Re: [PATCH 03/13] tracing: Move a printk out of ftrace_raw_reg_event_foo()

[Frederic Weisbecker <fweisbec@gmail.com>]

On Tue, Dec 08, 2009 at 03:49:27PM +0800, Li Zefan wrote:
> >> -static void ftrace_event_enable_disable(struct ftrace_event_call *call,
> >> +static int ftrace_event_enable_disable(struct ftrace_event_call *call,
> >>  					int enable)
> >>  {
> >> +	int ret = 0;
> >> +
> >>  	switch (enable) {
> >>  	case 0:
> >>  		if (call->enabled) {
> >> @@ -131,12 +133,19 @@ static void ftrace_event_enable_disable(struct ftrace_event_call *call,
> >>  		break;
> >>  	case 1:
> >>  		if (!call->enabled) {
> >> +			ret = call->regfunc(call);
> >> +			if (ret) {
> >> +				pr_info("event trace: Could not enable event "
> >> +					"%s\n", call->name);
> >> +				break;
> >> +			}
> >>  			call->enabled = 1;
> >>  			tracing_start_cmdline_record();
> >> -			call->regfunc(call);
> > 
> > That seems to open a tiny window during which we'll lose
> > some pid -> comm resolution records.
> > 
> > We should probably keep the previous call order.
> > 
> 
> Hm, actually the tiny window is already there in that
> tracing_stop_cmdline_record() is called before call->unregfunc()..
> 
> So I can send a fix-up patch later.


The presence of a previous race window doesn't justify we add
a new one :)

I'll fix that while applying the patch.

Thanks.