Re: [PATCH RFC net-next] trace: tcp: Add tracepoint for tcp_cwnd_reduction()

[Yonghong Song]

On 1/22/25 6:56 AM, Steven Rostedt wrote:
> On Wed, 22 Jan 2025 01:39:42 -0800
> Breno Leitao <leitao@xxxxxxxxxx> wrote:
>
> > Right, DECLARE_TRACE would solve my current problem, but, a056a5bed7fa
> > ("sched/debug: Export the newly added tracepoints") says "BPF doesn't
> > have infrastructure to access these bare tracepoints either.".
> >
> > Does BPF know how to attach to this bare tracepointers now?
> >
> > On the other side, it seems real tracepoints is getting more pervasive?
> > So, this current approach might be OK also?
> >
> > 	https://lore.kernel.org/bpf/20250118033723.GV1977892@ZenIV/T/#m4c2fb2d904e839b34800daf8578dff0b9abd69a0
> Thanks for the pointer. I didn't know this discussion was going on. I just
> asked to attend if this gets accepted. I'm only a 6 hour drive from
> Montreal anyway.
>
> > > You can see its use in include/trace/events/sched.h
> > I suppose I need to export the tracepointer with
> > EXPORT_TRACEPOINT_SYMBOL_GPL(), right?
> For modules to use them directly, yes. But there's other ways too.
>
> > I am trying to hack something as the following, but, I struggled to hook
> > BPF into it.
> Maybe you can use the iterator to search for the tracepoint.
>
> #include <linux/tracepoint.h>
>
> static void fct(struct tracepoint *tp, void *priv)
> {
> 	if (!tp->name || strcmp(tp->name, "<tracepoint_name>") != 0)
> 		return 0;
>
> 	// attach to tracepoint tp
> }
>
> [..]
> 	for_each_kernel_tracepoint(fct, NULL);
>
> This is how LTTng hooks to tracepoints.

The LTTng approach in the above needs a kernel module to enable and disable
the tracepoint and this is not a bpf-way to handle tracepoints.

So for bpf, we need a new UAPI to pass <tracepoint_name> from user
space to the kernel to attach to tracepoint tp since <tracepont_name> is not
available in trace_fs.

What is the criteria for a tracepoint to be a normal tp or a bare tp?


> -- Steve