[RFC PATCH] ARM: OMAP3630: Add generic machine descriptor

[Nishanth Menon]

On 09/20/2013 12:42 PM, Felipe Balbi wrote:
> Hi,
> 
> On Fri, Sep 20, 2013 at 10:16:48AM -0700, Olof Johansson wrote:
>>> On Fri, Sep 20, 2013 at 09:19:02AM -0700, Olof Johansson wrote:
>>>> On Fri, Sep 20, 2013 at 9:08 AM, Nishanth Menon <nm at ti.com> wrote:
>>>>> An alternative approach may be to (for all SoCs):
>>>>> 1. define every SoC entry - ti,omap3430 ti,omap3630...
>>>>> 2. have a generic omap3_init which uses "if (of_machine_is_compatible("ti,omap3630"))"
>>>>> to invoke the appropriate omap3xxx_init_early.
>>>>
>>>> Yes, this would be better, but you can do add a DT_MACHINE as in this
>>>> patch but have ti,omap3630 as the dt_compat table. Then there's no
>>>> need to add runtime checks.
>>>
>>> I was going to reply that adding of_machine_is_compatible("ti,omap3630")
>>> would help in some situations, but guess it's already tainted ;-)
>>
>> Oh, if it's just a few checks, then by all means go down that route. I
>> didn't look at the code to see how much it would be.
>>
>> But if a new DT_MACHINE is added, then it should definitely be based
>> on ti,omap3630 instead of listing all the boards.
This was more in the direction I was hoping to go.

> 
> the idea was to CPU compatible property to conditionally enable known
> erratas workarounds. In some cases, Revision register can't be trusted,
> so instead of creating per-errata DT properties (since that'd be
> describing the SW, in a way), I thought of using
> of_machine_is_compatible() checks, but that assumes CPU compatible is
> "correct".
I think the quirk handling is part of what Tony is attempting, and the
definition of these might help, I think..


-- 
Regards,
Nishanth Menon