b17b191b16255..e76a407cb2bb9 100644
@@ -2204,11 +2204,6 @@ static PyObject *pyrf_evsel__read(struct pyrf_evsel *pevsel,
 
 	CHECK_INITIALIZED(evsel, "evsel");
 
-	count_values = PyObject_New(struct pyrf_counts_values,
-							       &pyrf_counts_values__type);
-	if (!count_values)
-		return NULL;
-
 	if (!PyArg_ParseTuple(args, "ii", &cpu, &thread))
 		return NULL;
 

tools/perf/util/python.c static PyObject *pyrf_evsel__read(struct pyrf_evsel *pevsel, · line 2230

Quoted 9 lines
tools/perf/util/python.c +4 -0
@@ -2227,6 +2222,10 @@ static PyObject *pyrf_evsel__read(struct pyrf_evsel *pevsel,
 	if (evsel__ensure_counts(evsel))
 		return PyErr_NoMemory();
 
+	count_values = PyObject_New(struct pyrf_counts_values, &pyrf_counts_values__type);
+	if (!count_values)
+		return NULL;
+
 	/* Set up pointers to the old and newly read counter values. */
 	old_count = perf_counts(evsel->prev_raw_counts, cpu_idx, thread_idx);

[Severity: High]
This isn't a bug introduced by this patch, but I noticed a potential NULL pointer dereference here.

If evsel__ensure_counts() successfully allocates evsel->counts but fails to allocate evsel->prev_raw_counts, it returns -ENOMEM. If a user script handles the resulting Python MemoryError and retries the read operation, evsel__ensure_counts() will see that evsel->counts is no longer NULL and return success immediately, without allocating evsel->prev_raw_counts.

Will this cause perf_counts() to dereference a NULL prev_raw_counts pointer on the retry?

[Severity: High]
This is also a pre-existing issue, but does evsel->prev_raw_counts leak when the object is destroyed?

When a standalone Python evsel object is destroyed, its destruction path goes through pyrf_evsel__delete() -> evsel__put() -> evsel__exit(). While evsel__exit() properly calls evsel__free_counts() to free evsel->counts, it seems to miss freeing evsel->prev_raw_counts.

Could this unconditionally drain memory if Python evsel objects are used in a loop?

tools/perf/util/python.c static PyObject *pyrf_evsel__read(struct pyrf_evsel *pevsel, · line 2231

Quoted 1 line
tools/perf/util/python.c +0 -0
 	new_count = perf_counts(evsel->counts, cpu_idx, thread_idx);

[ ... ]

[Severity: Medium]
This is another pre-existing issue not caused by this patch, but looking at prepare_metric() in the same file, does this error path leak the duplicated string n?

tools/perf/util/python.c:prepare_metric() {
    ...
    char *n = strdup(evsel__metric_id(cur));

    if (!n)
        return -ENOMEM;
    ...
    ret = evsel__ensure_counts(cur);
    if (ret)
        return ret;
    ...
    ret = expr__add_id_val_source_count(pctx, n, val, source_count);
    if (ret)
        return ret;
    ...
}

If evsel__ensure_counts() or expr__add_id_val_source_count() returns an error, the function returns early without freeing n. Should we ensure n is freed on these error paths?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=14
author reviewer
#47 lore raw

PATCH v2 15/16 perf python: Fix memory leak in pyrf__metrics_cb

In reply to Ian Rogers

In pyrf__metrics_cb, PyDict_SetItem does not steal the reference of the key and value, so they need to be decref'ed after successful insertion to avoid memory leaks.

---
 tools/perf/util/python.c | 2 ++
 1 file changed, 2 insertions(+)
tools/perf/util/python.c +2 -0
index e76a407cb2bb..5471037d04c5 100644
@@ -3567,6 +3567,8 @@ static int pyrf__metrics_cb(const struct pmu_metric *pm,
 		Py_XDECREF(dict);
 		return -ENOMEM;
 	}
+	Py_DECREF(key);
+	Py_DECREF(value);
 
 	if (!add_to_dict(dict, "MetricName", pm->metric_name) ||
 	    !add_to_dict(dict, "PMU", pm->pmu) ||
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
author reviewer
#48 lore raw

PATCH v2 16/16 perf synthetic-events: Fix divide by zero in perf_event__synthesize_threads

In reply to Ian Rogers

If scandir() finds no matching tasks in /proc, n is 0. If thread_nr is > 1, we bypass the single-thread fast path and then clamp thread_nr to n, making it 0. This results in a divide by zero when calculating num_per_thread.

Handle n <= 1 early to use the single-thread fast path and prevent the crash.

---
 tools/perf/util/synthetic-events.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/synthetic-events.c +1 -1
index 35de41ad14d3..19c1f3eaf58d 100644
@@ -1058,7 +1058,7 @@ int perf_event__synthesize_threads(const struct perf_tool *tool,
 	else
 		thread_nr = nr_threads_synthesize;
 
-	if (thread_nr <= 1) {
+	if (thread_nr <= 1 || n <= 1) {
 		err = __perf_event__synthesize_threads(tool, process,
 						       machine,
 						       needs_mmap, mmap_data,
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
maintainer
#49 lore raw

Re: [PATCH v2 00/16] perf python: Fix python extension bugs from v19 review

In reply to Ian Rogers

Hi Ian,

Quoted 34 lines · Ian Rogers

Phase 1 of splitting the v19 review patches into smaller fixes. These patches address issues identified in the python extension during the review of the v19 script porting series, as well as fixing several pre-existing bugs and leaks.

v2:
- Addressed sashiko review feedback:
- Fixed 64-bit casting for counts_values by using T_ULONGLONG and PyLong_FromUnsignedLongLong.
- Replaced strict Py_TYPE checks with PyObject_TypeCheck for thread/cpu maps.
- Fixed a missing CHECK_INITIALIZED for pyrf_thread__get_cpu, pyrf_pmu__name, pyrf_pmu__events.
- Included <string.h> where memset was used.
- Fixed line splitting formatting on PyObject_New in pyrf_evsel__read.
- Added fixes for memory leaks and issues discovered during review:
- Fix memory leak of key and value in pyrf__metrics_cb
- Fix count_values memory leak in pyrf_evsel__read
- Fix sb_evlist leaks in both top and record paths.
- Fix potential divide by zero in perf_event__synthesize_threads

v1:
- Decomposed the single large "Fix python extension bugs" patch into 13 individual fixes.
- Added Fixes: tags for issues that date back to earlier commits.
- Addressed checkpatch warnings (line lengths, Signed-off-by, etc).

Ian Rogers (16):
perf script: Fix metric_evlist leak in script_find_metrics perf stat: Fix evsel_list leak in cmd_stat
perf tools: Fix sb_evlist leaks in top and record
perf python: Fix memory leak in pyrf_evlist__get_pollfd perf synthetic-events: Fix uninitialized pthread_join perf data: Fix directory file descriptor leak in perf_data__switch perf test: Fix skiplist leak in cmd_test
perf python: Check counts_values size in set_values perf python: Validate CPU and thread maps in pyrf_evsel__open

Quoted 8 lines

perf python: Validate attribute setters in pyrf_evsel perf python: Zero initialize perf_data in pyrf_data__init perf python: Add thread and PMU uninitialized checks perf python: Fix MetricGroup return type in perf.pyi perf python: Fix count_values memory leak in pyrf_evsel__read perf python: Fix memory leak in pyrf__metrics_cb perf synthetic-events: Fix divide by zero in
perf_event__synthesize_threads

Will you send v3? You don't need to fix all pre-existing issues but we should not introduce new bugs or regressions. I think some of Sashiko comments found them like data switch and pmu init checks.

Thanks,
Namhyung

Quoted 15 lines
 tools/perf/builtin-record.c        |   2 +
 tools/perf/builtin-script.c        |   2 +-
 tools/perf/builtin-stat.c          |   9 +-
 tools/perf/builtin-top.c           |   2 +
 tools/perf/python/perf.pyi         |   4 +-
 tools/perf/tests/builtin-test.c    |   5 +-
 tools/perf/util/data.c             |   3 +-
 tools/perf/util/python.c           | 160 ++++++++++++++++++++++++-----
 tools/perf/util/synthetic-events.c |   6 +-
 9 files changed, 153 insertions(+), 40 deletions(-)
Signature · 1 line
2.55.0.rc0.799.gd6f94ed593-goog
author reviewer
#50 lore raw

Re: [PATCH v2 00/16] perf python: Fix python extension bugs from v19 review

Quoted 47 lines · Namhyung Kim

Hi Ian,

Quoted 34 lines · Ian Rogers

Phase 1 of splitting the v19 review patches into smaller fixes. These patches address issues identified in the python extension during the review of the v19 script porting series, as well as fixing several pre-existing bugs and leaks.

v2:
- Addressed sashiko review feedback:
- Fixed 64-bit casting for counts_values by using T_ULONGLONG and PyLong_FromUnsignedLongLong.
- Replaced strict Py_TYPE checks with PyObject_TypeCheck for thread/cpu maps.
- Fixed a missing CHECK_INITIALIZED for pyrf_thread__get_cpu, pyrf_pmu__name, pyrf_pmu__events.
- Included <string.h> where memset was used.
- Fixed line splitting formatting on PyObject_New in pyrf_evsel__read.
- Added fixes for memory leaks and issues discovered during review:
- Fix memory leak of key and value in pyrf__metrics_cb
- Fix count_values memory leak in pyrf_evsel__read
- Fix sb_evlist leaks in both top and record paths.
- Fix potential divide by zero in perf_event__synthesize_threads

v1:
- Decomposed the single large "Fix python extension bugs" patch into 13 individual fixes.
- Added Fixes: tags for issues that date back to earlier commits.
- Addressed checkpatch warnings (line lengths, Signed-off-by, etc).

Ian Rogers (16):
perf script: Fix metric_evlist leak in script_find_metrics perf stat: Fix evsel_list leak in cmd_stat
perf tools: Fix sb_evlist leaks in top and record
perf python: Fix memory leak in pyrf_evlist__get_pollfd perf synthetic-events: Fix uninitialized pthread_join perf data: Fix directory file descriptor leak in perf_data__switch perf test: Fix skiplist leak in cmd_test
perf python: Check counts_values size in set_values perf python: Validate CPU and thread maps in pyrf_evsel__open

Quoted 8 lines

perf python: Validate attribute setters in pyrf_evsel perf python: Zero initialize perf_data in pyrf_data__init perf python: Add thread and PMU uninitialized checks perf python: Fix MetricGroup return type in perf.pyi perf python: Fix count_values memory leak in pyrf_evsel__read perf python: Fix memory leak in pyrf__metrics_cb perf synthetic-events: Fix divide by zero in
perf_event__synthesize_threads

Quoted 3 lines

Will you send v3? You don't need to fix all pre-existing issues but we should not introduce new bugs or regressions. I think some of Sashiko comments found them like data switch and pmu init checks.

Thanks Namhyung, I have a v3 coming. I'm working with Sashiko locally and trying to address as many issues as possible, including minor pre-existing ones—it's easier than remembering to clean them up later.

Ian

Quoted 18 lines

Thanks,
Namhyung

Quoted 15 lines
 tools/perf/builtin-record.c        |   2 +
 tools/perf/builtin-script.c        |   2 +-
 tools/perf/builtin-stat.c          |   9 +-
 tools/perf/builtin-top.c           |   2 +
 tools/perf/python/perf.pyi         |   4 +-
 tools/perf/tests/builtin-test.c    |   5 +-
 tools/perf/util/data.c             |   3 +-
 tools/perf/util/python.c           | 160 ++++++++++++++++++++++++-----
 tools/perf/util/synthetic-events.c |   6 +-
 9 files changed, 153 insertions(+), 40 deletions(-)

--
2.55.0.rc0.799.gd6f94ed593-goog

maintainer
#51 lore raw

Re: [PATCH v2 00/16] perf python: Fix python extension bugs from v19 review

Quoted 48 lines · Ian Rogers
Quoted 47 lines · Namhyung Kim

Hi Ian,

Quoted 42 lines · Ian Rogers

Phase 1 of splitting the v19 review patches into smaller fixes. These patches address issues identified in the python extension during the review of the v19 script porting series, as well as fixing several pre-existing bugs and leaks.

v2:
- Addressed sashiko review feedback:
- Fixed 64-bit casting for counts_values by using T_ULONGLONG and PyLong_FromUnsignedLongLong.
- Replaced strict Py_TYPE checks with PyObject_TypeCheck for thread/cpu maps.
- Fixed a missing CHECK_INITIALIZED for pyrf_thread__get_cpu, pyrf_pmu__name, pyrf_pmu__events.
- Included <string.h> where memset was used.
- Fixed line splitting formatting on PyObject_New in pyrf_evsel__read.
- Added fixes for memory leaks and issues discovered during review:
- Fix memory leak of key and value in pyrf__metrics_cb
- Fix count_values memory leak in pyrf_evsel__read
- Fix sb_evlist leaks in both top and record paths.
- Fix potential divide by zero in perf_event__synthesize_threads

v1:
- Decomposed the single large "Fix python extension bugs" patch into 13 individual fixes.
- Added Fixes: tags for issues that date back to earlier commits.
- Addressed checkpatch warnings (line lengths, Signed-off-by, etc).

Ian Rogers (16):
perf script: Fix metric_evlist leak in script_find_metrics perf stat: Fix evsel_list leak in cmd_stat
perf tools: Fix sb_evlist leaks in top and record
perf python: Fix memory leak in pyrf_evlist__get_pollfd perf synthetic-events: Fix uninitialized pthread_join perf data: Fix directory file descriptor leak in perf_data__switch perf test: Fix skiplist leak in cmd_test
perf python: Check counts_values size in set_values perf python: Validate CPU and thread maps in pyrf_evsel__open perf python: Validate attribute setters in pyrf_evsel perf python: Zero initialize perf_data in pyrf_data__init perf python: Add thread and PMU uninitialized checks perf python: Fix MetricGroup return type in perf.pyi perf python: Fix count_values memory leak in pyrf_evsel__read perf python: Fix memory leak in pyrf__metrics_cb
perf synthetic-events: Fix divide by zero in
perf_event__synthesize_threads

Quoted 7 lines
Quoted 3 lines

Will you send v3? You don't need to fix all pre-existing issues but we should not introduce new bugs or regressions. I think some of Sashiko comments found them like data switch and pmu init checks.

Thanks Namhyung, I have a v3 coming. I'm working with Sashiko locally and trying to address as many issues as possible, including minor pre-existing ones—it's easier than remembering to clean them up later.

Sounds good. Looking forward to v3.

Thanks,
Namhyung

author reviewer
#52 lore raw

v3 cover perf python: Fix python extension bugs from v19 review

In reply to Ian Rogers

Phase 1 of splitting the v19 review patches into smaller fixes. These patches address issues identified in the python extension during the review of the v19 script porting series, as well as fixing several pre-existing bugs and leaks.

v3:
- Addressed review feedback from Namhyung Kim and Sashiko AI:
- Dropped "perf data: Fix directory file descriptor leak in perf_data__switch" to avoid Use-After-Free during directory mode file rotation.
- Added list == NULL check in pyrf_counts_values_set_values to safely handle attribute deletion (del obj.values).
- Preserved errno before calling intlist__delete(skiplist) in cmd_test error path to prevent clobbering the return error code.
- Added CHECK_INITIALIZED to pyrf_pmu__repr to prevent NULL pointer dereferences on uninitialized PMU objects.
- Fixed duplicate patch title from v2.

v2:
- Addressed sashiko review feedback:
- Fixed 64-bit casting for counts_values by using T_ULONGLONG and PyLong_FromUnsignedLongLong.
- Replaced strict Py_TYPE checks with PyObject_TypeCheck for thread/cpu maps.
- Fixed a missing CHECK_INITIALIZED for pyrf_thread__get_cpu, pyrf_pmu__name, pyrf_pmu__events.
- Included <string.h> where memset was used.
- Fixed line splitting formatting on PyObject_New in pyrf_evsel__read.
- Added fixes for memory leaks and issues discovered during review:
- Fix memory leak of key and value in pyrf__metrics_cb
- Fix count_values memory leak in pyrf_evsel__read
- Fix sb_evlist leaks in both top and record paths.
- Fix potential divide by zero in perf_event__synthesize_threads

v1:
- Decomposed the single large "Fix python extension bugs" patch into 13 individual fixes.
- Added Fixes: tags for issues that date back to earlier commits.
- Addressed checkpatch warnings (line lengths, Signed-off-by, etc).

Ian Rogers (15):
perf script: Fix metric_evlist leak in script_find_metrics perf stat: Fix evsel_list leak in cmd_stat
perf tools: Fix sb_evlist leaks in top and record perf python: Fix memory leak in pyrf_evlist__get_pollfd perf synthetic-events: Fix uninitialized pthread_join perf test: Fix skiplist leak in cmd_test
perf python: Check counts_values size in set_values perf python: Validate CPU and thread maps in pyrf_evsel__open perf python: Validate attribute setters in pyrf_evsel perf python: Zero initialize perf_data in pyrf_data__init perf python: Add thread and PMU uninitialized checks perf python: Fix MetricGroup return type in perf.pyi perf python: Fix count_values memory leak in pyrf_evsel__read perf python: Fix memory leak in pyrf__metrics_cb perf synthetic-events: Fix divide by zero in
perf_event__synthesize_threads

 tools/perf/builtin-record.c        |   2 +
 tools/perf/builtin-script.c        |   2 +-
 tools/perf/builtin-stat.c          |   9 +-
 tools/perf/builtin-top.c           |   2 +
 tools/perf/python/perf.pyi         |   4 +-
 tools/perf/tests/builtin-test.c    |   9 +-
 tools/perf/util/python.c           | 166 ++++++++++++++++++++++++-----
 tools/perf/util/synthetic-events.c |   6 +-
 8 files changed, 161 insertions(+), 39 deletions(-)
Signature · 1 line
2.55.0.679.g6767b8d81c-goog
author reviewer
#53 lore raw

PATCH v3 1/15 perf script: Fix metric_evlist leak in script_find_metrics

Fix a memory leak in script_find_metrics() where metric_evlist is leaked when returning early on error paths.

Assisted-by Antigravity:gemini-3.1-pro
Fixes 3622990efaab ("perf script: Change metric format to use json metrics")
---
 tools/perf/builtin-script.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/builtin-script.c +1 -1
index f91d8b1fbd01..ad8ca08ceb5f 100644
@@ -2276,8 +2276,8 @@ static int script_find_metrics(const struct pmu_metric *pm,
 	}
 	pr_debug("Found metric '%s' whose evsels match those of in the perf data\n",
 		 pm->metric_name);
-	evlist__put(metric_evlist);
 out:
+	evlist__put(metric_evlist);
 	return 0;
 }
 
Signature · 1 line
2.55.0.679.g6767b8d81c-goog
#54 lore raw

Re: [PATCH v3 01/15] perf script: Fix metric_evlist leak in script_find_metrics

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] Prematurely setting metric_evsel->metric_id = NULL breaks subsequent metric event lookups, null-terminating the metric_events array and dropping events. --

--- Patch [1]: [PATCH v3 01/15] perf script: Fix metric_evlist leak in script_find_metrics --- commit 5fba9e29ce353d01c2f25b3b58a05d2fc79eb299
Author: Ian Rogers <[email protected]>

perf script: Fix metric_evlist leak in script_find_metrics

Fix a memory leak in script_find_metrics() where metric_evlist is leaked when returning early on error paths.

Quoted 3 lines
tools/perf/builtin-script.c +0 -0

[ ... ]

Quoted 4 lines
	evlist__for_each_entry(metric_evlist, metric_evsel) {
		struct evsel *script_evsel =
			map_metric_evsel_to_script_evsel(script_evlist, metric_evsel);
		struct metric_event *metric_me =
Quoted 8 lines
			metricgroup__lookup(evlist__metric_events(metric_evlist),
					    metric_evsel,
					    /*create=*/false);

		if (script_evsel->metric_id == NULL) {
			script_evsel->metric_id = metric_evsel->metric_id;
			metric_evsel->metric_id = NULL;
		}

[Severity: High]
This is a pre-existing issue, but could clearing metric_evsel->metric_id here interfere with mapping resolutions later in this loop?

[ ... ]

Quoted 10 lines
			list_splice_init(&metric_me->head, &script_me->head);
			list_for_each_entry(expr, &script_me->head, nd) {
				for (int i = 0; expr->metric_events[i]; i++) {
					expr->metric_events[i] =
						map_metric_evsel_to_script_evsel(script_evlist,
									expr->metric_events[i]);
				}
			}
		}
	}

When map_metric_evsel_to_script_evsel() evaluates the required events, would a NULL metric_id from the earlier transfer cause evsel__metric_id() to return unknown?

It looks like this might fail the ID comparison against the script event and return NULL, which would prematurely terminate the expr->metric_events array and cause events to be silently dropped.

[ ... ]

Quoted 5 lines
tools/perf/builtin-script.c +1 -1
-	evlist__put(metric_evlist);
 out:
+	evlist__put(metric_evlist);
 	return 0;
 }
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
author reviewer
#55 lore raw

PATCH v3 2/15 perf stat: Fix evsel_list leak in cmd_stat

In reply to Ian Rogers

Fix a memory leak in cmd_stat() where evsel_list is leaked if an error occurs while opening the output file.

Assisted-by Antigravity:gemini-3.1-pro
Fixes 361c99a661a7 ("perf evsel: Introduce perf_evlist")
---
 tools/perf/builtin-stat.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
tools/perf/builtin-stat.c +6 -3
index d68a2617a920..c50cf230f16e 100644
@@ -2735,7 +2735,8 @@ int cmd_stat(int argc, const char **argv)
 		output = fopen(output_name, mode);
 		if (!output) {
 			perror("failed to create output file");
-			return -1;
+			status = -1;
+			goto out;
 		}
 		if (!stat_config.json_output) {
 			clock_gettime(CLOCK_REALTIME, &tm);
@@ -2746,7 +2747,8 @@ int cmd_stat(int argc, const char **argv)
 		output = fdopen(output_fd, mode);
 		if (!output) {
 			perror("Failed opening logfd");
-			return -errno;
+			status = -errno;
+			goto out;
 		}
 	}
 
@@ -2755,7 +2757,8 @@ int cmd_stat(int argc, const char **argv)
 		parse_options_usage(stat_usage, stat_options, "o", 1);
 		parse_options_usage(NULL, stat_options, "log-fd", 0);
 		parse_options_usage(NULL, stat_options, "interval-clear", 0);
-		return -1;
+		status = -1;
+		goto out;
 	}
 
 	stat_config.output = output;
Signature · 1 line
2.55.0.679.g6767b8d81c-goog
author reviewer
#56 lore raw

PATCH v3 3/15 perf tools: Fix sb_evlist leaks in top and record

In reply to Ian Rogers

Fix a memory leak in cmd_top() where top.sb_evlist was not freed if evlist__add_bpf_sb_event() fails. Note that evlist__start_sb_thread() and evlist__stop_sb_thread() take ownership of the evlist and free it, so we must only free it if we fail before starting the thread. Also set top.sb_evlist to NULL to prevent use-after-free bugs.

Apply the same fix to builtin-record.c to avoid leaking rec->sb_evlist and calling pthread_join on uninitialized threads in the error path.

Assisted-by Antigravity:gemini-3.1-pro
Fixes b38d85ef49cf ("perf bpf: Decouple creating the evlist from adding the SB event")
---
 tools/perf/builtin-record.c | 2 ++
 tools/perf/builtin-top.c    | 2 ++
 2 files changed, 4 insertions(+)
tools/perf/builtin-record.c +2 -0
index f58d7e3c7879..a57987851cf0 100644
@@ -2258,6 +2258,8 @@ static int record__setup_sb_evlist(struct record *rec)
 
 		if (evlist__add_bpf_sb_event(rec->sb_evlist, perf_session__env(rec->session))) {
 			pr_err("Couldn't ask for PERF_RECORD_BPF_EVENT side band events.\n.");
+			evlist__put(rec->sb_evlist);
+			rec->sb_evlist = NULL;
 			return -1;
 		}
 	}
tools/perf/builtin-top.c +2 -0
index 5933c46ee137..570410599f1b 100644
@@ -1881,6 +1881,8 @@ int cmd_top(int argc, const char **argv)
 		if (evlist__add_bpf_sb_event(top.sb_evlist, &host_env)) {
 			pr_err("Couldn't ask for PERF_RECORD_BPF_EVENT side band events.\n.");
 			status = -EINVAL;
+			evlist__put(top.sb_evlist);
+			top.sb_evlist = NULL;
 			goto out_put_evlist;
 		}
 	}
Signature · 1 line
2.55.0.679.g6767b8d81c-goog
author reviewer
#57 lore raw

PATCH v3 4/15 perf python: Fix memory leak in pyrf_evlist__get_pollfd

In reply to Ian Rogers

Fix a Python list object leak in pyrf_evlist__get_pollfd() by adding a missing Py_DECREF on the error exit path.

Assisted-by Antigravity:gemini-3.1-pro
Fixes 877108e42b1b ("perf tools: Initial python binding")
---
 tools/perf/util/python.c | 3 +++
 1 file changed, 3 insertions(+)
tools/perf/util/python.c +3 -0
index af85c0771c44..c7ae1f85aec6 100644
@@ -2825,6 +2825,8 @@ static PyObject *pyrf_evlist__get_pollfd(struct pyrf_evlist *pevlist,
 
 	evlist = pevlist->evlist;
 	list = PyList_New(0);
+	if (!list)
+		return NULL;
 
 	for (i = 0; i < evlist__core(evlist)->pollfd.nr; ++i) {
 		PyObject *file;
@@ -2843,6 +2845,7 @@ static PyObject *pyrf_evlist__get_pollfd(struct pyrf_evlist *pevlist,
 
 	return list;
 free_list:
+	Py_XDECREF(list);
 	return PyErr_NoMemory();
 }
 
Signature · 1 line
2.55.0.679.g6767b8d81c-goog
#58 lore raw

Re: [PATCH v3 04/15] perf python: Fix memory leak in pyrf_evlist__get_pollfd

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Pre-existing issues:
- [High] This problem wasn't introduced by this patch, but missing NULL check on the return value of PyLong_FromLong in pyrf_counts_values_get_values leads to a crash.
- [High] This problem wasn't introduced by this patch, but there is a memory leak of key and value objects on the success path of PyDict_SetItem in pyrf__metrics_cb. --

--- Patch [4]: [PATCH v3 04/15] perf python: Fix memory leak in pyrf_evlist__get_pollfd --- commit 4d7f49a1892b79c6131283be18ebf7cc1b0fd24f
Author: Ian Rogers <[email protected]>

perf python: Fix memory leak in pyrf_evlist__get_pollfd

This patch fixes a Python list object leak in pyrf_evlist__get_pollfd() by adding a missing Py_DECREF on an error exit path. It also adds a check for list allocation failure.

Assisted-by Antigravity:gemini-3.1-pro
Fixes 877108e42b1b ("perf tools: Initial python binding")
Quoted 14 lines
tools/perf/util/python.c +2 -0
index af85c0771c448..c7ae1f85aec6d 100644
@@ -2825,6 +2825,8 @@ static PyObject *pyrf_evlist__get_pollfd(struct pyrf_evlist *pevlist,
 
 	evlist = pevlist->evlist;
 	list = PyList_New(0);
+	if (!list)
+		return NULL;
 
 	for (i = 0; i < evlist__core(evlist)->pollfd.nr; ++i) {
 		PyObject *file;

tools/perf/util/python.c static PyObject *pyrf_evlist__get_pollfd(struct pyrf_evlist *pevlist, · line 2850

Quoted 6 lines
tools/perf/util/python.c +1 -0
@@ -2843,6 +2845,7 @@ static PyObject *pyrf_evlist__get_pollfd(struct pyrf_evlist *pevlist,
 
 	return list;
 free_list:
+	Py_XDECREF(list);
 	return PyErr_NoMemory();
 }

[Severity: High]
This isn't a bug introduced by this patch, but I noticed a potential crash here in pyrf_counts_values_get_values() when memory is exhausted:

tools/perf/util/python.c:pyrf_counts_values_get_values() {
	...
	PyObject *vals = PyList_New(5);

	if (!vals)
		return NULL;
	for (int i = 0; i < 5; i++)
		PyList_SetItem(vals, i, PyLong_FromLong(self->values.values[i]));

	return vals;
}

Can PyLong_FromLong() return a null pointer during memory pressure?

If so, PyList_SetItem() will place the null pointer directly into the list's internal array without validation.

Could this cause operations like PyObject_Repr to later dereference the pointer and crash the Python interpreter abruptly instead of raising a MemoryError?

[Severity: High]
This is a pre-existing issue, but there appears to be a memory leak in pyrf__metrics_cb():

tools/perf/util/python.c:pyrf__metrics_cb() {
	...
	PyObject *key = dict ? PyUnicode_FromString("MetricGroup") : NULL;
	PyObject *value = key ? pyrf__metrics_groups(pm) : NULL;

	if (!value || PyDict_SetItem(dict, key, value) != 0) {
		Py_XDECREF(key);
		Py_XDECREF(value);
		Py_XDECREF(dict);
		return -ENOMEM;
	}

	if (!add_to_dict(dict, "MetricName", pm->metric_name) ||
	...
}

Does the success path of PyDict_SetItem leak the key and value objects?

The error path correctly drops the references, but if PyDict_SetItem succeeds, it does not steal the references.

Since this function is called as a callback in pmu_for_each_sys_metric(), will this unconditionally leak both objects for every valid system metric processed?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=4
author reviewer
#59 lore raw

PATCH v3 5/15 perf synthetic-events: Fix uninitialized pthread_join

In reply to Ian Rogers

In perf_event__synthesize_threads(), fix an uninitialized pthread_join() call when thread creation fails by only joining the successfully created threads.

Assisted-by Antigravity:gemini-3.1-pro
Fixes 340b47f510bb ("perf top: Implement multithreading for perf_event__synthesize_threads")
---
 tools/perf/util/synthetic-events.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tools/perf/util/synthetic-events.c +2 -2
index 75a32ae8ef62..3583a60cc487 100644
@@ -1221,8 +1221,8 @@ int perf_event__synthesize_threads(const struct perf_tool *tool,
 	}
 	err = 0;
 out_join:
-	for (i = 0; i < thread_nr; i++)
-		pthread_join(synthesize_threads[i], NULL);
+	for (j = 0; j < i; j++)
+		pthread_join(synthesize_threads[j], NULL);
 	free(args);
 free_threads:
 	free(synthesize_threads);
Signature · 1 line
2.55.0.679.g6767b8d81c-goog
author reviewer
#60 lore raw

PATCH v3 6/15 perf test: Fix skiplist leak in cmd_test

In reply to Ian Rogers

Fix a memory leak in cmd_test() where skiplist was not freed on exit paths.

Assisted-by Antigravity:gemini-3.1-pro
Fixes 2ae828786c65 ("perf test: Allow skipping tests")
---
 tools/perf/tests/builtin-test.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)
tools/perf/tests/builtin-test.c +7 -2
index 41dcfbf874f0..4d0784b16723 100644
@@ -1802,9 +1802,14 @@ int cmd_test(int argc, const char **argv)
 	rlimit__bump_memlock();
 
 	suites = build_suites();
-	if (!suites)
-		return errno ? -errno : -ENOMEM;
+	if (!suites) {
+		int err = errno;
+
+		intlist__delete(skiplist);
+		return err ? -err : -ENOMEM;
+	}
 	ret = __cmd_test(suites, argc, argv, skiplist);
 	free(suites);
+	intlist__delete(skiplist);
 	return ret;
 }
Signature · 1 line
2.55.0.679.g6767b8d81c-goog
author reviewer
#61 lore raw

PATCH v3 7/15 perf python: Check counts_values size in set_values

In reply to Ian Rogers

The set_values function incorrectly assumed the list contained exactly 5 elements. Add a check to prevent out-of-bounds access.

Fixes 877108e42b1b ("perf tools: Initial python binding")
---
 tools/perf/util/python.c | 38 ++++++++++++++++++++++++++++++--------
 1 file changed, 30 insertions(+), 8 deletions(-)
tools/perf/util/python.c +30 -8
index c7ae1f85aec6..f64d662bd3ee 100644
@@ -1881,11 +1881,11 @@ static void pyrf_counts_values__delete(struct pyrf_counts_values *pcounts_values
 	  0, help }
 
 static PyMemberDef pyrf_counts_values_members[] = {
-	counts_values_member_def(val, T_ULONG, "Value of event"),
-	counts_values_member_def(ena, T_ULONG, "Time for which enabled"),
-	counts_values_member_def(run, T_ULONG, "Time for which running"),
-	counts_values_member_def(id, T_ULONG, "Unique ID for an event"),
-	counts_values_member_def(lost, T_ULONG, "Num of lost samples"),
+	counts_values_member_def(val, T_ULONGLONG, "Value of event"),
+	counts_values_member_def(ena, T_ULONGLONG, "Time for which enabled"),
+	counts_values_member_def(run, T_ULONGLONG, "Time for which running"),
+	counts_values_member_def(id, T_ULONGLONG, "Unique ID for an event"),
+	counts_values_member_def(lost, T_ULONGLONG, "Num of lost samples"),
 	{ .name = NULL, },
 };
 
@@ -1895,8 +1895,15 @@ static PyObject *pyrf_counts_values_get_values(struct pyrf_counts_values *self,
 
 	if (!vals)
 		return NULL;
-	for (int i = 0; i < 5; i++)
-		PyList_SetItem(vals, i, PyLong_FromLong(self->values.values[i]));
+	for (int i = 0; i < 5; i++) {
+		PyObject *val = PyLong_FromUnsignedLongLong(self->values.values[i]);
+
+		if (!val) {
+			Py_DECREF(vals);
+			return NULL;
+		}
+		PyList_SetItem(vals, i, val);
+	}
 
 	return vals;
 }
@@ -1907,19 +1914,34 @@ static int pyrf_counts_values_set_values(struct pyrf_counts_values *self, PyObje
 	Py_ssize_t size;
 	PyObject *item = NULL;
 
+	if (list == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
 	if (!PyList_Check(list)) {
 		PyErr_SetString(PyExc_TypeError, "Value assigned must be a list");
 		return -1;
 	}
 
 	size = PyList_Size(list);
+	if (size != 5) {
+		PyErr_SetString(PyExc_ValueError, "List must have exactly 5 entries");
+		return -1;
+	}
+
 	for (Py_ssize_t i = 0; i < size; i++) {
+		unsigned long long val;
+
 		item = PyList_GetItem(list, i);
 		if (!PyLong_Check(item)) {
 			PyErr_SetString(PyExc_TypeError, "List members should be numbers");
 			return -1;
 		}
-		self->values.values[i] = PyLong_AsLong(item);
+		val = PyLong_AsUnsignedLongLong(item);
+		if (val == (unsigned long long)-1 && PyErr_Occurred())
+			return -1;
+		self->values.values[i] = val;
 	}
 
 	return 0;
Signature · 1 line
2.55.0.679.g6767b8d81c-goog
author reviewer
#62 lore raw

PATCH v3 8/15 perf python: Validate CPU and thread maps in pyrf_evsel__open

In reply to Ian Rogers

Add explicit Py_TYPE checks to ensure the arguments passed are actually of the correct pyrf_thread_map and pyrf_cpu_map types.

Fixes 877108e42b1b ("perf tools: Initial python binding")
---
 tools/perf/util/python.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)
tools/perf/util/python.c +12 -2
index f64d662bd3ee..a1334400874c 100644
@@ -2113,11 +2113,21 @@ static PyObject *pyrf_evsel__open(struct pyrf_evsel *pevsel,
 					 &pcpus, &pthreads, &group, &inherit))
 		return NULL;
 
-	if (pthreads != NULL && pthreads != Py_None)
+	if (pthreads != NULL && pthreads != Py_None) {
+		if (!PyObject_TypeCheck(pthreads, &pyrf_thread_map__type)) {
+			PyErr_SetString(PyExc_TypeError, "threads must be a thread_map");
+			return NULL;
+		}
 		threads = ((struct pyrf_thread_map *)pthreads)->threads;
+	}
 
-	if (pcpus != NULL && pcpus != Py_None)
+	if (pcpus != NULL && pcpus != Py_None) {
+		if (!PyObject_TypeCheck(pcpus, &pyrf_cpu_map__type)) {
+			PyErr_SetString(PyExc_TypeError, "cpus must be a cpu_map");
+			return NULL;
+		}
 		cpus = ((struct pyrf_cpu_map *)pcpus)->cpus;
+	}
 
 	evsel->core.attr.inherit = inherit;
 	/*
Signature · 1 line
2.55.0.679.g6767b8d81c-goog
author reviewer
#63 lore raw

PATCH v3 9/15 perf python: Validate attribute setters in pyrf_evsel

In reply to Ian Rogers

If val is NULL when setting an attribute, PyErr_SetString should be called as deleting the attribute isn't supported. In addition, ensure PyErr_Occurred is checked before setting the attribute to avoid setting a garbage value.

Fixes 877108e42b1b ("perf tools: Initial python binding")
---
 tools/perf/util/python.c | 89 ++++++++++++++++++++++++++++++++++------
 1 file changed, 77 insertions(+), 12 deletions(-)
tools/perf/util/python.c +77 -12
index a1334400874c..848912401c4f 100644
@@ -2301,6 +2301,11 @@ static int pyrf_evsel__set_tracking(PyObject *self, PyObject *val, void *closure
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
 	is_true = PyObject_IsTrue(val);
 	if (is_true < 0)
 		return -1;
@@ -2312,11 +2317,21 @@ static int pyrf_evsel__set_tracking(PyObject *self, PyObject *val, void *closure
 static int pyrf_evsel__set_attr_config(PyObject *self, PyObject *val, void *closure __maybe_unused)
 {
 	struct pyrf_evsel *pevsel = (void *)self;
+	unsigned long long new_val;
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
-	pevsel->evsel->core.attr.config = PyLong_AsUnsignedLongLong(val);
-	return PyErr_Occurred() ? -1 : 0;
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
+	new_val = PyLong_AsUnsignedLongLong(val);
+	if (PyErr_Occurred())
+		return -1;
+
+	pevsel->evsel->core.attr.config = new_val;
+	return 0;
 }
 
 static PyObject *pyrf_evsel__get_attr_config(PyObject *self, void *closure __maybe_unused)
@@ -2331,11 +2346,21 @@ static PyObject *pyrf_evsel__get_attr_config(PyObject *self, void *closure __may
 static int pyrf_evsel__set_attr_read_format(PyObject *self, PyObject *val, void *closure __maybe_unused)
 {
 	struct pyrf_evsel *pevsel = (void *)self;
+	unsigned long long new_val;
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
-	pevsel->evsel->core.attr.read_format = PyLong_AsUnsignedLongLong(val);
-	return PyErr_Occurred() ? -1 : 0;
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
+	new_val = PyLong_AsUnsignedLongLong(val);
+	if (PyErr_Occurred())
+		return -1;
+
+	pevsel->evsel->core.attr.read_format = new_val;
+	return 0;
 }
 
 static PyObject *pyrf_evsel__get_attr_read_format(PyObject *self, void *closure __maybe_unused)
@@ -2350,11 +2375,21 @@ static PyObject *pyrf_evsel__get_attr_read_format(PyObject *self, void *closure
 static int pyrf_evsel__set_attr_sample_period(PyObject *self, PyObject *val, void *closure __maybe_unused)
 {
 	struct pyrf_evsel *pevsel = (void *)self;
+	unsigned long long new_val;
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
-	pevsel->evsel->core.attr.sample_period = PyLong_AsUnsignedLongLong(val);
-	return PyErr_Occurred() ? -1 : 0;
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
+	new_val = PyLong_AsUnsignedLongLong(val);
+	if (PyErr_Occurred())
+		return -1;
+
+	pevsel->evsel->core.attr.sample_period = new_val;
+	return 0;
 }
 
 static PyObject *pyrf_evsel__get_attr_sample_period(PyObject *self, void *closure __maybe_unused)
@@ -2369,11 +2404,21 @@ static PyObject *pyrf_evsel__get_attr_sample_period(PyObject *self, void *closur
 static int pyrf_evsel__set_attr_sample_type(PyObject *self, PyObject *val, void *closure __maybe_unused)
 {
 	struct pyrf_evsel *pevsel = (void *)self;
+	unsigned long long new_val;
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
-	pevsel->evsel->core.attr.sample_type = PyLong_AsUnsignedLongLong(val);
-	return PyErr_Occurred() ? -1 : 0;
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
+	new_val = PyLong_AsUnsignedLongLong(val);
+	if (PyErr_Occurred())
+		return -1;
+
+	pevsel->evsel->core.attr.sample_type = new_val;
+	return 0;
 }
 
 static PyObject *pyrf_evsel__get_attr_sample_type(PyObject *self, void *closure __maybe_unused)
@@ -2397,11 +2442,21 @@ static PyObject *pyrf_evsel__get_attr_size(PyObject *self, void *closure __maybe
 static int pyrf_evsel__set_attr_type(PyObject *self, PyObject *val, void *closure __maybe_unused)
 {
 	struct pyrf_evsel *pevsel = (void *)self;
+	unsigned long new_val;
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
-	pevsel->evsel->core.attr.type = PyLong_AsUnsignedLong(val);
-	return PyErr_Occurred() ? -1 : 0;
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
+	new_val = PyLong_AsUnsignedLong(val);
+	if (PyErr_Occurred())
+		return -1;
+
+	pevsel->evsel->core.attr.type = new_val;
+	return 0;
 }
 
 static PyObject *pyrf_evsel__get_attr_type(PyObject *self, void *closure __maybe_unused)
@@ -2416,11 +2471,21 @@ static PyObject *pyrf_evsel__get_attr_type(PyObject *self, void *closure __maybe
 static int pyrf_evsel__set_attr_wakeup_events(PyObject *self, PyObject *val, void *closure __maybe_unused)
 {
 	struct pyrf_evsel *pevsel = (void *)self;
+	unsigned long new_val;
 
 	CHECK_INITIALIZED_INT(pevsel->evsel, "evsel");
 
-	pevsel->evsel->core.attr.wakeup_events = PyLong_AsUnsignedLong(val);
-	return PyErr_Occurred() ? -1 : 0;
+	if (val == NULL) {
+		PyErr_SetString(PyExc_TypeError, "cannot delete attribute");
+		return -1;
+	}
+
+	new_val = PyLong_AsUnsignedLong(val);
+	if (PyErr_Occurred())
+		return -1;
+
+	pevsel->evsel->core.attr.wakeup_events = new_val;
+	return 0;
 }
 
 static PyObject *pyrf_evsel__get_attr_wakeup_events(PyObject *self, void *closure __maybe_unused)
Signature · 1 line
2.55.0.679.g6767b8d81c-goog
author reviewer
#64 lore raw

PATCH v3 10/15 perf python: Zero initialize perf_data in pyrf_data__init

In reply to Ian Rogers

Replace path clearing with memset so the entire struct is zeroed, preventing uninitialized fields from causing errors later.

Fixes 4cd0142f7dec ("perf python: Add wrapper for perf_data file abstraction")
---
 tools/perf/util/python.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
tools/perf/util/python.c +2 -1
index 848912401c4f..c70b151428f7 100644
@@ -3,6 +3,7 @@
 #include <Python.h>
 
 #include <inttypes.h>
+#include <string.h>
 
 #include <linux/err.h>
 #include <poll.h>
@@ -3627,7 +3628,7 @@ static int pyrf_data__init(struct pyrf_data *pdata, PyObject *args, PyObject *kw
 	if (pdata->data.open)
 		perf_data__close(&pdata->data);
 	free((char *)pdata->data.path);
-	pdata->data.path = NULL;
+	memset(&pdata->data, 0, sizeof(pdata->data));
 
 	if (fd != -1) {
 		struct stat st;
Signature · 1 line
2.55.0.679.g6767b8d81c-goog
author reviewer
#65 lore raw

PATCH v3 11/15 perf python: Add thread and PMU uninitialized checks

In reply to Ian Rogers

Add CHECK_INITIALIZED checks to the thread attribute getters (get_pid, get_tid, get_ppid) to prevent crashes if they are accessed before being properly initialized.

Fixes 3b96bf7af60d ("perf python: Add python session abstraction wrapping perf's session")
---
 tools/perf/util/python.c | 8 ++++++++
 1 file changed, 8 insertions(+)
tools/perf/util/python.c +8 -0
index c70b151428f7..fe5afce17333 100644
@@ -1678,6 +1678,7 @@ static PyObject *pyrf_pmu__name(PyObject *self)
 {
 	struct pyrf_pmu *ppmu = (void *)self;
 
+	CHECK_INITIALIZED(ppmu->pmu, "pmu");
 	return PyUnicode_FromString(ppmu->pmu->name);
 }
 
@@ -1730,6 +1731,8 @@ static PyObject *pyrf_pmu__events(PyObject *self)
 	PyObject *py_list = PyList_New(0);
 	int ret;
 
+	CHECK_INITIALIZED(ppmu->pmu, "pmu");
+
 	if (!py_list)
 		return NULL;
 
@@ -1750,6 +1753,7 @@ static PyObject *pyrf_pmu__repr(PyObject *self)
 {
 	struct pyrf_pmu *ppmu = (void *)self;
 
+	CHECK_INITIALIZED(ppmu->pmu, "pmu");
 	return PyUnicode_FromFormat("pmu(%s)", ppmu->pmu->name);
 }
 
@@ -3764,21 +3768,25 @@ static PyMethodDef pyrf_thread__methods[] = {
 
 static PyObject *pyrf_thread__get_pid(struct pyrf_thread *pthread, void *closure __maybe_unused)
 {
+	CHECK_INITIALIZED(pthread->thread, "thread");
 	return PyLong_FromLong(thread__pid(pthread->thread));
 }
 
 static PyObject *pyrf_thread__get_tid(struct pyrf_thread *pthread, void *closure __maybe_unused)
 {
+	CHECK_INITIALIZED(pthread->thread, "thread");
 	return PyLong_FromLong(thread__tid(pthread->thread));
 }
 
 static PyObject *pyrf_thread__get_ppid(struct pyrf_thread *pthread, void *closure __maybe_unused)
 {
+	CHECK_INITIALIZED(pthread->thread, "thread");
 	return PyLong_FromLong(thread__ppid(pthread->thread));
 }
 
 static PyObject *pyrf_thread__get_cpu(struct pyrf_thread *pthread, void *closure __maybe_unused)
 {
+	CHECK_INITIALIZED(pthread->thread, "thread");
 	return PyLong_FromLong(thread__cpu(pthread->thread));
 }
 
Signature · 1 line
2.55.0.679.g6767b8d81c-goog
#66 lore raw

Re: [PATCH v3 11/15] perf python: Add thread and PMU uninitialized checks

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Medium] The newly added `CHECK_INITIALIZED(ppmu->pmu, "pmu")` in `pyrf_pmu__events()` causes a Python…

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] The newly added CHECK_INITIALIZED(ppmu->pmu, "pmu") in pyrf_pmu__events() causes a Python object memory leak of py_list on the error path. --

--- Patch [11]: [PATCH v3 11/15] perf python: Add thread and PMU uninitialized checks --- commit a381b82bb0eb00b93be525bfa1c265b574c03a5a
Author: Ian Rogers <[email protected]>

perf python: Add thread and PMU uninitialized checks

This commit adds CHECK_INITIALIZED guards to thread and PMU getters to prevent crashes when accessed before proper initialization.

Quoted 4 lines
tools/perf/util/python.c +0 -0
index c70b151428f76..fe5afce173330 100644

[ ... ]

tools/perf/util/python.c static PyObject *pyrf_pmu__events(PyObject *self) · line 1737

Quoted 8 lines
tools/perf/util/python.c +2 -0
@@ -1730,6 +1731,8 @@ static PyObject *pyrf_pmu__events(PyObject *self)
 	PyObject *py_list = PyList_New(0);
 	int ret;
 
+	CHECK_INITIALIZED(ppmu->pmu, "pmu");
+
 	if (!py_list)
 		return NULL;

[Severity: Medium]
Does this code leak py_list if CHECK_INITIALIZED fails and returns early?

Since py_list is allocated with PyList_New(0) just before the initialization check, it appears this Python object is left orphaned if the macro evaluates to true and forces an early return.

Could the CHECK_INITIALIZED macro be moved to the beginning of pyrf_pmu__events() before the PyList_New(0) allocation to avoid this leak?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=11
maintainer
#67 lore raw

Re: [PATCH v3 11/15] perf python: Add thread and PMU uninitialized checks

Hello,

Quoted 32 lines · On Sun, Aug 09, 2026 at 07:28:09AM +0000, [email protected] wrote:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] The newly added CHECK_INITIALIZED(ppmu->pmu, "pmu") in pyrf_pmu__events() causes a Python object memory leak of py_list on the error path. --

--- Patch [11]: [PATCH v3 11/15] perf python: Add thread and PMU uninitialized checks --- commit a381b82bb0eb00b93be525bfa1c265b574c03a5a
Author: Ian Rogers <[email protected]>

perf python: Add thread and PMU uninitialized checks

This commit adds CHECK_INITIALIZED guards to thread and PMU getters to prevent crashes when accessed before proper initialization.

Quoted 4 lines
tools/perf/util/python.c +0 -0
index c70b151428f76..fe5afce173330 100644

[ ... ]

tools/perf/util/python.c static PyObject *pyrf_pmu__events(PyObject *self) · line 1737

Quoted 8 lines
tools/perf/util/python.c +2 -0
@@ -1730,6 +1731,8 @@ static PyObject *pyrf_pmu__events(PyObject *self)
 	PyObject *py_list = PyList_New(0);
 	int ret;
 
+	CHECK_INITIALIZED(ppmu->pmu, "pmu");
+
 	if (!py_list)
 		return NULL;

[Severity: Medium]
Does this code leak py_list if CHECK_INITIALIZED fails and returns early?

Quoted 6 lines

Since py_list is allocated with PyList_New(0) just before the initialization check, it appears this Python object is left orphaned if the macro evaluates to true and forces an early return.

Could the CHECK_INITIALIZED macro be moved to the beginning of pyrf_pmu__events() before the PyList_New(0) allocation to avoid this leak?

It seems this is the only issue introduced in this series. As it's trivial, I'll apply the following change for Ian.

Thanks,
Namhyung

tools/perf/util/python.c +2 -1
index fe5afce1733306be..36c5f4b0d0bb15e7 100644
@@ -1728,11 +1728,12 @@ static int pyrf_pmu__events_cb(void *state, struct pmu_event_info *info)
 static PyObject *pyrf_pmu__events(PyObject *self)
 {
        struct pyrf_pmu *ppmu = (void *)self;
-       PyObject *py_list = PyList_New(0);
+       PyObject *py_list;
        int ret;
 
        CHECK_INITIALIZED(ppmu->pmu, "pmu");
 
+       py_list = PyList_New(0);
        if (!py_list)
                return NULL;
 
author reviewer
#68 lore raw

Re: [PATCH v3 11/15] perf python: Add thread and PMU uninitialized checks

Quoted 40 lines · Namhyung Kim

Hello,

Quoted 36 lines · On Sun, Aug 09, 2026 at 07:28:09AM +0000, [email protected] wrote:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] The newly added CHECK_INITIALIZED(ppmu->pmu, "pmu") in pyrf_pmu__events() causes a Python object memory leak of py_list on the error path. --

--- Patch [11]: [PATCH v3 11/15] perf python: Add thread and PMU uninitialized checks --- commit a381b82bb0eb00b93be525bfa1c265b574c03a5a
Author: Ian Rogers <[email protected]>

perf python: Add thread and PMU uninitialized checks

This commit adds CHECK_INITIALIZED guards to thread and PMU getters to prevent crashes when accessed before proper initialization.

Quoted 4 lines
tools/perf/util/python.c +0 -0
index c70b151428f76..fe5afce173330 100644

[ ... ]

tools/perf/util/python.c static PyObject *pyrf_pmu__events(PyObject *self) · line 1737

Quoted 8 lines
tools/perf/util/python.c +2 -0
@@ -1730,6 +1731,8 @@ static PyObject *pyrf_pmu__events(PyObject *self)
    PyObject *py_list = PyList_New(0);
    int ret;
 
+   CHECK_INITIALIZED(ppmu->pmu, "pmu");
+
    if (!py_list)
            return NULL;

[Severity: Medium]
Does this code leak py_list if CHECK_INITIALIZED fails and returns early?

Since py_list is allocated with PyList_New(0) just before the initialization check, it appears this Python object is left orphaned if the macro evaluates to true and forces an early return.

Quoted 5 lines
Quoted 2 lines

Could the CHECK_INITIALIZED macro be moved to the beginning of pyrf_pmu__events() before the PyList_New(0) allocation to avoid this leak?

It seems this is the only issue introduced in this series. As it's trivial, I'll apply the following change for Ian.

Looks good to me. Thanks,
Ian

Quoted 23 lines

Thanks,
Namhyung

tools/perf/util/python.c +2 -1
index fe5afce1733306be..36c5f4b0d0bb15e7 100644
@@ -1728,11 +1728,12 @@ static int pyrf_pmu__events_cb(void *state, struct pmu_event_info *info)
 static PyObject *pyrf_pmu__events(PyObject *self)
 {
        struct pyrf_pmu *ppmu = (void *)self;
-       PyObject *py_list = PyList_New(0);
+       PyObject *py_list;
        int ret;
 
        CHECK_INITIALIZED(ppmu->pmu, "pmu");
 
+       py_list = PyList_New(0);
        if (!py_list)
                return NULL;
 
author reviewer
#69 lore raw

PATCH v3 12/15 perf python: Fix MetricGroup return type in perf.pyi

In reply to Ian Rogers

The metrics() function can return a dictionary where the value is either a string or a list of strings, so the type signature in the stub file should be Union[str, List[str]].

Fixes 430da3cd03b4 ("perf python: Add perf.pyi stubs file")
---
 tools/perf/python/perf.pyi | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tools/perf/python/perf.pyi +2 -2
index f266c9baffd0..58b0b3ed819d 100644
@@ -1,5 +1,5 @@
 """Type stubs for the perf Python module."""
-from typing import Callable, Dict, List, Optional, Any, Iterator
+from typing import Callable, Dict, List, Optional, Any, Iterator, Union
 
 def config_get(name: str) -> Optional[str]:
     """Get a configuration value from perf config.
@@ -12,7 +12,7 @@ def config_get(name: str) -> Optional[str]:
     """
     ...
 
-def metrics() -> List[Dict[str, str]]:
+def metrics() -> List[Dict[str, Union[str, List[str]]]]:
     """Get a list of available metrics.
 
     Returns:
Signature · 1 line
2.55.0.679.g6767b8d81c-goog
#70 lore raw

Re: [PATCH v3 12/15] perf python: Fix MetricGroup return type in perf.pyi

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] Memory leak of PyObject references 'key' and 'value' in pyrf__metrics_cb upon success. --

--- Patch [12]: [PATCH v3 12/15] perf python: Fix MetricGroup return type in perf.pyi --- commit f9b984697afa89938af9de90c9da302a78dffa89
Author: Ian Rogers <[email protected]>

perf python: Fix MetricGroup return type in perf.pyi

The metrics() function can return a dictionary where the value is either a string or a list of strings, so the type signature in the stub file should be Union[str, List[str]].

Quoted 2 lines
tools/perf/python/perf.pyi +0 -0

[ ... ]

tools/perf/python/perf.pyi def config_get(name: str) -> Optional[str]: · line 15

Quoted 2 lines
tools/perf/python/perf.pyi +1 -1
-def metrics() -> List[Dict[str, str]]:
+def metrics() -> List[Dict[str, Union[str, List[str]]]]:

[Severity: High]
This isn't a bug introduced by this patch, but does the underlying C implementation of the metrics function leak the 'key' and 'value' objects on success?

Looking at pyrf__metrics_cb() in tools/perf/util/python.c, it appears PyDict_SetItem is used without decrementing the reference counts afterward:

	PyObject *key = dict ? PyUnicode_FromString("MetricGroup") : NULL;
	PyObject *value = key ? pyrf__metrics_groups(pm) : NULL;

	if (!value || PyDict_SetItem(dict, key, value) != 0) {
		Py_XDECREF(key);
		Py_XDECREF(value);
		Py_XDECREF(dict);
		return -ENOMEM;
	}

Since PyDict_SetItem doesn't steal references, wouldn't the caller need to decrement the reference counts for both 'key' and 'value' after a successful insertion? Since they are only decremented in the error path, does this result in a memory leak every time a metric group is processed?

Quoted 3 lines

"""Get a list of available metrics.

Returns:

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=12
author reviewer
#71 lore raw

PATCH v3 13/15 perf python: Fix count_values memory leak in pyrf_evsel__read

In reply to Ian Rogers

In pyrf_evsel__read, if PyArg_ParseTuple fails, the allocated count_values is leaked. Move the allocation of count_values after the PyArg_ParseTuple call to prevent the memory leak.

Fixes 739621f65702 ("perf python: Add evsel read method")
---
 tools/perf/util/python.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)
tools/perf/util/python.c +4 -5
index fe5afce17333..71431c9c2e76 100644
@@ -2210,11 +2210,6 @@ static PyObject *pyrf_evsel__read(struct pyrf_evsel *pevsel,
 
 	CHECK_INITIALIZED(evsel, "evsel");
 
-	count_values = PyObject_New(struct pyrf_counts_values,
-							       &pyrf_counts_values__type);
-	if (!count_values)
-		return NULL;
-
 	if (!PyArg_ParseTuple(args, "ii", &cpu, &thread))
 		return NULL;
 
@@ -2233,6 +2228,10 @@ static PyObject *pyrf_evsel__read(struct pyrf_evsel *pevsel,
 	if (evsel__ensure_counts(evsel))
 		return PyErr_NoMemory();
 
+	count_values = PyObject_New(struct pyrf_counts_values, &pyrf_counts_values__type);
+	if (!count_values)
+		return NULL;
+
 	/* Set up pointers to the old and newly read counter values. */
 	old_count = perf_counts(evsel->prev_raw_counts, cpu_idx, thread_idx);
 	new_count = perf_counts(evsel->counts, cpu_idx, thread_idx);
Signature · 1 line
2.55.0.679.g6767b8d81c-goog
author reviewer
#72 lore raw

PATCH v3 14/15 perf python: Fix memory leak in pyrf__metrics_cb

In reply to Ian Rogers

In pyrf__metrics_cb, PyDict_SetItem does not steal the reference of the key and value, so they need to be decref'ed after successful insertion to avoid memory leaks.

Fixes 47b3e95728eb ("perf python: Add metrics function")
---
 tools/perf/util/python.c | 2 ++
 1 file changed, 2 insertions(+)
tools/perf/util/python.c +2 -0
index 71431c9c2e76..077e0f5d9fcd 100644
@@ -3574,6 +3574,8 @@ static int pyrf__metrics_cb(const struct pmu_metric *pm,
 		Py_XDECREF(dict);
 		return -ENOMEM;
 	}
+	Py_DECREF(key);
+	Py_DECREF(value);
 
 	if (!add_to_dict(dict, "MetricName", pm->metric_name) ||
 	    !add_to_dict(dict, "PMU", pm->pmu) ||
Signature · 1 line
2.55.0.679.g6767b8d81c-goog
author reviewer
#73 lore raw

PATCH v3 15/15 perf synthetic-events: Fix divide by zero in perf_event__synthesize_threads

In reply to Ian Rogers

If scandir() finds no matching tasks in /proc, n is 0. If thread_nr is > 1, we bypass the single-thread fast path and then clamp thread_nr to n, making it 0. This results in a divide by zero when calculating num_per_thread.

Handle n <= 1 early to use the single-thread fast path and prevent the crash.

Fixes 340b47f510bb ("perf top: Implement multithreading for perf_event__synthesize_threads")
---
 tools/perf/util/synthetic-events.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/synthetic-events.c +1 -1
index 3583a60cc487..0c150193cca8 100644
@@ -1175,7 +1175,7 @@ int perf_event__synthesize_threads(const struct perf_tool *tool,
 	else
 		thread_nr = nr_threads_synthesize;
 
-	if (thread_nr <= 1) {
+	if (thread_nr <= 1 || n <= 1) {
 		err = __perf_event__synthesize_threads(tool, process,
 						       machine,
 						       needs_mmap, mmap_data,
Signature · 1 line
2.55.0.679.g6767b8d81c-goog
maintainer Applied
#74 lore raw

Re: [PATCH v3 00/15] perf python: Fix python extension bugs from v19 review

In reply to Ian Rogers
Quoted 17 lines · Ian Rogers

Phase 1 of splitting the v19 review patches into smaller fixes. These patches address issues identified in the python extension during the review of the v19 script porting series, as well as fixing several pre-existing bugs and leaks.

v3:
- Addressed review feedback from Namhyung Kim and Sashiko AI:
- Dropped "perf data: Fix directory file descriptor leak in perf_data__switch" to avoid Use-After-Free during directory mode file rotation.
- Added list == NULL check in pyrf_counts_values_set_values to safely handle attribute deletion (del obj.values).
- Preserved errno before calling intlist__delete(skiplist) in cmd_test error path to prevent clobbering the return error code.
- Added CHECK_INITIALIZED to pyrf_pmu__repr to prevent NULL pointer dereferences on uninitialized PMU objects.
- Fixed duplicate patch title from v2.

Quoted 1 line

[...]

Applied to perf-tools-next, thanks!

Best regards,
Namhyung