Tracepoints and Timestamps

Tracepoints and Timestamps

[Abhishek Karoliya]

Hi,

  I am remote debuging a multi threaded app on a powerpc e500v2 dual
processor running linux 2.6.32 while the host gdb is x86_64 linux
machine. The version of gdb is 7.3.1. On hitting a tracepoint, I want
to collect the timestamp. I am wondering is there is any way of doing
it currently on this gdb. If not, I was wondering if I can collect the
timebase registers (tbl and tbu). The current register dump does not
show the contents of these register on e500v2 although I have seen an
entry for other powerpc processor.

Is there a way I can quickly add support for these registers. If
someone can point me to file/doc/previous patches to do this I will be
grateful.

 Cheers!
Abhi

Re: Tracepoints and Timestamps

[Hui Zhu]

If you use the gdbserver that include in GDB source, you can access
special tvariable $trace_timestamp.

If not, I think maybe you can add a special tvariable that access to
timebase registers (tbl and tbu)  to gdbserver.  Then you can collect
it when you want it.

Thanks,
Hui


On Sat, Nov 5, 2011 at 03:05, Abhishek Karoliya <abhikaro@gmail.com> wrote:
> Hi,
>
>   I am remote debuging a multi threaded app on a powerpc e500v2 dual
> processor running linux 2.6.32 while the host gdb is x86_64 linux
> machine. The version of gdb is 7.3.1. On hitting a tracepoint, I want
> to collect the timestamp. I am wondering is there is any way of doing
> it currently on this gdb. If not, I was wondering if I can collect the
> timebase registers (tbl and tbu). The current register dump does not
> show the contents of these register on e500v2 although I have seen an
> entry for other powerpc processor.
>
> Is there a way I can quickly add support for these registers. If
> someone can point me to file/doc/previous patches to do this I will be
> grateful.
>
>  Cheers!
> Abhi
>

Re: Tracepoints and Timestamps

[Abhishek Karoliya]

Hi Hui,

   Thanks for the pointer ... However after going through the
gdbserver source code, I realise that there is no tracepoint support
for linux-ppc. So now I have a even bigger problem on my hand.

Does anyone know if I can ppc tracepoint support being actively
developed and if I can lay my hands on this experimental code.

Cheers!
Abhi

On Mon, Nov 7, 2011 at 6:40 AM, Hui Zhu <teawater@gmail.com> wrote:
> If you use the gdbserver that include in GDB source, you can access
> special tvariable $trace_timestamp.
>
> If not, I think maybe you can add a special tvariable that access to
> timebase registers (tbl and tbu)  to gdbserver.  Then you can collect
> it when you want it.
>
> Thanks,
> Hui
>
>
> On Sat, Nov 5, 2011 at 03:05, Abhishek Karoliya <abhikaro@gmail.com> wrote:
>> Hi,
>>
>>   I am remote debuging a multi threaded app on a powerpc e500v2 dual
>> processor running linux 2.6.32 while the host gdb is x86_64 linux
>> machine. The version of gdb is 7.3.1. On hitting a tracepoint, I want
>> to collect the timestamp. I am wondering is there is any way of doing
>> it currently on this gdb. If not, I was wondering if I can collect the
>> timebase registers (tbl and tbu). The current register dump does not
>> show the contents of these register on e500v2 although I have seen an
>> entry for other powerpc processor.
>>
>> Is there a way I can quickly add support for these registers. If
>> someone can point me to file/doc/previous patches to do this I will be
>> grateful.
>>
>>  Cheers!
>> Abhi
>>
>

Re: Tracepoints and Timestamps

[Stan Shebs]

On 11/7/11 6:53 AM, Abhishek Karoliya wrote:
> Hi Hui,
>
>     Thanks for the pointer ... However after going through the
> gdbserver source code, I realise that there is no tracepoint support
> for linux-ppc. So now I have a even bigger problem on my hand.
>
> Does anyone know if I can ppc tracepoint support being actively
> developed and if I can lay my hands on this experimental code.
>

The tracepoint support in GDBserver is pretty generic, and if you don't 
need fast tracepoints, there's not really anything that is arch-specific.

I would suggest enabling target_supports_tracepoints for ppc and letting 
us know how it goes.  If it "just works", then we could change the 
target_supports_tracepoints test to make it unconditional - fast 
tracepoint support has its own availability check at runtime, it could 
subsume the architecture check.

Stan
stan@codesourcery.com

Re: Tracepoints and Timestamps

[Pedro Alves]

On Monday 07 November 2011 16:15:17, Stan Shebs wrote:
> On 11/7/11 6:53 AM, Abhishek Karoliya wrote:
> > Hi Hui,
> >
> >     Thanks for the pointer ... However after going through the
> > gdbserver source code, I realise that there is no tracepoint support
> > for linux-ppc. So now I have a even bigger problem on my hand.
> >
> > Does anyone know if I can ppc tracepoint support being actively
> > developed and if I can lay my hands on this experimental code.
> >
> 
> The tracepoint support in GDBserver is pretty generic, and if you don't 
> need fast tracepoints, there's not really anything that is arch-specific.
> 
> I would suggest enabling target_supports_tracepoints for ppc and letting 
> us know how it goes.  If it "just works", then we could change the 
> target_supports_tracepoints test to make it unconditional.

We can't.  The backend needs to be able to step over breakpoints
so it can step over tracepoints without gdb involvement.  Which means, it
either needs to support hardware single-step (ppc does), or, it'll need to
know how to software single-step.  No gdbserver port can do the latter.
And it should ideally switch to using gdbserver side breakpoints (z0/Z0
packets, the insert_point/remove_point hooks), so breakpoints and
tracepoints can coexist.

linux-ppc-low.c should install the supports_tracepoints hook like
linux-x86-low.c does:

static int
x86_supports_tracepoints (void)
{
  return 1;
}

and insert_point/remote_point hooks, like x86_insert_point/x86_remove_point.

The Z0 insert/remove code support would ideally be moved to the gdbserver
core side, controlled by a new "use software breakpoints" hook added to
gdbserver's target vector, so that bit of code could be reused.

-- 
Pedro Alves