[Xenomai] Gdb freezes when stepping through rt_task_spawn.

[Xenomai] Gdb freezes when stepping through rt_task_spawn.

[Marco Tessore]

Hello,
I have a problem, you've probably already dealt with in the past:
when I debug with gdb / gdbserver, if I execute step by step while I 
invoke the call rt_task_spawn, gdb freezes with no chance of recovery.

Since I have seen that you have already faced the problem a few years 
ago, I was wondering if you had found a solution.

kind regards
Marco Tessore

Re: [Xenomai] Gdb freezes when stepping through rt_task_spawn.

[Gilles Chanteperdrix]

On 06/10/2014 11:18 AM, Marco Tessore wrote:
> Hello,
> I have a problem, you've probably already dealt with in the past:
> when I debug with gdb / gdbserver, if I execute step by step while I 
> invoke the call rt_task_spawn, gdb freezes with no chance of recovery.
> 
> Since I have seen that you have already faced the problem a few years 
> ago, I was wondering if you had found a solution.

Did you try xenomai 2.6 git? An issue was fixed very recently with
regard to gdb (it is actually the latest commit on this branch).


-- 
                                                                Gilles.

Re: [Xenomai] Gdb freezes when stepping through rt_task_spawn.

[Marco Tessore]

Thank you, I will try that commit.
But I have another question about it:
does the commit involve the patch to the kernel?
Since my kernel was patched using i-pipe git for the vanilla 3.10, but 
applied to a customized version of 3.11,
if the commit involves the kernel patch, it can be difficult for me to 
upgrade.

I have found another problem with gdb:
when I have multiple tasks pending on rt_sem_p,
if I change the state of breakpoints, the function can return error code 
-38.
May be that the same commit can resolve both problems.

Thank you very much
Marco Tessore

Il 10/06/2014 22:02, Gilles Chanteperdrix ha scritto:
> On 06/10/2014 11:18 AM, Marco Tessore wrote:
>> Hello,
>> I have a problem, you've probably already dealt with in the past:
>> when I debug with gdb / gdbserver, if I execute step by step while I
>> invoke the call rt_task_spawn, gdb freezes with no chance of recovery.
>>
>> Since I have seen that you have already faced the problem a few years
>> ago, I was wondering if you had found a solution.
> Did you try xenomai 2.6 git? An issue was fixed very recently with
> regard to gdb (it is actually the latest commit on this branch).
>
>

Re: [Xenomai] Gdb freezes when stepping through rt_task_spawn.

[Philippe Gerum]

On 06/12/2014 03:38 PM, Marco Tessore wrote:
> Thank you, I will try that commit.
> But I have another question about it:
> does the commit involve the patch to the kernel?

Yes.

> Since my kernel was patched using i-pipe git for the vanilla 3.10, but
> applied to a customized version of 3.11,
> if the commit involves the kernel patch, it can be difficult for me to
> upgrade.
>

There is no other option.

> I have found another problem with gdb:
> when I have multiple tasks pending on rt_sem_p,
> if I change the state of breakpoints, the function can return error code
> -38.
> May be that the same commit can resolve both problems.

It most likely does.

>
> Thank you very much
> Marco Tessore
>
> Il 10/06/2014 22:02, Gilles Chanteperdrix ha scritto:
>> On 06/10/2014 11:18 AM, Marco Tessore wrote:
>>> Hello,
>>> I have a problem, you've probably already dealt with in the past:
>>> when I debug with gdb / gdbserver, if I execute step by step while I
>>> invoke the call rt_task_spawn, gdb freezes with no chance of recovery.
>>>
>>> Since I have seen that you have already faced the problem a few years
>>> ago, I was wondering if you had found a solution.
>> Did you try xenomai 2.6 git? An issue was fixed very recently with
>> regard to gdb (it is actually the latest commit on this branch).
>>
>>
>
>
> _______________________________________________
> Xenomai mailing list
> Xenomai@xenomai.org
> http://www.xenomai.org/mailman/listinfo/xenomai
>


-- 
Philippe.

Re: [Xenomai] Gdb freezes when stepping through rt_task_spawn.

[Gilles Chanteperdrix]

On 06/12/2014 03:38 PM, Marco Tessore wrote:
> Thank you, I will try that commit.
> But I have another question about it:
> does the commit involve the patch to the kernel?
> Since my kernel was patched using i-pipe git for the vanilla 3.10, but 
> applied to a customized version of 3.11,
> if the commit involves the kernel patch, it can be difficult for me to 
> upgrade.

If you kept the ipipe kernel separated from Xenomai kernel sources, the
change is a change in xenomai sources, so the update should be painless.
If you put the xenomai-patched kernel under source control, then it will
be harder.

-- 
                                                                Gilles.

Re: [Xenomai] Gdb freezes when stepping through rt_task_spawn.

[Philippe Gerum]

On 06/12/2014 07:58 PM, Gilles Chanteperdrix wrote:
> On 06/12/2014 03:38 PM, Marco Tessore wrote:
>> Thank you, I will try that commit.
>> But I have another question about it:
>> does the commit involve the patch to the kernel?
>> Since my kernel was patched using i-pipe git for the vanilla 3.10, but
>> applied to a customized version of 3.11,
>> if the commit involves the kernel patch, it can be difficult for me to
>> upgrade.
>
> If you kept the ipipe kernel separated from Xenomai kernel sources, the
> change is a change in xenomai sources, so the update should be painless.
> If you put the xenomai-patched kernel under source control, then it will
> be harder.
>

ENOSYS would rather be symptomatic of this bug which is ARM-specific:
http://git.xenomai.org/ipipe.git/commit/?h=ipipe-3.10&id=e410a5e77323d53277d625e5dafe166cdcd64afb

OTOH, the patch below fixes a spurious EINTR error within the task 
creation syscall, which may well end up in a lockup:
http://git.xenomai.org/xenomai-2.6.git/commit/?id=589882956280d5cb4fdc181a5fcd5ae1188ab6ed

This is one of the situations where knowing the CPU architecture 
involved would help a lot.

-- 
Philippe.

Re: [Xenomai] Gdb freezes when stepping through rt_task_spawn.

[Gilles Chanteperdrix]

On 06/12/2014 10:51 PM, Philippe Gerum wrote:
> On 06/12/2014 07:58 PM, Gilles Chanteperdrix wrote:
>> On 06/12/2014 03:38 PM, Marco Tessore wrote:
>>> Thank you, I will try that commit.
>>> But I have another question about it:
>>> does the commit involve the patch to the kernel?
>>> Since my kernel was patched using i-pipe git for the vanilla 3.10, but
>>> applied to a customized version of 3.11,
>>> if the commit involves the kernel patch, it can be difficult for me to
>>> upgrade.
>>
>> If you kept the ipipe kernel separated from Xenomai kernel sources, the
>> change is a change in xenomai sources, so the update should be painless.
>> If you put the xenomai-patched kernel under source control, then it will
>> be harder.
>>
> 
> ENOSYS would rather be symptomatic of this bug which is ARM-specific:
> http://git.xenomai.org/ipipe.git/commit/?h=ipipe-3.10&id=e410a5e77323d53277d625e5dafe166cdcd64afb

Indeed. This fix for 3.10 does not seem right though. The real fix is 
to move the local_restart jump label, as in:

http://git.xenomai.org/ipipe.git/commit/?h=ipipe-3.14&id=d7a5e4762535f55bfff8eafbf1c1dde214d09bd6

> 
> OTOH, the patch below fixes a spurious EINTR error within the task 
> creation syscall, which may well end up in a lockup:
> http://git.xenomai.org/xenomai-2.6.git/commit/?id=589882956280d5cb4fdc181a5fcd5ae1188ab6ed
> 
> This is one of the situations where knowing the CPU architecture 
> involved would help a lot.

Yes, indeed, should have directed to the "Request for information" page.


-- 
                                                                Gilles.

Re: [Xenomai] Gdb freezes when stepping through rt_task_spawn.

[Philippe Gerum]

On 06/12/2014 11:45 PM, Gilles Chanteperdrix wrote:
> On 06/12/2014 10:51 PM, Philippe Gerum wrote:
>> On 06/12/2014 07:58 PM, Gilles Chanteperdrix wrote:
>>> On 06/12/2014 03:38 PM, Marco Tessore wrote:
>>>> Thank you, I will try that commit.
>>>> But I have another question about it:
>>>> does the commit involve the patch to the kernel?
>>>> Since my kernel was patched using i-pipe git for the vanilla 3.10, but
>>>> applied to a customized version of 3.11,
>>>> if the commit involves the kernel patch, it can be difficult for me to
>>>> upgrade.
>>>
>>> If you kept the ipipe kernel separated from Xenomai kernel sources, the
>>> change is a change in xenomai sources, so the update should be painless.
>>> If you put the xenomai-patched kernel under source control, then it will
>>> be harder.
>>>
>>
>> ENOSYS would rather be symptomatic of this bug which is ARM-specific:
>> http://git.xenomai.org/ipipe.git/commit/?h=ipipe-3.10&id=e410a5e77323d53277d625e5dafe166cdcd64afb
>
> Indeed. This fix for 3.10 does not seem right though. The real fix is
> to move the local_restart jump label, as in:
>
> http://git.xenomai.org/ipipe.git/commit/?h=ipipe-3.14&id=d7a5e4762535f55bfff8eafbf1c1dde214d09bd6

Will you issue a supplemental patch?

-- 
Philippe.

Re: [Xenomai] Gdb freezes when stepping through rt_task_spawn.

[Gilles Chanteperdrix]

On 06/13/2014 07:28 AM, Philippe Gerum wrote:
> On 06/12/2014 11:45 PM, Gilles Chanteperdrix wrote:
>> On 06/12/2014 10:51 PM, Philippe Gerum wrote:
>>> On 06/12/2014 07:58 PM, Gilles Chanteperdrix wrote:
>>>> On 06/12/2014 03:38 PM, Marco Tessore wrote:
>>>>> Thank you, I will try that commit.
>>>>> But I have another question about it:
>>>>> does the commit involve the patch to the kernel?
>>>>> Since my kernel was patched using i-pipe git for the vanilla 3.10, but
>>>>> applied to a customized version of 3.11,
>>>>> if the commit involves the kernel patch, it can be difficult for me to
>>>>> upgrade.
>>>>
>>>> If you kept the ipipe kernel separated from Xenomai kernel sources, the
>>>> change is a change in xenomai sources, so the update should be painless.
>>>> If you put the xenomai-patched kernel under source control, then it will
>>>> be harder.
>>>>
>>>
>>> ENOSYS would rather be symptomatic of this bug which is ARM-specific:
>>> http://git.xenomai.org/ipipe.git/commit/?h=ipipe-3.10&id=e410a5e77323d53277d625e5dafe166cdcd64afb
>>
>> Indeed. This fix for 3.10 does not seem right though. The real fix is
>> to move the local_restart jump label, as in:
>>
>> http://git.xenomai.org/ipipe.git/commit/?h=ipipe-3.14&id=d7a5e4762535f55bfff8eafbf1c1dde214d09bd6
> 
> Will you issue a supplemental patch?
> 
Done, I have pushed my branch which has other changes, merged with the
current state of the ipipe-3.10 branch.

-- 
                                                                Gilles.