[U-Boot] [PATCH v3 1/3] cmd_sf: add 'release' command

[Valentin Longchamp]

On 25/11/2015 04:52, Stefan Roese wrote:
> Hi Valentine,
> 
> On 24.11.2015 17:02, Valentin Longchamp wrote:
> 
> <snip>
> 
>>>>> Does your board use driver model from SPI and SPI flash? If not I
>>>>> think that should be the first step.
>>>>>
>>>>
>>>> No we don't. Could you please elaborate on how this would cover this use case
>>>> and should be the first step ?
>>>>
>>>> I am open to other ways to cover this use case of ours, especially since this
>>>> was done more than 2 years ago and u-boot has changed since then. However I
>>>> don't see the direct link between the driver model and how it would allow to
>>>> make sure spi_flash_free() is called in our u-boot env scripts.
>>>
>>> In driver model you would have a remove() method for your SPI driver
>>> which does the required pinmux changing.
>>
>> OK, when looking at SPI flash, SPI controller driver and the driver model, I
>> have found out why we require this "release" command.
>>
>> So the SPI subsystem and its drivers (with or without DM support) make sure that
>> spi_claim_bus/release_bus are called before/after the accesses. This should
>> cover the pinmux configuration stuff.
>>
>> In our case, it was not sufficient since drivers/spi/kirkwood_spi.c does the
>> pinmux configuration at 2 places: spi_claim_bus/release_bus for the IO
>> (MOSI/MISO/CLK) AND spi_setup_slave/free_slave for the chip select.
>>
>> Since spi_free_slave is not always called after spi_setup_slave (for instance
>> after a sf probe command on the u-boot prompt), the CS pin was not always given
>> back to the NAND flash controller that's why there was a need for a sf release
>> command in our case.
>>
>> I have now moved all the pinmux configuration for kirkwood_spi into
>> spi_claim_bus/release_bus and the need for this sf release command is not
>> necessary anymore.
>>
>> I am going to test this a bit more and send a new patch series which will not
>> require this release command.
>>
>>>
>>> There is a detailed howto in doc/driver-model showing how to port your
>>> driver over. Please let me know if you need any help/ideas.
>>
>> I have had a look at it and it is a great help on porting a driver to this
>> model. That's something we plan to do for our boards eventually.
> 
> So did you already port the kirkwood SPI driver to DM? I'm asking, since
> I also do have a patch for this in the queue. I'll hopefully be able
> to post it by end of this week.

No I didn't. I was just looking at what should be done for that. For my above
problem I want to send a patch to the current driver without porting it at all.

> 
> BTW: I would love to see the Kirkwood platform to be moved over to
> arch-mvebu at some time, which already supports DM.
> 

Tell me if you're doing something in this area, I may give some help or test a
few things, according to the remaining time besides my current projects. The
Kirkwood platform would definitely be better if this was done.

Valentin