[PATCH v1] perf lock: Add --duration-filter option

[Chun-Tse Shao]

This patch introduces the `--duration-filter` option, allows users to
exclude lock contention samples with durations shorter than the
specified filter value.

Signed-off-by: Chun-Tse Shao <ctshao@xxxxxxxxxx>
Cc: Namhyung Kim <namhyung@xxxxxxxxxx>
---
 tools/perf/Documentation/perf-lock.txt        |  3 +++
 tools/perf/builtin-lock.c                     |  3 +++
 tools/perf/util/bpf_lock_contention.c         | 22 ++++++++++++++-----
 .../perf/util/bpf_skel/lock_contention.bpf.c  |  7 ++++++
 tools/perf/util/lock-contention.h             |  1 +
 5 files changed, 31 insertions(+), 5 deletions(-)

diff --git a/tools/perf/Documentation/perf-lock.txt b/tools/perf/Documentation/perf-lock.txt
index 859dc11a7372..1f57f5fc59e0 100644
--- a/tools/perf/Documentation/perf-lock.txt
+++ b/tools/perf/Documentation/perf-lock.txt
@@ -216,6 +216,9 @@ CONTENTION OPTIONS
 --cgroup-filter=<value>::
 	Show lock contention only in the given cgroups (comma separated list).
 
+--duration-filter=<value>::
+  Filter out lock contention samples which durations less than the specified
+  value (default: 0). The unit is nanoseconds (ns).
 
 SEE ALSO
 --------
diff --git a/tools/perf/builtin-lock.c b/tools/perf/builtin-lock.c
index 05e7bc30488a..d7b454e712bf 100644
--- a/tools/perf/builtin-lock.c
+++ b/tools/perf/builtin-lock.c
@@ -60,6 +60,7 @@ static int stack_skip = CONTENTION_STACK_SKIP;
 static int print_nr_entries = INT_MAX / 2;
 static const char *output_name = NULL;
 static FILE *lock_output;
+static int duration_filter;
 
 static struct lock_filter filters;
 
@@ -2004,6 +2005,7 @@ static int __cmd_contention(int argc, const char **argv)
 		.save_callstack = needs_callstack(),
 		.owner = show_lock_owner,
 		.cgroups = RB_ROOT,
+		.duration_filter = duration_filter,
 	};
 
 	lockhash_table = calloc(LOCKHASH_SIZE, sizeof(*lockhash_table));
@@ -2580,6 +2582,7 @@ int cmd_lock(int argc, const char **argv)
 	OPT_BOOLEAN(0, "lock-cgroup", &show_lock_cgroups, "show lock stats by cgroup"),
 	OPT_CALLBACK('G', "cgroup-filter", NULL, "CGROUPS",
 		     "Filter specific cgroups", parse_cgroup_filter),
+	OPT_INTEGER(0, "duration-filter", &duration_filter, "Filter samples by duration"),
 	OPT_PARENT(lock_options)
 	};
 
diff --git a/tools/perf/util/bpf_lock_contention.c b/tools/perf/util/bpf_lock_contention.c
index 5af8f6d1bc95..7b982a3e4000 100644
--- a/tools/perf/util/bpf_lock_contention.c
+++ b/tools/perf/util/bpf_lock_contention.c
@@ -203,6 +203,7 @@ int lock_contention_prepare(struct lock_contention *con)
 	skel->rodata->aggr_mode = con->aggr_mode;
 	skel->rodata->needs_callstack = con->save_callstack;
 	skel->rodata->lock_owner = con->owner;
+	skel->rodata->duration_filter = con->duration_filter;
 
 	if (con->aggr_mode == LOCK_AGGR_CGROUP || con->filters->nr_cgrps) {
 		if (cgroup_is_v2("perf_event"))
@@ -568,12 +569,23 @@ struct lock_stat *pop_owner_stack_trace(struct lock_contention *con)
 	if (stack_trace == NULL)
 		goto out_err;
 
-	if (bpf_map_get_next_key(stacks_fd, NULL, stack_trace))
-		goto out_err;
+	/*
+	 * `owner_stacks` contains stacks recorded in `contention_begin()` that either never reached
+	 * `contention_end()` or were filtered out and not stored in `owner_stat`. We skip if we
+	 * cannot find corresponding `contention_data` in `owner_stat` with the given `stack_id`.
+	 */
+	while (true) {
+		if (bpf_map_get_next_key(stacks_fd, NULL, stack_trace))
+			goto out_err;
+
+		bpf_map_lookup_elem(stacks_fd, stack_trace, &stack_id);
+		ckey.stack_id = stack_id;
+		if (bpf_map_lookup_elem(stat_fd, &ckey, &cdata) == 0)
+			break;
 
-	bpf_map_lookup_elem(stacks_fd, stack_trace, &stack_id);
-	ckey.stack_id = stack_id;
-	bpf_map_lookup_elem(stat_fd, &ckey, &cdata);
+		/* Can not find `contention_data`, delete and skip. */
+		bpf_map_delete_elem(stacks_fd, stack_trace);
+	}
 
 	st = zalloc(sizeof(struct lock_stat));
 	if (!st)
diff --git a/tools/perf/util/bpf_skel/lock_contention.bpf.c b/tools/perf/util/bpf_skel/lock_contention.bpf.c
index 69be7a4234e0..26ddc0f21378 100644
--- a/tools/perf/util/bpf_skel/lock_contention.bpf.c
+++ b/tools/perf/util/bpf_skel/lock_contention.bpf.c
@@ -176,6 +176,7 @@ const volatile int stack_skip;
 const volatile int lock_owner;
 const volatile int use_cgroup_v2;
 const volatile int max_stack;
+const volatile int duration_filter;
 
 /* determine the key of lock stat */
 const volatile int aggr_mode;
@@ -457,6 +458,9 @@ static inline void update_contention_data(struct contention_data *data, u64 dura
 
 static inline void update_owner_stat(u32 id, u64 duration, u32 flags)
 {
+	if (duration < duration_filter)
+		return;
+
 	struct contention_key key = {
 		.stack_id = id,
 		.pid = 0,
@@ -707,6 +711,9 @@ int contention_end(u64 *ctx)
 		}
 	}
 skip_owner:
+	if (duration < duration_filter)
+		goto out;
+
 	switch (aggr_mode) {
 	case LOCK_AGGR_CALLER:
 		key.stack_id = pelem->stack_id;
diff --git a/tools/perf/util/lock-contention.h b/tools/perf/util/lock-contention.h
index b5d916aa49df..97042e6d8b10 100644
--- a/tools/perf/util/lock-contention.h
+++ b/tools/perf/util/lock-contention.h
@@ -149,6 +149,7 @@ struct lock_contention {
 	int owner;
 	int nr_filtered;
 	bool save_callstack;
+	int duration_filter;
 };
 
 struct option;
-- 
2.49.0.604.gff1f9ca942-goog