Hi, On Thu, Jun 29, 2023 at 1:40 AM Petr Mladek <pmladek@xxxxxxxx> wrote: > > On Wed 2023-06-28 22:07:17, Randy Dunlap wrote: > > > > > > On 6/27/23 21:27, Stephen Rothwell wrote: > > > Hi all, > > > > > > Please do *not* add any v5.6 related stuff to your linux-next included > > > branches until after v5.5-rc1 has been released. > > > > > > Changes since 20230627: > > > > > > > on powerpc64, when > > # CONFIG_HARDLOCKUP_DETECTOR is not set: > > > > ../arch/powerpc/platforms/pseries/mobility.c: In function 'pseries_migrate_partition': > > ../arch/powerpc/platforms/pseries/mobility.c:753:17: error: implicit declaration of function 'watchdog_hardlockup_set_timeout_pct'; did you mean 'watchdog_hardlockup_stop'? [-Werror=implicit-function-declaration] > > 753 | watchdog_hardlockup_set_timeout_pct(factor); > > | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ > > | watchdog_hardlockup_stop > > cc1: all warnings being treated as errors > > > > > > Full randconfig file is attached. > > > > > > If I blatantly modify include/linux/nmi.h with these comments, it builds OK: > > > > //#if defined(CONFIG_HARDLOCKUP_DETECTOR_ARCH) || defined(CONFIG_HARDLOCKUP_DETECTOR_SPARC64) > > #include <asm/nmi.h> > > //#endif > > This is well describes the probem. > > The decision whether to include asm/nmi.h should not depend only on > CONFIG_HARDLOCKUP_DETECTOR_* when it is needed also for other stuff. > > Well, it could not be always included because it exists only > on 4 architectures: > > $> ls -1 arch/*/include/asm/nmi.h > arch/powerpc/include/asm/nmi.h > arch/s390/include/asm/nmi.h > arch/sparc/include/asm/nmi.h > arch/x86/include/asm/nmi.h > > and include/asm-generic/nmi.h. does not exist. > > I see two reasonable solutions: > > 1. We either create an empty "include/asm-generic/nmi.h" and > always include asm/nmi.h in linux/nmi.h. > > It looks is a bit weird but it might be the easiest solution > unless it creates another problems elsewhere. > > > 2. Explicitely include asm/nmi.h where needed, in > arch/powerpc/platforms/pseries/mobility.c and maybe somewhere as > well. > > It is already done in many cases: > > $> git grep asm/nmi.h > arch/Kconfig: asm/nmi.h, and defines its own arch_touch_nmi_watchdog(). > arch/powerpc/kernel/mce.c:#include <asm/nmi.h> > arch/powerpc/kernel/traps.c:#include <asm/nmi.h> > arch/powerpc/kernel/watchdog.c:#include <asm/nmi.h> > arch/s390/kernel/entry.S:#include <asm/nmi.h> > arch/s390/kernel/idle.c:#include <asm/nmi.h> > arch/s390/kernel/machine_kexec.c:#include <asm/nmi.h> > arch/s390/kernel/nmi.c:#include <asm/nmi.h> > arch/s390/kernel/process.c:#include <asm/nmi.h> > arch/s390/kernel/setup.c:#include <asm/nmi.h> > arch/s390/kernel/smp.c:#include <asm/nmi.h> > arch/s390/kvm/interrupt.c:#include <asm/nmi.h> > arch/s390/kvm/kvm-s390.c:#include <asm/nmi.h> > arch/s390/kvm/vsie.c:#include <asm/nmi.h> > arch/sparc/kernel/pcr.c:#include <asm/nmi.h> > arch/sparc/kernel/perf_event.c:#include <asm/nmi.h> > arch/x86/events/amd/core.c:#include <asm/nmi.h> > arch/x86/events/amd/ibs.c:#include <asm/nmi.h> > arch/x86/events/core.c:#include <asm/nmi.h> > arch/x86/kernel/alternative.c:#include <asm/nmi.h> > arch/x86/kernel/apic/hw_nmi.c:#include <asm/nmi.h> > arch/x86/kernel/cpu/mce/inject.c:#include <asm/nmi.h> > arch/x86/kernel/cpu/mshyperv.c:#include <asm/nmi.h> > arch/x86/kernel/cpu/perfctr-watchdog.c:#include <asm/nmi.h> > arch/x86/kernel/crash.c:#include <asm/nmi.h> > arch/x86/kernel/kgdb.c:#include <asm/nmi.h> > arch/x86/kernel/nmi.c:#include <asm/nmi.h> > arch/x86/kernel/nmi_selftest.c:#include <asm/nmi.h> > arch/x86/kernel/process.c:#include <asm/nmi.h> > arch/x86/kernel/reboot.c:#include <asm/nmi.h> > arch/x86/kernel/smp.c:#include <asm/nmi.h> > arch/x86/kernel/smpboot.c:#include <asm/nmi.h> > arch/x86/kernel/tsc.c:#include <asm/nmi.h> > arch/x86/platform/uv/uv_nmi.c:#include <asm/nmi.h> > drivers/char/ipmi/ipmi_watchdog.c:#include <asm/nmi.h> > drivers/edac/igen6_edac.c:#include <asm/nmi.h> > drivers/s390/cio/cio.c:#include <asm/nmi.h> > drivers/watchdog/hpwdt.c:#include <asm/nmi.h> > include/linux/nmi.h:#include <asm/nmi.h> > include/linux/nmi.h:#include <asm/nmi.h> > kernel/sysctl.c:#include <asm/nmi.h> > > I would personally prefer the 2nd variant. Yes, this seems right to me. I'm always inclined to get build failures fixed ASAP, so I've posted a patch for it: https://lore.kernel.org/r/20230629124500.1.I55e2f4e7903d686c4484cb23c033c6a9e1a9d4c4@changeid > I am going to check other locations where it might be needed. If you find anything else, we can either do a v2 or another patch. -Doug
