Re: [RFC][PATCH] tracing: Warn and notify if tracepoints are not loaded due to module taint

[Mathieu Desnoyers <mathieu.desnoyers@efficios.com>]

----- Original Message -----
> From: "Steven Rostedt" <rostedt@goodmis.org>
> To: "Mathieu Desnoyers" <mathieu.desnoyers@efficios.com>
> Cc: "Peter Zijlstra" <peterz@infradead.org>, "LKML" <linux-kernel@vger.kernel.org>, "Ingo Molnar" <mingo@kernel.org>,
> "Rusty Russell" <rusty@rustcorp.com.au>, "Frederic Weisbecker" <fweisbec@gmail.com>, "Andrew Morton"
> <akpm@linux-foundation.org>
> Sent: Wednesday, February 26, 2014 11:15:42 AM
> Subject: Re: [RFC][PATCH] tracing: Warn and notify if tracepoints are not loaded due to module taint
> 
> On Wed, 26 Feb 2014 12:48:12 +0000 (UTC)
> Mathieu Desnoyers <mathieu.desnoyers@efficios.com> wrote:
> 
> > > How about instead of a WARN, you use a normal KERN_ERR printk(). There's
> > > no point to the entire WARN state dump, that's needlessly verbose.
> > > 
> > > When you have a normal error print you can have as many as are required
> > > and put the mod name back in.
> > 
> > The good old printk KERN_ERR is a very good idea. I agree that WARN() is
> > too verbose for our needs here.
> 
> Actually, it's not so bad for the WARN() after my last patch to only
> allocate (or even process tracepoints) if mod->num_tracepionts is
> greater than zero. I didn't realize you were wasting memory for all
> modules that were loaded.
> 
> My fear with the KERN_ERR is that it wont be noticeable enough. Where
> as a stack dump is something that will catch people's attention.
> 
> And as Rusty has said, if you are loading a module that is forced, or
> something strange, it is broken. The failure of loading the tracepoints
> of a module is a bug if the module happens to have tracepoints.
> 
> After the MOD_SIG fix, any failure should be a big banner bug. Either
> they are using a forced module with tracepoints that should not be
> loaded. Or they have tracepoints is a non-GPL module (which is also a
> big no-no).

Agreed that after the skip for modules containing 0 tracepoints, it gets
much more specific. I like that.

So then a WARN_ON() that prints the specific module name involved would
be the way to go ?

Thanks,

Mathieu

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com