Re: [RFC PATCH 3/9] of: mtd: add NAND timings retrieval support

[boris brezillon <b.brezillon@overkiz.com>]

On 08/01/2014 20:13, Jason Gunthorpe wrote:
> On Wed, Jan 08, 2014 at 08:00:02PM +0100, boris brezillon wrote:
>> Hello Jason,
>>
>> Le 08/01/2014 19:34, Jason Gunthorpe a ?crit :
>>> On Wed, Jan 08, 2014 at 03:21:58PM +0100, Boris BREZILLON wrote:
>>>
>>>> +int of_get_nand_timings(struct device_node *np, struct nand_timings *timings)
>>>> +{
>>>> +	memset(timings, 0, sizeof(*timings));
>>>> +	of_property_read_u32(np, "tCLS-min", &timings->tCLS_min);
>>>> +	of_property_read_u32(np, "tCLH-min", &timings->tCLH_min);
>>>> +	of_property_read_u32(np, "tCS-min", &timings->tCS_min);
>>> [..]
>>>
>>> A while ago when discussing another controller it was pointed out
>>> these values are all auto-probable directly from the NAND via a ONFI
>>> defined GET FEATURE @0x01 query, and adding these timings to the DT
>>> was NAK'd..
>>>
>>> Basically you set the interface to the slowest ONFI timing mode, do
>>> the GET FEATURE to the NAND chip and then increase the interface speed
>>> to the highest mutually supported ONFI mode.
>>> Is there some reason you need to encode this in the DT?
>> What if the NAND does not support the ONFI interface (and this is
>> exactly the case for the NAND available on the cubietruck board:
>> H27UCG8T2ATR).
> Sounds like a good reason to me!
>
> You might want to check if you can boil down the DT timings from the
> huge list to just an ONFI mode number..

Sure, but the sunxi driver needs at least 19 of them...

>
> I'd echo Rob's comments, the property needs to include the units
> in the name, and I strongly recommend picoseconds for these
> values.

Agreed, picosecond is a more future-proof unit.

>
> Also, you might want to check that the ONFI names for these parameters
> are used, not a vendor specific name to try and avoid confusion.

I'll check it.

Thanks.

Best Regards,

Boris

>
> Jason