Date: Mon, 16 Mar 2009 15:23:05 -0400
From: Mathieu Desnoyers <compudj@krystal.dyndns.org>
To: Steven Rostedt <rostedt@goodmis.org>
Cc: Jason Baron <jbaron@redhat.com>, mingo@elte.hu,
       linux-kernel@vger.kernel.org, acme@ghostprotocols.net,
       fweisbec@gmail.com, fche@redhat.com, peterz@infradead.org
Subject: Re: [Patch 2/2] tracepoints for softirq entry/exit - tracepoints
Message-ID: <20090316192305.GB11878@Krystal>
References: <20090312183603.GC3352@redhat.com> <20090314025701.GC22526@Krystal> <alpine.DEB.2.00.0903160934010.26431@gandalf.stny.rr.com> <20090316183739.GA10292@Krystal> <alpine.DEB.2.00.0903161441210.27978@gandalf.stny.rr.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Disposition: inline
In-Reply-To: <alpine.DEB.2.00.0903161441210.27978@gandalf.stny.rr.com>
User-Agent: Mutt/1.5.18 (2008-05-17)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2040
Lines: 50

* Steven Rostedt (rostedt@goodmis.org) wrote:
> 
> On Mon, 16 Mar 2009, Mathieu Desnoyers wrote:
> > > > 
> > > > The softirq tracepoints are a good idea indeed (I have similar ones in
> > > > the LTTng tree). My main concern is about the fact that you output the
> > > > softirq name in plain text to the trace buffers. I would rather prefer
> > > > to save only the softirq (h-vec) into the trace and dump the mapping
> > > > (h-vec) to name only once, so we can save precious trace bytes.
> > > 
> > > The TP_FMT is only used by those tracers that want to use it. Any tracer 
> > > can still hook directly to the trace point and do what every they want.
> > > 
> > > -- Steve
> > > 
> > 
> > By doing so, you are removing the ability to use the TP_FMT information
> > to perform high-speed system-wide tracing. I thought the goal was to
> > create a unified buffering, but sadly I don't see the high-speed
> > requirements being part of that plan.
> 
> TP_FMT has nothing to do with the unified buffering. The unified buffer 
> does not even know about it. But if you want high-speed event tracing, 
> that is what the TRACE_EVENT was created for.
> 
> The TRACE_FORMAT was made for things that will be recording string 
> information anyway, and recording a string into the buffer via memcpy or a 
> sprintf format (binary printk) doesn't make much difference.
> 

Are you saying that dynamically parsing a format string in a binary
printk has the same performance impact as a memcpy ? I would be very
interested to see your benchmarks.

Mathieu

> Then trace points for entry and exit does not fall into that category, and 
> should be represented by a TRACE_EVENT instead.
> 
> -- Steve
> 

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/