Re: [PATCH 00/16] mtd: spi-nor: aspeed: AST2600 support and extensions

[Boris Brezillon <boris.brezillon@xxxxxxxxxxxxx>]

On Fri, 11 Oct 2019 11:29:49 +0200
Cédric Le Goater <clg@xxxxxxxx> wrote:

> On 11/10/2019 08:45, Boris Brezillon wrote:
> > On Thu, 10 Oct 2019 23:47:45 +0000
> > Joel Stanley <joel@xxxxxxxxx> wrote:
> >   
> >> On Wed, 9 Oct 2019 at 20:56, Boris Brezillon
> >> <boris.brezillon@xxxxxxxxxxxxx> wrote:  
> >>>
> >>> Hi Cedric,
> >>>
> >>> On Fri,  4 Oct 2019 13:59:03 +0200
> >>> Cédric Le Goater <clg@xxxxxxxx> wrote:
> >>>    
> >>>> Hello,
> >>>>
> >>>> This series first extends the support for the Aspeed AST2500 and
> >>>> AST2400 SMC driver. It adds Dual Data support and read training giving
> >>>> the best read settings for a given chip. Support for the new AST2600
> >>>> SoC is added at the end.
> >>>>
> >>>> I understand that a new spi_mem framework exists and I do have an
> >>>> experimental driver using it. But unfortunately, it is difficult to
> >>>> integrate the read training. The Aspeed constraints are not compatible
> >>>> and i haven't had the time to extend the current framework.    
> >>>
> >>> Hm, I don't think that's a good reason to push new features to the
> >>> existing driver, especially since I asked others to migrate their
> >>> drivers to spi-mem in the past. I do understand your concerns, and I'll
> >>> let the SPI NOR/MTD maintainers make the final call, but I think it'd
> >>> be better for the SPI MEM ecosystem to think about this link-training
> >>> API (Vignesh needs it for the Cadence driver IIRC) rather than pushing
> >>> this kind of feature to spi-nor controller drivers.    
> >>
> >> As Cedric mentioned, the OpenBMC project has been shipping the read
> >> training code for the ast2400/ast2400 for several years now. It would
> >> be great to see it in mainline.
> >>
> >> I think it's reasonable to ask for the driver to be moved to the
> >> spi-mem subsystem once it has the required APIs.  
> > 
> > Except it won't have the necessary APIs unless someone works on it, and
> > adding this feature to existing spi-nor drivers won't help achieving
> > this goal.  
> 
> 
> What would you suggest ? Something like the patch below which would
> call a 'train' operation at the end of spi_add_device().

This has been discussed in the past with Vignesh, but I can't find the
thread where this discussion happened. My understanding was that link
training would use a command with well-known output (is there a
dedicated SPI NOR command for that?) and test different clk settings
until it finds one that works.

> 
> Also, when doing read training, we might need to know some lowlevel 
> characteristics of the chip being trained. Should we offer a way 
> to grab the probed m25p80 device and give access to the underlying 
> 'struct spi_nor' ? 
> 
>   static struct spi_nor *spi_get_pnor(struct spi_device *spi)
>   {
> 	struct spi_mem *spimem = spi_get_drvdata(spi);
> 	struct m25p *flash = spi_mem_get_drvdata(spimem);
> 
> 	return flash ? &flash->spi_nor : NULL;
>   }
> 
> Yeah, it's hideous. I just want to raise the issue.

Oh no. We definitely don't want to expose the spi_nor chip to the
spi_mem layer, but, if needed, we can add more fields to spi_mem and
let the spi_mem driver fill them. We just need to figure out what's
really needed.

> 
> Thanks,
> 
> C. 
> 
> 
> From b34297e6b991ff051bc1e16103d14b2a05c81827 Mon Sep 17 00:00:00 2001
> From: =?UTF-8?q?C=C3=A9dric=20Le=20Goater?= <clg@xxxxxxxx>
> Date: Fri, 11 Oct 2019 11:09:33 +0200
> Subject: [PATCH] spi: core: Add a device link training operation
> MIME-Version: 1.0
> Content-Type: text/plain; charset=UTF-8
> Content-Transfer-Encoding: 8bit
> 
> Signed-off-by: Cédric Le Goater <clg@xxxxxxxx>
> ---
>  include/linux/spi/spi.h |  4 ++++
>  drivers/spi/spi.c       | 23 +++++++++++++++++++++++
>  2 files changed, 27 insertions(+)
> 
> diff --git a/include/linux/spi/spi.h b/include/linux/spi/spi.h
> index af4f265d0f67..950b39304807 100644
> --- a/include/linux/spi/spi.h
> +++ b/include/linux/spi/spi.h
> @@ -409,6 +409,7 @@ static inline void spi_unregister_driver(struct spi_driver *sdrv)
>   * @fw_translate_cs: If the boot firmware uses different numbering scheme
>   *	what Linux expects, this optional hook can be used to translate
>   *	between the two.
> + * @train : perform device link training
>   *
>   * Each SPI controller can communicate with one or more @spi_device
>   * children.  These make a small bus, sharing MOSI, MISO and SCK signals
> @@ -604,6 +605,9 @@ struct spi_controller {
>  	void			*dummy_tx;
>  
>  	int (*fw_translate_cs)(struct spi_controller *ctlr, unsigned cs);
> +
> +	int			(*train)(struct spi_device *spi);

Was more thinking of something like:

	int (*link_setup)(struct spi_mem *mem,
			  struct spi_mem_op *op_template,
			  ...);

where the op_template would potentially differ depending on the type of
memory (NOR, NAND, SRAM?). I also don't know what other params would be
needed to do the link training.

BTW, this hook should be in the spi_mem_controller_ops struct not in
spi_controller.

> +
>  };
>  
>  static inline void *spi_controller_get_devdata(struct spi_controller *ctlr)
> diff --git a/drivers/spi/spi.c b/drivers/spi/spi.c
> index 75ac046cae52..759a66d74822 100644
> --- a/drivers/spi/spi.c
> +++ b/drivers/spi/spi.c
> @@ -542,6 +542,22 @@ static int spi_dev_check(struct device *dev, void *data)
>  	return 0;
>  }
>  
> +/**
> + * spi_train - link training of SPI device
> + * @spi: the device whose being trained
> + *
> + * Return: zero on success, else a negative error code.
> + */
> +static int spi_train(struct spi_device *spi)
> +{
> +	int		status = 0;
> +
> +	if (spi->controller->train)
> +		status = spi->controller->train(spi);
> +
> +	return status;
> +}
> +
>  /**
>   * spi_add_device - Add spi_device allocated with spi_alloc_device
>   * @spi: spi_device to register
> @@ -606,6 +622,13 @@ int spi_add_device(struct spi_device *spi)
>  	else
>  		dev_dbg(dev, "registered child %s\n", dev_name(&spi->dev));
>  
> +	status = spi_train(spi);
> +	if (status < 0) {
> +		dev_err(dev, "can't train %s, status %d\n",
> +				dev_name(&spi->dev), status);
> +		goto done;
> +	}
> +
>  done:
>  	mutex_unlock(&spi_add_lock);
>  	return status;


______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/