On Tue, 2019-12-03 at 16:51 +0000, Jonathan Cameron wrote:
> On Tue, 3 Dec 2019 16:38:50 +0000
> Mark Brown <broonie@xxxxxxxxxx> wrote:
>
> > On Sun, Dec 01, 2019 at 11:40:32AM +0000, Jonathan Cameron wrote:
> >
> > > +CC Mark as we probably need a more general view point on
> > > the question of whether SPI mode should be enforced by binding
> > > or in the driver.
> >
> > Not sure I see the question here, I think I was missing a bit of
> > the conversation? It's perfectly fine for a driver to specify a
> > mode, if the hardware always uses some unusual mode then there's
> > no sense in forcing every single DT to set the same mode. On the
> > other hand if there's some configuration for the driver that was
> > handling some board specific configuration that there's already
> > some generic SPI support for setting then it seems odd to have a
> > custom driver specific configuration mechanism.
> >
>
> If the driver picks a mode because that's what it says on the datasheet
> it prevents odd board configurations from working. The question
> becomes whether it makes sense in general to assume those odd board
> conditions don't exist until we actually have one, or to assume that
> they might and push the burden on to all DT files.
>
> Traditionally in IIO at least we've mostly taken the view the DT
> should be right and complete and had bindings state what normal
> parameters must be for it to work (assuming no inverters etc)
>
> If we encode it in the driver, and we later meet such a board we
> end up with a custom dance to query the DT parameters again and
> only override if present.
>
> We can't rely on the core SPI handling because I don't think
> there is any means of specifying a default.
>
> We can adopt the view that in general these weird boards with inverters
> are weird and just handle them when they occur. Sounds like that is your
> preference, at least for new parts.
>
> For old ones we have no idea if there are boards out there using
> them with inverters so easiest is probably to just carry on putting them
> in the DT bindings.
There might be a few other options, which would require some SPI OF change.
One example (for spi-cpha):
if (of_property_read_u32(nc, "spi-cpha", &tmp) == 0) {
spi->mode |= SPI_CPHA_OVERRIDE;
if (tmp)
spi->mode |= SPI_CPHA;
} else
if (of_property_read_bool(nc, "spi-cpha"))
spi->mode |= SPI_CPHA;
Or another option could be:
if (of_property_read_bool(nc, "spi-cpha-override")) {
spi->mode |= SPI_CPHA_OVERRIDE;
if
(of_property_read_bool(nc, "spi-cpha"))
spi->mode |= SPI_CPHA;
Naturally, this would require that spi_setup() checks SPI_CPHA_OVERRIDE and
doesn't set SPI_CPHA if SPI_CPHA_OVERRIDE is set.
Or maybe, a more complete solution would be an "spi-mode-conv" driver.
Similar to the fixed-factor-clock clk driver, which just does a computation
based on values from the DT.
To tell the truth, this would be a great idea, because we have something
like a passive 3-wire-to-4-wire HDL converter. This requires that the
driver be configured in 3-wire mode, the SPI controller in normal 4-wire.
That's because the SPI framework does a validation of the supported modes
(for the SPI controller) and invalidates what the device wants (which is
very reasonable).
An "spi-mode-conv" driver would also handle this 3-wire-4-wire dance, and
the level inversions, and other (similar) things.
Thoughts?
Alex
>
> Jonathan
>
>
