[PATCH 2/3] ARM: Orion: Bind the orion bridge interrupt controller through DT

[Sebastian Hesselbarth]

On 01/10/2013 07:07 AM, Andrew Lunn wrote:
> On Wed, Jan 09, 2013 at 08:47:34PM -0500, Jason Cooper wrote:
>> Sebastian,
>>
>> On Tue, Dec 11, 2012 at 11:55:35AM -0700, Jason Gunthorpe wrote:
>>> On Tue, Dec 11, 2012 at 08:13:33AM +0100, Andrew Lunn wrote:
>>>> On Mon, Dec 10, 2012 at 11:58:13PM -0700, Jason Gunthorpe wrote:
>>>>>>> +	bridge_irq = irq_of_parse_and_map(np, 0);
>>>>>>> +	/* FIXME: irq_of_parse_and_map returns 0 on error, but on Dove the
>>>>>>> +	 * bridge IRQ is 0.
>>>>>>> +	if (!bridge_irq)
>>>>>>> +		return -ENODEV;*/
>>>>>>> +
>>>>>>> +	return orion_bridge_irq_init(bridge_irq, -1, base, np);
>>>>>>
>>>>>> So does this mean for Dove it will currently always fail?
>>>>>
>>>>> I guess that is hard to read without syntax hi-lighting, the 'if' is
>>>>> also commented out.
>>>>>
>>>>> I was hopefull it would work as is, but looking deeper at
>>>>> irq_of_parse_and_map and its call tree makes me doubtfull now..
>>>>>
>>>>> Fixing irq_of_parse_and_map is way to big a job for me :(
>>>>
>>>> Hi Jason
>>>>
>>>> https://lwn.net/Articles/470820/
>>>>
>>>> I don't think its fixable.
>>>
>>> Oh right. That.
>>>
>>> Looking at this some more, I think it should work, but not for very
>>> good reasons.. irq_of_parse_and_map will return 0 on dove. It returns
>>> 0 because it is going through its error paths because it saw a 0 deep
>>> inside, but it still returns 0.
>>>
>>> orion_bridge_irq_init will work with 0 as a bridge_irq, we know this
>>> because the timer code was using those code paths for a long time
>>>
>>>> We need to see how other machine remap 0 to some other value and
>>>> implement that for Dove.
>>>
>>> It looks like it is straightforward to adjust Dove's
>>> get_irqnr_and_base to add 1 to all the interrupt numbers, adjust
>>> irqs.h to add 1, and adjust the irq setup. There would be no
>>> performance penalty to doing this, but testing would be need to be
>>> sure 100% of changes were made. If the above works it would be better
>>> to do that sort of change after, or maybe after the non-DT code is
>>> purged...
>>>
>>> Kirkwood is already OK WRT the 0 IRQ, it is handled internally to
>>> get_irqnr_and_base.
>>>
>>> So, I will suggest to please test on dove as-is...
>>
>> Did you have a chance to test this on dove?
>
> Hi Jason
>
> We discussed this a little and I prefer Sebastians implementation.
>
> Sebastian, do you have time to work on your code and get it ready?

Jason, Andrew,

we chose to jump to irqchip instead of fixing old code. I thought
Jason Gunthorpe volunteered to get it ready? Moving to irqchip also
requires clksrc-orion or reimplementing orion into clksrc-mvebu.

I can do it but it depends on when it should be ready?

Sebastian