[PATCH v2 0/6] KVM: Send reschedule IPIs for non-waiting requests

From: Sean Christopherson

Date: Thu Nov 19 2026 - 18:57:09 EST


My apologies to the s390 folks for dragging you into KVM's mess. Y'all are
Cc'd because I want to use smp_send_reschedule() to fix a KVM bug, and KVM
can be built as a module on s390. I'm hoping to get an Ack on patch 1,
definitely feel free to ignore the rest.

For the KVM changes, fix two bugs with using the SMP function call framework,
by switching to smp_send_reschedule() for non-waiting requests:

1. Making SMP function calls with IRQs disable can lead to deadlock.
2. Batching IPIs for SMP function calls using the per-CPU cpumask can corrupt
the mask if KVM kicks CPUs from hard IRQ context, as arm64 does when
handling KVM_IRQFD signals.

Sean Christopherson (6):
s390/smp: Export arch_smp_send_reschedule() for KVM
KVM: Add struct and helpers to handle kicking multiple (v)CPUs
KVM: Add dedicated flag to track if multi-vCPU request needs at least
one kick
KVM: Don't kick offline CPUs when making a multi-vCPU request
KVM: Send reschedule IPIs instead of SMP function calls for
non-waiting requests
KVM: Extract and dedup the kick code to check if a CPU needs an IPI

arch/s390/kernel/smp.c | 1 +
virt/kvm/kvm_main.c | 101 ++++++++++++++++++++++++-----------------
2 files changed, 61 insertions(+), 41 deletions(-)


base-commit: d21eef2084a9af87ac2a783bc7c2b175ba114e4e
--
2.56.1.708.g77570a6a1d-goog