Re: [PATCH 0/5] [RFC] binary reading of ftrace ring buffers

[Peter Zijlstra <peterz@infradead.org>]

On Wed, 2009-03-04 at 12:00 -0500, Mathieu Desnoyers wrote:

> > TRACE_EVENT_FORMAT(sched_switch,
> > 	TPPROTO(struct rq *rq, struct task_struct *prev,
> > 		struct task_struct *next),
> > 	TPARGS(rq, prev, next),
> > 	TPFMT("task %s:%d ==> %s:%d",
> > 	      prev->comm, prev->pid, next->comm, next->pid),
> > 	TRACE_STRUCT(
> > 		TRACE_FIELD(pid_t, prev_pid, prev->pid)
> > 		TRACE_FIELD(int, prev_prio, prev->prio)
> > 		TRACE_FIELD_SPECIAL(char next_comm[TASK_COMM_LEN],
> > 				    next_comm,
> > 				    TPCMD(memcpy(TRACE_ENTRY->next_comm,
> > 						 next->comm,
> > 						 TASK_COMM_LEN)))
> > 		TRACE_FIELD(pid_t, next_pid, next->pid)
> > 		TRACE_FIELD(int, next_prio, next->prio)
> > 	),
> > 	TPRAWFMT("prev %d:%d ==> next %s:%d:%d")
> > 	);
> > 
> 
> I fear that putting these user-visible format strings in tracepoint
> header files will create a big maintainability issue.
> 
> I'll post the LTTng patchset in a jiffy, where the format string
> awareness is done in a tracer-specific module. I don't understand why
> Peter Z. is not yelling against your tracepoints modifications : they
> are actually presenting to userspace an interface that is meant to
> eventually change.
> 
> I used a separate layer for format string presentation for this very
> purpose : I don't want to tie the kernel code instrumentation
> (tracepoints) to any kind of user-visible API.

Not only that, its butt-ugly to boot.

I'm just not sure what to say anymore.. people seem insistent on doing
this, look at the recent google trace and this zedtrace thing.

What I can live with is if we tie it 1:1 to our code structure by means
of code generation, in that case we can simply argue its a direct
representation of our code, and we don't have any API rules on that.

Arnaldo proposed using sparse and tree rewriting to accomplish this.

The above however creates these artificial things that can arguably be
kept static, therefore one could defend that expectation.

What's worse, it doesn't even come with version information :-)