+Thomas Bogendoerfer +linux-mips On 07/02/2023 14:40, Mark Brown wrote: > On Tue, Feb 07, 2023 at 10:56:04AM +0100, Guillaume Tucker wrote: >> On 07/02/2023 09:53, Ingo Molnar wrote: > >>>> 4 cc1: error: ‘-mloongson-mmi’ must be used with ‘-mhard-float’ >>> >>> ... but this exact same error was 'reported' a year ago on January 22: >>> >>>> 2 cc1: error: ‘-mloongson-mmi’ must be used with ‘-mhard-float’ > >>> So these regression reports are useless in this form and they clutter >>> people's inboxes. Is any person reading them and acting on them to make >>> sure these emails are sensible? > >> About the actual kernel build error, I guess it's a shame nobody >> has fixed this yet. I'll take a look what the root cause might >> be. If it's a KernelCI build configuration issue e.g. using >> wrong compiler flags then we can fix that, otherwise it's >> probably something to report more directly to some MIPS >> maintainers. > > Honestly at this point I think we should just drop the affected MIPS > configurations at this point, as Ingo says they've been failing for > so long with nobody caring. Maybe the MIPS people just haven't seen this. I've added Thomas and the linux-mips list to confirm. After some investigation, it turns out the error happens when doing "make modules_install". Here's the issue: * modules_install is listed in "no-compiler-targets" in the top-level Makefile * as a result, scripts/Makefile.compiler is not included * arch/mips/loongson64/Platform requires the "cc-option" function to add -mnon-loongson-mmi * since "cc-option" is not defined when just doing "make modules_install", the flag is not added and the error mentioned above occurs GitHub issue: https://github.com/kernelci/kernelci-project/issues/176 Here's a hack to prove this point, need-compiler is defined in the top-level Makefile so it shouldn't be used here but this "fixes" the problem: diff --git a/arch/mips/Makefile b/arch/mips/Makefile index 490dea07d4e0..024f62dbef76 100644 --- a/arch/mips/Makefile +++ b/arch/mips/Makefile @@ -317,10 +317,12 @@ KBUILD_CFLAGS += -fno-asynchronous-unwind-tables KBUILD_LDFLAGS += -m $(ld-emul) ifdef CONFIG_MIPS +ifdef need-compiler CHECKFLAGS += $(shell $(CC) $(KBUILD_CFLAGS) -dM -E -x c /dev/null | \ grep -E -vw '__GNUC_(MINOR_|PATCHLEVEL_)?_' | \ sed -e "s/^\#define /-D'/" -e "s/ /'='/" -e "s/$$/'/" -e 's/\$$/&&/g') endif +endif OBJCOPYFLAGS += --remove-section=.reginfo I guess another way would be to unconditionally add the options to the cflags, in fact there are other places where this appears to be done. I'm not sure which GCC or Clang versions support it or not, so that may not work in practice. diff --git a/arch/mips/loongson64/Platform b/arch/mips/loongson64/Platform index 473404cae1c4..b7b2db13f1a2 100644 --- a/arch/mips/loongson64/Platform +++ b/arch/mips/loongson64/Platform @@ -12,7 +12,7 @@ endif # Some -march= flags enable MMI instructions, and GCC complains about that # support being enabled alongside -msoft-float. Thus explicitly disable MMI. -cflags-y += $(call cc-option,-mno-loongson-mmi) +cflags-y += -mno-loongson-mmi # # Loongson Machines' Support This was reproduced using GCC 10 with the latest kernelci/gcc-10:mips-kselftest-kernelci Docker image running as root. Is this something someone familiar with MIPS would like to fix properly? If not then please confirm that we should just drop the loongson2k_defconfig builds from KernelCI. Is this kernel config actually still maintained in mainline? Thanks, Guillaume
