Messages in this thread | | | Date | Mon, 24 Aug 2009 14:13:22 +0800 | From | Li Zefan <> | Subject | Re: system gets stuck in a lock during boot |
| |
Peter Zijlstra wrote: > On Mon, 2009-08-24 at 10:41 +0800, Li Zefan wrote: > >>> @@ -940,7 +940,7 @@ event_create_dir(struct ftrace_event_call *call, struct dentr >>> entry = trace_create_file("enable", 0644, call->dir, call, >>> enable); >>> >>> - if (call->id) >>> + if (call->id && call->profile_enable) >> We do an extra check on ->profile_enable, shouldn't cause bug.. >> >>> entry = trace_create_file("id", 0444, call->dir, call, >>> id); >> Any way, I don't think this commit does the right thing: >> >> - If CONFIG_EVENT_PROFILE=y, we'll create events/<dir>/<event>/id, >> except events/ftrace/<event>/id. >> >> - if CONFIG_EVENT_PROFILE=n, there's no 'id' file at all! >> >> I think it's better to skip ftrace/ dir in perf tool code, instead of >> skipping creating id files in ftrace code. > > No, it does do the right thing. Your patch breaks things because not all > tracepoints are created through TRACE_EVENT() and will thus not have > their profile_enable/disable hooks set. > > By giving them an ID file, there is no way to distinguish good from bad > tracepoints. >
But removing the id file from events/ftrace/ might break some ftrace binary parsers?
And this commit makes 'id' file disapeared with CONFIG_EVENT_PROFILE=n.
> Expempting ftrace is no solid solution, suppose someone else does a > TRACE_EVENT_FORMAT() tracepoint, how would you know you could use it as > a profile event source? >
Agree.
> The id files really must stay conditional. >
I don't think it's a good idea to connect it with perfcounter this way. Wouldn't adding a 'profilable' file more intuitive?
> Aside from that, you only add #ifdef fuzz back which with his config is > moot, and shouldn't result in anything but a slightly bigger structure > to begin with. >
Yeah, actually I still fail to see where the bug is, maybe the commit that Justin bisected down is not the real culprit..
| |