Re: [PATCH v5 12/13] spi: mxic: Use spi_mem_generic_supports_op()

[Boris Brezillon <boris.brezillon@collabora.com>]

On Wed, 15 Dec 2021 18:44:26 +0100
Miquel Raynal <miquel.raynal@bootlin.com> wrote:

> Hi Boris,
> 
> boris.brezillon@collabora.com wrote on Tue, 14 Dec 2021 17:24:10 +0100:
> 
> > On Tue, 14 Dec 2021 12:41:39 +0100
> > Miquel Raynal <miquel.raynal@bootlin.com> wrote:
> >   
> > > This driver can be simplified a little bit by using
> > > spi_mem_generic_supports_op() instead of the
> > > spi_mem_default/dtr_supports_op() couple. The all_false boolean is
> > > inverted to become a dtr boolean, which checks if at least one of the
> > > operation member uses dtr mode. The idea behind this change is to
> > > simplify the introduction of the pipelined ECC engine.
> > > 
> > > Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
> > > ---
> > >  drivers/spi/spi-mxic.c | 10 +++-------
> > >  1 file changed, 3 insertions(+), 7 deletions(-)
> > > 
> > > diff --git a/drivers/spi/spi-mxic.c b/drivers/spi/spi-mxic.c
> > > index 485a7f2afb44..5e71aa630504 100644
> > > --- a/drivers/spi/spi-mxic.c
> > > +++ b/drivers/spi/spi-mxic.c
> > > @@ -452,7 +452,7 @@ static ssize_t mxic_spi_mem_dirmap_write(struct spi_mem_dirmap_desc *desc,
> > >  static bool mxic_spi_mem_supports_op(struct spi_mem *mem,
> > >  				     const struct spi_mem_op *op)
> > >  {
> > > -	bool all_false;
> > > +	struct spi_mem_controller_caps caps = {};
> > >  
> > >  	if (op->data.buswidth > 8 || op->addr.buswidth > 8 ||
> > >  	    op->dummy.buswidth > 8 || op->cmd.buswidth > 8)
> > > @@ -465,13 +465,9 @@ static bool mxic_spi_mem_supports_op(struct spi_mem *mem,
> > >  	if (op->addr.nbytes > 7)
> > >  		return false;
> > >  
> > > -	all_false = !op->cmd.dtr && !op->addr.dtr && !op->dummy.dtr &&
> > > -		    !op->data.dtr;
> > > +	caps.dtr = op->cmd.dtr || op->addr.dtr || op->dummy.dtr || op->data.dtr;    
> > 
> > Are you sure that's what you want to do? spi_mem_controller_caps is
> > supposed to encode the controller capabilities, not whether the
> > operation contains a DTR cycle or not. I'd expect this caps object to be
> > statically defined, with possibly one instance per-compat if the caps
> > depend on the HW revision.  
> 
> In order to keep the series easy to review I decided to go for the
> following approach:
> * Introduce the spi_mem_generic_supports_op_helper() which takes a
>   capabilities structure. This helper gathers all the checks from
>   spi_mem_default_supports_op() and spi_mem_dtr_supports_op(). These
>   two helpers now call the new one with either a NULL pointer in the
>   former case, or a structure with the .dtr parameter set to true in
>   the latter.
> * Change the API of spi_mem_default_supports_op(), this involves
>   updating many different drivers so this change does only that in a
>   very transparent way, with no functional changes at all. All the
>   drivers provide a NULL parameter for the capabilities structure.
> * Actually make use of the new parameter of
>   spi_mem_default_supports_op() in the drivers Cadence and Macronix,
>   which do have DTR support. This kills the spi_mem_dtr_supports_op()
>   helper.
> * Kill the temporary spi_mem_generic_supports_op() helper by moving
>   all the logic back into spi_mem_default_supports_op().
> 
> This approach is really straightforward and easily bisectable if
> needed.

There's also a second option that doesn't involve patching existing
users: add a spi_mem_controller_caps to the spi_controller struct, and
check this instance in your spi_mem_default_supports_op()
implementation. Note that the buswidth check done in the generic
helper is already based on caps exposed by the controller
through spi_controller.mode_bits ({RX/TX}_{DUAL,QUAD,OCTAL} bits).

> While working on this, I fixed the check we discussed on IRC
> about the command parameter when in a DTR operation. I also reverted
> the logic in the various checks, as you suggested.
> 
> Thanks,
> Miquèl