On Fri, Apr 16, 2021 at 11:55:45AM +0100, Joe Burmeister wrote:

> I think I've stubbled into an issue with some SPI drivers.

Like Lukas says you need to copy the maintainers of those drivers.

> Even if the driver did enforce its limit, the returned status could be ignored.
> Patch for that also attached.

See submitting-patches.rst for how to submit a patch to the kernel,
Lukas flagged up most of the issues here.

> So question is, what is the fix. Currently num_chipselect is not only not enforced
> but changed to what is in the device tree by_spi_get_gpio_numbers and
> spi_get_gpio_descs. Some drivers may not really have a limit anyway.

It's really driver specific if it needs or uses the property so any
issues should be at the driver level.  Ideally the property should be
completely redundant, we can figure out how many chip selects we need to
handle based on knowing what devices are connected and what the
controller is but the binding was standardised ages ago.