On Mon, Feb 03, 2025 at 09:18:33AM +0100, Arnd Bergmann wrote: > On Mon, Feb 3, 2025, at 08:26, Thomas Weißschuh wrote: > > Building lib/iomap.o on UM triggers warnings about missing prototypes. > > These prototypes should be defined by asm-generic/iomap.h, depending on > > other symbols. For example "ioread64_lo_hi" is based on "readq". > > However the generic variants of those tested symbols are defined in > > asm-generic/io.h, only after asm-generic/iomap.h has already been > > included, breaking the ifdef logic. > > Sorry I never took the time to fix this so far. > > > --- a/include/asm-generic/io.h > > +++ b/include/asm-generic/io.h > > @@ -13,10 +13,6 @@ > > #include <linux/types.h> > > #include <linux/instruction_pointer.h> > > > > -#ifdef CONFIG_GENERIC_IOMAP > > -#include <asm-generic/iomap.h> > > -#endif > > - > > #include <asm/mmiowb.h> > > #include <asm-generic/pci_iomap.h> > > > > @@ -1250,4 +1246,8 @@ extern int devmem_is_allowed(unsigned long pfn); > > > > #endif /* __KERNEL__ */ > > > > +#ifdef CONFIG_GENERIC_IOMAP > > +#include <asm-generic/iomap.h> > > +#endif > > + > > #endif /* __ASM_GENERIC_IO_H */ > > I have not tried it yet, but I suspect this is not the correct > fix here. Unfortunately the indirect header inclusions in this > file are way too complicated with corner cases in various > architectures. How much testing have you given your patch > across other targets? I think the last time we tried to address > it, we broke mips or parisc. It was build-tested on 0day. I also gave it some light boot testing on kunit/qemu. (Neither on mips or parisc)
