tracing: Allow to disable cmdline recording

We found that even enabling a single trace event that will rarely be
triggered can add big overhead to context switch.

(lmbench context switch test)
 -------------------------------------------------
 2p/0K 2p/16K 2p/64K 8p/16K 8p/64K 16p/16K 16p/64K
 ctxsw  ctxsw  ctxsw ctxsw  ctxsw   ctxsw   ctxsw
------ ------ ------ ------ ------ ------- -------
  2.19   2.3   2.21   2.56   2.13     2.54    2.07
  2.39   2.51  2.35   2.75   2.27     2.81    2.24

The overhead is 6% ~ 11%.

It's because when a trace event is enabled 3 tracepoints (sched_switch,
sched_wakeup, sched_wakeup_new) will be activated to map pid to cmdname.

We'd like to avoid this overhead, so add a trace option '(no)record-cmd'
to allow to disable cmdline recording.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
LKML-Reference: <4C2D57F4.2050204@cn.fujitsu.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

4 files changed, 41 insertions, 5 deletions

diff --git a/include/linux/ftrace_event.h b/include/linux/ftrace_event.h
index 01df7ca4ead7..2b7b1395b4d5 100644
--- a/include/linux/ftrace_event.h
+++ b/include/linux/ftrace_event.h
@@ -152,11 +152,13 @@ extern int ftrace_event_reg(struct ftrace_event_call *event,
 enum {
         TRACE_EVENT_FL_ENABLED_BIT,
         TRACE_EVENT_FL_FILTERED_BIT,
+        TRACE_EVENT_FL_RECORDED_CMD_BIT,
 };
 
 enum {
-        TRACE_EVENT_FL_ENABLED  = (1 << TRACE_EVENT_FL_ENABLED_BIT),
-        TRACE_EVENT_FL_FILTERED = (1 << TRACE_EVENT_FL_FILTERED_BIT),
+        TRACE_EVENT_FL_ENABLED          = (1 << TRACE_EVENT_FL_ENABLED_BIT),
+        TRACE_EVENT_FL_FILTERED         = (1 << TRACE_EVENT_FL_FILTERED_BIT),
+        TRACE_EVENT_FL_RECORDED_CMD     = (1 << TRACE_EVENT_FL_RECORDED_CMD_BIT),
 };
 
 struct ftrace_event_call {
@@ -174,6 +176,7 @@ struct ftrace_event_call {
          * 32 bit flags:
          *   bit 1:             enabled
          *   bit 2:             filter_active
+         *   bit 3:             enabled cmd record
          *
          * Changes to flags must hold the event_mutex.
          *
diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c
index 8683dec6946b..af9042977c08 100644
--- a/kernel/trace/trace.c
+++ b/kernel/trace/trace.c
@@ -344,7 +344,7 @@ static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
 /* trace_flags holds trace_options default values */
 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
         TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
-        TRACE_ITER_GRAPH_TIME;
+        TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD;
 
 static int trace_stop_count;
 static DEFINE_SPINLOCK(tracing_start_lock);
@@ -428,6 +428,7 @@ static const char *trace_options[] = {
         "latency-format",
         "sleep-time",
         "graph-time",
+        "record-cmd",
         NULL
 };
 
@@ -2561,6 +2562,9 @@ static void set_tracer_flags(unsigned int mask, int enabled)
                 trace_flags |= mask;
         else
                 trace_flags &= ~mask;
+
+        if (mask == TRACE_ITER_RECORD_CMD)
+                trace_event_enable_cmd_record(enabled);
 }
 
 static ssize_t
diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index 84d3f123e86f..7778f067fc8b 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h
@@ -591,6 +591,7 @@ enum trace_iterator_flags {
         TRACE_ITER_LATENCY_FMT          = 0x20000,
         TRACE_ITER_SLEEP_TIME           = 0x40000,
         TRACE_ITER_GRAPH_TIME           = 0x80000,
+        TRACE_ITER_RECORD_CMD           = 0x100000,
 };
 
 /*
@@ -723,6 +724,8 @@ filter_check_discard(struct ftrace_event_call *call, void *rec,
         return 0;
 }
 
+extern void trace_event_enable_cmd_record(bool enable);
+
 extern struct mutex event_mutex;
 extern struct list_head ftrace_events;
 
diff --git a/kernel/trace/trace_events.c b/kernel/trace/trace_events.c
index e8e6043f4d29..09b4fa6e4d3b 100644
--- a/kernel/trace/trace_events.c
+++ b/kernel/trace/trace_events.c
@@ -170,6 +170,26 @@ int ftrace_event_reg(struct ftrace_event_call *call, enum trace_reg type)
 }
 EXPORT_SYMBOL_GPL(ftrace_event_reg);
 
+void trace_event_enable_cmd_record(bool enable)
+{
+        struct ftrace_event_call *call;
+
+        mutex_lock(&event_mutex);
+        list_for_each_entry(call, &ftrace_events, list) {
+                if (!(call->flags & TRACE_EVENT_FL_ENABLED))
+                        continue;
+
+                if (enable) {
+                        tracing_start_cmdline_record();
+                        call->flags |= TRACE_EVENT_FL_RECORDED_CMD;
+                } else {
+                        tracing_stop_cmdline_record();
+                        call->flags &= ~TRACE_EVENT_FL_RECORDED_CMD;
+                }
+        }
+        mutex_unlock(&event_mutex);
+}
+
 static int ftrace_event_enable_disable(struct ftrace_event_call *call,
                                         int enable)
 {
@@ -179,13 +199,19 @@ static int ftrace_event_enable_disable(struct ftrace_event_call *call,
         case 0:
                 if (call->flags & TRACE_EVENT_FL_ENABLED) {
                         call->flags &= ~TRACE_EVENT_FL_ENABLED;
-                        tracing_stop_cmdline_record();
+                        if (call->flags & TRACE_EVENT_FL_RECORDED_CMD) {
+                                tracing_stop_cmdline_record();
+                                call->flags &= ~TRACE_EVENT_FL_RECORDED_CMD;
+                        }
                         call->class->reg(call, TRACE_REG_UNREGISTER);
                 }
                 break;
         case 1:
                 if (!(call->flags & TRACE_EVENT_FL_ENABLED)) {
-                        tracing_start_cmdline_record();
+                        if (trace_flags & TRACE_ITER_RECORD_CMD) {
+                                tracing_start_cmdline_record();
+                                call->flags |= TRACE_EVENT_FL_RECORDED_CMD;
+                        }
                         ret = call->class->reg(call, TRACE_REG_REGISTER);
                         if (ret) {
                                 tracing_stop_cmdline_record();