[PATCH/RFC 12/16] perf tools: Reduce lock contention when processing events

[Namhyung Kim <namhyung@kernel.org>]

When multi-thread is enabled, the machine->threads_lock is contented
as all worker threads try to grab the writer lock using the
machine__findnew_thread().  Usually, the thread they're looking for is
in the tree so they only need the reader lock though.

Thus try machine__find_thread() first, and then fallback to the
'findnew' API.  This will improve the performance.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
 tools/perf/util/event.c   |  7 +++++--
 tools/perf/util/machine.c | 19 ++++++++++++++++---
 2 files changed, 21 insertions(+), 5 deletions(-)

diff --git a/tools/perf/util/event.c b/tools/perf/util/event.c
index 8b10621b415c..cefd43f8ec66 100644
--- a/tools/perf/util/event.c
+++ b/tools/perf/util/event.c
@@ -993,10 +993,13 @@ int perf_event__preprocess_sample(const union perf_event *event,
 				  struct perf_sample *sample)
 {
 	u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
-	struct thread *thread = machine__findnew_thread(machine, sample->pid,
-							sample->tid);
+	struct thread *thread = machine__find_thread(machine, sample->pid,
+						     sample->tid);
 
 	if (thread == NULL)
+		thread = machine__findnew_thread(machine, sample->pid,
+						 sample->tid);
+	if (thread == NULL)
 		return -1;
 
 	dump_printf(" ... thread: %s:%d\n", thread__comm_str(thread), thread->tid);
diff --git a/tools/perf/util/machine.c b/tools/perf/util/machine.c
index f5882b8c8db9..6d7349501f26 100644
--- a/tools/perf/util/machine.c
+++ b/tools/perf/util/machine.c
@@ -467,7 +467,7 @@ struct comm *machine__thread_exec_comm(struct machine *machine,
 int machine__process_comm_event(struct machine *machine, union perf_event *event,
 				struct perf_sample *sample)
 {
-	struct thread *thread = machine__findnew_thread(machine,
+	struct thread *thread = machine__find_thread(machine,
 							event->comm.pid,
 							event->comm.tid);
 	bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
@@ -479,6 +479,10 @@ int machine__process_comm_event(struct machine *machine, union perf_event *event
 	if (dump_trace)
 		perf_event__fprintf_comm(event, stdout);
 
+	if (thread == NULL) {
+		thread = machine__findnew_thread(machine, event->comm.pid,
+						 event->comm.tid);
+	}
 	if (thread == NULL ||
 	    __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
 		dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
@@ -1314,8 +1318,13 @@ int machine__process_mmap2_event(struct machine *machine,
 		return 0;
 	}
 
-	thread = machine__findnew_thread(machine, event->mmap2.pid,
+	thread = machine__find_thread(machine, event->mmap2.pid,
 					event->mmap2.tid);
+	if (thread == NULL) {
+		thread = machine__findnew_thread(machine,
+						 event->mmap2.pid,
+						 event->mmap2.tid);
+	}
 	if (thread == NULL)
 		goto out_problem;
 
@@ -1368,8 +1377,12 @@ int machine__process_mmap_event(struct machine *machine, union perf_event *event
 		return 0;
 	}
 
-	thread = machine__findnew_thread(machine, event->mmap.pid,
+	thread = machine__find_thread(machine, event->mmap.pid,
 					 event->mmap.tid);
+	if (thread == NULL) {
+		thread = machine__findnew_thread(machine, event->mmap.pid,
+						 event->mmap.tid);
+	}
 	if (thread == NULL)
 		goto out_problem;
 
-- 
2.6.2