Gautam Menghani <gau...@linux.ibm.com> writes:
> As per the kernel documentation[1], hardlockup detector should be
> disabled in KVM guests as it may give false positives. On PPC, hardlockup
> detector is broken inside KVM guests because disable_hardlockup_detector()
Isn't it the opposite? Inside KVM guests, the hardlockup detector should
be *disabled*, but it's not it's *enabled*, due to this bug.
ie. it's not broken, it's working, but that's the bug.
> is marked as early_initcall and it uses is_kvm_guest(), which is
> initialized by check_kvm_guest() later during boot as it is a
> core_initcall. check_kvm_guest() is also called in pSeries_smp_probe(),
> which is called before initcalls, but it is skipped if KVM guest does
> not have doorbell support or if the guest is launched with SMT=1.
I'm wondering how no one has noticed. Most KVM guests have SMT=1.
> Move the check_kvm_guest() call in pSeries_smp_probe() to the initial
> part of function before doorbell/SMT checks so that "kvm_guest" static
> key is initialized by the time disable_hardlockup_detector() runs.
check_kvm_guest() is safe to be called multiple times so
disable_hardlockup_detector() should just call it before it calls
is_kvm_guest(). That should avoid future breakage when the order of
calls changes, or someone refactors pSeries_smp_probe().
Can you identify the commit that broke this and include a Fixes: tag
please.
cheers
> [1]: Documentation/admin-guide/sysctl/kernel.rst
>
> Signed-off-by: Gautam Menghani <gau...@linux.ibm.com>
> ---
> arch/powerpc/platforms/pseries/smp.c | 9 +++++++--
> 1 file changed, 7 insertions(+), 2 deletions(-)
>
> diff --git a/arch/powerpc/platforms/pseries/smp.c
> b/arch/powerpc/platforms/pseries/smp.c
> index c597711ef20a..516c7bfec933 100644
> --- a/arch/powerpc/platforms/pseries/smp.c
> +++ b/arch/powerpc/platforms/pseries/smp.c
> @@ -199,6 +199,13 @@ static __init void pSeries_smp_probe(void)
> else
> xics_smp_probe();
>
> + /*
> + * Make sure this is called regardless of doorbell/SMT status, as
> + * we disable hardlockup detector in an early_initcall where we need to
> + * know KVM status for disabling hardlockup detector in KVM guests.
> + */
> + check_kvm_guest();
> +
> /* No doorbell facility, must use the interrupt controller for IPIs */
> if (!cpu_has_feature(CPU_FTR_DBELL))
> return;
> @@ -207,8 +214,6 @@ static __init void pSeries_smp_probe(void)
> if (!cpu_has_feature(CPU_FTR_SMT))
> return;
>
> - check_kvm_guest();
> -
> if (is_kvm_guest()) {
> /*
> * KVM emulates doorbells by disabling FSCR[MSGP] so msgsndp
> --
> 2.47.0
