Hi Mark,

On Tuesday 29 October 2013 10:23:31 Mark Brown wrote:

On Tue, Oct 29, 2013 at 06:05:53PM +0100, Laurent Pinchart wrote:

quoted

The first one is that I can't compile-test all those drivers on all
architectures. The spi-sh-msiof driver, for instance, uses
io(read|write)(16|

Which architectures are these and is there not a symbol we can depend on
for them?

arch/cris for instance. We can use readl/writel instead (maybe it would be 
time to rationalize and document the I/O accessors across all architectures, 
but that's another topic).

My point is that there might be other issues that I won't be able to easily 
catch. This would break compilation for everybody for no reason, as the 
drivers are useless on non-SuperH, non-ARM platforms. That's why I believe 
COMPILE_TEST would be a better option as a first step.

quoted

32) which are not available on all architectures. There might be other
similar problems that I can't catch, and I don't want to introduce build
breakages in the kernel.

This is easy enough to handle if we do run into issues, it seems better to
get things available than try to step through architecture by architecture.

quoted

The second reason is that, as the IP cores have never been used on
anything but SuperH and ARM, I don't like the idea of clobbering the
config process with drivers that are useless on the target architecture.
Now that we have a COMPILE_TEST Kconfig option, my preference would thus
go to SUPERH || ARM || COMPILE_TEST over no dependency at all.

That's not what you did, though - you're not adding COMPILE_TEST.

No, but I can add it :-) If this gets agreed upon, I'll respin the series with 
|| COMPILE_TEST.

-- 
Regards,

Laurent Pinchart