Re: [RFC v3 PATCH 1/3] dt: device tree bindings for DDR memories

[Aneesh V <aneesh@ti.com>]

On Saturday 21 January 2012 12:58 PM, Olof Johansson wrote:
> On Thu, Jan 19, 2012 at 12:56 PM, Aneesh V<aneesh@ti.com>  wrote:
>> Hi Olof,
>>
>>
>> On Friday 20 January 2012 01:01 AM, Olof Johansson wrote:
>>>
>>> Hi,
>>>
>>> Sorry for the delay in responding, I know you pinged me about it
>>> yesterday.
>>>
>>> On Thu, Jan 19, 2012 at 6:31 AM, Aneesh V<aneesh@ti.com>    wrote:
>>>>
>>>> device tree bindings for LPDDR2 SDRAM memories compliant
>>>> to JESD209-2 standard.
>>>>
>>>> The 'lpddr2' binding in-turn uses another binding 'lpddr2-timings'
>>>> for specifying the AC timing parameters of the memory device at
>>>> different speed-bins.
>>>
>>>
>>> As I just commented on the thread with Mike, I think we would be
>>> better off sticking to embedding a standard JEDEC SPD structure in the
>>> device tree. It's not large (128-256 bytes depending on memory type),
>>> and it's clearly defined and used all over the industry.
>>>
>>> It also has the benefit of reusing parsing code if you ever end up
>>> with a system that uses DIMMs for memory, thus needing to parse the
>>> SPD on said modules.
>>
>>
>> I did mention in the previous thread why SPD doesn't work for us ([1] and
>> [2]). Let me repeat the key points here.
>
> Ah, sorry. Missed it in the chain of replies.
>
>> 1. I couldn't find an SPD addendum for LPDDR2 from the JEDEC website.
>> 2. This seems to indicate that SPD is not used for LPDDR2 devices.
>
> Bummer. I'm guessing most applications where LPDDR* is used won't be
> suitable for modular memory, so there's not the same need for SPD.
>
>> 3. I tried to see if I can fit the DDR3 or DDR2 SPD for our needs. But
>> some of the AC timing parameters needed by our controller are not
>> available in those layouts.
>
> Are those properties of the memory, or a combination of memory and
> board properties? I think it still makes sense for the memories that
> do have it to use the SPD format and extend with additional
> properties, at least if it's only a few additional properties needed.

They are AC timing parameters defined by the spec such as tCKESR, tFAW
etc, nothing specific to OMAP.

As for other memories, we do not intend to support any at the moment.
In my initial version I had bindings for DDR3, because we intended to
have DDR3 support in the driver. But we have since decided to drop DDR3
support for the following reasons:

1. According to the DDR3 spec, the operating frequency range for the
speed-bins is limited (unlike LPDDR2). Scaling DDR frequency in this
small range doesn't make sense for us (this may be the case for other
platforms too).

2. Unlike LPDDR2, DDR3 doesn't have a mechanism for polling the
temperature from the device and derate timings.

If DVFS and thermal handling are not relevant for DDR3, having a kernel
driver doesn't make sense. One-time settings in the bootloader are good
enough.

So, I have made my binding only for LPDDR2 and removed DDR3 parts from
it.


>
>> I don't see any option other than defining a new binding for LPDDR2.
>
> Yeah, agreed.

Thanks,
Aneesh