[U-Boot] [PATCH v4 19/20] SPL: NAND: Enhance drivers/mtd/nand/nand_spl_simple.c

[Scott Wood]

On 08/27/2012 12:50 PM, Tom Rini wrote:
> On Mon, Aug 27, 2012 at 12:14:30PM -0500, Scott Wood wrote:
>> On 08/27/2012 12:07 PM, Tom Rini wrote:
>>> On Mon, Aug 27, 2012 at 11:16:45AM -0500, Scott Wood wrote:
>>>> On 08/27/2012 09:37 AM, Tom Rini wrote:
>>>>> On 08/24/2012 05:09 PM, Scott Wood wrote:
>>>>>> What is the benefit of putting this in nand_spl_simple.c versus another
>>>>>> file?  What if someone wants to use this with a different NAND boot
>>>>>> implementation?
>>>>>
>>>>> I would start by questioning the need of a 3rd SPL framework.
>>>>
>>>> The "simple" driver does not work for all hardware.  This is why we have
>>>> nand_spl/nand_boot_fsl_elbc.c and others in addition to
>>>> nand_spl/nand_boot.c.  It's not a "3rd SPL framework", just a different
>>>> NAND implementation.
>>>
>>> The question boils down to, what are your size constraints?  I guess
>>> what I'm saying is, if it's <4kb, it's not using this file nor the
>>> framework.
>>
>> 4K SPLs will use nand_spl_simple.c.  It is pretty much a copy of
>> nand_spl/nand_boot.c which 4K SPLs use, and Wolfgang is insisting that
>> no new boards be added to nand_spl, so they must use the new SPL (even
>> if there are no new 4xx boards, presumably such a stance by Wolfgang
>> indicates a desire to see nand_spl go away entirely at some point).
>>
>>> If we've got more than 4kb to work with, it's using the
>>> framework (with changes if needed, of course) and I guess we could move
>>> the function to common/spl/spl_nand.c and add
>>> drivers/mtd/nand/nand_spl_fsl_elbc.c and so on.  Now that I've had more
>>> coffee, do I follow your suggestion right?
>>
>> I think so.  eLBC is 4K-limited, but IFC is similar and can do an 8K SPL
>> (though we currently don't), and who knows what controllers will come
>> along in the future.
> 
> When do you plan to try and do the conversion? :) 

I started a conversion of an eLBC board recently, but ran into some bugs
that I couldn't squash by the end of the merge window -- at which point
the timeslice expiration hit and its priority dropped.

I may be able to resume next week (this week is Linux Plumbers).

> I kludged (but think it would still work) hawkboard to 887 bytes over 4kb and I see bamboo is
> 736 bytes under, leaving a 151 byte gap (in this very quick and somewhat
> silly SWAG).  So maybe we can use this framework for 4KB systems.

Perhaps for some of them.  How much does the framework add?

> And, I'll split things out for now so we can move past this.

OK, thanks.

-Scott