Re: dynamic ftrace - graph

[Michal Simek <monstr@monstr.eu>]

Steven Rostedt wrote:
> On Fri, 2009-12-11 at 16:28 +0100, Michal Simek wrote:
>> Steven Rostedt wrote:
>>> On Fri, 2009-12-11 at 15:40 +0100, Michal Simek wrote:
>>>> Steven Rostedt wrote:
>>>>> Hey, if they work and they are only in your arch, then I'm fine with
>>>>> them ;-)
>>>> yeah. There is one just small hook to recordmcount.pl. :-)
>>> I may need to give an acked-by. If I did not yet, could you send me that
>>> specific patch.
>> I sent that patches to LKML last month - you were cc'd.
>> http://lkml.org/lkml/2009/11/23/48
> 
> They are probably still in the "todo" pile ;-)

:-)

> 
>> You should get it in your mailbox
>>
>> That part is pretty easy. :-)
>>
>> diff --git a/scripts/recordmcount.pl b/scripts/recordmcount.pl
>> index 090d300..b30dabd 100755
>> --- a/scripts/recordmcount.pl
>> +++ b/scripts/recordmcount.pl
>> @@ -245,6 +245,9 @@ if ($arch eq "x86_64") {
>>       $ld .= " -m elf64_sparc";
>>       $cc .= " -m64";
>>       $objcopy .= " -O elf64-sparc";
>> +} elsif ($arch eq "microblaze") {
>> +    # Microblaze calls '_mcount' instead of plain 'mcount'.
>> +    $mcount_regex = "^\\s*([0-9a-fA-F]+):.*\\s_mcount\$";
>>   } else {
>>       die "Arch $arch is not supported with CONFIG_FTRACE_MCOUNT_RECORD";
>>   }
> 
> If that's all it is, then here:

yes.
> 
> Acked-by: Steven Rostedt <rostedt@goodmis.org>

I don't expect it.:-)

> 
>>
>>>
>>>> we will see. Currently I have small problem only with irqsoff but I 
>>>> think this don't have connection with ftrace - only with lockdep I 
>>>> think. :-)
>>> You have lockdep working?  If you get lockdep working you should get the
>>> irqsoff tracer for free.
>> I have some problems when I turn on lockdep debug (CONFIG_DEBUG_LOCKDEP) 
>> - kernel freeze for currently unknown reason.
>> I saw that some arch has some part of code in entry.S but I haven't had 
>> a time to look at them - maybe needs some changes there too.
> 
> Yeah, anytime you disable interrupts or enable them (including when the
> hardware does it from entering an exception), you need to call into the
> lockdep infrastructure (trace_hardiqs_on/off){_caller}.

yes that's missing.

> 
> 
>> As you can see below locking testsuite output looks fine but irqsoff 
>> tracer freeze the kernel. Function tracer is ok.
>>
>> There could be a problem with some arch specific functions which 
>> shouldn't be compiled with -pg.
> 
> Does it still freeze if you don't enable the function tracer?

look below.

> 
>> Any suggestions? :-)
>>
> 
> 
>> # cd /
>> # mkdir debug
>> # mount -t debugfs none /debug
>> # cat /debug/tracing/available_tracers
>> irqsoff function sched_switch nop
>> # echo irqsoff > /debug/tracing/current_tracer
>>
>> and kernel freeze. :-(
> 
> Try it without configuring function tracer, or you could also do:

uclinux login: root
Password:
Login incorrect
uclinux login: root
Password:
# cd /
# mkdir debug
# mount -t debugfs none /debug/
# cat /debug/tracing/available_tracers
irqsoff sched_switch nop
# echo irqsoff > /debug/tracing/current_tracer


Also freeze - the problem is that low level code. I will look at it next 
week.

Thanks,
Michal

> 
> echo 0 > /proc/sys/ftrace_enabled
> 
> That will also disable the function tracer.
> 
> -- Steve
> 
> 


-- 
Michal Simek, Ing. (M.Eng)
w: www.monstr.eu p: +42-0-721842854
Maintainer of Linux kernel 2.6 Microblaze Linux - http://www.monstr.eu/fdt/
Microblaze U-BOOT custodian