[PATCH] selftests/damon: add regression test for damos_walk() vs kdamond exit race

Sailesh Nandanavanam posted 1 patch 8 hours ago
There is a newer version of this series
tools/testing/selftests/damon/Makefile        |  1 +
.../sysfs_damos_walk_kdamond_exit_race.py     | 47 +++++++++++++++++++
2 files changed, 48 insertions(+)
create mode 100644 tools/testing/selftests/damon/sysfs_damos_walk_kdamond_exit_race.py
[PATCH] selftests/damon: add regression test for damos_walk() vs kdamond exit race
Posted by Sailesh Nandanavanam 8 hours ago
Add a regression test that verifies damos_walk() does not deadlock
when racing with kdamond_fn() exit.

When kdamond_fn() finishes its main loop, it cancels remaining
damos_walk() requests and unsets damon_ctx->kdamond. Without the fix
in commit 33c3f6c2b48c, damos_walk() could be called right after
cancellation but before kdamond pointer unset, causing it to wait
forever for handling that never comes.

The test starts kdamond monitoring a short-lived process, waits for
the process to exit naturally triggering kdamond termination, then
rapidly calls update_schemes_tried_regions to hit the race window.
Without the fix this would hang indefinitely.

Fixes: 33c3f6c2b48c ("mm/damon/core: fix damos_walk() vs kdamond_fn() exit race")
Signed-off-by: Sailesh Nandanavanam <saileshnandanavanam@gmail.com>
---
 tools/testing/selftests/damon/Makefile        |  1 +
 .../sysfs_damos_walk_kdamond_exit_race.py     | 47 +++++++++++++++++++
 2 files changed, 48 insertions(+)
 create mode 100644 tools/testing/selftests/damon/sysfs_damos_walk_kdamond_exit_race.py

diff --git a/tools/testing/selftests/damon/Makefile b/tools/testing/selftests/damon/Makefile
index 2180c328a825..60c83d6c318e 100644
--- a/tools/testing/selftests/damon/Makefile
+++ b/tools/testing/selftests/damon/Makefile
@@ -20,6 +20,7 @@ TEST_PROGS += sysfs_update_removed_scheme_dir.sh
 TEST_PROGS += sysfs_update_schemes_tried_regions_hang.py
 TEST_PROGS += sysfs_memcg_path_leak.sh
 TEST_PROGS += sysfs_no_op_commit_break.py
+TEST_PROGS += sysfs_damos_walk_kdamond_exit_race.py
 
 EXTRA_CLEAN = __pycache__
 
diff --git a/tools/testing/selftests/damon/sysfs_damos_walk_kdamond_exit_race.py b/tools/testing/selftests/damon/sysfs_damos_walk_kdamond_exit_race.py
new file mode 100644
index 000000000000..9b9d66a23f51
--- /dev/null
+++ b/tools/testing/selftests/damon/sysfs_damos_walk_kdamond_exit_race.py
@@ -0,0 +1,47 @@
+#!/usr/bin/env python3
+# SPDX-License-Identifier: GPL-2.0
+#
+# Regression test for damos_walk() vs kdamond_fn() exit race.
+import subprocess
+import time
+import _damon_sysfs
+def main():
+    proc = subprocess.Popen(['sleep', '0.3'])
+    kdamonds = _damon_sysfs.Kdamonds([_damon_sysfs.Kdamond(
+            contexts=[_damon_sysfs.DamonCtx(
+                ops='vaddr',
+                targets=[_damon_sysfs.DamonTarget(pid=proc.pid)],
+                schemes=[_damon_sysfs.Damos(
+                    action='stat',
+                    access_pattern=_damon_sysfs.DamosAccessPattern(
+                        nr_accesses=[0, 200]))]
+                )]
+            )])
+    err = kdamonds.start()
+    if err is not None:
+        print('kdamond start failed: %s' % err)
+        exit(1)
+    proc.wait()
+    completed = False
+    timeout = time.time() + 5
+    while time.time() < timeout:
+        err = kdamonds.kdamonds[0].update_schemes_tried_regions()
+        if err is not None:
+            completed = True
+            break
+        try:
+            with open('/sys/kernel/mm/damon/admin/kdamonds/0/state') as f:
+                state = f.read().strip()
+            if state == 'off':
+                completed = True
+                break
+        except:
+            completed = True
+            break
+    if not completed:
+        print('FAIL: test timed out - possible damos_walk/kdamond exit race deadlock')
+        kdamonds.stop()
+        exit(1)
+    print('PASS: damos_walk() vs kdamond exit race not triggered')
+if __name__ == '__main__':
+    main()
-- 
2.34.1