Re: [PATCH 6/6] spi: octeon: Add thunderx driver

[Mark Brown]

On Wed, Jul 27, 2016 at 11:25:10AM -0700, David Daney wrote:
> On 07/27/2016 11:12 AM, Mark Brown wrote:
> > On Mon, Jul 25, 2016 at 09:31:15AM -0700, David Daney wrote:

> > > ARCH_THUNDER needs to die, so perhaps it should be (ARM64 || COMPILE_TEST)
> > > && PCI && 64BIT if you really want to hide it from non-arm64 kernel configs.

> > It does?  Why?

> It adds clutter.  If we build a generic kernel, we first must select all the
> ARCH_*, then go back and select the devices we want.  Not much of a value
> add.

The value comes when moving to a new kernel version and updating your
config or doing a new board - if you're building for a specific system
then you get fewer new driver prompts (especially if you've got a
smaller number of hotpluggable buses enabled).  This is the much more
common case for people doing kernel configuration, it did bother people
a lot in the past.

> Better to just directly select the devices and remove this middle ARCH_*
> layer.

It's one option every time a new vendor appears rather than every time a
new driver appears - it's a useful quality of life improvement for
people who are doing system customization that is fairly painless for
those doing generic kernels (who can just enable everything).

> Also who is responsible for making sure the proper ARCH_* constraints are
> maintained?  If we remove ARCH_THUNDER, no need to worry about this.

People using the devices or writing drivers...  this really isn't
particularly challenging and it's not like it's hard to fix if people
notice issues.

Attachment: signature.asc
Description: PGP signature