[Xenomai-help] IRQ handler argument corrupt when shared interrupts are enabled (Xenomai 2.4.10)

[Xenomai-help] IRQ handler argument corrupt when shared interrupts are enabled (Xenomai 2.4.10)

[Henri Roosen]

Xenomai 2.4.10:

IRQ handler argument is not the same as the argument I provided when
registering the IRQ handler. Problem only occurs when shared interrupt
support is enabled in the kernel config. NOTE: even though the
registered IRQ handler not a shared interrupt handler (flags = 0).

The problem is not reproducable on Xenomai 2.5.0.

So I was looking for the fix code in the 2.5.0 but couldn't find it.
Is this a know problem that has only been fixed on 2.5.0, or does the
bug just not show in 2.5.0? Could someone point out which commit on
2.5.0 fixes the problem?

Thanks,
Henri.

Re: [Xenomai-help] IRQ handler argument corrupt when shared interrupts are enabled (Xenomai 2.4.10)

[Jan Kiszka]

Henri Roosen wrote:
> Xenomai 2.4.10:
> 
> IRQ handler argument is not the same as the argument I provided when
> registering the IRQ handler. Problem only occurs when shared interrupt
> support is enabled in the kernel config. NOTE: even though the
> registered IRQ handler not a shared interrupt handler (flags = 0).

What API are you referring to, RTDM, native, or nucleus? What's your arch?

> 
> The problem is not reproducable on Xenomai 2.5.0.
> 
> So I was looking for the fix code in the 2.5.0 but couldn't find it.
> Is this a know problem that has only been fixed on 2.5.0, or does the
> bug just not show in 2.5.0? Could someone point out which commit on
> 2.5.0 fixes the problem?

I don't recall such a problem or any potential fix ATM.

Jan

-- 
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux

Re: [Xenomai-help] IRQ handler argument corrupt when shared interrupts are enabled (Xenomai 2.4.10)

[Henri Roosen]

On Fri, Jan 15, 2010 at 10:43 AM, Jan Kiszka <jan.kiszka@domain.hid> wrote:
> Henri Roosen wrote:
>> Xenomai 2.4.10:
>>
>> IRQ handler argument is not the same as the argument I provided when
>> registering the IRQ handler. Problem only occurs when shared interrupt
>> support is enabled in the kernel config. NOTE: even though the
>> registered IRQ handler not a shared interrupt handler (flags = 0).
>
> What API are you referring to, RTDM, native, or nucleus? What's your arch?

I am using the RTDM API on x86, kernel 2.6.30.10.

>
>>
>> The problem is not reproducable on Xenomai 2.5.0.
>>
>> So I was looking for the fix code in the 2.5.0 but couldn't find it.
>> Is this a know problem that has only been fixed on 2.5.0, or does the
>> bug just not show in 2.5.0? Could someone point out which commit on
>> 2.5.0 fixes the problem?
>
> I don't recall such a problem or any potential fix ATM.

Thanks Jan, that is good to know. I'll dig into it a little further...
Glad it is very reproducable over here ;-)

>
> Jan
>
> --
> Siemens AG, Corporate Technology, CT T DE IT 1
> Corporate Competence Center Embedded Linux
>

Re: [Xenomai-help] IRQ handler argument corrupt when shared interrupts are enabled (Xenomai 2.4.10)

[Jan Kiszka]

Henri Roosen wrote:
> On Fri, Jan 15, 2010 at 10:43 AM, Jan Kiszka <jan.kiszka@domain.hid> wrote:
>> Henri Roosen wrote:
>>> Xenomai 2.4.10:
>>>
>>> IRQ handler argument is not the same as the argument I provided when
>>> registering the IRQ handler. Problem only occurs when shared interrupt
>>> support is enabled in the kernel config. NOTE: even though the
>>> registered IRQ handler not a shared interrupt handler (flags = 0).
>> What API are you referring to, RTDM, native, or nucleus? What's your arch?
> 
> I am using the RTDM API on x86, kernel 2.6.30.10.
> 
>>> The problem is not reproducable on Xenomai 2.5.0.
>>>
>>> So I was looking for the fix code in the 2.5.0 but couldn't find it.
>>> Is this a know problem that has only been fixed on 2.5.0, or does the
>>> bug just not show in 2.5.0? Could someone point out which commit on
>>> 2.5.0 fixes the problem?
>> I don't recall such a problem or any potential fix ATM.
> 
> Thanks Jan, that is good to know. I'll dig into it a little further...
> Glad it is very reproducable over here ;-)
> 

Given that the size of xnintr_t aka rtdm_irq_t changes under
CONFIG_XENO_OPT_SHIRQ, your driver build could be inconsistent.
Alternatively, you face some memory corruption of the xnintr_t object
after registration, and that happens to hit the cookie on 2.4.10 while
it may wreck something else on 2.5. I would dig in this directions first
as I think the core service itself is fairly stable.

Jan

-- 
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux

Re: [Xenomai-help] IRQ handler argument corrupt when shared interrupts are enabled (Xenomai 2.4.10)

[Henri Roosen]

Ok, found out it was an issue in my build environment: The modules
were build against the wrong kernel tree. This explains the corruption
of the xnintr_t object.

Sorry for the inconvenience and thanks for the replies.

Thanks,
Henri.

On Fri, Jan 15, 2010 at 11:06 AM, Jan Kiszka <jan.kiszka@domain.hid> wrote:
> Henri Roosen wrote:
>> On Fri, Jan 15, 2010 at 10:43 AM, Jan Kiszka <jan.kiszka@domain.hid> wrote:
>>> Henri Roosen wrote:
>>>> Xenomai 2.4.10:
>>>>
>>>> IRQ handler argument is not the same as the argument I provided when
>>>> registering the IRQ handler. Problem only occurs when shared interrupt
>>>> support is enabled in the kernel config. NOTE: even though the
>>>> registered IRQ handler not a shared interrupt handler (flags = 0).
>>> What API are you referring to, RTDM, native, or nucleus? What's your arch?
>>
>> I am using the RTDM API on x86, kernel 2.6.30.10.
>>
>>>> The problem is not reproducable on Xenomai 2.5.0.
>>>>
>>>> So I was looking for the fix code in the 2.5.0 but couldn't find it.
>>>> Is this a know problem that has only been fixed on 2.5.0, or does the
>>>> bug just not show in 2.5.0? Could someone point out which commit on
>>>> 2.5.0 fixes the problem?
>>> I don't recall such a problem or any potential fix ATM.
>>
>> Thanks Jan, that is good to know. I'll dig into it a little further...
>> Glad it is very reproducable over here ;-)
>>
>
> Given that the size of xnintr_t aka rtdm_irq_t changes under
> CONFIG_XENO_OPT_SHIRQ, your driver build could be inconsistent.
> Alternatively, you face some memory corruption of the xnintr_t object
> after registration, and that happens to hit the cookie on 2.4.10 while
> it may wreck something else on 2.5. I would dig in this directions first
> as I think the core service itself is fairly stable.
>
> Jan
>
> --
> Siemens AG, Corporate Technology, CT T DE IT 1
> Corporate Competence Center Embedded Linux
>