lkml.org 
[lkml]   [2009]   [Feb]   [25]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 2/4] tracing: add event trace infrastructure

On Wed, 25 Feb 2009, Andrew Morton wrote:

> On Wed, 25 Feb 2009 11:00:53 +0200 Pekka Enberg <penberg@cs.helsinki.fi> wrote:
> Parse this:
>
> irqsoff latency trace v1.1.5 on 2.6.26-rc8
> --------------------------------------------------------------------
> latency: 97 us, #3/3, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2)
> -----------------
> | task: swapper-0 (uid:0 nice:0 policy:0 rt_prio:0)
> -----------------
> => started at: apic_timer_interrupt
> => ended at: do_softirq
>
> # _------=> CPU#
> # / _-----=> irqs-off
> # | / _----=> need-resched
> # || / _---=> hardirq/softirq
> # ||| / _--=> preempt-depth
> # |||| /
> # ||||| delay
> # cmd pid ||||| time | caller
> # \ / ||||| \ | /
> <idle>-0 0d..1 0us+: trace_hardirqs_off_thunk (apic_timer_interrupt)
> <idle>-0 0d.s. 97us : __do_softirq (do_softirq)
> <idle>-0 0d.s1 98us : trace_hardirqs_on (do_softirq)
>
> your time starts now.

Note, that is a legacy format, that works great for the latency tracers.
Those include (the example you used) irqsoff, the preempt off, wake up
latencies. This gives us a nice listing of why we are hitting a latency.
All other plugins generally use the trace file, although they could also
use that file too.

As for your English comment. I showed the header to my wife, and she had
no idea what irqs-off means, nor need-resched, nor preempt-depth ;-)
They are not quite English as they are technical terms. Most of the names
are hard coded into the kernel too.

Yeah, maybe that file (latency_trace) is a bit too much. I for one love
it. It is great to load a kernel on some remote box, and run the irqs off
latency tracer to see where the interrupts are disabled for the longest
time. This format is really nice because it shows me when we are in an
interrupt, or interrupts are disabled, and when the task should have been
rescheduled.

This has help find places that we miss a preemption check too.

If I had to also include a user space program, I would need to tell the
customer how to install it, and run it. Just telling someone to mount a
directory, run your test, cat a file to some temp file and send me the
temp file is much easier. Because most admins know how to do such things.
No learning curve, except what to mount and what to cat.

I've given a few talks on ftrace. And when I'm approached afterward by
people (usually the ones not paying attention to the speech ;-) asking
what utils they need to install. You should see the smile on their faces
when I tell them, "All you need is what is in busybox". Then they say,
"No, really? What do I need to use ftrace?" I again tell them, a kernel
with ftrace and a minimum requirement of what is in busybox: cat, grep,
echo. That is what makes ftrace powerful and robust.

Once I break that "busybox" rule. ftrace will be used by only a quarter of
those that use it now (and probably less).

-- Steve






\
 
 \ /
  Last update: 2009-02-25 15:43    [W:0.124 / U:0.516 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site