Re: [PATCH 3/7] KVM: PPC: Explicitly force IPIs when directly kicking vCPUs

From: Sean Christopherson <seanjc@google.com>
To: Gautam Menghani <gautam@linux.ibm.com>
Cc: Marc Zyngier <maz@kernel.org>, Oliver Upton <oupton@kernel.org>,
	 Tianrui Zhao <zhaotianrui@loongson.cn>,
	Bibo Mao <maobibo@loongson.cn>,
	 Huacai Chen <chenhuacai@kernel.org>,
	Madhavan Srinivasan <maddy@linux.ibm.com>,
	 Anup Patel <anup@brainfault.org>, Paul Walmsley <pjw@kernel.org>,
	 Palmer Dabbelt <palmer@dabbelt.com>,
	Albert Ou <aou@eecs.berkeley.edu>,
	 Paolo Bonzini <pbonzini@redhat.com>,
	Fuad Tabba <fuad.tabba@linux.dev>,
	 Joey Gouly <joey.gouly@arm.com>,
	Steffen Eiden <seiden@linux.ibm.com>,
	 Suzuki K Poulose <suzuki.poulose@arm.com>,
	Zenghui Yu <yuzenghui@huawei.com>,
	 Nicholas Piggin <npiggin@gmail.com>,
	Atish Patra <atish.patra@linux.dev>,
	 Alexandre Ghiti <alex@ghiti.fr>,
	linux-arm-kernel@lists.infradead.org,  kvmarm@lists.linux.dev,
	kvm@vger.kernel.org, loongarch@lists.linux.dev,
	 linux-mips@vger.kernel.org, linuxppc-dev@lists.ozlabs.org,
	 kvm-riscv@lists.infradead.org, linux-riscv@lists.infradead.org,
	 linux-kernel@vger.kernel.org,
	Stefan Teodorescu <fane@google.com>
Subject: Re: [PATCH 3/7] KVM: PPC: Explicitly force IPIs when directly kicking vCPUs
Date: Fri, 30 Oct 2026 11:57:10 -0700	[thread overview]
Message-ID: <auTohixp3V5-vNsB@google.com> (raw)
In-Reply-To: <auStM0rbFol79h5o@li-fbf77acc-29ba-11b2-a85c-e8e934f50a46.ibm.com>

On Fri, Oct 30, 2026, Gautam Menghani wrote:
> On Wed, Oct 28, 2026 at 12:02:55PM -0700, Sean Christopherson wrote:
> > Explicitly force an IPI when invoking kvm_vcpu_kick() directly instead of
> > relying on kvm_arch_vcpu_should_kick() to unconditionally return 1/true.
> > This will allow dropping kvm_arch_vcpu_should_kick() in favor of
> > kvm_request_needs_ipi().
> > 
> > No functional change intended (1 || 1 == true).
> > 
> > Signed-off-by: Sean Christopherson <seanjc@google.com>
> > ---
> >  arch/powerpc/include/asm/kvm_ppc.h | 13 ++++++++++++-
> >  arch/powerpc/kvm/book3s.c          |  4 ++--
> >  arch/powerpc/kvm/book3s_pr.c       |  2 +-
> >  arch/powerpc/kvm/booke.c           |  6 +++---
> >  arch/powerpc/kvm/e500_emulate.c    |  2 +-
> >  arch/powerpc/kvm/powerpc.c         |  2 +-
> >  6 files changed, 20 insertions(+), 9 deletions(-)
> > 
> > diff --git a/arch/powerpc/include/asm/kvm_ppc.h b/arch/powerpc/include/asm/kvm_ppc.h
> > index 169ea6a7fbad..bbe7378ad0ef 100644
> > --- a/arch/powerpc/include/asm/kvm_ppc.h
> > +++ b/arch/powerpc/include/asm/kvm_ppc.h
> > @@ -30,6 +30,17 @@
> >  #endif
> >  #include <asm/inst.h>
> >  
> > +/*
> > + * TODO: For unknown reasons, KVM PPC has historically ignored vcpu->mode when
> > + *       determining if kicking a vCPU needs an IPI.  Either document why that
> > + *       behavior is necessary, or drop this helper and the @force param from
> > + *       __kvm_vcpu_kick().
> > + */
> > +static inline void kvmppc_vcpu_kick(struct kvm_vcpu *vcpu)
> > +{
> > +	__kvm_vcpu_kick(vcpu, false, true);
> > +}
> > +
> 
> AFAICS, there is no reason for PPC to always return 1. I was looking at
> the old patches on mailing list and found this - https://lore.kernel.org/kvm/0FEE585F-5E21-47F2-A3F0-AD0DE05EBB43@suse.de
> 
> ```
> How about we return 1 for kvm_arch_vcpu_in_guest_mode() on !x86 always, just
> like it's done today basically? Then we can worry about needless IPIs later
> and don't regress from the respective current kick implementations.
> ```

FWIW, I found that as well, but there are a metric ton of kvm_vcpu_kick() calls
that have been added since then, and they scared me.

> However, I'll do some testing and get back by mid next week to make sure
> there are no suprises.

Thanks!  If we can drop the PPC behavior, that would be wonderful!  Though if
we do that, it should be for 7.5 in a follow-up (this series really should go
into 7.4 due to the SNP issue).