[PATCH] RFC: mtd/spi-nor: add checking for the spi_nor_read

[Michal Suchanek]

On 13 August 2015 at 11:03, Hou Zhiqiang <B48286 at freescale.com> wrote:
> Hello,
>
> I want to add the 4-Byte addressing flashes support for the Freescale eSPI
> controller dirver. Appreciate your suggestions.
>
> Background:
> In the Freescale eSPI controller dirver, if the transaction length exceed
> 64KiB, the contents of the transaction was touched to update the command
> with assumed the 3-Byte address width. It is a workaround, due to the
> Freescale eSPI controller have a hardware limit that the maximum one-time
> transaction length is 64KiB, that isn't consistent with many other vendors'
> SPI controllers, and so far the SPI flash driver assume the SPI controller
> have the ability to complete the specified transaction length at one-time.
>
> But this workaround is only suit for 3-Byte addressing slaves, as time goes
> by, there are 4-Byte address width SPI flashes, so this assumption isn't
> correct and this workaround discombobulate the controller driver layer and
> protocol driver layer. So, this workaround should be removed and the SPI
> client driver should ensure the transaction completion.
>
> There are two solutions:
> 1. In spi-nor framework, compare the retlen with the transaction length
> specified, if the retlen less than the transaction length and with no error
> number returned, re-initiate the transaction with the updated address.
> Advantage:
> It won't affect other SPI controllers without this limit.
> Disadvantage:
> It is not a systematic solution.

This can probably break transactions that are performed on non-flash
slaves but look like flash read command.

>
> 2. Add quirk support, something like i2c quirk.
> Advantage:
> It is a systematic solution.
> Disadvantage:
> The quirk mechanism is only useful for Freescale eSPI controller, but it will
> affect whole SPI framework.

I added this quirk in m25p80.c since my current SPI master driver is
just broken so I tried to work around that. While the quirk itself is
rejected as non-syetematic this series that adds the required support
for checking the return value of the SPI master driver transfer
functions.

So you have options:

2a: add SPI master quirk and check it in m25p80.c and truncate the
transfer in m25p80.c

2b: truncate the transfer in SPI master driver and just return the
number of bytes transferred

This patch series should handle fragmenting the transfers in spi-nor.c

http://lists.infradead.org/pipermail/linux-mtd/2015-July/060466.html

3: if your controller has manual CS driving support fragment the
transfer in your SPI master driver into chunks the controler can
transfer in one go.

Thanks

Michal