Re: [PATCH 1/3] tracing/kprobes: Avoid perf_trace_buf_*() if ->perf_events is empty

["zhangwei(Jovi)" <jovi.zhangwei@huawei.com>]

On 2013/6/17 1:21, Oleg Nesterov wrote:
> perf_trace_buf_prepare() + perf_trace_buf_submit() make no sense
> if this task/CPU has no active counters. Change kprobe_perf_func()
> and kretprobe_perf_func() to check call->perf_events beforehand
> and return if this list is empty.
> 
> For example, "perf record -e some_probe -p1". Only /sbin/init will
> report, all other threads which hit the same probe will do
> perf_trace_buf_prepare/perf_trace_buf_submit just to realize that
> nobody wants perf_swevent_event().
> 
> Signed-off-by: Oleg Nesterov <oleg@redhat.com>

Good point, I think we also need to change other places in below patch.

After applied the patch, perf_tp_event() function call reduced a lots
when using task based perf tracing.


-----------------------------------


tracing: Avoid perf_trace_buf_*() if ->perf_events is empty


Signed-off-by: zhangwei(Jovi) <jovi.zhangwei@huawei.com>
---
 include/trace/ftrace.h        |    5 ++++-
 kernel/trace/trace_syscalls.c |   10 ++++++++--
 2 files changed, 12 insertions(+), 3 deletions(-)

diff --git a/include/trace/ftrace.h b/include/trace/ftrace.h
index 19edd7f..5d340f5 100644
--- a/include/trace/ftrace.h
+++ b/include/trace/ftrace.h
@@ -659,6 +659,10 @@ perf_trace_##call(void *__data, proto)					\
 	int __data_size;						\
 	int rctx;							\
 									\
+	head = this_cpu_ptr(event_call->perf_events);			\
+	if (hlist_empty(head))						\
+		return;							\
+									\
 	perf_fetch_caller_regs(&__regs);				\
 									\
 	__data_size = ftrace_get_offsets_##call(&__data_offsets, args); \
@@ -679,7 +683,6 @@ perf_trace_##call(void *__data, proto)					\
 									\
 	{ assign; }							\
 									\
-	head = this_cpu_ptr(event_call->perf_events);			\
 	perf_trace_buf_submit(entry, __entry_size, rctx, __addr,	\
 		__count, &__regs, head, __task);			\
 }
diff --git a/kernel/trace/trace_syscalls.c b/kernel/trace/trace_syscalls.c
index 8f2ac73..28debf4 100644
--- a/kernel/trace/trace_syscalls.c
+++ b/kernel/trace/trace_syscalls.c
@@ -553,6 +553,10 @@ static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
 	if (!sys_data)
 		return;

+	head = this_cpu_ptr(sys_data->enter_event->perf_events);
+	if (hlist_empty(head))
+		return;
+
 	/* get the size after alignment with the u32 buffer size field */
 	size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 	size = ALIGN(size + sizeof(u32), sizeof(u64));
@@ -571,7 +575,6 @@ static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
 	syscall_get_arguments(current, regs, 0, sys_data->nb_args,
 			       (unsigned long *)&rec->args);

-	head = this_cpu_ptr(sys_data->enter_event->perf_events);
 	perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head, NULL);
 }

@@ -629,6 +632,10 @@ static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
 	if (!sys_data)
 		return;

+	head = this_cpu_ptr(sys_data->exit_event->perf_events);
+	if (hlist_empty(head))
+		return;
+
 	/* We can probably do that at build time */
 	size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
 	size -= sizeof(u32);
@@ -649,7 +656,6 @@ static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
 	rec->nr = syscall_nr;
 	rec->ret = syscall_get_return_value(current, regs);

-	head = this_cpu_ptr(sys_data->exit_event->perf_events);
 	perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head, NULL);
 }

-- 
1.7.9.7