Re: [PATCH 14/14] tracing: fix regression of perf function tracing

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

On 2013/3/27 21:09, Steven Rostedt wrote:
> On Wed, 2013-03-27 at 17:44 +0800, zhangwei(Jovi) wrote:
>> From: "zhangwei(Jovi)" <jovi.zhangwei@huawei.com>
>>
>> Using perf command: perf stat -e ftrace:function ls
>>
>> this will cause kernel warning and oops.
>>
>> [  797.828904] ------------[ cut here ]------------
>> [  797.828946] WARNING: at include/linux/ftrace.h:209 ftrace_ops_control_func+0xb1/0xc0()
>> [  797.829065] Pid: 6086, comm: perf Not tainted 3.8.0-rc4+ #88
>> [  797.829066] Call Trace:
>> [  797.829078]  [<c0447a42>] warn_slowpath_common+0x72/0xa0
>> [  797.829080]  [<c04bd6d1>] ? ftrace_ops_control_func+0xb1/0xc0
>> [  797.829082]  [<c04bd6d1>] ? ftrace_ops_control_func+0xb1/0xc0
>> [  797.829083]  [<c04b7453>] ? synchronize_sched+0x3/0x50
>> [  797.829085]  [<c04be50f>] ? __unregister_ftrace_function+0x8f/0x170
>> [  797.829087]  [<c0447a92>] warn_slowpath_null+0x22/0x30
>> [  797.829089]  [<c04bd6d1>] ftrace_ops_control_func+0xb1/0xc0
>> [  797.829099]  [<c08eba6b>] ftrace_call+0x5/0xb
>> [  797.829100]  [<c04b7458>] ? synchronize_sched+0x8/0x50
>> [  797.829102]  [<c04be50f>] __unregister_ftrace_function+0x8f/0x170
>> [  797.829104]  [<c04c00cf>] unregister_ftrace_function+0x1f/0x50
>> [  797.829109]  [<c04d31bd>] perf_ftrace_event_register+0x9d/0x140
>> [  797.829111]  [<c04d304b>] perf_trace_destroy+0x2b/0x50
>> [  797.829117]  [<c04db3c8>] tp_perf_event_destroy+0x8/0x10
>> [  797.829119]  [<c04dd672>] free_event+0x42/0x110
>> [  797.829121]  [<c04de446>] perf_event_release_kernel+0x56/0x90
>> [  797.829122]  [<c04de4fc>] put_event+0x7c/0xa0
>> [  797.829124]  [<c04de5cb>] perf_release+0xb/0x10
>> [  797.829128]  [<c0532ae6>] __fput+0xc6/0x1f0
>> [  797.829130]  [<c0532c1d>] ____fput+0xd/0x10
>> [  797.829134]  [<c04645b1>] task_work_run+0x81/0xa0
>> [  797.829142]  [<c0412819>] do_notify_resume+0x59/0x90
>> [  797.829150]  [<c08e4745>] work_notifysig+0x30/0x37
>> [  797.829152] ---[ end trace 4dbd63f12b55163f ]---
>>
>> This bug was introduced by below commit(in 3.8-rc4):
>>         commit 0a016409e42f273415f8225ddf2c58eb2df88034
>>         Author: Steven Rostedt <srostedt@redhat.com>
>>         Date:   Fri Nov 2 17:03:03 2012 -0400
>>
>>         ftrace: Optimize the function tracer list loop
>>
>> When variable op is ftrace_list_end, it cannot pass control ops checking,
>> so that loop optimize is not suit for ftrace_control_list, change it back.
>>
> 
> Thanks for finding the bug, but this isn't the fix.
> 
>> Signed-off-by: zhangwei(Jovi) <jovi.zhangwei@huawei.com>
>> Cc: stable@vger.kernel.org
>> ---
>>  kernel/trace/ftrace.c |    7 +++++--
>>  1 file changed, 5 insertions(+), 2 deletions(-)
>>
>> diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
>> index 2577082..2899974 100644
>> --- a/kernel/trace/ftrace.c
>> +++ b/kernel/trace/ftrace.c
>> @@ -4185,11 +4185,14 @@ ftrace_ops_control_func(unsigned long ip, unsigned long parent_ip,
>>  	 */
>>  	preempt_disable_notrace();
>>  	trace_recursion_set(TRACE_CONTROL_BIT);
>> -	do_for_each_ftrace_op(op, ftrace_control_list) {
> 
> I'd much rather keep the do_for_each_op() as it keeps everything
> consistent, and I don't have to worry about this if I change things in
> the future.
I changed this on thought of this would save a duplicate checking in hot path.
But never mind, it's a only tiny change, either way don't have big issue.
> 
> I'll fix this so that it wont process stub functions.
> 
> Thanks!
> 
> -- Steve
> 
> 
>> +	op = rcu_dereference_raw(ftrace_control_list);
>> +	while (op != &ftrace_list_end) {
>>  		if (!ftrace_function_local_disabled(op) &&
>>  		    ftrace_ops_test(op, ip))
>>  			op->func(ip, parent_ip, op, regs);
>> -	} while_for_each_ftrace_op(op);
>> +
>> +		op = rcu_dereference_raw(op->next);
>> +	}
>>  	trace_recursion_clear(TRACE_CONTROL_BIT);
>>  	preempt_enable_notrace();
>>  }
> 
> 
> 
> .
>