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

[Valentin Longchamp <valentin.longchamp@keymile.com>]

On 24/11/2015 02:49, Simon Glass wrote:
> Hi Valentine,
> 
> On 23 November 2015 at 02:19, Valentin Longchamp
> <valentin.longchamp@keymile.com> wrote:
>> Hi Simon,
>>
>> On 20/11/2015 18:19, Simon Glass wrote:
>>> Hi,
>>>
>>> On 20 November 2015 at 03:13, Valentin Longchamp
>>> <valentin.longchamp@keymile.com> wrote:
>>>> On 19/11/2015 17:57, Jagan Teki wrote:
>>>>> On 13 November 2015 at 18:55, Valentin Longchamp
>>>>> <valentin.longchamp@keymile.com> wrote:
>>>>>> The release command is the pendant of the probe command. This command
>>>>>> allows to call spi_flash_free from the command line. This may be
>>>>>> necessary for some boards where sf probe does change the state of the
>>>>>> hardware (like with some pin multiplexing changes for instance).
>>>>>
>>>>> So you want to change the state of pin multiplexing on your board with
>>>>> connected slave devices example: spi nor flash is it? what exactly the
>>>>> need of releasing? why can't we use pin multiplexing changes like
>>>>> selecting or deselecting particular lines through driver or from board
>>>>> files itself.
>>>>
>>>> That's our use case yes. Let me explain you it again in detail. Some of the
>>>> signals used to access the NAND Flash and the SPI NOR are shared. At reset, they
>>>> are available for the SPI NOR, since u-boot is in there and the CPU then
>>>> accesses it.
>>>>
>>>> In an usual boot sequence, the SPI NOR is accessed first (copying u-boot to the
>>>> RAM, reading out the environment) and so the pins are configured in hardware at
>>>> boot time for accessing the SPI NOR. After that, they are configured to access
>>>> the NAND where the kernel and filesystem are stored to boot Linux thanks to
>>>> env_relocate_spec() calling spi_flash_free() on exit in conjunction with [1]
>>>>
>>>> Now in the case where the boot sequence is interrupted and some accesses are
>>>> done to the SPI NOR, the pins are changed again to SPI NOR to perform these
>>>> accesses. But we have to make sure that the pins are configured back to NAND by
>>>> calling spi_flash_free() after these accesses and that's why I introduced the
>>>> release() function.
>>>>
>>>> In our case, there are 2 types of such accesses:
>>>> - environment variables write: this is the first patch of the series. It simply
>>>> adds calls to spi_flash_free() at function exit no only in env_relocate_spec()
>>>> but also in saveenv() so that the behavior here is coherent for the whole env_sf
>>>> file (spi_flash_probe() at function start, spi_flash_free() at function exit).
>>>> - updating u-boot: this is solved for us with the last 2 patches of the series.
>>>> The first one just adds a sf release command that does the opposite/cleanup to
>>>> sf probe and the second patch just calls this command in our scripts where
>>>> u-boot is updated/the SPI NOR is written.
>>>>
>>>> We are *indeed* using pin multiplexing changes, in our case, they are
>>>> implemented in the spi controller driver: drivers/spi/kirkwood_spi.c. To be very
>>>> specific, in our case this sf release command allows to explicitely call
>>>> spi_flash_free() which calls spi_free_slave(), which in our case
>>>> (kirkwood_spi.c) sets the pins back to their previous configuration.
>>>
>>> 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.

Valentin