Re: [Xenomai-help] IRQ conflict issues with and ESD PMC card.

[Jan Kiszka <jan.kiszka@domain.hid>]

Stefan Kisdaroczi wrote:
> Salut Gilles,
> 
> i need to jump in.
> 
> Am 28.06.2010 14:18, schrieb Gilles Chanteperdrix:
>> Nape Ibrahim Lentsoane wrote:
>>> Hello,
>>>
>>> I've been having this problem for some time now and I hope someone from
>>> the Xenomai community can help.
>>> Basically, my kernel does not want to boot with PMC card. I have managed
>>> to use console redirection of messages
>>> to the serial port and I can see the error messages. From what I can
>>> see, there is a problem with '/xnintr_irq_handler/' due
>>> to this message at boot time: /Xenomai: xnintr_irq_handler: IRQ10 not
>>> handled. Disabling IRQ line./
>> As repeated many times, disabling ACPI has been known to cause such
>> issues. So, you should keep ACPI enabled, only ACPI_PROCESSOR needs to
>> be disabled.
>>
>> For IRQ conflicts, see:
>> http://www.xenomai.org/index.php/FAQs#What_can_I_do_if_Xenomai_and_Linux_devices_share_the_same_IRQ.3F
> 
> I had to consult this wiki already multiple times (every time I get a new box...)
> It's ok, but there is not really a solution. Currently I have one pci-card
> and a rtdm driver. I got my unshared realtime IRQ after disabling the soundcard.
> 
> At the end of the year I need to be able to use three of these cards and
> five realtime serial connections in one machine. I already have sleepless
> nights ;-) and hope I understood something wrong...
> 
> AFAIK with 2.6.31 "threaded IRQ's" were merged to mainline. Would it be possible
> now for IPIPE to pass down the IRQ's to linux so that the IRQ is handled, but
> the IRQ-Thread is not started ? Or something like ?
> 

It's still not that simple. Threaded IRQs per se do not solve the issue,
neither in upstream nor for Xenomai.

What they do, though, is to lay the ground for slowing enabling RT-safe
IRQ-sharing in more and more Linux drivers: Given a proper split-up
between top-half and threaded handler, you now have the means (with
upstream) or at least the blueprint (for Xenomai) how to quickly silence
an IRQ line at device level. For Xenomai, these handler would have to be
made co-scheduling-safe and could then be registered against RTDM as
shared RT-IRQs. Still, this issue cannot be addressed automatically, it
has to be developed driver by driver.

The better news is that modern hardware now more and more provides MSI,
and that's an inherently unshared IRQ source.

Jan

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