Re: [PATCH] spi-nor: Add support for Atmel Dataflash memories

[Boris Brezillon]

Hi Nicolas,

On Mon, 7 May 2018 10:23:56 +0200
Nicolas Ferre <nicolas.ferre@xxxxxxxxxxxxx> wrote:

> On 04/05/2018 at 20:38, Boris Brezillon wrote:
> > Hi Radu,
> > 
> > Sorry for the late reply.
> > 
> > On Wed, 28 Feb 2018 13:55:01 +0200
> > Radu Pirea <radu.pirea@xxxxxxxxxxxxx> wrote:
> >   
> >> This patch add support in spi-nor for allmost all dataflash memories
> >> supported by old mtd_dataflash driver.  
> > 
> > Those devices clearly use a different instruction set, so I don't think
> > they fit in this framework. Can you tell us why you want to move
> > dataflash support to the SPI NOR framework. I think I know why, but I'd
> > like to get your version. My guess is that some people want to connect
> > dataflash chips to the Atmel QSPI controller, and it's not supported
> > right now because the Atmel QSPI controller implements the SPI-NOR
> > interface and not the generic SPI one, thus preventing anything that
> > is not a SPI NOR from being connected to this controller.
> > 
> > If I'm right, then the solution is to convert the QSPI driver to the
> > spi-mem interface [1] and move it to drivers/spi/.  
> 
> No, I we didn't think about this. Dataflash is not so popular those days 
> and we don't want to revive it anyway. Our QSPI driver has already a lot 
> of things to handle in QSPI-related topics to not mix it with oldies ;-)
> 
> The rationale behind this work is to get rid of the very old dataflash 
> standalone driver and benefit from the whole spi-nor infrastructure like 
> cache coherency management and DMA handling (which were broken in the 
> old dataflash driver in recent kernels).

Still don't think that's a good move, especially since those flashes are
using a completely different instruction set and are not exactly
behaving like SPI NORs. If we need some of the spi-nor logic to help
handle dataflash chips in a more efficient/safe way, then those bits
should be exposed as helpers at the MTD level instead of turning
spi-nor into a Frankenstein framework.

Regards,

Boris