Re: [PATCH v4 04/10] mtd: pxa3xx_nand: rework flash detection and timing setup

[Ezequiel Garcia <ezequiel.garcia@free-electrons.com>]

On 04/16/2015 10:41 AM, Sebastian Hesselbarth wrote:
> On 04/16/2015 03:10 PM, Ezequiel Garcia wrote:
>> On 04/15/2015 04:11 PM, Sebastian Hesselbarth wrote:
>>> On 15.04.2015 19:24, Antoine Tenart wrote:
>>>> Rework the pxa3xx_nand driver to allow using functions exported by the
>>>> nand framework to detect the flash and to configure the timings.
>>>>
>>>> Because this driver supports some non-ONFI devices, we also keep the
>>>> custom timing setup of this driver so these devices won't break.
>>>>
>>>> Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
>>>> ---
>>> [...]
>>> How about we get rid of the driver specific timings completely
>>> and pick up the best onfi timing match instead? The nand_ids table
>>> allows for a default_onfi_timing parameter even if onfi itself is
>>> not supported.
>>>
>>> For generic flash, i.e. no specific entry in the nand_ids table,
>>> we either choose onfi mode 0 (most conservative) or an even slower
>>> one.
>>>
>>
>> I think Robert mentioned [1] that using "ONFI default timings" on
>> non-ONFI devices didn't work for him.
>>
>> [1] https://lkml.org/lkml/2015/3/8/124
> 
> Ok, I see. But there is still the option to pass board specific
> timings with driver's platform_data. We could use
> 
> (a) pdata timings if passed
> (b) onfi timings if available
> (c) equivalent onfi timings if set
> (d) conservative equivalent onfi timings otherwise
> 

Right, using platform_data sounds like a nice compromise solution.
I'm willing to accept this series with the current timing rework; and
leave the timing setup in-driver replacement for followup patches.

> All we need is a function to convert sdr_timings to sane driver
> timings. And we really need to split this patch into tiny pieces
> otherwise it is not reviewable - or at least I need a full overview
> about the driver first.
> 

I think that's a bit of a different issue. This patch seems to be doing
two things: it removes the in-driver flash detection *and* reworks
timing setup.

How about we split this in two or even three patches? Along these lines:
1) introduce timing helpers, 2) rework timing setup, 3) remove in-driver
flash detection. Not sure how feasible it is.

> Also, as soon as Robert moves pxa3xx boards fully to DT, we'll loose
> the pdata timings option above. *sigh*
> 

Well, such move would include proper timing DT properties for non-ONFI
devices.
-- 
Ezequiel García, Free Electrons
Embedded Linux, Kernel and Android Engineering
http://free-electrons.com