How to write marker info in MMIO trace from kernel

How to write marker info in MMIO trace from kernel

[Larry Finger]

Stephen,

I hope you are the correct person for this question.

I would like to write information into the MMIO trace file similar to
what the user space write to /sys/.../trace_marker would accomplish,
however, I want to do it in place of printk statements in the driver I'm
tracing. Do you have any suggestions on how to do this? I have tools to
merge the trace dump with dmesg output, but the latter lags behind and I
do not get good correlation.

Thanks,

Larry

Re: How to write marker info in MMIO trace from kernel

[Steven Rostedt]

On Tue, 2010-05-04 at 13:36 -0500, Larry Finger wrote:
> Stephen,
> 
> I hope you are the correct person for this question.
> 
> I would like to write information into the MMIO trace file similar to
> what the user space write to /sys/.../trace_marker would accomplish,
> however, I want to do it in place of printk statements in the driver I'm
> tracing. Do you have any suggestions on how to do this? I have tools to
> merge the trace dump with dmesg output, but the latter lags behind and I
> do not get good correlation.

You can use "trace_printk()". I think that is what you are looking for.

-- Steve

Re: How to write marker info in MMIO trace from kernel

[Larry Finger]

On 05/04/2010 01:42 PM, Steven Rostedt wrote:
> On Tue, 2010-05-04 at 13:36 -0500, Larry Finger wrote:
>> Stephen,
>>
>> I hope you are the correct person for this question.
>>
>> I would like to write information into the MMIO trace file similar to
>> what the user space write to /sys/.../trace_marker would accomplish,
>> however, I want to do it in place of printk statements in the driver I'm
>> tracing. Do you have any suggestions on how to do this? I have tools to
>> merge the trace dump with dmesg output, but the latter lags behind and I
>> do not get good correlation.
> 
> You can use "trace_printk()". I think that is what you are looking for.

Thanks for the info. One other question - when I write only mmiotrace to
the current_tracer, the trace_printk() stuff does not end up in the
buffer. I can get it by using the "sysprof" tracer, but that has a lot
of stuff I do not want. Did I miss some configuration? My list of
available tracers is "blk kmemtrace mmiotrace wakeup_rt wakeup function
sysprof sched_switch nop".

Thanks,

Larry

Re: How to write marker info in MMIO trace from kernel

[Steven Rostedt]

On Tue, 2010-05-04 at 17:27 -0500, Larry Finger wrote:

> Thanks for the info. One other question - when I write only mmiotrace to
> the current_tracer, the trace_printk() stuff does not end up in the
> buffer. I can get it by using the "sysprof" tracer, but that has a lot
> of stuff I do not want. Did I miss some configuration? My list of
> available tracers is "blk kmemtrace mmiotrace wakeup_rt wakeup function
> sysprof sched_switch nop".

Ug, that's because the mmiotrace handles its own prints and ignores
anything it does not know about. Here, try this patch:

-- Steve

diff --git a/kernel/trace/trace_mmiotrace.c b/kernel/trace/trace_mmiotrace.c
index 017fa37..8652f28 100644
--- a/kernel/trace/trace_mmiotrace.c
+++ b/kernel/trace/trace_mmiotrace.c
@@ -282,7 +282,7 @@ static enum print_line_t mmio_print_line(struct trace_iterator *iter)
 	case TRACE_PRINT:
 		return mmio_print_mark(iter);
 	default:
-		return TRACE_TYPE_HANDLED; /* ignore unknown entries */
+		return TRACE_TYPE_UNHANDLED;
 	}
 }
 

Re: How to write marker info in MMIO trace from kernel

[Larry Finger]

On 05/04/2010 05:58 PM, Steven Rostedt wrote:
> On Tue, 2010-05-04 at 17:27 -0500, Larry Finger wrote:
> 
>> Thanks for the info. One other question - when I write only mmiotrace to
>> the current_tracer, the trace_printk() stuff does not end up in the
>> buffer. I can get it by using the "sysprof" tracer, but that has a lot
>> of stuff I do not want. Did I miss some configuration? My list of
>> available tracers is "blk kmemtrace mmiotrace wakeup_rt wakeup function
>> sysprof sched_switch nop".
> 
> Ug, that's because the mmiotrace handles its own prints and ignores
> anything it does not know about. Here, try this patch:
> 
> -- Steve
> 
> diff --git a/kernel/trace/trace_mmiotrace.c b/kernel/trace/trace_mmiotrace.c
> index 017fa37..8652f28 100644
> --- a/kernel/trace/trace_mmiotrace.c
> +++ b/kernel/trace/trace_mmiotrace.c
> @@ -282,7 +282,7 @@ static enum print_line_t mmio_print_line(struct trace_iterator *iter)
>  	case TRACE_PRINT:
>  		return mmio_print_mark(iter);
>  	default:
> -		return TRACE_TYPE_HANDLED; /* ignore unknown entries */
> +		return TRACE_TYPE_UNHANDLED;
>  	}
>  }

That did the trick.

Thanks a lot,

Larry