[Xenomai] Problem with ipipe-core-3.2.21-x86-1.patch?

[Xenomai] Problem with ipipe-core-3.2.21-x86-1.patch?

[Jeff Webb]

I am trying to build xenomai from the xenomai-2.6.git repo under ubuntu 12.04 on a x86_64 desktop machine.  When I try to build a xenomai-enabled kernel from the 3.2.21 sources, the build fails with a compile error.  (Sorry, I didn't save the error message, but I can repeat the process if necessary.)  In order to get things to compile, I made the following change to ipipe-core-3.2.21-x86-1.patch.  Is this a bug?  If so, I'm wondering why no one else has run into this one yet.  I think I built a kernel without any compiler errors under ubuntu 10.04 using this patch, but I'm not 100% certain about that.

-Jeff

diff --git a/ksrc/arch/x86/patches/ipipe-core-3.2.21-x86-1.patch b/ksrc/arch/x86/patches/ipipe-core-3.2.21-x86-1.patch
index 18b23be..58bdbf4 100644
--- a/ksrc/arch/x86/patches/ipipe-core-3.2.21-x86-1.patch
+++ b/ksrc/arch/x86/patches/ipipe-core-3.2.21-x86-1.patch
@@ -1751,12 +1751,12 @@ index 6d939d7..9bbe781 100644
   #ifdef CONFIG_SMP
   	chip->irq_set_affinity = ir_ioapic_set_affinity;
  +#ifdef CONFIG_IPIPE
-+	chip->irq_move = move_xxapic_irq,
++	chip->irq_move = move_xxapic_irq;
  +#endif
  +#endif
  +#ifdef CONFIG_IPIPE
-+	chip->irq_hold = hold_ioapic_irq,
-+	chip->irq_release = release_ioapic_irq,
++	chip->irq_hold = hold_ioapic_irq;
++	chip->irq_release = release_ioapic_irq;
   #endif
   }
   #endif /* CONFIG_IRQ_REMAP */

Re: [Xenomai] Problem with ipipe-core-3.2.21-x86-1.patch?

[Gilles Chanteperdrix]

On 10/16/2012 12:08 AM, Jeff Webb wrote:

> I am trying to build xenomai from the xenomai-2.6.git repo under
> ubuntu 12.04 on a x86_64 desktop machine.  When I try to build a
> xenomai-enabled kernel from the 3.2.21 sources, the build fails with
> a compile error.  (Sorry, I didn't save the error message, but I can
> repeat the process if necessary.)  In order to get things to compile,
> I made the following change to ipipe-core-3.2.21-x86-1.patch.  Is
> this a bug?  If so, I'm wondering why no one else has run into this
> one yet.  I think I built a kernel without any compiler errors under
> ubuntu 10.04 using this patch, but I'm not 100% certain about that.


Yep, it has been reported already, and is fixed in the 3.4 branch, not
backported in 3.2 yet.

-- 
                                                                Gilles.

Re: [Xenomai] Problem with ipipe-core-3.2.21-x86-1.patch?

[Jeff Webb]

On 10/15/2012 05:10 PM, Gilles Chanteperdrix wrote:
> On 10/16/2012 12:08 AM, Jeff Webb wrote:
>
>> I am trying to build xenomai from the xenomai-2.6.git repo under
>> ubuntu 12.04 on a x86_64 desktop machine.  When I try to build a
>> xenomai-enabled kernel from the 3.2.21 sources, the build fails with
>> a compile error.  (Sorry, I didn't save the error message, but I can
>> repeat the process if necessary.)  In order to get things to compile,
>> I made the following change to ipipe-core-3.2.21-x86-1.patch.  Is
>> this a bug?  If so, I'm wondering why no one else has run into this
>> one yet.  I think I built a kernel without any compiler errors under
>> ubuntu 10.04 using this patch, but I'm not 100% certain about that.
>
>
> Yep, it has been reported already, and is fixed in the 3.4 branch, not
> backported in 3.2 yet.
>

Thanks, Gilles.  Sorry, I missed that report.

Re: [Xenomai] Problem with ipipe-core-3.2.21-x86-1.patch?

[Leopold Palomo-Avellaneda]

A Dimarts 16 Octubre 2012, Gilles Chanteperdrix va escriure:
> On 10/16/2012 12:08 AM, Jeff Webb wrote:
> 
> > I am trying to build xenomai from the xenomai-2.6.git repo under
> > ubuntu 12.04 on a x86_64 desktop machine.  When I try to build a
> > xenomai-enabled kernel from the 3.2.21 sources, the build fails with
> > a compile error.  (Sorry, I didn't save the error message, but I can
> > repeat the process if necessary.)  In order to get things to compile,
> > I made the following change to ipipe-core-3.2.21-x86-1.patch.  Is
> > this a bug?  If so, I'm wondering why no one else has run into this
> > one yet.  I think I built a kernel without any compiler errors under
> > ubuntu 10.04 using this patch, but I'm not 100% certain about that.
> 
> 
> Yep, it has been reported already, and is fixed in the 3.4 branch, not
> backported in 3.2 yet.

hi,

1) does it means that this patch could be considered a backport to 3.2?

2) we use orocos and their xenomai part. We have found that when we run an 
application in xenomai, although the framework could do it, there's no 
migration between cpus, and all is executed in cpu0. The main developer have 
commented that a some test isolating xenomai should be done. 

I have seen that the patch touch something of affinity. Some one you have 
experienced some trouble when you run a multithread application and all the 
threads are executed in cpu0 and no migration is done?

Regards,

Leo



-- 
--
Linux User 152692
Catalonia

Re: [Xenomai] Problem with ipipe-core-3.2.21-x86-1.patch?

[Gilles Chanteperdrix]

On 10/16/2012 01:15 PM, Leopold Palomo-Avellaneda wrote:
> A Dimarts 16 Octubre 2012, Gilles Chanteperdrix va escriure:
>> On 10/16/2012 12:08 AM, Jeff Webb wrote:
>>
>>> I am trying to build xenomai from the xenomai-2.6.git repo under
>>> ubuntu 12.04 on a x86_64 desktop machine.  When I try to build a
>>> xenomai-enabled kernel from the 3.2.21 sources, the build fails with
>>> a compile error.  (Sorry, I didn't save the error message, but I can
>>> repeat the process if necessary.)  In order to get things to compile,
>>> I made the following change to ipipe-core-3.2.21-x86-1.patch.  Is
>>> this a bug?  If so, I'm wondering why no one else has run into this
>>> one yet.  I think I built a kernel without any compiler errors under
>>> ubuntu 10.04 using this patch, but I'm not 100% certain about that.
>>
>>
>> Yep, it has been reported already, and is fixed in the 3.4 branch, not
>> backported in 3.2 yet.
> 
> hi,
> 
> 1) does it means that this patch could be considered a backport to 3.2?
> 
> 2) we use orocos and their xenomai part. We have found that when we run an 
> application in xenomai, although the framework could do it, there's no 
> migration between cpus, and all is executed in cpu0. The main developer have 
> commented that a some test isolating xenomai should be done. 
> 
> I have seen that the patch touch something of affinity. Some one you have 
> experienced some trouble when you run a multithread application and all the 
> threads are executed in cpu0 and no migration is done?

A migration is detrimental to determinism, so Xenomai takes the easy way
out of this dilemna: absolutely no migration is done automatically. So,
if you want to run a thread on a different cpu, you have to set the
affinity yourself. It has always been that way.

What this patch touches is automatic migration of IRQs, which is
necessary, when as far as I understood, the IOAPIC does not do load
balancing correctly.

-- 
					    Gilles.

Re: [Xenomai] Problem with ipipe-core-3.2.21-x86-1.patch?

[Leopold Palomo-Avellaneda]

A Dimarts 16 Octubre 2012, Gilles Chanteperdrix va escriure:
> On 10/16/2012 01:15 PM, Leopold Palomo-Avellaneda wrote:
> > A Dimarts 16 Octubre 2012, Gilles Chanteperdrix va escriure:
> >> On 10/16/2012 12:08 AM, Jeff Webb wrote:
> >>
> >>> I am trying to build xenomai from the xenomai-2.6.git repo under
> >>> ubuntu 12.04 on a x86_64 desktop machine.  When I try to build a
> >>> xenomai-enabled kernel from the 3.2.21 sources, the build fails with
> >>> a compile error.  (Sorry, I didn't save the error message, but I can
> >>> repeat the process if necessary.)  In order to get things to compile,
> >>> I made the following change to ipipe-core-3.2.21-x86-1.patch.  Is
> >>> this a bug?  If so, I'm wondering why no one else has run into this
> >>> one yet.  I think I built a kernel without any compiler errors under
> >>> ubuntu 10.04 using this patch, but I'm not 100% certain about that.
> >>
> >>
> >> Yep, it has been reported already, and is fixed in the 3.4 branch, not
> >> backported in 3.2 yet.
> > 
> > hi,
> > 
> > 1) does it means that this patch could be considered a backport to 3.2?
> > 
> > 2) we use orocos and their xenomai part. We have found that when we run an 
> > application in xenomai, although the framework could do it, there's no 
> > migration between cpus, and all is executed in cpu0. The main developer 
have 
> > commented that a some test isolating xenomai should be done. 
> > 
> > I have seen that the patch touch something of affinity. Some one you have 
> > experienced some trouble when you run a multithread application and all 
the 
> > threads are executed in cpu0 and no migration is done?
> 
> A migration is detrimental to determinism, so Xenomai takes the easy way
> out of this dilemna: absolutely no migration is done automatically. So,
> if you want to run a thread on a different cpu, you have to set the
> affinity yourself. It has always been that way.

We do it using the Orocos api, but it doesn't work, so we don't know if it's a 
Orocos problem or Xenomai problem.

> 
> What this patch touches is automatic migration of IRQs, which is
> necessary, when as far as I understood, the IOAPIC does not do load
> balancing correctly.

I will test it.

Regards,

Leo



-- 
--
Linux User 152692
Catalonia

Re: [Xenomai] Problem with ipipe-core-3.2.21-x86-1.patch?

[Gilles Chanteperdrix]

On 10/16/2012 03:16 PM, Leopold Palomo-Avellaneda wrote:
> A Dimarts 16 Octubre 2012, Gilles Chanteperdrix va escriure:
>> On 10/16/2012 01:15 PM, Leopold Palomo-Avellaneda wrote:
>>> A Dimarts 16 Octubre 2012, Gilles Chanteperdrix va escriure:
>>>> On 10/16/2012 12:08 AM, Jeff Webb wrote:
>>>>
>>>>> I am trying to build xenomai from the xenomai-2.6.git repo under
>>>>> ubuntu 12.04 on a x86_64 desktop machine.  When I try to build a
>>>>> xenomai-enabled kernel from the 3.2.21 sources, the build fails with
>>>>> a compile error.  (Sorry, I didn't save the error message, but I can
>>>>> repeat the process if necessary.)  In order to get things to compile,
>>>>> I made the following change to ipipe-core-3.2.21-x86-1.patch.  Is
>>>>> this a bug?  If so, I'm wondering why no one else has run into this
>>>>> one yet.  I think I built a kernel without any compiler errors under
>>>>> ubuntu 10.04 using this patch, but I'm not 100% certain about that.
>>>>
>>>>
>>>> Yep, it has been reported already, and is fixed in the 3.4 branch, not
>>>> backported in 3.2 yet.
>>>
>>> hi,
>>>
>>> 1) does it means that this patch could be considered a backport to 3.2?
>>>
>>> 2) we use orocos and their xenomai part. We have found that when we run an 
>>> application in xenomai, although the framework could do it, there's no 
>>> migration between cpus, and all is executed in cpu0. The main developer 
> have 
>>> commented that a some test isolating xenomai should be done. 
>>>
>>> I have seen that the patch touch something of affinity. Some one you have 
>>> experienced some trouble when you run a multithread application and all 
> the 
>>> threads are executed in cpu0 and no migration is done?
>>
>> A migration is detrimental to determinism, so Xenomai takes the easy way
>> out of this dilemna: absolutely no migration is done automatically. So,
>> if you want to run a thread on a different cpu, you have to set the
>> affinity yourself. It has always been that way.
> 
> We do it using the Orocos api, but it doesn't work, so we don't know if it's a 
> Orocos problem or Xenomai problem.

What is the return value of sched_setaffinity?

-- 
					    Gilles.

Re: [Xenomai] Problem with ipipe-core-3.2.21-x86-1.patch?

[Leopold Palomo-Avellaneda]

Tomorrow I will try to test and know that data.

Regards, 


Leo

Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org> ha escrit:

>On 10/16/2012 03:16 PM, Leopold Palomo-Avellaneda wrote:
>> A Dimarts 16 Octubre 2012, Gilles Chanteperdrix va escriure:
>>> On 10/16/2012 01:15 PM, Leopold Palomo-Avellaneda wrote:
>>>> A Dimarts 16 Octubre 2012, Gilles Chanteperdrix va escriure:
>>>>> On 10/16/2012 12:08 AM, Jeff Webb wrote:
>>>>>
>>>>>> I am trying to build xenomai from the xenomai-2.6.git repo under
>>>>>> ubuntu 12.04 on a x86_64 desktop machine.  When I try to build a
>>>>>> xenomai-enabled kernel from the 3.2.21 sources, the build fails with
>>>>>> a compile error.  (Sorry, I didn't save the error message, but I can
>>>>>> repeat the process if necessary.)  In order to get things to compile,
>>>>>> I made the following change to ipipe-core-3.2.21-x86-1.patch.  Is
>>>>>> this a bug?  If so, I'm wondering why no one else has run into this
>>>>>> one yet.  I think I built a kernel without any compiler errors under
>>>>>> ubuntu 10.04 using this patch, but I'm not 100% certain about that.
>>>>>
>>>>>
>>>>> Yep, it has been reported already, and is fixed in the 3.4 branch, not
>>>>> backported in 3.2 yet.
>>>>
>>>> hi,
>>>>
>>>> 1) does it means that this patch could be considered a backport to 3.2?
>>>>
>>>> 2) we use orocos and their xenomai part. We have found that when we run an 
>>>> application in xenomai, although the framework could do it, there's no 
>>>> migration between cpus, and all is executed in cpu0. The main developer 
>> have 
>>>> commented that a some test isolating xenomai should be done. 
>>>>
>>>> I have seen that the patch touch something of affinity. Some one you have 
>>>> experienced some trouble when you run a multithread application and all 
>> the 
>>>> threads are executed in cpu0 and no migration is done?
>>>
>>> A migration is detrimental to determinism, so Xenomai takes the easy way
>>> out of this dilemna: absolutely no migration is done automatically. So,
>>> if you want to run a thread on a different cpu, you have to set the
>>> affinity yourself. It has always been that way.
>> 
>> We do it using the Orocos api, but it doesn't work, so we don't know if it's a 
>> Orocos problem or Xenomai problem.
>
>What is the return value of sched_setaffinity?
>
>-- 
>					    Gilles.