On Thu, Jan 23, 2020 at 10:31:20AM +0100, Krzysztof Kozlowski wrote:

On Thu, Jan 23, 2020 at 10:12:28AM +0100, Wolfram Sang wrote:

quoted

quoted

 config I2C_ZX2967
 	tristate "ZTE ZX2967 I2C support"
-	depends on ARCH_ZX
-	default y
+	depends on ARCH_ZX || (COMPILE_TEST && (ARC || ARM || ARM64 || M68K || RISCV || SUPERH || SPARC))
+	# COMPILE_TEST needs architectures with readsX()/writesX() primitives

The list of archs neither looks pretty nor very maintainable. My
suggestion is that we leave this out of COMPILE_TEST until we have
something like ARCH_HAS_READS or something. What do you think?

Indeed it does not look good. However having compile testing allows
kbuild to run sparse and smatch which already started pointing minor
issues in existing drivers.

Yeah... pros and cons... I don't have a strong opinion to keep it. Since
patch is important, maybe let's just skip this part?

Yeah, let's skip it for now. If you or someone is keen on having it,
something like ARCH_HAS_READS (<- needs better name) should be
introduced so we can use it here. But that can/should be handled
incrementally.