Re: [PATCH v2 1/2] tracing: Move tracepoint callbacks into DEFINE

[Frederic Weisbecker <fweisbec@gmail.com>]

On Fri, Aug 21, 2009 at 04:06:44PM -0400, Jason Baron wrote:
> On Fri, Aug 21, 2009 at 12:34:28PM -0700, Josh Stone wrote:
> > On 08/21/2009 10:52 AM, Jason Baron wrote:
> > > this means that when CONFIG_EVENT_TRACING is set, the 'generic' syscall
> > > enter/exit will show up as events in the debugfs, but enabling them
> > > wouldn't do anything. I think we should simply drop the
> > > 'CONFIG_FTRACE_SYSCALLS' 'ifdef' and 'else' clause. That will give us
> > > what we want - tying these callbacks directly to tracepoint.
> > 
> > But only x86 and s390 have TIF_SYSCALL_FTRACE, while kernel/tracepoint.c
> > must still compile everywhere that has CONFIG_TRACEPOINTS=y.
> > 
> 
> ok.
> 
> > Maybe it would be better to make that #ifdef TIF_SYSCALL_FTRACE, and
> > then also #ifdef the TRACE_EVENT declaration, so it will only show up on
> > kernels that actually support it.
> > 
> 
> hmm...maybe then we move these functions to an arch specific file, such
> as arch/x86{s390}/kernel/ftrace.c, making the code contigent on
> CONFIG_TRACEPOINTS? (I think that required less ifdefs...).


But the code of these functions is generic and should then keep beeing
factorized.

We could put them in kernel/trace/trace_syscalls.c and keep ftrace related
things inside CONFIG_TRACE_SYSCALLS and the above inside
CONFIG_TRACEPOINT + CONFIG_HAVE_SYSCALL_TRACING (then also change
the Makefile).

Or if you find that too dirty, may be create a separate file in kernel/trace/.


> 
> > Also, since this event is now usable outside of ftrace, would you object
> > to renaming the flag TIF_SYSCALL_TRACEPOINT?
> > 
> 
> sure.
> 
> thanks,
> 
> -Jason