Re: [PATCHv3 0/2] tracing: function graph output for preempt/irqs-off tracers

[Jiri Olsa <jolsa@redhat.com>]

On Fri, Mar 26, 2010 at 10:53:59AM -0400, Steven Rostedt wrote:
> On Fri, 2010-03-26 at 13:32 +0100, Jiri Olsa wrote:

SNIP

> 
> Getting better, but I still have some issues ;-)
> 
> Without graph, header looks like this: (old way)
> 
> # tracer: irqsoff
> #
> # irqsoff latency trace v1.1.5 on 2.6.34-rc2
> # --------------------------------------------------------------------
> # latency: 326 us, #73/73, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0
> #P:4)
> #    -----------------
> #    | task: swapper-0 (uid:0 nice:0 policy:0 rt_prio:0)
> #    -----------------
> #  => started at: save_args
> #  => ended at:   __do_softirq
> #
> #
> #                  _------=> CPU#            
> #                 / _-----=> irqs-off        
> #                | / _----=> need-resched    
> #                || / _---=> hardirq/softirq 
> #                ||| / _--=> preempt-depth   
> #                |||| /_--=> lock-depth       
> #                |||||/     delay             
> #  cmd     pid   |||||| time  |   caller      
> #     \   /      ||||||   \   |   /           
>   <idle>-0       1d..1.    1us+: trace_hardirqs_off_thunk <-save_args
> 
> 
> Lots of good info.

right, I missed the print_trace_header call..

SNIP

>  # echo 0 > /debug/tracing/tracing_enabled
>  # echo 0 > /debug/tracing/option/display-graph
>  # cat /debug/tracing/trace
> 
> # tracer: irqsoff
> irqbalan-2672    0d..2.   55us+: Unknown type 13
> irqbalan-2672    0d.h2.   62us+: Unknown type 13

I forgot the max_tr buffer is actually the one displayed,
so it needs reset as well when the display-graph option
is switched on/off.


> 
> 
> I think you can still do the "event" part, without effecting the way the
> function graph outputs normally. I would not have given up on that
> method.  You don't need to worry about it processing other events,
> because when you register it to write as an event, it will only be
> called when a function graph event was found. It will not be processing
> other events. Only when the tracer itself overrides the default writing
> will it do so.

The events would be called only for TRACE_GRAPH_RET, TRACE_GRAPH_ENT entries and
not for others, thats right.

However it's the graph ouput code that outputs other events' text
within "/*" and "*/".

So using the event way, all other events would be printed as normal
events(standard lines not alligned) with the standard header...
not like comments, as they are in the function_graph tracer.

I thought it'd be good for graph output to stay the same in irqsoff
tracer as in function_graph tracer.. if that is not the concern
the event way would be probably nicer :)

I'm sending updated patchset with above 2 fixies right away,
I can do/resend the event way later if needed.

thanks,
jirka