Hi Marc,

On 08/04/2011 11:47 AM, Marc Kleine-Budde wrote:
> On 08/04/2011 10:25 AM, Wolfgang Grandegger wrote:
>> On 08/04/2011 10:09 AM, Wolfgang Grandegger wrote:
>>> Hi Robin,
>>>
>>> On 08/04/2011 05:23 AM, Robin Holt wrote:
>>>> I have spent some more time working on this.  I took their code from the
>>>> April BSP and am reworking it to be a much cleaner (IMHO) implementation.
>>>> Hopefully I will have a set of patches for you tomorrow.
>>>>
>>>> Right now, I am down to the point of getting the _probe function
>>>> abstracted off.  I think I will get it fairly clean when I am completed.
>>>
>>> Not sure what are you trying to do? What is missing in the mainline
>>> Flexcan driver are the device tree bindings. Unfortunately, Freescale's
>>> driver is not a good approach. I think it can be done much simpler,
>>> especially with the new device-tree integration available in v3.0.
>>> Robin, you have the mainline kernel running for a P1010 board, right?
>>
>> You patches are a good starting point for discussion...
> 
> Have you already posted them?
> 
> From my point of view there are two problems:
> - OF tree integration (iomem, irq, clock)
> - endianess
> 
> The fsl driver assumes that OF tree means a big endian hardware, this is
> not true for arm, where we have OF tree now.

Yes. But I was also thinking about a smart integration of platform and
of platform bindings as we need this sooner than later for other drivers
as well. As you have already realized, already since 2.6.35 there is a
common platform_driver_register() and "of_node" is now part of "struct
device" allowing a much nice implementation. I will look for a good
example in the kernel...

Wolfgang.
_______________________________________________
Socketcan-users mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/socketcan-users

Reply via email to