On 29/10/2022 04.49, Krzysztof Kozlowski wrote: > On 28/10/2022 14:51, Sasha Finkelstein wrote: >> On Fri, 28 Oct 2022 at 20:54, Krzysztof Kozlowski >> <krzysztof.kozlowski@xxxxxxxxxx> wrote: >>> >>> On 28/10/2022 12:52, Sasha Finkelstein wrote: >>>> +config PWM_APPLE >>>> + tristate "Apple SoC PWM support" >>>> + depends on ARCH_APPLE || (COMPILE_TEST && 64BIT) >>> >>> Why this code cannot be build on 32-bit? >> It uses 64-bit divisions, which causes it to fail to build on 32-bit >> mips. It should not be a >> problem, since this hardware is only present on 64-bit SoCs. > > Does not matter, code should be portable and buildable on 32-bit. If it > does not build then your code is not correct. This statement does not apply in general. There are plenty of drivers which cannot reasonably build for 32-bit, and make no sense because no 32-bit hardware exists that could use them. Examples include anything that accesses 64-bit registers on 64-bit SoCs the normal way, and further anything that touches CPU stuff like system registers. In *this* case, if the only issue is some 64-bit math, then yes, it should be made to build on 32-bit (especially since this is likely to also work for older 32-bit Apple SoCs). But the (COMPILE_TEST && 64BIT) pattern is definitely valid in other cases, and I've been adding it lately to shut up the kernel test bot since it makes no sense to compile test a whole pile of our drivers on 32-bit architectures - they fundamentally can't compile without adding pointless hypothetical broken fluff to the driver like split MMIO accesses (which often can't work on real hardware), and it serves no purpose. - Hector
