Re: [ANNOUNCE] New utility: 'trace'

From: Frederic Weisbecker <fweisbec@gmail.com>
To: "Ted Ts'o" <tytso@mit.edu>
Cc: Peter Zijlstra <peterz@infradead.org>,
	Ingo Molnar <mingo@elte.hu>, Darren Hart <dvhart@linux.intel.com>,
	Thomas Gleixner <tglx@linutronix.de>,
	LKML <linux-kernel@vger.kernel.org>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	Steven Rostedt <rostedt@goodmis.org>,
	Arjan van de Ven <arjan@infradead.org>,
	Arnaldo Carvalho de Melo <acme@redhat.com>,
	Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>,
	Tom Zanussi <tzanussi@gmail.com>,
	Mathieu Desnoyers <mathieu.desnoyers@efficios.com>,
	Li Zefan <lizf@cn.fujitsu.com>, Jason Baron <jbaron@redhat.com>,
	"David S. Miller" <davem@davemloft.net>,
	Christoph Hellwig <hch@lst.de>, Pekka Enberg <penberg@kernel.org>,
	Lai Jiangshan <laijs@cn.fujitsu.com>,
	Eric Dumazet <eric.dumazet@gmail.com>
Subject: Re: [ANNOUNCE] New utility: 'trace'
Date: Wed, 17 Nov 2010 19:29:02 +0100	[thread overview]
Message-ID: <20101117182859.GB5352@nowhere> (raw)
In-Reply-To: <20101117181340.GF3290@thunk.org>

On Wed, Nov 17, 2010 at 01:13:40PM -0500, Ted Ts'o wrote:
> On Wed, Nov 17, 2010 at 03:10:33PM +0100, Frederic Weisbecker wrote:
> > Yeah I have a strange workflow. I'm working on that CPU isolation thing
> > and I have dozens of trace_printk all over the place for tons of
> > things. And everytime I remove one to unwind some output or to focus
> > on another one, I often have to restore it later because I need it
> > again. Usually I even just comment it out instead of removing it.
> 
> What I do for my file system development work is to drop in
> trace_printk's everywhere, since they are lightweight and don't slow
> down the system much.  Then when the system wedges, I use sysrq-z to
> get a "flight data recorder" printout of what happened up til the
> system hung.
> 
> I love the fact that the ring buffer is in "overwrite" mode (aka
> flight data recorder mode), and hope this doesn't go away.
> 
> Per line filtering is also great, but very often when I'm interacting
> with the block I/O layer, if something screws up, what happens is "and
> then whole machine locks up", and sysrq-z is literally all I have. 

Yeah all agreed.

Steve proposed to keep the current trace_printk() implementation that relies
on ftrace but rename in into ftrace_printk(). So that we can develop a new
trace_printk() based on trace_event interface and in the meantime keep the
old version in case something messes up with the new thing.